2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3 * Copyright (C) 2004-2007 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/swap.h>
21 #include <linux/gfs2_ondisk.h>
22 #include <linux/lm_interface.h>
23 #include <linux/backing-dev.h>
24 #include <linux/pagevec.h>
33 #include "ops_address.h"
42 static void gfs2_page_add_databufs(struct gfs2_inode
*ip
, struct page
*page
,
43 unsigned int from
, unsigned int to
)
45 struct buffer_head
*head
= page_buffers(page
);
46 unsigned int bsize
= head
->b_size
;
47 struct buffer_head
*bh
;
48 unsigned int start
, end
;
50 for (bh
= head
, start
= 0; bh
!= head
|| !start
;
51 bh
= bh
->b_this_page
, start
= end
) {
53 if (end
<= from
|| start
>= to
)
55 if (gfs2_is_jdata(ip
))
56 set_buffer_uptodate(bh
);
57 gfs2_trans_add_bh(ip
->i_gl
, bh
, 0);
62 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
64 * @lblock: The block number to look up
65 * @bh_result: The buffer head to return the result in
66 * @create: Non-zero if we may add block to the file
71 static int gfs2_get_block_noalloc(struct inode
*inode
, sector_t lblock
,
72 struct buffer_head
*bh_result
, int create
)
76 error
= gfs2_block_map(inode
, lblock
, bh_result
, 0);
79 if (!buffer_mapped(bh_result
))
84 static int gfs2_get_block_direct(struct inode
*inode
, sector_t lblock
,
85 struct buffer_head
*bh_result
, int create
)
87 return gfs2_block_map(inode
, lblock
, bh_result
, 0);
91 * gfs2_writepage_common - Common bits of writepage
92 * @page: The page to be written
93 * @wbc: The writeback control
95 * Returns: 1 if writepage is ok, otherwise an error code or zero if no error.
98 static int gfs2_writepage_common(struct page
*page
,
99 struct writeback_control
*wbc
)
101 struct inode
*inode
= page
->mapping
->host
;
102 struct gfs2_inode
*ip
= GFS2_I(inode
);
103 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
104 loff_t i_size
= i_size_read(inode
);
105 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
109 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_held_excl(ip
->i_gl
)))
112 if (current
->journal_info
)
114 /* Is the page fully outside i_size? (truncate in progress) */
115 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
116 if (page
->index
> end_index
|| (page
->index
== end_index
&& !offset
)) {
117 page
->mapping
->a_ops
->invalidatepage(page
, 0);
122 redirty_page_for_writepage(wbc
, page
);
129 * gfs2_writeback_writepage - Write page for writeback mappings
131 * @wbc: The writeback control
135 static int gfs2_writeback_writepage(struct page
*page
,
136 struct writeback_control
*wbc
)
140 ret
= gfs2_writepage_common(page
, wbc
);
144 ret
= mpage_writepage(page
, gfs2_get_block_noalloc
, wbc
);
146 ret
= block_write_full_page(page
, gfs2_get_block_noalloc
, wbc
);
151 * gfs2_ordered_writepage - Write page for ordered data files
152 * @page: The page to write
153 * @wbc: The writeback control
157 static int gfs2_ordered_writepage(struct page
*page
,
158 struct writeback_control
*wbc
)
160 struct inode
*inode
= page
->mapping
->host
;
161 struct gfs2_inode
*ip
= GFS2_I(inode
);
164 ret
= gfs2_writepage_common(page
, wbc
);
168 if (!page_has_buffers(page
)) {
169 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
170 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
172 gfs2_page_add_databufs(ip
, page
, 0, inode
->i_sb
->s_blocksize
-1);
173 return block_write_full_page(page
, gfs2_get_block_noalloc
, wbc
);
177 * __gfs2_jdata_writepage - The core of jdata writepage
178 * @page: The page to write
179 * @wbc: The writeback control
181 * This is shared between writepage and writepages and implements the
182 * core of the writepage operation. If a transaction is required then
183 * PageChecked will have been set and the transaction will have
184 * already been started before this is called.
