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> |
e2e40f2c | 23 | #include <linux/uio.h> |
774016b2 | 24 | #include <trace/events/writeback.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" | |
b3b94faa DT |
33 | #include "quota.h" |
34 | #include "trans.h" | |
18ec7d5c | 35 | #include "rgrp.h" |
cd81a4ba | 36 | #include "super.h" |
5c676f6d | 37 | #include "util.h" |
4340fe62 | 38 | #include "glops.h" |
b3b94faa | 39 | |
ba7f7290 | 40 | |
b120193e BP |
41 | static void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, |
42 | unsigned int from, unsigned int to) | |
ba7f7290 SW |
43 | { |
44 | struct buffer_head *head = page_buffers(page); | |
45 | unsigned int bsize = head->b_size; | |
46 | struct buffer_head *bh; | |
47 | unsigned int start, end; | |
48 | ||
49 | for (bh = head, start = 0; bh != head || !start; | |
50 | bh = bh->b_this_page, start = end) { | |
51 | end = start + bsize; | |
52 | if (end <= from || start >= to) | |
53 | continue; | |
ddf4b426 BM |
54 | if (gfs2_is_jdata(ip)) |
55 | set_buffer_uptodate(bh); | |
350a9b0a | 56 | gfs2_trans_add_data(ip->i_gl, bh); |
ba7f7290 SW |
57 | } |
58 | } | |
59 | ||
b3b94faa | 60 | /** |
7a6bbacb | 61 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
62 | * @inode: The inode |
63 | * @lblock: The block number to look up | |
64 | * @bh_result: The buffer head to return the result in | |
65 | * @create: Non-zero if we may add block to the file | |
66 | * | |
67 | * Returns: errno | |
68 | */ | |
69 | ||
7a6bbacb SW |
70 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
71 | struct buffer_head *bh_result, int create) | |
b3b94faa | 72 | { |
b3b94faa DT |
73 | int error; |
74 | ||
e9e1ef2b | 75 | error = gfs2_block_map(inode, lblock, bh_result, 0); |
b3b94faa DT |
76 | if (error) |
77 | return error; | |
de986e85 | 78 | if (!buffer_mapped(bh_result)) |
7a6bbacb SW |
79 | return -EIO; |
80 | return 0; | |
b3b94faa DT |
81 | } |
82 | ||
7a6bbacb SW |
83 | static int gfs2_get_block_direct(struct inode *inode, sector_t lblock, |
84 | struct buffer_head *bh_result, int create) | |
623d9355 | 85 | { |
e9e1ef2b | 86 | return gfs2_block_map(inode, lblock, bh_result, 0); |
623d9355 | 87 | } |
7a6bbacb | 88 | |
b3b94faa | 89 | /** |
9ff8ec32 SW |
90 | * gfs2_writepage_common - Common bits of writepage |
91 | * @page: The page to be written | |
92 | * @wbc: The writeback control | |
b3b94faa | 93 | * |
9ff8ec32 | 94 | * Returns: 1 if writepage is ok, otherwise an error code or zero if no error. |
b3b94faa DT |
95 | */ |
96 | ||
9ff8ec32 SW |
97 | static int gfs2_writepage_common(struct page *page, |
98 | struct writeback_control *wbc) | |
b3b94faa | 99 | { |
18ec7d5c | 100 | struct inode *inode = page->mapping->host; |
f4387149 SW |
101 | struct gfs2_inode *ip = GFS2_I(inode); |
102 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c SW |
103 | loff_t i_size = i_size_read(inode); |
104 | pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
105 | unsigned offset; | |
b3b94faa | 106 | |
9ff8ec32 SW |
107 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
108 | goto out; | |
5c676f6d | 109 | if (current->journal_info) |
9ff8ec32 | 110 | goto redirty; |
18ec7d5c | 111 | /* Is the page fully outside i_size? (truncate in progress) */ |
9ff8ec32 | 112 | offset = i_size & (PAGE_CACHE_SIZE-1); |
d2d7b8a2 | 113 | if (page->index > end_index || (page->index == end_index && !offset)) { |
d47992f8 | 114 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_CACHE_SIZE); |
9ff8ec32 | 115 | goto out; |
b3b94faa | 116 | } |
9ff8ec32 SW |
117 | return 1; |
118 | redirty: | |
119 | redirty_page_for_writepage(wbc, page); | |
120 | out: | |
121 | unlock_page(page); | |
122 | return 0; | |
123 | } | |
124 | ||
125 | /** | |
9d358143 | 126 | * gfs2_writepage - Write page for writeback mappings |
9ff8ec32 SW |
127 | * @page: The page |
128 | * @wbc: The writeback control | |
129 | * | |
130 | */ | |
131 | ||
9d358143 | 132 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) |
9ff8ec32 SW |
133 | { |
134 | int ret; | |
135 | ||
136 | ret = gfs2_writepage_common(page, wbc); | |
137 | if (ret <= 0) | |
138 | return ret; | |
139 | ||
30116ff6 | 140 | return nobh_writepage(page, gfs2_get_block_noalloc, wbc); |
9ff8ec32 SW |
141 | } |
142 | ||
b8e7cbb6 SW |
143 | /** |
144 | * __gfs2_jdata_writepage - The core of jdata writepage | |
145 | * @page: The page to write | |
146 | * @wbc: The writeback control | |
147 | * | |
148 | * This is shared between writepage and writepages and implements the | |
149 | * core of the writepage operation. If a transaction is required then | |
150 | * PageChecked will have been set and the transaction will have | |
151 | * already been started before this is called. | |
152 | */ | |
153 | ||
154 | static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
155 | { | |
156 | struct inode *inode = page->mapping->host; | |
157 | struct gfs2_inode *ip = GFS2_I(inode); | |
158 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
159 | ||
160 | if (PageChecked(page)) { | |
161 | ClearPageChecked(page); | |
162 | if (!page_has_buffers(page)) { | |
163 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
164 | (1 << BH_Dirty)|(1 << BH_Uptodate)); | |
165 | } | |
166 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize-1); | |
167 | } | |
168 | return block_write_full_page(page, gfs2_get_block_noalloc, wbc); | |
169 | } | |
170 | ||
9ff8ec32 SW |
171 | /** |
172 | * gfs2_jdata_writepage - Write complete page | |
173 | * @page: Page to write | |
1272574b | 174 | * @wbc: The writeback control |
9ff8ec32 SW |
175 | * |
176 | * Returns: errno | |
177 | * | |
178 | */ | |
179 | ||
180 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
181 | { | |
182 | struct inode *inode = page->mapping->host; | |
