Merge git://git.kernel.org/pub/scm/linux/kernel/git/dtor/input
[linux-2.6-block.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{
7276b3b0 68 return gfs2_block_map(inode, lblock, create, bh_result, 32);
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
7a6bbacb 86 error = gfs2_block_map(inode, lblock, 0, bh_result, 1);
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{
56965536 97 return gfs2_block_map(inode, lblock, 0, bh_result, 32);
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
fd88de56
SW
159static int zero_readpage(struct page *page)
160{
161 void *kaddr;
162
163 kaddr = kmap_atomic(page, KM_USER0);
164 memset(kaddr, 0, PAGE_CACHE_SIZE);
c312c4fd 165 kunmap_atomic(kaddr, KM_USER0);
fd88de56
SW
166
167 SetPageUptodate(page);
168
169 return 0;
170}
171
b3b94faa
DT
172/**
173 * stuffed_readpage - Fill in a Linux page with stuffed file data
174 * @ip: the inode
175 * @page: the page
176 *
177 * Returns: errno
178 */
179
180static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
181{
182 struct buffer_head *dibh;
183 void *kaddr;
184 int error;
185
fd88de56
SW
186 /* Only the first page of a stuffed file might contain data */
187 if (unlikely(page->index))
188 return zero_readpage(page);
189
b3b94faa
DT
190 error = gfs2_meta_inode_buffer(ip, &dibh);
191 if (error)
192 return error;
193
5c4e9e03 194 kaddr = kmap_atomic(page, KM_USER0);
fd88de56 195 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
b3b94faa 196 ip->i_di.di_size);
fd88de56 197 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
c312c4fd 198 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
199
200 brelse(dibh);
201
202 SetPageUptodate(page);
203
204 return 0;
205}
206
b3b94faa 207
b3b94faa
DT
208/**
209 * gfs2_readpage - readpage with locking
18ec7d5c
SW
210 * @file: The file to read a page for. N.B. This may be NULL if we are
211 * reading an internal file.
b3b94faa
DT
212 * @page: The page to read
213 *
214 * Returns: errno
215 */
216
217static int gfs2_readpage(struct file *file, struct page *page)
218{
feaa7bba
SW
219 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
220 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
dc41aeed 221 struct gfs2_file *gf = NULL;
18ec7d5c 222 struct gfs2_holder gh;
b3b94faa 223 int error;
59a1cc6b 224 int do_unlock = 0;
b3b94faa 225
dd538c83 226 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b 227 if (file) {
dc41aeed 228 gf = file->private_data;
59a1cc6b 229 if (test_bit(GFF_EXLOCK, &gf->f_flags))
dc41aeed 230 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
59a1cc6b
SW
231 goto skip_lock;
232 }
fe1bdedc 233 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|GL_AOP, &gh);
59a1cc6b 234 do_unlock = 1;
61a30dcb 235 error = gfs2_glock_nq_m_atime(1, &gh);
fd88de56 236 if (unlikely(error))
61a30dcb
SW
237 goto out_unlock;
238 }
b3b94faa 239
59a1cc6b 240skip_lock:
18ec7d5c 241 if (gfs2_is_stuffed(ip)) {
fd88de56
SW
242 error = stuffed_readpage(ip, page);
243 unlock_page(page);
b3b94faa 244 } else
18ec7d5c 245 error = mpage_readpage(page, gfs2_get_block);
b3b94faa
DT
246
247 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
248 error = -EIO;
249
07903c02 250 if (do_unlock) {
61a30dcb
SW
251 gfs2_glock_dq_m(1, &gh);
252 gfs2_holder_uninit(&gh);
253 }
18ec7d5c 254out:
b3b94faa 255 return error;
18ec7d5c
SW
256out_unlock:
257 unlock_page(page);
59a1cc6b 258 if (do_unlock)
fd88de56
SW
259 gfs2_holder_uninit(&gh);
260 goto out;
261}
262
fd88de56
SW
263/**
264 * gfs2_readpages - Read a bunch of pages at once
265 *
266 * Some notes:
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.
