Commit | Line | Data |
---|---|---|
7336d0e6 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
b3b94faa DT |
2 | /* |
3 | * Copyright (C) Sistina Software, Inc. 1997-2003 All rights reserved. | |
7eabb77e | 4 | * Copyright (C) 2004-2008 Red Hat, Inc. All rights reserved. |
b3b94faa DT |
5 | */ |
6 | ||
7 | #include <linux/sched.h> | |
8 | #include <linux/slab.h> | |
9 | #include <linux/spinlock.h> | |
10 | #include <linux/completion.h> | |
11 | #include <linux/buffer_head.h> | |
12 | #include <linux/pagemap.h> | |
fd88de56 | 13 | #include <linux/pagevec.h> |
9b124fbb | 14 | #include <linux/mpage.h> |
d1665e41 | 15 | #include <linux/fs.h> |
a8d638e3 | 16 | #include <linux/writeback.h> |
7765ec26 | 17 | #include <linux/swap.h> |
5c676f6d | 18 | #include <linux/gfs2_ondisk.h> |
47e83b50 | 19 | #include <linux/backing-dev.h> |
e2e40f2c | 20 | #include <linux/uio.h> |
774016b2 | 21 | #include <trace/events/writeback.h> |
64bc06bb | 22 | #include <linux/sched/signal.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" |
64bc06bb | 37 | #include "aops.h" |
b3b94faa | 38 | |
ba7f7290 | 39 | |
64bc06bb AG |
40 | void gfs2_page_add_databufs(struct gfs2_inode *ip, struct page *page, |
41 | unsigned int from, unsigned int len) | |
ba7f7290 SW |
42 | { |
43 | struct buffer_head *head = page_buffers(page); | |
44 | unsigned int bsize = head->b_size; | |
45 | struct buffer_head *bh; | |
88b65ce5 | 46 | unsigned int to = from + len; |
ba7f7290 SW |
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; | |
88b65ce5 | 52 | if (end <= from) |
ba7f7290 | 53 | continue; |
88b65ce5 AG |
54 | if (start >= to) |
55 | break; | |
845802b1 | 56 | set_buffer_uptodate(bh); |
350a9b0a | 57 | gfs2_trans_add_data(ip->i_gl, bh); |
ba7f7290 SW |
58 | } |
59 | } | |
60 | ||
b3b94faa | 61 | /** |
7a6bbacb | 62 | * gfs2_get_block_noalloc - Fills in a buffer head with details about a block |
b3b94faa DT |
63 | * @inode: The inode |
64 | * @lblock: The block number to look up | |
65 | * @bh_result: The buffer head to return the result in | |
66 | * @create: Non-zero if we may add block to the file | |
67 | * | |
68 | * Returns: errno | |
69 | */ | |
70 | ||
7a6bbacb SW |
71 | static int gfs2_get_block_noalloc(struct inode *inode, sector_t lblock, |
72 | struct buffer_head *bh_result, int create) | |
b3b94faa | 73 | { |
b3b94faa DT |
74 | int error; |
75 | ||
e9e1ef2b | 76 | error = gfs2_block_map(inode, lblock, bh_result, 0); |
b3b94faa DT |
77 | if (error) |
78 | return error; | |
de986e85 | 79 | if (!buffer_mapped(bh_result)) |
7a6bbacb SW |
80 | return -EIO; |
81 | return 0; | |
b3b94faa DT |
82 | } |
83 | ||
b3b94faa | 84 | /** |
59c01c50 CH |
85 | * gfs2_writepage - Write page for writeback mappings |
86 | * @page: The page | |
9ff8ec32 | 87 | * @wbc: The writeback control |
b3b94faa | 88 | */ |
59c01c50 | 89 | static int gfs2_writepage(struct page *page, struct writeback_control *wbc) |
b3b94faa | 90 | { |
18ec7d5c | 91 | struct inode *inode = page->mapping->host; |
f4387149 SW |
92 | struct gfs2_inode *ip = GFS2_I(inode); |
93 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
18ec7d5c | 94 | loff_t i_size = i_size_read(inode); |
09cbfeaf | 95 | pgoff_t end_index = i_size >> PAGE_SHIFT; |
18ec7d5c | 96 | unsigned offset; |
b3b94faa | 97 | |
9ff8ec32 SW |
98 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
99 | goto out; | |
5c676f6d | 100 | if (current->journal_info) |
9ff8ec32 | 101 | goto redirty; |
18ec7d5c | 102 | /* Is the page fully outside i_size? (truncate in progress) */ |
09cbfeaf | 103 | offset = i_size & (PAGE_SIZE-1); |
d2d7b8a2 | 104 | if (page->index > end_index || (page->index == end_index && !offset)) { |
09cbfeaf | 105 | page->mapping->a_ops->invalidatepage(page, 0, PAGE_SIZE); |
9ff8ec32 | 106 | goto out; |
b3b94faa | 107 | } |
59c01c50 CH |
108 | |
109 | return nobh_writepage(page, gfs2_get_block_noalloc, wbc); | |
110 | ||
9ff8ec32 SW |
111 | redirty: |
112 | redirty_page_for_writepage(wbc, page); | |
113 | out: | |
114 | unlock_page(page); | |
115 | return 0; | |
