]>
Commit | Line | Data |
---|---|---|
b3b94faa DT |
1 | /* |
2 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
7eabb77e | 3 | * Copyright (C) 2004-2008 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> |
7765ec26 | 20 | #include <linux/swap.h> |
5c676f6d | 21 | #include <linux/gfs2_ondisk.h> |
7d308590 | 22 | #include <linux/lm_interface.h> |
47e83b50 | 23 | #include <linux/backing-dev.h> |
b8e7cbb6 | 24 | #include <linux/pagevec.h> |
b3b94faa DT |
25 | |
26 | #include "gfs2.h" | |
5c676f6d | 27 | #include "incore.h" |
b3b94faa DT |
28 | #include "bmap.h" |
29 | #include "glock.h" | |
30 | #include "inode.h" | |
b3b94faa DT |
31 | #include "log.h" |
32 | #include "meta_io.h" | |
33 | #include "ops_address.h" | |
b3b94faa DT |
34 | #include "quota.h" |
35 | #include "trans.h" | |
18ec7d5c | 36 | #include "rgrp.h" |
cd81a4ba | 37 | #include "super.h" |
5c676f6d | 38 | #include "util.h" |
4340fe62 | 39 | #include "glops.h" |
b3b94faa | 40 | |
ba7f7290 SW |
41 | |
42 | static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, | |
43 | unsigned int from, unsigned int to) | |
44 | { | |
45 | struct buffer_head *head = page_buffers(page); | |
46 | unsigned int bsize = head->b_size; | |
47 | struct buffer_head *bh; | |
48 | unsigned int start, end; | |
49 | ||
50 | for (bh = head, start = 0; bh != head || !start; | |
51 | bh = bh->b_this_page, start = end) { | |
52 | end = start + bsize; | |
53 | if (end <= from || start >= to) | |
54 | continue; | |
ddf4b426 BM |
55 | if (gfs2_is_jdata(ip)) |
56 | set_buffer_uptodate(bh); | |
ba7f7290 SW |
57 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
58 | } | |
59 | } | |
60 | ||
b3b94faa | 61 | /** |
7a6bbacb | 62 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
63 | * @inode: The inode |
64 | * @lblock: The block number to look up | |
65 | * @bh_result: The buffer head to return the result in | |
66 | * @create: Non-zero if we may add block to the file | |
67 | * | |
68 | * Returns: errno | |
69 | */ | |
70 | ||
7a6bbacb SW |
71 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
72 | struct buffer_head *bh_result, int create) | |
b3b94faa | 73 | { |
b3b94faa DT |
74 | int error; |
75 | ||
e9e1ef2b | 76 | error = gfs2_block_map(inode, lblock, bh_result, 0); |
b3b94faa DT |
77 | if (error) |
78 | return error; | |
de986e85 | 79 | if (!buffer_mapped(bh_result)) |
7a6bbacb SW |
80 | return -EIO; |
81 | return 0; | |
b3b94faa DT |
82 | } |
83 | ||
7a6bbacb SW |
84 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
85 | struct buffer_head *bh_result, int create) | |
623d9355 | 86 | { |
e9e1ef2b | 87 | return gfs2_block_map(inode, lblock, bh_result, 0); |
623d9355 | 88 | } |
7a6bbacb | 89 | |
b3b94faa | 90 | /** |
9ff8ec32 SW |
91 | * gfs2_writepage_common - Common bits of writepage |
92 | * @page: The page to be written | |
93 | * @wbc: The writeback control | |
b3b94faa | 94 | * |
9ff8ec32 | 95 | * Returns: 1 if writepage is ok, otherwise an error code or zero if no error. |
b3b94faa DT |
96 | */ |
97 | ||
9ff8ec32 SW |
98 | static int gfs2_writepage_common(struct page *page, |
99 | struct writeback_control *wbc) | |
b3b94faa | 100 | { |
18ec7d5c | 101 | struct inode *inode = page->mapping->host; |
f4387149 SW |
102 | struct gfs2_inode *ip = GFS2_I(inode); |
103 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
104 | loff_t i_size = i_size_read(inode); |
105 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
106 | unsigned offset; | |
b3b94faa | 107 | |
9ff8ec32 SW |
108 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
109 | goto out; | |
5c676f6d | 110 | if (current->journal_info) |
9ff8ec32 | 111 | goto redirty; |
18ec7d5c | 112 | /* Is the page fully outside i_size? (truncate in progress) */ |
9ff8ec32 | 113 | offset = i_size & (PAGE_CACHE_SIZE-1); |
d2d7b8a2 | 114 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 115 | page->mapping->a_ops->invalidatepage(page, 0); |
9ff8ec32 | 116 | goto out; |
b3b94faa | 117 | } |
9ff8ec32 SW |
118 | return 1; |
119 | redirty: | |
120 | redirty_page_for_writepage(wbc, page); | |
121 | out: | |
122 | unlock_page(page); | |
123 | return 0; | |
124 | } | |
125 | ||
126 | /** | |
127 | * gfs2_writeback_writepage - Write page for writeback mappings | |
128 | * @page: The page | |
129 | * @wbc: The writeback control | |
130 | * | |
131 | */ | |
132 | ||
133 | static int gfs2_writeback_writepage(struct page *page, | |
134 | struct writeback_control *wbc) | |
135 | { | |
136 | int ret; | |
137 | ||
138 | ret = gfs2_writepage_common(page, wbc); | |
139 | if (ret <= 0) | |
140 | return ret; | |
141 | ||
142 | ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc); | |
143 | if (ret == -EAGAIN) | |
144 | ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
145 | return ret; | |
146 | } | |
147 | ||
148 | /** | |
149 | * gfs2_ordered_writepage - Write page for ordered data files | |
150 | * @page: The page to write | |
151 | * @wbc: The writeback control | |
152 | * | |
153 | */ | |
154 | ||
155 | static int gfs2_ordered_writepage(struct page *page, | |
156 | struct writeback_control *wbc) | |
157 | { | |
158 | struct inode *inode = page->mapping->host; | |
