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 | ||
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); | |
2164f9b9 | 94 | struct iomap_writepage_ctx wpc = { }; |
b3b94faa | 95 | |
9ff8ec32 SW |
96 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
97 | goto out; | |
5c676f6d | 98 | if (current->journal_info) |
9ff8ec32 | 99 | goto redirty; |
2164f9b9 | 100 | return iomap_writepage(page, wbc, &wpc, &gfs2_writeback_ops); |
59c01c50 | 101 | |
9ff8ec32 SW |
102 | redirty: |
103 | redirty_page_for_writepage(wbc, page); | |
104 | out: | |
105 | unlock_page(page); | |
106 | return 0; | |
107 | } | |
108 | ||
21b6924b BP |
109 | /** |
110 | * gfs2_write_jdata_page - gfs2 jdata-specific version of block_write_full_page | |
111 | * @page: The page to write | |
112 | * @wbc: The writeback control | |
113 | * | |
114 | * This is the same as calling block_write_full_page, but it also | |
fd4c5748 BM |
115 | * writes pages outside of i_size |
116 | */ | |
21b6924b BP |
117 | static int gfs2_write_jdata_page(struct page *page, |
118 | struct writeback_control *wbc) | |
fd4c5748 BM |
119 | { |
120 | struct inode * const inode = page->mapping->host; | |
121 | loff_t i_size = i_size_read(inode); | |
122 | const pgoff_t end_index = i_size >> PAGE_SHIFT; | |
123 | unsigned offset; | |
124 | ||
125 | /* | |
126 | * The page straddles i_size. It must be zeroed out on each and every | |
127 | * writepage invocation because it may be mmapped. "A file is mapped | |
128 | * in multiples of the page size. For a file that is not a multiple of | |
129 | * the page size, the remaining memory is zeroed when mapped, and | |
130 | * writes to that region are not written out to the file." | |
131 | */ | |
f3b64b57 | 132 | offset = i_size & (PAGE_SIZE - 1); |
fd4c5748 BM |
133 | if (page->index == end_index && offset) |
134 | zero_user_segment(page, offset, PAGE_SIZE); | |
135 | ||
21b6924b | 136 | return __block_write_full_page(inode, page, gfs2_get_block_noalloc, wbc, |
fd4c5748 BM |
137 | end_buffer_async_write); |
138 | } | |
139 | ||
b8e7cbb6 SW |
140 | /** |
141 | * __gfs2_jdata_writepage - The core of jdata writepage | |
142 | * @page: The page to write | |
143 | * @wbc: The writeback control | |
144 | * | |
145 | * This is shared between writepage and writepages and implements the | |
146 | * core of the writepage operation. If a transaction is required then | |
147 | * PageChecked will have been set and the transaction will have | |
148 | * already been started before this is called. | |
149 | */ | |
150 | ||
151 | static int __gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
152 | { | |
153 | struct inode *inode = page->mapping->host; | |
154 | struct gfs2_inode *ip = GFS2_I(inode); | |
155 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
156 | ||
157 | if (PageChecked(page)) { | |
158 | ClearPageChecked(page); | |
159 | if (!page_has_buffers(page)) { | |
160 | create_empty_buffers(page, inode->i_sb->s_blocksize, | |
47a9a527 | 161 | BIT(BH_Dirty)|BIT(BH_Uptodate)); |
b8e7cbb6 | 162 | } |
88b65ce5 | 163 | gfs2_page_add_databufs(ip, page, 0, sdp->sd_vfs->s_blocksize); |
b8e7cbb6 | 164 | } |
21b6924b | 165 | return gfs2_write_jdata_page(page, wbc); |
b8e7cbb6 SW |
166 | } |
167 | ||
9ff8ec32 SW |
168 | /** |
169 | * gfs2_jdata_writepage - Write complete page | |
170 | * @page: Page to write | |
1272574b | 171 | * @wbc: The writeback control |