116 | } | |
117 | ||
fd4c5748 BM |
118 | /* This is the same as calling block_write_full_page, but it also |
119 | * writes pages outside of i_size | |
120 | */ | |
c548a1c1 AP |
121 | static int gfs2_write_full_page(struct page *page, get_block_t *get_block, |
122 | struct writeback_control *wbc) | |
fd4c5748 BM |
123 | { |
124 | struct inode * const inode = page->mapping->host; | |
125 | loff_t i_size = i_size_read(inode); | |
126 | const pgoff_t end_index = i_size >> PAGE_SHIFT; | |
127 | unsigned offset; | |
128 | ||
129 | /* | |
130 | * The page straddles i_size. It must be zeroed out on each and every | |
131 | * writepage invocation because it may be mmapped. "A file is mapped | |
132 | * in multiples of the page size. For a file that is not a multiple of | |
133 | * the page size, the remaining memory is zeroed when mapped, and | |
134 | * writes to that region are not written out to the file." | |
135 | */ | |
f3b64b57 | 136 | offset = i_size & (PAGE_SIZE - 1); |
fd4c5748 BM |
137 | if (page->index == end_index && offset) |
138 | zero_user_segment(page, offset, PAGE_SIZE); | |
139 | ||
140 | return __block_write_full_page(inode, page, get_block, wbc, | |
141 | end_buffer_async_write); | |
142 | } | |
143 | ||
b8e7cbb6 SW |
144 | /** |
145 | * __gfs2_jdata_writepage - The core of jdata writepage | |
146 | * @page: The page to write | |
147 | * @wbc: The writeback control | |
148 | * | |
149 | * This is shared between writepage and writepages and implements the | |
150 | * core of the writepage operation. If a transaction is required then | |
151 | * PageChecked will have been set and the transaction will have | |
152 | * already been started before this is called. | |
153 | */ | |
154 | ||
155 | static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
156 | { | |
157 | struct inode *inode = page->mapping->host; | |
158 | struct gfs2_inode *ip = GFS2_I(inode); | |
159 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
160 | ||
161 | if (PageChecked(page)) { | |
162 | ClearPageChecked(page); | |
163 | if (!page_has_buffers(page)) { | |
164 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
47a9a527 | 165 | BIT(BH_Dirty)|BIT(BH_Uptodate)); |
b8e7cbb6 | 166 | } |
88b65ce5 | 167 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize); |
b8e7cbb6 | 168 | } |
fd4c5748 | 169 | return gfs2_write_full_page(page, gfs2_get_block_noalloc, wbc); |
b8e7cbb6 SW |
170 | } |
171 | ||
9ff8ec32 SW |
172 | /** |
173 | * gfs2_jdata_writepage - Write complete page | |
174 | * @page: Page to write | |
1272574b | 175 | * @wbc: The writeback control |
9ff8ec32 SW |
176 | * |
177 | * Returns: errno | |
178 | * | |
179 | */ | |
180 | ||
181 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
182 | { | |
183 | struct inode *inode = page->mapping->host; | |
fd4c5748 | 184 | struct gfs2_inode *ip = GFS2_I(inode); |
9ff8ec32 | 185 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9ff8ec32 | 186 | |
fd4c5748 BM |
187 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
188 | goto out; | |
189 | if (PageChecked(page) || current->journal_info) | |
190 | goto out_ignore; | |
e556280d | 191 | return __gfs2_jdata_writepage(page, wbc); |
18ec7d5c SW |
192 | |
193 | out_ignore: | |
194 | redirty_page_for_writepage(wbc, page); | |
fd4c5748 | 195 | out: |
18ec7d5c SW |
196 | unlock_page(page); |
197 | return 0; | |
b3b94faa DT |
198 | } |
199 | ||
a8d638e3 | 200 | /** |
45138990 | 201 | * gfs2_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
202 | * @mapping: The mapping to write |
203 | * @wbc: Write-back control | |
204 | * | |
45138990 | 205 | * Used for both ordered and writeback modes. |
a8d638e3 | 206 | */ |
45138990 SW |
207 | static int gfs2_writepages(struct address_space *mapping, |
208 | struct writeback_control *wbc) | |
a8d638e3 | 209 | { |
b066a4ee AD |
210 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); |
211 | int ret = mpage_writepages(mapping, wbc, gfs2_get_block_noalloc); | |
212 | ||
213 | /* | |
214 | * Even if we didn't write any pages here, we might still be holding | |
