]> bbs.cooldavid.org Git - net-next-2.6.git/blame - fs/gfs2/ops_address.c
[GFS2] Fix DIO deadlock
[net-next-2.6.git] / fs / gfs2 / ops_address.c
CommitLineData
b3b94faa
DT
1/*
2 * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved.
3a8a9a10 3 * Copyright (C) 2004-2006 Red Hat, Inc. All rights reserved.
b3b94faa
DT
4 *
5 * This copyrighted material is made available to anyone wishing to use,
6 * modify, copy, or redistribute it subject to the terms and conditions
e9fc2aa0 7 * of the GNU General Public License version 2.
b3b94faa
DT
8 */
9
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>
fd88de56 16#include <linux/pagevec.h>
9b124fbb 17#include <linux/mpage.h>
d1665e41 18#include <linux/fs.h>
5c676f6d 19#include <linux/gfs2_ondisk.h>
7d308590 20#include <linux/lm_interface.h>
b3b94faa
DT
21
22#include "gfs2.h"
5c676f6d 23#include "incore.h"
b3b94faa
DT
24#include "bmap.h"
25#include "glock.h"
26#include "inode.h"
b3b94faa
DT
27#include "log.h"
28#include "meta_io.h"
29#include "ops_address.h"
b3b94faa
DT
30#include "quota.h"
31#include "trans.h"
18ec7d5c 32#include "rgrp.h"
61a30dcb 33#include "ops_file.h"
5c676f6d 34#include "util.h"
4340fe62 35#include "glops.h"
b3b94faa 36
ba7f7290
SW
37
38static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
39 unsigned int from, unsigned int to)
40{
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;
45
46 for (bh = head, start = 0; bh != head || !start;
47 bh = bh->b_this_page, start = end) {
48 end = start + bsize;
49 if (end <= from || start >= to)
50 continue;
51 gfs2_trans_add_bh(ip->i_gl, bh, 0);
52 }
53}
54
b3b94faa 55/**
4ff14670 56 * gfs2_get_block - Fills in a buffer head with details about a block
b3b94faa
DT
57 * @inode: The inode
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
61 *
62 * Returns: errno
63 */
64
4ff14670
SW
65int gfs2_get_block(struct inode *inode, sector_t lblock,
66 struct buffer_head *bh_result, int create)
b3b94faa 67{
23591256 68 return gfs2_block_map(inode, lblock, create, bh_result);
b3b94faa
DT
69}
70
71/**
7a6bbacb 72 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
b3b94faa
DT
73 * @inode: The inode
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
77 *
78 * Returns: errno
79 */
80
7a6bbacb
SW
81static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
82 struct buffer_head *bh_result, int create)
b3b94faa 83{
b3b94faa
DT
84 int error;
85
23591256 86 error = gfs2_block_map(inode, lblock, 0, bh_result);
b3b94faa
DT
87 if (error)
88 return error;
7a6bbacb
SW
89 if (bh_result->b_blocknr == 0)
90 return -EIO;
91 return 0;
b3b94faa
DT
92}
93
7a6bbacb
SW
94static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
95 struct buffer_head *bh_result, int create)
623d9355 96{
23591256 97 return gfs2_block_map(inode, lblock, 0, bh_result);
623d9355 98}
7a6bbacb 99
b3b94faa
DT
100/**
101 * gfs2_writepage - Write complete page
102 * @page: Page to write
103 *
104 * Returns: errno
105 *
18ec7d5c
SW
106 * Some of this is copied from block_write_full_page() although we still
107 * call it to do most of the work.
