]>
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> |
47e83b50 | 22 | #include <linux/backing-dev.h> |
b3b94faa DT |
23 | |
24 | #include "gfs2.h" | |
5c676f6d | 25 | #include "incore.h" |
b3b94faa DT |
26 | #include "bmap.h" |
27 | #include "glock.h" | |
28 | #include "inode.h" | |
b3b94faa DT |
29 | #include "log.h" |
30 | #include "meta_io.h" | |
b3b94faa DT |
31 | #include "quota.h" |
32 | #include "trans.h" | |
18ec7d5c | 33 | #include "rgrp.h" |
cd81a4ba | 34 | #include "super.h" |
5c676f6d | 35 | #include "util.h" |
4340fe62 | 36 | #include "glops.h" |
b3b94faa | 37 | |
ba7f7290 SW |
38 | |
39 | static 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; | |
ddf4b426 BM |
52 | if (gfs2_is_jdata(ip)) |
53 | set_buffer_uptodate(bh); | |
ba7f7290 SW |
54 | gfs2_trans_add_bh(ip->i_gl, bh, 0); |
55 | } | |
56 | } | |
57 | ||
b3b94faa | 58 | /** |
7a6bbacb | 59 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
60 | * @inode: The inode |
61 | * @lblock: The block number to look up | |
62 | * @bh_result: The buffer head to return the result in | |
63 | * @create: Non-zero if we may add block to the file | |
64 | * | |
65 | * Returns: errno | |
66 | */ | |
67 | ||
7a6bbacb SW |
68 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
69 | struct buffer_head *bh_result, int create) | |
b3b94faa | 70 | { |
b3b94faa DT |
71 | int error; |
72 | ||
e9e1ef2b | 73 | error = gfs2_block_map(inode, lblock, bh_result, 0); |
b3b94faa DT |
74 | if (error) |
75 | return error; | |
de986e85 | 76 | if (!buffer_mapped(bh_result)) |
7a6bbacb SW |
77 | return -EIO; |
78 | return 0; | |
b3b94faa DT |
79 | } |
80 | ||
7a6bbacb SW |
81 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
82 | struct buffer_head *bh_result, int create) | |
623d9355 | 83 | { |
e9e1ef2b | 84 | return gfs2_block_map(inode, lblock, bh_result, 0); |
623d9355 | 85 | } |
7a6bbacb | 86 | |
b3b94faa | 87 | /** |
9ff8ec32 SW |
88 | * gfs2_writepage_common - Common bits of writepage |
89 | * @page: The page to be written | |
90 | * @wbc: The writeback control | |
b3b94faa | 91 | * |
9ff8ec32 | 92 | * Returns: 1 if writepage is ok, otherwise an error code or zero if no error. |
b3b94faa DT |
93 | */ |
94 | ||
9ff8ec32 SW |
95 | static int gfs2_writepage_common(struct page *page, |
96 | struct writeback_control *wbc) | |
b3b94faa | 97 | { |
18ec7d5c | 98 | struct inode *inode = page->mapping->host; |
f4387149 SW |
99 | struct gfs2_inode *ip = GFS2_I(inode); |
100 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
101 | loff_t i_size = i_size_read(inode); |
102 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
103 | unsigned offset; | |
b3b94faa | 104 | |
9ff8ec32 SW |
105 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
106 | goto out; | |
5c676f6d | 107 | if (current->journal_info) |
9ff8ec32 | 108 | goto redirty; |
18ec7d5c | 109 | /* Is the page fully outside i_size? (truncate in progress) */ |
9ff8ec32 | 110 | offset = i_size & (PAGE_CACHE_SIZE-1); |
d2d7b8a2 | 111 | if (page->index > end_index || (page->index == end_index && !offset)) { |
18ec7d5c | 112 | page->mapping->a_ops->invalidatepage(page, 0); |
9ff8ec32 | 113 | goto out; |
b3b94faa | 114 | } |
9ff8ec32 SW |
115 | return 1; |
116 | redirty: | |
117 | redirty_page_for_writepage(wbc, page); | |
118 | out: | |
119 | unlock_page(page); | |
120 | return 0; | |
121 | } | |
122 | ||
123 | /** | |
124 | * gfs2_writeback_writepage - Write page for writeback mappings | |
125 | * @page: The page | |
126 | * @wbc: The writeback control | |
127 | * | |
128 | */ | |
129 | ||
130 | static int gfs2_writeback_writepage(struct page *page, | |
131 | struct writeback_control *wbc) | |
132 | { | |
133 | int ret; | |
134 | ||
135 | ret = gfs2_writepage_common(page, wbc); | |
136 | if (ret <= 0) | |
137 | return ret; | |
138 | ||
139 | ret = mpage_writepage(page, gfs2_get_block_noalloc, wbc); | |
140 | if (ret == -EAGAIN) | |
141 | ret = block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
142 | return ret; | |
143 | } | |
144 | ||
145 | /** | |
146 | * gfs2_ordered_writepage - Write page for ordered data files | |
147 | * @page: The page to write | |
148 | * @wbc: The writeback control | |
149 | * | |
150 | */ | |
151 | ||
152 | static int gfs2_ordered_writepage(struct page *page, | |
153 | struct writeback_control *wbc) | |
154 | { | |
155 | struct inode *inode = page->mapping->host; | |
156 | struct gfs2_inode *ip = GFS2_I(inode); | |
157 | int ret; | |
158 | ||
159 | ret = gfs2_writepage_common(page, wbc); | |
160 | if (ret <= 0) | |
161 | return ret; | |
162 | ||
163 | if (!page_has_buffers(page)) { | |
164 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
165 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
166 | } | |
167 | gfs2_page_add_databufs(ip, page, 0, inode->i_sb->s_blocksize-1); | |
168 | return block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
169 | } | |
170 | ||
b8e7cbb6 SW |
171 | /** |
172 | * __gfs2_jdata_writepage - The core of jdata writepage | |
173 | * @page: The page to write | |
174 | * @wbc: The writeback control | |
175 | * | |
176 | * This is shared between writepage and writepages and implements the | |
177 | * core of the writepage operation. If a transaction is required then | |
178 | * PageChecked will have been set and the transaction will have | |
179 | * already been started before this is called. | |