215 | * dirty pages in the ail. We forcibly flush the ail because we don't | |
216 | * want balance_dirty_pages() to loop indefinitely trying to write out | |
217 | * pages held in the ail that it can't find. | |
218 | */ | |
219 | if (ret == 0) | |
220 | set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags); | |
221 | ||
222 | return ret; | |
a8d638e3 SW |
223 | } |
224 | ||
b8e7cbb6 SW |
225 | /** |
226 | * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages | |
227 | * @mapping: The mapping | |
228 | * @wbc: The writeback control | |
b8e7cbb6 SW |
229 | * @pvec: The vector of pages |
230 | * @nr_pages: The number of pages to write | |
1272574b | 231 | * @done_index: Page index |
b8e7cbb6 SW |
232 | * |
233 | * Returns: non-zero if loop should terminate, zero otherwise | |
234 | */ | |
235 | ||
236 | static int gfs2_write_jdata_pagevec(struct address_space *mapping, | |
237 | struct writeback_control *wbc, | |
238 | struct pagevec *pvec, | |
9aa01593 | 239 | int nr_pages, |
774016b2 | 240 | pgoff_t *done_index) |
b8e7cbb6 SW |
241 | { |
242 | struct inode *inode = mapping->host; | |
243 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
45eb0504 | 244 | unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits); |
b8e7cbb6 SW |
245 | int i; |
246 | int ret; | |
247 | ||
20b95bf2 | 248 | ret = gfs2_trans_begin(sdp, nrblocks, nrblocks); |
b8e7cbb6 SW |
249 | if (ret < 0) |
250 | return ret; | |
251 | ||
252 | for(i = 0; i < nr_pages; i++) { | |
253 | struct page *page = pvec->pages[i]; | |
254 | ||
774016b2 SW |
255 | *done_index = page->index; |
256 | ||
b8e7cbb6 SW |
257 | lock_page(page); |
258 | ||
259 | if (unlikely(page->mapping != mapping)) { | |
774016b2 | 260 | continue_unlock: |
b8e7cbb6 SW |
261 | unlock_page(page); |
262 | continue; | |
263 | } | |
264 | ||
774016b2 SW |
265 | if (!PageDirty(page)) { |
266 | /* someone wrote it for us */ | |
267 | goto continue_unlock; | |
b8e7cbb6 SW |
268 | } |
269 | ||
774016b2 SW |
270 | if (PageWriteback(page)) { |
271 | if (wbc->sync_mode != WB_SYNC_NONE) | |
272 | wait_on_page_writeback(page); | |
273 | else | |
274 | goto continue_unlock; | |
b8e7cbb6 SW |
275 | } |
276 | ||
774016b2 SW |
277 | BUG_ON(PageWriteback(page)); |
278 | if (!clear_page_dirty_for_io(page)) | |
279 | goto continue_unlock; | |
280 | ||
de1414a6 | 281 | trace_wbc_writepage(wbc, inode_to_bdi(inode)); |
b8e7cbb6 SW |
282 | |
283 | ret = __gfs2_jdata_writepage(page, wbc); | |
774016b2 SW |
284 | if (unlikely(ret)) { |
285 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
286 | unlock_page(page); | |
287 | ret = 0; | |
288 | } else { | |
289 | ||
290 | /* | |
291 | * done_index is set past this page, | |
292 | * so media errors will not choke | |
293 | * background writeout for the entire | |
294 | * file. This has consequences for | |
295 | * range_cyclic semantics (ie. it may | |
296 | * not be suitable for data integrity | |
297 | * writeout). | |
298 | */ | |
299 | *done_index = page->index + 1; | |
300 | ret = 1; | |
301 | break; | |
302 | } | |
303 | } | |
b8e7cbb6 | 304 | |
774016b2 SW |
305 | /* |
306 | * We stop writing back only if we are not doing | |
307 | * integrity sync. In case of integrity sync we have to | |
308 | * keep going until we have written all the pages | |
309 | * we tagged for writeback prior to entering this loop. | |
310 | */ | |
311 | if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) { | |
b8e7cbb6 | 312 | ret = 1; |
774016b2 SW |
313 | break; |
314 | } | |
315 | ||
b8e7cbb6 SW |
316 | } |
317 | gfs2_trans_end(sdp); | |
318 | return ret; | |
319 | } | |
320 | ||
321 | /** | |
322 | * gfs2_write_cache_jdata - Like write_cache_pages but different | |
323 | * @mapping: The mapping to write | |
324 | * @wbc: The writeback control | |
b8e7cbb6 SW |
325 | * |
326 | * The reason that we use our own function here is that we need to | |
327 | * start transactions before we grab page locks. This allows us | |
328 | * to get the ordering right. | |
329 | */ | |
330 | ||