b3b94faa
DT
108 */
109
110static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
111{
18ec7d5c 112 struct inode *inode = page->mapping->host;
f4387149
SW
113 struct gfs2_inode *ip = GFS2_I(inode);
114 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
115 loff_t i_size = i_size_read(inode);
116 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
117 unsigned offset;
b3b94faa 118 int error;
18ec7d5c 119 int done_trans = 0;
b3b94faa 120
b3b94faa
DT
121 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
122 unlock_page(page);
123 return -EIO;
124 }
5c676f6d 125 if (current->journal_info)
18ec7d5c
SW
126 goto out_ignore;
127
128 /* Is the page fully outside i_size? (truncate in progress) */
129 offset = i_size & (PAGE_CACHE_SIZE-1);
d2d7b8a2 130 if (page->index > end_index || (page->index == end_index && !offset)) {
18ec7d5c 131 page->mapping->a_ops->invalidatepage(page, 0);
b3b94faa 132 unlock_page(page);
18ec7d5c 133 return 0; /* don't care */
b3b94faa
DT
134 }
135
18ec7d5c
SW
136 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
137 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
138 if (error)
139 goto out_ignore;
f4387149
SW
140 if (!page_has_buffers(page)) {
141 create_empty_buffers(page, inode->i_sb->s_blocksize,
142 (1 << BH_Dirty)|(1 << BH_Uptodate));
143 }
18ec7d5c
SW
144 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
145 done_trans = 1;
146 }
7a6bbacb 147 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
18ec7d5c
SW
148 if (done_trans)
149 gfs2_trans_end(sdp);
b3b94faa 150 gfs2_meta_cache_flush(ip);
b3b94faa 151 return error;
18ec7d5c
SW
152
153out_ignore:
154 redirty_page_for_writepage(wbc, page);
155 unlock_page(page);
156 return 0;
b3b94faa
DT
157}
158
159/**
160 * stuffed_readpage - Fill in a Linux page with stuffed file data
161 * @ip: the inode
162 * @page: the page
163 *
164 * Returns: errno
165 */
166
167static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
168{
169 struct buffer_head *dibh;
170 void *kaddr;
171 int error;
172
61057c6b 173 BUG_ON(page->index);
fd88de56 174
b3b94faa
DT
175 error = gfs2_meta_inode_buffer(ip, &dibh);
176 if (error)
177 return error;
178
5c4e9e03 179 kaddr = kmap_atomic(page, KM_USER0);
fd88de56 180 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
b3b94faa 181 ip->i_di.di_size);
fd88de56 182 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
c312c4fd 183 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
184
185 brelse(dibh);
186
187 SetPageUptodate(page);
188
189 return 0;
190}
191
b3b94faa 192
b3b94faa
DT
193/**
194 * gfs2_readpage - readpage with locking
18ec7d5c
SW
195 * @file: The file to read a page for. N.B. This may be NULL if we are
196 * reading an internal file.
b3b94faa
DT
197 * @page: The page to read
198 *
199 * Returns: errno
200 */
201
202static int gfs2_readpage(struct file *file, struct page *page)
203{
feaa7bba
SW
204 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
205 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
dc41aeed 206 struct gfs2_file *gf = NULL;
18ec7d5c 207 struct gfs2_holder gh;
b3b94faa 208 int error;
59a1cc6b 209 int do_unlock = 0;
b3b94faa 210
dd538c83 211 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b 212 if (file) {
dc41aeed 213 gf = file->private_data;
59a1cc6b 214 if (test_bit(GFF_EXLOCK, &gf->f_flags))
dc41aeed 215 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
59a1cc6b
SW
216 goto skip_lock;
217 }
2ca99501 218 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
59a1cc6b 219 do_unlock = 1;
dcd24799 220 error = gfs2_glock_nq_atime(&gh);
fd88de56 221 if (unlikely(error))
61a30dcb
SW
222 goto out_unlock;
223 }
b3b94faa 224
59a1cc6b 225skip_lock:
18ec7d5c 226 if (gfs2_is_stuffed(ip)) {
fd88de56
SW
227 error = stuffed_readpage(ip, page);
228 unlock_page(page);
b3b94faa 229 } else
18ec7d5c 230 error = mpage_readpage(page, gfs2_get_block);
b3b94faa
DT
231
232 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
233 error = -EIO;
234
07903c02 235 if (do_unlock) {
61a30dcb
SW
236 gfs2_glock_dq_m(1, &gh);
237 gfs2_holder_uninit(&gh);
238 }
18ec7d5c 239out:
b3b94faa 240 return error;
18ec7d5c 241out_unlock:
2ca99501
SW
242 if (error == GLR_TRYFAILED)
243 error = AOP_TRUNCATED_PAGE;
18ec7d5c 244 unlock_page(page);
59a1cc6b 245 if (do_unlock)
fd88de56
SW
246 gfs2_holder_uninit(&gh);
247 goto out;
248}
249
fd88de56
SW
250/**
251 * gfs2_readpages - Read a bunch of pages at once
252 *
253 * Some notes:
254 * 1. This is only for readahead, so we can simply ignore any things
255 * which are slightly inconvenient (such as locking conflicts between
256 * the page lock and the glock) and return having done no I/O. Its
257 * obviously not something we'd want to do on too regular a basis.