180 | */ | |
181 | ||
182 | static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
183 | { | |
184 | struct inode *inode = page->mapping->host; | |
185 | struct gfs2_inode *ip = GFS2_I(inode); | |
186 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
187 | ||
188 | if (PageChecked(page)) { | |
189 | ClearPageChecked(page); | |
190 | if (!page_has_buffers(page)) { | |
191 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
192 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
193 | } | |
194 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); | |
195 | } | |
196 | return block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
197 | } | |
198 | ||
9ff8ec32 SW |
199 | /** |
200 | * gfs2_jdata_writepage - Write complete page | |
201 | * @page: Page to write | |
202 | * | |
203 | * Returns: errno | |
204 | * | |
205 | */ | |
206 | ||
207 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
208 | { | |
209 | struct inode *inode = page->mapping->host; | |
9ff8ec32 | 210 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1bb7322f | 211 | int ret; |
9ff8ec32 SW |
212 | int done_trans = 0; |
213 | ||
bf36a713 | 214 | if (PageChecked(page)) { |
b8e7cbb6 SW |
215 | if (wbc->sync_mode != WB_SYNC_ALL) |
216 | goto out_ignore; | |
1bb7322f SW |
217 | ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); |
218 | if (ret) | |
18ec7d5c | 219 | goto out_ignore; |
18ec7d5c SW |
220 | done_trans = 1; |
221 | } | |
1bb7322f SW |
222 | ret = gfs2_writepage_common(page, wbc); |
223 | if (ret > 0) | |
224 | ret = __gfs2_jdata_writepage(page, wbc); | |
18ec7d5c SW |
225 | if (done_trans) |
226 | gfs2_trans_end(sdp); | |
1bb7322f | 227 | return ret; |
18ec7d5c SW |
228 | |
229 | out_ignore: | |
230 | redirty_page_for_writepage(wbc, page); | |
231 | unlock_page(page); | |
232 | return 0; | |
b3b94faa DT |
233 | } |
234 | ||
a8d638e3 | 235 | /** |
5561093e | 236 | * gfs2_writeback_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
237 | * @mapping: The mapping to write |
238 | * @wbc: Write-back control | |
239 | * | |
5561093e | 240 | * For the data=writeback case we can already ignore buffer heads |
a8d638e3 SW |
241 | * and write whole extents at once. This is a big reduction in the |
242 | * number of I/O requests we send and the bmap calls we make in this case. | |
243 | */ | |
5561093e SW |
244 | static int gfs2_writeback_writepages(struct address_space *mapping, |
245 | struct writeback_control *wbc) | |
a8d638e3 | 246 | { |
5561093e | 247 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); |
a8d638e3 SW |
248 | } |
249 | ||
b8e7cbb6 SW |
250 | /** |
251 | * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages | |
252 | * @mapping: The mapping | |
253 | * @wbc: The writeback control | |
254 | * @writepage: The writepage function to call for each page | |
255 | * @pvec: The vector of pages | |
256 | * @nr_pages: The number of pages to write | |
257 | * | |
258 | * Returns: non-zero if loop should terminate, zero otherwise | |
259 | */ | |
260 | ||
261 | static int gfs2_write_jdata_pagevec(struct address_space *mapping, | |
262 | struct writeback_control *wbc, | |
263 | struct pagevec *pvec, | |
264 | int nr_pages, pgoff_t end) | |
265 | { | |
266 | struct inode *inode = mapping->host; | |
267 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
268 | loff_t i_size = i_size_read(inode); | |
269 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
270 | unsigned offset = i_size & (PAGE_CACHE_SIZE-1); | |
271 | unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize); | |
272 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
273 | int i; | |
274 | int ret; | |
275 | ||
20b95bf2 | 276 | ret = gfs2_trans_begin(sdp, nrblocks, nrblocks); |
b8e7cbb6 SW |
277 | if (ret < 0) |
278 | return ret; | |
279 | ||
280 | for(i = 0; i < nr_pages; i++) { | |
281 | struct page *page = pvec->pages[i]; | |
282 | ||
283 | lock_page(page); | |
284 | ||
285 | if (unlikely(page->mapping != mapping)) { | |
286 | unlock_page(page); | |
287 | continue; | |
288 | } | |
289 | ||
290 | if (!wbc->range_cyclic && page->index > end) { | |
291 | ret = 1; | |
292 | unlock_page(page); | |
293 | continue; | |
294 | } | |
295 | ||
296 | if (wbc->sync_mode != WB_SYNC_NONE) | |
297 | wait_on_page_writeback(page); | |
298 | ||
299 | if (PageWriteback(page) || | |
300 | !clear_page_dirty_for_io(page)) { | |
301 | unlock_page(page); | |
302 | continue; | |
303 | } | |
304 | ||
305 | /* Is the page fully outside i_size? (truncate in progress) */ | |
306 | if (page->index > end_index || (page->index == end_index && !offset)) { | |
307 | page->mapping->a_ops->invalidatepage(page, 0); | |
308 | unlock_page(page); | |
309 | continue; | |
310 | } | |
311 | ||
312 | ret = __gfs2_jdata_writepage(page, wbc); | |
313 | ||
314 | if (ret || (--(wbc->nr_to_write) <= 0)) | |
315 | ret = 1; | |
316 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
317 | wbc->encountered_congestion = 1; | |
318 | ret = 1; | |
319 | } | |
320 | ||
321 | } | |
322 | gfs2_trans_end(sdp); | |
323 | return ret; | |
324 | } | |
325 | ||
326 | /** | |
327 | * gfs2_write_cache_jdata - Like write_cache_pages but different | |
328 | * @mapping: The mapping to write | |
329 | * @wbc: The writeback control | |
330 | * @writepage: The writepage function to call | |
331 | * @data: The data to pass to writepage | |
