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)) |
4e79e3f0 | 80 | return -ENODATA; |
7a6bbacb | 81 | return 0; |
b3b94faa DT |
82 | } |
83 | ||
21b6924b BP |
84 | /** |
85 | * gfs2_write_jdata_page - gfs2 jdata-specific version of block_write_full_page | |
86 | * @page: The page to write | |
87 | * @wbc: The writeback control | |
88 | * | |
89 | * This is the same as calling block_write_full_page, but it also | |
fd4c5748 BM |
90 | * writes pages outside of i_size |
91 | */ | |
21b6924b BP |
92 | static int gfs2_write_jdata_page(struct page *page, |
93 | struct writeback_control *wbc) | |
fd4c5748 BM |
94 | { |
95 | struct inode * const inode = page->mapping->host; | |
96 | loff_t i_size = i_size_read(inode); | |
97 | const pgoff_t end_index = i_size >> PAGE_SHIFT; | |
98 | unsigned offset; | |
99 | ||
100 | /* | |
101 | * The page straddles i_size. It must be zeroed out on each and every | |
102 | * writepage invocation because it may be mmapped. "A file is mapped | |
103 | * in multiples of the page size. For a file that is not a multiple of | |
104 | * the page size, the remaining memory is zeroed when mapped, and | |
105 | * writes to that region are not written out to the file." | |
106 | */ | |
f3b64b57 | 107 | offset = i_size & (PAGE_SIZE - 1); |
fd4c5748 BM |
108 | if (page->index == end_index && offset) |
109 | zero_user_segment(page, offset, PAGE_SIZE); | |
110 | ||
21b6924b | 111 | return __block_write_full_page(inode, page, gfs2_get_block_noalloc, wbc, |
fd4c5748 BM |
112 | end_buffer_async_write); |
113 | } | |
114 | ||
b8e7cbb6 SW |
115 | /** |
116 | * __gfs2_jdata_writepage - The core of jdata writepage | |
117 | * @page: The page to write | |
118 | * @wbc: The writeback control | |
119 | * | |
120 | * This is shared between writepage and writepages and implements the | |
121 | * core of the writepage operation. If a transaction is required then | |
122 | * PageChecked will have been set and the transaction will have | |
123 | * already been started before this is called. | |
124 | */ | |
125 | ||
126 | static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
127 | { | |
128 | struct inode *inode = page->mapping->host; | |
129 | struct gfs2_inode *ip = GFS2_I(inode); | |
130 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
131 | ||
132 | if (PageChecked(page)) { | |
133 | ClearPageChecked(page); | |
134 | if (!page_has_buffers(page)) { | |
135 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
47a9a527 | 136 | BIT(BH_Dirty)|BIT(BH_Uptodate)); |
b8e7cbb6 | 137 | } |
88b65ce5 | 138 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize); |
b8e7cbb6 | 139 | } |
21b6924b | 140 | return gfs2_write_jdata_page(page, wbc); |
b8e7cbb6 SW |
141 | } |
142 | ||
9ff8ec32 SW |
143 | /** |
144 | * gfs2_jdata_writepage - Write complete page | |
145 | * @page: Page to write | |
1272574b | 146 | * @wbc: The writeback control |
9ff8ec32 SW |
147 | * |
148 | * Returns: errno | |
149 | * | |
150 | */ | |
151 | ||
152 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
153 | { | |
154 | struct inode *inode = page->mapping->host; | |
fd4c5748 | 155 | struct gfs2_inode *ip = GFS2_I(inode); |
9ff8ec32 | 156 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9ff8ec32 | 157 | |
fd4c5748 BM |
158 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