331 | static int gfs2_write_cache_jdata(struct address_space *mapping, | |
332 | struct writeback_control *wbc) | |
333 | { | |
b8e7cbb6 SW |
334 | int ret = 0; |
335 | int done = 0; | |
336 | struct pagevec pvec; | |
337 | int nr_pages; | |
774016b2 | 338 | pgoff_t uninitialized_var(writeback_index); |
b8e7cbb6 SW |
339 | pgoff_t index; |
340 | pgoff_t end; | |
774016b2 SW |
341 | pgoff_t done_index; |
342 | int cycled; | |
b8e7cbb6 | 343 | int range_whole = 0; |
10bbd235 | 344 | xa_mark_t tag; |
b8e7cbb6 | 345 | |
86679820 | 346 | pagevec_init(&pvec); |
b8e7cbb6 | 347 | if (wbc->range_cyclic) { |
774016b2 SW |
348 | writeback_index = mapping->writeback_index; /* prev offset */ |
349 | index = writeback_index; | |
350 | if (index == 0) | |
351 | cycled = 1; | |
352 | else | |
353 | cycled = 0; | |
b8e7cbb6 SW |
354 | end = -1; |
355 | } else { | |
09cbfeaf KS |
356 | index = wbc->range_start >> PAGE_SHIFT; |
357 | end = wbc->range_end >> PAGE_SHIFT; | |
b8e7cbb6 SW |
358 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
359 | range_whole = 1; | |
774016b2 | 360 | cycled = 1; /* ignore range_cyclic tests */ |
b8e7cbb6 | 361 | } |
774016b2 SW |
362 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
363 | tag = PAGECACHE_TAG_TOWRITE; | |
364 | else | |
365 | tag = PAGECACHE_TAG_DIRTY; | |
b8e7cbb6 SW |
366 | |
367 | retry: | |
774016b2 SW |
368 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
369 | tag_pages_for_writeback(mapping, index, end); | |
370 | done_index = index; | |
371 | while (!done && (index <= end)) { | |
d2bc5b3c | 372 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 373 | tag); |
774016b2 SW |
374 | if (nr_pages == 0) |
375 | break; | |
376 | ||
9aa01593 | 377 | ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index); |
b8e7cbb6 SW |
378 | if (ret) |
379 | done = 1; | |
380 | if (ret > 0) | |
381 | ret = 0; | |
b8e7cbb6 SW |
382 | pagevec_release(&pvec); |
383 | cond_resched(); | |
384 | } | |
385 | ||
774016b2 | 386 | if (!cycled && !done) { |
b8e7cbb6 | 387 | /* |
774016b2 | 388 | * range_cyclic: |
b8e7cbb6 SW |
389 | * We hit the last page and there is more work to be done: wrap |
390 | * back to the start of the file | |
391 | */ | |
774016b2 | 392 | cycled = 1; |
b8e7cbb6 | 393 | index = 0; |
774016b2 | 394 | end = writeback_index - 1; |
b8e7cbb6 SW |
395 | goto retry; |
396 | } | |
397 | ||
398 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
774016b2 SW |
399 | mapping->writeback_index = done_index; |
400 | ||
b8e7cbb6 SW |
401 | return ret; |
402 | } | |
403 | ||
404 | ||
405 | /** | |
406 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk | |
407 | * @mapping: The mapping to write | |
408 | * @wbc: The writeback control | |
409 | * | |
410 | */ | |
411 | ||
412 | static int gfs2_jdata_writepages(struct address_space *mapping, | |
413 | struct writeback_control *wbc) | |
414 | { | |
415 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
416 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
417 | int ret; | |
418 | ||
419 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
420 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { | |
805c0907 BP |
421 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | |
422 | GFS2_LFC_JDATA_WPAGES); | |
b8e7cbb6 SW |
423 | ret = gfs2_write_cache_jdata(mapping, wbc); |
424 | } | |
425 | return ret; | |
426 | } | |
427 | ||
b3b94faa DT |
428 | /** |
429 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
430 | * @ip: the inode | |
431 | * @page: the page | |
432 | * | |
433 | * Returns: errno | |
434 | */ | |
378b6cbf | 435 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
436 | { |
437 | struct buffer_head *dibh; | |
602c89d2 | 438 | u64 dsize = i_size_read(&ip->i_inode); |
b3b94faa DT |
439 | void *kaddr; |
440 | int error; | |
441 | ||
bf126aee | 442 | /* |
3c18ddd1 | 443 | * Due to the order of unstuffing files and ->fault(), we can be |
bf126aee SW |
444 | * asked for a zero page in the case of a stuffed file being extended, |
445 | * so we need to supply one here. It doesn't happen often. | |
