]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/gfs2/ops_address.c
[GFS2] Add writepages for "data=writeback" mounts
[mirror_ubuntu-jammy-kernel.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>
a8d638e3 19#include <linux/writeback.h>
5c676f6d 20#include <linux/gfs2_ondisk.h>
7d308590 21#include <linux/lm_interface.h>
b3b94faa
DT
22
23#include "gfs2.h"
5c676f6d 24#include "incore.h"
b3b94faa
DT
25#include "bmap.h"
26#include "glock.h"
27#include "inode.h"
b3b94faa
DT
28#include "log.h"
29#include "meta_io.h"
30#include "ops_address.h"
b3b94faa
DT
31#include "quota.h"
32#include "trans.h"
18ec7d5c 33#include "rgrp.h"
61a30dcb 34#include "ops_file.h"
5c676f6d 35#include "util.h"
4340fe62 36#include "glops.h"
b3b94faa 37
ba7f7290
SW
38
39static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page,
40 unsigned int from, unsigned int to)
41{
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;
46
47 for (bh = head, start = 0; bh != head || !start;
48 bh = bh->b_this_page, start = end) {
49 end = start + bsize;
50 if (end <= from || start >= to)
51 continue;
52 gfs2_trans_add_bh(ip->i_gl, bh, 0);
53 }
54}
55
b3b94faa 56/**
4ff14670 57 * gfs2_get_block - Fills in a buffer head with details about a block
b3b94faa
DT
58 * @inode: The inode
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
62 *
63 * Returns: errno
64 */
65
4ff14670
SW
66int gfs2_get_block(struct inode *inode, sector_t lblock,
67 struct buffer_head *bh_result, int create)
b3b94faa 68{
23591256 69 return gfs2_block_map(inode, lblock, create, bh_result);
b3b94faa
DT
70}
71
72/**
7a6bbacb 73 * gfs2_get_block_noalloc - Fills in a buffer head with details about a block
b3b94faa
DT
74 * @inode: The inode
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
78 *
79 * Returns: errno
80 */
81
7a6bbacb
SW
82static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock,
83 struct buffer_head *bh_result, int create)
b3b94faa 84{
b3b94faa
DT
85 int error;
86
23591256 87 error = gfs2_block_map(inode, lblock, 0, bh_result);
b3b94faa
DT
88 if (error)
89 return error;
7a6bbacb
SW
90 if (bh_result->b_blocknr == 0)
91 return -EIO;
92 return 0;
b3b94faa
DT
93}
94
7a6bbacb
SW
95static int gfs2_get_block_direct(struct inode *inode, sector_t lblock,
96 struct buffer_head *bh_result, int create)
623d9355 97{
23591256 98 return gfs2_block_map(inode, lblock, 0, bh_result);
623d9355 99}
7a6bbacb 100
b3b94faa
DT
101/**
102 * gfs2_writepage - Write complete page
103 * @page: Page to write
104 *
105 * Returns: errno
106 *
18ec7d5c
SW
107 * Some of this is copied from block_write_full_page() although we still
108 * call it to do most of the work.
b3b94faa
DT
109 */
110
111static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
112{
18ec7d5c 113 struct inode *inode = page->mapping->host;
f4387149
SW
114 struct gfs2_inode *ip = GFS2_I(inode);
115 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
116 loff_t i_size = i_size_read(inode);
117 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
118 unsigned offset;
b3b94faa 119 int error;
18ec7d5c 120 int done_trans = 0;
b3b94faa 121
b3b94faa
DT
122 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
123 unlock_page(page);
124 return -EIO;
125 }
5c676f6d 126 if (current->journal_info)
18ec7d5c
SW
127 goto out_ignore;
128
129 /* Is the page fully outside i_size? (truncate in progress) */
130 offset = i_size & (PAGE_CACHE_SIZE-1);
d2d7b8a2 131 if (page->index > end_index || (page->index == end_index && !offset)) {
18ec7d5c 132 page->mapping->a_ops->invalidatepage(page, 0);
b3b94faa 133 unlock_page(page);
18ec7d5c 134 return 0; /* don't care */
b3b94faa
DT
135 }
136
18ec7d5c
SW
137 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
138 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
139 if (error)
140 goto out_ignore;
f4387149
SW
141 if (!page_has_buffers(page)) {
142 create_empty_buffers(page, inode->i_sb->s_blocksize,
143 (1 << BH_Dirty)|(1 << BH_Uptodate));
144 }
18ec7d5c
SW
145 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
146 done_trans = 1;
147 }
7a6bbacb 148 error = block_write_full_page(page, gfs2_get_block_noalloc, wbc);
18ec7d5c
SW
149 if (done_trans)
150 gfs2_trans_end(sdp);
b3b94faa 151 gfs2_meta_cache_flush(ip);
b3b94faa 152 return error;
18ec7d5c
SW
153
154out_ignore:
155 redirty_page_for_writepage(wbc, page);
156 unlock_page(page);
157 return 0;
b3b94faa
DT
158}
159
a8d638e3
SW
160/**
161 * gfs2_writepages - Write a bunch of dirty pages back to disk
162 * @mapping: The mapping to write
163 * @wbc: Write-back control
164 *
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).