9ff8ec32 SW |
172 | * |
173 | * Returns: errno | |
174 | * | |
175 | */ | |
176 | ||
177 | static int gfs2_jdata_writepage(struct page *page, struct writeback_control *wbc) | |
178 | { | |
179 | struct inode *inode = page->mapping->host; | |
fd4c5748 | 180 | struct gfs2_inode *ip = GFS2_I(inode); |
9ff8ec32 | 181 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
9ff8ec32 | 182 | |
fd4c5748 BM |
183 | if (gfs2_assert_withdraw(sdp, gfs2_glock_is_held_excl(ip->i_gl))) |
184 | goto out; | |
185 | if (PageChecked(page) || current->journal_info) | |
186 | goto out_ignore; | |
e556280d | 187 | return __gfs2_jdata_writepage(page, wbc); |
18ec7d5c SW |
188 | |
189 | out_ignore: | |
190 | redirty_page_for_writepage(wbc, page); | |
fd4c5748 | 191 | out: |
18ec7d5c SW |
192 | unlock_page(page); |
193 | return 0; | |
b3b94faa DT |
194 | } |
195 | ||
a8d638e3 | 196 | /** |
45138990 | 197 | * gfs2_writepages - Write a bunch of dirty pages back to disk |
a8d638e3 SW |
198 | * @mapping: The mapping to write |
199 | * @wbc: Write-back control | |
200 | * | |
45138990 | 201 | * Used for both ordered and writeback modes. |
a8d638e3 | 202 | */ |
45138990 SW |
203 | static int gfs2_writepages(struct address_space *mapping, |
204 | struct writeback_control *wbc) | |
a8d638e3 | 205 | { |
b066a4ee | 206 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); |
2164f9b9 CH |
207 | struct iomap_writepage_ctx wpc = { }; |
208 | int ret; | |
b066a4ee AD |
209 | |
210 | /* | |
211 | * Even if we didn't write any pages here, we might still be holding | |
212 | * dirty pages in the ail. We forcibly flush the ail because we don't | |
213 | * want balance_dirty_pages() to loop indefinitely trying to write out | |
214 | * pages held in the ail that it can't find. | |
215 | */ | |
2164f9b9 | 216 | ret = iomap_writepages(mapping, wbc, &wpc, &gfs2_writeback_ops); |
b066a4ee AD |
217 | if (ret == 0) |
218 | set_bit(SDF_FORCE_AIL_FLUSH, &sdp->sd_flags); | |
b066a4ee | 219 | return ret; |
a8d638e3 SW |
220 | } |
221 | ||
b8e7cbb6 SW |
222 | /** |
223 | * gfs2_write_jdata_pagevec - Write back a pagevec's worth of pages | |
224 | * @mapping: The mapping | |
225 | * @wbc: The writeback control | |
b8e7cbb6 SW |
226 | * @pvec: The vector of pages |
227 | * @nr_pages: The number of pages to write | |
1272574b | 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, | |
9aa01593 | 236 | int nr_pages, |
774016b2 | 237 | pgoff_t *done_index) |
b8e7cbb6 SW |
238 | { |
239 | struct inode *inode = mapping->host; | |
240 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
45eb0504 | 241 | unsigned nrblocks = nr_pages * (PAGE_SIZE >> inode->i_blkbits); |
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 | *done_index = page->index; |
253 | ||
b8e7cbb6 SW |
254 | lock_page(page); |
255 | ||
256 | if (unlikely(page->mapping != mapping)) { | |
774016b2 | 257 | continue_unlock: |
b8e7cbb6 SW |
258 | unlock_page(page); |
259 | continue; | |
260 | } | |
261 | ||
774016b2 SW |
262 | if (!PageDirty(page)) { |
263 | /* someone wrote it for us */ | |
264 | goto continue_unlock; | |
b8e7cbb6 SW |
265 | } |
266 | ||
774016b2 SW |
267 | if (PageWriteback(page)) { |
268 | if (wbc->sync_mode != WB_SYNC_NONE) | |
269 | wait_on_page_writeback(page); | |
270 | else | |
271 | goto continue_unlock; | |
b8e7cbb6 SW |
272 | } |
273 | ||
774016b2 SW |
274 | BUG_ON(PageWriteback(page)); |
