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/pagemap.h>
16 #include <linux/pagevec.h>
17 #include <linux/mpage.h>
19 #include <linux/writeback.h>
20 #include <linux/gfs2_ondisk.h>
21 #include <linux/lm_interface.h>
30 #include "ops_address.h"
39 static void gfs2_page_add_databufs(struct gfs2_inode
*ip
, struct page
*page
,
40 unsigned int from
, unsigned int to
)
42 struct buffer_head
*head
= page_buffers(page
);
43 unsigned int bsize
= head
->b_size
;
44 struct buffer_head
*bh
;
45 unsigned int start
, end
;
47 for (bh
= head
, start
= 0; bh
!= head
|| !start
;
48 bh
= bh
->b_this_page
, start
= end
) {
50 if (end
<= from
|| start
>= to
)
52 gfs2_trans_add_bh(ip
->i_gl
, bh
, 0);
57 * gfs2_get_block - Fills in a buffer head with details about a block
59 * @lblock: The block number to look up
60 * @bh_result: The buffer head to return the result in
61 * @create: Non-zero if we may add block to the file
66 int gfs2_get_block(struct inode
*inode
, sector_t lblock
,
67 struct buffer_head
*bh_result
, int create
)
69 return gfs2_block_map(inode
, lblock
, create
, bh_result
);
73 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
75 * @lblock: The block number to look up
76 * @bh_result: The buffer head to return the result in
77 * @create: Non-zero if we may add block to the file
82 static int gfs2_get_block_noalloc(struct inode
*inode
, sector_t lblock
,
83 struct buffer_head
*bh_result
, int create
)
87 error
= gfs2_block_map(inode
, lblock
, 0, bh_result
);
90 if (bh_result
->b_blocknr
== 0)
95 static int gfs2_get_block_direct(struct inode
*inode
, sector_t lblock
,
96 struct buffer_head
*bh_result
, int create
)
98 return gfs2_block_map(inode
, lblock
, 0, bh_result
);
102 * gfs2_writepage - Write complete page
103 * @page: Page to write
107 * Some of this is copied from block_write_full_page() although we still
108 * call it to do most of the work.
111 static int gfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
113 struct inode
*inode
= page
->mapping
->host
;
114 struct gfs2_inode
*ip
= GFS2_I(inode
);
115 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
116 loff_t i_size
= i_size_read(inode
);
117 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
122 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_held_excl(ip
->i_gl
))) {
126 if (current
->journal_info
)
129 /* Is the page fully outside i_size? (truncate in progress) */
130 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
131 if (page
->index
> end_index
|| (page
->index
== end_index
&& !offset
)) {
132 page
->mapping
->a_ops
->invalidatepage(page
, 0);
134 return 0; /* don't care */
137 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
|| gfs2_is_jdata(ip
)) {
138 error
= gfs2_trans_begin(sdp
, RES_DINODE
+ 1, 0);
141 if (!page_has_buffers(page
)) {
142 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
143 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
145 gfs2_page_add_databufs(ip
, page
, 0, sdp
->sd_vfs
->s_blocksize
-1);
148 error
= block_write_full_page(page
, gfs2_get_block_noalloc
, wbc
);
151 gfs2_meta_cache_flush(ip
);
155 redirty_page_for_writepage(wbc
, page
);
161 * gfs2_writepages - Write a bunch of dirty pages back to disk
162 * @mapping: The mapping to write
163 * @wbc: Write-back control
165 * For journaled files and/or ordered writes this just falls back to the
166 * kernel's default writepages path for now. We will probably want to change
167 * that eventually (i.e. when we look at allocate on flush).
169 * For the data=writeback case though we can already ignore buffer heads
170 * and write whole extents at once. This is a big reduction in the
171 * number of I/O requests we send and the bmap calls we make in this case.
