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>
15 #include <linux/pagemap.h>
16 #include <linux/mpage.h>
18 #include <asm/semaphore.h>
26 #include "ops_address.h"
34 * gfs2_get_block - Fills in a buffer head with details about a block
36 * @lblock: The block number to look up
37 * @bh_result: The buffer head to return the result in
38 * @create: Non-zero if we may add block to the file
43 int gfs2_get_block(struct inode
*inode
, sector_t lblock
,
44 struct buffer_head
*bh_result
, int create
)
46 struct gfs2_inode
*ip
= get_v2ip(inode
);
51 error
= gfs2_block_map(ip
, lblock
, &new, &dblock
, NULL
);
58 map_bh(bh_result
, inode
->i_sb
, dblock
);
60 set_buffer_new(bh_result
);
66 * get_block_noalloc - Fills in a buffer head with details about a block
68 * @lblock: The block number to look up
69 * @bh_result: The buffer head to return the result in
70 * @create: Non-zero if we may add block to the file
75 static int get_block_noalloc(struct inode
*inode
, sector_t lblock
,
76 struct buffer_head
*bh_result
, int create
)
78 struct gfs2_inode
*ip
= get_v2ip(inode
);
83 error
= gfs2_block_map(ip
, lblock
, &new, &dblock
, NULL
);
88 map_bh(bh_result
, inode
->i_sb
, dblock
);
89 else if (gfs2_assert_withdraw(ip
->i_sbd
, !create
))
95 static int get_blocks(struct inode
*inode
, sector_t lblock
,
96 unsigned long max_blocks
, struct buffer_head
*bh_result
,
99 struct gfs2_inode
*ip
= get_v2ip(inode
);
105 error
= gfs2_block_map(ip
, lblock
, &new, &dblock
, &extlen
);
112 map_bh(bh_result
, inode
->i_sb
, dblock
);
114 set_buffer_new(bh_result
);
116 if (extlen
> max_blocks
)
118 bh_result
->b_size
= extlen
<< inode
->i_blkbits
;
123 static int get_blocks_noalloc(struct inode
*inode
, sector_t lblock
,
124 unsigned long max_blocks
,
125 struct buffer_head
*bh_result
, int create
)
127 struct gfs2_inode
*ip
= get_v2ip(inode
);
133 error
= gfs2_block_map(ip
, lblock
, &new, &dblock
, &extlen
);
138 map_bh(bh_result
, inode
->i_sb
, dblock
);
139 if (extlen
> max_blocks
)
141 bh_result
->b_size
= extlen
<< inode
->i_blkbits
;
142 } else if (gfs2_assert_withdraw(ip
->i_sbd
, !create
))
149 * gfs2_writepage - Write complete page
150 * @page: Page to write
154 * Some of this is copied from block_write_full_page() although we still
155 * call it to do most of the work.
158 static int gfs2_writepage(struct page
*page
, struct writeback_control
*wbc
)
160 struct inode
*inode
= page
->mapping
->host
;
161 struct gfs2_inode
*ip
= get_v2ip(page
->mapping
->host
);
162 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
163 loff_t i_size
= i_size_read(inode
);
164 pgoff_t end_index
= i_size
>> PAGE_CACHE_SHIFT
;
169 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_held_excl(ip
->i_gl
))) {
176 /* Is the page fully outside i_size? (truncate in progress) */
177 offset
= i_size
& (PAGE_CACHE_SIZE
-1);
178 if (page
->index
>= end_index
+1 || !offset
) {
179 page
->mapping
->a_ops
->invalidatepage(page
, 0);
181 return 0; /* don't care */
184 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
|| gfs2_is_jdata(ip
)) {
185 error
= gfs2_trans_begin(sdp
, RES_DINODE
+ 1, 0);
188 gfs2_page_add_databufs(ip
, page
, 0, sdp
->sd_vfs
->s_blocksize
-1);
191 error
= block_write_full_page(page
, get_block_noalloc
, wbc
);
194 gfs2_meta_cache_flush(ip
);
198 redirty_page_for_writepage(wbc
, page
);
204 * stuffed_readpage - Fill in a Linux page with stuffed file data
211 static int stuffed_readpage(struct gfs2_inode
*ip
, struct page
*page
)
213 struct buffer_head
*dibh
;
217 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
221 kaddr
= kmap_atomic(page
, KM_USER0
);
222 memcpy((char *)kaddr
,
223 dibh
->b_data
+ sizeof(struct gfs2_dinode
),
225 memset((char *)kaddr
+ ip
->i_di
.di_size
,
227 PAGE_CACHE_SIZE
- ip
->i_di
.di_size
);
228 kunmap_atomic(page
, KM_USER0
);
232 SetPageUptodate(page
);
237 static int zero_readpage(struct page
*page
)
241 kaddr
= kmap_atomic(page
, KM_USER0
);
242 memset(kaddr
, 0, PAGE_CACHE_SIZE
);
243 kunmap_atomic(page
, KM_USER0
);
245 SetPageUptodate(page
);
252 * gfs2_readpage - readpage with locking
253 * @file: The file to read a page for. N.B. This may be NULL if we are
254 * reading an internal file.