9ff8ec32 | 183 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1bb7322f | 184 | int ret; |
9ff8ec32 SW |
185 | int done_trans = 0; |
186 | ||
bf36a713 | 187 | if (PageChecked(page)) { |
b8e7cbb6 SW |
188 | if (wbc->sync_mode != WB_SYNC_ALL) |
189 | goto out_ignore; | |
1bb7322f SW |
190 | ret = gfs2_trans_begin(sdp, RES_DINODE + 1, 0); |
191 | if (ret) | |
18ec7d5c | 192 | goto out_ignore; |
18ec7d5c SW |
193 | done_trans = 1; |
194 | } | |
1bb7322f SW |
195 | ret = gfs2_writepage_common(page, wbc); |
196 | if (ret > 0) | |
197 | ret = __gfs2_jdata_writepage(page, wbc); | |
18ec7d5c SW |
198 | if (done_trans) |
199 | gfs2_trans_end(sdp); | |
1bb7322f | 200 | return ret; |
18ec7d5c SW |
201 | |
202 | out_ignore: | |
203 | redirty_page_for_writepage(wbc, page); | |
204 | unlock_page(page); | |
205 | return 0; | |
b3b94faa DT |
206 | } |
207 | ||
a8d638e3 | 208 | /** |
45138990 | 209 | * gfs2_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
210 | * @mapping: The mapping to write |
211 | * @wbc: Write-back control | |
212 | * | |
45138990 | 213 | * Used for both ordered and writeback modes. |
a8d638e3 | 214 | */ |
45138990 SW |
215 | static int gfs2_writepages(struct address_space *mapping, |
216 | struct writeback_control *wbc) | |
a8d638e3 | 217 | { |
5561093e | 218 | return mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); |
a8d638e3 SW |
219 | } |
220 | ||
b8e7cbb6 SW |
221 | /** |
222 | * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages | |
223 | * @mapping: The mapping | |
224 | * @wbc: The writeback control | |
b8e7cbb6 SW |
225 | * @pvec: The vector of pages |
226 | * @nr_pages: The number of pages to write | |
1272574b FF |
227 | * @end: End position |
228 | * @done_index: Page index | |
b8e7cbb6 SW |
229 | * |
230 | * Returns: non-zero if loop should terminate, zero otherwise | |
231 | */ | |
232 | ||
233 | static int gfs2_write_jdata_pagevec(struct address_space *mapping, | |
234 | struct writeback_control *wbc, | |
235 | struct pagevec *pvec, | |
774016b2 SW |
236 | int nr_pages, pgoff_t end, |
237 | pgoff_t *done_index) | |
b8e7cbb6 SW |
238 | { |
239 | struct inode *inode = mapping->host; | |
240 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b8e7cbb6 | 241 | unsigned nrblocks = nr_pages * (PAGE_CACHE_SIZE/inode->i_sb->s_blocksize); |
b8e7cbb6 SW |
242 | int i; |
243 | int ret; | |
244 | ||
20b95bf2 | 245 | ret = gfs2_trans_begin(sdp, nrblocks, nrblocks); |
b8e7cbb6 SW |
246 | if (ret < 0) |
247 | return ret; | |
248 | ||
249 | for(i = 0; i < nr_pages; i++) { | |
250 | struct page *page = pvec->pages[i]; | |
251 | ||
774016b2 SW |
252 | /* |
253 | * At this point, the page may be truncated or | |
254 | * invalidated (changing page->mapping to NULL), or | |
255 | * even swizzled back from swapper_space to tmpfs file | |
256 | * mapping. However, page->index will not change | |
257 | * because we have a reference on the page. | |
258 | */ | |
259 | if (page->index > end) { | |
260 | /* | |
261 | * can't be range_cyclic (1st pass) because | |
262 | * end == -1 in that case. | |
263 | */ | |
264 | ret = 1; | |
265 | break; | |
266 | } | |
267 | ||
268 | *done_index = page->index; | |
269 | ||
b8e7cbb6 SW |
270 | lock_page(page); |
271 | ||
272 | if (unlikely(page->mapping != mapping)) { | |
774016b2 | 273 | continue_unlock: |
b8e7cbb6 SW |
274 | unlock_page(page); |
275 | continue; | |
276 | } | |
277 | ||
774016b2 SW |
278 | if (!PageDirty(page)) { |
279 | /* someone wrote it for us */ | |
280 | goto continue_unlock; | |
b8e7cbb6 SW |
281 | } |
282 | ||
774016b2 SW |
283 | if (PageWriteback(page)) { |
284 | if (wbc->sync_mode != WB_SYNC_NONE) | |
285 | wait_on_page_writeback(page); | |
286 | else | |
287 | goto continue_unlock; | |
b8e7cbb6 SW |
288 | } |
289 | ||
774016b2 SW |
290 | BUG_ON(PageWriteback(page)); |
291 | if (!clear_page_dirty_for_io(page)) | |
292 | goto continue_unlock; | |
293 | ||
de1414a6 | 294 | trace_wbc_writepage(wbc, inode_to_bdi(inode)); |
b8e7cbb6 SW |
295 | |
296 | ret = __gfs2_jdata_writepage(page, wbc); | |
774016b2 SW |
297 | if (unlikely(ret)) { |
298 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
299 | unlock_page(page); | |
300 | ret = 0; | |
301 | } else { | |
302 | ||
303 | /* | |
304 | * done_index is set past this page, | |
305 | * so media errors will not choke | |
306 | * background writeout for the entire | |
307 | * file. This has consequences for | |
308 | * range_cyclic semantics (ie. it may | |
309 | * not be suitable for data integrity | |
310 | * writeout). | |
311 | */ | |
312 | *done_index = page->index + 1; | |
313 | ret = 1; | |
314 | break; | |
315 | } | |
316 | } | |
b8e7cbb6 | 317 | |
774016b2 SW |
318 | /* |
319 | * We stop writing back only if we are not doing | |
320 | * integrity sync. In case of integrity sync we have to | |
321 | * keep going until we have written all the pages | |
322 | * we tagged for writeback prior to entering this loop. | |
323 | */ | |
324 | if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) { | |
b8e7cbb6 | 325 | ret = 1; |
774016b2 SW |
326 | break; |
327 | } | |
328 | ||
b8e7cbb6 SW |
329 | } |
330 | gfs2_trans_end(sdp); | |
331 | return ret; | |
332 | } | |
333 | ||
334 | /** | |
335 | * gfs2_write_cache_jdata - Like write_cache_pages but different | |
336 | * @mapping: The mapping to write | |
337 | * @wbc: The writeback control | |
b8e7cbb6 SW |
338 | * |
339 | * The reason that we use our own function here is that we need to | |
340 | * start transactions before we grab page locks. This allows us | |
341 | * to get the ordering right. | |
342 | */ | |
343 | ||
344 | static int gfs2_write_cache_jdata(struct address_space *mapping, | |
345 | struct writeback_control *wbc) | |
346 | { | |
b8e7cbb6 SW |
347 | int ret = 0; |
348 | int done = 0; | |
