Commit | Line | Data |
---|---|---|
457c8996 | 1 | // SPDX-License-Identifier: GPL-2.0-only |
1da177e4 LT |
2 | /* |
3 | * linux/fs/buffer.c | |
4 | * | |
5 | * Copyright (C) 1991, 1992, 2002 Linus Torvalds | |
6 | */ | |
7 | ||
8 | /* | |
9 | * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95 | |
10 | * | |
11 | * Removed a lot of unnecessary code and simplified things now that | |
12 | * the buffer cache isn't our primary cache - Andrew Tridgell 12/96 | |
13 | * | |
14 | * Speed up hash, lru, and free list operations. Use gfp() for allocating | |
15 | * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM | |
16 | * | |
17 | * Added 32k buffer block sizes - these are required older ARM systems. - RMK | |
18 | * | |
19 | * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de> | |
20 | */ | |
21 | ||
1da177e4 | 22 | #include <linux/kernel.h> |
f361bf4a | 23 | #include <linux/sched/signal.h> |
1da177e4 LT |
24 | #include <linux/syscalls.h> |
25 | #include <linux/fs.h> | |
ae259a9c | 26 | #include <linux/iomap.h> |
1da177e4 LT |
27 | #include <linux/mm.h> |
28 | #include <linux/percpu.h> | |
29 | #include <linux/slab.h> | |
16f7e0fe | 30 | #include <linux/capability.h> |
1da177e4 LT |
31 | #include <linux/blkdev.h> |
32 | #include <linux/file.h> | |
33 | #include <linux/quotaops.h> | |
34 | #include <linux/highmem.h> | |
630d9c47 | 35 | #include <linux/export.h> |
bafc0dba | 36 | #include <linux/backing-dev.h> |
1da177e4 LT |
37 | #include <linux/writeback.h> |
38 | #include <linux/hash.h> | |
39 | #include <linux/suspend.h> | |
40 | #include <linux/buffer_head.h> | |
55e829af | 41 | #include <linux/task_io_accounting_ops.h> |
1da177e4 | 42 | #include <linux/bio.h> |
1da177e4 LT |
43 | #include <linux/cpu.h> |
44 | #include <linux/bitops.h> | |
45 | #include <linux/mpage.h> | |
fb1c8f93 | 46 | #include <linux/bit_spinlock.h> |
29f3ad7d | 47 | #include <linux/pagevec.h> |
f745c6f5 | 48 | #include <linux/sched/mm.h> |
5305cb83 | 49 | #include <trace/events/block.h> |
31fb992c | 50 | #include <linux/fscrypt.h> |
4fa512ce | 51 | #include <linux/fsverity.h> |
1da177e4 | 52 | |
2b211dc0 BD |
53 | #include "internal.h" |
54 | ||
1da177e4 | 55 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); |
5bdf402a RHI |
56 | static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, |
57 | struct writeback_control *wbc); | |
1da177e4 LT |
58 | |
59 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) | |
60 | ||
f0059afd TH |
61 | inline void touch_buffer(struct buffer_head *bh) |
62 | { | |
5305cb83 | 63 | trace_block_touch_buffer(bh); |
03c5f331 | 64 | folio_mark_accessed(bh->b_folio); |
f0059afd TH |
65 | } |
66 | EXPORT_SYMBOL(touch_buffer); | |
67 | ||
fc9b52cd | 68 | void __lock_buffer(struct buffer_head *bh) |
1da177e4 | 69 | { |
74316201 | 70 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 LT |
71 | } |
72 | EXPORT_SYMBOL(__lock_buffer); | |
73 | ||
fc9b52cd | 74 | void unlock_buffer(struct buffer_head *bh) |
1da177e4 | 75 | { |
51b07fc3 | 76 | clear_bit_unlock(BH_Lock, &bh->b_state); |
4e857c58 | 77 | smp_mb__after_atomic(); |
1da177e4 LT |
78 | wake_up_bit(&bh->b_state, BH_Lock); |
79 | } | |
1fe72eaa | 80 | EXPORT_SYMBOL(unlock_buffer); |
1da177e4 | 81 | |
b4597226 | 82 | /* |
520f301c MWO |
83 | * Returns if the folio has dirty or writeback buffers. If all the buffers |
84 | * are unlocked and clean then the folio_test_dirty information is stale. If | |
85 | * any of the buffers are locked, it is assumed they are locked for IO. | |
b4597226 | 86 | */ |
520f301c | 87 | void buffer_check_dirty_writeback(struct folio *folio, |
b4597226 MG |
88 | bool *dirty, bool *writeback) |
89 | { | |
90 | struct buffer_head *head, *bh; | |
91 | *dirty = false; | |
92 | *writeback = false; | |
93 | ||
520f301c | 94 | BUG_ON(!folio_test_locked(folio)); |
b4597226 | 95 | |
520f301c MWO |
96 | head = folio_buffers(folio); |
97 | if (!head) | |
b4597226 MG |
98 | return; |
99 | ||
520f301c | 100 | if (folio_test_writeback(folio)) |
b4597226 MG |
101 | *writeback = true; |
102 | ||
b4597226 MG |
103 | bh = head; |
104 | do { | |
105 | if (buffer_locked(bh)) | |
106 | *writeback = true; | |
107 | ||
108 | if (buffer_dirty(bh)) | |
109 | *dirty = true; | |
110 | ||
111 | bh = bh->b_this_page; | |
112 | } while (bh != head); | |
113 | } | |
114 | EXPORT_SYMBOL(buffer_check_dirty_writeback); | |
115 | ||
1da177e4 LT |
116 | /* |
117 | * Block until a buffer comes unlocked. This doesn't stop it | |
118 | * from becoming locked again - you have to lock it yourself | |
119 | * if you want to preserve its state. | |
120 | */ | |
121 | void __wait_on_buffer(struct buffer_head * bh) | |
122 | { | |
74316201 | 123 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 | 124 | } |
1fe72eaa | 125 | EXPORT_SYMBOL(__wait_on_buffer); |
1da177e4 | 126 | |
b744c2ac | 127 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
1da177e4 | 128 | { |
432f16e6 RE |
129 | if (!test_bit(BH_Quiet, &bh->b_state)) |
130 | printk_ratelimited(KERN_ERR | |
a1c6f057 DM |
131 | "Buffer I/O error on dev %pg, logical block %llu%s\n", |
132 | bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); | |
1da177e4 LT |
133 | } |
134 | ||
135 | /* | |
68671f35 DM |
136 | * End-of-IO handler helper function which does not touch the bh after |
137 | * unlocking it. | |
138 | * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but | |
139 | * a race there is benign: unlock_buffer() only use the bh's address for | |
140 | * hashing after unlocking the buffer, so it doesn't actually touch the bh | |
141 | * itself. | |
1da177e4 | 142 | */ |
68671f35 | 143 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
1da177e4 LT |
144 | { |
145 | if (uptodate) { | |
146 | set_buffer_uptodate(bh); | |
147 | } else { | |
70246286 | 148 | /* This happens, due to failed read-ahead attempts. */ |
1da177e4 LT |
149 | clear_buffer_uptodate(bh); |
150 | } | |
151 | unlock_buffer(bh); | |
68671f35 DM |
152 | } |
153 | ||
154 | /* | |
155 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and | |
79f59784 | 156 | * unlock the buffer. |
68671f35 DM |
157 | */ |
158 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) | |
159 | { | |
160 | __end_buffer_read_notouch(bh, uptodate); | |
1da177e4 LT |
161 | put_bh(bh); |
162 | } | |
1fe72eaa | 163 | EXPORT_SYMBOL(end_buffer_read_sync); |
1da177e4 LT |
164 | |
165 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) | |
166 | { | |
1da177e4 LT |
167 | if (uptodate) { |
168 | set_buffer_uptodate(bh); | |
169 | } else { | |
432f16e6 | 170 | buffer_io_error(bh, ", lost sync page write"); |
87354e5d | 171 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
172 | clear_buffer_uptodate(bh); |
173 | } | |
174 | unlock_buffer(bh); | |
175 | put_bh(bh); | |
176 | } | |
1fe72eaa | 177 | EXPORT_SYMBOL(end_buffer_write_sync); |
1da177e4 | 178 | |
1da177e4 LT |
179 | /* |
180 | * Various filesystems appear to want __find_get_block to be non-blocking. | |
181 | * But it's the page lock which protects the buffers. To get around this, | |
182 | * we get exclusion from try_to_free_buffers with the blockdev mapping's | |
183 | * private_lock. | |
184 | * | |
b93b0163 | 185 | * Hack idea: for the blockdev mapping, private_lock contention |
1da177e4 | 186 | * may be quite high. This code could TryLock the page, and if that |
b93b0163 | 187 | * succeeds, there is no need to take private_lock. |
1da177e4 LT |
188 | */ |
189 | static struct buffer_head * | |
385fd4c5 | 190 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
1da177e4 LT |
191 | { |
192 | struct inode *bd_inode = bdev->bd_inode; | |
