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/gfs2_ondisk.h>
20 #include <linux/lm_interface.h>
29 #include "ops_address.h"
38 static void gfs2_page_add_databufs(struct gfs2_inode
*ip
, struct page
*page
,
39 unsigned int from
, unsigned int to
)
41 struct buffer_head
*head
= page_buffers(page
);
42 unsigned int bsize
= head
->b_size
;
43 struct buffer_head
*bh
;
44 unsigned int start
, end
;
46 for (bh
= head
, start
= 0; bh
!= head
|| !start
;
47 bh
= bh
->b_this_page
, start
= end
) {
49 if (end
<= from
|| start
>= to
)
51 gfs2_trans_add_bh(ip
->i_gl
, bh
, 0);
56 * gfs2_get_block - Fills in a buffer head with details about a block
58 * @lblock: The block number to look up
59 * @bh_result: The buffer head to return the result in
60 * @create: Non-zero if we may add block to the file
65 int gfs2_get_block(struct inode
*inode
, sector_t lblock
,
66 struct buffer_head
*bh_result
, int create
)
68 return gfs2_block_map(inode
, lblock
, create
, bh_result
);
72 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
74 * @lblock: The block number to look up
75 * @bh_result: The buffer head to return the result in
76 * @create: Non-zero if we may add block to the file
81 static int gfs2_get_block_noalloc(struct inode
*inode
, sector_t lblock
,
82 struct buffer_head
*bh_result
, int create
)
86 error
= gfs2_block_map(inode
, lblock
, 0, bh_result
);
89 if (bh_result
->b_blocknr
== 0)
94 static int gfs2_get_block_direct(struct inode
*inode
, sector_t lblock
,
95 struct buffer_head
*bh_result
, int create
)
97 return gfs2_block_map(inode
, lblock
, 0, bh_result
);
101 * gfs2_writepage - Write complete page
102 * @page: Page to write
106 * Some of this is copied from block_write_full_page() although we still
107 * call it to do most of the work.
110 static int gfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
112 struct inode
*inode
= page
->mapping
->host
;
113 struct gfs2_inode
*ip
= GFS2_I(inode
);
114 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
115 loff_t i_size
= i_size_read(inode
);
116 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
121 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_held_excl(ip
->i_gl
))) {
125 if (current
->journal_info
)
128 /* Is the page fully outside i_size? (truncate in progress) */
129 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
130 if (page
->index
> end_index
|| (page
->index
== end_index
&& !offset
)) {
131 page
->mapping
->a_ops
->invalidatepage(page
, 0);
133 return 0; /* don't care */
136 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
|| gfs2_is_jdata(ip
)) {
137 error
= gfs2_trans_begin(sdp
, RES_DINODE
+ 1, 0);
140 if (!page_has_buffers(page
)) {
141 create_empty_buffers(page
, inode
->i_sb
->s_blocksize
,
142 (1 << BH_Dirty
)|(1 << BH_Uptodate
));
144 gfs2_page_add_databufs(ip
, page
, 0, sdp
->sd_vfs
->s_blocksize
-1);
147 error
= block_write_full_page(page
, gfs2_get_block_noalloc
, wbc
);
150 gfs2_meta_cache_flush(ip
);
154 redirty_page_for_writepage(wbc
, page
);
159 static int zero_readpage(struct page
*page
)
163 kaddr
= kmap_atomic(page
, KM_USER0
);
164 memset(kaddr
, 0, PAGE_CACHE_SIZE
);
165 kunmap_atomic(kaddr
, KM_USER0
);
167 SetPageUptodate(page
);
173 * stuffed_readpage - Fill in a Linux page with stuffed file data
180 static int stuffed_readpage(struct gfs2_inode
*ip
, struct page
*page
)
182 struct buffer_head
*dibh
;
186 /* Only the first page of a stuffed file might contain data */
187 if (unlikely(page
->index
))
188 return zero_readpage(page
);
190 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
194 kaddr
= kmap_atomic(page
, KM_USER0
);
195 memcpy(kaddr
, dibh
->b_data
+ sizeof(struct gfs2_dinode
),
197 memset(kaddr
+ ip
->i_di
.di_size
, 0, PAGE_CACHE_SIZE
- ip
->i_di
.di_size
);
198 kunmap_atomic(kaddr
, KM_USER0
);
202 SetPageUptodate(page
);
209 * gfs2_readpage - readpage with locking
210 * @file: The file to read a page for. N.B. This may be NULL if we are
211 * reading an internal file.