275 | if (!clear_page_dirty_for_io(page)) | |
276 | goto continue_unlock; | |
277 | ||
de1414a6 | 278 | trace_wbc_writepage(wbc, inode_to_bdi(inode)); |
b8e7cbb6 SW |
279 | |
280 | ret = __gfs2_jdata_writepage(page, wbc); | |
774016b2 SW |
281 | if (unlikely(ret)) { |
282 | if (ret == AOP_WRITEPAGE_ACTIVATE) { | |
283 | unlock_page(page); | |
284 | ret = 0; | |
285 | } else { | |
286 | ||
287 | /* | |
288 | * done_index is set past this page, | |
289 | * so media errors will not choke | |
290 | * background writeout for the entire | |
291 | * file. This has consequences for | |
292 | * range_cyclic semantics (ie. it may | |
293 | * not be suitable for data integrity | |
294 | * writeout). | |
295 | */ | |
296 | *done_index = page->index + 1; | |
297 | ret = 1; | |
298 | break; | |
299 | } | |
300 | } | |
b8e7cbb6 | 301 | |
774016b2 SW |
302 | /* |
303 | * We stop writing back only if we are not doing | |
304 | * integrity sync. In case of integrity sync we have to | |
305 | * keep going until we have written all the pages | |
306 | * we tagged for writeback prior to entering this loop. | |
307 | */ | |
308 | if (--wbc->nr_to_write <= 0 && wbc->sync_mode == WB_SYNC_NONE) { | |
b8e7cbb6 | 309 | ret = 1; |
774016b2 SW |
310 | break; |
311 | } | |
312 | ||
b8e7cbb6 SW |
313 | } |
314 | gfs2_trans_end(sdp); | |
315 | return ret; | |
316 | } | |
317 | ||
318 | /** | |
319 | * gfs2_write_cache_jdata - Like write_cache_pages but different | |
320 | * @mapping: The mapping to write | |
321 | * @wbc: The writeback control | |
b8e7cbb6 SW |
322 | * |
323 | * The reason that we use our own function here is that we need to | |
324 | * start transactions before we grab page locks. This allows us | |
325 | * to get the ordering right. | |
326 | */ | |
327 | ||
328 | static int gfs2_write_cache_jdata(struct address_space *mapping, | |
329 | struct writeback_control *wbc) | |
330 | { | |
b8e7cbb6 SW |
331 | int ret = 0; |
332 | int done = 0; | |
333 | struct pagevec pvec; | |
334 | int nr_pages; | |
3f649ab7 | 335 | pgoff_t writeback_index; |
b8e7cbb6 SW |
336 | pgoff_t index; |
337 | pgoff_t end; | |
774016b2 SW |
338 | pgoff_t done_index; |
339 | int cycled; | |
b8e7cbb6 | 340 | int range_whole = 0; |
10bbd235 | 341 | xa_mark_t tag; |
b8e7cbb6 | 342 | |
86679820 | 343 | pagevec_init(&pvec); |
b8e7cbb6 | 344 | if (wbc->range_cyclic) { |
774016b2 SW |
345 | writeback_index = mapping->writeback_index; /* prev offset */ |
346 | index = writeback_index; | |
347 | if (index == 0) | |
348 | cycled = 1; | |
349 | else | |
350 | cycled = 0; | |
b8e7cbb6 SW |
351 | end = -1; |
352 | } else { | |
09cbfeaf KS |
353 | index = wbc->range_start >> PAGE_SHIFT; |
354 | end = wbc->range_end >> PAGE_SHIFT; | |
b8e7cbb6 SW |
355 | if (wbc->range_start == 0 && wbc->range_end == LLONG_MAX) |
356 | range_whole = 1; | |
774016b2 | 357 | cycled = 1; /* ignore range_cyclic tests */ |
b8e7cbb6 | 358 | } |
774016b2 SW |
359 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
360 | tag = PAGECACHE_TAG_TOWRITE; | |
361 | else | |
362 | tag = PAGECACHE_TAG_DIRTY; | |
b8e7cbb6 SW |
363 | |
364 | retry: | |
774016b2 SW |
365 | if (wbc->sync_mode == WB_SYNC_ALL || wbc->tagged_writepages) |
366 | tag_pages_for_writeback(mapping, index, end); | |
367 | done_index = index; | |
368 | while (!done && (index <= end)) { | |
d2bc5b3c | 369 | nr_pages = pagevec_lookup_range_tag(&pvec, mapping, &index, end, |