187 static int __gfs2_jdata_writepage(struct page
*page
, struct writeback_control
*wbc
)
189 struct inode
*inode
= page
->mapping
->host
;
190 struct gfs2_inode
*ip
= GFS2_I(inode
);
191 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
193 if (PageChecked(page
)) {
194 ClearPageChecked(page
);
195 if (!page_has_buffers(page
)) {
196 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
197 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
199 gfs2_page_add_databufs(ip
, page
, 0, sdp
->sd_vfs
->s_blocksize
-1);
201 return block_write_full_page(page
, gfs2_get_block_noalloc
, wbc
);
205 * gfs2_jdata_writepage - Write complete page
206 * @page: Page to write
212 static int gfs2_jdata_writepage(struct page
*page
, struct writeback_control
*wbc
)
214 struct inode
*inode
= page
->mapping
->host
;
215 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
219 error
= gfs2_writepage_common(page
, wbc
);
223 if (PageChecked(page
)) {
224 if (wbc
->sync_mode
!= WB_SYNC_ALL
)
226 error
= gfs2_trans_begin(sdp
, RES_DINODE
+ 1, 0);
231 error
= __gfs2_jdata_writepage(page
, wbc
);
237 redirty_page_for_writepage(wbc
, page
);
243 * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk
244 * @mapping: The mapping to write
245 * @wbc: Write-back control
247 * For the data=writeback case we can already ignore buffer heads
248 * and write whole extents at once. This is a big reduction in the
249 * number of I/O requests we send and the bmap calls we make in this case.
251 static int gfs2_writeback_writepages(struct address_space
*mapping
,
252 struct writeback_control
*wbc
)
254 return mpage_writepages(mapping
, wbc
, gfs2_get_block_noalloc
);
258 * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages
259 * @mapping: The mapping
260 * @wbc: The writeback control
261 * @writepage: The writepage function to call for each page
262 * @pvec: The vector of pages
263 * @nr_pages: The number of pages to write
265 * Returns: non-zero if loop should terminate, zero otherwise
268 static int gfs2_write_jdata_pagevec(struct address_space
*mapping
,
269 struct writeback_control
*wbc
,
270 struct pagevec
*pvec
,
271 int nr_pages
, pgoff_t end
)
273 struct inode
*inode
= mapping
->host
;
274 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
275 loff_t i_size
= i_size_read(inode
);
276 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
277 unsigned offset
= i_size
& (PAGE_CACHE_SIZE
-1);
278 unsigned nrblocks
= nr_pages
* (PAGE_CACHE_SIZE
/inode
->i_sb
->s_blocksize
);
279 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
283 ret
= gfs2_trans_begin(sdp
, nrblocks
, 0);
287 for(i
= 0; i
< nr_pages
; i
++) {
288 struct page
*page
= pvec
->pages
[i
];
292 if (unlikely(page
->mapping
!= mapping
)) {
297 if (!wbc
->range_cyclic
&& page
->index
> end
) {
303 if (wbc
->sync_mode
!= WB_SYNC_NONE
)
304 wait_on_page_writeback(page
);
306 if (PageWriteback(page
) ||
307 !clear_page_dirty_for_io(page
)) {
312 /* Is the page fully outside i_size? (truncate in progress) */
313 if (page
->index
> end_index
|| (page
->index
== end_index
&& !offset
)) {
314 page
->mapping
->a_ops
->invalidatepage(page
, 0);
319 ret
= __gfs2_jdata_writepage(page
, wbc
);
321 if (ret
|| (--(wbc
->nr_to_write
) <= 0))
323 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
324 wbc
->encountered_congestion
= 1;
334 * gfs2_write_cache_jdata - Like write_cache_pages but different
335 * @mapping: The mapping to write
336 * @wbc: The writeback control
337 * @writepage: The writepage function to call
338 * @data: The data to pass to writepage
340 * The reason that we use our own function here is that we need to
341 * start transactions before we grab page locks. This allows us
342 * to get the ordering right.