277 */
278static int gfs2_readpages(struct file *file, struct address_space *mapping,
279 struct list_head *pages, unsigned nr_pages)
280{
281 struct inode *inode = mapping->host;
feaa7bba
SW
282 struct gfs2_inode *ip = GFS2_I(inode);
283 struct gfs2_sbd *sdp = GFS2_SB(inode);
fd88de56
SW
284 struct gfs2_holder gh;
285 unsigned page_idx;
286 int ret;
59a1cc6b 287 int do_unlock = 0;
fd88de56 288
dd538c83 289 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b
SW
290 if (file) {
291 struct gfs2_file *gf = file->private_data;
292 if (test_bit(GFF_EXLOCK, &gf->f_flags))
293 goto skip_lock;
294 }
fd88de56
SW
295 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
296 LM_FLAG_TRY_1CB|GL_ATIME|GL_AOP, &gh);
59a1cc6b 297 do_unlock = 1;
fd88de56 298 ret = gfs2_glock_nq_m_atime(1, &gh);
907b9bce 299 if (ret == GLR_TRYFAILED)
fd88de56
SW
300 goto out_noerror;
301 if (unlikely(ret))
302 goto out_unlock;
303 }
59a1cc6b 304skip_lock:
fd88de56
SW
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++) {
ffeb874b
SW
309 struct page *page = list_entry(pages->prev, struct page, lru);
310 prefetchw(&page->flags);
fd88de56
SW
311 list_del(&page->lru);
312 if (!add_to_page_cache(page, mapping,
313 page->index, GFP_KERNEL)) {
314 ret = stuffed_readpage(ip, page);
315 unlock_page(page);
316 if (!pagevec_add(&lru_pvec, page))
317 __pagevec_lru_add(&lru_pvec);
ffeb874b
SW
318 } else {
319 page_cache_release(page);
fd88de56 320 }
fd88de56
SW
321 }
322 pagevec_lru_add(&lru_pvec);
323 ret = 0;
324 } else {
325 /* What we really want to do .... */
326 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
327 }
328
59a1cc6b 329 if (do_unlock) {
fd88de56
SW
330 gfs2_glock_dq_m(1, &gh);
331 gfs2_holder_uninit(&gh);
332 }
333out:
334 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
335 ret = -EIO;
336 return ret;
337out_noerror:
338 ret = 0;
339out_unlock:
340 /* unlock all pages, we can't do any I/O right now */
341 for (page_idx = 0; page_idx < nr_pages; page_idx++) {
ffeb874b 342 struct page *page = list_entry(pages->prev, struct page, lru);
fd88de56
SW
343 list_del(&page->lru);
344 unlock_page(page);
345 page_cache_release(page);
346 }
59a1cc6b 347 if (do_unlock)
fd88de56 348 gfs2_holder_uninit(&gh);
18ec7d5c 349 goto out;
b3b94faa
DT
350}
351
352/**
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)
358 *
359 * Returns: errno
360 */
361
362static int gfs2_prepare_write(struct file *file, struct page *page,
363 unsigned from, unsigned to)
364{
feaa7bba
SW
365 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
366 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c
SW
367 unsigned int data_blocks, ind_blocks, rblocks;
368 int alloc_required;
b3b94faa 369 int error = 0;
18ec7d5c
SW
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;
52ae7b79
RC
373 unsigned int write_len = to - from;
374
b3b94faa 375
fe1bdedc 376 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|GL_AOP, &ip->i_gh);
18ec7d5c
SW
377 error = gfs2_glock_nq_m_atime(1, &ip->i_gh);
378 if (error)
379 goto out_uninit;
b3b94faa 380
52ae7b79 381 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
18ec7d5c 382
52ae7b79 383 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
18ec7d5c
SW
384 if (error)
385 goto out_unlock;
b3b94faa 386
18ec7d5c 387
f5c54804 388 ip->i_alloc.al_requested = 0;
18ec7d5c
SW
389 if (alloc_required) {
390 al = gfs2_alloc_get(ip);
391
392 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
393 if (error)
394 goto out_alloc_put;
395
396 error = gfs2_quota_check(ip, ip->i_di.di_uid, ip->i_di.di_gid);
397 if (error)
398 goto out_qunlock;
399
400 al->al_requested = data_blocks + ind_blocks;
401 error = gfs2_inplace_reserve(ip);
402 if (error)
403 goto out_qunlock;
404 }
405
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;
411
412 error = gfs2_trans_begin(sdp, rblocks, 0);
413 if (error)
414 goto out;
415
416 if (gfs2_is_stuffed(ip)) {
417 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