159 | goto out; | |
160 | if (PageChecked(page) || current->journal_info) | |
161 | goto out_ignore; | |
e556280d | 162 | return __gfs2_jdata_writepage(page, wbc); |
18ec7d5c SW |
163 | |
164 | out_ignore: | |
165 | redirty_page_for_writepage(wbc, page); | |
fd4c5748 | 166 | out: |
18ec7d5c SW |
167 | unlock_page(page); |
168 | return 0; | |
b3b94faa DT |
169 | } |
170 | ||
a8d638e3 | 171 | /** |
45138990 | 172 | * gfs2_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
173 | * @mapping: The mapping to write |
174 | * @wbc: Write-back control | |
175 | * | |
45138990 | 176 | * Used for both ordered and writeback modes. |
a8d638e3 | 177 | */ |
45138990 SW |
178 | static int gfs2_writepages(struct address_space *mapping, |
179 | struct writeback_control *wbc) | |
a8d638e3 | 180 | { |
b066a4ee | 181 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); |
2164f9b9 CH |
182 | struct iomap_writepage_ctx wpc = { }; |
183 | int ret; | |
b066a4ee AD |
184 | |
185 | /* | |
186 | * Even if we didn't write any pages here, we might still be holding | |
187 | * dirty pages in the ail. We forcibly flush the ail because we don't | |
188 | * want balance_dirty_pages() to loop indefinitely trying to write out | |
189 | * pages held in the ail that it can't find. | |
190 | */ | |
2164f9b9 | 191 | ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops); |
b066a4ee AD |
192 | if (ret == 0) |
193 | set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags); | |
b066a4ee | 194 | return ret; |
a8d638e3 SW |
195 | } |
196 | ||
b8e7cbb6 SW |
197 | /** |
198 | * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages | |
199 | * @mapping: The mapping | |
200 | * @wbc: The writeback control | |
b8e7cbb6 SW |
201 | * @pvec: The vector of pages |
202 | * @nr_pages: The number of pages to write | |
1272574b | 203 | * @done_index: Page index |
b8e7cbb6 SW |
204 | * |
205 | * Returns: non-zero if loop should terminate, zero otherwise | |
206 | */ | |
207 | ||
208 | static int gfs2_write_jdata_pagevec(struct address_space *mapping, | |
209 | struct writeback_control *wbc, | |
210 | struct pagevec *pvec, | |
9aa01593 | 211 | int nr_pages, |
774016b2 | 212 | pgoff_t *done_index) |
b8e7cbb6 SW |
213 | { |
214 | struct inode *inode = mapping->host; | |
215 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
45eb0504 | 216 | unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits); |
b8e7cbb6 SW |
217 | int i; |
218 | int ret; | |
219 | ||
20b95bf2 | 220 | ret = gfs2_trans_begin(sdp, nrblocks, nrblocks); |
b8e7cbb6 SW |
221 | if (ret < 0) |
222 | return ret; | |
223 | ||
224 | for(i = 0; i < nr_pages; i++) { | |
225 | struct page *page = pvec->pages[i]; | |
226 | ||
774016b2 SW |
227 | *done_index = page->index; |
228 | ||
b8e7cbb6 SW |
229 | lock_page(page); |
230 | ||
231 | if (unlikely(page->mapping != mapping)) { | |
774016b2 | 232 | continue_unlock: |
b8e7cbb6 SW |
233 | unlock_page(page); |
234 | continue; | |
235 | } | |
236 | ||
774016b2 SW |
237 | if (!PageDirty(page)) { |
238 | /* someone wrote it for us */ | |
239 | goto continue_unlock; | |
b8e7cbb6 SW |
240 | } |
241 | ||
774016b2 SW |
242 | if (PageWriteback(page)) { |
243 | if (wbc->sync_mode != WB_SYNC_NONE) | |
244 | wait_on_page_writeback(page); | |
245 | else | |
246 | goto continue_unlock; | |
b8e7cbb6 SW |