193 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
194 | struct buffer_head *ret = NULL; | |
195 | pgoff_t index; | |
196 | struct buffer_head *bh; | |
197 | struct buffer_head *head; | |
198 | struct page *page; | |
199 | int all_mapped = 1; | |
43636c80 | 200 | static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); |
1da177e4 | 201 | |
09cbfeaf | 202 | index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); |
2457aec6 | 203 | page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); |
1da177e4 LT |
204 | if (!page) |
205 | goto out; | |
206 | ||
207 | spin_lock(&bd_mapping->private_lock); | |
208 | if (!page_has_buffers(page)) | |
209 | goto out_unlock; | |
210 | head = page_buffers(page); | |
211 | bh = head; | |
212 | do { | |
97f76d3d NK |
213 | if (!buffer_mapped(bh)) |
214 | all_mapped = 0; | |
215 | else if (bh->b_blocknr == block) { | |
1da177e4 LT |
216 | ret = bh; |
217 | get_bh(bh); | |
218 | goto out_unlock; | |
219 | } | |
1da177e4 LT |
220 | bh = bh->b_this_page; |
221 | } while (bh != head); | |
222 | ||
223 | /* we might be here because some of the buffers on this page are | |
224 | * not mapped. This is due to various races between | |
225 | * file io on the block device and getblk. It gets dealt with | |
226 | * elsewhere, don't buffer_error if we had some unmapped buffers | |
227 | */ | |
43636c80 TH |
228 | ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); |
229 | if (all_mapped && __ratelimit(&last_warned)) { | |
230 | printk("__find_get_block_slow() failed. block=%llu, " | |
231 | "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " | |
232 | "device %pg blocksize: %d\n", | |
233 | (unsigned long long)block, | |
234 | (unsigned long long)bh->b_blocknr, | |
235 | bh->b_state, bh->b_size, bdev, | |
236 | 1 << bd_inode->i_blkbits); | |
1da177e4 LT |
237 | } |
238 | out_unlock: | |
239 | spin_unlock(&bd_mapping->private_lock); | |
09cbfeaf | 240 | put_page(page); |
1da177e4 LT |
241 | out: |
242 | return ret; | |
243 | } | |
244 | ||
1da177e4 LT |
245 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) |
246 | { | |
1da177e4 | 247 | unsigned long flags; |
a3972203 | 248 | struct buffer_head *first; |
1da177e4 | 249 | struct buffer_head *tmp; |
2e2dba15 MWO |
250 | struct folio *folio; |
251 | int folio_uptodate = 1; | |
1da177e4 LT |
252 | |
253 | BUG_ON(!buffer_async_read(bh)); | |
254 | ||
2e2dba15 | 255 | folio = bh->b_folio; |
1da177e4 LT |
256 | if (uptodate) { |
257 | set_buffer_uptodate(bh); | |
258 | } else { | |
259 | clear_buffer_uptodate(bh); | |
432f16e6 | 260 | buffer_io_error(bh, ", async page read"); |
2e2dba15 | 261 | folio_set_error(folio); |
1da177e4 LT |
262 | } |
263 | ||
264 | /* | |
265 | * Be _very_ careful from here on. Bad things can happen if | |
266 | * two buffer heads end IO at almost the same time and both | |
267 | * decide that the page is now completely done. | |
268 | */ | |
2e2dba15 | 269 | first = folio_buffers(folio); |
f1e67e35 | 270 | spin_lock_irqsave(&first->b_uptodate_lock, flags); |
1da177e4 LT |
271 | clear_buffer_async_read(bh); |
272 | unlock_buffer(bh); | |
273 | tmp = bh; | |
274 | do { | |
275 | if (!buffer_uptodate(tmp)) | |
2e2dba15 | 276 | folio_uptodate = 0; |
1da177e4 LT |
277 | if (buffer_async_read(tmp)) { |
278 | BUG_ON(!buffer_locked(tmp)); | |
279 | goto still_busy; | |
280 | } | |
281 | tmp = tmp->b_this_page; | |
282 | } while (tmp != bh); | |
f1e67e35 | 283 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
284 | |
285 | /* | |
6e8e79fc MWO |
286 | * If all of the buffers are uptodate then we can set the page |
287 | * uptodate. | |
1da177e4 | 288 | */ |
2e2dba15 MWO |
289 | if (folio_uptodate) |
290 | folio_mark_uptodate(folio); | |
291 | folio_unlock(folio); | |
1da177e4 LT |
292 | return; |
293 | ||
294 | still_busy: | |
f1e67e35 | 295 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
296 | return; |
297 | } | |
298 | ||
4fa512ce | 299 | struct postprocess_bh_ctx { |
31fb992c EB |
300 | struct work_struct work; |
301 | struct buffer_head *bh; | |
302 | }; | |
303 | ||
4fa512ce EB |
304 | static void verify_bh(struct work_struct *work) |
305 | { | |
306 | struct postprocess_bh_ctx *ctx = | |
307 | container_of(work, struct postprocess_bh_ctx, work); | |
308 | struct buffer_head *bh = ctx->bh; | |
309 | bool valid; | |
310 | ||
5d0f0e57 EB |
311 | valid = fsverity_verify_blocks(page_folio(bh->b_page), bh->b_size, |
312 | bh_offset(bh)); | |
4fa512ce EB |
313 | end_buffer_async_read(bh, valid); |
314 | kfree(ctx); | |
315 | } | |
316 | ||
317 | static bool need_fsverity(struct buffer_head *bh) | |
318 | { | |
319 | struct page *page = bh->b_page; | |
320 | struct inode *inode = page->mapping->host; | |
321 | ||
322 | return fsverity_active(inode) && | |
323 | /* needed by ext4 */ | |
324 | page->index < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); | |
325 | } | |
326 | ||
31fb992c EB |
327 | static void decrypt_bh(struct work_struct *work) |
328 | { | |
4fa512ce EB |
329 | struct postprocess_bh_ctx *ctx = |
330 | container_of(work, struct postprocess_bh_ctx, work); | |
31fb992c EB |
331 | struct buffer_head *bh = ctx->bh; |
332 | int err; | |
333 | ||
51e4e315 EB |
334 | err = fscrypt_decrypt_pagecache_blocks(page_folio(bh->b_page), |
335 | bh->b_size, bh_offset(bh)); | |
4fa512ce EB |
336 | if (err == 0 && need_fsverity(bh)) { |
337 | /* | |
338 | * We use different work queues for decryption and for verity | |
339 | * because verity may require reading metadata pages that need | |
340 | * decryption, and we shouldn't recurse to the same workqueue. | |
341 | */ | |
342 | INIT_WORK(&ctx->work, verify_bh); | |
343 | fsverity_enqueue_verify_work(&ctx->work); | |
344 | return; | |
345 | } | |
31fb992c EB |
346 | end_buffer_async_read(bh, err == 0); |
347 | kfree(ctx); | |
348 | } | |
349 | ||
350 | /* | |
2c69e205 | 351 | * I/O completion handler for block_read_full_folio() - pages |
31fb992c EB |
352 | * which come unlocked at the end of I/O. |
353 | */ | |
354 | static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate) | |
355 | { | |
3822a7c4 | 356 | struct inode *inode = bh->b_folio->mapping->host; |
4fa512ce EB |
357 | bool decrypt = fscrypt_inode_uses_fs_layer_crypto(inode); |
358 | bool verify = need_fsverity(bh); | |
359 | ||
360 | /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */ | |
361 | if (uptodate && (decrypt || verify)) { | |
362 | struct postprocess_bh_ctx *ctx = | |
363 | kmalloc(sizeof(*ctx), GFP_ATOMIC); | |
31fb992c EB |
364 | |
365 | if (ctx) { | |
31fb992c | 366 | ctx->bh = bh; |
4fa512ce EB |
367 | if (decrypt) { |
368 | INIT_WORK(&ctx->work, decrypt_bh); | |
369 | fscrypt_enqueue_decrypt_work(&ctx->work); | |
370 | } else { | |
371 | INIT_WORK(&ctx->work, verify_bh); | |
372 | fsverity_enqueue_verify_work(&ctx->work); | |
373 | } | |
31fb992c EB |
374 | return; |
375 | } | |
376 | uptodate = 0; | |
377 | } | |
378 | end_buffer_async_read(bh, uptodate); | |
379 | } | |
380 | ||
1da177e4 LT |
381 | /* |
382 | * Completion handler for block_write_full_page() - pages which are unlocked | |
383 | * during I/O, and which have PageWriteback cleared upon I/O completion. | |
384 | */ | |
35c80d5f | 385 | void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
1da177e4 | 386 | { |
1da177e4 | 387 | unsigned long flags; |
a3972203 | 388 | struct buffer_head *first; |
1da177e4 | 389 | struct buffer_head *tmp; |
743ed81e | 390 | struct folio *folio; |
1da177e4 LT |
391 | |
392 | BUG_ON(!buffer_async_write(bh)); | |
393 | ||
743ed81e | 394 | folio = bh->b_folio; |
1da177e4 LT |
395 | if (uptodate) { |
396 | set_buffer_uptodate(bh); | |