173 static int gfs2_writepages(struct address_space
*mapping
,
174 struct writeback_control
*wbc
)
176 struct inode
*inode
= mapping
->host
;
177 struct gfs2_inode
*ip
= GFS2_I(inode
);
178 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
180 if (sdp
->sd_args
.ar_data
== GFS2_DATA_WRITEBACK
&& !gfs2_is_jdata(ip
))
181 return mpage_writepages(mapping
, wbc
, gfs2_get_block_noalloc
);
183 return generic_writepages(mapping
, wbc
);
187 * stuffed_readpage - Fill in a Linux page with stuffed file data
194 static int stuffed_readpage(struct gfs2_inode
*ip
, struct page
*page
)
196 struct buffer_head
*dibh
;
201 * Due to the order of unstuffing files and ->nopage(), we can be
202 * asked for a zero page in the case of a stuffed file being extended,
203 * so we need to supply one here. It doesn't happen often.
205 if (unlikely(page
->index
)) {
206 kaddr
= kmap_atomic(page
, KM_USER0
);
207 memset(kaddr
, 0, PAGE_CACHE_SIZE
);
208 kunmap_atomic(kaddr
, KM_USER0
);
209 flush_dcache_page(page
);
210 SetPageUptodate(page
);
214 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
218 kaddr
= kmap_atomic(page
, KM_USER0
);
219 memcpy(kaddr
, dibh
->b_data
+ sizeof(struct gfs2_dinode
),
221 memset(kaddr
+ ip
->i_di
.di_size
, 0, PAGE_CACHE_SIZE
- ip
->i_di
.di_size
);
222 kunmap_atomic(kaddr
, KM_USER0
);
223 flush_dcache_page(page
);
225 SetPageUptodate(page
);
232 * gfs2_readpage - readpage with locking
233 * @file: The file to read a page for. N.B. This may be NULL if we are
234 * reading an internal file.
235 * @page: The page to read
240 static int gfs2_readpage(struct file
*file
, struct page
*page
)
242 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
243 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
244 struct gfs2_file
*gf
= NULL
;
245 struct gfs2_holder gh
;
249 if (likely(file
!= &gfs2_internal_file_sentinel
)) {
251 gf
= file
->private_data
;
252 if (test_bit(GFF_EXLOCK
, &gf
->f_flags
))
253 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
256 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
|LM_FLAG_TRY_1CB
, &gh
);
258 error
= gfs2_glock_nq_atime(&gh
);
264 if (gfs2_is_stuffed(ip
)) {
265 error
= stuffed_readpage(ip
, page
);
268 error
= mpage_readpage(page
, gfs2_get_block
);
270 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
274 gfs2_glock_dq_m(1, &gh
);
275 gfs2_holder_uninit(&gh
);
281 if (error
== GLR_TRYFAILED
) {
282 error
= AOP_TRUNCATED_PAGE
;
286 gfs2_holder_uninit(&gh
);
291 * gfs2_readpages - Read a bunch of pages at once
294 * 1. This is only for readahead, so we can simply ignore any things
295 * which are slightly inconvenient (such as locking conflicts between
296 * the page lock and the glock) and return having done no I/O. Its
297 * obviously not something we'd want to do on too regular a basis.
298 * Any I/O we ignore at this time will be done via readpage later.
299 * 2. We don't handle stuffed files here we let readpage do the honours.
300 * 3. mpage_readpages() does most of the heavy lifting in the common case.
301 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
302 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
303 * well as read-ahead.