345 static int gfs2_write_cache_jdata(struct address_space
*mapping
,
346 struct writeback_control
*wbc
)
348 struct backing_dev_info
*bdi
= mapping
->backing_dev_info
;
358 if (wbc
->nonblocking
&& bdi_write_congested(bdi
)) {
359 wbc
->encountered_congestion
= 1;
363 pagevec_init(&pvec
, 0);
364 if (wbc
->range_cyclic
) {
365 index
= mapping
->writeback_index
; /* Start from prev offset */
368 index
= wbc
->range_start
>> PAGE_CACHE_SHIFT
;
369 end
= wbc
->range_end
>> PAGE_CACHE_SHIFT
;
370 if (wbc
->range_start
== 0 && wbc
->range_end
== LLONG_MAX
)
376 while (!done
&& (index
<= end
) &&
377 (nr_pages
= pagevec_lookup_tag(&pvec
, mapping
, &index
,
379 min(end
- index
, (pgoff_t
)PAGEVEC_SIZE
-1) + 1))) {
381 ret
= gfs2_write_jdata_pagevec(mapping
, wbc
, &pvec
, nr_pages
, end
);
387 pagevec_release(&pvec
);
391 if (!scanned
&& !done
) {
393 * We hit the last page and there is more work to be done: wrap
394 * back to the start of the file
401 if (wbc
->range_cyclic
|| (range_whole
&& wbc
->nr_to_write
> 0))
402 mapping
->writeback_index
= index
;
408 * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk
409 * @mapping: The mapping to write
410 * @wbc: The writeback control
414 static int gfs2_jdata_writepages(struct address_space
*mapping
,
415 struct writeback_control
*wbc
)
417 struct gfs2_inode
*ip
= GFS2_I(mapping
->host
);
418 struct gfs2_sbd
*sdp
= GFS2_SB(mapping
->host
);
421 ret
= gfs2_write_cache_jdata(mapping
, wbc
);
422 if (ret
== 0 && wbc
->sync_mode
== WB_SYNC_ALL
) {
423 gfs2_log_flush(sdp
, ip
->i_gl
);
424 ret
= gfs2_write_cache_jdata(mapping
, wbc
);
430 * stuffed_readpage - Fill in a Linux page with stuffed file data
437 static int stuffed_readpage(struct gfs2_inode
*ip
, struct page
*page
)
439 struct buffer_head
*dibh
;
444 * Due to the order of unstuffing files and ->nopage(), we can be
445 * asked for a zero page in the case of a stuffed file being extended,
446 * so we need to supply one here. It doesn't happen often.
448 if (unlikely(page
->index
)) {
449 zero_user_page(page
, 0, PAGE_CACHE_SIZE
, KM_USER0
);
453 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
457 kaddr
= kmap_atomic(page
, KM_USER0
);
458 memcpy(kaddr
, dibh
->b_data
+ sizeof(struct gfs2_dinode
),
460 memset(kaddr
+ ip
->i_di
.di_size
, 0, PAGE_CACHE_SIZE
- ip
->i_di
.di_size
);
461 kunmap_atomic(kaddr
, KM_USER0
);
462 flush_dcache_page(page
);
464 SetPageUptodate(page
);
471 * __gfs2_readpage - readpage
472 * @file: The file to read a page for
473 * @page: The page to read
475 * This is the core of gfs2's readpage. Its used by the internal file
476 * reading code as in that case we already hold the glock. Also its
477 * called by gfs2_readpage() once the required lock has been granted.
481 static int __gfs2_readpage(void *file
, struct page
*page
)
483 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
484 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
487 if (gfs2_is_stuffed(ip
)) {
488 error
= stuffed_readpage(ip
, page
);
491 error
= mpage_readpage(page
, gfs2_block_map
);
494 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
501 * gfs2_readpage - read a page of a file
502 * @file: The file to read
503 * @page: The page of the file
505 * This deals with the locking required. We use a trylock in order to
506 * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE
507 * in the event that we are unable to get the lock.