258 * Any I/O we ignore at this time will be done via readpage later.
259 * 2. We have to handle stuffed files here too.
260 * 3. mpage_readpages() does most of the heavy lifting in the common case.
261 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
262 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
263 * well as read-ahead.
264 */
265static int gfs2_readpages(struct file *file, struct address_space *mapping,
266 struct list_head *pages, unsigned nr_pages)
267{
268 struct inode *inode = mapping->host;
feaa7bba
SW
269 struct gfs2_inode *ip = GFS2_I(inode);
270 struct gfs2_sbd *sdp = GFS2_SB(inode);
fd88de56
SW
271 struct gfs2_holder gh;
272 unsigned page_idx;
273 int ret;
59a1cc6b 274 int do_unlock = 0;
fd88de56 275
dd538c83 276 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b
SW
277 if (file) {
278 struct gfs2_file *gf = file->private_data;
279 if (test_bit(GFF_EXLOCK, &gf->f_flags))
280 goto skip_lock;
281 }
fd88de56 282 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
2ca99501 283 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
59a1cc6b 284 do_unlock = 1;
dcd24799 285 ret = gfs2_glock_nq_atime(&gh);
907b9bce 286 if (ret == GLR_TRYFAILED)
fd88de56
SW
287 goto out_noerror;
288 if (unlikely(ret))
289 goto out_unlock;
290 }
59a1cc6b 291skip_lock:
fd88de56
SW
292 if (gfs2_is_stuffed(ip)) {
293 struct pagevec lru_pvec;
294 pagevec_init(&lru_pvec, 0);
295 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
ffeb874b
SW
296 struct page *page = list_entry(pages->prev, struct page, lru);
297 prefetchw(&page->flags);
fd88de56
SW
298 list_del(&page->lru);
299 if (!add_to_page_cache(page, mapping,
300 page->index, GFP_KERNEL)) {
301 ret = stuffed_readpage(ip, page);
302 unlock_page(page);
303 if (!pagevec_add(&lru_pvec, page))
304 __pagevec_lru_add(&lru_pvec);
ffeb874b
SW
305 } else {
306 page_cache_release(page);
fd88de56 307 }
fd88de56
SW
308 }
309 pagevec_lru_add(&lru_pvec);
310 ret = 0;
311 } else {
312 /* What we really want to do .... */
313 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
314 }
315
59a1cc6b 316 if (do_unlock) {
fd88de56
SW
317 gfs2_glock_dq_m(1, &gh);
318 gfs2_holder_uninit(&gh);
319 }
320out:
321 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
322 ret = -EIO;
323 return ret;
324out_noerror:
325 ret = 0;
326out_unlock:
59a1cc6b 327 if (do_unlock)
fd88de56 328 gfs2_holder_uninit(&gh);
18ec7d5c 329 goto out;
b3b94faa
DT
330}
331
332/**
333 * gfs2_prepare_write - Prepare to write a page to a file
334 * @file: The file to write to
335 * @page: The page which is to be prepared for writing
336 * @from: From (byte range within page)
337 * @to: To (byte range within page)
338 *
339 * Returns: errno
340 */
341
342static int gfs2_prepare_write(struct file *file, struct page *page,
343 unsigned from, unsigned to)
344{
feaa7bba
SW
345 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
