]> git.proxmox.com Git - mirror_ubuntu-artful-kernel.git/blame - fs/gfs2/ops_address.c
[GFS2] use lib/parser for parsing mount options
[mirror_ubuntu-artful-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 */
a2cf8222
AB
173static int gfs2_writepages(struct address_space *mapping,
174 struct writeback_control *wbc)
a8d638e3
SW
175{
176 struct inode *inode = mapping->host;
177 struct gfs2_inode *ip = GFS2_I(inode);
178 struct gfs2_sbd *sdp = GFS2_SB(inode);
179
180 if (sdp->sd_args.ar_data == GFS2_DATA_WRITEBACK && !gfs2_is_jdata(ip))
181 return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc);
182
183 return generic_writepages(mapping, wbc);
184}
185
b3b94faa
DT
186/**
187 * stuffed_readpage - Fill in a Linux page with stuffed file data
188 * @ip: the inode
189 * @page: the page
190 *
191 * Returns: errno
192 */
193
194static int stuffed_readpage(struct gfs2_inode *ip, struct page *page)
195{
196 struct buffer_head *dibh;
197 void *kaddr;
198 int error;
199
61057c6b 200 BUG_ON(page->index);
fd88de56 201
b3b94faa
DT
202 error = gfs2_meta_inode_buffer(ip, &dibh);
203 if (error)
204 return error;
205
5c4e9e03 206 kaddr = kmap_atomic(page, KM_USER0);
fd88de56 207 memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode),
b3b94faa 208 ip->i_di.di_size);
fd88de56 209 memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size);
c312c4fd 210 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
211
212 brelse(dibh);
213
214 SetPageUptodate(page);
215
216 return 0;
217}
218
b3b94faa 219
b3b94faa
DT
220/**
221 * gfs2_readpage - readpage with locking
18ec7d5c
SW
222 * @file: The file to read a page for. N.B. This may be NULL if we are
223 * reading an internal file.
b3b94faa
DT
224 * @page: The page to read
225 *
226 * Returns: errno
227 */
228
229static int gfs2_readpage(struct file *file, struct page *page)
230{
feaa7bba
SW
231 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
232 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
dc41aeed 233 struct gfs2_file *gf = NULL;
18ec7d5c 234 struct gfs2_holder gh;
b3b94faa 235 int error;
59a1cc6b 236 int do_unlock = 0;
b3b94faa 237
dd538c83 238 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b 239 if (file) {
dc41aeed 240 gf = file->private_data;
59a1cc6b 241 if (test_bit(GFF_EXLOCK, &gf->f_flags))
dc41aeed 242 /* gfs2_sharewrite_nopage has grabbed the ip->i_gl already */
59a1cc6b
SW
243 goto skip_lock;
244 }
2ca99501 245 gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME|LM_FLAG_TRY_1CB, &gh);
59a1cc6b 246 do_unlock = 1;
dcd24799 247 error = gfs2_glock_nq_atime(&gh);
fd88de56 248 if (unlikely(error))
61a30dcb
SW
249 goto out_unlock;
250 }
b3b94faa 251
59a1cc6b 252skip_lock:
18ec7d5c 253 if (gfs2_is_stuffed(ip)) {
fd88de56
SW
254 error = stuffed_readpage(ip, page);
255 unlock_page(page);
b3b94faa 256 } else
18ec7d5c 257 error = mpage_readpage(page, gfs2_get_block);
b3b94faa
DT
258
259 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
260 error = -EIO;
261
07903c02 262 if (do_unlock) {
61a30dcb
SW
263 gfs2_glock_dq_m(1, &gh);
264 gfs2_holder_uninit(&gh);
265 }
18ec7d5c 266out:
b3b94faa 267 return error;
18ec7d5c
SW
268out_unlock:
269 unlock_page(page);
a13cbe37
JW
270 if (error == GLR_TRYFAILED) {
271 error = AOP_TRUNCATED_PAGE;
272 yield();
273 }
59a1cc6b 274 if (do_unlock)
fd88de56
SW
275 gfs2_holder_uninit(&gh);
276 goto out;
277}
278
fd88de56
SW
279/**
280 * gfs2_readpages - Read a bunch of pages at once
281 *
282 * Some notes:
283 * 1. This is only for readahead, so we can simply ignore any things
284 * which are slightly inconvenient (such as locking conflicts between
285 * the page lock and the glock) and return having done no I/O. Its
286 * obviously not something we'd want to do on too regular a basis.