168 *
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.
172 */
173int gfs2_writepages(struct address_space *mapping, struct writeback_control *wbc)
174{
175 struct inode *inode = mapping->host;
176 struct gfs2_inode *ip = GFS2_I(inode);
177 struct gfs2_sbd *sdp = GFS2_SB(inode);
178
179 if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip))
180 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
181
182 return generic_writepages(mapping, wbc);
183}
184
b3b94faa
DT
185/**
186 * stuffed_readpage - Fill in a Linux page with stuffed file data
187 * @ip: the inode
188 * @page: the page
189 *
190 * Returns: errno
191 */
192
193static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
194{
195 struct buffer_head *dibh;
196 void *kaddr;
197 int error;
198
61057c6b 199 BUG_ON(page->index);
fd88de56 200
b3b94faa
DT
201 error = gfs2_meta_inode_buffer(ip, &dibh);
202 if (error)
203 return error;
204
5c4e9e03 205 kaddr = kmap_atomic(page, KM_USER0);
fd88de56 206 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
b3b94faa 207 ip->i_di.di_size);
fd88de56 208 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
c312c4fd 209 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
210
211 brelse(dibh);
212
213 SetPageUptodate(page);
214
215 return 0;
216}
217
b3b94faa 218
b3b94faa
DT
219/**
220 * gfs2_readpage - readpage with locking
18ec7d5c
SW
221 * @file: The file to read a page for. N.B. This may be NULL if we are
222 * reading an internal file.
b3b94faa
DT
223 * @page: The page to read
224 *
225 * Returns: errno
226 */
227
228static int gfs2_readpage(struct file *file, struct page *page)
229{
feaa7bba
SW
230 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
231 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
dc41aeed 232 struct gfs2_file *gf = NULL;
18ec7d5c 233 struct gfs2_holder gh;
b3b94faa 234 int error;
59a1cc6b 235 int do_unlock = 0;
b3b94faa 236
dd538c83 237 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b 238 if (file) {
dc41aeed 239 gf = file->private_data;
59a1cc6b 240 if (test_bit(GFF_EXLOCK, &gf->f_flags))
dc41aeed 241 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
59a1cc6b
SW
242 goto skip_lock;
243 }
2ca99501 244 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
59a1cc6b 245 do_unlock = 1;
dcd24799 246 error = gfs2_glock_nq_atime(&gh);
fd88de56 247 if (unlikely(error))
61a30dcb
SW
248 goto out_unlock;
249 }
b3b94faa 250
59a1cc6b 251skip_lock:
18ec7d5c 252 if (gfs2_is_stuffed(ip)) {
fd88de56
SW
253 error = stuffed_readpage(ip, page);
254 unlock_page(page);
b3b94faa 255 } else
18ec7d5c 256 error = mpage_readpage(page, gfs2_get_block);
b3b94faa
DT
257
258 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
259 error = -EIO;
260
07903c02 261 if (do_unlock) {
61a30dcb
SW
262 gfs2_glock_dq_m(1, &gh);
263 gfs2_holder_uninit(&gh);
264 }
18ec7d5c 265out:
b3b94faa 266 return error;
18ec7d5c 267out_unlock:
2ca99501
SW
268 if (error == GLR_TRYFAILED)
269 error = AOP_TRUNCATED_PAGE;
18ec7d5c 270 unlock_page(page);
59a1cc6b 271 if (do_unlock)
fd88de56
SW
272 gfs2_holder_uninit(&gh);
273 goto out;
274}
275
fd88de56
SW
276/**
277 * gfs2_readpages - Read a bunch of pages at once
278 *
279 * Some notes:
280 * 1. This is only for readahead, so we can simply ignore any things
281 * which are slightly inconvenient (such as locking conflicts between
282 * the page lock and the glock) and return having done no I/O. Its
283 * obviously not something we'd want to do on too regular a basis.