159 | struct gfs2_inode *ip = GFS2_I(inode); | |
160 | int ret; | |
161 | ||
162 | ret = gfs2_writepage_common(page, wbc); | |
163 | if (ret <= 0) | |
164 | return ret; | |
165 | ||
166 | if (!page_has_buffers(page)) { | |
167 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
168 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
169 | } | |
170 | gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1); | |
171 | return block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
172 | } | |
173 | ||
b8e7cbb6 SW |
174 | /** |
175 | * __gfs2_jdata_writepage - The core of jdata writepage | |
176 | * @page: The page to write | |
177 | * @wbc: The writeback control | |
178 | * | |
179 | * This is shared between writepage and writepages and implements the | |
180 | * core of the writepage operation. If a transaction is required then | |
181 | * PageChecked will have been set and the transaction will have | |
182 | * already been started before this is called. | |
183 | */ | |
184 | ||
185 | static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
186 | { | |
187 | struct inode *inode = page->mapping->host; | |
188 | struct gfs2_inode *ip = GFS2_I(inode); | |
189 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
190 | ||
191 | if (PageChecked(page)) { | |
192 | ClearPageChecked(page); | |
193 | if (!page_has_buffers(page)) { | |
194 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
195 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
196 | } | |
197 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); | |
198 | } | |
199 | return block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
200 | } | |
201 | ||
9ff8ec32 SW |
202 | /** |
203 | * gfs2_jdata_writepage - Write complete page | |
204 | * @page: Page to write | |
205 | * | |
206 | * Returns: errno | |
207 | * | |
208 | */ | |
209 | ||
210 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
211 | { | |
212 | struct inode *inode = page->mapping->host; | |
9ff8ec32 SW |
213 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
214 | int error; | |
215 | int done_trans = 0; | |
216 | ||
217 | error = gfs2_writepage_common(page, wbc); | |
218 | if (error <= 0) | |
219 | return error; | |
b3b94faa | 220 | |
bf36a713 | 221 | if (PageChecked(page)) { |
b8e7cbb6 SW |
222 | if (wbc->sync_mode != WB_SYNC_ALL) |
223 | goto out_ignore; | |
18ec7d5c SW |
224 | error = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); |
225 | if (error) | |
226 | goto out_ignore; | |
18ec7d5c SW |
227 | done_trans = 1; |
228 | } | |
b8e7cbb6 | 229 | error = __gfs2_jdata_writepage(page, wbc); |
18ec7d5c SW |
230 | if (done_trans) |
231 | gfs2_trans_end(sdp); | |
b3b94faa | 232 | return error; |
18ec7d5c SW |
233 | |
234 | out_ignore: | |
235 | redirty_page_for_writepage(wbc, page); | |
236 | unlock_page(page); | |
237 | return 0; | |
b3b94faa DT |
238 | } |
239 | ||
a8d638e3 | 240 | /** |
5561093e | 241 | * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
242 | * @mapping: The mapping to write |
243 | * @wbc: Write-back control | |
244 | * | |
5561093e | 245 | * For the data=writeback case we can already ignore buffer heads |
a8d638e3 SW |
246 | * and write whole extents at once. This is a big reduction in the |
247 | * number of I/O requests we send and the bmap calls we make in this case. | |
248 | */ | |
5561093e SW |
249 | static int gfs2_writeback_writepages(struct address_space *mapping, |
250 | struct writeback_control *wbc) | |
a8d638e3 | 251 | { |
5561093e | 252 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); |
a8d638e3 SW |
253 | } |
254 | ||
b8e7cbb6 SW |
255 | /** |
256 | * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages | |
257 | * @mapping: The mapping | |
258 | * @wbc: The writeback control | |
259 | * @writepage: The writepage function to call for each page | |
260 | * @pvec: The vector of pages | |
261 | * @nr_pages: The number of pages to write | |
262 | * | |
263 | * Returns: non-zero if loop should terminate, zero otherwise | |
264 | */ | |
265 | ||
266 | static int gfs2_write_jdata_pagevec(struct address_space *mapping, | |
267 | struct writeback_control *wbc, | |
268 | struct pagevec *pvec, | |
269 | int nr_pages, pgoff_t end) | |
270 | { | |
271 | struct inode *inode = mapping->host; | |
272 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
273 | loff_t i_size = i_size_read(inode); | |
274 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
275 | unsigned offset = i_size & (PAGE_CACHE_SIZE-1); | |
276 | unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize); | |
277 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
278 | int i; | |
279 | int ret; | |
280 | ||
281 | ret = gfs2_trans_begin(sdp, nrblocks, 0); | |
282 | if (ret < 0) | |
283 | return ret; | |
284 | ||
285 | for(i = 0; i < nr_pages; i++) { | |
286 | struct page *page = pvec->pages[i]; | |
287 | ||
288 | lock_page(page); | |
289 | ||
290 | if (unlikely(page->mapping != mapping)) { | |
291 | unlock_page(page); | |
292 | continue; | |
293 | } | |
294 | ||
295 | if (!wbc->range_cyclic && page->index > end) { | |
296 | ret = 1; | |
297 | unlock_page(page); | |
298 | continue; | |
299 | } | |
300 | ||
301 | if (wbc->sync_mode != WB_SYNC_NONE) | |
