]> git.proxmox.com Git - mirror_ubuntu-jammy-kernel.git/blame - fs/gfs2/ops_address.c
[GFS2] Fix lack of buffers in writepage bug
[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
7 * of the GNU General Public License v.2.
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>
b3b94faa
DT
20
21#include "gfs2.h"
5c676f6d
SW
22#include "lm_interface.h"
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{
b3b94faa
DT
68 int new = create;
69 uint64_t dblock;
70 int error;
fd88de56 71 int boundary;
b3b94faa 72
fd88de56 73 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
b3b94faa
DT
74 if (error)
75 return error;
76
77 if (!dblock)
78 return 0;
79
80 map_bh(bh_result, inode->i_sb, dblock);
81 if (new)
82 set_buffer_new(bh_result);
fd88de56
SW
83 if (boundary)
84 set_buffer_boundary(bh_result);
b3b94faa
DT
85
86 return 0;
87}
88
89/**
90 * get_block_noalloc - Fills in a buffer head with details about a block
91 * @inode: The inode
92 * @lblock: The block number to look up
93 * @bh_result: The buffer head to return the result in
94 * @create: Non-zero if we may add block to the file
95 *
96 * Returns: errno
97 */
98
99static int get_block_noalloc(struct inode *inode, sector_t lblock,
100 struct buffer_head *bh_result, int create)
101{
b3b94faa
DT
102 int new = 0;
103 uint64_t dblock;
104 int error;
fd88de56 105 int boundary;
b3b94faa 106
fd88de56 107 error = gfs2_block_map(inode, lblock, &new, &dblock, &boundary);
b3b94faa
DT
108 if (error)
109 return error;
110
111 if (dblock)
112 map_bh(bh_result, inode->i_sb, dblock);
feaa7bba 113 else if (gfs2_assert_withdraw(GFS2_SB(inode), !create))
b3b94faa 114 error = -EIO;
fd88de56
SW
115 if (boundary)
116 set_buffer_boundary(bh_result);
b3b94faa
DT
117
118 return error;
119}
120
b3b94faa
DT
121/**
122 * gfs2_writepage - Write complete page
123 * @page: Page to write
124 *
125 * Returns: errno
126 *
18ec7d5c
SW
127 * Some of this is copied from block_write_full_page() although we still
128 * call it to do most of the work.
b3b94faa
DT
129 */
130
131static int gfs2_writepage(struct page *page, struct writeback_control *wbc)
132{
18ec7d5c 133 struct inode *inode = page->mapping->host;
f4387149
SW
134 struct gfs2_inode *ip = GFS2_I(inode);
135 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
136 loff_t i_size = i_size_read(inode);
137 pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
138 unsigned offset;
b3b94faa 139 int error;
18ec7d5c 140 int done_trans = 0;
b3b94faa 141
b3b94faa
DT
142 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) {
143 unlock_page(page);
144 return -EIO;
145 }
5c676f6d 146 if (current->journal_info)
18ec7d5c
SW
147 goto out_ignore;
148
149 /* Is the page fully outside i_size? (truncate in progress) */
150 offset = i_size & (PAGE_CACHE_SIZE-1);
d2d7b8a2 151 if (page->index > end_index || (page->index == end_index && !offset)) {
18ec7d5c 152 page->mapping->a_ops->invalidatepage(page, 0);
b3b94faa 153 unlock_page(page);
18ec7d5c 154 return 0; /* don't care */
b3b94faa
DT
155 }
156
18ec7d5c
SW
157 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED || gfs2_is_jdata(ip)) {
158 error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0);
159 if (error)
160 goto out_ignore;
f4387149
SW
161 if (!page_has_buffers(page)) {
162 create_empty_buffers(page, inode->i_sb->s_blocksize,
163 (1 << BH_Dirty)|(1 << BH_Uptodate));
164 }
18ec7d5c
SW
165 gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1);
166 done_trans = 1;
167 }
18ec7d5c
SW