332 | * | |
333 | * The reason that we use our own function here is that we need to | |
334 | * start transactions before we grab page locks. This allows us | |
335 | * to get the ordering right. | |
336 | */ | |
337 | ||
338 | static int gfs2_write_cache_jdata(struct address_space *mapping, | |
339 | struct writeback_control *wbc) | |
340 | { | |
341 | struct backing_dev_info *bdi = mapping->backing_dev_info; | |
342 | int ret = 0; | |
343 | int done = 0; | |
344 | struct pagevec pvec; | |
345 | int nr_pages; | |
346 | pgoff_t index; | |
347 | pgoff_t end; | |
348 | int scanned = 0; | |
349 | int range_whole = 0; | |
350 | ||
351 | if (wbc->nonblocking && bdi_write_congested(bdi)) { | |
352 | wbc->encountered_congestion = 1; | |
353 | return 0; | |
354 | } | |
355 | ||
356 | pagevec_init(&pvec, 0); | |
357 | if (wbc->range_cyclic) { | |
358 | index = mapping->writeback_index; /* Start from prev offset */ | |
359 | end = -1; | |
360 | } else { | |
361 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
362 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
363 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
364 | range_whole = 1; | |
365 | scanned = 1; | |
366 | } | |
367 | ||
368 | retry: | |
369 | while (!done && (index <= end) && | |
370 | (nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, | |
371 | PAGECACHE_TAG_DIRTY, | |
372 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1))) { | |
373 | scanned = 1; | |
374 | ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end); | |
375 | if (ret) | |
376 | done = 1; | |
377 | if (ret > 0) | |
378 | ret = 0; | |
379 | ||
380 | pagevec_release(&pvec); | |
381 | cond_resched(); | |
382 | } | |
383 | ||
384 | if (!scanned && !done) { | |
385 | /* | |
386 | * We hit the last page and there is more work to be done: wrap | |
387 | * back to the start of the file | |
388 | */ | |
389 | scanned = 1; | |
390 | index = 0; | |
391 | goto retry; | |
392 | } | |
393 | ||
394 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
395 | mapping->writeback_index = index; | |
396 | return ret; | |
397 | } | |
398 | ||
399 | ||
400 | /** | |
401 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk | |
402 | * @mapping: The mapping to write | |
403 | * @wbc: The writeback control | |
404 | * | |
405 | */ | |
406 | ||
407 | static int gfs2_jdata_writepages(struct address_space *mapping, | |
408 | struct writeback_control *wbc) | |
409 | { | |
410 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
411 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
412 | int ret; | |
413 | ||
414 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
415 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { | |
416 | gfs2_log_flush(sdp, ip->i_gl); | |
417 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
418 | } | |
419 | return ret; | |
420 | } | |
421 | ||
b3b94faa DT |
422 | /** |
423 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
424 | * @ip: the inode | |
425 | * @page: the page | |
426 | * | |
427 | * Returns: errno | |
428 | */ | |
429 | ||
430 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
431 | { | |
432 | struct buffer_head *dibh; | |
433 | void *kaddr; | |
434 | int error; | |
435 | ||
bf126aee | 436 | /* |
3c18ddd1 | 437 | * Due to the order of unstuffing files and ->fault(), we can be |
bf126aee SW |
438 | * asked for a zero page in the case of a stuffed file being extended, |
439 | * so we need to supply one here. It doesn't happen often. | |
440 | */ | |
441 | if (unlikely(page->index)) { | |
eebd2aa3 | 442 | zero_user(page, 0, PAGE_CACHE_SIZE); |
0a7ab79c | 443 | SetPageUptodate(page); |
bf126aee SW |
444 | return 0; |
445 | } | |
fd88de56 | 446 | |
b3b94faa DT |
447 | error = gfs2_meta_inode_buffer(ip, &dibh); |
448 | if (error) | |
449 | return error; | |
450 | ||
5c4e9e03 | 451 | kaddr = kmap_atomic(page, KM_USER0); |
fd88de56 | 452 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), |
c9e98886 SW |
453 | ip->i_disksize); |
454 | memset(kaddr + ip->i_disksize, 0, PAGE_CACHE_SIZE - ip->i_disksize); | |
c312c4fd | 455 | kunmap_atomic(kaddr, KM_USER0); |
bf126aee | 456 | flush_dcache_page(page); |
b3b94faa | 457 | brelse(dibh); |
b3b94faa DT |
458 | SetPageUptodate(page); |
459 | ||
460 | return 0; | |
461 | } | |
462 | ||
b3b94faa | 463 | |
b3b94faa | 464 | /** |
51ff87bd SW |
465 | * __gfs2_readpage - readpage |
466 | * @file: The file to read a page for | |
b3b94faa DT |
467 | * @page: The page to read |
468 | * | |
51ff87bd SW |
469 | * This is the core of gfs2's readpage. Its used by the internal file |
470 | * reading code as in that case we already hold the glock. Also its | |
471 | * called by gfs2_readpage() once the required lock has been granted. | |
472 | * | |
b3b94faa DT |
473 | */ |
474 | ||
51ff87bd | 475 | static int __gfs2_readpage(void *file, struct page *page) |
b3b94faa | 476 | { |
feaa7bba SW |
477 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
478 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
b3b94faa DT |
479 | int error; |
480 | ||
18ec7d5c | 481 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
482 | error = stuffed_readpage(ip, page); |
483 | unlock_page(page); | |
51ff87bd | 484 | } else { |
e9e1ef2b | 485 | error = mpage_readpage(page, gfs2_block_map); |
51ff87bd | 486 | } |
b3b94faa DT |
487 | |