247 | } |
248 | ||
774016b2 SW |
249 | BUG_ON(PageWriteback(page)); |
250 | if (!clear_page_dirty_for_io(page)) | |
251 | goto continue_unlock; | |
252 | ||
de1414a6 | 253 | trace_wbc_writepage(wbc, inode_to_bdi(inode)); |
b8e7cbb6 SW |
254 | |
255 | ret = __gfs2_jdata_writepage(page, wbc); | |
774016b2 SW |
256 | if (unlikely(ret)) { |
257 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
258 | unlock_page(page); | |
259 | ret = 0; | |
260 | } else { | |
261 | ||
262 | /* | |
263 | * done_index is set past this page, | |
264 | * so media errors will not choke | |
265 | * background writeout for the entire | |
266 | * file. This has consequences for | |
267 | * range_cyclic semantics (ie. it may | |
268 | * not be suitable for data integrity | |
269 | * writeout). | |
270 | */ | |
271 | *done_index = page->index + 1; | |
272 | ret = 1; | |
273 | break; | |
274 | } | |
275 | } | |
b8e7cbb6 | 276 | |
774016b2 SW |
277 | /* |
278 | * We stop writing back only if we are not doing | |
279 | * integrity sync. In case of integrity sync we have to | |
280 | * keep going until we have written all the pages | |
281 | * we tagged for writeback prior to entering this loop. | |
282 | */ | |
283 | if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) { | |
b8e7cbb6 | 284 | ret = 1; |
774016b2 SW |
285 | break; |
286 | } | |
287 | ||
b8e7cbb6 SW |
288 | } |
289 | gfs2_trans_end(sdp); | |
290 | return ret; | |
291 | } | |
292 | ||
293 | /** | |
294 | * gfs2_write_cache_jdata - Like write_cache_pages but different | |
295 | * @mapping: The mapping to write | |
296 | * @wbc: The writeback control | |
b8e7cbb6 SW |
297 | * |
298 | * The reason that we use our own function here is that we need to | |
299 | * start transactions before we grab page locks. This allows us | |
300 | * to get the ordering right. | |
301 | */ | |
302 | ||
303 | static int gfs2_write_cache_jdata(struct address_space *mapping, | |
304 | struct writeback_control *wbc) | |
305 | { | |
b8e7cbb6 SW |
306 | int ret = 0; |
307 | int done = 0; | |
308 | struct pagevec pvec; | |
309 | int nr_pages; | |
3f649ab7 | 310 | pgoff_t writeback_index; |
b8e7cbb6 SW |
311 | pgoff_t index; |
312 | pgoff_t end; | |
774016b2 SW |
313 | pgoff_t done_index; |
314 | int cycled; | |
b8e7cbb6 | 315 | int range_whole = 0; |
10bbd235 | 316 | xa_mark_t tag; |
b8e7cbb6 | 317 | |
86679820 | 318 | pagevec_init(&pvec); |
b8e7cbb6 | 319 | if (wbc->range_cyclic) { |
774016b2 SW |
320 | writeback_index = mapping->writeback_index; /* prev offset */ |
321 | index = writeback_index; | |
322 | if (index == 0) | |
323 | cycled = 1; | |
324 | else | |
325 | cycled = 0; | |
b8e7cbb6 SW |
326 | end = -1; |
327 | } else { | |
09cbfeaf KS |
328 | index = wbc->range_start >> PAGE_SHIFT; |
329 | end = wbc->range_end >> PAGE_SHIFT; | |
b8e7cbb6 SW |
330 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
331 | range_whole = 1; | |
774016b2 | 332 | cycled = 1; /* ignore range_cyclic tests */ |
b8e7cbb6 | 333 | } |
774016b2 SW |
334 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
335 | tag = PAGECACHE_TAG_TOWRITE; | |
336 | else | |
337 | tag = PAGECACHE_TAG_DIRTY; | |
b8e7cbb6 SW |
338 | |
339 | retry: | |
774016b2 SW |
340 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
341 | tag_pages_for_writeback(mapping, index, end); | |