168 error = block_write_full_page(page, get_block_noalloc, wbc);
169 if (done_trans)
170 gfs2_trans_end(sdp);
b3b94faa 171 gfs2_meta_cache_flush(ip);
b3b94faa 172 return error;
18ec7d5c
SW
173
174out_ignore:
175 redirty_page_for_writepage(wbc, page);
176 unlock_page(page);
177 return 0;
b3b94faa
DT
178}
179
fd88de56
SW
180static int zero_readpage(struct page *page)
181{
182 void *kaddr;
183
184 kaddr = kmap_atomic(page, KM_USER0);
185 memset(kaddr, 0, PAGE_CACHE_SIZE);
186 kunmap_atomic(page, KM_USER0);
187
188 SetPageUptodate(page);
189
190 return 0;
191}
192
b3b94faa
DT
193/**
194 * stuffed_readpage - Fill in a Linux page with stuffed file data
195 * @ip: the inode
196 * @page: the page
197 *
198 * Returns: errno
199 */
200
201static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
202{
203 struct buffer_head *dibh;
204 void *kaddr;
205 int error;
206
fd88de56
SW
207 /* Only the first page of a stuffed file might contain data */
208 if (unlikely(page->index))
209 return zero_readpage(page);
210
b3b94faa
DT
211 error = gfs2_meta_inode_buffer(ip, &dibh);
212 if (error)
213 return error;
214
5c4e9e03 215 kaddr = kmap_atomic(page, KM_USER0);
fd88de56 216 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
b3b94faa 217 ip->i_di.di_size);
fd88de56 218 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
5c4e9e03 219 kunmap_atomic(page, KM_USER0);
b3b94faa
DT
220
221 brelse(dibh);
222
223 SetPageUptodate(page);
224
225 return 0;
226}
227
b3b94faa 228
b3b94faa
DT
229/**
230 * gfs2_readpage - readpage with locking
18ec7d5c
SW
231 * @file: The file to read a page for. N.B. This may be NULL if we are
232 * reading an internal file.
b3b94faa
DT
233 * @page: The page to read
234 *
235 * Returns: errno
236 */
237
238static int gfs2_readpage(struct file *file, struct page *page)
239{
feaa7bba
SW
240 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
241 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c 242 struct gfs2_holder gh;
b3b94faa 243 int error;
59a1cc6b 244 int do_unlock = 0;
b3b94faa 245
fd88de56 246 if (likely(file != &gfs2_internal_file_sentinal)) {
59a1cc6b
SW
247 if (file) {
248 struct gfs2_file *gf = file->private_data;
249 if (test_bit(GFF_EXLOCK, &gf->f_flags))
250 goto skip_lock;
251 }
fe1bdedc 252 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
59a1cc6b 253 do_unlock = 1;
61a30dcb 254 error = gfs2_glock_nq_m_atime(1, &gh);
fd88de56 255 if (unlikely(error))
61a30dcb
SW
256 goto out_unlock;
257 }
b3b94faa 258
59a1cc6b 259skip_lock:
18ec7d5c 260 if (gfs2_is_stuffed(ip)) {
fd88de56
SW
261 error = stuffed_readpage(ip, page);
262 unlock_page(page);
b3b94faa 263 } else
18ec7d5c 264 error = mpage_readpage(page, gfs2_get_block);
b3b94faa
DT
265
266 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
267 error = -EIO;
268
61a30dcb
SW
269 if (file != &gfs2_internal_file_sentinal) {
270 gfs2_glock_dq_m(1, &gh);
271 gfs2_holder_uninit(&gh);
272 }
18ec7d5c 273out:
b3b94faa 274 return error;
18ec7d5c
SW
275out_unlock:
276 unlock_page(page);
59a1cc6b 277 if (do_unlock)
fd88de56
SW
278 gfs2_holder_uninit(&gh);
279 goto out;
280}
281
fd88de56
SW
282/**
283 * gfs2_readpages - Read a bunch of pages at once
284 *
285 * Some notes:
286 * 1. This is only for readahead, so we can simply ignore any things
287 * which are slightly inconvenient (such as locking conflicts between
288 * the page lock and the glock) and return having done no I/O. Its
289 * obviously not something we'd want to do on too regular a basis.