349 | struct pagevec pvec; | |
350 | int nr_pages; | |
774016b2 | 351 | pgoff_t uninitialized_var(writeback_index); |
b8e7cbb6 SW |
352 | pgoff_t index; |
353 | pgoff_t end; | |
774016b2 SW |
354 | pgoff_t done_index; |
355 | int cycled; | |
b8e7cbb6 | 356 | int range_whole = 0; |
774016b2 | 357 | int tag; |
b8e7cbb6 | 358 | |
b8e7cbb6 SW |
359 | pagevec_init(&pvec, 0); |
360 | if (wbc->range_cyclic) { | |
774016b2 SW |
361 | writeback_index = mapping->writeback_index; /* prev offset */ |
362 | index = writeback_index; | |
363 | if (index == 0) | |
364 | cycled = 1; | |
365 | else | |
366 | cycled = 0; | |
b8e7cbb6 SW |
367 | end = -1; |
368 | } else { | |
369 | index = wbc->range_start >> PAGE_CACHE_SHIFT; | |
370 | end = wbc->range_end >> PAGE_CACHE_SHIFT; | |
371 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) | |
372 | range_whole = 1; | |
774016b2 | 373 | cycled = 1; /* ignore range_cyclic tests */ |
b8e7cbb6 | 374 | } |
774016b2 SW |
375 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
376 | tag = PAGECACHE_TAG_TOWRITE; | |
377 | else | |
378 | tag = PAGECACHE_TAG_DIRTY; | |
b8e7cbb6 SW |
379 | |
380 | retry: | |
774016b2 SW |
381 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
382 | tag_pages_for_writeback(mapping, index, end); | |
383 | done_index = index; | |
384 | while (!done && (index <= end)) { | |
385 | nr_pages = pagevec_lookup_tag(&pvec, mapping, &index, tag, | |
386 | min(end - index, (pgoff_t)PAGEVEC_SIZE-1) + 1); | |
387 | if (nr_pages == 0) | |
388 | break; | |
389 | ||
390 | ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, end, &done_index); | |
b8e7cbb6 SW |
391 | if (ret) |
392 | done = 1; | |
393 | if (ret > 0) | |
394 | ret = 0; | |
b8e7cbb6 SW |
395 | pagevec_release(&pvec); |
396 | cond_resched(); | |
397 | } | |
398 | ||
774016b2 | 399 | if (!cycled && !done) { |
b8e7cbb6 | 400 | /* |
774016b2 | 401 | * range_cyclic: |
b8e7cbb6 SW |
402 | * We hit the last page and there is more work to be done: wrap |
403 | * back to the start of the file | |
404 | */ | |
774016b2 | 405 | cycled = 1; |
b8e7cbb6 | 406 | index = 0; |
774016b2 | 407 | end = writeback_index - 1; |
b8e7cbb6 SW |
408 | goto retry; |
409 | } | |
410 | ||
411 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
774016b2 SW |
412 | mapping->writeback_index = done_index; |
413 | ||
b8e7cbb6 SW |
414 | return ret; |
415 | } | |
416 | ||
417 | ||
418 | /** | |
419 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk | |
420 | * @mapping: The mapping to write | |
421 | * @wbc: The writeback control | |
422 | * | |
423 | */ | |
424 | ||
425 | static int gfs2_jdata_writepages(struct address_space *mapping, | |
426 | struct writeback_control *wbc) | |
427 | { | |
428 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
429 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
430 | int ret; | |
431 | ||
432 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
433 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { | |
24972557 | 434 | gfs2_log_flush(sdp, ip->i_gl, NORMAL_FLUSH); |
b8e7cbb6 SW |
435 | ret = gfs2_write_cache_jdata(mapping, wbc); |
436 | } | |
437 | return ret; | |
438 | } | |
439 | ||
b3b94faa DT |
440 | /** |
441 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
442 | * @ip: the inode | |
443 | * @page: the page | |
444 | * | |
445 | * Returns: errno | |
446 | */ | |
447 | ||
448 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) | |
449 | { | |
450 | struct buffer_head *dibh; | |
602c89d2 | 451 | u64 dsize = i_size_read(&ip->i_inode); |
b3b94faa DT |
452 | void *kaddr; |
453 | int error; | |
454 | ||
bf126aee | 455 | /* |
3c18ddd1 | 456 | * Due to the order of unstuffing files and ->fault(), we can be |
bf126aee SW |
457 | * asked for a zero page in the case of a stuffed file being extended, |
458 | * so we need to supply one here. It doesn't happen often. | |
459 | */ | |
460 | if (unlikely(page->index)) { | |
eebd2aa3 | 461 | zero_user(page, 0, PAGE_CACHE_SIZE); |
0a7ab79c | 462 | SetPageUptodate(page); |
bf126aee SW |
463 | return 0; |
464 | } | |
fd88de56 | 465 | |
b3b94faa DT |
466 | error = gfs2_meta_inode_buffer(ip, &dibh); |
467 | if (error) | |
468 | return error; | |
469 | ||
d9349285 | 470 | kaddr = kmap_atomic(page); |
602c89d2 SW |
471 | if (dsize > (dibh->b_size - sizeof(struct gfs2_dinode))) |
472 | dsize = (dibh->b_size - sizeof(struct gfs2_dinode)); | |
473 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); | |
474 | memset(kaddr + dsize, 0, PAGE_CACHE_SIZE - dsize); | |
d9349285 | 475 | kunmap_atomic(kaddr); |
bf126aee | 476 | flush_dcache_page(page); |
b3b94faa | 477 | brelse(dibh); |
b3b94faa DT |
478 | SetPageUptodate(page); |
479 | ||
480 | return 0; | |
481 | } | |
482 | ||
b3b94faa | 483 | |
b3b94faa | 484 | /** |
51ff87bd SW |
485 | * __gfs2_readpage - readpage |
486 | * @file: The file to read a page for | |
b3b94faa DT |
487 | * @page: The page to read |
488 | * | |
51ff87bd SW |
489 | * This is the core of gfs2's readpage. Its used by the internal file |
490 | * reading code as in that case we already hold the glock. Also its | |
491 | * called by gfs2_readpage() once the required lock has been granted. | |
492 | * | |
b3b94faa DT |
493 | */ |
494 | ||
51ff87bd | 495 | static int __gfs2_readpage(void *file, struct page *page) |
b3b94faa | 496 | { |
feaa7bba SW |
497 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
498 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
b3b94faa DT |
499 | int error; |
500 | ||
18ec7d5c | 501 | if (gfs2_is_stuffed(ip)) { |
fd88de56 SW |
502 | error = stuffed_readpage(ip, page); |
503 | unlock_page(page); | |
51ff87bd | 504 | } else { |
e9e1ef2b | 505 | error = mpage_readpage(page, gfs2_block_map); |
51ff87bd | 506 | } |
b3b94faa DT |
507 | |
508 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) | |