67fd707f | 370 | tag); |
774016b2 SW |
371 | if (nr_pages == 0) |
372 | break; | |
373 | ||
9aa01593 | 374 | ret = gfs2_write_jdata_pagevec(mapping, wbc, &pvec, nr_pages, &done_index); |
b8e7cbb6 SW |
375 | if (ret) |
376 | done = 1; | |
377 | if (ret > 0) | |
378 | ret = 0; | |
b8e7cbb6 SW |
379 | pagevec_release(&pvec); |
380 | cond_resched(); | |
381 | } | |
382 | ||
774016b2 | 383 | if (!cycled && !done) { |
b8e7cbb6 | 384 | /* |
774016b2 | 385 | * range_cyclic: |
b8e7cbb6 SW |
386 | * We hit the last page and there is more work to be done: wrap |
387 | * back to the start of the file | |
388 | */ | |
774016b2 | 389 | cycled = 1; |
b8e7cbb6 | 390 | index = 0; |
774016b2 | 391 | end = writeback_index - 1; |
b8e7cbb6 SW |
392 | goto retry; |
393 | } | |
394 | ||
395 | if (wbc->range_cyclic || (range_whole && wbc->nr_to_write > 0)) | |
774016b2 SW |
396 | mapping->writeback_index = done_index; |
397 | ||
b8e7cbb6 SW |
398 | return ret; |
399 | } | |
400 | ||
401 | ||
402 | /** | |
403 | * gfs2_jdata_writepages - Write a bunch of dirty pages back to disk | |
404 | * @mapping: The mapping to write | |
405 | * @wbc: The writeback control | |
406 | * | |
407 | */ | |
408 | ||
409 | static int gfs2_jdata_writepages(struct address_space *mapping, | |
410 | struct writeback_control *wbc) | |
411 | { | |
412 | struct gfs2_inode *ip = GFS2_I(mapping->host); | |
413 | struct gfs2_sbd *sdp = GFS2_SB(mapping->host); | |
414 | int ret; | |
415 | ||
416 | ret = gfs2_write_cache_jdata(mapping, wbc); | |
417 | if (ret == 0 && wbc->sync_mode == WB_SYNC_ALL) { | |
805c0907 BP |
418 | gfs2_log_flush(sdp, ip->i_gl, GFS2_LOG_HEAD_FLUSH_NORMAL | |
419 | GFS2_LFC_JDATA_WPAGES); | |
b8e7cbb6 SW |
420 | ret = gfs2_write_cache_jdata(mapping, wbc); |
421 | } | |
422 | return ret; | |
423 | } | |
424 | ||
b3b94faa DT |
425 | /** |
426 | * stuffed_readpage - Fill in a Linux page with stuffed file data | |
427 | * @ip: the inode | |
428 | * @page: the page | |
429 | * | |
430 | * Returns: errno | |
431 | */ | |
378b6cbf | 432 | static int stuffed_readpage(struct gfs2_inode *ip, struct page *page) |
b3b94faa DT |
433 | { |
434 | struct buffer_head *dibh; | |
602c89d2 | 435 | u64 dsize = i_size_read(&ip->i_inode); |
b3b94faa DT |
436 | void *kaddr; |
437 | int error; | |
438 | ||
bf126aee | 439 | /* |
3c18ddd1 | 440 | * Due to the order of unstuffing files and ->fault(), we can be |
bf126aee SW |
441 | * asked for a zero page in the case of a stuffed file being extended, |
442 | * so we need to supply one here. It doesn't happen often. | |
443 | */ | |
444 | if (unlikely(page->index)) { | |
09cbfeaf | 445 | zero_user(page, 0, PAGE_SIZE); |
0a7ab79c | 446 | SetPageUptodate(page); |
bf126aee SW |
447 | return 0; |
448 | } | |
fd88de56 | 449 | |
b3b94faa DT |
450 | error = gfs2_meta_inode_buffer(ip, &dibh); |
451 | if (error) | |
452 | return error; | |
453 | ||
d9349285 | 454 | kaddr = kmap_atomic(page); |
235628c5 AG |
455 | if (dsize > gfs2_max_stuffed_size(ip)) |
456 | dsize = gfs2_max_stuffed_size(ip); | |
602c89d2 | 457 | memcpy(kaddr, dibh->b_data + sizeof(struct gfs2_dinode), dsize); |
09cbfeaf | 458 | memset(kaddr + dsize, 0, PAGE_SIZE - dsize); |
d9349285 | 459 | kunmap_atomic(kaddr); |
bf126aee | 460 | flush_dcache_page(page); |
b3b94faa | 461 | brelse(dibh); |
b3b94faa DT |
462 | SetPageUptodate(page); |
463 | ||