488 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
51ff87bd | 489 | return -EIO; |
b3b94faa | 490 | |
51ff87bd SW |
491 | return error; |
492 | } | |
493 | ||
494 | /** | |
495 | * gfs2_readpage - read a page of a file | |
496 | * @file: The file to read | |
497 | * @page: The page of the file | |
498 | * | |
01b7c7ae SW |
499 | * This deals with the locking required. We have to unlock and |
500 | * relock the page in order to get the locking in the right | |
501 | * order. | |
51ff87bd SW |
502 | */ |
503 | ||
504 | static int gfs2_readpage(struct file *file, struct page *page) | |
505 | { | |
01b7c7ae SW |
506 | struct address_space *mapping = page->mapping; |
507 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
6802e340 | 508 | struct gfs2_holder gh; |
51ff87bd SW |
509 | int error; |
510 | ||
01b7c7ae | 511 | unlock_page(page); |
719ee344 SW |
512 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
513 | error = gfs2_glock_nq(&gh); | |
01b7c7ae | 514 | if (unlikely(error)) |
6802e340 | 515 | goto out; |
01b7c7ae SW |
516 | error = AOP_TRUNCATED_PAGE; |
517 | lock_page(page); | |
518 | if (page->mapping == mapping && !PageUptodate(page)) | |
519 | error = __gfs2_readpage(file, page); | |
520 | else | |
521 | unlock_page(page); | |
6802e340 | 522 | gfs2_glock_dq(&gh); |
18ec7d5c | 523 | out: |
6802e340 | 524 | gfs2_holder_uninit(&gh); |
01b7c7ae SW |
525 | if (error && error != AOP_TRUNCATED_PAGE) |
526 | lock_page(page); | |
51ff87bd SW |
527 | return error; |
528 | } | |
529 | ||
530 | /** | |
531 | * gfs2_internal_read - read an internal file | |
532 | * @ip: The gfs2 inode | |
533 | * @ra_state: The readahead state (or NULL for no readahead) | |
534 | * @buf: The buffer to fill | |
535 | * @pos: The file position | |
536 | * @size: The amount to read | |
537 | * | |
538 | */ | |
539 | ||
540 | int gfs2_internal_read(struct gfs2_inode *ip, struct file_ra_state *ra_state, | |
541 | char *buf, loff_t *pos, unsigned size) | |
542 | { | |
543 | struct address_space *mapping = ip->i_inode.i_mapping; | |
544 | unsigned long index = *pos / PAGE_CACHE_SIZE; | |
545 | unsigned offset = *pos & (PAGE_CACHE_SIZE - 1); | |
546 | unsigned copied = 0; | |
547 | unsigned amt; | |
548 | struct page *page; | |
549 | void *p; | |
550 | ||
551 | do { | |
552 | amt = size - copied; | |
553 | if (offset + size > PAGE_CACHE_SIZE) | |
554 | amt = PAGE_CACHE_SIZE - offset; | |
555 | page = read_cache_page(mapping, index, __gfs2_readpage, NULL); | |
556 | if (IS_ERR(page)) | |
557 | return PTR_ERR(page); | |
558 | p = kmap_atomic(page, KM_USER0); | |
559 | memcpy(buf + copied, p + offset, amt); | |
560 | kunmap_atomic(p, KM_USER0); | |
561 | mark_page_accessed(page); | |
562 | page_cache_release(page); | |
563 | copied += amt; | |
564 | index++; | |
565 | offset = 0; | |
566 | } while(copied < size); | |
567 | (*pos) += size; | |
568 | return size; | |
fd88de56 SW |
569 | } |
570 | ||
fd88de56 SW |
571 | /** |
572 | * gfs2_readpages - Read a bunch of pages at once | |
573 | * | |
574 | * Some notes: | |
575 | * 1. This is only for readahead, so we can simply ignore any things | |
576 | * which are slightly inconvenient (such as locking conflicts between | |
577 | * the page lock and the glock) and return having done no I/O. Its | |
578 | * obviously not something we'd want to do on too regular a basis. | |
579 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 580 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 | 581 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
e9e1ef2b | 582 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
fd88de56 | 583 | */ |
3cc3f710 | 584 | |
fd88de56 SW |
585 | static int gfs2_readpages(struct file *file, struct address_space *mapping, |
586 | struct list_head *pages, unsigned nr_pages) | |
587 | { | |
588 | struct inode *inode = mapping->host; | |
feaa7bba SW |
589 | struct gfs2_inode *ip = GFS2_I(inode); |
590 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 591 | struct gfs2_holder gh; |
3cc3f710 | 592 | int ret; |
fd88de56 | 593 | |
719ee344 SW |
594 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
595 | ret = gfs2_glock_nq(&gh); | |
51ff87bd | 596 | if (unlikely(ret)) |
3cc3f710 | 597 | goto out_uninit; |
e1d5b18a | 598 | if (!gfs2_is_stuffed(ip)) |
e9e1ef2b | 599 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map); |
3cc3f710 SW |
600 | gfs2_glock_dq(&gh); |
601 | out_uninit: | |
602 | gfs2_holder_uninit(&gh); | |
fd88de56 SW |
603 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
604 | ret = -EIO; | |
605 | return ret; | |
b3b94faa DT |
606 | } |
607 | ||
608 | /** | |
7765ec26 | 609 | * gfs2_write_begin - Begin to write to a file |
b3b94faa | 610 | * @file: The file to write to |
7765ec26 SW |
611 | * @mapping: The mapping in which to write |
612 | * @pos: The file offset at which to start writing | |
613 | * @len: Length of the write | |
614 | * @flags: Various flags | |
615 | * @pagep: Pointer to return the page | |
616 | * @fsdata: Pointer to return fs data (unused by GFS2) | |
b3b94faa DT |
617 | * |
618 | * Returns: errno | |
619 | */ | |
620 | ||
7765ec26 SW |
621 | static int gfs2_write_begin(struct file *file, struct address_space *mapping, |
622 | loff_t pos, unsigned len, unsigned flags, | |
623 | struct page **pagep, void **fsdata) | |
b3b94faa | 624 | { |
7765ec26 SW |