287 * Any I/O we ignore at this time will be done via readpage later.
e1d5b18a 288 * 2. We don't handle stuffed files here we let readpage do the honours.
fd88de56
SW
289 * 3. mpage_readpages() does most of the heavy lifting in the common case.
290 * 4. gfs2_get_block() is relied upon to set BH_Boundary in the right places.
291 * 5. We use LM_FLAG_TRY_1CB here, effectively we then have lock-ahead as
292 * well as read-ahead.
293 */
294static int gfs2_readpages(struct file *file, struct address_space *mapping,
295 struct list_head *pages, unsigned nr_pages)
296{
297 struct inode *inode = mapping->host;
feaa7bba
SW
298 struct gfs2_inode *ip = GFS2_I(inode);
299 struct gfs2_sbd *sdp = GFS2_SB(inode);
fd88de56 300 struct gfs2_holder gh;
e1d5b18a 301 int ret = 0;
59a1cc6b 302 int do_unlock = 0;
fd88de56 303
dd538c83 304 if (likely(file != &gfs2_internal_file_sentinel)) {
59a1cc6b
SW
305 if (file) {
306 struct gfs2_file *gf = file->private_data;
307 if (test_bit(GFF_EXLOCK, &gf->f_flags))
308 goto skip_lock;
309 }
fd88de56 310 gfs2_holder_init(ip->i_gl, LM_ST_SHARED,
2ca99501 311 LM_FLAG_TRY_1CB|GL_ATIME, &gh);
59a1cc6b 312 do_unlock = 1;
dcd24799 313 ret = gfs2_glock_nq_atime(&gh);
907b9bce 314 if (ret == GLR_TRYFAILED)
fd88de56
SW
315 goto out_noerror;
316 if (unlikely(ret))
317 goto out_unlock;
318 }
59a1cc6b 319skip_lock:
e1d5b18a 320 if (!gfs2_is_stuffed(ip))
fd88de56 321 ret = mpage_readpages(mapping, pages, nr_pages, gfs2_get_block);
fd88de56 322
59a1cc6b 323 if (do_unlock) {
fd88de56
SW
324 gfs2_glock_dq_m(1, &gh);
325 gfs2_holder_uninit(&gh);
326 }
327out:
328 if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags)))
329 ret = -EIO;
330 return ret;
331out_noerror:
332 ret = 0;
333out_unlock:
59a1cc6b 334 if (do_unlock)
fd88de56 335 gfs2_holder_uninit(&gh);
18ec7d5c 336 goto out;
b3b94faa
DT
337}
338
339/**
340 * gfs2_prepare_write - Prepare to write a page to a file
341 * @file: The file to write to
342 * @page: The page which is to be prepared for writing
343 * @from: From (byte range within page)
344 * @to: To (byte range within page)
345 *
346 * Returns: errno
347 */
348
349static int gfs2_prepare_write(struct file *file, struct page *page,
350 unsigned from, unsigned to)
351{
feaa7bba
SW
352 struct gfs2_inode *ip = GFS2_I(page->mapping->host);
353 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
18ec7d5c
SW
354 unsigned int data_blocks, ind_blocks, rblocks;
355 int alloc_required;
b3b94faa 356 int error = 0;
18ec7d5c
SW
357 loff_t pos = ((loff_t)page->index << PAGE_CACHE_SHIFT) + from;
358 loff_t end = ((loff_t)page->index << PAGE_CACHE_SHIFT) + to;
359 struct gfs2_alloc *al;
52ae7b79
RC
360 unsigned int write_len = to - from;
361
b3b94faa 362
2ca99501 363 gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME|LM_FLAG_TRY_1CB, &ip->i_gh);
dcd24799 364 error = gfs2_glock_nq_atime(&ip->i_gh);
2ca99501 365 if (unlikely(error)) {
2d72e710
SW
366 if (error == GLR_TRYFAILED) {
367 unlock_page(page);
2ca99501 368 error = AOP_TRUNCATED_PAGE;
a13cbe37 369 yield();
2d72e710 370 }
18ec7d5c 371 goto out_uninit;
2ca99501 372 }
b3b94faa 373
52ae7b79 374 gfs2_write_calc_reserv(ip, write_len, &data_blocks, &ind_blocks);
18ec7d5c 375
52ae7b79 376 error = gfs2_write_alloc_required(ip, pos, write_len, &alloc_required);
18ec7d5c
SW
377 if (error)
378 goto out_unlock;