464 | return 0; | |
465 | } | |
466 | ||
e9b5b23e MWO |
467 | /** |
468 | * gfs2_read_folio - read a folio from a file | |
469 | * @file: The file to read | |
470 | * @folio: The folio in the file | |
471 | */ | |
472 | static int gfs2_read_folio(struct file *file, struct folio *folio) | |
b3b94faa | 473 | { |
e9b5b23e | 474 | struct inode *inode = folio->mapping->host; |
2164f9b9 CH |
475 | struct gfs2_inode *ip = GFS2_I(inode); |
476 | struct gfs2_sbd *sdp = GFS2_SB(inode); | |
b3b94faa DT |
477 | int error; |
478 | ||
2164f9b9 | 479 | if (!gfs2_is_jdata(ip) || |
e9b5b23e | 480 | (i_blocksize(inode) == PAGE_SIZE && !folio_buffers(folio))) { |
7479c505 | 481 | error = iomap_read_folio(folio, &gfs2_iomap_ops); |
f95cbb44 | 482 | } else if (gfs2_is_stuffed(ip)) { |
e9b5b23e MWO |
483 | error = stuffed_readpage(ip, &folio->page); |
484 | folio_unlock(folio); | |
51ff87bd | 485 | } else { |
f132ab7d | 486 | error = mpage_read_folio(folio, gfs2_block_map); |
51ff87bd | 487 | } |
b3b94faa | 488 | |
eb43e660 | 489 | if (unlikely(gfs2_withdrawn(sdp))) |
51ff87bd | 490 | return -EIO; |
b3b94faa | 491 | |
51ff87bd SW |
492 | return error; |
493 | } | |
494 | ||
51ff87bd SW |
495 | /** |
496 | * gfs2_internal_read - read an internal file | |
497 | * @ip: The gfs2 inode | |
51ff87bd SW |
498 | * @buf: The buffer to fill |
499 | * @pos: The file position | |
500 | * @size: The amount to read | |
501 | * | |
502 | */ | |
503 | ||
4306629e AP |
504 | int gfs2_internal_read(struct gfs2_inode *ip, char *buf, loff_t *pos, |
505 | unsigned size) | |
51ff87bd SW |
506 | { |
507 | struct address_space *mapping = ip->i_inode.i_mapping; | |
45eb0504 | 508 | unsigned long index = *pos >> PAGE_SHIFT; |
09cbfeaf | 509 | unsigned offset = *pos & (PAGE_SIZE - 1); |
51ff87bd SW |
510 | unsigned copied = 0; |
511 | unsigned amt; | |
512 | struct page *page; | |
513 | void *p; | |
514 | ||
515 | do { | |
516 | amt = size - copied; | |
09cbfeaf KS |
517 | if (offset + size > PAGE_SIZE) |
518 | amt = PAGE_SIZE - offset; | |
e9b5b23e | 519 | page = read_cache_page(mapping, index, gfs2_read_folio, NULL); |
51ff87bd SW |
520 | if (IS_ERR(page)) |
521 | return PTR_ERR(page); | |
d9349285 | 522 | p = kmap_atomic(page); |
51ff87bd | 523 | memcpy(buf + copied, p + offset, amt); |
d9349285 | 524 | kunmap_atomic(p); |
09cbfeaf | 525 | put_page(page); |
51ff87bd SW |
526 | copied += amt; |
527 | index++; | |
528 | offset = 0; | |
529 | } while(copied < size); | |
530 | (*pos) += size; | |
531 | return size; | |
fd88de56 SW |
532 | } |
533 | ||
fd88de56 | 534 | /** |
d4388340 | 535 | * gfs2_readahead - Read a bunch of pages at once |
c551f66c | 536 | * @rac: Read-ahead control structure |
fd88de56 SW |
537 | * |
538 | * Some notes: | |
539 | * 1. This is only for readahead, so we can simply ignore any things | |
540 | * which are slightly inconvenient (such as locking conflicts between | |
541 | * the page lock and the glock) and return having done no I/O. Its | |
542 | * obviously not something we'd want to do on too regular a basis. | |
543 | * Any I/O we ignore at this time will be done via readpage later. | |
e1d5b18a | 544 | * 2. We don't handle stuffed files here we let readpage do the honours. |
d4388340 | 545 | * 3. mpage_readahead() does most of the heavy lifting in the common case. |
e9e1ef2b | 546 | * 4. gfs2_block_map() is relied upon to set BH_Boundary in the right places. |