625 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
626 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
1946f70a | 627 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
7ed122e4 | 628 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
18ec7d5c | 629 | int alloc_required; |
b3b94faa | 630 | int error = 0; |
18ec7d5c | 631 | struct gfs2_alloc *al; |
7765ec26 SW |
632 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
633 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
634 | unsigned to = from + len; | |
635 | struct page *page; | |
52ae7b79 | 636 | |
719ee344 SW |
637 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); |
638 | error = gfs2_glock_nq(&ip->i_gh); | |
7765ec26 | 639 | if (unlikely(error)) |
18ec7d5c | 640 | goto out_uninit; |
1946f70a BM |
641 | if (&ip->i_inode == sdp->sd_rindex) { |
642 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
643 | GL_NOCACHE, &m_ip->i_gh); | |
644 | if (unlikely(error)) { | |
645 | gfs2_glock_dq(&ip->i_gh); | |
646 | goto out_uninit; | |
647 | } | |
648 | } | |
b3b94faa | 649 | |
7765ec26 | 650 | error = gfs2_write_alloc_required(ip, pos, len, &alloc_required); |
18ec7d5c | 651 | if (error) |
c41d4f09 | 652 | goto out_unlock; |
18ec7d5c | 653 | |
7ed122e4 SW |
654 | if (alloc_required || gfs2_is_jdata(ip)) |
655 | gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks); | |
656 | ||
18ec7d5c SW |
657 | if (alloc_required) { |
658 | al = gfs2_alloc_get(ip); | |
182fe5ab CG |
659 | if (!al) { |
660 | error = -ENOMEM; | |
661 | goto out_unlock; | |
662 | } | |
18ec7d5c | 663 | |
d82661d9 | 664 | error = gfs2_quota_lock_check(ip); |
18ec7d5c SW |
665 | if (error) |
666 | goto out_alloc_put; | |
667 | ||
18ec7d5c SW |
668 | al->al_requested = data_blocks + ind_blocks; |
669 | error = gfs2_inplace_reserve(ip); | |
670 | if (error) | |
671 | goto out_qunlock; | |
672 | } | |
673 | ||
674 | rblocks = RES_DINODE + ind_blocks; | |
675 | if (gfs2_is_jdata(ip)) | |
676 | rblocks += data_blocks ? data_blocks : 1; | |
677 | if (ind_blocks || data_blocks) | |
678 | rblocks += RES_STATFS + RES_QUOTA; | |
1946f70a BM |
679 | if (&ip->i_inode == sdp->sd_rindex) |
680 | rblocks += 2 * RES_STATFS; | |
18ec7d5c | 681 | |
16615be1 SW |
682 | error = gfs2_trans_begin(sdp, rblocks, |
683 | PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); | |
18ec7d5c | 684 | if (error) |
a867bb28 | 685 | goto out_trans_fail; |
18ec7d5c | 686 | |
c41d4f09 | 687 | error = -ENOMEM; |
e4fefbac | 688 | flags |= AOP_FLAG_NOFS; |
54566b2c | 689 | page = grab_cache_page_write_begin(mapping, index, flags); |
c41d4f09 SW |
690 | *pagep = page; |
691 | if (unlikely(!page)) | |
692 | goto out_endtrans; | |
693 | ||
18ec7d5c | 694 | if (gfs2_is_stuffed(ip)) { |
c41d4f09 | 695 | error = 0; |
7765ec26 | 696 | if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { |
f25ef0c1 | 697 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
698 | if (error == 0) |
699 | goto prepare_write; | |
c41d4f09 | 700 | } else if (!PageUptodate(page)) { |
b3b94faa | 701 | error = stuffed_readpage(ip, page); |
c41d4f09 | 702 | } |
5c4e9e03 | 703 | goto out; |
18ec7d5c SW |
704 | } |
705 | ||
5c4e9e03 | 706 | prepare_write: |
e9e1ef2b | 707 | error = block_prepare_write(page, from, to, gfs2_block_map); |
18ec7d5c | 708 | out: |
c41d4f09 SW |
709 | if (error == 0) |
710 | return 0; | |
711 | ||
712 | page_cache_release(page); | |
713 | if (pos + len > ip->i_inode.i_size) | |
714 | vmtruncate(&ip->i_inode, ip->i_inode.i_size); | |
715 | out_endtrans: | |
716 | gfs2_trans_end(sdp); | |
a867bb28 | 717 | out_trans_fail: |
c41d4f09 SW |
718 | if (alloc_required) { |
719 | gfs2_inplace_release(ip); | |
18ec7d5c | 720 | out_qunlock: |
c41d4f09 | 721 | gfs2_quota_unlock(ip); |
18ec7d5c | 722 | out_alloc_put: |
c41d4f09 SW |
723 | gfs2_alloc_put(ip); |
724 | } | |
18ec7d5c | 725 | out_unlock: |
1946f70a BM |
726 | if (&ip->i_inode == sdp->sd_rindex) { |
727 | gfs2_glock_dq(&m_ip->i_gh); | |
728 | gfs2_holder_uninit(&m_ip->i_gh); | |
729 | } | |
c41d4f09 | 730 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 731 | out_uninit: |
c41d4f09 | 732 | gfs2_holder_uninit(&ip->i_gh); |
b3b94faa DT |
733 | return error; |
734 | } | |
735 | ||
7ae8fa84 RP |
736 | /** |
737 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
738 | * @inode: the rindex inode | |
739 | */ | |
740 | static void adjust_fs_space(struct inode *inode) | |
741 | { | |
742 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
1946f70a BM |
743 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
744 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); | |
7ae8fa84 RP |
745 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
746 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
1946f70a | 747 | struct buffer_head *m_bh, *l_bh; |
7ae8fa84 RP |
748 | u64 fs_total, new_free; |
749 | ||
750 | /* Total up the file system space, according to the latest rindex. */ | |
751 | fs_total = gfs2_ri_total(sdp); | |
1946f70a BM |
752 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) |
753 | return; | |
7ae8fa84 RP |
754 | |
755 | spin_lock(&sdp->sd_statfs_spin); | |
1946f70a BM |
756 | gfs2_statfs_change_in(m_sc, m_bh->b_data + |
757 | sizeof(struct gfs2_dinode)); | |
7ae8fa84 RP |
758 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) |
759 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