302 | wait_on_page_writeback(page); | |
303 | ||
304 | if (PageWriteback(page) || | |
305 | !clear_page_dirty_for_io(page)) { | |
306 | unlock_page(page); | |
307 | continue; | |
308 | } | |
309 | ||
310 | /* Is the page fully outside i_size? (truncate in progress) */ | |
311 | if (page->index > end_index || (page->index == end_index && !offset)) { | |
312 | page->mapping->a_ops->invalidatepage(page, 0); | |
313 | unlock_page(page); | |
314 | continue; | |
315 | } | |
316 | ||
317 | ret = __gfs2_jdata_writepage(page, wbc); | |
318 | ||
319 | if (ret || (--(wbc->nr_to_write) <= 0)) | |
320 | ret = 1; | |
321 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
322 | wbc->encountered_congestion = 1; | |
323 | ret = 1; | |
324 | } | |
325 | ||
326 | } | |
327 | gfs2_trans_end(sdp); | |
328 | return ret; | |
329 | } | |
330 | ||
331 | /** | |
332 | * gfs2_write_cache_jdata - Like write_cache_pages but different | |
333 | * @mapping: The mapping to write | |
334 | * @wbc: The writeback control | |
335 | * @writepage: The writepage function to call | |
336 | * @data: The data to pass to writepage | |
337 | * | |
338 | * The reason that we use our own function here is that we need to | |
339 | * start transactions before we grab page locks. This allows us | |
340 | * to get the ordering right. | |
341 | */ | |
342 | ||
343 | static int gfs2_write_cache_jdata(struct address_space *mapping, | |
344 | struct writeback_control *wbc) | |
345 | { | |
346 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
347 | int ret = 0; | |
348 | int done = 0; | |
349 | struct pagevec pvec; | |
350 | int nr_pages; | |
351 | pgoff_t index; | |
352 | pgoff_t end; | |
353 | int scanned = 0; | |
354 | int range_whole = 0; | |
355 | ||
356 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
357 | wbc->encountered_congestion = 1; | |
358 | return 0; | |
359 | } | |
360 | ||
361 | pagevec_init(&pvec, 0); | |
362 | if (wbc->range_cyclic) { | |
363 | index = mapping->writeback_index; /* Start from prev offset */ | |
364 | end = -1; | |
365 | } else { | |
366 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
367 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
368 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
369 | range_whole = 1; | |
370 | scanned = 1; | |
371 | } | |
372 | ||
373 | retry: | |
374 | while (!done && (index <= end) && | |
375 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
376 | PAGECACHE_TAG_DIRTY, | |
377 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
378 | scanned = 1; | |
379 | ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end); | |
380 | if (ret) | |
381 | done = 1; | |
382 | if (ret > 0) | |
383 | ret = 0; | |
384 | ||
385 | pagevec_release(&pvec); | |
386 | cond_resched(); | |
387 | } | |
388 | ||
389 | if (!scanned && !done) { | |
390 | /* | |
391 | * We hit the last page and there is more work to be done: wrap | |
392 | * back to the start of the file | |
393 | */ | |
394 | scanned = 1; | |
395 | index = 0; | |
396 | goto retry; | |
397 | } | |
398 | ||
399 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
400 | mapping->writeback_index = index; | |
401 | return ret; | |
402 | } | |
403 | ||
404 | ||
405 | /** | |
406 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk | |
407 | * @mapping: The mapping to write | |
408 | * @wbc: The writeback control | |
409 | * | |
410 | */ | |
411 | ||
412 | static int gfs2_jdata_writepages(struct address_space *mapping, | |
413 | struct writeback_control *wbc) | |
414 | { | |
415 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
416 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
417 | int ret; | |
418 | ||
419 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
420 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { | |
421 | gfs2_log_flush(sdp, ip->i_gl); | |
422 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
423 | } | |
424 | return ret; | |
425 | } | |
426 | ||
b3b94faa DT |
427 | /** |
428 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
429 | * @ip: the inode | |
430 | * @page: the page | |
431 | * | |
432 | * Returns: errno | |
433 | */ | |
434 | ||
435 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
436 | { | |
437 | struct buffer_head *dibh; | |
438 | void *kaddr; | |
439 | int error; | |
440 | ||
bf126aee SW |
441 | /* |
442 | * Due to the order of unstuffing files and ->nopage(), we can be | |
443 | * asked for a zero page in the case of a stuffed file being extended, | |
444 | * so we need to supply one here. It doesn't happen often. | |
445 | */ | |
446 | if (unlikely(page->index)) { | |
eebd2aa3 | 447 | zero_user(page, 0, PAGE_CACHE_SIZE); |
bf126aee SW |
448 | return 0; |
449 | } | |
fd88de56 | 450 | |
b3b94faa DT |
451 | error = gfs2_meta_inode_buffer(ip, &dibh); |
452 | if (error) | |
453 | return error; | |
454 | ||
5c4e9e03 | 455 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 456 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
b3b94faa | 457 | ip->i_di.di_size); |
fd88de56 | 458 | memset(kaddr + ip->i_di.di_size, 0, PAGE_CACHE_SIZE - ip->i_di.di_size); |
c312c4fd | 459 | kunmap_atomic(kaddr, KM_USER0); |
bf126aee | 460 | flush_dcache_page(page); |