397 | } else { | |
432f16e6 | 398 | buffer_io_error(bh, ", lost async page write"); |
87354e5d | 399 | mark_buffer_write_io_error(bh); |
1da177e4 | 400 | clear_buffer_uptodate(bh); |
743ed81e | 401 | folio_set_error(folio); |
1da177e4 LT |
402 | } |
403 | ||
743ed81e | 404 | first = folio_buffers(folio); |
f1e67e35 | 405 | spin_lock_irqsave(&first->b_uptodate_lock, flags); |
a3972203 | 406 | |
1da177e4 LT |
407 | clear_buffer_async_write(bh); |
408 | unlock_buffer(bh); | |
409 | tmp = bh->b_this_page; | |
410 | while (tmp != bh) { | |
411 | if (buffer_async_write(tmp)) { | |
412 | BUG_ON(!buffer_locked(tmp)); | |
413 | goto still_busy; | |
414 | } | |
415 | tmp = tmp->b_this_page; | |
416 | } | |
f1e67e35 | 417 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
743ed81e | 418 | folio_end_writeback(folio); |
1da177e4 LT |
419 | return; |
420 | ||
421 | still_busy: | |
f1e67e35 | 422 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
423 | return; |
424 | } | |
1fe72eaa | 425 | EXPORT_SYMBOL(end_buffer_async_write); |
1da177e4 LT |
426 | |
427 | /* | |
428 | * If a page's buffers are under async readin (end_buffer_async_read | |
429 | * completion) then there is a possibility that another thread of | |
430 | * control could lock one of the buffers after it has completed | |
431 | * but while some of the other buffers have not completed. This | |
432 | * locked buffer would confuse end_buffer_async_read() into not unlocking | |
433 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() | |
434 | * that this buffer is not under async I/O. | |
435 | * | |
436 | * The page comes unlocked when it has no locked buffer_async buffers | |
437 | * left. | |
438 | * | |
439 | * PageLocked prevents anyone starting new async I/O reads any of | |
440 | * the buffers. | |
441 | * | |
442 | * PageWriteback is used to prevent simultaneous writeout of the same | |
443 | * page. | |
444 | * | |
445 | * PageLocked prevents anyone from starting writeback of a page which is | |
446 | * under read I/O (PageWriteback is only ever set against a locked page). | |
447 | */ | |
448 | static void mark_buffer_async_read(struct buffer_head *bh) | |
449 | { | |
31fb992c | 450 | bh->b_end_io = end_buffer_async_read_io; |
1da177e4 LT |
451 | set_buffer_async_read(bh); |
452 | } | |
453 | ||
1fe72eaa HS |
454 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
455 | bh_end_io_t *handler) | |
1da177e4 | 456 | { |
35c80d5f | 457 | bh->b_end_io = handler; |
1da177e4 LT |
458 | set_buffer_async_write(bh); |
459 | } | |
35c80d5f CM |
460 | |
461 | void mark_buffer_async_write(struct buffer_head *bh) | |
462 | { | |
463 | mark_buffer_async_write_endio(bh, end_buffer_async_write); | |
464 | } | |
1da177e4 LT |
465 | EXPORT_SYMBOL(mark_buffer_async_write); |
466 | ||
467 | ||
468 | /* | |
469 | * fs/buffer.c contains helper functions for buffer-backed address space's | |
470 | * fsync functions. A common requirement for buffer-based filesystems is | |
471 | * that certain data from the backing blockdev needs to be written out for | |
472 | * a successful fsync(). For example, ext2 indirect blocks need to be | |
473 | * written back and waited upon before fsync() returns. | |
474 | * | |
475 | * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), | |
476 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the | |
477 | * management of a list of dependent buffers at ->i_mapping->private_list. | |
478 | * | |
479 | * Locking is a little subtle: try_to_free_buffers() will remove buffers | |
480 | * from their controlling inode's queue when they are being freed. But | |
481 | * try_to_free_buffers() will be operating against the *blockdev* mapping | |
482 | * at the time, not against the S_ISREG file which depends on those buffers. | |
483 | * So the locking for private_list is via the private_lock in the address_space | |
484 | * which backs the buffers. Which is different from the address_space | |
485 | * against which the buffers are listed. So for a particular address_space, | |
486 | * mapping->private_lock does *not* protect mapping->private_list! In fact, | |
487 | * mapping->private_list will always be protected by the backing blockdev's | |
488 | * ->private_lock. | |
489 | * | |
490 | * Which introduces a requirement: all buffers on an address_space's | |
491 | * ->private_list must be from the same address_space: the blockdev's. | |
492 | * | |
493 | * address_spaces which do not place buffers at ->private_list via these | |
494 | * utility functions are free to use private_lock and private_list for | |
495 | * whatever they want. The only requirement is that list_empty(private_list) | |
496 | * be true at clear_inode() time. | |
497 | * | |
498 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The | |
499 | * filesystems should do that. invalidate_inode_buffers() should just go | |
500 | * BUG_ON(!list_empty). | |
501 | * | |
502 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should | |
503 | * take an address_space, not an inode. And it should be called | |
504 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being | |
505 | * queued up. | |
506 | * | |
507 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the | |
508 | * list if it is already on a list. Because if the buffer is on a list, | |
509 | * it *must* already be on the right one. If not, the filesystem is being | |
510 | * silly. This will save a ton of locking. But first we have to ensure | |
511 | * that buffers are taken *off* the old inode's list when they are freed | |
512 | * (presumably in truncate). That requires careful auditing of all | |
513 | * filesystems (do it inside bforget()). It could also be done by bringing | |
514 | * b_inode back. | |
515 | */ | |
516 | ||
517 | /* | |
518 | * The buffer's backing address_space's private_lock must be held | |
519 | */ | |
dbacefc9 | 520 | static void __remove_assoc_queue(struct buffer_head *bh) |
1da177e4 LT |
521 | { |
522 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 523 | WARN_ON(!bh->b_assoc_map); |
58ff407b | 524 | bh->b_assoc_map = NULL; |
1da177e4 LT |
525 | } |
526 | ||
527 | int inode_has_buffers(struct inode *inode) | |
528 | { | |
529 | return !list_empty(&inode->i_data.private_list); | |
530 | } | |
531 | ||
532 | /* | |
533 | * osync is designed to support O_SYNC io. It waits synchronously for | |
534 | * all already-submitted IO to complete, but does not queue any new | |
535 | * writes to the disk. | |
536 | * | |
79f59784 ZY |
537 | * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer |
538 | * as you dirty the buffers, and then use osync_inode_buffers to wait for | |
1da177e4 LT |
539 | * completion. Any other dirty buffers which are not yet queued for |
540 | * write will not be flushed to disk by the osync. | |
541 | */ | |
542 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) | |
543 | { | |
544 | struct buffer_head *bh; | |
545 | struct list_head *p; | |
546 | int err = 0; | |
547 | ||
548 | spin_lock(lock); | |
549 | repeat: | |
550 | list_for_each_prev(p, list) { | |
551 | bh = BH_ENTRY(p); | |
552 | if (buffer_locked(bh)) { | |
553 | get_bh(bh); | |
554 | spin_unlock(lock); | |
555 | wait_on_buffer(bh); | |
556 | if (!buffer_uptodate(bh)) | |
557 | err = -EIO; | |
558 | brelse(bh); | |
559 | spin_lock(lock); | |
560 | goto repeat; | |
561 | } | |
562 | } | |
563 | spin_unlock(lock); | |
564 | return err; | |
565 | } | |
566 | ||
08fdc8a0 | 567 | void emergency_thaw_bdev(struct super_block *sb) |
c2d75438 | 568 | { |
040f04bd | 569 | while (sb->s_bdev && !thaw_bdev(sb->s_bdev)) |