212 * @page: The page to read
217 static int gfs2_readpage(struct file
*file
, struct page
*page
)
219 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
220 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
221 struct gfs2_file
*gf
= NULL
;
222 struct gfs2_holder gh
;
226 if (likely(file
!= &gfs2_internal_file_sentinel
)) {
228 gf
= file
->private_data
;
229 if (test_bit(GFF_EXLOCK
, &gf
->f_flags
))
230 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
233 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
|GL_AOP
, &gh
);
235 error
= gfs2_glock_nq_m_atime(1, &gh
);
241 if (gfs2_is_stuffed(ip
)) {
242 error
= stuffed_readpage(ip
, page
);
245 error
= mpage_readpage(page
, gfs2_get_block
);
247 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
251 gfs2_glock_dq_m(1, &gh
);
252 gfs2_holder_uninit(&gh
);
259 gfs2_holder_uninit(&gh
);
264 * gfs2_readpages - Read a bunch of pages at once
267 * 1. This is only for readahead, so we can simply ignore any things
268 * which are slightly inconvenient (such as locking conflicts between
269 * the page lock and the glock) and return having done no I/O. Its
270 * obviously not something we'd want to do on too regular a basis.
271 * Any I/O we ignore at this time will be done via readpage later.
272 * 2. We have to handle stuffed files here too.
273 * 3. mpage_readpages() does most of the heavy lifting in the common case.
274 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
275 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
276 * well as read-ahead.
278 static int gfs2_readpages(struct file
*file
, struct address_space
*mapping
,
279 struct list_head
*pages
, unsigned nr_pages
)
281 struct inode
*inode
= mapping
->host
;
282 struct gfs2_inode
*ip
= GFS2_I(inode
);
283 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
284 struct gfs2_holder gh
;
289 if (likely(file
!= &gfs2_internal_file_sentinel
)) {
291 struct gfs2_file
*gf
= file
->private_data
;
292 if (test_bit(GFF_EXLOCK
, &gf
->f_flags
))
295 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
,
296 LM_FLAG_TRY_1CB
|GL_ATIME
|GL_AOP
, &gh
);
298 ret
= gfs2_glock_nq_m_atime(1, &gh
);
299 if (ret
== GLR_TRYFAILED
)
305 if (gfs2_is_stuffed(ip
)) {
306 struct pagevec lru_pvec
;
307 pagevec_init(&lru_pvec
, 0);
308 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
309 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
310 prefetchw(&page
->flags
);
311 list_del(&page
->lru
);
312 if (!add_to_page_cache(page
, mapping
,
313 page
->index
, GFP_KERNEL
)) {
314 ret
= stuffed_readpage(ip
, page
);
316 if (!pagevec_add(&lru_pvec
, page
))
317 __pagevec_lru_add(&lru_pvec
);
319 page_cache_release(page
);
322 pagevec_lru_add(&lru_pvec
);
325 /* What we really want to do .... */
326 ret
= mpage_readpages(mapping
, pages
, nr_pages
, gfs2_get_block
);
330 gfs2_glock_dq_m(1, &gh
);
331 gfs2_holder_uninit(&gh
);
334 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
340 /* unlock all pages, we can't do any I/O right now */
341 for (page_idx
= 0; page_idx
< nr_pages
; page_idx
++) {
342 struct page
*page
= list_entry(pages
->prev
, struct page
, lru
);
343 list_del(&page
->lru
);
345 page_cache_release(page
);
348 gfs2_holder_uninit(&gh