284 * Any I/O we ignore at this time will be done via readpage later.
e1d5b18a 285 * 2. We don't handle stuffed files here we let readpage do the honours.
fd88de56
SW
286 * 3. mpage_readpages() does most of the heavy lifting in the common case.
287 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
288 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
289 * well as read-ahead.
290 */
291static int gfs2_readpages(struct file *file, struct address_space *mapping,
292 struct list_head *pages, unsigned nr_pages)
293{
294 struct inode *inode = mapping->host;
feaa7bba
SW
295 struct gfs2_inode *ip = GFS2_I(inode);
296 struct gfs2_sbd *sdp = GFS2_SB(inode);
fd88de56 297 struct gfs2_holder gh;
e1d5b18a 298 int ret = 0;
59a1cc6b 299 int do_unlock = 0;
fd88de56 300
dd538c83 301 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b
SW
302 if (file) {
303 struct gfs2_file *gf = file->private_data;
304 if (test_bit(GFF_EXLOCK, &gf->f_flags))
305 goto skip_lock;
306 }
fd88de56 307 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
2ca99501 308 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
59a1cc6b 309 do_unlock = 1;
dcd24799 310 ret = gfs2_glock_nq_atime(&gh);
907b9bce 311 if (ret == GLR_TRYFAILED)
fd88de56
SW
312 goto out_noerror;
313 if (unlikely(ret))
314 goto out_unlock;
315 }
59a1cc6b 316skip_lock:
e1d5b18a 317 if (!gfs2_is_stuffed(ip))
fd88de56 318 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
fd88de56 319
59a1cc6b 320 if (do_unlock) {
fd88de56
SW
321 gfs2_glock_dq_m(1, &gh);
322 gfs2_holder_uninit(&gh);
323 }
324out:
325 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
326 ret = -EIO;
327 return ret;
328out_noerror:
329 ret = 0;
330out_unlock:
59a1cc6b 331 if (do_unlock)
fd88de56 332 gfs2_holder_uninit(&gh);
18ec7d5c 333 goto out;
b3b94faa
DT
334}
335
336/**
337 * gfs2_prepare_write - Prepare to write a page to a file
338 * @file: The file to write to
339 * @page: The page which is to be prepared for writing
340 * @from: From (byte range within page)
341 * @to: To (byte range within page)
342 *
343 * Returns: errno
344 */
345
346static int gfs2_prepare_write(struct file *file, struct page *page,
347 unsigned from, unsigned to)
348{
feaa7bba
SW
349 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
350 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c
SW
351 unsigned int data_blocks, ind_blocks, rblocks;
352 int alloc_required;
b3b94faa 353 int error = 0;
18ec7d5c
SW
354 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
355 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
356 struct gfs2_alloc *al;
52ae7b79
RC
357 unsigned int write_len = to - from;
358
b3b94faa 359
2ca99501 360 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
dcd24799 361 error = gfs2_glock_nq_atime(&ip->i_gh);
2ca99501
SW
362 if (unlikely(error)) {
363 if (error == GLR_TRYFAILED)
364 error = AOP_TRUNCATED_PAGE;
18ec7d5c 365 goto out_uninit;
2ca99501 366 }
b3b94faa 367
52ae7b79 368 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
18ec7d5c 369
52ae7b79 370 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
18ec7d5c
SW
371 if (error)
372 goto out_unlock;
b3b94faa 373
18ec7d5c 374
f5c54804 375 ip->i_alloc.al_requested = 0;
18ec7d5c
SW
376 if (alloc_required) {
377 al = gfs2_alloc_get(ip);
378
379 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
380 if (error)
381 goto out_alloc_put;
382
2933f925 383 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
18ec7d5c
SW
384 if (error)
385 goto out_qunlock;
386
387 al->al_requested = data_blocks + ind_blocks;
388 error = gfs2_inplace_reserve(ip);
389 if (error)
390 goto out_qunlock;