760 | else | |
761 | new_free = 0; | |
762 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
763 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
764 | (unsigned long long)new_free); | |
7ae8fa84 | 765 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
1946f70a BM |
766 | |
767 | if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0) | |
768 | goto out; | |
769 | update_statfs(sdp, m_bh, l_bh); | |
770 | brelse(l_bh); | |
771 | out: | |
772 | brelse(m_bh); | |
7ae8fa84 RP |
773 | } |
774 | ||
b3b94faa | 775 | /** |
7765ec26 SW |
776 | * gfs2_stuffed_write_end - Write end for stuffed files |
777 | * @inode: The inode | |
778 | * @dibh: The buffer_head containing the on-disk inode | |
779 | * @pos: The file position | |
780 | * @len: The length of the write | |
781 | * @copied: How much was actually copied by the VFS | |
782 | * @page: The page | |
783 | * | |
784 | * This copies the data from the page into the inode block after | |
785 | * the inode data structure itself. | |
786 | * | |
787 | * Returns: errno | |
788 | */ | |
789 | static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, | |
790 | loff_t pos, unsigned len, unsigned copied, | |
791 | struct page *page) | |
792 | { | |
793 | struct gfs2_inode *ip = GFS2_I(inode); | |
794 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1946f70a | 795 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
7765ec26 SW |
796 | u64 to = pos + copied; |
797 | void *kaddr; | |
798 | unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode); | |
799 | struct gfs2_dinode *di = (struct gfs2_dinode *)dibh->b_data; | |
800 | ||
801 | BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode))); | |
802 | kaddr = kmap_atomic(page, KM_USER0); | |
803 | memcpy(buf + pos, kaddr + pos, copied); | |
804 | memset(kaddr + pos + copied, 0, len - copied); | |
805 | flush_dcache_page(page); | |
806 | kunmap_atomic(kaddr, KM_USER0); | |
807 | ||
808 | if (!PageUptodate(page)) | |
809 | SetPageUptodate(page); | |
810 | unlock_page(page); | |
811 | page_cache_release(page); | |
812 | ||
7537d81a AD |
813 | if (copied) { |
814 | if (inode->i_size < to) { | |
815 | i_size_write(inode, to); | |
816 | ip->i_disksize = inode->i_size; | |
817 | } | |
818 | gfs2_dinode_out(ip, di); | |
7765ec26 SW |
819 | mark_inode_dirty(inode); |
820 | } | |
821 | ||
822 | if (inode == sdp->sd_rindex) | |
823 | adjust_fs_space(inode); | |
824 | ||
825 | brelse(dibh); | |
826 | gfs2_trans_end(sdp); | |
1946f70a BM |
827 | if (inode == sdp->sd_rindex) { |
828 | gfs2_glock_dq(&m_ip->i_gh); | |
829 | gfs2_holder_uninit(&m_ip->i_gh); | |
830 | } | |
7765ec26 SW |
831 | gfs2_glock_dq(&ip->i_gh); |
832 | gfs2_holder_uninit(&ip->i_gh); | |
833 | return copied; | |
834 | } | |
835 | ||
836 | /** | |
837 | * gfs2_write_end | |
b3b94faa | 838 | * @file: The file to write to |
7765ec26 SW |
839 | * @mapping: The address space to write to |
840 | * @pos: The file position | |
841 | * @len: The length of the data | |
842 | * @copied: | |
843 | * @page: The page that has been written | |
844 | * @fsdata: The fsdata (unused in GFS2) | |
845 | * | |
846 | * The main write_end function for GFS2. We have a separate one for | |
847 | * stuffed files as they are slightly different, otherwise we just | |
848 | * put our locking around the VFS provided functions. | |
b3b94faa DT |
849 | * |
850 | * Returns: errno | |
851 | */ | |
852 | ||
7765ec26 SW |
853 | static int gfs2_write_end(struct file *file, struct address_space *mapping, |
854 | loff_t pos, unsigned len, unsigned copied, | |
855 | struct page *page, void *fsdata) | |
b3b94faa DT |
856 | { |
857 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
858 | struct gfs2_inode *ip = GFS2_I(inode); |
859 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1946f70a | 860 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
18ec7d5c | 861 | struct buffer_head *dibh; |
6dbd8224 | 862 | struct gfs2_alloc *al = ip->i_alloc; |
7765ec26 SW |
863 | unsigned int from = pos & (PAGE_CACHE_SIZE - 1); |
864 | unsigned int to = from + len; | |
865 | int ret; | |
b3b94faa | 866 | |
7afd88d9 | 867 | BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL); |
18ec7d5c | 868 | |
7765ec26 SW |
869 | ret = gfs2_meta_inode_buffer(ip, &dibh); |
870 | if (unlikely(ret)) { | |
871 | unlock_page(page); | |
872 | page_cache_release(page); | |
873 | goto failed; | |
874 | } | |
18ec7d5c SW |
875 | |
876 | gfs2_trans_add_bh(ip->i_gl, dibh, 1); | |
b3b94faa | 877 | |
7765ec26 SW |
878 | if (gfs2_is_stuffed(ip)) |
879 | return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page); | |
b3b94faa | 880 | |
bf36a713 | 881 | if (!gfs2_is_writeback(ip)) |
7765ec26 | 882 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa | 883 | |
7765ec26 | 884 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
7537d81a AD |
885 | if (ret > 0) { |
886 | if (inode->i_size > ip->i_disksize) | |
887 | ip->i_disksize = inode->i_size; | |
888 | gfs2_dinode_out(ip, dibh->b_data); | |
9656b2c1 | 889 | mark_inode_dirty(inode); |
48516ced SW |
890 | } |
891 | ||
7ae8fa84 RP |
892 | if (inode == sdp->sd_rindex) |
893 | adjust_fs_space(inode); | |
894 | ||
18ec7d5c SW |
895 | brelse(dibh); |
896 | gfs2_trans_end(sdp); | |
7765ec26 | 897 | failed: |
6dbd8224 | 898 | if (al) { |
18ec7d5c SW |
899 | gfs2_inplace_release(ip); |
900 | gfs2_quota_unlock(ip); | |