);
353 * gfs2_prepare_write - Prepare to write a page to a file
354 * @file: The file to write to
355 * @page: The page which is to be prepared for writing
356 * @from: From (byte range within page)
357 * @to: To (byte range within page)
362 static int gfs2_prepare_write(struct file
*file
, struct page
*page
,
363 unsigned from
, unsigned to
)
365 struct gfs2_inode
*ip
= GFS2_I(page
->mapping
->host
);
366 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
367 unsigned int data_blocks
, ind_blocks
, rblocks
;
370 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + from
;
371 loff_t end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
372 struct gfs2_alloc
*al
;
373 unsigned int write_len
= to
- from
;
376 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_ATIME
|GL_AOP
, &ip
->i_gh
);
377 error
= gfs2_glock_nq_m_atime(1, &ip
->i_gh
);
381 gfs2_write_calc_reserv(ip
, write_len
, &data_blocks
, &ind_blocks
);
383 error
= gfs2_write_alloc_required(ip
, pos
, write_len
, &alloc_required
);
388 ip
->i_alloc
.al_requested
= 0;
389 if (alloc_required
) {
390 al
= gfs2_alloc_get(ip
);
392 error
= gfs2_quota_lock(ip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
396 error
= gfs2_quota_check(ip
, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
400 al
->al_requested
= data_blocks
+ ind_blocks
;
401 error
= gfs2_inplace_reserve(ip
);
406 rblocks
= RES_DINODE
+ ind_blocks
;
407 if (gfs2_is_jdata(ip
))
408 rblocks
+= data_blocks
? data_blocks
: 1;
409 if (ind_blocks
|| data_blocks
)
410 rblocks
+= RES_STATFS
+ RES_QUOTA
;
412 error
= gfs2_trans_begin(sdp
, rblocks
, 0);
416 if (gfs2_is_stuffed(ip
)) {
417 if (end
> sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
)) {
418 error
= gfs2_unstuff_dinode(ip
, page
);
421 } else if (!PageUptodate(page
))
422 error
= stuffed_readpage(ip
, page
);
427 error
= block_prepare_write(page
, from
, to
, gfs2_get_block
);
432 if (alloc_required
) {
433 gfs2_inplace_release(ip
);
435 gfs2_quota_unlock(ip
);
440 gfs2_glock_dq_m(1, &ip
->i_gh
);
442 gfs2_holder_uninit(&ip
->i_gh
);
449 * gfs2_commit_write - Commit write to a file
450 * @file: The file to write to
451 * @page: The page containing the data
452 * @from: From (byte range within page)
453 * @to: To (byte range within page)
458 static int gfs2_commit_write(struct file
*file
, struct page
*page
,
459 unsigned from
, unsigned to
)
461 struct inode
*inode
= page
->mapping
->host
;
462 struct gfs2_inode
*ip
= GFS2_I(inode
);
463 struct gfs2_sbd
*sdp
= GFS2_SB(inode
);
464 int error
= -EOPNOTSUPP
;
465 struct buffer_head
*dibh
;
466 struct gfs2_alloc
*al
= &ip
->i_alloc
;
467 struct gfs2_dinode
*di
;
469 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_locked_by_me(ip
->i_gl
)))
472 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
476 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
477 di
= (struct gfs2_dinode
*)dibh
->b_data
;
479 if (gfs2_is_stuffed(ip
)) {
483 file_size
= ((u64
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
485 kaddr
= kmap_atomic(page
, KM_USER0
);
486 memcpy(dibh
->b_data
+ sizeof(struct gfs2_dinode
) + from
,
487 kaddr
+ from
, to
- from
);
488 kunmap_atomic(kaddr
, KM_USER0
);
490 SetPageUptodate(page