f25ef0c1 418 error = gfs2_unstuff_dinode(ip, page);
5c4e9e03
SW
419 if (error == 0)
420 goto prepare_write;
421 } else if (!PageUptodate(page))
b3b94faa 422 error = stuffed_readpage(ip, page);
5c4e9e03 423 goto out;
18ec7d5c
SW
424 }
425
5c4e9e03 426prepare_write:
18ec7d5c
SW
427 error = block_prepare_write(page, from, to, gfs2_get_block);
428
429out:
430 if (error) {
431 gfs2_trans_end(sdp);
432 if (alloc_required) {
433 gfs2_inplace_release(ip);
434out_qunlock:
435 gfs2_quota_unlock(ip);
436out_alloc_put:
437 gfs2_alloc_put(ip);
438 }
439out_unlock:
440 gfs2_glock_dq_m(1, &ip->i_gh);
441out_uninit:
442 gfs2_holder_uninit(&ip->i_gh);
443 }
b3b94faa
DT
444
445 return error;
446}
447
448/**
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)
454 *
455 * Returns: errno
456 */
457
458static int gfs2_commit_write(struct file *file, struct page *page,
459 unsigned from, unsigned to)
460{
461 struct inode *inode = page->mapping->host;
feaa7bba
SW
462 struct gfs2_inode *ip = GFS2_I(inode);
463 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
464 int error = -EOPNOTSUPP;
465 struct buffer_head *dibh;
48516ced
SW
466 struct gfs2_alloc *al = &ip->i_alloc;
467 struct gfs2_dinode *di;
b3b94faa 468
18ec7d5c
SW
469 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
470 goto fail_nounlock;
471
472 error = gfs2_meta_inode_buffer(ip, &dibh);
473 if (error)
474 goto fail_endtrans;
475
476 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
48516ced 477 di = (struct gfs2_dinode *)dibh->b_data;
18ec7d5c 478
b3b94faa 479 if (gfs2_is_stuffed(ip)) {
cd915493 480 u64 file_size;
b3b94faa
DT
481 void *kaddr;
482
cd915493 483 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
b3b94faa 484
18ec7d5c 485 kaddr = kmap_atomic(page, KM_USER0);
b3b94faa 486 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
0bd5996a 487 kaddr + from, to - from);
c312c4fd 488 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
489
490 SetPageUptodate(page);
491
492 if (inode->i_size < file_size)
493 i_size_write(inode, file_size);
494 } else {
568f4c96
SW
495 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
496 gfs2_is_jdata(ip))
257f9b4e 497 gfs2_page_add_databufs(ip, page, from, to);
b3b94faa
DT
498 error = generic_commit_write(file, page, from, to);
499 if (error)
500 goto fail;
501 }
502
48516ced 503 if (ip->i_di.di_size < inode->i_size) {
18ec7d5c 504 ip->i_di.di_size = inode->i_size;
48516ced
SW
505 di->di_size = cpu_to_be64(inode->i_size);
506 }
507
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);
18ec7d5c 512
18ec7d5c
SW
513 brelse(dibh);
514 gfs2_trans_end(sdp);
515 if (al->al_requested) {
516 gfs2_inplace_release(ip);
517 gfs2_quota_unlock(ip);
518 gfs2_alloc_put(ip);
519 }
520 gfs2_glock_dq_m(1, &ip->i_gh);
521 gfs2_holder_uninit(&ip->i_gh);
b3b94faa
DT
522 return 0;
523
18ec7d5c
SW
524fail:
525 brelse(dibh);
526fail_endtrans:
527 gfs2_trans_end(sdp);
528 if (al->al_requested) {
529 gfs2_inplace_release(ip);
530 gfs2_quota_unlock(ip);
531 gfs2_alloc_put(ip);
532 }
533 gfs2_glock_dq_m(1, &ip->i_gh);
534 gfs2_holder_uninit(&ip->i_gh);
535fail_nounlock:
b3b94faa 536 ClearPageUptodate(page);
b3b94faa
DT
537 return error;
538}
539
540/**
541 * gfs2_bmap - Block map function
542 * @mapping: Address space info
543 * @lblock: The block to map
544 *
545 * Returns: The disk address for the block or 0 on hole or error
546 */
547
548static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
549{
feaa7bba 550 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa
DT
551 struct gfs2_holder i_gh;
552 sector_t dblock = 0;
553 int error;
554
b3b94faa
DT
555 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
556 if (error)
557 return 0;
558
559 if (!gfs2_is_stuffed(ip))
4ff14670 560 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
b3b94faa
DT
561
562 gfs2_glock_dq_uninit(&i_gh);
563
564 return dblock;
565}
566
567static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
568{