446 | */ | |
447 | if (unlikely(page->index)) { | |
09cbfeaf | 448 | zero_user(page, 0, PAGE_SIZE); |
0a7ab79c | 449 | SetPageUptodate(page); |
bf126aee SW |
450 | return 0; |
451 | } | |
fd88de56 | 452 | |
b3b94faa DT |
453 | error = gfs2_meta_inode_buffer(ip, &dibh); |
454 | if (error) | |
455 | return error; | |
456 | ||
d9349285 | 457 | kaddr = kmap_atomic(page); |
235628c5 AG |
458 | if (dsize > gfs2_max_stuffed_size(ip)) |
459 | dsize = gfs2_max_stuffed_size(ip); | |
602c89d2 | 460 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 461 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
d9349285 | 462 | kunmap_atomic(kaddr); |
bf126aee | 463 | flush_dcache_page(page); |
b3b94faa | 464 | brelse(dibh); |
b3b94faa DT |
465 | SetPageUptodate(page); |
466 | ||
467 | return 0; | |
468 | } | |
469 | ||
b3b94faa | 470 | |
b3b94faa | 471 | /** |
51ff87bd SW |
472 | * __gfs2_readpage - readpage |
473 | * @file: The file to read a page for | |
b3b94faa DT |
474 | * @page: The page to read |
475 | * | |
9db115a0 AG |
476 | * This is the core of gfs2's readpage. It's used by the internal file |
477 | * reading code as in that case we already hold the glock. Also it's | |
51ff87bd | 478 | * called by gfs2_readpage() once the required lock has been granted. |
b3b94faa DT |
479 | */ |
480 | ||
51ff87bd | 481 | static int __gfs2_readpage(void *file, struct page *page) |
b3b94faa | 482 | { |
feaa7bba SW |
483 | struct gfs2_inode *ip = GFS2_I(page->mapping->host); |
484 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); | |
f95cbb44 | 485 | |
b3b94faa DT |
486 | int error; |
487 | ||
f95cbb44 AG |
488 | if (i_blocksize(page->mapping->host) == PAGE_SIZE && |
489 | !page_has_buffers(page)) { | |
490 | error = iomap_readpage(page, &gfs2_iomap_ops); | |
491 | } else if (gfs2_is_stuffed(ip)) { | |
fd88de56 SW |
492 | error = stuffed_readpage(ip, page); |
493 | unlock_page(page); | |
51ff87bd | 494 | } else { |
e9e1ef2b | 495 | error = mpage_readpage(page, gfs2_block_map); |
51ff87bd | 496 | } |
b3b94faa | 497 | |
eb43e660 | 498 | if (unlikely(gfs2_withdrawn(sdp))) |
51ff87bd | 499 | return -EIO; |
b3b94faa | 500 | |
51ff87bd SW |
501 | return error; |
502 | } | |
503 | ||
504 | /** | |
505 | * gfs2_readpage - read a page of a file | |
506 | * @file: The file to read | |
507 | * @page: The page of the file | |
508 | * | |
01b7c7ae SW |
509 | * This deals with the locking required. We have to unlock and |
510 | * relock the page in order to get the locking in the right | |
511 | * order. | |
51ff87bd SW |
512 | */ |
513 | ||
514 | static int gfs2_readpage(struct file *file, struct page *page) | |
515 | { | |
01b7c7ae SW |
516 | struct address_space *mapping = page->mapping; |
517 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
6802e340 | 518 | struct gfs2_holder gh; |
51ff87bd SW |
519 | int error; |
520 | ||
01b7c7ae | 521 | unlock_page(page); |
719ee344 SW |
522 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
523 | error = gfs2_glock_nq(&gh); | |
01b7c7ae | 524 | if (unlikely(error)) |
6802e340 | 525 | goto out; |
01b7c7ae SW |
526 | error = AOP_TRUNCATED_PAGE; |
527 | lock_page(page); | |
528 | if (page->mapping == mapping && !PageUptodate(page)) | |
529 | error = __gfs2_readpage(file, page); | |
530 | else | |
531 | unlock_page(page); | |
6802e340 | 532 | gfs2_glock_dq(&gh); |
18ec7d5c | 533 | out: |
6802e340 | 534 | gfs2_holder_uninit(&gh); |
01b7c7ae SW |
535 | if (error && error != AOP_TRUNCATED_PAGE) |
536 | lock_page(page); | |
51ff87bd SW |
537 | return error; |
538 | } | |
539 | ||
540 | /** | |
541 | * gfs2_internal_read - read an internal file | |
542 | * @ip: The gfs2 inode | |
51ff87bd SW |
543 | * @buf: The buffer to fill |
544 | * @pos: The file position | |
545 | * @size: The amount to read | |
546 | * | |
547 | */ | |
548 | ||
4306629e AP |
549 | int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos, |
550 | unsigned size) | |
51ff87bd SW |
551 | { |
552 | struct address_space *mapping = ip->i_inode.i_mapping; | |