);
492 if (inode
->i_size
< file_size
)
493 i_size_write(inode
, file_size
);
495 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
||
497 gfs2_page_add_databufs(ip
, page
, from
, to
);
498 error
= generic_commit_write(file
, page
, from
, to
);
503 if (ip
->i_di
.di_size
< inode
->i_size
) {
504 ip
->i_di
.di_size
= inode
->i_size
;
505 di
->di_size
= cpu_to_be64(inode
->i_size
);
508 di
->di_mode
= cpu_to_be32(inode
->i_mode
);
509 di
->di_atime
= cpu_to_be64(inode
->i_atime
.tv_sec
);
510 di
->di_mtime
= cpu_to_be64(inode
->i_mtime
.tv_sec
);
511 di
->di_ctime
= cpu_to_be64(inode
->i_ctime
.tv_sec
);
515 if (al
->al_requested
) {
516 gfs2_inplace_release(ip
);
517 gfs2_quota_unlock(ip
);
520 gfs2_glock_dq_m(1, &ip
->i_gh
);
521 gfs2_holder_uninit(&ip
->i_gh
);
528 if (al
->al_requested
) {
529 gfs2_inplace_release(ip
);
530 gfs2_quota_unlock(ip
);
533 gfs2_glock_dq_m(1, &ip
->i_gh
);
534 gfs2_holder_uninit(&ip
->i_gh
);
536 ClearPageUptodate(page
);
541 * gfs2_bmap - Block map function
542 * @mapping: Address space info
543 * @lblock: The block to map
545 * Returns: The disk address for the block or 0 on hole or error
548 static sector_t
gfs2_bmap(struct address_space
*mapping
, sector_t lblock
)
550 struct gfs2_inode
*ip
= GFS2_I(mapping
->host
);
551 struct gfs2_holder i_gh
;
555 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
, &i_gh
);
559 if (!gfs2_is_stuffed(ip
))
560 dblock
= generic_block_bmap(mapping
, lblock
, gfs2_get_block
);
562 gfs2_glock_dq_uninit(&i_gh
);
567 static void discard_buffer(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
)
569 struct gfs2_bufdata
*bd
;
575 bh
->b_private
= NULL
;
577 gfs2_log_unlock(sdp
);
580 clear_buffer_dirty(bh
);
582 clear_buffer_mapped(bh
);
583 clear_buffer_req(bh
);
584 clear_buffer_new(bh
);
585 clear_buffer_delay(bh
);
589 static void gfs2_invalidatepage(struct page
*page
, unsigned long offset
)
591 struct gfs2_sbd
*sdp
= GFS2_SB(page
->mapping
->host
);
592 struct buffer_head
*head
, *bh
, *next
;
593 unsigned int curr_off
= 0;
595 BUG_ON(!PageLocked(page
));
596 if (!page_has_buffers(page
))
599 bh
= head
= page_buffers(page
);
601 unsigned int next_off
= curr_off
+ bh
->b_size
;
602 next
= bh
->b_this_page
;
604 if (offset
<= curr_off
)
605 discard_buffer(sdp
, bh
);
609 } while (bh
!= head
);
612 try_to_release_page(page
, 0);
617 static ssize_t
gfs2_direct_IO(int rw
, struct kiocb
*iocb
,
618 const struct iovec
*iov
, loff_t offset
,
619 unsigned long nr_segs
)
621 struct file
*file
= iocb
->ki_filp
;
622 struct inode
*inode
= file
->f_mapping
->host
;
623 struct gfs2_inode
*ip
= GFS2_I(inode
);
624 struct gfs2_holder gh
;
628 mutex_lock(&inode
->i_mutex
);
630 * Shared lock, even if its a write, since we do no allocation
631 * on this path. All we need change is atime.
633 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &gh
);
634 rv
= gfs2_glock_nq_m_atime(1, &gh
);
638 if (offset
> i_size_read(inode
))
642 * Should we return an error here? I can't see that O_DIRECT for
643 * a journaled file makes any sense. For now we'll silently fall
644 * back to buffered I/O, likewise we do the same for stuffed
645 * files since they are (a) small and (b) unaligned.