510 static int gfs2_readpage(struct file
*file
, struct page
*page
)
512 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
513 struct gfs2_holder gh
;
516 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
|LM_FLAG_TRY_1CB
, &gh
);
517 error
= gfs2_glock_nq_atime(&gh
);
518 if (unlikely(error
)) {
522 error
= __gfs2_readpage(file
, page
);
525 gfs2_holder_uninit(&gh
);
526 if (error
== GLR_TRYFAILED
) {
528 return AOP_TRUNCATED_PAGE
;
534 * gfs2_internal_read - read an internal file
535 * @ip: The gfs2 inode
536 * @ra_state: The readahead state (or NULL for no readahead)
537 * @buf: The buffer to fill
538 * @pos: The file position
539 * @size: The amount to read
543 int gfs2_internal_read(struct gfs2_inode
*ip
, struct file_ra_state
*ra_state
,
544 char *buf
, loff_t
*pos
, unsigned size
)
546 struct address_space
*mapping
= ip
->i_inode
.i_mapping
;
547 unsigned long index
= *pos
/ PAGE_CACHE_SIZE
;
548 unsigned offset
= *pos
& (PAGE_CACHE_SIZE
- 1);
556 if (offset
+ size
> PAGE_CACHE_SIZE
)
557 amt
= PAGE_CACHE_SIZE
- offset
;
558 page
= read_cache_page(mapping
, index
, __gfs2_readpage
, NULL
);
560 return PTR_ERR(page
);
561 p
= kmap_atomic(page
, KM_USER0
);
562 memcpy(buf
+ copied
, p
+ offset
, amt
);
563 kunmap_atomic(p
, KM_USER0
);
564 mark_page_accessed(page
);
565 page_cache_release(page
);
569 } while(copied
< size
);
575 * gfs2_readpages - Read a bunch of pages at once
578 * 1. This is only for readahead, so we can simply ignore any things
579 * which are slightly inconvenient (such as locking conflicts between
580 * the page lock and the glock) and return having done no I/O. Its
581 * obviously not something we'd want to do on too regular a basis.
582 * Any I/O we ignore at this time will be done via readpage later.
583 * 2. We don't handle stuffed files here we let readpage do the honours.
584 * 3. mpage_readpages() does most of the heavy lifting in the common case.
585 * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places.
588 static int gfs2_readpages(struct file
*file
, struct address_space
*mapping
,
589 struct list_head
*pages
, unsigned nr_pages
)
591 struct inode
*inode
= mapping
->host
;
592 struct gfs2_inode
*ip
= GFS2_I(inode
);
593 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
594 struct gfs2_holder gh
;
597 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &gh
);
598 ret
= gfs2_glock_nq_atime(&gh
);
601 if (!gfs2_is_stuffed(ip
))
602 ret
= mpage_readpages(mapping
, pages
, nr_pages
, gfs2_block_map
);
605 gfs2_holder_uninit(&gh
);
606 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
612 * gfs2_write_begin - Begin to write to a file
613 * @file: The file to write to
614 * @mapping: The mapping in which to write
615 * @pos: The file offset at which to start writing
616 * @len: Length of the write
617 * @flags: Various flags
618 * @pagep: Pointer to return the page
619 * @fsdata: Pointer to return fs data (unused by GFS2)
624 static int gfs2_write_begin(struct file
*file
, struct address_space
*mapping
,
625 loff_t pos
, unsigned len
, unsigned flags
,
626 struct page