b3b94faa | 461 | brelse(dibh); |
b3b94faa DT |
462 | SetPageUptodate(page); |
463 | ||
464 | return 0; | |
465 | } | |
466 | ||
b3b94faa | 467 | |
b3b94faa | 468 | /** |
51ff87bd SW |
469 | * __gfs2_readpage - readpage |
470 | * @file: The file to read a page for | |
b3b94faa DT |
471 | * @page: The page to read |
472 | * | |
51ff87bd SW |
473 | * This is the core of gfs2's readpage. Its used by the internal file |
474 | * reading code as in that case we already hold the glock. Also its | |
475 | * called by gfs2_readpage() once the required lock has been granted. | |
476 | * | |
b3b94faa DT |
477 | */ |
478 | ||
51ff87bd | 479 | static int __gfs2_readpage(void *file, struct page *page) |
b3b94faa | 480 | { |
feaa7bba SW |
481 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
482 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
b3b94faa DT |
483 | int error; |
484 | ||
18ec7d5c | 485 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
486 | error = stuffed_readpage(ip, page); |
487 | unlock_page(page); | |
51ff87bd | 488 | } else { |
e9e1ef2b | 489 | error = mpage_readpage(page, gfs2_block_map); |
51ff87bd | 490 | } |
b3b94faa DT |
491 | |
492 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
51ff87bd | 493 | return -EIO; |
b3b94faa | 494 | |
51ff87bd SW |
495 | return error; |
496 | } | |
497 | ||
498 | /** | |
499 | * gfs2_readpage - read a page of a file | |
500 | * @file: The file to read | |
501 | * @page: The page of the file | |
502 | * | |
3cc3f710 | 503 | * This deals with the locking required. We use a trylock in order to |
51ff87bd SW |
504 | * avoid the page lock / glock ordering problems returning AOP_TRUNCATED_PAGE |
505 | * in the event that we are unable to get the lock. | |
506 | */ | |
507 | ||
508 | static int gfs2_readpage(struct file *file, struct page *page) | |
509 | { | |
510 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); | |
7afd88d9 | 511 | struct gfs2_holder *gh; |
51ff87bd SW |
512 | int error; |
513 | ||
7afd88d9 SW |
514 | gh = gfs2_glock_is_locked_by_me(ip->i_gl); |
515 | if (!gh) { | |
516 | gh = kmalloc(sizeof(struct gfs2_holder), GFP_NOFS); | |
517 | if (!gh) | |
518 | return -ENOBUFS; | |
519 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, gh); | |
51ff87bd | 520 | unlock_page(page); |
7afd88d9 SW |
521 | error = gfs2_glock_nq_atime(gh); |
522 | if (likely(error != 0)) | |
523 | goto out; | |
524 | return AOP_TRUNCATED_PAGE; | |
61a30dcb | 525 | } |
51ff87bd | 526 | error = __gfs2_readpage(file, page); |
7afd88d9 | 527 | gfs2_glock_dq(gh); |
18ec7d5c | 528 | out: |
7afd88d9 SW |
529 | gfs2_holder_uninit(gh); |
530 | kfree(gh); | |
51ff87bd SW |
531 | return error; |
532 | } | |
533 | ||
534 | /** | |
535 | * gfs2_internal_read - read an internal file | |
536 | * @ip: The gfs2 inode | |
537 | * @ra_state: The readahead state (or NULL for no readahead) | |
538 | * @buf: The buffer to fill | |
539 | * @pos: The file position | |
540 | * @size: The amount to read | |
541 | * | |
542 | */ | |
543 | ||
544 | int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state, | |
545 | char *buf, loff_t *pos, unsigned size) | |
546 | { | |
547 | struct address_space *mapping = ip->i_inode.i_mapping; | |
548 | unsigned long index = *pos / PAGE_CACHE_SIZE; | |
549 | unsigned offset = *pos & (PAGE_CACHE_SIZE - 1); | |
550 | unsigned copied = 0; | |
551 | unsigned amt; | |
552 | struct page *page; | |
553 | void *p; | |
554 | ||
555 | do { | |
556 | amt = size - copied; | |
557 | if (offset + size > PAGE_CACHE_SIZE) | |
558 | amt = PAGE_CACHE_SIZE - offset; | |
559 | page = read_cache_page(mapping, index, __gfs2_readpage, NULL); | |
560 | if (IS_ERR(page)) | |
561 | return PTR_ERR(page); | |
562 | p = kmap_atomic(page, KM_USER0); | |
563 | memcpy(buf + copied, p + offset, amt); | |
564 | kunmap_atomic(p, KM_USER0); | |
565 | mark_page_accessed(page); | |
566 | page_cache_release(page); | |
567 | copied += amt; | |
568 | index++; | |
569 | offset = 0; | |
570 | } while(copied < size); | |
571 | (*pos) += size; | |
572 | return size; | |
fd88de56 SW |
573 | } |
574 | ||
fd88de56 SW |
575 | /** |
576 | * gfs2_readpages - Read a bunch of pages at once | |
577 | * | |
578 | * Some notes: | |
579 | * 1. This is only for readahead, so we can simply ignore any things | |
580 | * which are slightly inconvenient (such as locking conflicts between | |
581 | * the page lock and the glock) and return having done no I/O. Its | |
582 | * obviously not something we'd want to do on too regular a basis. | |
583 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 584 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 | 585 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
e9e1ef2b | 586 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
fd88de56 | 587 | */ |
3cc3f710 | 588 | |
fd88de56 SW |
589 | static int gfs2_readpages(struct file *file, struct address_space *mapping, |
590 | struct list_head *pages, unsigned nr_pages) | |
591 | { | |
592 | struct inode *inode = mapping->host; | |
feaa7bba SW |
593 | struct gfs2_inode *ip = GFS2_I(inode); |
594 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 595 | struct gfs2_holder gh; |