391 }
392
393 rblocks = RES_DINODE + ind_blocks;
394 if (gfs2_is_jdata(ip))
395 rblocks += data_blocks ? data_blocks : 1;
396 if (ind_blocks || data_blocks)
397 rblocks += RES_STATFS + RES_QUOTA;
398
399 error = gfs2_trans_begin(sdp, rblocks, 0);
400 if (error)
401 goto out;
402
403 if (gfs2_is_stuffed(ip)) {
404 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
f25ef0c1 405 error = gfs2_unstuff_dinode(ip, page);
5c4e9e03
SW
406 if (error == 0)
407 goto prepare_write;
408 } else if (!PageUptodate(page))
b3b94faa 409 error = stuffed_readpage(ip, page);
5c4e9e03 410 goto out;
18ec7d5c
SW
411 }
412
5c4e9e03 413prepare_write:
18ec7d5c
SW
414 error = block_prepare_write(page, from, to, gfs2_get_block);
415
416out:
417 if (error) {
418 gfs2_trans_end(sdp);
419 if (alloc_required) {
420 gfs2_inplace_release(ip);
421out_qunlock:
422 gfs2_quota_unlock(ip);
423out_alloc_put:
424 gfs2_alloc_put(ip);
425 }
426out_unlock:
427 gfs2_glock_dq_m(1, &ip->i_gh);
428out_uninit:
429 gfs2_holder_uninit(&ip->i_gh);
430 }
b3b94faa
DT
431
432 return error;
433}
434
435/**
436 * gfs2_commit_write - Commit write to a file
437 * @file: The file to write to
438 * @page: The page containing the data
439 * @from: From (byte range within page)
440 * @to: To (byte range within page)
441 *
442 * Returns: errno
443 */
444
445static int gfs2_commit_write(struct file *file, struct page *page,
446 unsigned from, unsigned to)
447{
448 struct inode *inode = page->mapping->host;
feaa7bba
SW
449 struct gfs2_inode *ip = GFS2_I(inode);
450 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
451 int error = -EOPNOTSUPP;
452 struct buffer_head *dibh;
48516ced
SW
453 struct gfs2_alloc *al = &ip->i_alloc;
454 struct gfs2_dinode *di;
b3b94faa 455
18ec7d5c
SW
456 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
457 goto fail_nounlock;
458
459 error = gfs2_meta_inode_buffer(ip, &dibh);
460 if (error)
461 goto fail_endtrans;
462
463 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
48516ced 464 di = (struct gfs2_dinode *)dibh->b_data;
18ec7d5c 465
b3b94faa 466 if (gfs2_is_stuffed(ip)) {
cd915493 467 u64 file_size;
b3b94faa
DT
468 void *kaddr;
469
cd915493 470 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
b3b94faa 471
18ec7d5c 472 kaddr = kmap_atomic(page, KM_USER0);
b3b94faa 473 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
0bd5996a 474 kaddr + from, to - from);
c312c4fd 475 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
476
477 SetPageUptodate(page);
478
ae619320 479 if (inode->i_size < file_size) {
b3b94faa 480 i_size_write(inode, file_size);
ae619320
SW
481 mark_inode_dirty(inode);
482 }
b3b94faa 483 } else {
568f4c96
SW
484 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
485 gfs2_is_jdata(ip))
257f9b4e 486 gfs2_page_add_databufs(ip, page, from, to);
b3b94faa
DT
487 error = generic_commit_write(file, page, from, to);
488 if (error)
489 goto fail;
490 }
491
48516ced 492 if (ip->i_di.di_size < inode->i_size) {
18ec7d5c 493 ip->i_di.di_size = inode->i_size;
48516ced
SW
494 di->di_size = cpu_to_be64(inode->i_size);
495 }
496
18ec7d5c
SW
497 brelse(dibh);
498 gfs2_trans_end(sdp);
499 if (al->al_requested) {
500 gfs2_inplace_release(ip);
501 gfs2_quota_unlock(ip);
502 gfs2_alloc_put(ip);
503 }
504 gfs2_glock_dq_m(1, &ip->i_gh);
505 gfs2_holder_uninit(&ip->i_gh);
b3b94faa
DT
506 return 0;
507
18ec7d5c
SW
508fail:
509 brelse(dibh);
510fail_endtrans:
511 gfs2_trans_end(sdp);
512 if (al->al_requested) {
513 gfs2_inplace_release(ip);
514 gfs2_quota_unlock(ip);
515 gfs2_alloc_put(ip);
516 }
517 gfs2_glock_dq_m(1, &ip->i_gh);
518 gfs2_holder_uninit(&ip->i_gh);