fd88de56 | 547 | */ |
3cc3f710 | 548 | |
d4388340 | 549 | static void gfs2_readahead(struct readahead_control *rac) |
fd88de56 | 550 | { |
d4388340 | 551 | struct inode *inode = rac->mapping->host; |
feaa7bba | 552 | struct gfs2_inode *ip = GFS2_I(inode); |
fd88de56 | 553 | |
2164f9b9 CH |
554 | if (gfs2_is_stuffed(ip)) |
555 | ; | |
556 | else if (gfs2_is_jdata(ip)) | |
d4388340 | 557 | mpage_readahead(rac, gfs2_block_map); |
2164f9b9 CH |
558 | else |
559 | iomap_readahead(rac, &gfs2_iomap_ops); | |
b3b94faa DT |
560 | } |
561 | ||
7ae8fa84 RP |
562 | /** |
563 | * adjust_fs_space - Adjusts the free space available due to gfs2_grow | |
564 | * @inode: the rindex inode | |
565 | */ | |
64bc06bb | 566 | void adjust_fs_space(struct inode *inode) |
7ae8fa84 | 567 | { |
d0a22a4b | 568 | struct gfs2_sbd *sdp = GFS2_SB(inode); |
1946f70a | 569 | struct gfs2_inode *m_ip = GFS2_I(sdp->sd_statfs_inode); |
7ae8fa84 RP |
570 | struct gfs2_statfs_change_host *m_sc = &sdp->sd_statfs_master; |
571 | struct gfs2_statfs_change_host *l_sc = &sdp->sd_statfs_local; | |
70c11ba8 | 572 | struct buffer_head *m_bh; |
7ae8fa84 RP |
573 | u64 fs_total, new_free; |
574 | ||
d0a22a4b AG |
575 | if (gfs2_trans_begin(sdp, 2 * RES_STATFS, 0) != 0) |
576 | return; | |
577 | ||
7ae8fa84 RP |
578 | /* Total up the file system space, according to the latest rindex. */ |
579 | fs_total = gfs2_ri_total(sdp); | |
1946f70a | 580 | if (gfs2_meta_inode_buffer(m_ip, &m_bh) != 0) |
d0a22a4b | 581 | goto out; |
7ae8fa84 RP |
582 | |
583 | spin_lock(&sdp->sd_statfs_spin); | |
1946f70a BM |
584 | gfs2_statfs_change_in(m_sc, m_bh->b_data + |
585 | sizeof(struct gfs2_dinode)); | |
7ae8fa84 RP |
586 | if (fs_total > (m_sc->sc_total + l_sc->sc_total)) |
587 | new_free = fs_total - (m_sc->sc_total + l_sc->sc_total); | |
588 | else | |
589 | new_free = 0; | |
590 | spin_unlock(&sdp->sd_statfs_spin); | |
6c53267f RP |
591 | fs_warn(sdp, "File system extended by %llu blocks.\n", |
592 | (unsigned long long)new_free); | |
7ae8fa84 | 593 | gfs2_statfs_change(sdp, new_free, new_free, 0); |
1946f70a | 594 | |
70c11ba8 | 595 | update_statfs(sdp, m_bh); |
1946f70a | 596 | brelse(m_bh); |
d0a22a4b AG |
597 | out: |
598 | sdp->sd_rindex_uptodate = 0; | |
599 | gfs2_trans_end(sdp); | |
7ae8fa84 RP |
600 | } |
601 | ||
e621900a MWO |
602 | static bool jdata_dirty_folio(struct address_space *mapping, |
603 | struct folio *folio) | |
8fb68595 | 604 | { |
6302d6f4 | 605 | if (current->journal_info) |
e621900a MWO |
606 | folio_set_checked(folio); |
607 | return block_dirty_folio(mapping, folio); | |
8fb68595 RP |
608 | } |
609 | ||
b3b94faa DT |
610 | /** |
611 | * gfs2_bmap - Block map function | |
612 | * @mapping: Address space info | |
613 | * @lblock: The block to map | |
614 | * | |
615 | * Returns: The disk address for the block or 0 on hole or error | |
616 | */ | |
617 | ||
618 | static sector_t gfs2_bmap(struct address_space *mapping, sector_t lblock) | |
619 | { | |
feaa7bba | 620 | struct gfs2_inode *ip = GFS2_I(mapping->host); |
b3b94faa DT |
621 | struct gfs2_holder i_gh; |
622 | sector_t dblock = 0; | |
623 | int error; | |
624 | ||
b3b94faa DT |
625 | error = gfs2_glock_nq_init(ip->i_gl, LM_ST_SHARED, LM_FLAG_ANY, &i_gh); |
626 | if (error) | |
627 | return 0; | |
628 | ||
629 | if (!gfs2_is_stuffed(ip)) | |
7770c93a | 630 | dblock = iomap_bmap(mapping, lblock, &gfs2_iomap_ops); |