3cc3f710 | 596 | int ret; |
fd88de56 | 597 | |
3cc3f710 | 598 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, GL_ATIME, &gh); |
51ff87bd | 599 | ret = gfs2_glock_nq_atime(&gh); |
51ff87bd | 600 | if (unlikely(ret)) |
3cc3f710 | 601 | goto out_uninit; |
e1d5b18a | 602 | if (!gfs2_is_stuffed(ip)) |
e9e1ef2b | 603 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map); |
3cc3f710 SW |
604 | gfs2_glock_dq(&gh); |
605 | out_uninit: | |
606 | gfs2_holder_uninit(&gh); | |
fd88de56 SW |
607 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
608 | ret = -EIO; | |
609 | return ret; | |
b3b94faa DT |
610 | } |
611 | ||
612 | /** | |
7765ec26 | 613 | * gfs2_write_begin - Begin to write to a file |
b3b94faa | 614 | * @file: The file to write to |
7765ec26 SW |
615 | * @mapping: The mapping in which to write |
616 | * @pos: The file offset at which to start writing | |
617 | * @len: Length of the write | |
618 | * @flags: Various flags | |
619 | * @pagep: Pointer to return the page | |
620 | * @fsdata: Pointer to return fs data (unused by GFS2) | |
b3b94faa DT |
621 | * |
622 | * Returns: errno | |
623 | */ | |
624 | ||
7765ec26 SW |
625 | static int gfs2_write_begin(struct file *file, struct address_space *mapping, |
626 | loff_t pos, unsigned len, unsigned flags, | |
627 | struct page **pagep, void **fsdata) | |
b3b94faa | 628 | { |
7765ec26 SW |
629 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
630 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
18ec7d5c SW |
631 | unsigned int data_blocks, ind_blocks, rblocks; |
632 | int alloc_required; | |
b3b94faa | 633 | int error = 0; |
18ec7d5c | 634 | struct gfs2_alloc *al; |
7765ec26 SW |
635 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
636 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
637 | unsigned to = from + len; | |
638 | struct page *page; | |
52ae7b79 | 639 | |
7765ec26 | 640 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, GL_ATIME, &ip->i_gh); |
dcd24799 | 641 | error = gfs2_glock_nq_atime(&ip->i_gh); |
7765ec26 | 642 | if (unlikely(error)) |
18ec7d5c | 643 | goto out_uninit; |
b3b94faa | 644 | |
7765ec26 | 645 | gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks); |
7765ec26 | 646 | error = gfs2_write_alloc_required(ip, pos, len, &alloc_required); |
18ec7d5c | 647 | if (error) |
c41d4f09 | 648 | goto out_unlock; |
18ec7d5c SW |
649 | |
650 | if (alloc_required) { | |
651 | al = gfs2_alloc_get(ip); | |
182fe5ab CG |
652 | if (!al) { |
653 | error = -ENOMEM; | |
654 | goto out_unlock; | |
655 | } | |
18ec7d5c SW |
656 | |
657 | error = gfs2_quota_lock(ip, NO_QUOTA_CHANGE, NO_QUOTA_CHANGE); | |
658 | if (error) | |
659 | goto out_alloc_put; | |
660 | ||
2933f925 | 661 | error = gfs2_quota_check(ip, ip->i_inode.i_uid, ip->i_inode.i_gid); |
18ec7d5c SW |
662 | if (error) |
663 | goto out_qunlock; | |
664 | ||
665 | al->al_requested = data_blocks + ind_blocks; | |
666 | error = gfs2_inplace_reserve(ip); | |
667 | if (error) | |
668 | goto out_qunlock; | |
669 | } | |
670 | ||
671 | rblocks = RES_DINODE + ind_blocks; | |
672 | if (gfs2_is_jdata(ip)) | |
673 | rblocks += data_blocks ? data_blocks : 1; | |
674 | if (ind_blocks || data_blocks) | |
675 | rblocks += RES_STATFS + RES_QUOTA; | |
676 | ||
16615be1 SW |
677 | error = gfs2_trans_begin(sdp, rblocks, |
678 | PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); | |
18ec7d5c | 679 | if (error) |
a867bb28 | 680 | goto out_trans_fail; |
18ec7d5c | 681 | |
c41d4f09 SW |
682 | error = -ENOMEM; |
683 | page = __grab_cache_page(mapping, index); | |
684 | *pagep = page; | |
685 | if (unlikely(!page)) | |
686 | goto out_endtrans; | |
687 | ||
18ec7d5c | 688 | if (gfs2_is_stuffed(ip)) { |
c41d4f09 | 689 | error = 0; |
7765ec26 | 690 | if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { |
f25ef0c1 | 691 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
692 | if (error == 0) |
693 | goto prepare_write; | |
c41d4f09 | 694 | } else if (!PageUptodate(page)) { |
b3b94faa | 695 | error = stuffed_readpage(ip, page); |
c41d4f09 | 696 | } |
5c4e9e03 | 697 | goto out; |
18ec7d5c SW |
698 | } |
699 | ||
5c4e9e03 | 700 | prepare_write: |
e9e1ef2b | 701 | error = block_prepare_write(page, from, to, gfs2_block_map); |
18ec7d5c | 702 | out: |
c41d4f09 SW |
703 | if (error == 0) |
704 | return 0; | |
705 | ||
706 | page_cache_release(page); | |
707 | if (pos + len > ip->i_inode.i_size) | |
708 | vmtruncate(&ip->i_inode, ip->i_inode.i_size); | |
709 | out_endtrans: | |
710 | gfs2_trans_end(sdp); | |
a867bb28 | 711 | out_trans_fail: |
c41d4f09 SW |
712 | if (alloc_required) { |
713 | gfs2_inplace_release(ip); | |
18ec7d5c | 714 | out_qunlock: |
c41d4f09 | 715 | gfs2_quota_unlock(ip); |
18ec7d5c | 716 | out_alloc_put: |
c41d4f09 SW |
717 | gfs2_alloc_put(ip); |
718 | } | |
18ec7d5c | 719 | out_unlock: |
c41d4f09 | 720 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 721 | out_uninit: |
c41d4f09 | 722 | gfs2_holder_uninit(&ip->i_gh); |
b3b94faa DT |
723 | return error; |
724 | } | |
725 | ||
7ae8fa84 RP |
726 | /** |
727 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
728 | * @inode: the rindex inode | |
729 | */ | |
730 | static void adjust_fs_space(struct inode *inode) | |