64fb4eb7 569 struct gfs2_bufdata *bd;
b3b94faa
DT
570
571 gfs2_log_lock(sdp);
5c676f6d 572 bd = bh->b_private;
64fb4eb7
SW
573 if (bd) {
574 bd->bd_bh = NULL;
5c676f6d 575 bh->b_private = NULL;
15d00c0b
SW
576 }
577 gfs2_log_unlock(sdp);
b3b94faa
DT
578
579 lock_buffer(bh);
580 clear_buffer_dirty(bh);
581 bh->b_bdev = NULL;
582 clear_buffer_mapped(bh);
583 clear_buffer_req(bh);
584 clear_buffer_new(bh);
585 clear_buffer_delay(bh);
586 unlock_buffer(bh);
587}
588
8628de05 589static void gfs2_invalidatepage(struct page *page, unsigned long offset)
b3b94faa 590{
15d00c0b 591 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
b3b94faa
DT
592 struct buffer_head *head, *bh, *next;
593 unsigned int curr_off = 0;
b3b94faa
DT
594
595 BUG_ON(!PageLocked(page));
596 if (!page_has_buffers(page))
8628de05 597 return;
b3b94faa
DT
598
599 bh = head = page_buffers(page);
600 do {
601 unsigned int next_off = curr_off + bh->b_size;
602 next = bh->b_this_page;
603
604 if (offset <= curr_off)
605 discard_buffer(sdp, bh);
606
607 curr_off = next_off;
608 bh = next;
609 } while (bh != head);
610
611 if (!offset)
8628de05 612 try_to_release_page(page, 0);
b3b94faa 613
8628de05 614 return;
b3b94faa
DT
615}
616
a9e5f4d0
SW
617static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
618 const struct iovec *iov, loff_t offset,
619 unsigned long nr_segs)
d1665e41
SW
620{
621 struct file *file = iocb->ki_filp;
622 struct inode *inode = file->f_mapping->host;
feaa7bba 623 struct gfs2_inode *ip = GFS2_I(inode);
d1665e41
SW
624 struct gfs2_holder gh;
625 int rv;
626
a9e5f4d0
SW
627 if (rw == READ)
628 mutex_lock(&inode->i_mutex);
d1665e41 629 /*
a9e5f4d0 630 * Shared lock, even if its a write, since we do no allocation
d1665e41
SW
631 * on this path. All we need change is atime.
632 */
633 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh);
634 rv = gfs2_glock_nq_m_atime(1, &gh);
635 if (rv)
636 goto out;
637
a9e5f4d0
SW
638 if (offset > i_size_read(inode))
639 goto out;
640
d1665e41
SW
641 /*
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.
646 */
647 if (gfs2_is_jdata(ip))
648 goto out;
649
650 if (gfs2_is_stuffed(ip))
651 goto out;
652
a9e5f4d0
SW
653 rv = blockdev_direct_IO_own_locking(rw, iocb, inode,
654 inode->i_sb->s_bdev,
655 iov, offset, nr_segs,
7a6bbacb 656 gfs2_get_block_direct, NULL);
d1665e41
SW
657out:
658 gfs2_glock_dq_m(1, &gh);
659 gfs2_holder_uninit(&gh);
a9e5f4d0
SW
660 if (rw == READ)
661 mutex_unlock(&inode->i_mutex);
d1665e41
SW
662
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
747 if (time_after_eq(jiffies, t)) {
748 stuck_releasepage(bh);
749 /* should we withdraw here? */
750 return 0;
4340fe62
SW
751 }
752
166afccd 753 yield();
4340fe62
SW
754 }
755
756 gfs2_assert_warn(sdp, !buffer_pinned(bh));
623d9355 757 gfs2_assert_warn(sdp, !buffer_dirty(bh));
4340fe62 758
623d9355 759 gfs2_log_lock(sdp);
4340fe62
SW
760 bd = bh->b_private;
761 if (bd) {
762 gfs2_assert_warn(sdp, bd->bd_bh == bh);
763 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
4340fe62 764 gfs2_assert_warn(sdp, !bd->bd_ail);
623d9355
SW
765 bd->bd_bh = NULL;
766 if (!list_empty(&bd->bd_le.le_list))
767 bd = NULL;
4340fe62
SW
768 bh->b_private = NULL;
769 }
623d9355
SW
770 gfs2_log_unlock(sdp);
771 if (bd)
772 kmem_cache_free(gfs2_bufdata_cachep, bd);
4340fe62
SW
773
774 bh = bh->b_this_page;
166afccd 775 } while (bh != head);
4340fe62 776
a9e5f4d0 777out:
4340fe62
SW
778 return try_to_free_buffers(page);
779}
780
66de045d 781const struct address_space_operations gfs2_file_aops = {
b3b94faa
DT
782 .writepage = gfs2_writepage,
783 .readpage = gfs2_readpage,
fd88de56 784 .readpages = gfs2_readpages,
b3b94faa
DT
785 .sync_page = block_sync_page,
786 .prepare_write = gfs2_prepare_write,
787 .commit_write = gfs2_commit_write,
788 .bmap = gfs2_bmap,
789 .invalidatepage = gfs2_invalidatepage,
4340fe62 790 .releasepage = gfs2_releasepage,
b3b94faa
DT
791 .direct_IO = gfs2_direct_IO,
792};
793