290 * Any I/O we ignore at this time will be done via readpage later.
291 * 2. We have to handle stuffed files here too.
292 * 3. mpage_readpages() does most of the heavy lifting in the common case.
293 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
294 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
295 * well as read-ahead.
296 */
297static int gfs2_readpages(struct file *file, struct address_space *mapping,
298 struct list_head *pages, unsigned nr_pages)
299{
300 struct inode *inode = mapping->host;
feaa7bba
SW
301 struct gfs2_inode *ip = GFS2_I(inode);
302 struct gfs2_sbd *sdp = GFS2_SB(inode);
fd88de56
SW
303 struct gfs2_holder gh;
304 unsigned page_idx;
305 int ret;
59a1cc6b 306 int do_unlock = 0;
fd88de56
SW
307
308 if (likely(file != &gfs2_internal_file_sentinal)) {
59a1cc6b
SW
309 if (file) {
310 struct gfs2_file *gf = file->private_data;
311 if (test_bit(GFF_EXLOCK, &gf->f_flags))
312 goto skip_lock;
313 }
fd88de56
SW
314 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
315 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
59a1cc6b 316 do_unlock = 1;
fd88de56
SW
317 ret = gfs2_glock_nq_m_atime(1, &gh);
318 if (ret == GLR_TRYFAILED)
319 goto out_noerror;
320 if (unlikely(ret))
321 goto out_unlock;
322 }
59a1cc6b 323skip_lock:
fd88de56
SW
324 if (gfs2_is_stuffed(ip)) {
325 struct pagevec lru_pvec;
326 pagevec_init(&lru_pvec, 0);
327 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
ffeb874b
SW
328 struct page *page = list_entry(pages->prev, struct page, lru);
329 prefetchw(&page->flags);
fd88de56
SW
330 list_del(&page->lru);
331 if (!add_to_page_cache(page, mapping,
332 page->index, GFP_KERNEL)) {
333 ret = stuffed_readpage(ip, page);
334 unlock_page(page);
335 if (!pagevec_add(&lru_pvec, page))
336 __pagevec_lru_add(&lru_pvec);
ffeb874b
SW
337 } else {
338 page_cache_release(page);
fd88de56 339 }
fd88de56
SW
340 }
341 pagevec_lru_add(&lru_pvec);
342 ret = 0;
343 } else {
344 /* What we really want to do .... */
345 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
346 }
347
59a1cc6b 348 if (do_unlock) {
fd88de56
SW
349 gfs2_glock_dq_m(1, &gh);
350 gfs2_holder_uninit(&gh);
351 }
352out:
353 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
354 ret = -EIO;
355 return ret;
356out_noerror:
357 ret = 0;
358out_unlock:
359 /* unlock all pages, we can't do any I/O right now */
360 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
ffeb874b 361 struct page *page = list_entry(pages->prev, struct page, lru);
fd88de56
SW
362 list_del(&page->lru);
363 unlock_page(page);
364 page_cache_release(page);
365 }
59a1cc6b 366 if (do_unlock)
fd88de56 367 gfs2_holder_uninit(&gh);
18ec7d5c 368 goto out;
b3b94faa
DT
369}
370
371/**
372 * gfs2_prepare_write - Prepare to write a page to a file
373 * @file: The file to write to
374 * @page: The page which is to be prepared for writing
375 * @from: From (byte range within page)
376 * @to: To (byte range within page)
377 *
378 * Returns: errno
379 */
380
381static int gfs2_prepare_write(struct file *file, struct page *page,
382 unsigned from, unsigned to)
383{
feaa7bba
SW
384 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
385 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c
SW
386 unsigned int data_blocks, ind_blocks, rblocks;
387 int alloc_required;
b3b94faa 388 int error = 0;
18ec7d5c
SW