731 | { | |
732 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
733 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; | |
734 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
735 | u64 fs_total, new_free; | |
736 | ||
737 | /* Total up the file system space, according to the latest rindex. */ | |
738 | fs_total = gfs2_ri_total(sdp); | |
739 | ||
740 | spin_lock(&sdp->sd_statfs_spin); | |
741 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) | |
742 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
743 | else | |
744 | new_free = 0; | |
745 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
746 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
747 | (unsigned long long)new_free); | |
7ae8fa84 RP |
748 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
749 | } | |
750 | ||
b3b94faa | 751 | /** |
7765ec26 SW |
752 | * gfs2_stuffed_write_end - Write end for stuffed files |
753 | * @inode: The inode | |
754 | * @dibh: The buffer_head containing the on-disk inode | |
755 | * @pos: The file position | |
756 | * @len: The length of the write | |
757 | * @copied: How much was actually copied by the VFS | |
758 | * @page: The page | |
759 | * | |
760 | * This copies the data from the page into the inode block after | |
761 | * the inode data structure itself. | |
762 | * | |
763 | * Returns: errno | |
764 | */ | |
765 | static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, | |
766 | loff_t pos, unsigned len, unsigned copied, | |
767 | struct page *page) | |
768 | { | |
769 | struct gfs2_inode *ip = GFS2_I(inode); | |
770 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
771 | u64 to = pos + copied; | |
772 | void *kaddr; | |
773 | unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode); | |
774 | struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data; | |
775 | ||
776 | BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode))); | |
777 | kaddr = kmap_atomic(page, KM_USER0); | |
778 | memcpy(buf + pos, kaddr + pos, copied); | |
779 | memset(kaddr + pos + copied, 0, len - copied); | |
780 | flush_dcache_page(page); | |
781 | kunmap_atomic(kaddr, KM_USER0); | |
782 | ||
783 | if (!PageUptodate(page)) | |
784 | SetPageUptodate(page); | |
785 | unlock_page(page); | |
786 | page_cache_release(page); | |
787 | ||
788 | if (inode->i_size < to) { | |
789 | i_size_write(inode, to); | |
790 | ip->i_di.di_size = inode->i_size; | |
791 | di->di_size = cpu_to_be64(inode->i_size); | |
792 | mark_inode_dirty(inode); | |
793 | } | |
794 | ||
795 | if (inode == sdp->sd_rindex) | |
796 | adjust_fs_space(inode); | |
797 | ||
798 | brelse(dibh); | |
799 | gfs2_trans_end(sdp); | |
800 | gfs2_glock_dq(&ip->i_gh); | |
801 | gfs2_holder_uninit(&ip->i_gh); | |
802 | return copied; | |
803 | } | |
804 | ||
805 | /** | |
806 | * gfs2_write_end | |
b3b94faa | 807 | * @file: The file to write to |
7765ec26 SW |
808 | * @mapping: The address space to write to |
809 | * @pos: The file position | |
810 | * @len: The length of the data | |
811 | * @copied: | |
812 | * @page: The page that has been written | |
813 | * @fsdata: The fsdata (unused in GFS2) | |
814 | * | |
815 | * The main write_end function for GFS2. We have a separate one for | |
816 | * stuffed files as they are slightly different, otherwise we just | |
817 | * put our locking around the VFS provided functions. | |
b3b94faa DT |
818 | * |
819 | * Returns: errno | |
820 | */ | |
821 | ||
7765ec26 SW |
822 | static int gfs2_write_end(struct file *file, struct address_space *mapping, |
823 | loff_t pos, unsigned len, unsigned copied, | |
824 | struct page *page, void *fsdata) | |
b3b94faa DT |
825 | { |
826 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
827 | struct gfs2_inode *ip = GFS2_I(inode); |
828 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c | 829 | struct buffer_head *dibh; |
6dbd8224 | 830 | struct gfs2_alloc *al = ip->i_alloc; |
48516ced | 831 | struct gfs2_dinode *di; |
7765ec26 SW |
832 | unsigned int from = pos & (PAGE_CACHE_SIZE - 1); |
833 | unsigned int to = from + len; | |
834 | int ret; | |
b3b94faa | 835 | |
7afd88d9 | 836 | BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL); |
18ec7d5c | 837 | |
7765ec26 SW |
838 | ret = gfs2_meta_inode_buffer(ip, &dibh); |
839 | if (unlikely(ret)) { | |
840 | unlock_page(page); | |
841 | page_cache_release(page); | |
842 | goto failed; | |
843 | } | |
18ec7d5c SW |
844 | |
845 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
b3b94faa | 846 | |
7765ec26 SW |
847 | if (gfs2_is_stuffed(ip)) |
848 | return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page); | |
b3b94faa | 849 | |
bf36a713 | 850 | if (!gfs2_is_writeback(ip)) |
7765ec26 | 851 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa | 852 | |
7765ec26 | 853 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
b3b94faa | 854 | |
9656b2c1 SW |
855 | if (likely(ret >= 0) && (inode->i_size > ip->i_di.di_size)) { |
856 | di = (struct gfs2_dinode *)dibh->b_data; | |
857 | ip->i_di.di_size = inode->i_size; | |
858 | di->di_size = cpu_to_be64(inode->i_size); | |
859 | mark_inode_dirty(inode); | |
48516ced SW |
860 | } |
861 | ||
7ae8fa84 RP |