901 | gfs2_alloc_put(ip); | |
902 | } | |
1946f70a BM |
903 | if (inode == sdp->sd_rindex) { |
904 | gfs2_glock_dq(&m_ip->i_gh); | |
905 | gfs2_holder_uninit(&m_ip->i_gh); | |
906 | } | |
7765ec26 | 907 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 908 | gfs2_holder_uninit(&ip->i_gh); |
7765ec26 | 909 | return ret; |
b3b94faa DT |
910 | } |
911 | ||
8fb68595 RP |
912 | /** |
913 | * gfs2_set_page_dirty - Page dirtying function | |
914 | * @page: The page to dirty | |
915 | * | |
916 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
917 | */ | |
918 | ||
919 | static int gfs2_set_page_dirty(struct page *page) | |
920 | { | |
5561093e | 921 | SetPageChecked(page); |
8fb68595 RP |
922 | return __set_page_dirty_buffers(page); |
923 | } | |
924 | ||
b3b94faa DT |
925 | /** |
926 | * gfs2_bmap - Block map function | |
927 | * @mapping: Address space info | |
928 | * @lblock: The block to map | |
929 | * | |
930 | * Returns: The disk address for the block or 0 on hole or error | |
931 | */ | |
932 | ||
933 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
934 | { | |
feaa7bba | 935 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
936 | struct gfs2_holder i_gh; |
937 | sector_t dblock = 0; | |
938 | int error; | |
939 | ||
b3b94faa DT |
940 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
941 | if (error) | |
942 | return 0; | |
943 | ||
944 | if (!gfs2_is_stuffed(ip)) | |
e9e1ef2b | 945 | dblock = generic_block_bmap(mapping, lblock, gfs2_block_map); |
b3b94faa DT |
946 | |
947 | gfs2_glock_dq_uninit(&i_gh); | |
948 | ||
949 | return dblock; | |
950 | } | |
951 | ||
d7b616e2 SW |
952 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
953 | { | |
954 | struct gfs2_bufdata *bd; | |
955 | ||
956 | lock_buffer(bh); | |
957 | gfs2_log_lock(sdp); | |
958 | clear_buffer_dirty(bh); | |
959 | bd = bh->b_private; | |
960 | if (bd) { | |
16615be1 SW |
961 | if (!list_empty(&bd->bd_le.le_list) && !buffer_pinned(bh)) |
962 | list_del_init(&bd->bd_le.le_list); | |
963 | else | |
964 | gfs2_remove_from_journal(bh, current->journal_info, 0); | |
d7b616e2 SW |
965 | } |
966 | bh->b_bdev = NULL; | |
967 | clear_buffer_mapped(bh); | |
968 | clear_buffer_req(bh); | |
969 | clear_buffer_new(bh); | |
970 | gfs2_log_unlock(sdp); | |
971 | unlock_buffer(bh); | |
972 | } | |
973 | ||
8628de05 | 974 | static void gfs2_invalidatepage(struct page *page, unsigned long offset) |
b3b94faa | 975 | { |
d7b616e2 SW |
976 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
977 | struct buffer_head *bh, *head; | |
978 | unsigned long pos = 0; | |
979 | ||
b3b94faa | 980 | BUG_ON(!PageLocked(page)); |
8fb68595 RP |
981 | if (offset == 0) |
982 | ClearPageChecked(page); | |
d7b616e2 SW |
983 | if (!page_has_buffers(page)) |
984 | goto out; | |
b3b94faa | 985 | |
d7b616e2 SW |
986 | bh = head = page_buffers(page); |
987 | do { | |
988 | if (offset <= pos) | |
989 | gfs2_discard(sdp, bh); | |
990 | pos += bh->b_size; | |
991 | bh = bh->b_this_page; | |
992 | } while (bh != head); | |
993 | out: | |
994 | if (offset == 0) | |
995 | try_to_release_page(page, 0); | |
b3b94faa DT |
996 | } |
997 | ||
c7b33834 SW |
998 | /** |
999 | * gfs2_ok_for_dio - check that dio is valid on this file | |
1000 | * @ip: The inode | |
1001 | * @rw: READ or WRITE | |
1002 | * @offset: The offset at which we are reading or writing | |
1003 | * | |
1004 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
1005 | * 1 (to accept the i/o request) | |
1006 | */ | |
1007 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, int rw, loff_t offset) | |
1008 | { | |
1009 | /* | |
1010 | * Should we return an error here? I can't see that O_DIRECT for | |
5561093e SW |
1011 | * a stuffed file makes any sense. For now we'll silently fall |
1012 | * back to buffered I/O | |
c7b33834 | 1013 | */ |
c7b33834 SW |
1014 | if (gfs2_is_stuffed(ip)) |
1015 | return 0; | |
1016 | ||
acb57a36 | 1017 | if (offset >= i_size_read(&ip->i_inode)) |
c7b33834 SW |
1018 | return 0; |
1019 | return 1; | |
1020 | } | |
1021 | ||
1022 | ||
1023 | ||
a9e5f4d0 SW |
1024 | static ssize_t gfs2_direct_IO(int rw, struct kiocb *iocb, |
1025 | const struct iovec *iov, loff_t offset, | |
1026 | unsigned long nr_segs) | |
d1665e41 SW |
1027 | { |
1028 | struct file *file = iocb->ki_filp; | |
1029 | struct inode *inode = file->f_mapping->host; | |
feaa7bba | 1030 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
1031 | struct gfs2_holder gh; |
1032 | int rv; | |
1033 | ||
1034 | /* | |
c7b33834 SW |
1035 | * Deferred lock, even if its a write, since we do no allocation |
1036 | * on this path. All we need change is atime, and this lock mode | |
1037 | * ensures that other nodes have flushed their buffered read caches | |
1038 | * (i.e. their page cache entries for this inode). We do not, | |
1039 | * unfortunately have the option of only flushing a range like | |
1040 | * the VFS does. | |
d1665e41 | 1041 | */ |
719ee344 SW |
1042 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); |
1043 | rv = gfs2_glock_nq(&gh); | |
d1665e41 | 1044 | if (rv) |
c7b33834 SW |
1045 | return rv; |
1046 | rv = gfs2_ok_for_dio(ip, rw, offset); | |
1047 | if (rv != 1) | |
1048 | goto out; /* dio not valid, fall back to buffered i/o */ | |
1049 | ||
1050 | rv = blockdev_direct_IO_no_locking(rw, iocb, inode, inode->i_sb->s_bdev, | |