389 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
390 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
391 struct gfs2_alloc *al;
b3b94faa 392
fe1bdedc 393 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
18ec7d5c
SW
394 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
395 if (error)
396 goto out_uninit;
b3b94faa 397
18ec7d5c
SW
398 gfs2_write_calc_reserv(ip, to - from, &data_blocks, &ind_blocks);
399
400 error = gfs2_write_alloc_required(ip, pos, from - to, &alloc_required);
401 if (error)
402 goto out_unlock;
b3b94faa 403
18ec7d5c
SW
404
405 if (alloc_required) {
406 al = gfs2_alloc_get(ip);
407
408 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
409 if (error)
410 goto out_alloc_put;
411
412 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
413 if (error)
414 goto out_qunlock;
415
416 al->al_requested = data_blocks + ind_blocks;
417 error = gfs2_inplace_reserve(ip);
418 if (error)
419 goto out_qunlock;
420 }
421
422 rblocks = RES_DINODE + ind_blocks;
423 if (gfs2_is_jdata(ip))
424 rblocks += data_blocks ? data_blocks : 1;
425 if (ind_blocks || data_blocks)
426 rblocks += RES_STATFS + RES_QUOTA;
427
428 error = gfs2_trans_begin(sdp, rblocks, 0);
429 if (error)
430 goto out;
431
432 if (gfs2_is_stuffed(ip)) {
433 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
f25ef0c1 434 error = gfs2_unstuff_dinode(ip, page);
5c4e9e03
SW
435 if (error == 0)
436 goto prepare_write;
437 } else if (!PageUptodate(page))
b3b94faa 438 error = stuffed_readpage(ip, page);
5c4e9e03 439 goto out;
18ec7d5c
SW
440 }
441
5c4e9e03 442prepare_write:
18ec7d5c
SW
443 error = block_prepare_write(page, from, to, gfs2_get_block);
444
445out:
446 if (error) {
447 gfs2_trans_end(sdp);
448 if (alloc_required) {
449 gfs2_inplace_release(ip);
450out_qunlock:
451 gfs2_quota_unlock(ip);
452out_alloc_put:
453 gfs2_alloc_put(ip);
454 }
455out_unlock:
456 gfs2_glock_dq_m(1, &ip->i_gh);
457out_uninit:
458 gfs2_holder_uninit(&ip->i_gh);
459 }
b3b94faa
DT
460
461 return error;
462}
463
464/**
465 * gfs2_commit_write - Commit write to a file
466 * @file: The file to write to
467 * @page: The page containing the data
468 * @from: From (byte range within page)
469 * @to: To (byte range within page)
470 *
471 * Returns: errno
472 */
473
474static int gfs2_commit_write(struct file *file, struct page *page,
475 unsigned from, unsigned to)
476{
477 struct inode *inode = page->mapping->host;
feaa7bba
SW
478 struct gfs2_inode *ip = GFS2_I(inode);
479 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
480 int error = -EOPNOTSUPP;
481 struct buffer_head *dibh;
482 struct gfs2_alloc *al = &ip->i_alloc;;
b3b94faa 483
18ec7d5c
SW
484 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
485 goto fail_nounlock;
486
487 error = gfs2_meta_inode_buffer(ip, &dibh);
488 if (error)
489 goto fail_endtrans;
490
491 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
492
b3b94faa 493 if (gfs2_is_stuffed(ip)) {
b3b94faa
DT
494 uint64_t file_size;
495 void *kaddr;
496
497 file_size = ((uint64_t)page->index << PAGE_CACHE_SHIFT) + to;
498
18ec7d5c 499 kaddr = kmap_atomic(page, KM_USER0);
b3b94faa 500 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
18ec7d5c
SW
501 (char *)kaddr + from, to - from);
502 kunmap_atomic(page, KM_USER0);
b3b94faa
DT
503
504 SetPageUptodate(page);
505
506 if (inode->i_size < file_size)