51ff87bd | 509 | return -EIO; |
b3b94faa | 510 | |
51ff87bd SW |
511 | return error; |
512 | } | |
513 | ||
514 | /** | |
515 | * gfs2_readpage - read a page of a file | |
516 | * @file: The file to read | |
517 | * @page: The page of the file | |
518 | * | |
01b7c7ae SW |
519 | * This deals with the locking required. We have to unlock and |
520 | * relock the page in order to get the locking in the right | |
521 | * order. | |
51ff87bd SW |
522 | */ |
523 | ||
524 | static int gfs2_readpage(struct file *file, struct page *page) | |
525 | { | |
01b7c7ae SW |
526 | struct address_space *mapping = page->mapping; |
527 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
6802e340 | 528 | struct gfs2_holder gh; |
51ff87bd SW |
529 | int error; |
530 | ||
01b7c7ae | 531 | unlock_page(page); |
719ee344 SW |
532 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
533 | error = gfs2_glock_nq(&gh); | |
01b7c7ae | 534 | if (unlikely(error)) |
6802e340 | 535 | goto out; |
01b7c7ae SW |
536 | error = AOP_TRUNCATED_PAGE; |
537 | lock_page(page); | |
538 | if (page->mapping == mapping && !PageUptodate(page)) | |
539 | error = __gfs2_readpage(file, page); | |
540 | else | |
541 | unlock_page(page); | |
6802e340 | 542 | gfs2_glock_dq(&gh); |
18ec7d5c | 543 | out: |
6802e340 | 544 | gfs2_holder_uninit(&gh); |
01b7c7ae SW |
545 | if (error && error != AOP_TRUNCATED_PAGE) |
546 | lock_page(page); | |
51ff87bd SW |
547 | return error; |
548 | } | |
549 | ||
550 | /** | |
551 | * gfs2_internal_read - read an internal file | |
552 | * @ip: The gfs2 inode | |
51ff87bd SW |
553 | * @buf: The buffer to fill |
554 | * @pos: The file position | |
555 | * @size: The amount to read | |
556 | * | |
557 | */ | |
558 | ||
4306629e AP |
559 | int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos, |
560 | unsigned size) | |
51ff87bd SW |
561 | { |
562 | struct address_space *mapping = ip->i_inode.i_mapping; | |
563 | unsigned long index = *pos / PAGE_CACHE_SIZE; | |
564 | unsigned offset = *pos & (PAGE_CACHE_SIZE - 1); | |
565 | unsigned copied = 0; | |
566 | unsigned amt; | |
567 | struct page *page; | |
568 | void *p; | |
569 | ||
570 | do { | |
571 | amt = size - copied; | |
572 | if (offset + size > PAGE_CACHE_SIZE) | |
573 | amt = PAGE_CACHE_SIZE - offset; | |
574 | page = read_cache_page(mapping, index, __gfs2_readpage, NULL); | |
575 | if (IS_ERR(page)) | |
576 | return PTR_ERR(page); | |
d9349285 | 577 | p = kmap_atomic(page); |
51ff87bd | 578 | memcpy(buf + copied, p + offset, amt); |
d9349285 | 579 | kunmap_atomic(p); |
51ff87bd SW |
580 | page_cache_release(page); |
581 | copied += amt; | |
582 | index++; | |
583 | offset = 0; | |
584 | } while(copied < size); | |
585 | (*pos) += size; | |
586 | return size; | |
fd88de56 SW |
587 | } |
588 | ||
fd88de56 SW |
589 | /** |
590 | * gfs2_readpages - Read a bunch of pages at once | |
1272574b FF |
591 | * @file: The file to read from |
592 | * @mapping: Address space info | |
593 | * @pages: List of pages to read | |
594 | * @nr_pages: Number of pages to read | |
fd88de56 SW |
595 | * |
596 | * Some notes: | |
597 | * 1. This is only for readahead, so we can simply ignore any things | |
598 | * which are slightly inconvenient (such as locking conflicts between | |
599 | * the page lock and the glock) and return having done no I/O. Its | |
600 | * obviously not something we'd want to do on too regular a basis. | |
601 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 602 | * 2. We don't handle stuffed files here we let readpage do the honours. |
fd88de56 | 603 | * 3. mpage_readpages() does most of the heavy lifting in the common case. |
e9e1ef2b | 604 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
fd88de56 | 605 | */ |
3cc3f710 | 606 | |
fd88de56 SW |
607 | static int gfs2_readpages(struct file *file, struct address_space *mapping, |
608 | struct list_head *pages, unsigned nr_pages) | |
609 | { | |
610 | struct inode *inode = mapping->host; | |
feaa7bba SW |
611 | struct gfs2_inode *ip = GFS2_I(inode); |
612 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
fd88de56 | 613 | struct gfs2_holder gh; |
3cc3f710 | 614 | int ret; |
fd88de56 | 615 | |
719ee344 SW |
616 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
617 | ret = gfs2_glock_nq(&gh); | |
51ff87bd | 618 | if (unlikely(ret)) |
3cc3f710 | 619 | goto out_uninit; |
e1d5b18a | 620 | if (!gfs2_is_stuffed(ip)) |
e9e1ef2b | 621 | ret = mpage_readpages(mapping, pages, nr_pages, gfs2_block_map); |
3cc3f710 SW |
622 | gfs2_glock_dq(&gh); |
623 | out_uninit: | |
624 | gfs2_holder_uninit(&gh); | |
fd88de56 SW |
625 | if (unlikely(test_bit(SDF_SHUTDOWN, &sdp->sd_flags))) |
626 | ret = -EIO; | |
627 | return ret; | |
b3b94faa DT |
628 | } |
629 | ||
630 | /** | |
7765ec26 | 631 | * gfs2_write_begin - Begin to write to a file |
b3b94faa | 632 | * @file: The file to write to |
7765ec26 SW |
633 | * @mapping: The mapping in which to write |
634 | * @pos: The file offset at which to start writing | |
635 | * @len: Length of the write | |
636 | * @flags: Various flags | |
637 | * @pagep: Pointer to return the page | |
638 | * @fsdata: Pointer to return fs data (unused by GFS2) | |
b3b94faa DT |
639 | * |
640 | * Returns: errno | |
641 | */ | |
642 | ||
7765ec26 SW |
643 | static int gfs2_write_begin(struct file *file, struct address_space *mapping, |
644 | loff_t pos, unsigned len, unsigned flags, | |
645 | struct page **pagep, void **fsdata) | |
b3b94faa | 646 | { |
7765ec26 SW |
647 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
648 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
1946f70a | 649 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
7ed122e4 | 650 | unsigned int data_blocks = 0, ind_blocks = 0, rblocks; |
71f890f7 | 651 | unsigned requested = 0; |
