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
;
202 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
206 kaddr
= kmap_atomic(page
, KM_USER0
);
207 memcpy(kaddr
, dibh
->b_data
+ sizeof(struct gfs2_dinode
),
209 memset(kaddr
+ ip
->i_di
.di_size
, 0, PAGE_CACHE_SIZE
- ip
->i_di
.di_size
);
210 kunmap_atomic(kaddr
, KM_USER0
);
214 SetPageUptodate(page
);
221 * gfs2_readpage - readpage with locking
222 * @file: The file to read a page for. N.B. This may be NULL if we are
223 * reading an internal file.
224 * @page: The page to read
229 static int gfs2_readpage(struct file
*file
, struct page
*page
)
231 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
232 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
233 struct gfs2_file
*gf
= NULL
;
234 struct gfs2_holder gh
;
238 if (likely(file
!= &gfs2_internal_file_sentinel
)) {
240 gf
= file
->private_data
;
241 if (test_bit(GFF_EXLOCK
, &gf
->f_flags
))
242 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
245 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
|LM_FLAG_TRY_1CB
, &gh
);
247 error
= gfs2_glock_nq_atime(&gh
);
253 if (gfs2_is_stuffed(ip
)) {
254 error
= stuffed_readpage(ip
, page
);
257 error
= mpage_readpage(page
, gfs2_get_block
);
259 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
263 gfs2_glock_dq_m(1, &gh
);
264 gfs2_holder_uninit(&gh
);
270 if (error
== GLR_TRYFAILED
) {
271 error
= AOP_TRUNCATED_PAGE
;
275 gfs2_holder_uninit(&gh
);
280 * gfs2_readpages - Read a bunch of pages at once
283 * 1. This is only for readahead, so we can simply ignore any things
284 * which are slightly inconvenient (such as locking conflicts between
285 * the page lock and the glock) and return having done no I/O. Its
286 * obviously not something we'd want to do on too regular a basis.
287 * Any I/O we ignore at this time will be done via readpage later.
288 * 2. We don't handle stuffed files here we let readpage do the honours.
289 * 3. mpage_readpages() does most of the heavy lifting in the common case.
290 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
291 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
292 * well as read-ahead.
294 static int gfs2_readpages(struct file
*file
, struct address_space
*mapping
,
295 struct list_head
*pages
, unsigned nr_pages
)
297 struct inode
*inode
= mapping
->host
;
298 struct gfs2_inode
*ip
= GFS2_I(inode
);
299 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
300 struct gfs2_holder gh
;
304 if (likely(file
!= &gfs2_internal_file_sentinel
)) {
306 struct gfs2_file
*gf
= file
->private_data
;
307 if (test_bit(GFF_EXLOCK
, &gf
->f_flags
))
310 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
,
311 LM_FLAG_TRY_1CB
|GL_ATIME
, &gh
);
313 ret
= gfs2_glock_nq_atime(&gh
);
314 if (ret
== GLR_TRYFAILED
)
320 if (!gfs2_is_stuffed(ip
))
321 ret
= mpage_readpages(mapping
, pages
, nr_pages
, gfs2_get_block
);
324 gfs2_glock_dq_m(1, &gh
);
325 gfs2_holder_uninit(&gh
);
328 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
335 gfs2_holder_uninit(&gh
);
340 * gfs2_prepare_write - Prepare to write a page to a file
341 * @file: The file to write to
342 * @page: The page which is to be prepared for writing
343 * @from: From (byte range within page)
344 * @to: To (byte range within page)
349 static int gfs2_prepare_write(struct file
*file