a1c6f057 | 570 | printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); |
01a05b33 | 571 | } |
c2d75438 | 572 | |
1da177e4 | 573 | /** |
78a4a50a | 574 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
67be2dd1 | 575 | * @mapping: the mapping which wants those buffers written |
1da177e4 LT |
576 | * |
577 | * Starts I/O against the buffers at mapping->private_list, and waits upon | |
578 | * that I/O. | |
579 | * | |
67be2dd1 MW |
580 | * Basically, this is a convenience function for fsync(). |
581 | * @mapping is a file or directory which needs those buffers to be written for | |
582 | * a successful fsync(). | |
1da177e4 LT |
583 | */ |
584 | int sync_mapping_buffers(struct address_space *mapping) | |
585 | { | |
252aa6f5 | 586 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
587 | |
588 | if (buffer_mapping == NULL || list_empty(&mapping->private_list)) | |
589 | return 0; | |
590 | ||
591 | return fsync_buffers_list(&buffer_mapping->private_lock, | |
592 | &mapping->private_list); | |
593 | } | |
594 | EXPORT_SYMBOL(sync_mapping_buffers); | |
595 | ||
596 | /* | |
597 | * Called when we've recently written block `bblock', and it is known that | |
598 | * `bblock' was for a buffer_boundary() buffer. This means that the block at | |
599 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's | |
600 | * dirty, schedule it for IO. So that indirects merge nicely with their data. | |
601 | */ | |
602 | void write_boundary_block(struct block_device *bdev, | |
603 | sector_t bblock, unsigned blocksize) | |
604 | { | |
605 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); | |
606 | if (bh) { | |
607 | if (buffer_dirty(bh)) | |
e7ea1129 | 608 | write_dirty_buffer(bh, 0); |
1da177e4 LT |
609 | put_bh(bh); |
610 | } | |
611 | } | |
612 | ||
613 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) | |
614 | { | |
615 | struct address_space *mapping = inode->i_mapping; | |
abc8a8a2 | 616 | struct address_space *buffer_mapping = bh->b_folio->mapping; |
1da177e4 LT |
617 | |
618 | mark_buffer_dirty(bh); | |
252aa6f5 RA |
619 | if (!mapping->private_data) { |
620 | mapping->private_data = buffer_mapping; | |
1da177e4 | 621 | } else { |
252aa6f5 | 622 | BUG_ON(mapping->private_data != buffer_mapping); |
1da177e4 | 623 | } |
535ee2fb | 624 | if (!bh->b_assoc_map) { |
1da177e4 LT |
625 | spin_lock(&buffer_mapping->private_lock); |
626 | list_move_tail(&bh->b_assoc_buffers, | |
627 | &mapping->private_list); | |
58ff407b | 628 | bh->b_assoc_map = mapping; |
1da177e4 LT |
629 | spin_unlock(&buffer_mapping->private_lock); |
630 | } | |
631 | } | |
632 | EXPORT_SYMBOL(mark_buffer_dirty_inode); | |
633 | ||
634 | /* | |
635 | * Add a page to the dirty page list. | |
636 | * | |
637 | * It is a sad fact of life that this function is called from several places | |
638 | * deeply under spinlocking. It may not sleep. | |
639 | * | |
640 | * If the page has buffers, the uptodate buffers are set dirty, to preserve | |
641 | * dirty-state coherency between the page and the buffers. It the page does | |
642 | * not have buffers then when they are later attached they will all be set | |
643 | * dirty. | |
644 | * | |
645 | * The buffers are dirtied before the page is dirtied. There's a small race | |
646 | * window in which a writepage caller may see the page cleanness but not the | |
647 | * buffer dirtiness. That's fine. If this code were to set the page dirty | |
648 | * before the buffers, a concurrent writepage caller could clear the page dirty | |
649 | * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean | |
650 | * page on the dirty page list. | |
651 | * | |
652 | * We use private_lock to lock against try_to_free_buffers while using the | |
653 | * page's buffer list. Also use this to protect against clean buffers being | |
654 | * added to the page after it was set dirty. | |
655 | * | |
656 | * FIXME: may need to call ->reservepage here as well. That's rather up to the | |
657 | * address_space though. | |
658 | */ | |
e621900a | 659 | bool block_dirty_folio(struct address_space *mapping, struct folio *folio) |
1da177e4 | 660 | { |
e621900a MWO |
661 | struct buffer_head *head; |
662 | bool newly_dirty; | |
1da177e4 LT |
663 | |
664 | spin_lock(&mapping->private_lock); | |
e621900a MWO |
665 | head = folio_buffers(folio); |
666 | if (head) { | |
1da177e4 LT |
667 | struct buffer_head *bh = head; |
668 | ||
669 | do { | |
670 | set_buffer_dirty(bh); | |
671 | bh = bh->b_this_page; | |
672 | } while (bh != head); | |
673 | } | |
c4843a75 | 674 | /* |
bcfe06bf | 675 | * Lock out page's memcg migration to keep PageDirty |
81f8c3a4 | 676 | * synchronized with per-memcg dirty page counters. |
c4843a75 | 677 | */ |
e621900a MWO |
678 | folio_memcg_lock(folio); |
679 | newly_dirty = !folio_test_set_dirty(folio); | |
1da177e4 LT |
680 | spin_unlock(&mapping->private_lock); |
681 | ||
a8e7d49a | 682 | if (newly_dirty) |
e621900a | 683 | __folio_mark_dirty(folio, mapping, 1); |
c4843a75 | 684 | |
e621900a | 685 | folio_memcg_unlock(folio); |
c4843a75 GT |
686 | |
687 | if (newly_dirty) | |
688 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
689 | ||
a8e7d49a | 690 | return newly_dirty; |
1da177e4 | 691 | } |
e621900a | 692 | EXPORT_SYMBOL(block_dirty_folio); |
1da177e4 LT |
693 | |
694 | /* | |
695 | * Write out and wait upon a list of buffers. | |
696 | * | |
697 | * We have conflicting pressures: we want to make sure that all | |
698 | * initially dirty buffers get waited on, but that any subsequently | |
699 | * dirtied buffers don't. After all, we don't want fsync to last | |
700 | * forever if somebody is actively writing to the file. | |
701 | * | |
702 | * Do this in two main stages: first we copy dirty buffers to a | |
703 | * temporary inode list, queueing the writes as we go. Then we clean | |
704 | * up, waiting for those writes to complete. | |
705 | * | |
706 | * During this second stage, any subsequent updates to the file may end | |
707 | * up refiling the buffer on the original inode's dirty list again, so | |
708 | * there is a chance we will end up with a buffer queued for write but | |
709 | * not yet completed on that list. So, as a final cleanup we go through | |
710 | * the osync code to catch these locked, dirty buffers without requeuing | |
711 | * any newly dirty buffers for write. | |
712 | */ | |
713 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) | |
714 | { | |
715 | struct buffer_head *bh; | |
716 | struct list_head tmp; | |
7eaceacc | 717 | struct address_space *mapping; |
1da177e4 | 718 | int err = 0, err2; |
4ee2491e | 719 | struct blk_plug plug; |
1da177e4 LT |
720 | |
721 | INIT_LIST_HEAD(&tmp); | |
4ee2491e | 722 | blk_start_plug(&plug); |
1da177e4 LT |
723 | |
724 | spin_lock(lock); | |
725 | while (!list_empty(list)) { | |
726 | bh = BH_ENTRY(list->next); | |
535ee2fb | 727 | mapping = bh->b_assoc_map; |
58ff407b | 728 | __remove_assoc_queue(bh); |
535ee2fb JK |
729 | /* Avoid race with mark_buffer_dirty_inode() which does |
730 | * a lockless check and we rely on seeing the dirty bit */ | |
731 | smp_mb(); | |
1da177e4 LT |
732 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
733 | list_add(&bh->b_assoc_buffers, &tmp); | |
535ee2fb | 734 | bh->b_assoc_map = mapping; |
1da177e4 LT |
735 | if (buffer_dirty(bh)) { |
736 | get_bh(bh); | |
737 | spin_unlock(lock); | |
738 | /* | |
739 | * Ensure any pending I/O completes so that | |
9cb569d6 CH |
740 | * write_dirty_buffer() actually writes the |
741 | * current contents - it is a noop if I/O is | |
742 | * still in flight on potentially older | |
743 | * contents. | |
1da177e4 | 744 | */ |