346 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c
SW
347 unsigned int data_blocks, ind_blocks, rblocks;
348 int alloc_required;
b3b94faa 349 int error = 0;
18ec7d5c
SW
350 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
351 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
352 struct gfs2_alloc *al;
52ae7b79
RC
353 unsigned int write_len = to - from;
354
b3b94faa 355
2ca99501 356 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
dcd24799 357 error = gfs2_glock_nq_atime(&ip->i_gh);
2ca99501
SW
358 if (unlikely(error)) {
359 if (error == GLR_TRYFAILED)
360 error = AOP_TRUNCATED_PAGE;
18ec7d5c 361 goto out_uninit;
2ca99501 362 }
b3b94faa 363
52ae7b79 364 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
18ec7d5c 365
52ae7b79 366 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
18ec7d5c
SW
367 if (error)
368 goto out_unlock;
b3b94faa 369
18ec7d5c 370
f5c54804 371 ip->i_alloc.al_requested = 0;
18ec7d5c
SW
372 if (alloc_required) {
373 al = gfs2_alloc_get(ip);
374
375 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
376 if (error)
377 goto out_alloc_put;
378
2933f925 379 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
18ec7d5c
SW
380 if (error)
381 goto out_qunlock;
382
383 al->al_requested = data_blocks + ind_blocks;
384 error = gfs2_inplace_reserve(ip);
385 if (error)
386 goto out_qunlock;
387 }
388
389 rblocks = RES_DINODE + ind_blocks;
390 if (gfs2_is_jdata(ip))
391 rblocks += data_blocks ? data_blocks : 1;
392 if (ind_blocks || data_blocks)
393 rblocks += RES_STATFS + RES_QUOTA;
394
395 error = gfs2_trans_begin(sdp, rblocks, 0);
396 if (error)
397 goto out;
398
399 if (gfs2_is_stuffed(ip)) {
400 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
f25ef0c1 401 error = gfs2_unstuff_dinode(ip, page);
5c4e9e03
SW
402 if (error == 0)
403 goto prepare_write;
404 } else if (!PageUptodate(page))
b3b94faa 405 error = stuffed_readpage(ip, page);
5c4e9e03 406 goto out;
18ec7d5c
SW
407 }
408
5c4e9e03 409prepare_write:
18ec7d5c
SW
410 error = block_prepare_write(page, from, to, gfs2_get_block);
411
412out:
413 if (error) {
414 gfs2_trans_end(sdp);
415 if (alloc_required) {
416 gfs2_inplace_release(ip);
417out_qunlock:
418 gfs2_quota_unlock(ip);
419out_alloc_put:
420 gfs2_alloc_put(ip);
421 }
422out_unlock:
423 gfs2_glock_dq_m(1, &ip->i_gh);
424out_uninit:
425 gfs2_holder_uninit(&ip->i_gh);
426 }
b3b94faa
DT
427
428 return error;
429}
430
431/**
432 * gfs2_commit_write - Commit write to a file
433 * @file: The file to write to
434 * @page: The page containing the data
435 * @from: From (byte range within page)
436 * @to: To (byte range within page)
437 *
438 * Returns: errno
439 */
440
441static int gfs2_commit_write(struct file *file, struct page *page,
442 unsigned from, unsigned to)
443{
444 struct inode *inode = page->mapping->host;
feaa7bba
SW