b3b94faa DT |
631 | |
632 | gfs2_glock_dq_uninit(&i_gh); | |
633 | ||
634 | return dblock; | |
635 | } | |
636 | ||
d7b616e2 SW |
637 | static void gfs2_discard(struct gfs2_sbd *sdp, struct buffer_head *bh) |
638 | { | |
639 | struct gfs2_bufdata *bd; | |
640 | ||
641 | lock_buffer(bh); | |
642 | gfs2_log_lock(sdp); | |
643 | clear_buffer_dirty(bh); | |
644 | bd = bh->b_private; | |
645 | if (bd) { | |
c0752aa7 BP |
646 | if (!list_empty(&bd->bd_list) && !buffer_pinned(bh)) |
647 | list_del_init(&bd->bd_list); | |
68942870 BP |
648 | else { |
649 | spin_lock(&sdp->sd_ail_lock); | |
68cd4ce2 | 650 | gfs2_remove_from_journal(bh, REMOVE_JDATA); |
68942870 BP |
651 | spin_unlock(&sdp->sd_ail_lock); |
652 | } | |
d7b616e2 SW |
653 | } |
654 | bh->b_bdev = NULL; | |
655 | clear_buffer_mapped(bh); | |
656 | clear_buffer_req(bh); | |
657 | clear_buffer_new(bh); | |
658 | gfs2_log_unlock(sdp); | |
659 | unlock_buffer(bh); | |
660 | } | |
661 | ||
5f4b2976 MWO |
662 | static void gfs2_invalidate_folio(struct folio *folio, size_t offset, |
663 | size_t length) | |
b3b94faa | 664 | { |
5f4b2976 MWO |
665 | struct gfs2_sbd *sdp = GFS2_SB(folio->mapping->host); |
666 | size_t stop = offset + length; | |
667 | int partial_page = (offset || length < folio_size(folio)); | |
d7b616e2 SW |
668 | struct buffer_head *bh, *head; |
669 | unsigned long pos = 0; | |
670 | ||
5f4b2976 | 671 | BUG_ON(!folio_test_locked(folio)); |
5c0bb97c | 672 | if (!partial_page) |
5f4b2976 MWO |
673 | folio_clear_checked(folio); |
674 | head = folio_buffers(folio); | |
675 | if (!head) | |
d7b616e2 | 676 | goto out; |
b3b94faa | 677 | |
5f4b2976 | 678 | bh = head; |
d7b616e2 | 679 | do { |
5c0bb97c LC |
680 | if (pos + bh->b_size > stop) |
681 | return; | |
682 | ||
d7b616e2 SW |
683 | if (offset <= pos) |
684 | gfs2_discard(sdp, bh); | |
685 | pos += bh->b_size; | |
686 | bh = bh->b_this_page; | |
687 | } while (bh != head); | |
688 | out: | |
5c0bb97c | 689 | if (!partial_page) |
5f4b2976 | 690 | filemap_release_folio(folio, 0); |
b3b94faa DT |
691 | } |
692 | ||
4340fe62 | 693 | /** |
e45c20d1 MWO |
694 | * gfs2_release_folio - free the metadata associated with a folio |
695 | * @folio: the folio that's being released | |
4340fe62 SW |
696 | * @gfp_mask: passed from Linux VFS, ignored by us |
697 | * | |
e45c20d1 | 698 | * Calls try_to_free_buffers() to free the buffers and put the folio if the |
0ebbe4f9 | 699 | * buffers can be released. |
4340fe62 | 700 | * |
e45c20d1 | 701 | * Returns: true if the folio was put or else false |
4340fe62 SW |
702 | */ |
703 | ||
e45c20d1 | 704 | bool gfs2_release_folio(struct folio *folio, gfp_t gfp_mask) |
4340fe62 | 705 | { |
e45c20d1 | 706 | struct address_space *mapping = folio->mapping; |
009d8518 | 707 | struct gfs2_sbd *sdp = gfs2_mapping2sbd(mapping); |
4340fe62 SW |
708 | struct buffer_head *bh, *head; |
709 | struct gfs2_bufdata *bd; | |
4340fe62 | 710 | |
e45c20d1 MWO |
711 | head = folio_buffers(folio); |
712 | if (!head) | |
713 | return false; | |
4340fe62 | 714 | |
1c185c02 | 715 | /* |
e45c20d1 MWO |
716 | * mm accommodates an old ext3 case where clean folios might |
717 | * not have had the dirty bit cleared. Thus, it can send actual | |
718 | * dirty folios to ->release_folio() via shrink_active_list(). | |
1c185c02 | 719 | * |
e45c20d1 MWO |
720 | * As a workaround, we skip folios that contain dirty buffers |