342 | done_index = index; | |
343 | while (!done && (index <= end)) { | |
d2bc5b3c | 344 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 345 | tag); |
774016b2 SW |
346 | if (nr_pages == 0) |
347 | break; | |
348 | ||
9aa01593 | 349 | ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index); |
b8e7cbb6 SW |
350 | if (ret) |
351 | done = 1; | |
352 | if (ret > 0) | |
353 | ret = 0; | |
b8e7cbb6 SW |
354 | pagevec_release(&pvec); |
355 | cond_resched(); | |
356 | } | |
357 | ||
774016b2 | 358 | if (!cycled && !done) { |
b8e7cbb6 | 359 | /* |
774016b2 | 360 | * range_cyclic: |
b8e7cbb6 SW |
361 | * We hit the last page and there is more work to be done: wrap |
362 | * back to the start of the file | |
363 | */ | |
774016b2 | 364 | cycled = 1; |
b8e7cbb6 | 365 | index = 0; |
774016b2 | 366 | end = writeback_index - 1; |
b8e7cbb6 SW |
367 | goto retry; |
368 | } | |
369 | ||
370 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
774016b2 SW |
371 | mapping->writeback_index = done_index; |
372 | ||
b8e7cbb6 SW |
373 | return ret; |
374 | } | |
375 | ||
376 | ||
377 | /** | |
378 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk | |
379 | * @mapping: The mapping to write | |
380 | * @wbc: The writeback control | |
381 | * | |
382 | */ | |
383 | ||
384 | static int gfs2_jdata_writepages(struct address_space *mapping, | |
385 | struct writeback_control *wbc) | |
386 | { | |
387 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
388 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
389 | int ret; | |
390 | ||
391 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
392 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { | |
805c0907 BP |
393 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | |
394 | GFS2_LFC_JDATA_WPAGES); | |
b8e7cbb6 SW |
395 | ret = gfs2_write_cache_jdata(mapping, wbc); |
396 | } | |
397 | return ret; | |
398 | } | |
399 | ||
b3b94faa DT |
400 | /** |
401 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
402 | * @ip: the inode | |
403 | * @page: the page | |
404 | * | |
405 | * Returns: errno | |
406 | */ | |
378b6cbf | 407 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
408 | { |
409 | struct buffer_head *dibh; | |
602c89d2 | 410 | u64 dsize = i_size_read(&ip->i_inode); |
b3b94faa DT |
411 | void *kaddr; |
412 | int error; | |
413 | ||
bf126aee | 414 | /* |
3c18ddd1 | 415 | * Due to the order of unstuffing files and ->fault(), we can be |
bf126aee SW |
416 | * asked for a zero page in the case of a stuffed file being extended, |
417 | * so we need to supply one here. It doesn't happen often. | |
418 | */ | |
419 | if (unlikely(page->index)) { | |
09cbfeaf | 420 | zero_user(page, 0, PAGE_SIZE); |
0a7ab79c | 421 | SetPageUptodate(page); |
bf126aee SW |
422 | return 0; |
423 | } | |
fd88de56 | 424 | |
b3b94faa DT |
425 | error = gfs2_meta_inode_buffer(ip, &dibh); |
426 | if (error) | |
427 | return error; | |
428 | ||
d9349285 | 429 | kaddr = kmap_atomic(page); |
602c89d2 | 430 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 431 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
d9349285 | 432 | kunmap_atomic(kaddr); |
bf126aee | 433 | flush_dcache_page(page); |
b3b94faa | 434 | brelse(dibh); |
b3b94faa DT |
435 | SetPageUptodate(page); |
436 | ||
437 | return 0; | |
438 | } | |
439 | ||