, struct page
*page
,
350 unsigned from
, unsigned to
)
352 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
353 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
354 unsigned int data_blocks
, ind_blocks
, rblocks
;
357 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + from
;
358 loff_t end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
359 struct gfs2_alloc
*al
;
360 unsigned int write_len
= to
- from
;
363 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_ATIME
|LM_FLAG_TRY_1CB
, &ip
->i_gh
);
364 error
= gfs2_glock_nq_atime(&ip
->i_gh
);
365 if (unlikely(error
)) {
366 if (error
== GLR_TRYFAILED
) {
368 error
= AOP_TRUNCATED_PAGE
;
374 gfs2_write_calc_reserv(ip
, write_len
, &data_blocks
, &ind_blocks
);
376 error
= gfs2_write_alloc_required(ip
, pos
, write_len
, &alloc_required
);
381 ip
->i_alloc
.al_requested
= 0;
382 if (alloc_required
) {
383 al
= gfs2_alloc_get(ip
);
385 error
= gfs2_quota_lock(ip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
389 error
= gfs2_quota_check(ip
, ip
->i_inode
.i_uid
, ip
->i_inode
.i_gid
);
393 al
->al_requested
= data_blocks
+ ind_blocks
;
394 error
= gfs2_inplace_reserve(ip
);
399 rblocks
= RES_DINODE
+ ind_blocks
;
400 if (gfs2_is_jdata(ip
))
401 rblocks
+= data_blocks
? data_blocks
: 1;
402 if (ind_blocks
|| data_blocks
)
403 rblocks
+= RES_STATFS
+ RES_QUOTA
;
405 error
= gfs2_trans_begin(sdp
, rblocks
, 0);
409 if (gfs2_is_stuffed(ip
)) {
410 if (end
> sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
)) {
411 error
= gfs2_unstuff_dinode(ip
, page
);
414 } else if (!PageUptodate(page
))
415 error
= stuffed_readpage(ip
, page
);
420 error
= block_prepare_write(page
, from
, to
, gfs2_get_block
);
425 if (alloc_required
) {
426 gfs2_inplace_release(ip
);
428 gfs2_quota_unlock(ip
);
433 gfs2_glock_dq_m(1, &ip
->i_gh
);
435 gfs2_holder_uninit(&ip
->i_gh
);
442 * gfs2_commit_write - Commit write to a file
443 * @file: The file to write to
444 * @page: The page containing the data
445 * @from: From (byte range within page)
446 * @to: To (byte range within page)
451 static int gfs2_commit_write(struct file
*file
, struct page
*page
,
452 unsigned from
, unsigned to
)
454 struct inode
*inode
= page
->mapping
->host
;
455 struct gfs2_inode
*ip
= GFS2_I(inode
);
456 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
457 int error
= -EOPNOTSUPP
;
458 struct buffer_head
*dibh
;
459 struct gfs2_alloc
*al
= &ip
->i_alloc
;
460 struct gfs2_dinode
*di
;
462 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_locked_by_me(ip
->i_gl
)))
465 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
469 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
470 di
= (struct gfs2_dinode
*)dibh
->b_data
;
472 if (gfs2_is_stuffed(ip
)) {
476 file_size
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
478 kaddr
= kmap_atomic(page
, KM_USER0
);
479 memcpy(dibh
->b_data
+ sizeof(struct gfs2_dinode
) + from
,
480 kaddr
+ from
, to
- from
);
481 kunmap_atomic(kaddr
, KM_USER0
);
483 SetPageUptodate(page
);
485 if (inode
->i_size
< file_size
) {
486 i_size_write(inode
, file_size
);
487 mark_inode_dirty(inode
);
490 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
||
492 gfs2_page_add_databufs(ip
, page
, from
, to
);
493 error
= generic_commit_write(file
, page
, from
, to
);
498 if (ip
->i_di
.di_size
< inode