70fd7614 | 745 | write_dirty_buffer(bh, REQ_SYNC); |
9cf6b720 JA |
746 | |
747 | /* | |
748 | * Kick off IO for the previous mapping. Note | |
749 | * that we will not run the very last mapping, | |
750 | * wait_on_buffer() will do that for us | |
751 | * through sync_buffer(). | |
752 | */ | |
1da177e4 LT |
753 | brelse(bh); |
754 | spin_lock(lock); | |
755 | } | |
756 | } | |
757 | } | |
758 | ||
4ee2491e JA |
759 | spin_unlock(lock); |
760 | blk_finish_plug(&plug); | |
761 | spin_lock(lock); | |
762 | ||
1da177e4 LT |
763 | while (!list_empty(&tmp)) { |
764 | bh = BH_ENTRY(tmp.prev); | |
1da177e4 | 765 | get_bh(bh); |
535ee2fb JK |
766 | mapping = bh->b_assoc_map; |
767 | __remove_assoc_queue(bh); | |
768 | /* Avoid race with mark_buffer_dirty_inode() which does | |
769 | * a lockless check and we rely on seeing the dirty bit */ | |
770 | smp_mb(); | |
771 | if (buffer_dirty(bh)) { | |
772 | list_add(&bh->b_assoc_buffers, | |
e3892296 | 773 | &mapping->private_list); |
535ee2fb JK |
774 | bh->b_assoc_map = mapping; |
775 | } | |
1da177e4 LT |
776 | spin_unlock(lock); |
777 | wait_on_buffer(bh); | |
778 | if (!buffer_uptodate(bh)) | |
779 | err = -EIO; | |
780 | brelse(bh); | |
781 | spin_lock(lock); | |
782 | } | |
783 | ||
784 | spin_unlock(lock); | |
785 | err2 = osync_buffers_list(lock, list); | |
786 | if (err) | |
787 | return err; | |
788 | else | |
789 | return err2; | |
790 | } | |
791 | ||
792 | /* | |
793 | * Invalidate any and all dirty buffers on a given inode. We are | |
794 | * probably unmounting the fs, but that doesn't mean we have already | |
795 | * done a sync(). Just drop the buffers from the inode list. | |
796 | * | |
797 | * NOTE: we take the inode's blockdev's mapping's private_lock. Which | |
798 | * assumes that all the buffers are against the blockdev. Not true | |
799 | * for reiserfs. | |
800 | */ | |
801 | void invalidate_inode_buffers(struct inode *inode) | |
802 | { | |
803 | if (inode_has_buffers(inode)) { | |
804 | struct address_space *mapping = &inode->i_data; | |
805 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 806 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
807 | |
808 | spin_lock(&buffer_mapping->private_lock); | |
809 | while (!list_empty(list)) | |
810 | __remove_assoc_queue(BH_ENTRY(list->next)); | |
811 | spin_unlock(&buffer_mapping->private_lock); | |
812 | } | |
813 | } | |
52b19ac9 | 814 | EXPORT_SYMBOL(invalidate_inode_buffers); |
1da177e4 LT |
815 | |
816 | /* | |
817 | * Remove any clean buffers from the inode's buffer list. This is called | |
818 | * when we're trying to free the inode itself. Those buffers can pin it. | |
819 | * | |
820 | * Returns true if all buffers were removed. | |
821 | */ | |
822 | int remove_inode_buffers(struct inode *inode) | |
823 | { | |
824 | int ret = 1; | |
825 | ||
826 | if (inode_has_buffers(inode)) { | |
827 | struct address_space *mapping = &inode->i_data; | |
828 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 829 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
830 | |
831 | spin_lock(&buffer_mapping->private_lock); | |
832 | while (!list_empty(list)) { | |
833 | struct buffer_head *bh = BH_ENTRY(list->next); | |
834 | if (buffer_dirty(bh)) { | |
835 | ret = 0; | |
836 | break; | |
837 | } | |
838 | __remove_assoc_queue(bh); | |
839 | } | |
840 | spin_unlock(&buffer_mapping->private_lock); | |
841 | } | |
842 | return ret; | |
843 | } | |
844 | ||
845 | /* | |
846 | * Create the appropriate buffers when given a page for data area and | |
847 | * the size of each buffer.. Use the bh->b_this_page linked list to | |
848 | * follow the buffers created. Return NULL if unable to create more | |
849 | * buffers. | |
850 | * | |
851 | * The retry flag is used to differentiate async IO (paging, swapping) | |
852 | * which may not fail from ordinary buffer allocations. | |
853 | */ | |
854 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, | |
640ab98f | 855 | bool retry) |
1da177e4 LT |
856 | { |
857 | struct buffer_head *bh, *head; | |
f745c6f5 | 858 | gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; |
1da177e4 | 859 | long offset; |
b87d8cef | 860 | struct mem_cgroup *memcg, *old_memcg; |
1da177e4 | 861 | |
640ab98f JA |
862 | if (retry) |
863 | gfp |= __GFP_NOFAIL; | |
864 | ||
6eeb104e JW |
865 | /* The page lock pins the memcg */ |
866 | memcg = page_memcg(page); | |
b87d8cef | 867 | old_memcg = set_active_memcg(memcg); |
f745c6f5 | 868 | |
1da177e4 LT |
869 | head = NULL; |
870 | offset = PAGE_SIZE; | |
871 | while ((offset -= size) >= 0) { | |
640ab98f | 872 | bh = alloc_buffer_head(gfp); |
1da177e4 LT |
873 | if (!bh) |
874 | goto no_grow; | |
875 | ||
1da177e4 LT |
876 | bh->b_this_page = head; |
877 | bh->b_blocknr = -1; | |
878 | head = bh; | |
879 | ||
1da177e4 LT |
880 | bh->b_size = size; |
881 | ||
882 | /* Link the buffer to its page */ | |
883 | set_bh_page(bh, page, offset); | |
1da177e4 | 884 | } |
f745c6f5 | 885 | out: |
b87d8cef | 886 | set_active_memcg(old_memcg); |
1da177e4 LT |
887 | return head; |
888 | /* | |
889 | * In case anything failed, we just free everything we got. | |
890 | */ | |
891 | no_grow: | |
892 | if (head) { | |
893 | do { | |
894 | bh = head; | |
895 | head = head->b_this_page; | |
896 | free_buffer_head(bh); | |
897 | } while (head); | |
898 | } | |
899 | ||
f745c6f5 | 900 | goto out; |
1da177e4 LT |
901 | } |
902 | EXPORT_SYMBOL_GPL(alloc_page_buffers); | |
903 | ||
904 | static inline void | |
905 | link_dev_buffers(struct page *page, struct buffer_head *head) | |
906 | { | |
907 | struct buffer_head *bh, *tail; | |
908 | ||
909 | bh = head; | |
910 | do { | |
911 | tail = bh; | |
912 | bh = bh->b_this_page; | |
913 | } while (bh); | |
914 | tail->b_this_page = head; | |
45dcfc27 | 915 | attach_page_private(page, head); |
1da177e4 LT |
916 | } |
917 | ||
bbec0270 LT |
918 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
919 | { | |
920 | sector_t retval = ~((sector_t)0); | |
b86058f9 | 921 | loff_t sz = bdev_nr_bytes(bdev); |
bbec0270 LT |
922 | |
923 | if (sz) { | |
924 | unsigned int sizebits = blksize_bits(size); | |
925 | retval = (sz >> sizebits); | |
926 | } | |
927 | return retval; | |
928 | } | |
929 | ||
1da177e4 LT |
930 | /* |
931 | * Initialise the state of a blockdev page's buffers. | |
932 | */ | |
676ce6d5 | 933 | static sector_t |
1da177e4 LT |
934 | init_page_buffers(struct page *page, struct block_device *bdev, |
935 | sector_t block, int size) | |
936 | { | |
937 | struct buffer_head *head = page_buffers(page); | |
938 | struct buffer_head *bh = head; | |
939 | int uptodate = PageUptodate(page); | |
bcd1d063 | 940 | sector_t end_block = blkdev_max_block(bdev, size); |
1da177e4 LT |
941 | |
942 | do { | |
943 | if (!buffer_mapped(bh)) { | |
01950a34 EB |
944 | bh->b_end_io = NULL; |
945 | bh->b_private = NULL; | |
1da177e4 LT |
946 | bh->b_bdev = bdev; |
947 | bh->b_blocknr = block; | |
948 | if (uptodate) | |
949 | set_buffer_uptodate(bh); | |
080399aa JM |
950 | if (block < end_block) |
951 | set_buffer_mapped(bh); | |
1da177e4 LT |
952 | } |
953 | block++; | |
954 | bh = bh->b_this_page; | |
955 | } while (bh != head); | |
676ce6d5 HD |
956 | |
957 | /* | |
958 | * Caller needs to validate requested block against end of device. | |
959 | */ | |
960 | return end_block; | |
1da177e4 LT |
961 | } |
962 | ||
963 | /* | |
964 | * Create the page-cache page that contains the requested block. | |
965 | * | |
676ce6d5 | 966 | * This is used purely for blockdev mappings. |
1da177e4 | 967 | */ |
676ce6d5 | 968 | static int |
1da177e4 | 969 | grow_dev_page(struct block_device *bdev, sector_t block, |