e9b5b23e MWO |
440 | /** |
441 | * gfs2_read_folio - read a folio from a file | |
442 | * @file: The file to read | |
443 | * @folio: The folio in the file | |
444 | */ | |
445 | static int gfs2_read_folio(struct file *file, struct folio *folio) | |
b3b94faa | 446 | { |
e9b5b23e | 447 | struct inode *inode = folio->mapping->host; |
2164f9b9 CH |
448 | struct gfs2_inode *ip = GFS2_I(inode); |
449 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa DT |
450 | int error; |
451 | ||
2164f9b9 | 452 | if (!gfs2_is_jdata(ip) || |
e9b5b23e | 453 | (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) { |
7479c505 | 454 | error = iomap_read_folio(folio, &gfs2_iomap_ops); |
f95cbb44 | 455 | } else if (gfs2_is_stuffed(ip)) { |
e9b5b23e MWO |
456 | error = stuffed_readpage(ip, &folio->page); |
457 | folio_unlock(folio); | |
51ff87bd | 458 | } else { |
f132ab7d | 459 | error = mpage_read_folio(folio, gfs2_block_map); |
51ff87bd | 460 | } |
b3b94faa | 461 | |
eb43e660 | 462 | if (unlikely(gfs2_withdrawn(sdp))) |
51ff87bd | 463 | return -EIO; |
b3b94faa | 464 | |
51ff87bd SW |
465 | return error; |
466 | } | |
467 | ||
51ff87bd SW |
468 | /** |
469 | * gfs2_internal_read - read an internal file | |
470 | * @ip: The gfs2 inode | |
51ff87bd SW |
471 | * @buf: The buffer to fill |
472 | * @pos: The file position | |
473 | * @size: The amount to read | |
474 | * | |
475 | */ | |
476 | ||
4306629e AP |
477 | int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos, |
478 | unsigned size) | |
51ff87bd SW |
479 | { |
480 | struct address_space *mapping = ip->i_inode.i_mapping; | |
45eb0504 | 481 | unsigned long index = *pos >> PAGE_SHIFT; |
09cbfeaf | 482 | unsigned offset = *pos & (PAGE_SIZE - 1); |
51ff87bd SW |
483 | unsigned copied = 0; |
484 | unsigned amt; | |
485 | struct page *page; | |
486 | void *p; | |
487 | ||
488 | do { | |
489 | amt = size - copied; | |
09cbfeaf KS |
490 | if (offset + size > PAGE_SIZE) |
491 | amt = PAGE_SIZE - offset; | |
e9b5b23e | 492 | page = read_cache_page(mapping, index, gfs2_read_folio, NULL); |
51ff87bd SW |
493 | if (IS_ERR(page)) |
494 | return PTR_ERR(page); | |
d9349285 | 495 | p = kmap_atomic(page); |
51ff87bd | 496 | memcpy(buf + copied, p + offset, amt); |
d9349285 | 497 | kunmap_atomic(p); |
09cbfeaf | 498 | put_page(page); |
51ff87bd SW |
499 | copied += amt; |
500 | index++; | |
501 | offset = 0; | |
502 | } while(copied < size); | |
503 | (*pos) += size; | |
504 | return size; | |
fd88de56 SW |
505 | } |
506 | ||
fd88de56 | 507 | /** |
d4388340 | 508 | * gfs2_readahead - Read a bunch of pages at once |
c551f66c | 509 | * @rac: Read-ahead control structure |
fd88de56 SW |
510 | * |
511 | * Some notes: | |
512 | * 1. This is only for readahead, so we can simply ignore any things | |
513 | * which are slightly inconvenient (such as locking conflicts between | |
514 | * the page lock and the glock) and return having done no I/O. Its | |
515 | * obviously not something we'd want to do on too regular a basis. | |
516 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 517 | * 2. We don't handle stuffed files here we let readpage do the honours. |
d4388340 | 518 | * 3. mpage_readahead() does most of the heavy lifting in the common case. |
e9e1ef2b | 519 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
fd88de56 | 520 | */ |