519fail_nounlock:
b3b94faa 520 ClearPageUptodate(page);
b3b94faa
DT
521 return error;
522}
523
524/**
525 * gfs2_bmap - Block map function
526 * @mapping: Address space info
527 * @lblock: The block to map
528 *
529 * Returns: The disk address for the block or 0 on hole or error
530 */
531
532static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
533{
feaa7bba 534 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa
DT
535 struct gfs2_holder i_gh;
536 sector_t dblock = 0;
537 int error;
538
b3b94faa
DT
539 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
540 if (error)
541 return 0;
542
543 if (!gfs2_is_stuffed(ip))
4ff14670 544 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
b3b94faa
DT
545
546 gfs2_glock_dq_uninit(&i_gh);
547
548 return dblock;
549}
550
551static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
552{
64fb4eb7 553 struct gfs2_bufdata *bd;
b3b94faa
DT
554
555 gfs2_log_lock(sdp);
5c676f6d 556 bd = bh->b_private;
64fb4eb7
SW
557 if (bd) {
558 bd->bd_bh = NULL;
5c676f6d 559 bh->b_private = NULL;
15d00c0b
SW
560 }
561 gfs2_log_unlock(sdp);
b3b94faa
DT
562
563 lock_buffer(bh);
564 clear_buffer_dirty(bh);
565 bh->b_bdev = NULL;
566 clear_buffer_mapped(bh);
567 clear_buffer_req(bh);
568 clear_buffer_new(bh);
569 clear_buffer_delay(bh);
570 unlock_buffer(bh);
571}
572
8628de05 573static void gfs2_invalidatepage(struct page *page, unsigned long offset)
b3b94faa 574{
15d00c0b 575 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
b3b94faa
DT
576 struct buffer_head *head, *bh, *next;
577 unsigned int curr_off = 0;
b3b94faa
DT
578
579 BUG_ON(!PageLocked(page));
580 if (!page_has_buffers(page))
8628de05 581 return;
b3b94faa
DT
582
583 bh = head = page_buffers(page);
584 do {
585 unsigned int next_off = curr_off + bh->b_size;
586 next = bh->b_this_page;
587
588 if (offset <= curr_off)
589 discard_buffer(sdp, bh);
590
591 curr_off = next_off;
592 bh = next;
593 } while (bh != head);
594
595 if (!offset)
8628de05 596 try_to_release_page(page, 0);
b3b94faa 597
8628de05 598 return;
b3b94faa
DT
599}
600
c7b33834
SW
601/**
602 * gfs2_ok_for_dio - check that dio is valid on this file
603 * @ip: The inode
604 * @rw: READ or WRITE
605 * @offset: The offset at which we are reading or writing
606 *
607 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
608 * 1 (to accept the i/o request)
609 */
610static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
611{
612 /*
613 * Should we return an error here? I can't see that O_DIRECT for
614 * a journaled file makes any sense. For now we'll silently fall
615 * back to buffered I/O, likewise we do the same for stuffed
616 * files since they are (a) small and (b) unaligned.
617 */
618 if (gfs2_is_jdata(ip))
619 return 0;
620
621 if (gfs2_is_stuffed(ip))
622 return 0;
623
624 if (offset > i_size_read(&ip->i_inode))
625 return 0;
626 return 1;
627}
628
629
630
a9e5f4d0
SW
631static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
632 const struct iovec *iov, loff_t offset,
633 unsigned long nr_segs)
d1665e41
SW
634{
635 struct file *file = iocb->ki_filp;
636 struct inode *inode = file->f_mapping->host;
feaa7bba 637 struct gfs2_inode *ip = GFS2_I(inode);
d1665e41
SW
638 struct gfs2_holder gh;
639 int rv;
640
641 /*
c7b33834
SW
642 * Deferred lock, even if its a write, since we do no allocation
643 * on this path. All we need change is atime, and this lock mode
644 * ensures that other nodes have flushed their buffered read caches
645 * (i.e. their page cache entries for this inode). We do not,
646 * unfortunately have the option of only flushing a range like
647 * the VFS does.