b3b94faa 379
18ec7d5c 380
f5c54804 381 ip->i_alloc.al_requested = 0;
18ec7d5c
SW
382 if (alloc_required) {
383 al = gfs2_alloc_get(ip);
384
385 error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE);
386 if (error)
387 goto out_alloc_put;
388
2933f925 389 error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid);
18ec7d5c
SW
390 if (error)
391 goto out_qunlock;
392
393 al->al_requested = data_blocks + ind_blocks;
394 error = gfs2_inplace_reserve(ip);
395 if (error)
396 goto out_qunlock;
397 }
398
399 rblocks = RES_DINODE + ind_blocks;
400 if (gfs2_is_jdata(ip))
401 rblocks += data_blocks ? data_blocks : 1;
402 if (ind_blocks || data_blocks)
403 rblocks += RES_STATFS + RES_QUOTA;
404
405 error = gfs2_trans_begin(sdp, rblocks, 0);
406 if (error)
407 goto out;
408
409 if (gfs2_is_stuffed(ip)) {
410 if (end > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) {
f25ef0c1 411 error = gfs2_unstuff_dinode(ip, page);
5c4e9e03
SW
412 if (error == 0)
413 goto prepare_write;
414 } else if (!PageUptodate(page))
b3b94faa 415 error = stuffed_readpage(ip, page);
5c4e9e03 416 goto out;
18ec7d5c
SW
417 }
418
5c4e9e03 419prepare_write:
18ec7d5c
SW
420 error = block_prepare_write(page, from, to, gfs2_get_block);
421
422out:
423 if (error) {
424 gfs2_trans_end(sdp);
425 if (alloc_required) {
426 gfs2_inplace_release(ip);
427out_qunlock:
428 gfs2_quota_unlock(ip);
429out_alloc_put:
430 gfs2_alloc_put(ip);
431 }
432out_unlock:
433 gfs2_glock_dq_m(1, &ip->i_gh);
434out_uninit:
435 gfs2_holder_uninit(&ip->i_gh);
436 }
b3b94faa
DT
437
438 return error;
439}
440
441/**
442 * gfs2_commit_write - Commit write to a file
443 * @file: The file to write to
444 * @page: The page containing the data
445 * @from: From (byte range within page)
446 * @to: To (byte range within page)
447 *
448 * Returns: errno
449 */
450
451static int gfs2_commit_write(struct file *file, struct page *page,
452 unsigned from, unsigned to)
453{
454 struct inode *inode = page->mapping->host;
feaa7bba
SW
455 struct gfs2_inode *ip = GFS2_I(inode);
456 struct gfs2_sbd *sdp = GFS2_SB(inode);
18ec7d5c
SW
457 int error = -EOPNOTSUPP;
458 struct buffer_head *dibh;
48516ced
SW
459 struct gfs2_alloc *al = &ip->i_alloc;
460 struct gfs2_dinode *di;
b3b94faa 461
18ec7d5c
SW
462 if (gfs2_assert_withdraw(sdp, gfs2_glock_is_locked_by_me(ip->i_gl)))
463 goto fail_nounlock;
464
465 error = gfs2_meta_inode_buffer(ip, &dibh);
466 if (error)
467 goto fail_endtrans;
468
469 gfs2_trans_add_bh(ip->i_gl, dibh, 1);
48516ced 470 di = (struct gfs2_dinode *)dibh->b_data;
18ec7d5c 471
b3b94faa 472 if (gfs2_is_stuffed(ip)) {
cd915493 473 u64 file_size;
b3b94faa
DT
474 void *kaddr;
475
cd915493 476 file_size = ((u64)page->index << PAGE_CACHE_SHIFT) + to;
b3b94faa 477
18ec7d5c 478 kaddr = kmap_atomic(page, KM_USER0);
b3b94faa 479 memcpy(dibh->b_data + sizeof(struct gfs2_dinode) + from,
0bd5996a 480 kaddr + from, to - from);
c312c4fd 481 kunmap_atomic(kaddr, KM_USER0);
b3b94faa
DT
482
483 SetPageUptodate(page);
484
ae619320 485 if (inode->i_size < file_size) {
b3b94faa 486 i_size_write(inode, file_size);
ae619320
SW
487 mark_inode_dirty(inode);
488 }
b3b94faa 489 } else {
568f4c96
SW
490 if (sdp->sd_args.ar_data == GFS2_DATA_ORDERED ||
491 gfs2_is_jdata(ip))
257f9b4e 492 gfs2_page_add_databufs(ip, page, from, to);
b3b94faa
DT