18ec7d5c | 652 | int alloc_required; |
b3b94faa | 653 | int error = 0; |
7765ec26 SW |
654 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
655 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); | |
7765ec26 | 656 | struct page *page; |
52ae7b79 | 657 | |
719ee344 SW |
658 | gfs2_holder_init(ip->i_gl, LM_ST_EXCLUSIVE, 0, &ip->i_gh); |
659 | error = gfs2_glock_nq(&ip->i_gh); | |
7765ec26 | 660 | if (unlikely(error)) |
18ec7d5c | 661 | goto out_uninit; |
1946f70a BM |
662 | if (&ip->i_inode == sdp->sd_rindex) { |
663 | error = gfs2_glock_nq_init(m_ip->i_gl, LM_ST_EXCLUSIVE, | |
664 | GL_NOCACHE, &m_ip->i_gh); | |
665 | if (unlikely(error)) { | |
666 | gfs2_glock_dq(&ip->i_gh); | |
667 | goto out_uninit; | |
668 | } | |
669 | } | |
b3b94faa | 670 | |
461cb419 | 671 | alloc_required = gfs2_write_alloc_required(ip, pos, len); |
18ec7d5c | 672 | |
7ed122e4 SW |
673 | if (alloc_required || gfs2_is_jdata(ip)) |
674 | gfs2_write_calc_reserv(ip, len, &data_blocks, &ind_blocks); | |
675 | ||
18ec7d5c | 676 | if (alloc_required) { |
7b9cff46 | 677 | struct gfs2_alloc_parms ap = { .aflags = 0, }; |
b8fbf471 AD |
678 | requested = data_blocks + ind_blocks; |
679 | ap.target = requested; | |
680 | error = gfs2_quota_lock_check(ip, &ap); | |
18ec7d5c | 681 | if (error) |
5407e242 | 682 | goto out_unlock; |
18ec7d5c | 683 | |
7b9cff46 | 684 | error = gfs2_inplace_reserve(ip, &ap); |
18ec7d5c SW |
685 | if (error) |
686 | goto out_qunlock; | |
687 | } | |
688 | ||
689 | rblocks = RES_DINODE + ind_blocks; | |
690 | if (gfs2_is_jdata(ip)) | |
691 | rblocks += data_blocks ? data_blocks : 1; | |
692 | if (ind_blocks || data_blocks) | |
693 | rblocks += RES_STATFS + RES_QUOTA; | |
1946f70a BM |
694 | if (&ip->i_inode == sdp->sd_rindex) |
695 | rblocks += 2 * RES_STATFS; | |
bf97b673 | 696 | if (alloc_required) |
71f890f7 | 697 | rblocks += gfs2_rg_blocks(ip, requested); |
18ec7d5c | 698 | |
16615be1 SW |
699 | error = gfs2_trans_begin(sdp, rblocks, |
700 | PAGE_CACHE_SIZE/sdp->sd_sb.sb_bsize); | |
18ec7d5c | 701 | if (error) |
a867bb28 | 702 | goto out_trans_fail; |
18ec7d5c | 703 | |
c41d4f09 | 704 | error = -ENOMEM; |
e4fefbac | 705 | flags |= AOP_FLAG_NOFS; |
54566b2c | 706 | page = grab_cache_page_write_begin(mapping, index, flags); |
c41d4f09 SW |
707 | *pagep = page; |
708 | if (unlikely(!page)) | |
709 | goto out_endtrans; | |
710 | ||
18ec7d5c | 711 | if (gfs2_is_stuffed(ip)) { |
c41d4f09 | 712 | error = 0; |
7765ec26 | 713 | if (pos + len > sdp->sd_sb.sb_bsize - sizeof(struct gfs2_dinode)) { |
f25ef0c1 | 714 | error = gfs2_unstuff_dinode(ip, page); |
5c4e9e03 SW |
715 | if (error == 0) |
716 | goto prepare_write; | |
c41d4f09 | 717 | } else if (!PageUptodate(page)) { |
b3b94faa | 718 | error = stuffed_readpage(ip, page); |
c41d4f09 | 719 | } |
5c4e9e03 | 720 | goto out; |
18ec7d5c SW |
721 | } |
722 | ||
5c4e9e03 | 723 | prepare_write: |
ebdec241 | 724 | error = __block_write_begin(page, from, len, gfs2_block_map); |
18ec7d5c | 725 | out: |
c41d4f09 SW |
726 | if (error == 0) |
727 | return 0; | |
728 | ||
6c474f7b | 729 | unlock_page(page); |
c41d4f09 | 730 | page_cache_release(page); |
15c6fd97 | 731 | |
ff8f33c8 | 732 | gfs2_trans_end(sdp); |
c41d4f09 | 733 | if (pos + len > ip->i_inode.i_size) |
ff8f33c8 SW |
734 | gfs2_trim_blocks(&ip->i_inode); |
735 | goto out_trans_fail; | |
736 | ||
c41d4f09 SW |
737 | out_endtrans: |
738 | gfs2_trans_end(sdp); | |
a867bb28 | 739 | out_trans_fail: |
c41d4f09 SW |
740 | if (alloc_required) { |
741 | gfs2_inplace_release(ip); | |
18ec7d5c | 742 | out_qunlock: |
c41d4f09 | 743 | gfs2_quota_unlock(ip); |
c41d4f09 | 744 | } |
18ec7d5c | 745 | out_unlock: |
1946f70a BM |
746 | if (&ip->i_inode == sdp->sd_rindex) { |
747 | gfs2_glock_dq(&m_ip->i_gh); | |
748 | gfs2_holder_uninit(&m_ip->i_gh); | |
749 | } | |
c41d4f09 | 750 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 751 | out_uninit: |
c41d4f09 | 752 | gfs2_holder_uninit(&ip->i_gh); |
b3b94faa DT |
753 | return error; |
754 | } | |
755 | ||
7ae8fa84 RP |
756 | /** |
757 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
758 | * @inode: the rindex inode | |
759 | */ | |
760 | static void adjust_fs_space(struct inode *inode) | |
761 | { | |
762 | struct gfs2_sbd *sdp = inode->i_sb->s_fs_info; | |
1946f70a BM |
763 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
764 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); | |
7ae8fa84 RP |
765 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
766 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
1946f70a | 767 | struct buffer_head *m_bh, *l_bh; |
7ae8fa84 RP |
768 | u64 fs_total, new_free; |
769 | ||
770 | /* Total up the file system space, according to the latest rindex. */ | |
771 | fs_total = gfs2_ri_total(sdp); | |
1946f70a BM |
772 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) |
773 | return; | |
7ae8fa84 RP |
774 | |
775 | spin_lock(&sdp->sd_statfs_spin); | |
1946f70a BM |
776 | gfs2_statfs_change_in(m_sc, m_bh->b_data + |
777 | sizeof(struct gfs2_dinode)); | |
7ae8fa84 RP |
778 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) |
779 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
780 | else | |
781 | new_free = 0; | |
782 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
783 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
784 | (unsigned long long)new_free); | |
7ae8fa84 | 785 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
1946f70a BM |
786 | |
787 | if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0) | |
788 | goto out; | |
789 | update_statfs(sdp, m_bh, l_bh); | |
790 | brelse(l_bh); | |
791 | out: | |
792 | brelse(m_bh); | |
7ae8fa84 RP |
793 | } |
794 | ||
b3b94faa | 795 | /** |
7765ec26 SW |
796 | * gfs2_stuffed_write_end - Write end for stuffed files |
797 | * @inode: The inode | |