305 static int gfs2_readpages(struct file
*file
, struct address_space
*mapping
,
306 struct list_head
*pages
, unsigned nr_pages
)
308 struct inode
*inode
= mapping
->host
;
309 struct gfs2_inode
*ip
= GFS2_I(inode
);
310 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
311 struct gfs2_holder gh
;
315 if (likely(file
!= &gfs2_internal_file_sentinel
)) {
317 struct gfs2_file
*gf
= file
->private_data
;
318 if (test_bit(GFF_EXLOCK
, &gf
->f_flags
))
321 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
,
322 LM_FLAG_TRY_1CB
|GL_ATIME
, &gh
);
324 ret
= gfs2_glock_nq_atime(&gh
);
325 if (ret
== GLR_TRYFAILED
)
331 if (!gfs2_is_stuffed(ip
))
332 ret
= mpage_readpages(mapping
, pages
, nr_pages
, gfs2_get_block
);
335 gfs2_glock_dq_m(1, &gh
);
336 gfs2_holder_uninit(&gh
);
339 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
346 gfs2_holder_uninit(&gh
);
351 * gfs2_prepare_write - Prepare to write a page to a file
352 * @file: The file to write to
353 * @page: The page which is to be prepared for writing
354 * @from: From (byte range within page)
355 * @to: To (byte range within page)
360 static int gfs2_prepare_write(struct file
*file
, struct page
*page
,
361 unsigned from
, unsigned to
)
363 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
364 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
365 unsigned int data_blocks
, ind_blocks
, rblocks
;
368 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + from
;
369 loff_t end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
370 struct gfs2_alloc
*al
;
371 unsigned int write_len
= to
- from
;
374 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_ATIME
|LM_FLAG_TRY_1CB
, &ip
->i_gh
);
375 error
= gfs2_glock_nq_atime(&ip
->i_gh
);
376 if (unlikely(error
)) {
377 if (error
== GLR_TRYFAILED
) {
379 error
= AOP_TRUNCATED_PAGE
;
385 gfs2_write_calc_reserv(ip
, write_len
, &data_blocks
, &ind_blocks
);
387 error
= gfs2_write_alloc_required(ip
, pos
, write_len
, &alloc_required
);
392 ip
->i_alloc
.al_requested
= 0;
393 if (alloc_required
) {
394 al
= gfs2_alloc_get(ip
);
396 error
= gfs2_quota_lock(ip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
400 error
= gfs2_quota_check(ip
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
404 al
->al_requested
= data_blocks
+ ind_blocks
;
405 error
= gfs2_inplace_reserve(ip
);
410 rblocks
= RES_DINODE
+ ind_blocks
;
411 if (gfs2_is_jdata(ip
))
412 rblocks
+= data_blocks
? data_blocks
: 1;
413 if (ind_blocks
|| data_blocks
)
414 rblocks
+= RES_STATFS
+ RES_QUOTA
;
416 error
= gfs2_trans_begin(sdp
, rblocks
, 0);
420 if (gfs2_is_stuffed(ip
)) {
421 if (end
> sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
)) {
422 error
= gfs2_unstuff_dinode(ip
, page
);
425 } else if (!PageUptodate(page
))
426 error
= stuffed_readpage(ip
, page
);
431 error
= block_prepare_write(page
, from
, to
, gfs2_get_block
);
436 if (alloc_required
) {
437 gfs2_inplace_release(ip
);
439 gfs2_quota_unlock(ip
);
444 gfs2_glock_dq_m(1, &ip
->i_gh
);
446 gfs2_holder_uninit(&ip
->i_gh
);
453 * gfs2_commit_write - Commit write to a file
454 * @file: The file to write to
455 * @page: The page containing the data
456 * @from: From (byte range within page)
457 * @to: To (byte range within page)
462 static int gfs2_commit_write(struct file
*file
, struct page
*page
,
463 unsigned from
, unsigned to
)
465 struct inode
*inode
= page
->mapping
->host
;
466 struct gfs2_inode
*ip
= GFS2_I(inode
);
467 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
468 int error
= -EOPNOTSUPP
;
469 struct buffer_head
*dibh
;
470 struct gfs2_alloc
*al
= &ip
->i_alloc
;
471 struct gfs2_dinode
*di
;
473 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_locked_by_me(ip