->i_size
) {
499 ip
->i_di
.di_size
= inode
->i_size
;
500 di
->di_size
= cpu_to_be64(inode
->i_size
);
505 if (al
->al_requested
) {
506 gfs2_inplace_release(ip
);
507 gfs2_quota_unlock(ip
);
510 gfs2_glock_dq_m(1, &ip
->i_gh
);
511 gfs2_holder_uninit(&ip
->i_gh
);
518 if (al
->al_requested
) {
519 gfs2_inplace_release(ip
);
520 gfs2_quota_unlock(ip
);
523 gfs2_glock_dq_m(1, &ip
->i_gh
);
524 gfs2_holder_uninit(&ip
->i_gh
);
526 ClearPageUptodate(page
);
531 * gfs2_bmap - Block map function
532 * @mapping: Address space info
533 * @lblock: The block to map
535 * Returns: The disk address for the block or 0 on hole or error
538 static sector_t
gfs2_bmap(struct address_space
*mapping
, sector_t lblock
)
540 struct gfs2_inode
*ip
= GFS2_I(mapping
->host
);
541 struct gfs2_holder i_gh
;
545 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
, &i_gh
);
549 if (!gfs2_is_stuffed(ip
))
550 dblock
= generic_block_bmap(mapping
, lblock
, gfs2_get_block
);
552 gfs2_glock_dq_uninit(&i_gh
);
557 static void discard_buffer(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
)
559 struct gfs2_bufdata
*bd
;
565 bh
->b_private
= NULL
;
567 gfs2_log_unlock(sdp
);
570 clear_buffer_dirty(bh
);
572 clear_buffer_mapped(bh
);
573 clear_buffer_req(bh
);
574 clear_buffer_new(bh
);
575 clear_buffer_delay(bh
);
579 static void gfs2_invalidatepage(struct page
*page
, unsigned long offset
)
581 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
582 struct buffer_head
*head
, *bh
, *next
;
583 unsigned int curr_off
= 0;
585 BUG_ON(!PageLocked(page
));
586 if (!page_has_buffers(page
))
589 bh
= head
= page_buffers(page
);
591 unsigned int next_off
= curr_off
+ bh
->b_size
;
592 next
= bh
->b_this_page
;
594 if (offset
<= curr_off
)
595 discard_buffer(sdp
, bh
);
599 } while (bh
!= head
);
602 try_to_release_page(page
, 0);
608 * gfs2_ok_for_dio - check that dio is valid on this file
611 * @offset: The offset at which we are reading or writing
613 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
614 * 1 (to accept the i/o request)
616 static int gfs2_ok_for_dio(struct gfs2_inode
*ip
, int rw
, loff_t offset
)
619 * Should we return an error here? I can't see that O_DIRECT for
620 * a journaled file makes any sense. For now we'll silently fall
621 * back to buffered I/O, likewise we do the same for stuffed
622 * files since they are (a) small and (b) unaligned.
624 if (gfs2_is_jdata(ip
))
627 if (gfs2_is_stuffed(ip
))
630 if (offset
> i_size_read(&ip
->i_inode
))
637 static ssize_t
gfs2_direct_IO(int rw
, struct kiocb
*iocb
,
638 const struct iovec
*iov
, loff_t offset
,
639 unsigned long nr_segs
)
641 struct file
*file
= iocb
->ki_filp
;
642 struct inode
*inode
= file
->f_mapping
->host
;
643 struct gfs2_inode
*ip
= GFS2_I(inode
);
644 struct gfs2_holder gh
;
648 * Deferred lock, even if its a write, since we do no allocation
649 * on this path. All we need change is atime, and this lock mode
650 * ensures that other nodes have flushed their buffered read caches
651 * (i.e. their page cache entries for this inode). We do not,
652 * unfortunately have the option of only flushing a range like
655 gfs2_holder_init(ip
->i_gl
, LM_ST_DEFERRED
, GL_ATIME
, &gh
);
656 rv
= gfs2_glock_nq_atime(&gh
);
659 rv
= gfs2_ok_for_dio(ip
, rw
, offset
);
661 goto out
; /* dio not valid, fall back to buffered i/o */