**pagep
, void **fsdata
)
628 struct gfs2_inode
*ip
= GFS2_I(mapping
->host
);
629 struct gfs2_sbd
*sdp
= GFS2_SB(mapping
->host
);
630 unsigned int data_blocks
, ind_blocks
, rblocks
;
633 struct gfs2_alloc
*al
;
634 pgoff_t index
= pos
>> PAGE_CACHE_SHIFT
;
635 unsigned from
= pos
& (PAGE_CACHE_SIZE
- 1);
636 unsigned to
= from
+ len
;
639 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_ATIME
, &ip
->i_gh
);
640 error
= gfs2_glock_nq_atime(&ip
->i_gh
);
644 gfs2_write_calc_reserv(ip
, len
, &data_blocks
, &ind_blocks
);
645 error
= gfs2_write_alloc_required(ip
, pos
, len
, &alloc_required
);
649 ip
->i_alloc
.al_requested
= 0;
650 if (alloc_required
) {
651 al
= gfs2_alloc_get(ip
);
653 error
= gfs2_quota_lock(ip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
657 error
= gfs2_quota_check(ip
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
661 al
->al_requested
= data_blocks
+ ind_blocks
;
662 error
= gfs2_inplace_reserve(ip
);
667 rblocks
= RES_DINODE
+ ind_blocks
;
668 if (gfs2_is_jdata(ip
))
669 rblocks
+= data_blocks
? data_blocks
: 1;
670 if (ind_blocks
|| data_blocks
)
671 rblocks
+= RES_STATFS
+ RES_QUOTA
;
673 error
= gfs2_trans_begin(sdp
, rblocks
,
674 PAGE_CACHE_SIZE
/sdp
->sd_sb
.sb_bsize
);
679 page
= __grab_cache_page(mapping
, index
);
684 if (gfs2_is_stuffed(ip
)) {
686 if (pos
+ len
> sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
)) {
687 error
= gfs2_unstuff_dinode(ip
, page
);
690 } else if (!PageUptodate(page
)) {
691 error
= stuffed_readpage(ip
, page
);
697 error
= block_prepare_write(page
, from
, to
, gfs2_block_map
);
702 page_cache_release(page
);
703 if (pos
+ len
> ip
->i_inode
.i_size
)
704 vmtruncate(&ip
->i_inode
, ip
->i_inode
.i_size
);
708 if (alloc_required
) {
709 gfs2_inplace_release(ip
);
711 gfs2_quota_unlock(ip
);
716 gfs2_glock_dq(&ip
->i_gh
);
718 gfs2_holder_uninit(&ip
->i_gh
);
723 * adjust_fs_space - Adjusts the free space available due to gfs2_grow
724 * @inode: the rindex inode
726 static void adjust_fs_space(struct inode
*inode
)
728 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
729 struct gfs2_statfs_change_host
*m_sc
= &sdp
->sd_statfs_master
;
730 struct gfs2_statfs_change_host
*l_sc
= &sdp
->sd_statfs_local
;
731 u64 fs_total
, new_free
;
733 /* Total up the file system space, according to the latest rindex. */
734 fs_total
= gfs2_ri_total(sdp
);
736 spin_lock(&sdp
->sd_statfs_spin
);
737 if (fs_total
> (m_sc
->sc_total
+ l_sc
->sc_total
))
738 new_free
= fs_total
- (m_sc
->sc_total
+ l_sc
->sc_total
);
741 spin_unlock(&sdp
->sd_statfs_spin
);
742 fs_warn(sdp
, "File system extended by %llu blocks.\n",
743 (unsigned long long)new_free
);
744 gfs2_statfs_change(sdp
, new_free
, new_free
, 0);
748 * gfs2_stuffed_write_end - Write end for stuffed files
750 * @dibh: The buffer_head containing the on-disk inode
751 * @pos: The file position
752 * @len: The length of the write
753 * @copied: How much was actually copied by the VFS
756 * This copies the data from the page into the inode block after
757 * the inode data structure itself.