->i_gl
)))
476 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
480 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
481 di
= (struct gfs2_dinode
*)dibh
->b_data
;
483 if (gfs2_is_stuffed(ip
)) {
487 file_size
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
489 kaddr
= kmap_atomic(page
, KM_USER0
);
490 memcpy(dibh
->b_data
+ sizeof(struct gfs2_dinode
) + from
,
491 kaddr
+ from
, to
- from
);
492 kunmap_atomic(kaddr
, KM_USER0
);
494 SetPageUptodate(page
);
496 if (inode
->i_size
< file_size
) {
497 i_size_write(inode
, file_size
);
498 mark_inode_dirty(inode
);
501 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
||
503 gfs2_page_add_databufs(ip
, page
, from
, to
);
504 error
= generic_commit_write(file
, page
, from
, to
);
509 if (ip
->i_di
.di_size
< inode
->i_size
) {
510 ip
->i_di
.di_size
= inode
->i_size
;
511 di
->di_size
= cpu_to_be64(inode
->i_size
);
516 if (al
->al_requested
) {
517 gfs2_inplace_release(ip
);
518 gfs2_quota_unlock(ip
);
522 gfs2_glock_dq_m(1, &ip
->i_gh
);
524 gfs2_holder_uninit(&ip
->i_gh
);
531 if (al
->al_requested
) {
532 gfs2_inplace_release(ip
);
533 gfs2_quota_unlock(ip
);
537 gfs2_glock_dq_m(1, &ip
->i_gh
);
539 gfs2_holder_uninit(&ip
->i_gh
);
541 ClearPageUptodate(page
);
546 * gfs2_bmap - Block map function
547 * @mapping: Address space info
548 * @lblock: The block to map
550 * Returns: The disk address for the block or 0 on hole or error
553 static sector_t
gfs2_bmap(struct address_space
*mapping
, sector_t lblock
)
555 struct gfs2_inode
*ip
= GFS2_I(mapping
->host
);
556 struct gfs2_holder i_gh
;
560 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
, &i_gh
);
564 if (!gfs2_is_stuffed(ip
))
565 dblock
= generic_block_bmap(mapping
, lblock
, gfs2_get_block
);
567 gfs2_glock_dq_uninit(&i_gh
);
572 static void discard_buffer(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
)
574 struct gfs2_bufdata
*bd
;
580 bh
->b_private
= NULL
;
582 gfs2_log_unlock(sdp
);
585 clear_buffer_dirty(bh
);
587 clear_buffer_mapped(bh
);
588 clear_buffer_req(bh
);
589 clear_buffer_new(bh
);
590 clear_buffer_delay(bh
);
594 static void gfs2_invalidatepage(struct page
*page
, unsigned long offset
)
596 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
597 struct buffer_head
*head
, *bh
, *next
;
598 unsigned int curr_off
= 0;
600 BUG_ON(!PageLocked(page
));
601 if (!page_has_buffers(page
))
604 bh
= head
= page_buffers(page
);
606 unsigned int next_off
= curr_off
+ bh
->b_size
;
607 next
= bh
->b_this_page
;
609 if (offset
<= curr_off
)
610 discard_buffer(sdp
, bh
);
614 } while (bh
!= head
);
617 try_to_release_page(page
, 0);
623 * gfs2_ok_for_dio - check that dio is valid on this file
626 * @offset: The offset at which we are reading or writing
628 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
629 * 1 (to accept the i/o request)
631 static int gfs2_ok_for_dio(struct gfs2_inode
*ip
, int rw
, loff_t offset
)
634 * Should we return an error here? I can't see that O_DIRECT for
635 * a journaled file makes any sense. For now we'll silently fall
636 * back to buffered I/O, likewise we do the same for stuffed
637 * files since they are (a) small and (b) unaligned.
639 if (gfs2_is_jdata(ip
))
642 if (gfs2_is_stuffed(ip
))
645 if (offset
> i_size_read(&ip
->i_inode
))
652 static ssize_t
gfs2_direct_IO(int rw
, struct kiocb
*iocb
,
653 const struct iovec
*iov
, loff_t offset
,
654 unsigned long nr_segs
)
656 struct file
*file
= iocb
->ki_filp
;
657 struct inode
*inode
= file
->f_mapping
->host
;
658 struct gfs2_inode
*ip
= GFS2_I(inode
);
659 struct gfs2_holder gh
;
663 * Deferred lock, even if its a write, since we do no allocation
664 * on this path. All we need change is atime, and this lock mode