647 if (gfs2_is_jdata(ip
))
650 if (gfs2_is_stuffed(ip
))
653 rv
= blockdev_direct_IO_own_locking(rw
, iocb
, inode
,
655 iov
, offset
, nr_segs
,
656 gfs2_get_block_direct
, NULL
);
658 gfs2_glock_dq_m(1, &gh
);
659 gfs2_holder_uninit(&gh
);
661 mutex_unlock(&inode
->i_mutex
);
667 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
668 * @bh: the buffer we're stuck on
672 static void stuck_releasepage(struct buffer_head
*bh
)
674 struct inode
*inode
= bh
->b_page
->mapping
->host
;
675 struct gfs2_sbd
*sdp
= inode
->i_sb
->s_fs_info
;
676 struct gfs2_bufdata
*bd
= bh
->b_private
;
677 struct gfs2_glock
*gl
;
678 static unsigned limit
= 0;
684 fs_warn(sdp
, "stuck in gfs2_releasepage() %p\n", inode
);
685 fs_warn(sdp
, "blkno = %llu, bh->b_count = %d\n",
686 (unsigned long long)bh
->b_blocknr
, atomic_read(&bh
->b_count
));
687 fs_warn(sdp
, "pinned = %u\n", buffer_pinned(bh
));
688 fs_warn(sdp
, "bh->b_private = %s\n", (bd
) ? "!NULL" : "NULL");
695 fs_warn(sdp
, "gl = (%u, %llu)\n",
696 gl
->gl_name
.ln_type
, (unsigned long long)gl
->gl_name
.ln_number
);
698 fs_warn(sdp
, "bd_list_tr = %s, bd_le.le_list = %s\n",
699 (list_empty(&bd
->bd_list_tr
)) ? "no" : "yes",
700 (list_empty(&bd
->bd_le
.le_list
)) ? "no" : "yes");
702 if (gl
->gl_ops
== &gfs2_inode_glops
) {
703 struct gfs2_inode
*ip
= gl
->gl_object
;
709 fs_warn(sdp
, "ip = %llu %llu\n",
710 (unsigned long long)ip
->i_num
.no_formal_ino
,
711 (unsigned long long)ip
->i_num
.no_addr
);
713 for (x
= 0; x
< GFS2_MAX_META_HEIGHT
; x
++)
714 fs_warn(sdp
, "ip->i_cache[%u] = %s\n",
715 x
, (ip
->i_cache
[x
]) ? "!NULL" : "NULL");
720 * gfs2_releasepage - free the metadata associated with a page
721 * @page: the page that's being released
722 * @gfp_mask: passed from Linux VFS, ignored by us
724 * Call try_to_free_buffers() if the buffers in this page can be
730 int gfs2_releasepage(struct page
*page
, gfp_t gfp_mask
)
732 struct inode
*aspace
= page
->mapping
->host
;
733 struct gfs2_sbd
*sdp
= aspace
->i_sb
->s_fs_info
;
734 struct buffer_head
*bh
, *head
;
735 struct gfs2_bufdata
*bd
;
736 unsigned long t
= jiffies
+ gfs2_tune_get(sdp
, gt_stall_secs
) * HZ
;
738 if (!page_has_buffers(page
))
741 head
= bh
= page_buffers(page
);
743 while (atomic_read(&bh
->b_count
)) {
744 if (!atomic_read(&aspace
->i_writecount
))
747 if (time_after_eq(jiffies
, t
)) {
748 stuck_releasepage(bh
);
749 /* should we withdraw here? */
756 gfs2_assert_warn(sdp
, !buffer_pinned(bh
));
757 gfs2_assert_warn(sdp
, !buffer_dirty(bh
));
762 gfs2_assert_warn(sdp
, bd
->bd_bh
== bh
);
763 gfs2_assert_warn(sdp
, list_empty(&bd
->bd_list_tr
));
764 gfs2_assert_warn(sdp
, !bd
->bd_ail
);
766 if (!list_empty(&bd
->bd_le
.le_list
))
768 bh
->b_private
= NULL
;
770 gfs2_log_unlock(sdp
);
772 kmem_cache_free(gfs2_bufdata_cachep
, bd
);
774 bh
= bh
->b_this_page
;
775 } while (bh
!= head
);
778 return try_to_free_buffers(page
);
781 const struct address_space_operations gfs2_file_aops
= {
782 .writepage
= gfs2_writepage
,
783 .readpage
= gfs2_readpage
,
784 .readpages
= gfs2_readpages
,
785 .sync_page
= block_sync_page
,
786 .prepare_write
= gfs2_prepare_write
,
787 .commit_write
= gfs2_commit_write
,
789 .invalidatepage
= gfs2_invalidatepage
,
790 .releasepage
= gfs2_releasepage
,
791 .direct_IO
= gfs2_direct_IO
,