45eb0504 | 553 | unsigned long index = *pos >> PAGE_SHIFT; |
09cbfeaf | 554 | unsigned offset = *pos & (PAGE_SIZE - 1); |
51ff87bd SW |
555 | unsigned copied = 0; |
556 | unsigned amt; | |
557 | struct page *page; | |
558 | void *p; | |
559 | ||
560 | do { | |
561 | amt = size - copied; | |
09cbfeaf KS |
562 | if (offset + size > PAGE_SIZE) |
563 | amt = PAGE_SIZE - offset; | |
51ff87bd SW |
564 | page = read_cache_page(mapping, index, __gfs2_readpage, NULL); |
565 | if (IS_ERR(page)) | |
566 | return PTR_ERR(page); | |
d9349285 | 567 | p = kmap_atomic(page); |
51ff87bd | 568 | memcpy(buf + copied, p + offset, amt); |
d9349285 | 569 | kunmap_atomic(p); |
09cbfeaf | 570 | put_page(page); |
51ff87bd SW |
571 | copied += amt; |
572 | index++; | |
573 | offset = 0; | |
574 | } while(copied < size); | |
575 | (*pos) += size; | |
576 | return size; | |
fd88de56 SW |
577 | } |
578 | ||
fd88de56 | 579 | /** |
d4388340 | 580 | * gfs2_readahead - Read a bunch of pages at once |
1272574b FF |
581 | * @file: The file to read from |
582 | * @mapping: Address space info | |
583 | * @pages: List of pages to read | |
584 | * @nr_pages: Number of pages to read | |
fd88de56 SW |
585 | * |
586 | * Some notes: | |
587 | * 1. This is only for readahead, so we can simply ignore any things | |
588 | * which are slightly inconvenient (such as locking conflicts between | |
589 | * the page lock and the glock) and return having done no I/O. Its | |
590 | * obviously not something we'd want to do on too regular a basis. | |
591 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 592 | * 2. We don't handle stuffed files here we let readpage do the honours. |
d4388340 | 593 | * 3. mpage_readahead() does most of the heavy lifting in the common case. |
e9e1ef2b | 594 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
fd88de56 | 595 | */ |
3cc3f710 | 596 | |
d4388340 | 597 | static void gfs2_readahead(struct readahead_control *rac) |
fd88de56 | 598 | { |
d4388340 | 599 | struct inode *inode = rac->mapping->host; |
feaa7bba | 600 | struct gfs2_inode *ip = GFS2_I(inode); |
fd88de56 | 601 | struct gfs2_holder gh; |
fd88de56 | 602 | |
719ee344 | 603 | gfs2_holder_init(ip->i_gl, LM_ST_SHARED, 0, &gh); |
d4388340 | 604 | if (gfs2_glock_nq(&gh)) |
3cc3f710 | 605 | goto out_uninit; |
e1d5b18a | 606 | if (!gfs2_is_stuffed(ip)) |
d4388340 | 607 | mpage_readahead(rac, gfs2_block_map); |
3cc3f710 SW |
608 | gfs2_glock_dq(&gh); |
609 | out_uninit: | |
610 | gfs2_holder_uninit(&gh); | |
b3b94faa DT |
611 | } |
612 | ||
7ae8fa84 RP |
613 | /** |
614 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
615 | * @inode: the rindex inode | |
616 | */ | |
64bc06bb | 617 | void adjust_fs_space(struct inode *inode) |
7ae8fa84 | 618 | { |
d0a22a4b | 619 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1946f70a BM |
620 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
621 | struct gfs2_inode *l_ip = GFS2_I(sdp->sd_sc_inode); | |
7ae8fa84 RP |
622 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
623 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
1946f70a | 624 | struct buffer_head *m_bh, *l_bh; |
7ae8fa84 RP |
625 | u64 fs_total, new_free; |
626 | ||
d0a22a4b AG |
627 | if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0) |
628 | return; | |
629 | ||
7ae8fa84 RP |
630 | /* Total up the file system space, according to the latest rindex. */ |
631 | fs_total = gfs2_ri_total(sdp); | |
1946f70a | 632 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) |
d0a22a4b | 633 | goto out; |
7ae8fa84 RP |
634 | |
635 | spin_lock(&sdp->sd_statfs_spin); | |
1946f70a BM |
636 | gfs2_statfs_change_in(m_sc, m_bh->b_data + |
637 | sizeof(struct gfs2_dinode)); | |
7ae8fa84 RP |
638 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) |
639 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
640 | else | |
641 | new_free = 0; | |
642 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
643 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
644 | (unsigned long long)new_free); | |
7ae8fa84 | 645 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