d1665e41 648 */
c7b33834 649 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
dcd24799 650 rv = gfs2_glock_nq_atime(&gh);
d1665e41 651 if (rv)
c7b33834
SW
652 return rv;
653 rv = gfs2_ok_for_dio(ip, rw, offset);
654 if (rv != 1)
655 goto out; /* dio not valid, fall back to buffered i/o */
656
657 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
658 iov, offset, nr_segs,
659 gfs2_get_block_direct, NULL);
d1665e41
SW
660out:
661 gfs2_glock_dq_m(1, &gh);
662 gfs2_holder_uninit(&gh);
d1665e41
SW
663 return rv;
664}
665
4340fe62
SW
666/**
667 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
668 * @bh: the buffer we're stuck on
669 *
670 */
671
672static void stuck_releasepage(struct buffer_head *bh)
673{
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;
166afccd
SW
678static unsigned limit = 0;
679
0bd5996a 680 if (limit > 3)
166afccd 681 return;
0bd5996a 682 limit++;
4340fe62
SW
683
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");
689
690 if (!bd)
691 return;
692
693 gl = bd->bd_gl;
694
907b9bce 695 fs_warn(sdp, "gl = (%u, %llu)\n",
4340fe62
SW
696 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
697
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");
701
702 if (gl->gl_ops == &gfs2_inode_glops) {
703 struct gfs2_inode *ip = gl->gl_object;
704 unsigned int x;
705
706 if (!ip)
707 return;
708
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);
712
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");
716 }
717}
718
719/**
623d9355 720 * gfs2_releasepage - free the metadata associated with a page
4340fe62
SW
721 * @page: the page that's being released
722 * @gfp_mask: passed from Linux VFS, ignored by us
723 *
724 * Call try_to_free_buffers() if the buffers in this page can be
725 * released.
726 *
727 * Returns: 0
728 */
729
730int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
731{
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;
166afccd 736 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
4340fe62
SW
737
738 if (!page_has_buffers(page))
739 goto out;
740
741 head = bh = page_buffers(page);
742 do {
4340fe62 743 while (atomic_read(&bh->b_count)) {
166afccd
SW
744 if (!atomic_read(&aspace->i_writecount))
745 return 0;
746
61057c6b
RC
747 if (!(gfp_mask & __GFP_WAIT))
748 return 0;
749
166afccd
SW
750 if (time_after_eq(jiffies, t)) {
751 stuck_releasepage(bh);
752 /* should we withdraw here? */
753 return 0;
4340fe62
SW
754 }
755
166afccd 756 yield();
4340fe62
SW
757 }
758
759 gfs2_assert_warn(sdp, !buffer_pinned(bh));
623d9355 760 gfs2_assert_warn(sdp, !buffer_dirty(bh));
4340fe62 761
623d9355 762 gfs2_log_lock(sdp);
4340fe62
SW
763 bd = bh->b_private;
764 if (bd) {
765 gfs2_assert_warn(sdp, bd->bd_bh == bh);
766 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
4340fe62 767 gfs2_assert_warn(sdp, !bd->bd_ail);
623d9355
SW
768 bd->bd_bh = NULL;
769 if (!list_empty(&bd->bd_le.le_list))
770 bd = NULL;
4340fe62
SW
771 bh->b_private = NULL;
772 }
623d9355
SW
773 gfs2_log_unlock(sdp);
774 if (bd)
775 kmem_cache_free(gfs2_bufdata_cachep, bd);
4340fe62
SW
776
777 bh = bh->b_this_page;
166afccd 778 } while (bh != head);
4340fe62 779
a9e5f4d0 780out:
4340fe62
SW
781 return try_to_free_buffers(page);
782}
783
66de045d 784const struct address_space_operations gfs2_file_aops = {
b3b94faa 785 .writepage = gfs2_writepage,
a8d638e3 786 .writepages = gfs2_writepages,
b3b94faa 787 .readpage = gfs2_readpage,
fd88de56 788 .readpages = gfs2_readpages,
b3b94faa
DT
789 .sync_page = block_sync_page,
790 .prepare_write = gfs2_prepare_write,
791 .commit_write = gfs2_commit_write,
792 .bmap = gfs2_bmap,
793 .invalidatepage = gfs2_invalidatepage,
4340fe62 794 .releasepage = gfs2_releasepage,
b3b94faa
DT
795 .direct_IO = gfs2_direct_IO,
796};
797