862 | if (inode == sdp->sd_rindex) |
863 | adjust_fs_space(inode); | |
864 | ||
18ec7d5c SW |
865 | brelse(dibh); |
866 | gfs2_trans_end(sdp); | |
7765ec26 | 867 | failed: |
6dbd8224 | 868 | if (al) { |
18ec7d5c SW |
869 | gfs2_inplace_release(ip); |
870 | gfs2_quota_unlock(ip); | |
871 | gfs2_alloc_put(ip); | |
872 | } | |
7765ec26 | 873 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 874 | gfs2_holder_uninit(&ip->i_gh); |
7765ec26 | 875 | return ret; |
b3b94faa DT |
876 | } |
877 | ||
8fb68595 RP |
878 | /** |
879 | * gfs2_set_page_dirty - Page dirtying function | |
880 | * @page: The page to dirty | |
881 | * | |
882 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
883 | */ | |
884 | ||
885 | static int gfs2_set_page_dirty(struct page *page) | |
886 | { | |
5561093e | 887 | SetPageChecked(page); |
8fb68595 RP |
888 | return __set_page_dirty_buffers(page); |
889 | } | |
890 | ||
b3b94faa DT |
891 | /** |
892 | * gfs2_bmap - Block map function | |
893 | * @mapping: Address space info | |
894 | * @lblock: The block to map | |
895 | * | |
896 | * Returns: The disk address for the block or 0 on hole or error | |
897 | */ | |
898 | ||
899 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
900 | { | |
feaa7bba | 901 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
902 | struct gfs2_holder i_gh; |
903 | sector_t dblock = 0; | |
904 | int error; | |
905 | ||
b3b94faa DT |
906 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
907 | if (error) | |
908 | return 0; | |
909 | ||
910 | if (!gfs2_is_stuffed(ip)) | |
e9e1ef2b | 911 | dblock = generic_block_bmap(mapping, lblock, gfs2_block_map); |
b3b94faa DT |
912 | |
913 | gfs2_glock_dq_uninit(&i_gh); | |
914 | ||
915 | return dblock; | |
916 | } | |
917 | ||
d7b616e2 SW |
918 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
919 | { | |
920 | struct gfs2_bufdata *bd; | |
921 | ||
922 | lock_buffer(bh); | |
923 | gfs2_log_lock(sdp); | |
924 | clear_buffer_dirty(bh); | |
925 | bd = bh->b_private; | |
926 | if (bd) { | |
16615be1 SW |
927 | if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh)) |
928 | list_del_init(&bd->bd_le.le_list); | |
929 | else | |
930 | gfs2_remove_from_journal(bh, current->journal_info, 0); | |
d7b616e2 SW |
931 | } |
932 | bh->b_bdev = NULL; | |
933 | clear_buffer_mapped(bh); | |
934 | clear_buffer_req(bh); | |
935 | clear_buffer_new(bh); | |
936 | gfs2_log_unlock(sdp); | |
937 | unlock_buffer(bh); | |
938 | } | |
939 | ||
8628de05 | 940 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 941 | { |
d7b616e2 SW |
942 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
943 | struct buffer_head *bh, *head; | |
944 | unsigned long pos = 0; | |
945 | ||
b3b94faa | 946 | BUG_ON(!PageLocked(page)); |
8fb68595 RP |
947 | if (offset == 0) |
948 | ClearPageChecked(page); | |
d7b616e2 SW |
949 | if (!page_has_buffers(page)) |
950 | goto out; | |
b3b94faa | 951 | |
d7b616e2 SW |
952 | bh = head = page_buffers(page); |
953 | do { | |
954 | if (offset <= pos) | |
955 | gfs2_discard(sdp, bh); | |
956 | pos += bh->b_size; | |
957 | bh = bh->b_this_page; | |
958 | } while (bh != head); | |
959 | out: | |
960 | if (offset == 0) | |
961 | try_to_release_page(page, 0); | |
b3b94faa DT |
962 | } |
963 | ||
c7b33834 SW |
964 | /** |
965 | * gfs2_ok_for_dio - check that dio is valid on this file | |
966 | * @ip: The inode | |
967 | * @rw: READ or WRITE | |
968 | * @offset: The offset at which we are reading or writing | |
969 | * | |
970 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
971 | * 1 (to accept the i/o request) | |
972 | */ | |
973 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
974 | { | |
975 | /* | |
976 | * Should we return an error here? I can't see that O_DIRECT for | |
5561093e SW |
977 | * a stuffed file makes any sense. For now we'll silently fall |
978 | * back to buffered I/O | |
c7b33834 | 979 | */ |
c7b33834 SW |
980 | if (gfs2_is_stuffed(ip)) |
981 | return 0; | |
982 | ||
983 | if (offset > i_size_read(&ip->i_inode)) | |
984 | return 0; | |
985 | return 1; | |
986 | } | |
987 | ||
988 | ||
989 | ||
a9e5f4d0 SW |
990 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
991 | const struct iovec *iov, loff_t offset, | |
992 | unsigned long nr_segs) | |
d1665e41 SW |
993 | { |
994 | struct file *file = iocb->ki_filp; | |
995 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 996 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
997 | struct gfs2_holder gh; |
998 | int rv; | |
999 | ||
1000 | /* | |
c7b33834 SW |
1001 | * Deferred lock, even if its a write, since we do no allocation |
1002 | * on this path. All we need change is atime, and this lock mode | |
1003 | * ensures that other nodes have flushed their buffered read caches | |
1004 | * (i.e. their page cache entries for this inode). We do not, | |
1005 | * unfortunately have the option of only flushing a range like | |
1006 | * the VFS does. | |
d1665e41 | 1007 | */ |
c7b33834 | 1008 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, GL_ATIME, &gh); |
dcd24799 | 1009 | rv = gfs2_glock_nq_atime(&gh); |
d1665e41 | 1010 | if (rv) |
c7b33834 SW |
1011 | return rv; |
1012 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