3b5e6454 | 970 | pgoff_t index, int size, int sizebits, gfp_t gfp) |
1da177e4 LT |
971 | { |
972 | struct inode *inode = bdev->bd_inode; | |
973 | struct page *page; | |
974 | struct buffer_head *bh; | |
676ce6d5 | 975 | sector_t end_block; |
c4b4c2a7 | 976 | int ret = 0; |
84235de3 | 977 | gfp_t gfp_mask; |
1da177e4 | 978 | |
c62d2555 | 979 | gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; |
3b5e6454 | 980 | |
84235de3 JW |
981 | /* |
982 | * XXX: __getblk_slow() can not really deal with failure and | |
983 | * will endlessly loop on improvised global reclaim. Prefer | |
984 | * looping in the allocator rather than here, at least that | |
985 | * code knows what it's doing. | |
986 | */ | |
987 | gfp_mask |= __GFP_NOFAIL; | |
988 | ||
989 | page = find_or_create_page(inode->i_mapping, index, gfp_mask); | |
1da177e4 | 990 | |
e827f923 | 991 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
992 | |
993 | if (page_has_buffers(page)) { | |
994 | bh = page_buffers(page); | |
995 | if (bh->b_size == size) { | |
676ce6d5 | 996 | end_block = init_page_buffers(page, bdev, |
f2d5a944 AA |
997 | (sector_t)index << sizebits, |
998 | size); | |
676ce6d5 | 999 | goto done; |
1da177e4 | 1000 | } |
68189fef | 1001 | if (!try_to_free_buffers(page_folio(page))) |
1da177e4 LT |
1002 | goto failed; |
1003 | } | |
1004 | ||
1005 | /* | |
1006 | * Allocate some buffers for this page | |
1007 | */ | |
94dc24c0 | 1008 | bh = alloc_page_buffers(page, size, true); |
1da177e4 LT |
1009 | |
1010 | /* | |
1011 | * Link the page to the buffers and initialise them. Take the | |
1012 | * lock to be atomic wrt __find_get_block(), which does not | |
1013 | * run under the page lock. | |
1014 | */ | |
1015 | spin_lock(&inode->i_mapping->private_lock); | |
1016 | link_dev_buffers(page, bh); | |
f2d5a944 AA |
1017 | end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, |
1018 | size); | |
1da177e4 | 1019 | spin_unlock(&inode->i_mapping->private_lock); |
676ce6d5 HD |
1020 | done: |
1021 | ret = (block < end_block) ? 1 : -ENXIO; | |
1da177e4 | 1022 | failed: |
1da177e4 | 1023 | unlock_page(page); |
09cbfeaf | 1024 | put_page(page); |
676ce6d5 | 1025 | return ret; |
1da177e4 LT |
1026 | } |
1027 | ||
1028 | /* | |
1029 | * Create buffers for the specified block device block's page. If | |
1030 | * that page was dirty, the buffers are set dirty also. | |
1da177e4 | 1031 | */ |
858119e1 | 1032 | static int |
3b5e6454 | 1033 | grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) |
1da177e4 | 1034 | { |
1da177e4 LT |
1035 | pgoff_t index; |
1036 | int sizebits; | |
1037 | ||
90432e60 | 1038 | sizebits = PAGE_SHIFT - __ffs(size); |
1da177e4 | 1039 | index = block >> sizebits; |
1da177e4 | 1040 | |
e5657933 AM |
1041 | /* |
1042 | * Check for a block which wants to lie outside our maximum possible | |
1043 | * pagecache index. (this comparison is done using sector_t types). | |
1044 | */ | |
1045 | if (unlikely(index != block >> sizebits)) { | |
e5657933 | 1046 | printk(KERN_ERR "%s: requested out-of-range block %llu for " |
a1c6f057 | 1047 | "device %pg\n", |
8e24eea7 | 1048 | __func__, (unsigned long long)block, |
a1c6f057 | 1049 | bdev); |
e5657933 AM |
1050 | return -EIO; |
1051 | } | |
676ce6d5 | 1052 | |
1da177e4 | 1053 | /* Create a page with the proper size buffers.. */ |
3b5e6454 | 1054 | return grow_dev_page(bdev, block, index, size, sizebits, gfp); |
1da177e4 LT |
1055 | } |
1056 | ||
0026ba40 | 1057 | static struct buffer_head * |
3b5e6454 GK |
1058 | __getblk_slow(struct block_device *bdev, sector_t block, |
1059 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1060 | { |
1061 | /* Size must be multiple of hard sectorsize */ | |
e1defc4f | 1062 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
1da177e4 LT |
1063 | (size < 512 || size > PAGE_SIZE))) { |
1064 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", | |
1065 | size); | |
e1defc4f MP |
1066 | printk(KERN_ERR "logical block size: %d\n", |
1067 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1068 | |
1069 | dump_stack(); | |
1070 | return NULL; | |
1071 | } | |
1072 | ||
676ce6d5 HD |
1073 | for (;;) { |
1074 | struct buffer_head *bh; | |
1075 | int ret; | |
1da177e4 LT |
1076 | |
1077 | bh = __find_get_block(bdev, block, size); | |
1078 | if (bh) | |
1079 | return bh; | |
676ce6d5 | 1080 | |
3b5e6454 | 1081 | ret = grow_buffers(bdev, block, size, gfp); |
676ce6d5 HD |
1082 | if (ret < 0) |
1083 | return NULL; | |
1da177e4 LT |
1084 | } |
1085 | } | |
1086 | ||
1087 | /* | |
1088 | * The relationship between dirty buffers and dirty pages: | |
1089 | * | |
1090 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and | |
ec82e1c1 | 1091 | * the page is tagged dirty in the page cache. |
1da177e4 LT |
1092 | * |
1093 | * At all times, the dirtiness of the buffers represents the dirtiness of | |
1094 | * subsections of the page. If the page has buffers, the page dirty bit is | |
1095 | * merely a hint about the true dirty state. | |
1096 | * | |
1097 | * When a page is set dirty in its entirety, all its buffers are marked dirty | |
1098 | * (if the page has buffers). | |
1099 | * | |
1100 | * When a buffer is marked dirty, its page is dirtied, but the page's other | |
1101 | * buffers are not. | |
1102 | * | |
1103 | * Also. When blockdev buffers are explicitly read with bread(), they | |
1104 | * individually become uptodate. But their backing page remains not | |
1105 | * uptodate - even if all of its buffers are uptodate. A subsequent | |
2c69e205 MWO |
1106 | * block_read_full_folio() against that folio will discover all the uptodate |
1107 | * buffers, will set the folio uptodate and will perform no I/O. | |
1da177e4 LT |
1108 | */ |
1109 | ||
1110 | /** | |
1111 | * mark_buffer_dirty - mark a buffer_head as needing writeout | |
67be2dd1 | 1112 | * @bh: the buffer_head to mark dirty |
1da177e4 | 1113 | * |
ec82e1c1 MW |
1114 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set |
1115 | * its backing page dirty, then tag the page as dirty in the page cache | |
1116 | * and then attach the address_space's inode to its superblock's dirty | |
1da177e4 LT |
1117 | * inode list. |
1118 | * | |
abc8a8a2 | 1119 | * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->private_lock, |
b93b0163 | 1120 | * i_pages lock and mapping->host->i_lock. |
1da177e4 | 1121 | */ |
fc9b52cd | 1122 | void mark_buffer_dirty(struct buffer_head *bh) |
1da177e4 | 1123 | { |
787d2214 | 1124 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
1be62dc1 | 1125 | |
5305cb83 TH |
1126 | trace_block_dirty_buffer(bh); |
1127 | ||
1be62dc1 LT |
1128 | /* |
1129 | * Very *carefully* optimize the it-is-already-dirty case. | |
1130 | * | |
1131 | * Don't let the final "is it dirty" escape to before we | |
1132 | * perhaps modified the buffer. | |
1133 | */ | |
1134 | if (buffer_dirty(bh)) { | |
1135 | smp_mb(); | |
1136 | if (buffer_dirty(bh)) | |
1137 | return; | |
1138 | } | |
1139 | ||
a8e7d49a | 1140 | if (!test_set_buffer_dirty(bh)) { |
cf1d3417 | 1141 | struct folio *folio = bh->b_folio; |
c4843a75 | 1142 | struct address_space *mapping = NULL; |
c4843a75 | 1143 | |
cf1d3417 MWO |
1144 | folio_memcg_lock(folio); |
1145 | if (!folio_test_set_dirty(folio)) { | |
1146 | mapping = folio->mapping; | |
8e9d78ed | 1147 | if (mapping) |
cf1d3417 | 1148 | __folio_mark_dirty(folio, mapping, 0); |
8e9d78ed | 1149 | } |
cf1d3417 | 1150 | folio_memcg_unlock(folio); |
c4843a75 GT |
1151 | if (mapping) |
1152 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
a8e7d49a | 1153 | } |
1da177e4 | 1154 | } |
1fe72eaa | 1155 | EXPORT_SYMBOL(mark_buffer_dirty); |