255 * @page: The page to read
260 static int gfs2_readpage(struct file
*file
, struct page
*page
)
262 struct gfs2_inode
*ip
= get_v2ip(page
->mapping
->host
);
263 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
264 struct gfs2_holder gh
;
267 if (file
!= &gfs2_internal_file_sentinal
) {
268 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &gh
);
269 error
= gfs2_glock_nq_m_atime(1, &gh
);
274 if (gfs2_is_stuffed(ip
)) {
276 error
= stuffed_readpage(ip
, page
);
279 error
= zero_readpage(page
);
281 error
= mpage_readpage(page
, gfs2_get_block
);
283 if (unlikely(test_bit(SDF_SHUTDOWN
, &sdp
->sd_flags
)))
286 if (file
!= &gfs2_internal_file_sentinal
) {
287 gfs2_glock_dq_m(1, &gh
);
288 gfs2_holder_uninit(&gh
);
298 * gfs2_prepare_write - Prepare to write a page to a file
299 * @file: The file to write to
300 * @page: The page which is to be prepared for writing
301 * @from: From (byte range within page)
302 * @to: To (byte range within page)
307 static int gfs2_prepare_write(struct file
*file
, struct page
*page
,
308 unsigned from
, unsigned to
)
310 struct gfs2_inode
*ip
= get_v2ip(page
->mapping
->host
);
311 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
312 unsigned int data_blocks
, ind_blocks
, rblocks
;
315 loff_t pos
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + from
;
316 loff_t end
= ((loff_t
)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
317 struct gfs2_alloc
*al
;
319 gfs2_holder_init(ip
->i_gl
, LM_ST_EXCLUSIVE
, GL_ATIME
, &ip
->i_gh
);
320 error
= gfs2_glock_nq_m_atime(1, &ip
->i_gh
);
324 gfs2_write_calc_reserv(ip
, to
- from
, &data_blocks
, &ind_blocks
);
326 error
= gfs2_write_alloc_required(ip
, pos
, from
- to
, &alloc_required
);
331 if (alloc_required
) {
332 al
= gfs2_alloc_get(ip
);
334 error
= gfs2_quota_lock(ip
, NO_QUOTA_CHANGE
, NO_QUOTA_CHANGE
);
338 error
= gfs2_quota_check(ip
, ip
->i_di
.di_uid
, ip
->i_di
.di_gid
);
342 al
->al_requested
= data_blocks
+ ind_blocks
;
343 error
= gfs2_inplace_reserve(ip
);
348 rblocks
= RES_DINODE
+ ind_blocks
;
349 if (gfs2_is_jdata(ip
))
350 rblocks
+= data_blocks
? data_blocks
: 1;
351 if (ind_blocks
|| data_blocks
)
352 rblocks
+= RES_STATFS
+ RES_QUOTA
;
354 error
= gfs2_trans_begin(sdp
, rblocks
, 0);
358 if (gfs2_is_stuffed(ip
)) {
359 if (end
> sdp
->sd_sb
.sb_bsize
- sizeof(struct gfs2_dinode
)) {
360 error
= gfs2_unstuff_dinode(ip
, gfs2_unstuffer_page
,
364 } else if (!PageUptodate(page
))
365 error
= stuffed_readpage(ip
, page
);
370 error
= block_prepare_write(page
, from
, to
, gfs2_get_block
);
375 if (alloc_required
) {
376 gfs2_inplace_release(ip
);
378 gfs2_quota_unlock(ip
);
383 gfs2_glock_dq_m(1, &ip
->i_gh
);