507 i_size_write(inode, file_size);
508 } else {
568f4c96
SW
509 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
510 gfs2_is_jdata(ip))
257f9b4e 511 gfs2_page_add_databufs(ip, page, from, to);
b3b94faa
DT
512 error = generic_commit_write(file, page, from, to);
513 if (error)
514 goto fail;
515 }
516
18ec7d5c
SW
517 if (ip->i_di.di_size < inode->i_size)
518 ip->i_di.di_size = inode->i_size;
519
520 gfs2_dinode_out(&ip->i_di, dibh->b_data);
521 brelse(dibh);
522 gfs2_trans_end(sdp);
523 if (al->al_requested) {
524 gfs2_inplace_release(ip);
525 gfs2_quota_unlock(ip);
526 gfs2_alloc_put(ip);
527 }
528 gfs2_glock_dq_m(1, &ip->i_gh);
529 gfs2_holder_uninit(&ip->i_gh);
b3b94faa
DT
530 return 0;
531
18ec7d5c
SW
532fail:
533 brelse(dibh);
534fail_endtrans:
535 gfs2_trans_end(sdp);
536 if (al->al_requested) {
537 gfs2_inplace_release(ip);
538 gfs2_quota_unlock(ip);
539 gfs2_alloc_put(ip);
540 }
541 gfs2_glock_dq_m(1, &ip->i_gh);
542 gfs2_holder_uninit(&ip->i_gh);
543fail_nounlock:
b3b94faa 544 ClearPageUptodate(page);
b3b94faa
DT
545 return error;
546}
547
548/**
549 * gfs2_bmap - Block map function
550 * @mapping: Address space info
551 * @lblock: The block to map
552 *
553 * Returns: The disk address for the block or 0 on hole or error
554 */
555
556static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
557{
feaa7bba 558 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa
DT
559 struct gfs2_holder i_gh;
560 sector_t dblock = 0;
561 int error;
562
b3b94faa
DT
563 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
564 if (error)
565 return 0;
566
567 if (!gfs2_is_stuffed(ip))
4ff14670 568 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
b3b94faa
DT
569
570 gfs2_glock_dq_uninit(&i_gh);
571
572 return dblock;
573}
574
575static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
576{
64fb4eb7 577 struct gfs2_bufdata *bd;
b3b94faa
DT
578
579 gfs2_log_lock(sdp);
5c676f6d 580 bd = bh->b_private;
64fb4eb7
SW
581 if (bd) {
582 bd->bd_bh = NULL;
5c676f6d 583 bh->b_private = NULL;
b3b94faa
DT
584 gfs2_log_unlock(sdp);
585 brelse(bh);
586 } else
587 gfs2_log_unlock(sdp);
588
589 lock_buffer(bh);
590 clear_buffer_dirty(bh);
591 bh->b_bdev = NULL;
592 clear_buffer_mapped(bh);
593 clear_buffer_req(bh);
594 clear_buffer_new(bh);
595 clear_buffer_delay(bh);
596 unlock_buffer(bh);
597}
598
8628de05 599static void gfs2_invalidatepage(struct page *page, unsigned long offset)
b3b94faa 600{
5c676f6d 601 struct gfs2_sbd *sdp = page->mapping->host->i_sb->s_fs_info;
b3b94faa
DT
602 struct buffer_head *head, *bh, *next;
603 unsigned int curr_off = 0;
b3b94faa
DT
604
605 BUG_ON(!PageLocked(page));
606 if (!page_has_buffers(page))
8628de05 607 return;
b3b94faa
DT
608
609 bh = head = page_buffers(page);
610 do {
611 unsigned int next_off = curr_off + bh->b_size;
612 next = bh->b_this_page;
613
614 if (offset <= curr_off)
615 discard_buffer(sdp, bh);
616
617 curr_off = next_off;
618 bh = next;
619 } while (bh != head);
620
621 if (!offset)
8628de05 622 try_to_release_page(page, 0);
b3b94faa 623
8628de05 624 return;
b3b94faa
DT
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
a9e5f4d0
SW
637 if (rw == READ)
638 mutex_lock(&inode->i_mutex);
d1665e41 639 /*
a9e5f4d0 640 * Shared lock, even if its a write, since we do no allocation
d1665e41
SW
641 * on this path. All we need change is atime.