3cc3f710 | 521 | |
d4388340 | 522 | static void gfs2_readahead(struct readahead_control *rac) |
fd88de56 | 523 | { |
d4388340 | 524 | struct inode *inode = rac->mapping->host; |
feaa7bba | 525 | struct gfs2_inode *ip = GFS2_I(inode); |
fd88de56 | 526 | |
2164f9b9 CH |
527 | if (gfs2_is_stuffed(ip)) |
528 | ; | |
529 | else if (gfs2_is_jdata(ip)) | |
d4388340 | 530 | mpage_readahead(rac, gfs2_block_map); |
2164f9b9 CH |
531 | else |
532 | iomap_readahead(rac, &gfs2_iomap_ops); | |
b3b94faa DT |
533 | } |
534 | ||
7ae8fa84 RP |
535 | /** |
536 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
537 | * @inode: the rindex inode | |
538 | */ | |
64bc06bb | 539 | void adjust_fs_space(struct inode *inode) |
7ae8fa84 | 540 | { |
d0a22a4b | 541 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1946f70a | 542 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
7ae8fa84 RP |
543 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
544 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
70c11ba8 | 545 | struct buffer_head *m_bh; |
7ae8fa84 RP |
546 | u64 fs_total, new_free; |
547 | ||
d0a22a4b AG |
548 | if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0) |
549 | return; | |
550 | ||
7ae8fa84 RP |
551 | /* Total up the file system space, according to the latest rindex. */ |
552 | fs_total = gfs2_ri_total(sdp); | |
1946f70a | 553 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) |
d0a22a4b | 554 | goto out; |
7ae8fa84 RP |
555 | |
556 | spin_lock(&sdp->sd_statfs_spin); | |
1946f70a BM |
557 | gfs2_statfs_change_in(m_sc, m_bh->b_data + |
558 | sizeof(struct gfs2_dinode)); | |
7ae8fa84 RP |
559 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) |
560 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
561 | else | |
562 | new_free = 0; | |
563 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
564 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
565 | (unsigned long long)new_free); | |
7ae8fa84 | 566 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
1946f70a | 567 | |
70c11ba8 | 568 | update_statfs(sdp, m_bh); |
1946f70a | 569 | brelse(m_bh); |
d0a22a4b AG |
570 | out: |
571 | sdp->sd_rindex_uptodate = 0; | |
572 | gfs2_trans_end(sdp); | |
7ae8fa84 RP |
573 | } |
574 | ||
e621900a MWO |
575 | static bool jdata_dirty_folio(struct address_space *mapping, |
576 | struct folio *folio) | |
8fb68595 | 577 | { |
6302d6f4 | 578 | if (current->journal_info) |
e621900a MWO |
579 | folio_set_checked(folio); |
580 | return block_dirty_folio(mapping, folio); | |
8fb68595 RP |
581 | } |
582 | ||
b3b94faa DT |
583 | /** |
584 | * gfs2_bmap - Block map function | |
585 | * @mapping: Address space info | |
586 | * @lblock: The block to map | |
587 | * | |
588 | * Returns: The disk address for the block or 0 on hole or error | |
589 | */ | |
590 | ||
591 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
592 | { | |
feaa7bba | 593 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
594 | struct gfs2_holder i_gh; |
595 | sector_t dblock = 0; | |
596 | int error; | |
597 | ||
b3b94faa DT |
598 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
599 | if (error) | |
600 | return 0; | |
601 | ||
602 | if (!gfs2_is_stuffed(ip)) | |
7770c93a | 603 | dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops); |
b3b94faa DT |
604 | |