721 | * below. Once ->release_folio isn't called on dirty folios | |
722 | * anymore, we can warn on dirty buffers like we used to here | |
723 | * again. | |
1c185c02 AG |
724 | */ |
725 | ||
bb3b0e3d | 726 | gfs2_log_lock(sdp); |
e45c20d1 | 727 | bh = head; |
4340fe62 | 728 | do { |
bb3b0e3d SW |
729 | if (atomic_read(&bh->b_count)) |
730 | goto cannot_release; | |
731 | bd = bh->b_private; | |
16ca9412 | 732 | if (bd && bd->bd_tr) |
bb3b0e3d | 733 | goto cannot_release; |
1c185c02 AG |
734 | if (buffer_dirty(bh) || WARN_ON(buffer_pinned(bh))) |
735 | goto cannot_release; | |
bb3b0e3d | 736 | bh = bh->b_this_page; |
e45c20d1 | 737 | } while (bh != head); |
4340fe62 | 738 | |
e45c20d1 | 739 | bh = head; |
bb3b0e3d | 740 | do { |
4340fe62 SW |
741 | bd = bh->b_private; |
742 | if (bd) { | |
743 | gfs2_assert_warn(sdp, bd->bd_bh == bh); | |
e4f29206 | 744 | bd->bd_bh = NULL; |
4340fe62 | 745 | bh->b_private = NULL; |
019dd669 BP |
746 | /* |
747 | * The bd may still be queued as a revoke, in which | |
748 | * case we must not dequeue nor free it. | |
749 | */ | |
750 | if (!bd->bd_blkno && !list_empty(&bd->bd_list)) | |
751 | list_del_init(&bd->bd_list); | |
752 | if (list_empty(&bd->bd_list)) | |
753 | kmem_cache_free(gfs2_bufdata_cachep, bd); | |
e4f29206 | 754 | } |
4340fe62 SW |
755 | |
756 | bh = bh->b_this_page; | |
166afccd | 757 | } while (bh != head); |
e4f29206 | 758 | gfs2_log_unlock(sdp); |
4340fe62 | 759 | |
68189fef | 760 | return try_to_free_buffers(folio); |
8f065d36 | 761 | |
bb3b0e3d SW |
762 | cannot_release: |
763 | gfs2_log_unlock(sdp); | |
e45c20d1 | 764 | return false; |
4340fe62 SW |
765 | } |
766 | ||
eadd7535 | 767 | static const struct address_space_operations gfs2_aops = { |
9d358143 | 768 | .writepage = gfs2_writepage, |
45138990 | 769 | .writepages = gfs2_writepages, |
f132ab7d | 770 | .read_folio = gfs2_read_folio, |
d4388340 | 771 | .readahead = gfs2_readahead, |
187c82cb | 772 | .dirty_folio = filemap_dirty_folio, |
8597447d | 773 | .release_folio = iomap_release_folio, |
d82354f6 | 774 | .invalidate_folio = iomap_invalidate_folio, |
5561093e | 775 | .bmap = gfs2_bmap, |
967bcc91 | 776 | .direct_IO = noop_direct_IO, |
2ec810d5 | 777 | .migrate_folio = filemap_migrate_folio, |
2164f9b9 | 778 | .is_partially_uptodate = iomap_is_partially_uptodate, |
aa261f54 | 779 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
780 | }; |
781 | ||
5561093e | 782 | static const struct address_space_operations gfs2_jdata_aops = { |
9ff8ec32 | 783 | .writepage = gfs2_jdata_writepage, |
b8e7cbb6 | 784 | .writepages = gfs2_jdata_writepages, |
f132ab7d | 785 | .read_folio = gfs2_read_folio, |
d4388340 | 786 | .readahead = gfs2_readahead, |
e621900a | 787 | .dirty_folio = jdata_dirty_folio, |
5561093e | 788 | .bmap = gfs2_bmap, |
5f4b2976 | 789 | .invalidate_folio = gfs2_invalidate_folio, |
e45c20d1 | 790 | .release_folio = gfs2_release_folio, |
229615de | 791 | .is_partially_uptodate = block_is_partially_uptodate, |
aa261f54 | 792 | .error_remove_page = generic_error_remove_page, |
5561093e SW |
793 | }; |
794 | ||
795 | void gfs2_set_aops(struct inode *inode) | |
796 | { | |
eadd7535 | 797 | if (gfs2_is_jdata(GFS2_I(inode))) |
977767a7 | 798 | inode->i_mapping->a_ops = &gfs2_jdata_aops; |
5561093e | 799 | else |
eadd7535 | 800 | inode->i_mapping->a_ops = &gfs2_aops; |
5561093e | 801 | } |