445 struct gfs2_inode *ip = GFS2_I(inode);
446 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
447 int error = -EOPNOTSUPP;
448 struct buffer_head *dibh;
48516ced
SW
449 struct gfs2_alloc *al = &ip->i_alloc;
450 struct gfs2_dinode *di;
b3b94faa 451
18ec7d5c
SW
452 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
453 goto fail_nounlock;
454
455 error = gfs2_meta_inode_buffer(ip, &dibh);
456 if (error)
457 goto fail_endtrans;
458
459 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
48516ced 460 di = (struct gfs2_dinode *)dibh->b_data;
18ec7d5c 461
b3b94faa 462 if (gfs2_is_stuffed(ip)) {
cd915493 463 u64 file_size;
b3b94faa
DT
464 void *kaddr;
465
cd915493 466 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
b3b94faa 467
18ec7d5c 468 kaddr = kmap_atomic(page, KM_USER0);
b3b94faa 469 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
0bd5996a 470 kaddr + from, to - from);
c312c4fd 471 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
472
473 SetPageUptodate(page);
474
ae619320 475 if (inode->i_size < file_size) {
b3b94faa 476 i_size_write(inode, file_size);
ae619320
SW
477 mark_inode_dirty(inode);
478 }
b3b94faa 479 } else {
568f4c96
SW
480 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
481 gfs2_is_jdata(ip))
257f9b4e 482 gfs2_page_add_databufs(ip, page, from, to);
b3b94faa
DT
483 error = generic_commit_write(file, page, from, to);
484 if (error)
485 goto fail;
486 }
487
48516ced 488 if (ip->i_di.di_size < inode->i_size) {
18ec7d5c 489 ip->i_di.di_size = inode->i_size;
48516ced
SW
490 di->di_size = cpu_to_be64(inode->i_size);
491 }
492
18ec7d5c
SW
493 brelse(dibh);
494 gfs2_trans_end(sdp);
495 if (al->al_requested) {
496 gfs2_inplace_release(ip);
497 gfs2_quota_unlock(ip);
498 gfs2_alloc_put(ip);
499 }
500 gfs2_glock_dq_m(1, &ip->i_gh);
501 gfs2_holder_uninit(&ip->i_gh);
b3b94faa
DT
502 return 0;
503
18ec7d5c
SW
504fail:
505 brelse(dibh);
506fail_endtrans:
507 gfs2_trans_end(sdp);
508 if (al->al_requested) {
509 gfs2_inplace_release(ip);
510 gfs2_quota_unlock(ip);
511 gfs2_alloc_put(ip);
512 }
513 gfs2_glock_dq_m(1, &ip->i_gh);
514 gfs2_holder_uninit(&ip->i_gh);
515fail_nounlock:
b3b94faa 516 ClearPageUptodate(page);
b3b94faa
DT
517 return error;
518}
519
520/**
521 * gfs2_bmap - Block map function
522 * @mapping: Address space info
523 * @lblock: The block to map
524 *
525 * Returns: The disk address for the block or 0 on hole or error
526 */
527
528static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
529{
feaa7bba 530 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa
DT
531 struct gfs2_holder i_gh;
532 sector_t dblock = 0;
533 int error;
534
b3b94faa
DT
535 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
536 if (error)
537 return 0;
538
539 if (!gfs2_is_stuffed(ip))
4ff14670 540 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
b3b94faa
DT
541
542 gfs2_glock_dq_uninit(&i_gh);