1946f70a BM |
646 | |
647 | if (gfs2_meta_inode_buffer(l_ip, &l_bh) != 0) | |
d0a22a4b | 648 | goto out2; |
1946f70a BM |
649 | update_statfs(sdp, m_bh, l_bh); |
650 | brelse(l_bh); | |
d0a22a4b | 651 | out2: |
1946f70a | 652 | brelse(m_bh); |
d0a22a4b AG |
653 | out: |
654 | sdp->sd_rindex_uptodate = 0; | |
655 | gfs2_trans_end(sdp); | |
7ae8fa84 RP |
656 | } |
657 | ||
8fb68595 | 658 | /** |
b9e03f18 | 659 | * jdata_set_page_dirty - Page dirtying function |
8fb68595 RP |
660 | * @page: The page to dirty |
661 | * | |
662 | * Returns: 1 if it dirtyed the page, or 0 otherwise | |
663 | */ | |
664 | ||
b9e03f18 | 665 | static int jdata_set_page_dirty(struct page *page) |
8fb68595 | 666 | { |
5561093e | 667 | SetPageChecked(page); |
8fb68595 RP |
668 | return __set_page_dirty_buffers(page); |
669 | } | |
670 | ||
b3b94faa DT |
671 | /** |
672 | * gfs2_bmap - Block map function | |
673 | * @mapping: Address space info | |
674 | * @lblock: The block to map | |
675 | * | |
676 | * Returns: The disk address for the block or 0 on hole or error | |
677 | */ | |
678 | ||
679 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
680 | { | |
feaa7bba | 681 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
682 | struct gfs2_holder i_gh; |
683 | sector_t dblock = 0; | |
684 | int error; | |
685 | ||
b3b94faa DT |
686 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
687 | if (error) | |
688 | return 0; | |
689 | ||
690 | if (!gfs2_is_stuffed(ip)) | |
7770c93a | 691 | dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops); |
b3b94faa DT |
692 | |
693 | gfs2_glock_dq_uninit(&i_gh); | |
694 | ||
695 | return dblock; | |
696 | } | |
697 | ||
d7b616e2 SW |
698 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
699 | { | |
700 | struct gfs2_bufdata *bd; | |
701 | ||
702 | lock_buffer(bh); | |
703 | gfs2_log_lock(sdp); | |
704 | clear_buffer_dirty(bh); | |
705 | bd = bh->b_private; | |
706 | if (bd) { | |
c0752aa7 BP |
707 | if (!list_empty(&bd->bd_list) && !buffer_pinned(bh)) |
708 | list_del_init(&bd->bd_list); | |
16615be1 | 709 | else |
68cd4ce2 | 710 | gfs2_remove_from_journal(bh, REMOVE_JDATA); |
d7b616e2 SW |
711 | } |
712 | bh->b_bdev = NULL; | |
713 | clear_buffer_mapped(bh); | |
714 | clear_buffer_req(bh); | |
715 | clear_buffer_new(bh); | |
716 | gfs2_log_unlock(sdp); | |
717 | unlock_buffer(bh); | |
718 | } | |
719 | ||
d47992f8 LC |
720 | static void gfs2_invalidatepage(struct page *page, unsigned int offset, |
721 | unsigned int length) | |
b3b94faa | 722 | { |
d7b616e2 | 723 | struct gfs2_sbd *sdp = GFS2_SB(page->mapping->host); |
5c0bb97c | 724 | unsigned int stop = offset + length; |
09cbfeaf | 725 | int partial_page = (offset || length < PAGE_SIZE); |
d7b616e2 SW |
726 | struct buffer_head *bh, *head; |
727 | unsigned long pos = 0; | |
728 | ||
b3b94faa | 729 | BUG_ON(!PageLocked(page)); |
5c0bb97c | 730 | if (!partial_page) |
8fb68595 | 731 | ClearPageChecked(page); |
d7b616e2 SW |
732 | if (!page_has_buffers(page)) |
733 | goto out; | |
b3b94faa | 734 | |
d7b616e2 SW |
735 | bh = head = page_buffers(page); |
736 | do { | |
5c0bb97c LC |
737 | if (pos + bh->b_size > stop) |
738 | return; | |
739 | ||
d7b616e2 SW |
740 | if (offset <= pos) |
741 | gfs2_discard(sdp, bh); | |
742 | pos += bh->b_size; | |
743 | bh = bh->b_this_page; | |
744 | } while (bh != head); | |
745 | out: | |
5c0bb97c | 746 | if (!partial_page) |
d7b616e2 | 747 | try_to_release_page(page, 0); |
b3b94faa DT |
748 | } |
749 | ||
4340fe62 | 750 | /** |
623d9355 | 751 | * gfs2_releasepage - free the metadata associated with a page |
4340fe62 SW |
752 | * @page: the page that's being released |
753 | * @gfp_mask: passed from Linux VFS, ignored by us | |
754 | * | |
0ebbe4f9 AG |
755 | * Calls try_to_free_buffers() to free the buffers and put the page if the |
756 | * buffers can be released. | |
4340fe62 | 757 | * |
0ebbe4f9 | 758 | * Returns: 1 if the page was put or else 0 |
4340fe62 SW |
759 | */ |
760 | ||
761 | int gfs2_releasepage(struct page *page, gfp_t gfp_mask) | |