663 rv
= blockdev_direct_IO_no_locking(rw
, iocb
, inode
, inode
->i_sb
->s_bdev
,
664 iov
, offset
, nr_segs
,
665 gfs2_get_block_direct
, NULL
);
667 gfs2_glock_dq_m(1, &gh
);
668 gfs2_holder_uninit(&gh
);
673 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
674 * @bh: the buffer we're stuck on
678 static void stuck_releasepage(struct buffer_head
*bh
)
680 struct inode
*inode
= bh
->b_page
->mapping
->host
;
681 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
682 struct gfs2_bufdata
*bd
= bh
->b_private
;
683 struct gfs2_glock
*gl
;
684 static unsigned limit
= 0;
690 fs_warn(sdp
, "stuck in gfs2_releasepage() %p\n", inode
);
691 fs_warn(sdp
, "blkno = %llu, bh->b_count = %d\n",
692 (unsigned long long)bh
->b_blocknr
, atomic_read(&bh
->b_count
));
693 fs_warn(sdp
, "pinned = %u\n", buffer_pinned(bh
));
694 fs_warn(sdp
, "bh->b_private = %s\n", (bd
) ? "!NULL" : "NULL");
701 fs_warn(sdp
, "gl = (%u, %llu)\n",
702 gl
->gl_name
.ln_type
, (unsigned long long)gl
->gl_name
.ln_number
);
704 fs_warn(sdp
, "bd_list_tr = %s, bd_le.le_list = %s\n",
705 (list_empty(&bd
->bd_list_tr
)) ? "no" : "yes",
706 (list_empty(&bd
->bd_le
.le_list
)) ? "no" : "yes");
708 if (gl
->gl_ops
== &gfs2_inode_glops
) {
709 struct gfs2_inode
*ip
= gl
->gl_object
;
715 fs_warn(sdp
, "ip = %llu %llu\n",
716 (unsigned long long)ip
->i_num
.no_formal_ino
,
717 (unsigned long long)ip
->i_num
.no_addr
);
719 for (x
= 0; x
< GFS2_MAX_META_HEIGHT
; x
++)
720 fs_warn(sdp
, "ip->i_cache[%u] = %s\n",
721 x
, (ip
->i_cache
[x
]) ? "!NULL" : "NULL");
726 * gfs2_releasepage - free the metadata associated with a page
727 * @page: the page that's being released
728 * @gfp_mask: passed from Linux VFS, ignored by us
730 * Call try_to_free_buffers() if the buffers in this page can be
736 int gfs2_releasepage(struct page
*page
, gfp_t gfp_mask
)
738 struct inode
*aspace
= page
->mapping
->host
;
739 struct gfs2_sbd
*sdp
= aspace
->i_sb
->s_fs_info
;
740 struct buffer_head
*bh
, *head
;
741 struct gfs2_bufdata
*bd
;
742 unsigned long t
= jiffies
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
;
744 if (!page_has_buffers(page
))
747 head
= bh
= page_buffers(page
);
749 while (atomic_read(&bh
->b_count
)) {
750 if (!atomic_read(&aspace
->i_writecount
))
753 if (!(gfp_mask
& __GFP_WAIT
))
756 if (time_after_eq(jiffies
, t
)) {
757 stuck_releasepage(bh
);
758 /* should we withdraw here? */
765 gfs2_assert_warn(sdp
, !buffer_pinned(bh
));
766 gfs2_assert_warn(sdp
, !buffer_dirty(bh
));
771 gfs2_assert_warn(sdp
, bd
->bd_bh
== bh
);
772 gfs2_assert_warn(sdp
, list_empty(&bd
->bd_list_tr
));
773 gfs2_assert_warn(sdp
, !bd
->bd_ail
);
775 if (!list_empty(&bd
->bd_le
.le_list
))
777 bh
->b_private
= NULL
;
779 gfs2_log_unlock(sdp
);
781 kmem_cache_free(gfs2_bufdata_cachep
, bd
);
783 bh
= bh
->b_this_page
;
784 } while (bh
!= head
);
787 return try_to_free_buffers(page
);
790 const struct address_space_operations gfs2_file_aops
= {
791 .writepage
= gfs2_writepage
,
792 .writepages
= gfs2_writepages
,
793 .readpage
= gfs2_readpage
,
794 .readpages
= gfs2_readpages
,
795 .sync_page
= block_sync_page
,
796 .prepare_write
= gfs2_prepare_write
,
797 .commit_write
= gfs2_commit_write
,
799 .invalidatepage
= gfs2_invalidatepage
,
800 .releasepage
= gfs2_releasepage
,
801 .direct_IO
= gfs2_direct_IO
,