761 static int gfs2_stuffed_write_end(struct inode
*inode
, struct buffer_head
*dibh
,
762 loff_t pos
, unsigned len
, unsigned copied
,
765 struct gfs2_inode
*ip
= GFS2_I(inode
);
766 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
767 u64 to
= pos
+ copied
;
769 unsigned char *buf
= dibh
->b_data
+ sizeof(struct gfs2_dinode
);
770 struct gfs2_dinode
*di
= (struct gfs2_dinode
*)dibh
->b_data
;
772 BUG_ON((pos
+ len
) > (dibh
->b_size
- sizeof(struct gfs2_dinode
)));
773 kaddr
= kmap_atomic(page
, KM_USER0
);
774 memcpy(buf
+ pos
, kaddr
+ pos
, copied
);
775 memset(kaddr
+ pos
+ copied
, 0, len
- copied
);
776 flush_dcache_page(page
);
777 kunmap_atomic(kaddr
, KM_USER0
);
779 if (!PageUptodate(page
))
780 SetPageUptodate(page
);
782 page_cache_release(page
);
784 if (inode
->i_size
< to
) {
785 i_size_write(inode
, to
);
786 ip
->i_di
.di_size
= inode
->i_size
;
787 di
->di_size
= cpu_to_be64(inode
->i_size
);
788 mark_inode_dirty(inode
);
791 if (inode
== sdp
->sd_rindex
)
792 adjust_fs_space(inode
);
796 gfs2_glock_dq(&ip
->i_gh
);
797 gfs2_holder_uninit(&ip
->i_gh
);
803 * @file: The file to write to
804 * @mapping: The address space to write to
805 * @pos: The file position
806 * @len: The length of the data
808 * @page: The page that has been written
809 * @fsdata: The fsdata (unused in GFS2)
811 * The main write_end function for GFS2. We have a separate one for
812 * stuffed files as they are slightly different, otherwise we just
813 * put our locking around the VFS provided functions.
818 static int gfs2_write_end(struct file
*file
, struct address_space
*mapping
,
819 loff_t pos
, unsigned len
, unsigned copied
,
820 struct page
*page
, void *fsdata
)
822 struct inode
*inode
= page
->mapping
->host
;
823 struct gfs2_inode
*ip
= GFS2_I(inode
);
824 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
825 struct buffer_head
*dibh
;
826 struct gfs2_alloc
*al
= &ip
->i_alloc
;
827 struct gfs2_dinode
*di
;
828 unsigned int from
= pos
& (PAGE_CACHE_SIZE
- 1);
829 unsigned int to
= from
+ len
;
832 BUG_ON(gfs2_glock_is_locked_by_me(ip
->i_gl
) == 0);
834 ret
= gfs2_meta_inode_buffer(ip
, &dibh
);
837 page_cache_release(page
);
841 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
843 if (gfs2_is_stuffed(ip
))
844 return gfs2_stuffed_write_end(inode
, dibh
, pos
, len
, copied
, page
);
846 if (!gfs2_is_writeback(ip
))
847 gfs2_page_add_databufs(ip
, page
, from
, to
);
849 ret
= generic_write_end(file
, mapping
, pos
, len
, copied
, page
, fsdata
);
851 if (likely(ret
>= 0)) {
853 if ((pos
+ copied
) > inode
->i_size
) {
854 di
= (struct gfs2_dinode
*)dibh
->b_data
;
855 ip
->i_di
.di_size
= inode
->i_size
;
856 di
->di_size
= cpu_to_be64(inode
->i_size
);
857 mark_inode_dirty(inode
);
861 if (inode
== sdp
->sd_rindex
)
862 adjust_fs_space(inode
);
867 if (al
->al_requested
) {
868 gfs2_inplace_release(ip
);
869 gfs2_quota_unlock(ip
);
872 gfs2_glock_dq(&ip
->i_gh
);
873 gfs2_holder_uninit(&ip
->i_gh
);
878 * gfs2_set_page_dirty - Page dirtying function
879 * @page: The page to dirty
881 * Returns: 1 if it dirtyed the page, or 0 otherwise