798 | * @dibh: The buffer_head containing the on-disk inode | |
799 | * @pos: The file position | |
800 | * @len: The length of the write | |
801 | * @copied: How much was actually copied by the VFS | |
802 | * @page: The page | |
803 | * | |
804 | * This copies the data from the page into the inode block after | |
805 | * the inode data structure itself. | |
806 | * | |
807 | * Returns: errno | |
808 | */ | |
809 | static int gfs2_stuffed_write_end(struct inode *inode, struct buffer_head *dibh, | |
810 | loff_t pos, unsigned len, unsigned copied, | |
811 | struct page *page) | |
812 | { | |
813 | struct gfs2_inode *ip = GFS2_I(inode); | |
814 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1946f70a | 815 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
7765ec26 SW |
816 | u64 to = pos + copied; |
817 | void *kaddr; | |
818 | unsigned char *buf = dibh->b_data + sizeof(struct gfs2_dinode); | |
7765ec26 SW |
819 | |
820 | BUG_ON((pos + len) > (dibh->b_size - sizeof(struct gfs2_dinode))); | |
d9349285 | 821 | kaddr = kmap_atomic(page); |
7765ec26 SW |
822 | memcpy(buf + pos, kaddr + pos, copied); |
823 | memset(kaddr + pos + copied, 0, len - copied); | |
824 | flush_dcache_page(page); | |
d9349285 | 825 | kunmap_atomic(kaddr); |
7765ec26 SW |
826 | |
827 | if (!PageUptodate(page)) | |
828 | SetPageUptodate(page); | |
829 | unlock_page(page); | |
830 | page_cache_release(page); | |
831 | ||
7537d81a | 832 | if (copied) { |
a2e0f799 | 833 | if (inode->i_size < to) |
7537d81a | 834 | i_size_write(inode, to); |
7765ec26 SW |
835 | mark_inode_dirty(inode); |
836 | } | |
837 | ||
9ae3c6de | 838 | if (inode == sdp->sd_rindex) { |
7765ec26 | 839 | adjust_fs_space(inode); |
ca9248d8 | 840 | sdp->sd_rindex_uptodate = 0; |
9ae3c6de | 841 | } |
7765ec26 SW |
842 | |
843 | brelse(dibh); | |
844 | gfs2_trans_end(sdp); | |
1946f70a BM |
845 | if (inode == sdp->sd_rindex) { |
846 | gfs2_glock_dq(&m_ip->i_gh); | |
847 | gfs2_holder_uninit(&m_ip->i_gh); | |
848 | } | |
7765ec26 SW |
849 | gfs2_glock_dq(&ip->i_gh); |
850 | gfs2_holder_uninit(&ip->i_gh); | |
851 | return copied; | |
852 | } | |
853 | ||
854 | /** | |
855 | * gfs2_write_end | |
b3b94faa | 856 | * @file: The file to write to |
7765ec26 SW |
857 | * @mapping: The address space to write to |
858 | * @pos: The file position | |
859 | * @len: The length of the data | |
1272574b | 860 | * @copied: How much was actually copied by the VFS |
7765ec26 SW |
861 | * @page: The page that has been written |
862 | * @fsdata: The fsdata (unused in GFS2) | |
863 | * | |
864 | * The main write_end function for GFS2. We have a separate one for | |
865 | * stuffed files as they are slightly different, otherwise we just | |
866 | * put our locking around the VFS provided functions. | |
b3b94faa DT |
867 | * |
868 | * Returns: errno | |
869 | */ | |
870 | ||
7765ec26 SW |
871 | static int gfs2_write_end(struct file *file, struct address_space *mapping, |
872 | loff_t pos, unsigned len, unsigned copied, | |
873 | struct page *page, void *fsdata) | |
b3b94faa DT |
874 | { |
875 | struct inode *inode = page->mapping->host; | |
feaa7bba SW |
876 | struct gfs2_inode *ip = GFS2_I(inode); |
877 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
1946f70a | 878 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
18ec7d5c | 879 | struct buffer_head *dibh; |
7765ec26 SW |
880 | unsigned int from = pos & (PAGE_CACHE_SIZE - 1); |
881 | unsigned int to = from + len; | |
882 | int ret; | |
0c901809 BM |
883 | struct gfs2_trans *tr = current->journal_info; |
884 | BUG_ON(!tr); | |
b3b94faa | 885 | |
7afd88d9 | 886 | BUG_ON(gfs2_glock_is_locked_by_me(ip->i_gl) == NULL); |
18ec7d5c | 887 | |
7765ec26 SW |
888 | ret = gfs2_meta_inode_buffer(ip, &dibh); |
889 | if (unlikely(ret)) { | |
890 | unlock_page(page); | |
891 | page_cache_release(page); | |
892 | goto failed; | |
893 | } | |
18ec7d5c | 894 | |
7765ec26 SW |
895 | if (gfs2_is_stuffed(ip)) |
896 | return gfs2_stuffed_write_end(inode, dibh, pos, len, copied, page); | |
b3b94faa | 897 | |
bf36a713 | 898 | if (!gfs2_is_writeback(ip)) |
7765ec26 | 899 | gfs2_page_add_databufs(ip, page, from, to); |
b3b94faa | 900 | |
7765ec26 | 901 | ret = generic_write_end(file, mapping, pos, len, copied, page, fsdata); |
0c901809 BM |
902 | if (tr->tr_num_buf_new) |
903 | __mark_inode_dirty(inode, I_DIRTY_DATASYNC); | |
904 | else | |
905 | gfs2_trans_add_meta(ip->i_gl, dibh); | |
906 | ||
48516ced | 907 | |
9ae3c6de | 908 | if (inode == sdp->sd_rindex) { |
7ae8fa84 | 909 | adjust_fs_space(inode); |
ca9248d8 | 910 | sdp->sd_rindex_uptodate = 0; |
9ae3c6de | 911 | } |
7ae8fa84 | 912 | |
18ec7d5c | 913 | brelse(dibh); |
7765ec26 | 914 | failed: |
deab72d3 | 915 | gfs2_trans_end(sdp); |
71f890f7 | 916 | gfs2_inplace_release(ip); |
5407e242 | 917 | if (ip->i_res->rs_qa_qd_num) |
18ec7d5c | 918 | gfs2_quota_unlock(ip); |
1946f70a BM |
919 | if (inode == sdp->sd_rindex) { |
920 | gfs2_glock_dq(&m_ip->i_gh); | |
921 | gfs2_holder_uninit(&m_ip->i_gh); | |
922 | } | |
7765ec26 | 923 | gfs2_glock_dq(&ip->i_gh); |
18ec7d5c | 924 | gfs2_holder_uninit(&ip->i_gh); |
7765ec26 | 925 | return ret; |
b3b94faa DT |
926 | } |
927 | ||
8fb68595 RP |
928 | /** |
929 | * gfs2_set_page_dirty - Page dirtying function | |
930 | * @page: The page to dirty | |
931 | * | |
932 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
933 | */ | |
934 | ||
935 | static int gfs2_set_page_dirty(struct page *page) | |
936 | { | |
5561093e | 937 | SetPageChecked(page); |
8fb68595 RP |
938 | return __set_page_dirty_buffers(page); |
939 | } | |
940 | ||
b3b94faa DT |
941 | /** |
942 | * gfs2_bmap - Block map function | |
943 | * @mapping: Address space info | |
944 | * @lblock: The block to map | |
945 | * | |
946 | * Returns: The disk address for the block or 0 on hole or error | |