605 | gfs2_glock_dq_uninit(&i_gh); | |
606 | ||
607 | return dblock; | |
608 | } | |
609 | ||
d7b616e2 SW |
610 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
611 | { | |
612 | struct gfs2_bufdata *bd; | |
613 | ||
614 | lock_buffer(bh); | |
615 | gfs2_log_lock(sdp); | |
616 | clear_buffer_dirty(bh); | |
617 | bd = bh->b_private; | |
618 | if (bd) { | |
c0752aa7 BP |
619 | if (!list_empty(&bd->bd_list) && !buffer_pinned(bh)) |
620 | list_del_init(&bd->bd_list); | |
68942870 BP |
621 | else { |
622 | spin_lock(&sdp->sd_ail_lock); | |
68cd4ce2 | 623 | gfs2_remove_from_journal(bh, REMOVE_JDATA); |
68942870 BP |
624 | spin_unlock(&sdp->sd_ail_lock); |
625 | } | |
d7b616e2 SW |
626 | } |
627 | bh->b_bdev = NULL; | |
628 | clear_buffer_mapped(bh); | |
629 | clear_buffer_req(bh); | |
630 | clear_buffer_new(bh); | |
631 | gfs2_log_unlock(sdp); | |
632 | unlock_buffer(bh); | |
633 | } | |
634 | ||
5f4b2976 MWO |
635 | static void gfs2_invalidate_folio(struct folio *folio, size_t offset, |
636 | size_t length) | |
b3b94faa | 637 | { |
5f4b2976 MWO |
638 | struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host); |
639 | size_t stop = offset + length; | |
640 | int partial_page = (offset || length < folio_size(folio)); | |
d7b616e2 SW |
641 | struct buffer_head *bh, *head; |
642 | unsigned long pos = 0; | |
643 | ||
5f4b2976 | 644 | BUG_ON(!folio_test_locked(folio)); |
5c0bb97c | 645 | if (!partial_page) |
5f4b2976 MWO |
646 | folio_clear_checked(folio); |
647 | head = folio_buffers(folio); | |
648 | if (!head) | |
d7b616e2 | 649 | goto out; |
b3b94faa | 650 | |
5f4b2976 | 651 | bh = head; |
d7b616e2 | 652 | do { |
5c0bb97c LC |
653 | if (pos + bh->b_size > stop) |
654 | return; | |
655 | ||
d7b616e2 SW |
656 | if (offset <= pos) |
657 | gfs2_discard(sdp, bh); | |
658 | pos += bh->b_size; | |
659 | bh = bh->b_this_page; | |
660 | } while (bh != head); | |
661 | out: | |
5c0bb97c | 662 | if (!partial_page) |
5f4b2976 | 663 | filemap_release_folio(folio, 0); |
b3b94faa DT |
664 | } |
665 | ||
4340fe62 | 666 | /** |
e45c20d1 MWO |
667 | * gfs2_release_folio - free the metadata associated with a folio |
668 | * @folio: the folio that's being released | |
4340fe62 SW |
669 | * @gfp_mask: passed from Linux VFS, ignored by us |
670 | * | |
e45c20d1 | 671 | * Calls try_to_free_buffers() to free the buffers and put the folio if the |
0ebbe4f9 | 672 | * buffers can be released. |
4340fe62 | 673 | * |
e45c20d1 | 674 | * Returns: true if the folio was put or else false |
4340fe62 SW |
675 | */ |
676 | ||
e45c20d1 | 677 | bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask) |
4340fe62 | 678 | { |
e45c20d1 | 679 | struct address_space *mapping = folio->mapping; |
009d8518 | 680 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); |
4340fe62 SW |
681 | struct buffer_head *bh, *head; |
682 | struct gfs2_bufdata *bd; | |
4340fe62 | 683 | |
e45c20d1 MWO |
684 | head = folio_buffers(folio); |
685 | if (!head) | |
686 | return false; | |
4340fe62 | 687 | |
1c185c02 | 688 | /* |
e45c20d1 MWO |
689 | * mm accommodates an old ext3 case where clean folios might |
690 | * not have had the dirty bit cleared. Thus, it can send actual | |
691 | * dirty folios to ->release_folio() via shrink_active_list(). | |
1c185c02 | 692 | * |
e45c20d1 MWO |
693 | * As a workaround, we skip folios that contain dirty buffers |