884 static int gfs2_set_page_dirty(struct page
*page
)
886 SetPageChecked(page
);
887 return __set_page_dirty_buffers(page
);
891 * gfs2_bmap - Block map function
892 * @mapping: Address space info
893 * @lblock: The block to map
895 * Returns: The disk address for the block or 0 on hole or error
898 static sector_t
gfs2_bmap(struct address_space
*mapping
, sector_t lblock
)
900 struct gfs2_inode
*ip
= GFS2_I(mapping
->host
);
901 struct gfs2_holder i_gh
;
905 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
, &i_gh
);
909 if (!gfs2_is_stuffed(ip
))
910 dblock
= generic_block_bmap(mapping
, lblock
, gfs2_block_map
);
912 gfs2_glock_dq_uninit(&i_gh
);
917 static void gfs2_discard(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
)
919 struct gfs2_bufdata
*bd
;
923 clear_buffer_dirty(bh
);
926 if (!list_empty(&bd
->bd_le
.le_list
) && !buffer_pinned(bh
))
927 list_del_init(&bd
->bd_le
.le_list
);
929 gfs2_remove_from_journal(bh
, current
->journal_info
, 0);
932 clear_buffer_mapped(bh
);
933 clear_buffer_req(bh
);
934 clear_buffer_new(bh
);
935 gfs2_log_unlock(sdp
);
939 static void gfs2_invalidatepage(struct page
*page
, unsigned long offset
)
941 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
942 struct buffer_head
*bh
, *head
;
943 unsigned long pos
= 0;
945 BUG_ON(!PageLocked(page
));
947 ClearPageChecked(page
);
948 if (!page_has_buffers(page
))
951 bh
= head
= page_buffers(page
);
954 gfs2_discard(sdp
, bh
);
956 bh
= bh
->b_this_page
;
957 } while (bh
!= head
);
960 try_to_release_page(page
, 0);
964 * gfs2_ok_for_dio - check that dio is valid on this file
967 * @offset: The offset at which we are reading or writing
969 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
970 * 1 (to accept the i/o request)
972 static int gfs2_ok_for_dio(struct gfs2_inode
*ip
, int rw
, loff_t offset
)
975 * Should we return an error here? I can't see that O_DIRECT for
976 * a stuffed file makes any sense. For now we'll silently fall
977 * back to buffered I/O
979 if (gfs2_is_stuffed(ip
))
982 if (offset
> i_size_read(&ip
->i_inode
))
989 static ssize_t
gfs2_direct_IO(int rw
, struct kiocb
*iocb
,
990 const struct iovec
*iov
, loff_t offset
,
991 unsigned long nr_segs
)
993 struct file
*file
= iocb
->ki_filp
;
994 struct inode
*inode
= file
->f_mapping
->host
;
995 struct gfs2_inode
*ip
= GFS2_I(inode
);
996 struct gfs2_holder gh
;
1000 * Deferred lock, even if its a write, since we do no allocation
1001 * on this path. All we need change is atime, and this lock mode
1002 * ensures that other nodes have flushed their buffered read caches
1003 * (i.e. their page cache entries for this inode). We do not,
1004 * unfortunately have the option of only flushing a range like
1007 gfs2_holder_init(ip
->i_gl
, LM_ST_DEFERRED
, GL_ATIME
, &gh
);
1008 rv
= gfs2_glock_nq_atime(&gh
);
1011 rv
= gfs2_ok_for_dio(ip
, rw
, offset
);
1013 goto out
; /* dio not valid, fall back to buffered i/o */
1015 rv
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
1016 iov
, offset
, nr_segs
,
1017 gfs2_get_block_direct
, NULL
);
1019 gfs2_glock_dq_m(1, &gh
);