947 | */ | |
948 | ||
949 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
950 | { | |
feaa7bba | 951 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
952 | struct gfs2_holder i_gh; |
953 | sector_t dblock = 0; | |
954 | int error; | |
955 | ||
b3b94faa DT |
956 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
957 | if (error) | |
958 | return 0; | |
959 | ||
960 | if (!gfs2_is_stuffed(ip)) | |
e9e1ef2b | 961 | dblock = generic_block_bmap(mapping, lblock, gfs2_block_map); |
b3b94faa DT |
962 | |
963 | gfs2_glock_dq_uninit(&i_gh); | |
964 | ||
965 | return dblock; | |
966 | } | |
967 | ||
d7b616e2 SW |
968 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
969 | { | |
970 | struct gfs2_bufdata *bd; | |
971 | ||
972 | lock_buffer(bh); | |
973 | gfs2_log_lock(sdp); | |
974 | clear_buffer_dirty(bh); | |
975 | bd = bh->b_private; | |
976 | if (bd) { | |
c0752aa7 BP |
977 | if (!list_empty(&bd->bd_list) && !buffer_pinned(bh)) |
978 | list_del_init(&bd->bd_list); | |
16615be1 SW |
979 | else |
980 | gfs2_remove_from_journal(bh, current->journal_info, 0); | |
d7b616e2 SW |
981 | } |
982 | bh->b_bdev = NULL; | |
983 | clear_buffer_mapped(bh); | |
984 | clear_buffer_req(bh); | |
985 | clear_buffer_new(bh); | |
986 | gfs2_log_unlock(sdp); | |
987 | unlock_buffer(bh); | |
988 | } | |
989 | ||
d47992f8 LC |
990 | static void gfs2_invalidatepage(struct page *page, unsigned int offset, |
991 | unsigned int length) | |
b3b94faa | 992 | { |
d7b616e2 | 993 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
5c0bb97c LC |
994 | unsigned int stop = offset + length; |
995 | int partial_page = (offset || length < PAGE_CACHE_SIZE); | |
d7b616e2 SW |
996 | struct buffer_head *bh, *head; |
997 | unsigned long pos = 0; | |
998 | ||
b3b94faa | 999 | BUG_ON(!PageLocked(page)); |
5c0bb97c | 1000 | if (!partial_page) |
8fb68595 | 1001 | ClearPageChecked(page); |
d7b616e2 SW |
1002 | if (!page_has_buffers(page)) |
1003 | goto out; | |
b3b94faa | 1004 | |
d7b616e2 SW |
1005 | bh = head = page_buffers(page); |
1006 | do { | |
5c0bb97c LC |
1007 | if (pos + bh->b_size > stop) |
1008 | return; | |
1009 | ||
d7b616e2 SW |
1010 | if (offset <= pos) |
1011 | gfs2_discard(sdp, bh); | |
1012 | pos += bh->b_size; | |
1013 | bh = bh->b_this_page; | |
1014 | } while (bh != head); | |
1015 | out: | |
5c0bb97c | 1016 | if (!partial_page) |
d7b616e2 | 1017 | try_to_release_page(page, 0); |
b3b94faa DT |
1018 | } |
1019 | ||
c7b33834 SW |
1020 | /** |
1021 | * gfs2_ok_for_dio - check that dio is valid on this file | |
1022 | * @ip: The inode | |
c7b33834 SW |
1023 | * @offset: The offset at which we are reading or writing |
1024 | * | |
1025 | * Returns: 0 (to ignore the i/o request and thus fall back to buffered i/o) | |
1026 | * 1 (to accept the i/o request) | |
1027 | */ | |
6f673763 | 1028 | static int gfs2_ok_for_dio(struct gfs2_inode *ip, loff_t offset) |
c7b33834 SW |
1029 | { |
1030 | /* | |
1031 | * Should we return an error here? I can't see that O_DIRECT for | |
5561093e SW |
1032 | * a stuffed file makes any sense. For now we'll silently fall |
1033 | * back to buffered I/O | |
c7b33834 | 1034 | */ |
c7b33834 SW |
1035 | if (gfs2_is_stuffed(ip)) |
1036 | return 0; | |
1037 | ||
acb57a36 | 1038 | if (offset >= i_size_read(&ip->i_inode)) |
c7b33834 SW |
1039 | return 0; |
1040 | return 1; | |
1041 | } | |
1042 | ||
1043 | ||
1044 | ||
22c6186e OS |
1045 | static ssize_t gfs2_direct_IO(struct kiocb *iocb, struct iov_iter *iter, |
1046 | loff_t offset) | |
d1665e41 SW |
1047 | { |
1048 | struct file *file = iocb->ki_filp; | |
1049 | struct inode *inode = file->f_mapping->host; | |
dfd11184 | 1050 | struct address_space *mapping = inode->i_mapping; |
feaa7bba | 1051 | struct gfs2_inode *ip = GFS2_I(inode); |
d1665e41 SW |
1052 | struct gfs2_holder gh; |
1053 | int rv; | |
1054 | ||
1055 | /* | |
c7b33834 SW |
1056 | * Deferred lock, even if its a write, since we do no allocation |
1057 | * on this path. All we need change is atime, and this lock mode | |
1058 | * ensures that other nodes have flushed their buffered read caches | |
1059 | * (i.e. their page cache entries for this inode). We do not, | |
1060 | * unfortunately have the option of only flushing a range like | |
1061 | * the VFS does. | |
d1665e41 | 1062 | */ |
719ee344 SW |
1063 | gfs2_holder_init(ip->i_gl, LM_ST_DEFERRED, 0, &gh); |
1064 | rv = gfs2_glock_nq(&gh); | |
d1665e41 | 1065 | if (rv) |
c7b33834 | 1066 | return rv; |
6f673763 | 1067 | rv = gfs2_ok_for_dio(ip, offset); |
c7b33834 SW |
1068 | if (rv != 1) |
1069 | goto out; /* dio not valid, fall back to buffered i/o */ | |
1070 | ||
dfd11184 SW |
1071 | /* |
1072 | * Now since we are holding a deferred (CW) lock at this point, you | |
1073 | * might be wondering why this is ever needed. There is a case however | |
1074 | * where we've granted a deferred local lock against a cached exclusive | |
1075 | * glock. That is ok provided all granted local locks are deferred, but | |
1076 | * it also means that it is possible to encounter pages which are | |
1077 | * cached and possibly also mapped. So here we check for that and sort | |
1078 | * them out ahead of the dio. The glock state machine will take care of | |
1079 | * everything else. | |
1080 | * | |
1081 | * If in fact the cached glock state (gl->gl_state) is deferred (CW) in | |
1082 | * the first place, mapping->nr_pages will always be zero. | |
1083 | */ | |
1084 | if (mapping->nrpages) { | |
1085 | loff_t lstart = offset & (PAGE_CACHE_SIZE - 1); | |
a6cbcd4a | 1086 | loff_t len = iov_iter_count(iter); |
dfd11184 SW |
1087 | loff_t end = PAGE_ALIGN(offset + len) - 1; |
1088 | ||
1089 | rv = 0; | |
1090 | if (len == 0) | |
1091 | goto out; | |
1092 | if (test_and_clear_bit(GIF_SW_PAGED, &ip->i_flags)) | |