642 */
643 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
644 rv = gfs2_glock_nq_m_atime(1, &gh);
645 if (rv)
646 goto out;
647
a9e5f4d0
SW
648 if (offset > i_size_read(inode))
649 goto out;
650
d1665e41
SW
651 /*
652 * Should we return an error here? I can't see that O_DIRECT for
653 * a journaled file makes any sense. For now we'll silently fall
654 * back to buffered I/O, likewise we do the same for stuffed
655 * files since they are (a) small and (b) unaligned.
656 */
657 if (gfs2_is_jdata(ip))
658 goto out;
659
660 if (gfs2_is_stuffed(ip))
661 goto out;
662
a9e5f4d0
SW
663 rv = blockdev_direct_IO_own_locking(rw, iocb, inode,
664 inode->i_sb->s_bdev,
665 iov, offset, nr_segs,
666 gfs2_get_block, NULL);
d1665e41
SW
667out:
668 gfs2_glock_dq_m(1, &gh);
669 gfs2_holder_uninit(&gh);
a9e5f4d0
SW
670 if (rw == READ)
671 mutex_unlock(&inode->i_mutex);
d1665e41
SW
672
673 return rv;
674}
675
4340fe62
SW
676/**
677 * stuck_releasepage - We're stuck in gfs2_releasepage(). Print stuff out.
678 * @bh: the buffer we're stuck on
679 *
680 */
681
682static void stuck_releasepage(struct buffer_head *bh)
683{
684 struct inode *inode = bh->b_page->mapping->host;
685 struct gfs2_sbd *sdp = inode->i_sb->s_fs_info;
686 struct gfs2_bufdata *bd = bh->b_private;
687 struct gfs2_glock *gl;
688
689 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
690 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
691 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
692 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
693 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
694
695 if (!bd)
696 return;
697
698 gl = bd->bd_gl;
699
700 fs_warn(sdp, "gl = (%u, %llu)\n",
701 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
702
703 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
704 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
705 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
706
707 if (gl->gl_ops == &gfs2_inode_glops) {
708 struct gfs2_inode *ip = gl->gl_object;
709 unsigned int x;
710
711 if (!ip)
712 return;
713
714 fs_warn(sdp, "ip = %llu %llu\n",
715 (unsigned long long)ip->i_num.no_formal_ino,
716 (unsigned long long)ip->i_num.no_addr);
717
718 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
719 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
720 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
721 }
722}
723
724/**
725 * gfs2_aspace_releasepage - free the metadata associated with a page
726 * @page: the page that's being released
727 * @gfp_mask: passed from Linux VFS, ignored by us
728 *
729 * Call try_to_free_buffers() if the buffers in this page can be
730 * released.
731 *
732 * Returns: 0
733 */
734
735int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
736{
737 struct inode *aspace = page->mapping->host;
738 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
739 struct buffer_head *bh, *head;
740 struct gfs2_bufdata *bd;
741 unsigned long t;
742
743 if (!page_has_buffers(page))
744 goto out;
745
746 head = bh = page_buffers(page);
747 do {
748 t = jiffies;
749
750 while (atomic_read(&bh->b_count)) {
751 if (atomic_read(&aspace->i_writecount)) {
752 if (time_after_eq(jiffies, t +
753 gfs2_tune_get(sdp, gt_stall_secs) * HZ)) {
754 stuck_releasepage(bh);
755 t = jiffies;
756 }
757
758 yield();
759 continue;
760 }
761
762 return 0;
763 }
764
765 gfs2_assert_warn(sdp, !buffer_pinned(bh));
766
767 bd = bh->b_private;
768 if (bd) {
769 gfs2_assert_warn(sdp, bd->bd_bh == bh);
770 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
771 gfs2_assert_warn(sdp, list_empty(&bd->bd_le.le_list));
772 gfs2_assert_warn(sdp, !bd->bd_ail);
773 kmem_cache_free(gfs2_bufdata_cachep, bd);
774 bh->b_private = NULL;
775 }
776
777 bh = bh->b_this_page;
778 }
779 while (bh != head);
780
a9e5f4d0 781out:
4340fe62
SW
782 return try_to_free_buffers(page);
783}
784
66de045d 785const struct address_space_operations gfs2_file_aops = {
b3b94faa
DT
786 .writepage = gfs2_writepage,
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