665 * ensures that other nodes have flushed their buffered read caches
666 * (i.e. their page cache entries for this inode). We do not,
667 * unfortunately have the option of only flushing a range like
670 gfs2_holder_init(ip
->i_gl
, LM_ST_DEFERRED
, GL_ATIME
, &gh
);
671 rv
= gfs2_glock_nq_atime(&gh
);
674 rv
= gfs2_ok_for_dio(ip
, rw
, offset
);
676 goto out
; /* dio not valid, fall back to buffered i/o */
678 rv
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
679 iov
, offset
, nr_segs
,
680 gfs2_get_block_direct
, NULL
);
682 gfs2_glock_dq_m(1, &gh
);
683 gfs2_holder_uninit(&gh
);
688 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
689 * @bh: the buffer we're stuck on
693 static void stuck_releasepage(struct buffer_head
*bh
)
695 struct inode
*inode
= bh
->b_page
->mapping
->host
;
696 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
697 struct gfs2_bufdata
*bd
= bh
->b_private
;
698 struct gfs2_glock
*gl
;
699 static unsigned limit
= 0;
705 fs_warn(sdp
, "stuck in gfs2_releasepage() %p\n", inode
);
706 fs_warn(sdp
, "blkno = %llu, bh->b_count = %d\n",
707 (unsigned long long)bh
->b_blocknr
, atomic_read(&bh
->b_count
));
708 fs_warn(sdp
, "pinned = %u\n", buffer_pinned(bh
));
709 fs_warn(sdp
, "bh->b_private = %s\n", (bd
) ? "!NULL" : "NULL");
716 fs_warn(sdp
, "gl = (%u, %llu)\n",
717 gl
->gl_name
.ln_type
, (unsigned long long)gl
->gl_name
.ln_number
);
719 fs_warn(sdp
, "bd_list_tr = %s, bd_le.le_list = %s\n",
720 (list_empty(&bd
->bd_list_tr
)) ? "no" : "yes",
721 (list_empty(&bd
->bd_le
.le_list
)) ? "no" : "yes");
723 if (gl
->gl_ops
== &gfs2_inode_glops
) {
724 struct gfs2_inode
*ip
= gl
->gl_object
;
730 fs_warn(sdp
, "ip = %llu %llu\n",
731 (unsigned long long)ip
->i_num
.no_formal_ino
,
732 (unsigned long long)ip
->i_num
.no_addr
);
734 for (x
= 0; x
< GFS2_MAX_META_HEIGHT
; x
++)
735 fs_warn(sdp
, "ip->i_cache[%u] = %s\n",
736 x
, (ip
->i_cache
[x
]) ? "!NULL" : "NULL");
741 * gfs2_releasepage - free the metadata associated with a page
742 * @page: the page that's being released
743 * @gfp_mask: passed from Linux VFS, ignored by us
745 * Call try_to_free_buffers() if the buffers in this page can be
751 int gfs2_releasepage(struct page
*page
, gfp_t gfp_mask
)
753 struct inode
*aspace
= page
->mapping
->host
;
754 struct gfs2_sbd
*sdp
= aspace
->i_sb
->s_fs_info
;
755 struct buffer_head
*bh
, *head
;
756 struct gfs2_bufdata
*bd
;
757 unsigned long t
= jiffies
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
;
759 if (!page_has_buffers(page
))
762 head
= bh
= page_buffers(page
);
764 while (atomic_read(&bh
->b_count
)) {
765 if (!atomic_read(&aspace
->i_writecount
))
768 if (!(gfp_mask
& __GFP_WAIT
))
771 if (time_after_eq(jiffies
, t
)) {
772 stuck_releasepage(bh
);
773 /* should we withdraw here? */
780 gfs2_assert_warn(sdp
, !buffer_pinned(bh
));
781 gfs2_assert_warn(sdp
, !buffer_dirty(bh
));
786 gfs2_assert_warn(sdp
, bd
->bd_bh
== bh
);
787 gfs2_assert_warn(sdp
, list_empty(&bd
->bd_list_tr
));
788 gfs2_assert_warn(sdp
, !bd
->bd_ail
);
790 if (!list_empty(&bd
->bd_le
.le_list
))
792 bh
->b_private
= NULL
;
794 gfs2_log_unlock(sdp
);
796 kmem_cache_free(gfs2_bufdata_cachep
, bd
);
798 bh
= bh
->b_this_page
;
799 } while (bh
!= head
);
802 return try_to_free_buffers(page
);
805 const struct address_space_operations gfs2_file_aops
= {
806 .writepage
= gfs2_writepage
,
807 .writepages
= gfs2_writepages
,
808 .readpage
= gfs2_readpage
,
809 .readpages
= gfs2_readpages
,
810 .sync_page
= block_sync_page
,
811 .prepare_write
= gfs2_prepare_write
,
812 .commit_write
= gfs2_commit_write
,
814 .invalidatepage
= gfs2_invalidatepage
,
815 .releasepage
= gfs2_releasepage
,
816 .direct_IO
= gfs2_direct_IO
,