1013 | if (rv != 1) | |
1014 | goto out; /* dio not valid, fall back to buffered i/o */ | |
1015 | ||
1016 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
1017 | iov, offset, nr_segs, | |
1018 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
1019 | out: |
1020 | gfs2_glock_dq_m(1, &gh); | |
1021 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
1022 | return rv; |
1023 | } | |
1024 | ||
4340fe62 | 1025 | /** |
623d9355 | 1026 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
1027 | * @page: the page that's being released |
1028 | * @gfp_mask: passed from Linux VFS, ignored by us | |
1029 | * | |
1030 | * Call try_to_free_buffers() if the buffers in this page can be | |
1031 | * released. | |
1032 | * | |
1033 | * Returns: 0 | |
1034 | */ | |
1035 | ||
1036 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
1037 | { | |
1038 | struct inode *aspace = page->mapping->host; | |
1039 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
1040 | struct buffer_head *bh, *head; | |
1041 | struct gfs2_bufdata *bd; | |
4340fe62 SW |
1042 | |
1043 | if (!page_has_buffers(page)) | |
891ba6d4 | 1044 | return 0; |
4340fe62 | 1045 | |
bb3b0e3d | 1046 | gfs2_log_lock(sdp); |
4340fe62 SW |
1047 | head = bh = page_buffers(page); |
1048 | do { | |
bb3b0e3d SW |
1049 | if (atomic_read(&bh->b_count)) |
1050 | goto cannot_release; | |
1051 | bd = bh->b_private; | |
1052 | if (bd && bd->bd_ail) | |
1053 | goto cannot_release; | |
4340fe62 | 1054 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); |
623d9355 | 1055 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
bb3b0e3d SW |
1056 | bh = bh->b_this_page; |
1057 | } while(bh != head); | |
1058 | gfs2_log_unlock(sdp); | |
4340fe62 | 1059 | |
bb3b0e3d SW |
1060 | head = bh = page_buffers(page); |
1061 | do { | |
623d9355 | 1062 | gfs2_log_lock(sdp); |
4340fe62 SW |
1063 | bd = bh->b_private; |
1064 | if (bd) { | |
1065 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
1066 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
d7b616e2 SW |
1067 | if (!list_empty(&bd->bd_le.le_list)) { |
1068 | if (!buffer_pinned(bh)) | |
1069 | list_del_init(&bd->bd_le.le_list); | |
1070 | else | |
1071 | bd = NULL; | |
1072 | } | |
1073 | if (bd) | |
1074 | bd->bd_bh = NULL; | |
4340fe62 SW |
1075 | bh->b_private = NULL; |
1076 | } | |
623d9355 SW |
1077 | gfs2_log_unlock(sdp); |
1078 | if (bd) | |
1079 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
1080 | |
1081 | bh = bh->b_this_page; | |
166afccd | 1082 | } while (bh != head); |
4340fe62 | 1083 | |
4340fe62 | 1084 | return try_to_free_buffers(page); |
bb3b0e3d SW |
1085 | cannot_release: |
1086 | gfs2_log_unlock(sdp); | |
1087 | return 0; | |
4340fe62 SW |
1088 | } |
1089 | ||
5561093e | 1090 | static const struct address_space_operations gfs2_writeback_aops = { |
9ff8ec32 | 1091 | .writepage = gfs2_writeback_writepage, |
5561093e SW |
1092 | .writepages = gfs2_writeback_writepages, |
1093 | .readpage = gfs2_readpage, | |
1094 | .readpages = gfs2_readpages, | |
1095 | .sync_page = block_sync_page, | |
1096 | .write_begin = gfs2_write_begin, | |
1097 | .write_end = gfs2_write_end, | |
1098 | .bmap = gfs2_bmap, | |
1099 | .invalidatepage = gfs2_invalidatepage, | |
1100 | .releasepage = gfs2_releasepage, | |
1101 | .direct_IO = gfs2_direct_IO, | |
e5d9dc27 | 1102 | .migratepage = buffer_migrate_page, |
5561093e SW |
1103 | }; |
1104 | ||
1105 | static const struct address_space_operations gfs2_ordered_aops = { | |
9ff8ec32 | 1106 | .writepage = gfs2_ordered_writepage, |
b3b94faa | 1107 | .readpage = gfs2_readpage, |
fd88de56 | 1108 | .readpages = gfs2_readpages, |
b3b94faa | 1109 | .sync_page = block_sync_page, |
7765ec26 SW |
1110 | .write_begin = gfs2_write_begin, |
1111 | .write_end = gfs2_write_end, | |
8fb68595 | 1112 | .set_page_dirty = gfs2_set_page_dirty, |
b3b94faa DT |
1113 | .bmap = gfs2_bmap, |
1114 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 1115 | .releasepage = gfs2_releasepage, |
b3b94faa | 1116 | .direct_IO = gfs2_direct_IO, |
e5d9dc27 | 1117 | .migratepage = buffer_migrate_page, |
b3b94faa DT |
1118 | }; |
1119 | ||
5561093e | 1120 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 1121 | .writepage = gfs2_jdata_writepage, |
b8e7cbb6 | 1122 | .writepages = gfs2_jdata_writepages, |
5561093e SW |
1123 | .readpage = gfs2_readpage, |
1124 | .readpages = gfs2_readpages, | |
1125 | .sync_page = block_sync_page, | |
1126 | .write_begin = gfs2_write_begin, | |
1127 | .write_end = gfs2_write_end, | |
1128 | .set_page_dirty = gfs2_set_page_dirty, | |
1129 | .bmap = gfs2_bmap, | |
1130 | .invalidatepage = gfs2_invalidatepage, | |
1131 | .releasepage = gfs2_releasepage, | |
1132 | }; | |
1133 | ||
1134 | void gfs2_set_aops(struct inode *inode) | |
1135 | { | |
1136 | struct gfs2_inode *ip = GFS2_I(inode); | |
1137 | ||
1138 | if (gfs2_is_writeback(ip)) | |
1139 | inode->i_mapping->a_ops = &gfs2_writeback_aops; | |
1140 | else if (gfs2_is_ordered(ip)) | |
1141 | inode->i_mapping->a_ops = &gfs2_ordered_aops; | |
1142 | else if (gfs2_is_jdata(ip)) | |
1143 | inode->i_mapping->a_ops = &gfs2_jdata_aops; | |
1144 | else | |
1145 | BUG(); | |
1146 | } | |
1147 |