1da177e4 | 1156 | |
87354e5d JL |
1157 | void mark_buffer_write_io_error(struct buffer_head *bh) |
1158 | { | |
485e9605 JL |
1159 | struct super_block *sb; |
1160 | ||
87354e5d JL |
1161 | set_buffer_write_io_error(bh); |
1162 | /* FIXME: do we need to set this in both places? */ | |
abc8a8a2 MWO |
1163 | if (bh->b_folio && bh->b_folio->mapping) |
1164 | mapping_set_error(bh->b_folio->mapping, -EIO); | |
87354e5d JL |
1165 | if (bh->b_assoc_map) |
1166 | mapping_set_error(bh->b_assoc_map, -EIO); | |
485e9605 JL |
1167 | rcu_read_lock(); |
1168 | sb = READ_ONCE(bh->b_bdev->bd_super); | |
1169 | if (sb) | |
1170 | errseq_set(&sb->s_wb_err, -EIO); | |
1171 | rcu_read_unlock(); | |
87354e5d JL |
1172 | } |
1173 | EXPORT_SYMBOL(mark_buffer_write_io_error); | |
1174 | ||
1da177e4 LT |
1175 | /* |
1176 | * Decrement a buffer_head's reference count. If all buffers against a page | |
1177 | * have zero reference count, are clean and unlocked, and if the page is clean | |
1178 | * and unlocked then try_to_free_buffers() may strip the buffers from the page | |
1179 | * in preparation for freeing it (sometimes, rarely, buffers are removed from | |
1180 | * a page but it ends up not being freed, and buffers may later be reattached). | |
1181 | */ | |
1182 | void __brelse(struct buffer_head * buf) | |
1183 | { | |
1184 | if (atomic_read(&buf->b_count)) { | |
1185 | put_bh(buf); | |
1186 | return; | |
1187 | } | |
5c752ad9 | 1188 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
1da177e4 | 1189 | } |
1fe72eaa | 1190 | EXPORT_SYMBOL(__brelse); |
1da177e4 LT |
1191 | |
1192 | /* | |
1193 | * bforget() is like brelse(), except it discards any | |
1194 | * potentially dirty data. | |
1195 | */ | |
1196 | void __bforget(struct buffer_head *bh) | |
1197 | { | |
1198 | clear_buffer_dirty(bh); | |
535ee2fb | 1199 | if (bh->b_assoc_map) { |
abc8a8a2 | 1200 | struct address_space *buffer_mapping = bh->b_folio->mapping; |
1da177e4 LT |
1201 | |
1202 | spin_lock(&buffer_mapping->private_lock); | |
1203 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 1204 | bh->b_assoc_map = NULL; |
1da177e4 LT |
1205 | spin_unlock(&buffer_mapping->private_lock); |
1206 | } | |
1207 | __brelse(bh); | |
1208 | } | |
1fe72eaa | 1209 | EXPORT_SYMBOL(__bforget); |
1da177e4 LT |
1210 | |
1211 | static struct buffer_head *__bread_slow(struct buffer_head *bh) | |
1212 | { | |
1213 | lock_buffer(bh); | |
1214 | if (buffer_uptodate(bh)) { | |
1215 | unlock_buffer(bh); | |
1216 | return bh; | |
1217 | } else { | |
1218 | get_bh(bh); | |
1219 | bh->b_end_io = end_buffer_read_sync; | |
1420c4a5 | 1220 | submit_bh(REQ_OP_READ, bh); |
1da177e4 LT |
1221 | wait_on_buffer(bh); |
1222 | if (buffer_uptodate(bh)) | |
1223 | return bh; | |
1224 | } | |
1225 | brelse(bh); | |
1226 | return NULL; | |
1227 | } | |
1228 | ||
1229 | /* | |
1230 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). | |
1231 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their | |
1232 | * refcount elevated by one when they're in an LRU. A buffer can only appear | |
1233 | * once in a particular CPU's LRU. A single buffer can be present in multiple | |
1234 | * CPU's LRUs at the same time. | |
1235 | * | |
1236 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and | |
1237 | * sb_find_get_block(). | |
1238 | * | |
1239 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use | |
1240 | * a local interrupt disable for that. | |
1241 | */ | |
1242 | ||
86cf78d7 | 1243 | #define BH_LRU_SIZE 16 |
1da177e4 LT |
1244 | |
1245 | struct bh_lru { | |
1246 | struct buffer_head *bhs[BH_LRU_SIZE]; | |
1247 | }; | |
1248 | ||
1249 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; | |
1250 | ||
1251 | #ifdef CONFIG_SMP | |
1252 | #define bh_lru_lock() local_irq_disable() | |
1253 | #define bh_lru_unlock() local_irq_enable() | |
1254 | #else | |
1255 | #define bh_lru_lock() preempt_disable() | |
1256 | #define bh_lru_unlock() preempt_enable() | |
1257 | #endif | |
1258 | ||
1259 | static inline void check_irqs_on(void) | |
1260 | { | |
1261 | #ifdef irqs_disabled | |
1262 | BUG_ON(irqs_disabled()); | |
1263 | #endif | |
1264 | } | |
1265 | ||
1266 | /* | |
241f01fb EB |
1267 | * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is |
1268 | * inserted at the front, and the buffer_head at the back if any is evicted. | |
1269 | * Or, if already in the LRU it is moved to the front. | |
1da177e4 LT |
1270 | */ |
1271 | static void bh_lru_install(struct buffer_head *bh) | |
1272 | { | |
241f01fb EB |
1273 | struct buffer_head *evictee = bh; |
1274 | struct bh_lru *b; | |
1275 | int i; | |
1da177e4 LT |
1276 | |
1277 | check_irqs_on(); | |
c0226eb8 MK |
1278 | bh_lru_lock(); |
1279 | ||
8cc621d2 MK |
1280 | /* |
1281 | * the refcount of buffer_head in bh_lru prevents dropping the | |
1282 | * attached page(i.e., try_to_free_buffers) so it could cause | |
1283 | * failing page migration. | |
1284 | * Skip putting upcoming bh into bh_lru until migration is done. | |
1285 | */ | |
c0226eb8 MK |
1286 | if (lru_cache_disabled()) { |
1287 | bh_lru_unlock(); | |
8cc621d2 | 1288 | return; |
c0226eb8 | 1289 | } |
1da177e4 | 1290 | |
241f01fb EB |
1291 | b = this_cpu_ptr(&bh_lrus); |
1292 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1293 | swap(evictee, b->bhs[i]); | |
1294 | if (evictee == bh) { | |
1295 | bh_lru_unlock(); | |
1296 | return; | |
1da177e4 | 1297 | } |
1da177e4 | 1298 | } |
1da177e4 | 1299 | |
241f01fb EB |
1300 | get_bh(bh); |
1301 | bh_lru_unlock(); | |
1302 | brelse(evictee); | |
1da177e4 LT |
1303 | } |
1304 | ||
1305 | /* | |
1306 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. | |
1307 | */ | |
858119e1 | 1308 | static struct buffer_head * |
3991d3bd | 1309 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1310 | { |
1311 | struct buffer_head *ret = NULL; | |
3991d3bd | 1312 | unsigned int i; |
1da177e4 LT |
1313 | |
1314 | check_irqs_on(); | |
1315 | bh_lru_lock(); | |
1da177e4 | 1316 | for (i = 0; i < BH_LRU_SIZE; i++) { |
c7b92516 | 1317 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
1da177e4 | 1318 | |
9470dd5d ZB |
1319 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
1320 | bh->b_size == size) { | |
1da177e4 LT |
1321 | if (i) { |
1322 | while (i) { | |
c7b92516 CL |
1323 | __this_cpu_write(bh_lrus.bhs[i], |
1324 | __this_cpu_read(bh_lrus.bhs[i - 1])); | |
1da177e4 LT |
1325 | i--; |
1326 | } | |
c7b92516 | 1327 | __this_cpu_write(bh_lrus.bhs[0], bh); |
1da177e4 LT |
1328 | } |
1329 | get_bh(bh); | |
1330 | ret = bh; | |
1331 | break; | |
1332 | } | |
1333 | } | |
1334 | bh_lru_unlock(); | |
1335 | return ret; | |
1336 | } | |
1337 | ||
1338 | /* | |
1339 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh | |
1340 | * it in the LRU and mark it as accessed. If it is not present then return | |
1341 | * NULL | |
1342 | */ | |
1343 | struct buffer_head * | |
3991d3bd | 1344 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1345 | { |
1346 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); | |
1347 | ||
1348 | if (bh == NULL) { | |
2457aec6 | 1349 | /* __find_get_block_slow will mark the page accessed */ |
385fd4c5 | 1350 | bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1351 | if (bh) |
1352 | bh_lru_install(bh); | |
2457aec6 | 1353 | } else |
1da177e4 | 1354 | touch_buffer(bh); |
2457aec6 | 1355 | |
1da177e4 LT |
1356 | return bh; |
1357 | } | |
1358 | EXPORT_SYMBOL(__find_get_block); | |
1359 | ||
1360 | /* | |
3b5e6454 | 1361 | * __getblk_gfp() will locate (and, if necessary, create) the buffer_head |
1da177e4 LT |
1362 | * which corresponds to the passed block_device, block and size. The |