385 gfs2_holder_uninit(&ip
->i_gh
);
392 * gfs2_commit_write - Commit write to a file
393 * @file: The file to write to
394 * @page: The page containing the data
395 * @from: From (byte range within page)
396 * @to: To (byte range within page)
401 static int gfs2_commit_write(struct file
*file
, struct page
*page
,
402 unsigned from
, unsigned to
)
404 struct inode
*inode
= page
->mapping
->host
;
405 struct gfs2_inode
*ip
= get_v2ip(inode
);
406 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
407 int error
= -EOPNOTSUPP
;
408 struct buffer_head
*dibh
;
409 struct gfs2_alloc
*al
= &ip
->i_alloc
;;
411 if (gfs2_assert_withdraw(sdp
, gfs2_glock_is_locked_by_me(ip
->i_gl
)))
414 error
= gfs2_meta_inode_buffer(ip
, &dibh
);
418 gfs2_trans_add_bh(ip
->i_gl
, dibh
, 1);
420 if (gfs2_is_stuffed(ip
)) {
424 file_size
= ((uint64_t)page
->index
<< PAGE_CACHE_SHIFT
) + to
;
426 kaddr
= kmap_atomic(page
, KM_USER0
);
427 memcpy(dibh
->b_data
+ sizeof(struct gfs2_dinode
) + from
,
428 (char *)kaddr
+ from
, to
- from
);
429 kunmap_atomic(page
, KM_USER0
);
431 SetPageUptodate(page
);
433 if (inode
->i_size
< file_size
)
434 i_size_write(inode
, file_size
);
436 if (sdp
->sd_args
.ar_data
== GFS2_DATA_ORDERED
||
438 gfs2_page_add_databufs(ip
, page
, from
, to
);
439 error
= generic_commit_write(file
, page
, from
, to
);
444 if (ip
->i_di
.di_size
< inode
->i_size
)
445 ip
->i_di
.di_size
= inode
->i_size
;
447 gfs2_dinode_out(&ip
->i_di
, dibh
->b_data
);
450 if (al
->al_requested
) {
451 gfs2_inplace_release(ip
);
452 gfs2_quota_unlock(ip
);
455 gfs2_glock_dq_m(1, &ip
->i_gh
);
456 gfs2_holder_uninit(&ip
->i_gh
);
463 if (al
->al_requested
) {
464 gfs2_inplace_release(ip
);
465 gfs2_quota_unlock(ip
);
468 gfs2_glock_dq_m(1, &ip
->i_gh
);
469 gfs2_holder_uninit(&ip
->i_gh
);
471 ClearPageUptodate(page
);
476 * gfs2_bmap - Block map function
477 * @mapping: Address space info
478 * @lblock: The block to map
480 * Returns: The disk address for the block or 0 on hole or error
483 static sector_t
gfs2_bmap(struct address_space
*mapping
, sector_t lblock
)
485 struct gfs2_inode
*ip
= get_v2ip(mapping
->host
);
486 struct gfs2_holder i_gh
;
490 error
= gfs2_glock_nq_init(ip
->i_gl
, LM_ST_SHARED
, LM_FLAG_ANY
, &i_gh
);
494 if (!gfs2_is_stuffed(ip
))
495 dblock
= generic_block_bmap(mapping
, lblock
, gfs2_get_block
);
497 gfs2_glock_dq_uninit(&i_gh
);
502 static void discard_buffer(struct gfs2_sbd
*sdp
, struct buffer_head
*bh
)
504 struct gfs2_bufdata
*bd
;
511 gfs2_log_unlock(sdp
);
514 gfs2_log_unlock(sdp
);
517 clear_buffer_dirty(bh
);
519 clear_buffer_mapped(bh
);
520 clear_buffer_req(bh
);
521 clear_buffer_new(bh
);
522 clear_buffer_delay(bh