694 | * below. Once ->release_folio isn't called on dirty folios | |
695 | * anymore, we can warn on dirty buffers like we used to here | |
696 | * again. | |
1c185c02 AG |
697 | */ |
698 | ||
bb3b0e3d | 699 | gfs2_log_lock(sdp); |
e45c20d1 | 700 | bh = head; |
4340fe62 | 701 | do { |
bb3b0e3d SW |
702 | if (atomic_read(&bh->b_count)) |
703 | goto cannot_release; | |
704 | bd = bh->b_private; | |
16ca9412 | 705 | if (bd && bd->bd_tr) |
bb3b0e3d | 706 | goto cannot_release; |
1c185c02 AG |
707 | if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh))) |
708 | goto cannot_release; | |
bb3b0e3d | 709 | bh = bh->b_this_page; |
e45c20d1 | 710 | } while (bh != head); |
4340fe62 | 711 | |
e45c20d1 | 712 | bh = head; |
bb3b0e3d | 713 | do { |
4340fe62 SW |
714 | bd = bh->b_private; |
715 | if (bd) { | |
716 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
e4f29206 | 717 | bd->bd_bh = NULL; |
4340fe62 | 718 | bh->b_private = NULL; |
019dd669 BP |
719 | /* |
720 | * The bd may still be queued as a revoke, in which | |
721 | * case we must not dequeue nor free it. | |
722 | */ | |
723 | if (!bd->bd_blkno && !list_empty(&bd->bd_list)) | |
724 | list_del_init(&bd->bd_list); | |
725 | if (list_empty(&bd->bd_list)) | |
726 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
e4f29206 | 727 | } |
4340fe62 SW |
728 | |
729 | bh = bh->b_this_page; | |
166afccd | 730 | } while (bh != head); |
e4f29206 | 731 | gfs2_log_unlock(sdp); |
4340fe62 | 732 | |
68189fef | 733 | return try_to_free_buffers(folio); |
8f065d36 | 734 | |
bb3b0e3d SW |
735 | cannot_release: |
736 | gfs2_log_unlock(sdp); | |
e45c20d1 | 737 | return false; |
4340fe62 SW |
738 | } |
739 | ||
eadd7535 | 740 | static const struct address_space_operations gfs2_aops = { |
45138990 | 741 | .writepages = gfs2_writepages, |
f132ab7d | 742 | .read_folio = gfs2_read_folio, |
d4388340 | 743 | .readahead = gfs2_readahead, |
187c82cb | 744 | .dirty_folio = filemap_dirty_folio, |
8597447d | 745 | .release_folio = iomap_release_folio, |
d82354f6 | 746 | .invalidate_folio = iomap_invalidate_folio, |
5561093e | 747 | .bmap = gfs2_bmap, |
967bcc91 | 748 | .direct_IO = noop_direct_IO, |
2ec810d5 | 749 | .migrate_folio = filemap_migrate_folio, |
2164f9b9 | 750 | .is_partially_uptodate = iomap_is_partially_uptodate, |
aa261f54 | 751 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
752 | }; |
753 | ||
5561093e | 754 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 755 | .writepage = gfs2_jdata_writepage, |
b8e7cbb6 | 756 | .writepages = gfs2_jdata_writepages, |
f132ab7d | 757 | .read_folio = gfs2_read_folio, |
d4388340 | 758 | .readahead = gfs2_readahead, |
e621900a | 759 | .dirty_folio = jdata_dirty_folio, |
5561093e | 760 | .bmap = gfs2_bmap, |
5f4b2976 | 761 | .invalidate_folio = gfs2_invalidate_folio, |
e45c20d1 | 762 | .release_folio = gfs2_release_folio, |
229615de | 763 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 764 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
765 | }; |
766 | ||
767 | void gfs2_set_aops(struct inode *inode) | |
768 | { | |
eadd7535 | 769 | if (gfs2_is_jdata(GFS2_I(inode))) |
977767a7 | 770 | inode->i_mapping->a_ops = &gfs2_jdata_aops; |
5561093e | 771 | else |
eadd7535 | 772 | inode->i_mapping->a_ops = &gfs2_aops; |
5561093e | 773 | } |