493 error = generic_commit_write(file, page, from, to);
494 if (error)
495 goto fail;
496 }
497
48516ced 498 if (ip->i_di.di_size < inode->i_size) {
18ec7d5c 499 ip->i_di.di_size = inode->i_size;
48516ced
SW
500 di->di_size = cpu_to_be64(inode->i_size);
501 }
502
18ec7d5c
SW
503 brelse(dibh);
504 gfs2_trans_end(sdp);
505 if (al->al_requested) {
506 gfs2_inplace_release(ip);
507 gfs2_quota_unlock(ip);
508 gfs2_alloc_put(ip);
509 }
1de91390 510 unlock_page(page);
18ec7d5c 511 gfs2_glock_dq_m(1, &ip->i_gh);
1de91390 512 lock_page(page);
18ec7d5c 513 gfs2_holder_uninit(&ip->i_gh);
b3b94faa
DT
514 return 0;
515
18ec7d5c
SW
516fail:
517 brelse(dibh);
518fail_endtrans:
519 gfs2_trans_end(sdp);
520 if (al->al_requested) {
521 gfs2_inplace_release(ip);
522 gfs2_quota_unlock(ip);
523 gfs2_alloc_put(ip);
524 }
1de91390 525 unlock_page(page);
18ec7d5c 526 gfs2_glock_dq_m(1, &ip->i_gh);
1de91390 527 lock_page(page);
18ec7d5c
SW
528 gfs2_holder_uninit(&ip->i_gh);
529fail_nounlock:
b3b94faa 530 ClearPageUptodate(page);
b3b94faa
DT
531 return error;
532}
533
534/**
535 * gfs2_bmap - Block map function
536 * @mapping: Address space info
537 * @lblock: The block to map
538 *
539 * Returns: The disk address for the block or 0 on hole or error
540 */
541
542static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock)
543{
feaa7bba 544 struct gfs2_inode *ip = GFS2_I(mapping->host);
b3b94faa
DT
545 struct gfs2_holder i_gh;
546 sector_t dblock = 0;
547 int error;
548
b3b94faa
DT
549 error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh);
550 if (error)
551 return 0;
552
553 if (!gfs2_is_stuffed(ip))
4ff14670 554 dblock = generic_block_bmap(mapping, lblock, gfs2_get_block);
b3b94faa
DT
555
556 gfs2_glock_dq_uninit(&i_gh);
557
558 return dblock;
559}
560
561static void discard_buffer(struct gfs2_sbd *sdp, struct buffer_head *bh)
562{
64fb4eb7 563 struct gfs2_bufdata *bd;
b3b94faa
DT
564
565 gfs2_log_lock(sdp);
5c676f6d 566 bd = bh->b_private;
64fb4eb7
SW
567 if (bd) {
568 bd->bd_bh = NULL;
5c676f6d 569 bh->b_private = NULL;
15d00c0b
SW
570 }
571 gfs2_log_unlock(sdp);
b3b94faa
DT
572
573 lock_buffer(bh);
574 clear_buffer_dirty(bh);
575 bh->b_bdev = NULL;
576 clear_buffer_mapped(bh);
577 clear_buffer_req(bh);
578 clear_buffer_new(bh);
579 clear_buffer_delay(bh);
580 unlock_buffer(bh);
581}
582
8628de05 583static void gfs2_invalidatepage(struct page *page, unsigned long offset)
b3b94faa 584{
15d00c0b 585 struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host);
b3b94faa
DT
586 struct buffer_head *head, *bh, *next;
587 unsigned int curr_off = 0;
b3b94faa
DT
588
589 BUG_ON(!PageLocked(page));
590 if (!page_has_buffers(page))
8628de05 591 return;
b3b94faa
DT
592
593 bh = head = page_buffers(page);
594 do {
595 unsigned int next_off = curr_off + bh->b_size;
596 next = bh->b_this_page;
597
598 if (offset <= curr_off)
599 discard_buffer(sdp, bh);
600
601 curr_off = next_off;
602 bh = next;
603 } while (bh != head);
604
605 if (!offset)
8628de05 606 try_to_release_page(page, 0);
b3b94faa 607
8628de05 608 return;
b3b94faa
DT
609}
610
c7b33834
SW
611/**
612 * gfs2_ok_for_dio - check that dio is valid on this file
613 * @ip: The inode
614 * @rw: READ or WRITE
615 * @offset: The offset at which we are reading or writing
616 *
617 * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o)
618 * 1 (to accept the i/o request)