1093 | unmap_shared_mapping_range(ip->i_inode.i_mapping, offset, len); | |
1094 | rv = filemap_write_and_wait_range(mapping, lstart, end); | |
1095 | if (rv) | |
086352f1 | 1096 | goto out; |
6f673763 | 1097 | if (iov_iter_rw(iter) == WRITE) |
086352f1 | 1098 | truncate_inode_pages_range(mapping, lstart, end); |
dfd11184 SW |
1099 | } |
1100 | ||
17f8c842 OS |
1101 | rv = __blockdev_direct_IO(iocb, inode, inode->i_sb->s_bdev, iter, |
1102 | offset, gfs2_get_block_direct, NULL, NULL, 0); | |
d1665e41 | 1103 | out: |
8e711e10 | 1104 | gfs2_glock_dq(&gh); |
d1665e41 | 1105 | gfs2_holder_uninit(&gh); |
d1665e41 SW |
1106 | return rv; |
1107 | } | |
1108 | ||
4340fe62 | 1109 | /** |
623d9355 | 1110 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
1111 | * @page: the page that's being released |
1112 | * @gfp_mask: passed from Linux VFS, ignored by us | |
1113 | * | |
1114 | * Call try_to_free_buffers() if the buffers in this page can be | |
1115 | * released. | |
1116 | * | |
1117 | * Returns: 0 | |
1118 | */ | |
1119 | ||
1120 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
1121 | { | |
009d8518 SW |
1122 | struct address_space *mapping = page->mapping; |
1123 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); | |
4340fe62 SW |
1124 | struct buffer_head *bh, *head; |
1125 | struct gfs2_bufdata *bd; | |
4340fe62 SW |
1126 | |
1127 | if (!page_has_buffers(page)) | |
891ba6d4 | 1128 | return 0; |
4340fe62 | 1129 | |
bb3b0e3d | 1130 | gfs2_log_lock(sdp); |
380f7c65 | 1131 | spin_lock(&sdp->sd_ail_lock); |
4340fe62 SW |
1132 | head = bh = page_buffers(page); |
1133 | do { | |
bb3b0e3d SW |
1134 | if (atomic_read(&bh->b_count)) |
1135 | goto cannot_release; | |
1136 | bd = bh->b_private; | |
16ca9412 | 1137 | if (bd && bd->bd_tr) |
bb3b0e3d | 1138 | goto cannot_release; |
8f065d36 SW |
1139 | if (buffer_pinned(bh) || buffer_dirty(bh)) |
1140 | goto not_possible; | |
bb3b0e3d SW |
1141 | bh = bh->b_this_page; |
1142 | } while(bh != head); | |
380f7c65 | 1143 | spin_unlock(&sdp->sd_ail_lock); |
4340fe62 | 1144 | |
bb3b0e3d SW |
1145 | head = bh = page_buffers(page); |
1146 | do { | |
4340fe62 SW |
1147 | bd = bh->b_private; |
1148 | if (bd) { | |
1149 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
e4f29206 SW |
1150 | if (!list_empty(&bd->bd_list)) |
1151 | list_del_init(&bd->bd_list); | |
1152 | bd->bd_bh = NULL; | |
4340fe62 | 1153 | bh->b_private = NULL; |
623d9355 | 1154 | kmem_cache_free(gfs2_bufdata_cachep, bd); |
e4f29206 | 1155 | } |
4340fe62 SW |
1156 | |
1157 | bh = bh->b_this_page; | |
166afccd | 1158 | } while (bh != head); |
e4f29206 | 1159 | gfs2_log_unlock(sdp); |
4340fe62 | 1160 | |
4340fe62 | 1161 | return try_to_free_buffers(page); |
8f065d36 SW |
1162 | |
1163 | not_possible: /* Should never happen */ | |
1164 | WARN_ON(buffer_dirty(bh)); | |
1165 | WARN_ON(buffer_pinned(bh)); | |
bb3b0e3d | 1166 | cannot_release: |
380f7c65 | 1167 | spin_unlock(&sdp->sd_ail_lock); |
bb3b0e3d SW |
1168 | gfs2_log_unlock(sdp); |
1169 | return 0; | |
4340fe62 SW |
1170 | } |
1171 | ||
5561093e | 1172 | static const struct address_space_operations gfs2_writeback_aops = { |
9d358143 | 1173 | .writepage = gfs2_writepage, |
45138990 | 1174 | .writepages = gfs2_writepages, |
5561093e SW |
1175 | .readpage = gfs2_readpage, |
1176 | .readpages = gfs2_readpages, | |
5561093e SW |
1177 | .write_begin = gfs2_write_begin, |
1178 | .write_end = gfs2_write_end, | |
1179 | .bmap = gfs2_bmap, | |
1180 | .invalidatepage = gfs2_invalidatepage, | |
1181 | .releasepage = gfs2_releasepage, | |
1182 | .direct_IO = gfs2_direct_IO, | |
e5d9dc27 | 1183 | .migratepage = buffer_migrate_page, |
229615de | 1184 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1185 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
1186 | }; |
1187 | ||
1188 | static const struct address_space_operations gfs2_ordered_aops = { | |
9d358143 | 1189 | .writepage = gfs2_writepage, |
45138990 | 1190 | .writepages = gfs2_writepages, |
b3b94faa | 1191 | .readpage = gfs2_readpage, |
fd88de56 | 1192 | .readpages = gfs2_readpages, |
7765ec26 SW |
1193 | .write_begin = gfs2_write_begin, |
1194 | .write_end = gfs2_write_end, | |
8fb68595 | 1195 | .set_page_dirty = gfs2_set_page_dirty, |
b3b94faa DT |
1196 | .bmap = gfs2_bmap, |
1197 | .invalidatepage = gfs2_invalidatepage, | |
4340fe62 | 1198 | .releasepage = gfs2_releasepage, |
b3b94faa | 1199 | .direct_IO = gfs2_direct_IO, |
e5d9dc27 | 1200 | .migratepage = buffer_migrate_page, |
229615de | 1201 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1202 | .error_remove_page = generic_error_remove_page, |
b3b94faa DT |
1203 | }; |
1204 | ||
5561093e | 1205 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 1206 | .writepage = gfs2_jdata_writepage, |
b8e7cbb6 | 1207 | .writepages = gfs2_jdata_writepages, |
5561093e SW |
1208 | .readpage = gfs2_readpage, |
1209 | .readpages = gfs2_readpages, | |
5561093e SW |
1210 | .write_begin = gfs2_write_begin, |
1211 | .write_end = gfs2_write_end, | |
1212 | .set_page_dirty = gfs2_set_page_dirty, | |
1213 | .bmap = gfs2_bmap, | |
1214 | .invalidatepage = gfs2_invalidatepage, | |
1215 | .releasepage = gfs2_releasepage, | |
229615de | 1216 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 1217 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
1218 | }; |
1219 | ||
1220 | void gfs2_set_aops(struct inode *inode) | |
1221 | { | |
1222 | struct gfs2_inode *ip = GFS2_I(inode); | |
1223 | ||
1224 | if (gfs2_is_writeback(ip)) | |
1225 | inode->i_mapping->a_ops = &gfs2_writeback_aops; | |
1226 | else if (gfs2_is_ordered(ip)) | |
1227 | inode->i_mapping->a_ops = &gfs2_ordered_aops; | |
1228 | else if (gfs2_is_jdata(ip)) | |
1229 | inode->i_mapping->a_ops = &gfs2_jdata_aops; | |
1230 | else | |
1231 | BUG(); | |
1232 | } | |
1233 |