543
544 return dblock;
545}
546
547static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
548{
64fb4eb7 549 struct gfs2_bufdata *bd;
b3b94faa
DT
550
551 gfs2_log_lock(sdp);
5c676f6d 552 bd = bh->b_private;
64fb4eb7
SW
553 if (bd) {
554 bd->bd_bh = NULL;
5c676f6d 555 bh->b_private = NULL;
15d00c0b
SW
556 }
557 gfs2_log_unlock(sdp);
b3b94faa
DT
558
559 lock_buffer(bh);
560 clear_buffer_dirty(bh);
561 bh->b_bdev = NULL;
562 clear_buffer_mapped(bh);
563 clear_buffer_req(bh);
564 clear_buffer_new(bh);
565 clear_buffer_delay(bh);
566 unlock_buffer(bh);
567}
568
8628de05 569static void gfs2_invalidatepage(struct page *page, unsigned long offset)
b3b94faa 570{
15d00c0b 571 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
b3b94faa
DT
572 struct buffer_head *head, *bh, *next;
573 unsigned int curr_off = 0;
b3b94faa
DT
574
575 BUG_ON(!PageLocked(page));
576 if (!page_has_buffers(page))
8628de05 577 return;
b3b94faa
DT
578
579 bh = head = page_buffers(page);
580 do {
581 unsigned int next_off = curr_off + bh->b_size;
582 next = bh->b_this_page;
583
584 if (offset <= curr_off)
585 discard_buffer(sdp, bh);
586
587 curr_off = next_off;
588 bh = next;
589 } while (bh != head);
590
591 if (!offset)
8628de05 592 try_to_release_page(page, 0);
b3b94faa 593
8628de05 594 return;
b3b94faa
DT
595}
596
c7b33834
SW
597/**
598 * gfs2_ok_for_dio - check that dio is valid on this file
599 * @ip: The inode
600 * @rw: READ or WRITE
601 * @offset: The offset at which we are reading or writing
602 *
603 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
604 * 1 (to accept the i/o request)
605 */
606static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
607{
608 /*
609 * Should we return an error here? I can't see that O_DIRECT for
610 * a journaled file makes any sense. For now we'll silently fall
611 * back to buffered I/O, likewise we do the same for stuffed
612 * files since they are (a) small and (b) unaligned.
613 */
614 if (gfs2_is_jdata(ip))
615 return 0;
616
617 if (gfs2_is_stuffed(ip))
618 return 0;
619
620 if (offset > i_size_read(&ip->i_inode))
621 return 0;
622 return 1;
623}
624
625
626
a9e5f4d0
SW
627static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
628 const struct iovec *iov, loff_t offset,
629 unsigned long nr_segs)
d1665e41
SW
630{
631 struct file *file = iocb->ki_filp;
632 struct inode *inode = file->f_mapping->host;
feaa7bba 633 struct gfs2_inode *ip = GFS2_I(inode);
d1665e41
SW
634 struct gfs2_holder gh;
635 int rv;
636
637 /*
c7b33834
SW
638 * Deferred lock, even if its a write, since we do no allocation
639 * on this path. All we need change is atime, and this lock mode
640 * ensures that other nodes have flushed their buffered read caches
641 * (i.e. their page cache entries for this inode). We do not,
642 * unfortunately have the option of only flushing a range like
643 * the VFS does.