619 */
620static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset)
621{
622 /*
623 * Should we return an error here? I can't see that O_DIRECT for
624 * a journaled file makes any sense. For now we'll silently fall
625 * back to buffered I/O, likewise we do the same for stuffed
626 * files since they are (a) small and (b) unaligned.
627 */
628 if (gfs2_is_jdata(ip))
629 return 0;
630
631 if (gfs2_is_stuffed(ip))
632 return 0;
633
634 if (offset > i_size_read(&ip->i_inode))
635 return 0;
636 return 1;
637}
638
639
640
a9e5f4d0
SW
641static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb,
642 const struct iovec *iov, loff_t offset,
643 unsigned long nr_segs)
d1665e41
SW
644{
645 struct file *file = iocb->ki_filp;
646 struct inode *inode = file->f_mapping->host;
feaa7bba 647 struct gfs2_inode *ip = GFS2_I(inode);
d1665e41
SW
648 struct gfs2_holder gh;
649 int rv;
650
651 /*
c7b33834
SW
652 * Deferred lock, even if its a write, since we do no allocation
653 * on this path. All we need change is atime, and this lock mode
654 * ensures that other nodes have flushed their buffered read caches
655 * (i.e. their page cache entries for this inode). We do not,
656 * unfortunately have the option of only flushing a range like
657 * the VFS does.
d1665e41 658 */
c7b33834 659 gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh);
dcd24799 660 rv = gfs2_glock_nq_atime(&gh);
d1665e41 661 if (rv)
c7b33834
SW
662 return rv;
663 rv = gfs2_ok_for_dio(ip, rw, offset);
664 if (rv != 1)
665 goto out; /* dio not valid, fall back to buffered i/o */
666
667 rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev,
668 iov, offset, nr_segs,
669 gfs2_get_block_direct, NULL);
d1665e41
SW
670out:
671 gfs2_glock_dq_m(1, &gh);
672 gfs2_holder_uninit(&gh);
d1665e41
SW
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;
166afccd
SW
688static unsigned limit = 0;
689
0bd5996a 690 if (limit > 3)
166afccd 691 return;
0bd5996a 692 limit++;
4340fe62
SW
693
694 fs_warn(sdp, "stuck in gfs2_releasepage() %p\n", inode);
695 fs_warn(sdp, "blkno = %llu, bh->b_count = %d\n",
696 (unsigned long long)bh->b_blocknr, atomic_read(&bh->b_count));
697 fs_warn(sdp, "pinned = %u\n", buffer_pinned(bh));
698 fs_warn(sdp, "bh->b_private = %s\n", (bd) ? "!NULL" : "NULL");
699
700 if (!bd)
701 return;
702
703 gl = bd->bd_gl;
704
907b9bce 705 fs_warn(sdp, "gl = (%u, %llu)\n",
4340fe62
SW
706 gl->gl_name.ln_type, (unsigned long long)gl->gl_name.ln_number);
707
708 fs_warn(sdp, "bd_list_tr = %s, bd_le.le_list = %s\n",
709 (list_empty(&bd->bd_list_tr)) ? "no" : "yes",
710 (list_empty(&bd->bd_le.le_list)) ? "no" : "yes");
711
712 if (gl->gl_ops == &gfs2_inode_glops) {
713 struct gfs2_inode *ip = gl->gl_object;
714 unsigned int x;
715
716 if (!ip)
717 return;
718
719 fs_warn(sdp, "ip = %llu %llu\n",
720 (unsigned long long)ip->i_num.no_formal_ino,
721 (unsigned long long)ip->i_num.no_addr);
722
723 for (x = 0; x < GFS2_MAX_META_HEIGHT; x++)
724 fs_warn(sdp, "ip->i_cache[%u] = %s\n",
725 x, (ip->i_cache[x]) ? "!NULL" : "NULL");
726 }
727}
728
729/**
623d9355 730 * gfs2_releasepage - free the metadata associated with a page
4340fe62
SW
731 * @page: the page that's being released
732 * @gfp_mask: passed from Linux VFS, ignored by us
733 *
734 * Call try_to_free_buffers() if the buffers in this page can be
735 * released.