1020 gfs2_holder_uninit(&gh
);
1025 * gfs2_releasepage - free the metadata associated with a page
1026 * @page: the page that's being released
1027 * @gfp_mask: passed from Linux VFS, ignored by us
1029 * Call try_to_free_buffers() if the buffers in this page can be
1035 int gfs2_releasepage(struct page
*page
, gfp_t gfp_mask
)
1037 struct inode
*aspace
= page
->mapping
->host
;
1038 struct gfs2_sbd
*sdp
= aspace
->i_sb
->s_fs_info
;
1039 struct buffer_head
*bh
, *head
;
1040 struct gfs2_bufdata
*bd
;
1042 if (!page_has_buffers(page
))
1046 head
= bh
= page_buffers(page
);
1048 if (atomic_read(&bh
->b_count
))
1049 goto cannot_release
;
1051 if (bd
&& bd
->bd_ail
)
1052 goto cannot_release
;
1053 gfs2_assert_warn(sdp
, !buffer_pinned(bh
));
1054 gfs2_assert_warn(sdp
, !buffer_dirty(bh
));
1055 bh
= bh
->b_this_page
;
1056 } while(bh
!= head
);
1057 gfs2_log_unlock(sdp
);
1059 head
= bh
= page_buffers(page
);
1064 gfs2_assert_warn(sdp
, bd
->bd_bh
== bh
);
1065 gfs2_assert_warn(sdp
, list_empty(&bd
->bd_list_tr
));
1066 if (!list_empty(&bd
->bd_le
.le_list
)) {
1067 if (!buffer_pinned(bh
))
1068 list_del_init(&bd
->bd_le
.le_list
);
1074 bh
->b_private
= NULL
;
1076 gfs2_log_unlock(sdp
);
1078 kmem_cache_free(gfs2_bufdata_cachep
, bd
);
1080 bh
= bh
->b_this_page
;
1081 } while (bh
!= head
);
1083 return try_to_free_buffers(page
);
1085 gfs2_log_unlock(sdp
);
1089 static const struct address_space_operations gfs2_writeback_aops
= {
1090 .writepage
= gfs2_writeback_writepage
,
1091 .writepages
= gfs2_writeback_writepages
,
1092 .readpage
= gfs2_readpage
,
1093 .readpages
= gfs2_readpages
,
1094 .sync_page
= block_sync_page
,
1095 .write_begin
= gfs2_write_begin
,
1096 .write_end
= gfs2_write_end
,
1098 .invalidatepage
= gfs2_invalidatepage
,
1099 .releasepage
= gfs2_releasepage
,
1100 .direct_IO
= gfs2_direct_IO
,
1103 static const struct address_space_operations gfs2_ordered_aops
= {
1104 .writepage
= gfs2_ordered_writepage
,
1105 .readpage
= gfs2_readpage
,
1106 .readpages
= gfs2_readpages
,
1107 .sync_page
= block_sync_page
,
1108 .write_begin
= gfs2_write_begin
,
1109 .write_end
= gfs2_write_end
,
1110 .set_page_dirty
= gfs2_set_page_dirty
,
1112 .invalidatepage
= gfs2_invalidatepage
,
1113 .releasepage
= gfs2_releasepage
,
1114 .direct_IO
= gfs2_direct_IO
,
1117 static const struct address_space_operations gfs2_jdata_aops
= {
1118 .writepage
= gfs2_jdata_writepage
,
1119 .writepages
= gfs2_jdata_writepages
,
1120 .readpage
= gfs2_readpage
,
1121 .readpages
= gfs2_readpages
,
1122 .sync_page
= block_sync_page
,
1123 .write_begin
= gfs2_write_begin
,
1124 .write_end
= gfs2_write_end
,
1125 .set_page_dirty
= gfs2_set_page_dirty
,
1127 .invalidatepage
= gfs2_invalidatepage
,
1128 .releasepage
= gfs2_releasepage
,
1131 void gfs2_set_aops(struct inode
*inode
)
1133 struct gfs2_inode
*ip
= GFS2_I(inode
);
1135 if (gfs2_is_writeback(ip
))
1136 inode
->i_mapping
->a_ops
= &gfs2_writeback_aops
;
1137 else if (gfs2_is_ordered(ip
))
1138 inode
->i_mapping
->a_ops
= &gfs2_ordered_aops
;
1139 else if (gfs2_is_jdata(ip
))
1140 inode
->i_mapping
->a_ops
= &gfs2_jdata_aops
;