d1665e41 644 */
c7b33834 645 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
dcd24799 646 rv = gfs2_glock_nq_atime(&gh);
d1665e41 647 if (rv)
c7b33834
SW
648 return rv;
649 rv = gfs2_ok_for_dio(ip, rw, offset);
650 if (rv != 1)
651 goto out; /* dio not valid, fall back to buffered i/o */
652
653 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
654 iov, offset, nr_segs,
655 gfs2_get_block_direct, NULL);
d1665e41
SW
656out:
657 gfs2_glock_dq_m(1, &gh);
658 gfs2_holder_uninit(&gh);
d1665e41
SW
659 return rv;
660}
661
4340fe62
SW
662/**
663 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
664 * @bh: the buffer we're stuck on
665 *
666 */
667
668static void stuck_releasepage(struct buffer_head *bh)
669{
670 struct inode *inode = bh->b_page->mapping->host;
671 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
672 struct gfs2_bufdata *bd = bh->b_private;
673 struct gfs2_glock *gl;
166afccd
SW
674static unsigned limit = 0;
675
0bd5996a 676 if (limit > 3)
166afccd 677 return;
0bd5996a 678 limit++;
4340fe62
SW
679
680 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
681 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
682 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
683 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
684 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
685
686 if (!bd)
687 return;
688
689 gl = bd->bd_gl;
690
907b9bce 691 fs_warn(sdp, "gl = (%u, %llu)\n",
4340fe62
SW
692 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
693
694 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
695 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
696 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
697
698 if (gl->gl_ops == &gfs2_inode_glops) {
699 struct gfs2_inode *ip = gl->gl_object;
700 unsigned int x;
701
702 if (!ip)
703 return;
704
705 fs_warn(sdp, "ip = %llu %llu\n",
706 (unsigned long long)ip->i_num.no_formal_ino,
707 (unsigned long long)ip->i_num.no_addr);
708
709 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
710 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
711 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
712 }
713}
714
715/**
623d9355 716 * gfs2_releasepage - free the metadata associated with a page
4340fe62
SW
717 * @page: the page that's being released
718 * @gfp_mask: passed from Linux VFS, ignored by us
719 *
720 * Call try_to_free_buffers() if the buffers in this page can be
721 * released.
722 *
723 * Returns: 0
724 */
725
726int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
727{
728 struct inode *aspace = page->mapping->host;
729 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
730 struct buffer_head *bh, *head;
731 struct gfs2_bufdata *bd;
166afccd 732 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
4340fe62
SW
733
734 if (!page_has_buffers(page))
735 goto out;
736
737 head = bh = page_buffers(page);
738 do {
4340fe62 739 while (atomic_read(&bh->b_count)) {
166afccd
SW
740 if (!atomic_read(&aspace->i_writecount))
741 return 0;
742
61057c6b
RC
743 if (!(gfp_mask & __GFP_WAIT))
744 return 0;
745
166afccd
SW
746 if (time_after_eq(jiffies, t)) {
747 stuck_releasepage(bh);
748 /* should we withdraw here? */
749 return 0;
4340fe62
SW
750 }
751
166afccd 752 yield();
4340fe62
SW
753 }
754
755 gfs2_assert_warn(sdp, !buffer_pinned(bh));
623d9355 756 gfs2_assert_warn(sdp, !buffer_dirty(bh));
4340fe62 757
623d9355 758 gfs2_log_lock(sdp);
4340fe62
SW
759 bd = bh->b_private;
760 if (bd) {
761 gfs2_assert_warn(sdp, bd->bd_bh == bh);
762 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
4340fe62 763 gfs2_assert_warn(sdp, !bd->bd_ail);
623d9355
SW
764 bd->bd_bh = NULL;
765 if (!list_empty(&bd->bd_le.le_list))
766 bd = NULL;
4340fe62
SW
767 bh->b_private = NULL;
768 }
623d9355
SW
769 gfs2_log_unlock(sdp);
770 if (bd)
771 kmem_cache_free(gfs2_bufdata_cachep, bd);
4340fe62
SW
772
773 bh = bh->b_this_page;
166afccd 774 } while (bh != head);
4340fe62 775
a9e5f4d0 776out:
4340fe62
SW
777 return try_to_free_buffers(page);
778}
779
66de045d 780const struct address_space_operations gfs2_file_aops = {
b3b94faa
DT
781 .writepage = gfs2_writepage,
782 .readpage = gfs2_readpage,
fd88de56 783 .readpages = gfs2_readpages,
b3b94faa
DT
784 .sync_page = block_sync_page,
785 .prepare_write = gfs2_prepare_write,
786 .commit_write = gfs2_commit_write,
787 .bmap = gfs2_bmap,
788 .invalidatepage = gfs2_invalidatepage,
4340fe62 789 .releasepage = gfs2_releasepage,
b3b94faa
DT
790 .direct_IO = gfs2_direct_IO,
791};
792