1051 | iov, offset, nr_segs, | |
1052 | gfs2_get_block_direct, NULL); | |
d1665e41 SW |
1053 | out: |
1054 | gfs2_glock_dq_m(1, &gh); | |
1055 | gfs2_holder_uninit(&gh); | |
d1665e41 SW |
1056 | return rv; |
1057 | } | |
1058 | ||
4340fe62 | 1059 | /** |
623d9355 | 1060 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
1061 | * @page: the page that's being released |
1062 | * @gfp_mask: passed from Linux VFS, ignored by us | |
1063 | * | |
1064 | * Call try_to_free_buffers() if the buffers in this page can be | |
1065 | * released. | |
1066 | * | |
1067 | * Returns: 0 | |
1068 | */ | |
1069 | ||
1070 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
1071 | { | |
1072 | struct inode *aspace = page->mapping->host; | |
1073 | struct gfs2_sbd *sdp = aspace->i_sb->s_fs_info; | |
1074 | struct buffer_head *bh, *head; | |
1075 | struct gfs2_bufdata *bd; | |
4340fe62 SW |
1076 | |
1077 | if (!page_has_buffers(page)) | |
891ba6d4 | 1078 | return 0; |
4340fe62 | 1079 | |
bb3b0e3d | 1080 | gfs2_log_lock(sdp); |
4340fe62 SW |
1081 | head = bh = page_buffers(page); |
1082 | do { | |
bb3b0e3d SW |
1083 | if (atomic_read(&bh->b_count)) |
1084 | goto cannot_release; | |
1085 | bd = bh->b_private; | |
1086 | if (bd && bd->bd_ail) | |
1087 | goto cannot_release; | |
4340fe62 | 1088 | gfs2_assert_warn(sdp, !buffer_pinned(bh)); |
623d9355 | 1089 | gfs2_assert_warn(sdp, !buffer_dirty(bh)); |
bb3b0e3d SW |
1090 | bh = bh->b_this_page; |
1091 | } while(bh != head); | |
1092 | gfs2_log_unlock(sdp); | |
4340fe62 | 1093 | |
bb3b0e3d SW |
1094 | head = bh = page_buffers(page); |
1095 | do { | |
623d9355 | 1096 | gfs2_log_lock(sdp); |
4340fe62 SW |
1097 | bd = bh->b_private; |
1098 | if (bd) { | |
1099 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
1100 | gfs2_assert_warn(sdp, list_empty(&bd->bd_list_tr)); | |
d7b616e2 SW |
1101 | if (!list_empty(&bd->bd_le.le_list)) { |
1102 | if (!buffer_pinned(bh)) | |
1103 | list_del_init(&bd->bd_le.le_list); | |
1104 | else | |
1105 | bd = NULL; | |
1106 | } | |
1107 | if (bd) | |
1108 | bd->bd_bh = NULL; | |
4340fe62 SW |
1109 | bh->b_private = NULL; |
1110 | } | |
623d9355 SW |
1111 | gfs2_log_unlock(sdp); |
1112 | if (bd) | |
1113 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
4340fe62 SW |
1114 | |
1115 | bh = bh->b_this_page; | |
166afccd | 1116 | } while (bh != head); |
4340fe62 | 1117 | |
4340fe62 | 1118 | return try_to_free_buffers(page); |
bb3b0e3d SW |
1119 | cannot_release: |
1120 | gfs2_log_unlock(sdp); | |
1121 | return 0; | |
4340fe62 SW |
1122 | } |
1123 | ||
5561093e | 1124 | static const struct address_space_operations gfs2_writeback_aops = { |
9ff8ec32 | 1125 | .writepage = gfs2_writeback_writepage, |
5561093e SW |
1126 | .writepages = gfs2_writeback_writepages, |
1127 | .readpage = gfs2_readpage, | |
1128 | .readpages = gfs2_readpages, | |
1129 | .sync_page = block_sync_page, | |
1130 | .write_begin = gfs2_write_begin, | |
1131 | .write_end = gfs2_write_end, | |
1132 | .bmap = gfs2_bmap, | |
1133 | .invalidatepage = gfs2_invalidatepage, | |
1134 | .releasepage = gfs2_releasepage, | |
1135 | .direct_IO = gfs2_direct_IO, | |
e5d9dc27 | 1136 | .migratepage = buffer_migrate_page, |
229615de | 1137 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1138 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
1139 | }; |
1140 | ||
1141 | static const struct address_space_operations gfs2_ordered_aops = { | |
9ff8ec32 | 1142 | .writepage = gfs2_ordered_writepage, |
b3b94faa | 1143 | .readpage = gfs2_readpage, |
fd88de56 | 1144 | .readpages = gfs2_readpages, |
b3b94faa | 1145 | .sync_page = block_sync_page, |
7765ec26 SW |
1146 | .write_begin = gfs2_write_begin, |
1147 | .write_end = gfs2_write_end, | |
8fb68595 | 1148 | .set_page_dirty = gfs2_set_page_dirty, |
b3b94faa DT |
1149 | .bmap = gfs2_bmap, |
1150 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 1151 | .releasepage = gfs2_releasepage, |
b3b94faa | 1152 | .direct_IO = gfs2_direct_IO, |
e5d9dc27 | 1153 | .migratepage = buffer_migrate_page, |
229615de | 1154 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1155 | .error_remove_page = generic_error_remove_page, |
b3b94faa DT |
1156 | }; |
1157 | ||
5561093e | 1158 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 1159 | .writepage = gfs2_jdata_writepage, |
b8e7cbb6 | 1160 | .writepages = gfs2_jdata_writepages, |
5561093e SW |
1161 | .readpage = gfs2_readpage, |
1162 | .readpages = gfs2_readpages, | |
1163 | .sync_page = block_sync_page, | |
1164 | .write_begin = gfs2_write_begin, | |
1165 | .write_end = gfs2_write_end, | |
1166 | .set_page_dirty = gfs2_set_page_dirty, | |
1167 | .bmap = gfs2_bmap, | |
1168 | .invalidatepage = gfs2_invalidatepage, | |
1169 | .releasepage = gfs2_releasepage, | |
229615de | 1170 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1171 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
1172 | }; |
1173 | ||
1174 | void gfs2_set_aops(struct inode *inode) | |
1175 | { | |
1176 | struct gfs2_inode *ip = GFS2_I(inode); | |
1177 | ||
1178 | if (gfs2_is_writeback(ip)) | |
1179 | inode->i_mapping->a_ops = &gfs2_writeback_aops; | |
1180 | else if (gfs2_is_ordered(ip)) | |
1181 | inode->i_mapping->a_ops = &gfs2_ordered_aops; | |
1182 | else if (gfs2_is_jdata(ip)) | |
1183 | inode->i_mapping->a_ops = &gfs2_jdata_aops; | |
1184 | else | |
1185 | BUG(); | |
1186 | } | |
1187 |