762 | { | |
009d8518 SW |
763 | struct address_space *mapping = page->mapping; |
764 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); | |
4340fe62 SW |
765 | struct buffer_head *bh, *head; |
766 | struct gfs2_bufdata *bd; | |
4340fe62 SW |
767 | |
768 | if (!page_has_buffers(page)) | |
891ba6d4 | 769 | return 0; |
4340fe62 | 770 | |
1c185c02 AG |
771 | /* |
772 | * From xfs_vm_releasepage: mm accommodates an old ext3 case where | |
773 | * clean pages might not have had the dirty bit cleared. Thus, it can | |
774 | * send actual dirty pages to ->releasepage() via shrink_active_list(). | |
775 | * | |
776 | * As a workaround, we skip pages that contain dirty buffers below. | |
777 | * Once ->releasepage isn't called on dirty pages anymore, we can warn | |
778 | * on dirty buffers like we used to here again. | |
779 | */ | |
780 | ||
bb3b0e3d | 781 | gfs2_log_lock(sdp); |
380f7c65 | 782 | spin_lock(&sdp->sd_ail_lock); |
4340fe62 SW |
783 | head = bh = page_buffers(page); |
784 | do { | |
bb3b0e3d SW |
785 | if (atomic_read(&bh->b_count)) |
786 | goto cannot_release; | |
787 | bd = bh->b_private; | |
16ca9412 | 788 | if (bd && bd->bd_tr) |
bb3b0e3d | 789 | goto cannot_release; |
1c185c02 AG |
790 | if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh))) |
791 | goto cannot_release; | |
bb3b0e3d SW |
792 | bh = bh->b_this_page; |
793 | } while(bh != head); | |
380f7c65 | 794 | spin_unlock(&sdp->sd_ail_lock); |
4340fe62 | 795 | |
bb3b0e3d SW |
796 | head = bh = page_buffers(page); |
797 | do { | |
4340fe62 SW |
798 | bd = bh->b_private; |
799 | if (bd) { | |
800 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
e4f29206 | 801 | bd->bd_bh = NULL; |
4340fe62 | 802 | bh->b_private = NULL; |
019dd669 BP |
803 | /* |
804 | * The bd may still be queued as a revoke, in which | |
805 | * case we must not dequeue nor free it. | |
806 | */ | |
807 | if (!bd->bd_blkno && !list_empty(&bd->bd_list)) | |
808 | list_del_init(&bd->bd_list); | |
809 | if (list_empty(&bd->bd_list)) | |
810 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
e4f29206 | 811 | } |
4340fe62 SW |
812 | |
813 | bh = bh->b_this_page; | |
166afccd | 814 | } while (bh != head); |
e4f29206 | 815 | gfs2_log_unlock(sdp); |
4340fe62 | 816 | |
4340fe62 | 817 | return try_to_free_buffers(page); |
8f065d36 | 818 | |
bb3b0e3d | 819 | cannot_release: |
380f7c65 | 820 | spin_unlock(&sdp->sd_ail_lock); |
bb3b0e3d SW |
821 | gfs2_log_unlock(sdp); |
822 | return 0; | |
4340fe62 SW |
823 | } |
824 | ||
eadd7535 | 825 | static const struct address_space_operations gfs2_aops = { |
9d358143 | 826 | .writepage = gfs2_writepage, |
45138990 | 827 | .writepages = gfs2_writepages, |
5561093e | 828 | .readpage = gfs2_readpage, |
d4388340 | 829 | .readahead = gfs2_readahead, |
5561093e SW |
830 | .bmap = gfs2_bmap, |
831 | .invalidatepage = gfs2_invalidatepage, | |
832 | .releasepage = gfs2_releasepage, | |
967bcc91 | 833 | .direct_IO = noop_direct_IO, |
e5d9dc27 | 834 | .migratepage = buffer_migrate_page, |
229615de | 835 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 836 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
837 | }; |
838 | ||
5561093e | 839 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 840 | .writepage = gfs2_jdata_writepage, |
b8e7cbb6 | 841 | .writepages = gfs2_jdata_writepages, |
5561093e | 842 | .readpage = gfs2_readpage, |
d4388340 | 843 | .readahead = gfs2_readahead, |
b9e03f18 | 844 | .set_page_dirty = jdata_set_page_dirty, |
5561093e SW |
845 | .bmap = gfs2_bmap, |
846 | .invalidatepage = gfs2_invalidatepage, | |
847 | .releasepage = gfs2_releasepage, | |
229615de | 848 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 849 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
850 | }; |
851 | ||
852 | void gfs2_set_aops(struct inode *inode) | |
853 | { | |
eadd7535 | 854 | if (gfs2_is_jdata(GFS2_I(inode))) |
977767a7 | 855 | inode->i_mapping->a_ops = &gfs2_jdata_aops; |
5561093e | 856 | else |
eadd7535 | 857 | inode->i_mapping->a_ops = &gfs2_aops; |
5561093e | 858 | } |