736 *
737 * Returns: 0
738 */
739
740int gfs2_releasepage(struct page *page, gfp_t gfp_mask)
741{
742 struct inode *aspace = page->mapping->host;
743 struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info;
744 struct buffer_head *bh, *head;
745 struct gfs2_bufdata *bd;
166afccd 746 unsigned long t = jiffies + gfs2_tune_get(sdp, gt_stall_secs) * HZ;
4340fe62
SW
747
748 if (!page_has_buffers(page))
749 goto out;
750
751 head = bh = page_buffers(page);
752 do {
4340fe62 753 while (atomic_read(&bh->b_count)) {
166afccd
SW
754 if (!atomic_read(&aspace->i_writecount))
755 return 0;
756
61057c6b
RC
757 if (!(gfp_mask & __GFP_WAIT))
758 return 0;
759
166afccd
SW
760 if (time_after_eq(jiffies, t)) {
761 stuck_releasepage(bh);
762 /* should we withdraw here? */
763 return 0;
4340fe62
SW
764 }
765
166afccd 766 yield();
4340fe62
SW
767 }
768
769 gfs2_assert_warn(sdp, !buffer_pinned(bh));
623d9355 770 gfs2_assert_warn(sdp, !buffer_dirty(bh));
4340fe62 771
623d9355 772 gfs2_log_lock(sdp);
4340fe62
SW
773 bd = bh->b_private;
774 if (bd) {
775 gfs2_assert_warn(sdp, bd->bd_bh == bh);
776 gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr));
4340fe62 777 gfs2_assert_warn(sdp, !bd->bd_ail);
623d9355
SW
778 bd->bd_bh = NULL;
779 if (!list_empty(&bd->bd_le.le_list))
780 bd = NULL;
4340fe62
SW
781 bh->b_private = NULL;
782 }
623d9355
SW
783 gfs2_log_unlock(sdp);
784 if (bd)
785 kmem_cache_free(gfs2_bufdata_cachep, bd);
4340fe62
SW
786
787 bh = bh->b_this_page;
166afccd 788 } while (bh != head);
4340fe62 789
a9e5f4d0 790out:
4340fe62
SW
791 return try_to_free_buffers(page);
792}
793
66de045d 794const struct address_space_operations gfs2_file_aops = {
b3b94faa 795 .writepage = gfs2_writepage,
a8d638e3 796 .writepages = gfs2_writepages,
b3b94faa 797 .readpage = gfs2_readpage,
fd88de56 798 .readpages = gfs2_readpages,
b3b94faa
DT
799 .sync_page = block_sync_page,
800 .prepare_write = gfs2_prepare_write,
801 .commit_write = gfs2_commit_write,
802 .bmap = gfs2_bmap,
803 .invalidatepage = gfs2_invalidatepage,
4340fe62 804 .releasepage = gfs2_releasepage,
b3b94faa
DT
805 .direct_IO = gfs2_direct_IO,
806};
807