);
526 static int gfs2_invalidatepage(struct page
*page
, unsigned long offset
)
528 struct gfs2_sbd
*sdp
= get_v2sdp(page
->mapping
->host
->i_sb
);
529 struct buffer_head
*head
, *bh
, *next
;
530 unsigned int curr_off
= 0;
533 BUG_ON(!PageLocked(page
));
534 if (!page_has_buffers(page
))
537 bh
= head
= page_buffers(page
);
539 unsigned int next_off
= curr_off
+ bh
->b_size
;
540 next
= bh
->b_this_page
;
542 if (offset
<= curr_off
)
543 discard_buffer(sdp
, bh
);
547 } while (bh
!= head
);
550 ret
= try_to_release_page(page
, 0);
555 static ssize_t
gfs2_direct_IO_write(struct kiocb
*iocb
, const struct iovec
*iov
,
556 loff_t offset
, unsigned long nr_segs
)
558 struct file
*file
= iocb
->ki_filp
;
559 struct inode
*inode
= file
->f_mapping
->host
;
560 struct gfs2_inode
*ip
= get_v2ip(inode
);
561 struct gfs2_holder gh
;
565 * Shared lock, even though its write, since we do no allocation
566 * on this path. All we need change is atime.
568 gfs2_holder_init(ip
->i_gl
, LM_ST_SHARED
, GL_ATIME
, &gh
);
569 rv
= gfs2_glock_nq_m_atime(1, &gh
);
574 * Should we return an error here? I can't see that O_DIRECT for
575 * a journaled file makes any sense. For now we'll silently fall
576 * back to buffered I/O, likewise we do the same for stuffed
577 * files since they are (a) small and (b) unaligned.
579 if (gfs2_is_jdata(ip
))
582 if (gfs2_is_stuffed(ip
))
585 rv
= __blockdev_direct_IO(WRITE
, iocb
, inode
, inode
->i_sb
->s_bdev
,
586 iov
, offset
, nr_segs
, get_blocks_noalloc
,
587 NULL
, DIO_OWN_LOCKING
);
589 gfs2_glock_dq_m(1, &gh
);
590 gfs2_holder_uninit(&gh
);
598 * This is called with a shared lock already held for the read path.
599 * Currently, no locks are held when the write path is called.
601 static ssize_t
gfs2_direct_IO(int rw
, struct kiocb
*iocb
,
602 const struct iovec
*iov
, loff_t offset
,
603 unsigned long nr_segs
)
605 struct file
*file
= iocb
->ki_filp
;
606 struct inode
*inode
= file
->f_mapping
->host
;
607 struct gfs2_inode
*ip
= get_v2ip(inode
);
608 struct gfs2_sbd
*sdp
= ip
->i_sbd
;
611 return gfs2_direct_IO_write(iocb
, iov
, offset
, nr_segs
);
613 if (gfs2_assert_warn(sdp
, gfs2_glock_is_locked_by_me(ip
->i_gl
)) ||
614 gfs2_assert_warn(sdp
, !gfs2_is_stuffed(ip
)))
617 return __blockdev_direct_IO(READ
, iocb
, inode
, inode
->i_sb
->s_bdev
, iov
,
618 offset
, nr_segs
, get_blocks
, NULL
,
622 struct address_space_operations gfs2_file_aops
= {
623 .writepage
= gfs2_writepage
,
624 .readpage
= gfs2_readpage
,
625 .sync_page
= block_sync_page
,
626 .prepare_write
= gfs2_prepare_write
,
627 .commit_write
= gfs2_commit_write
,
629 .invalidatepage
= gfs2_invalidatepage
,
630 .direct_IO
= gfs2_direct_IO
,