1363 | * returned buffer has its reference count incremented. | |
1364 | * | |
3b5e6454 GK |
1365 | * __getblk_gfp() will lock up the machine if grow_dev_page's |
1366 | * try_to_free_buffers() attempt is failing. FIXME, perhaps? | |
1da177e4 LT |
1367 | */ |
1368 | struct buffer_head * | |
3b5e6454 GK |
1369 | __getblk_gfp(struct block_device *bdev, sector_t block, |
1370 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1371 | { |
1372 | struct buffer_head *bh = __find_get_block(bdev, block, size); | |
1373 | ||
1374 | might_sleep(); | |
1375 | if (bh == NULL) | |
3b5e6454 | 1376 | bh = __getblk_slow(bdev, block, size, gfp); |
1da177e4 LT |
1377 | return bh; |
1378 | } | |
3b5e6454 | 1379 | EXPORT_SYMBOL(__getblk_gfp); |
1da177e4 LT |
1380 | |
1381 | /* | |
1382 | * Do async read-ahead on a buffer.. | |
1383 | */ | |
3991d3bd | 1384 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1385 | { |
1386 | struct buffer_head *bh = __getblk(bdev, block, size); | |
a3e713b5 | 1387 | if (likely(bh)) { |
e7ea1129 | 1388 | bh_readahead(bh, REQ_RAHEAD); |
a3e713b5 AM |
1389 | brelse(bh); |
1390 | } | |
1da177e4 LT |
1391 | } |
1392 | EXPORT_SYMBOL(__breadahead); | |
1393 | ||
1394 | /** | |
3b5e6454 | 1395 | * __bread_gfp() - reads a specified block and returns the bh |
67be2dd1 | 1396 | * @bdev: the block_device to read from |
1da177e4 LT |
1397 | * @block: number of block |
1398 | * @size: size (in bytes) to read | |
3b5e6454 GK |
1399 | * @gfp: page allocation flag |
1400 | * | |
1da177e4 | 1401 | * Reads a specified block, and returns buffer head that contains it. |
3b5e6454 GK |
1402 | * The page cache can be allocated from non-movable area |
1403 | * not to prevent page migration if you set gfp to zero. | |
1da177e4 LT |
1404 | * It returns NULL if the block was unreadable. |
1405 | */ | |
1406 | struct buffer_head * | |
3b5e6454 GK |
1407 | __bread_gfp(struct block_device *bdev, sector_t block, |
1408 | unsigned size, gfp_t gfp) | |
1da177e4 | 1409 | { |
3b5e6454 | 1410 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); |
1da177e4 | 1411 | |
a3e713b5 | 1412 | if (likely(bh) && !buffer_uptodate(bh)) |
1da177e4 LT |
1413 | bh = __bread_slow(bh); |
1414 | return bh; | |
1415 | } | |
3b5e6454 | 1416 | EXPORT_SYMBOL(__bread_gfp); |
1da177e4 | 1417 | |
8cc621d2 MK |
1418 | static void __invalidate_bh_lrus(struct bh_lru *b) |
1419 | { | |
1420 | int i; | |
1421 | ||
1422 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1423 | brelse(b->bhs[i]); | |
1424 | b->bhs[i] = NULL; | |
1425 | } | |
1426 | } | |
1da177e4 LT |
1427 | /* |
1428 | * invalidate_bh_lrus() is called rarely - but not only at unmount. | |
1429 | * This doesn't race because it runs in each cpu either in irq | |
1430 | * or with preempt disabled. | |
1431 | */ | |
1432 | static void invalidate_bh_lru(void *arg) | |
1433 | { | |
1434 | struct bh_lru *b = &get_cpu_var(bh_lrus); | |
1da177e4 | 1435 | |
8cc621d2 | 1436 | __invalidate_bh_lrus(b); |
1da177e4 LT |
1437 | put_cpu_var(bh_lrus); |
1438 | } | |
42be35d0 | 1439 | |
8cc621d2 | 1440 | bool has_bh_in_lru(int cpu, void *dummy) |
42be35d0 GBY |
1441 | { |
1442 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); | |
1443 | int i; | |
1da177e4 | 1444 | |
42be35d0 GBY |
1445 | for (i = 0; i < BH_LRU_SIZE; i++) { |
1446 | if (b->bhs[i]) | |
1d706679 | 1447 | return true; |
42be35d0 GBY |
1448 | } |
1449 | ||
1d706679 | 1450 | return false; |
42be35d0 GBY |
1451 | } |
1452 | ||
f9a14399 | 1453 | void invalidate_bh_lrus(void) |
1da177e4 | 1454 | { |
cb923159 | 1455 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); |
1da177e4 | 1456 | } |
9db5579b | 1457 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
1da177e4 | 1458 | |
243418e3 MK |
1459 | /* |
1460 | * It's called from workqueue context so we need a bh_lru_lock to close | |
1461 | * the race with preemption/irq. | |
1462 | */ | |
1463 | void invalidate_bh_lrus_cpu(void) | |
8cc621d2 MK |
1464 | { |
1465 | struct bh_lru *b; | |
1466 | ||
1467 | bh_lru_lock(); | |
243418e3 | 1468 | b = this_cpu_ptr(&bh_lrus); |
8cc621d2 MK |
1469 | __invalidate_bh_lrus(b); |
1470 | bh_lru_unlock(); | |
1471 | } | |
1472 | ||
1da177e4 LT |
1473 | void set_bh_page(struct buffer_head *bh, |
1474 | struct page *page, unsigned long offset) | |
1475 | { | |
1476 | bh->b_page = page; | |
e827f923 | 1477 | BUG_ON(offset >= PAGE_SIZE); |
1da177e4 LT |
1478 | if (PageHighMem(page)) |
1479 | /* | |
1480 | * This catches illegal uses and preserves the offset: | |
1481 | */ | |
1482 | bh->b_data = (char *)(0 + offset); | |
1483 | else | |
1484 | bh->b_data = page_address(page) + offset; | |
1485 | } | |
1486 | EXPORT_SYMBOL(set_bh_page); | |
1487 | ||
1488 | /* | |
1489 | * Called when truncating a buffer on a page completely. | |
1490 | */ | |
e7470ee8 MG |
1491 | |
1492 | /* Bits that are cleared during an invalidate */ | |
1493 | #define BUFFER_FLAGS_DISCARD \ | |
1494 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ | |
1495 | 1 << BH_Delay | 1 << BH_Unwritten) | |
1496 | ||
858119e1 | 1497 | static void discard_buffer(struct buffer_head * bh) |
1da177e4 | 1498 | { |
b0192296 | 1499 | unsigned long b_state; |
e7470ee8 | 1500 | |
1da177e4 LT |
1501 | lock_buffer(bh); |
1502 | clear_buffer_dirty(bh); | |
1503 | bh->b_bdev = NULL; | |
b0192296 UB |
1504 | b_state = READ_ONCE(bh->b_state); |
1505 | do { | |
1506 | } while (!try_cmpxchg(&bh->b_state, &b_state, | |
1507 | b_state & ~BUFFER_FLAGS_DISCARD)); | |
1da177e4 LT |
1508 | unlock_buffer(bh); |
1509 | } | |
1510 | ||
1da177e4 | 1511 | /** |
7ba13abb MWO |
1512 | * block_invalidate_folio - Invalidate part or all of a buffer-backed folio. |
1513 | * @folio: The folio which is affected. | |
d47992f8 LC |
1514 | * @offset: start of the range to invalidate |
1515 | * @length: length of the range to invalidate | |
1da177e4 | 1516 | * |
7ba13abb | 1517 | * block_invalidate_folio() is called when all or part of the folio has been |
814e1d25 | 1518 | * invalidated by a truncate operation. |
1da177e4 | 1519 | * |
7ba13abb | 1520 | * block_invalidate_folio() does not have to release all buffers, but it must |
1da177e4 LT |
1521 | * ensure that no dirty buffer is left outside @offset and that no I/O |
1522 | * is underway against any of the blocks which are outside the truncation | |
1523 | * point. Because the caller is about to free (and possibly reuse) those | |
1524 | * blocks on-disk. | |
1525 | */ | |
7ba13abb | 1526 | void block_invalidate_folio(struct folio *folio, size_t offset, size_t length) |
1da177e4 LT |
1527 | { |
1528 | struct buffer_head *head, *bh, *next; | |
7ba13abb MWO |
1529 | size_t curr_off = 0; |
1530 | size_t stop = length + offset; | |
1da177e4 | 1531 | |
7ba13abb | 1532 | BUG_ON(!folio_test_locked(folio)); |
1da177e4 | 1533 | |
d47992f8 LC |
1534 | /* |
1535 | * Check for overflow | |
1536 | */ | |
7ba13abb MWO |
1537 | BUG_ON(stop > folio_size(folio) || stop < length); |
1538 | ||
1539 | head = folio_buffers(folio); | |
1540 | if (!head) | |
1541 | return; | |
d47992f8 | 1542 | |
1da177e4 LT |
1543 | bh = head; |
1544 | do { | |
7ba13abb | 1545 | size_t next_off = curr_off + bh->b_size; |
1da177e4 LT |
1546 | next = bh->b_this_page; |
1547 | ||
d47992f8 LC |
1548 | /* |
1549 | * Are we still fully in range ? | |
1550 | */ | |
1551 | if (next_off > stop) | |
1552 | goto out; | |
1553 | ||
1da177e4 LT |
1554 | /* |
1555 | * is this block fully invalidated? | |
1556 | */ | |
1557 | if (offset <= curr_off) | |
1558 | discard_buffer(bh); | |
1559 | curr_off = next_off; | |
1560 | bh = next; | |
1561 | } while (bh != head); | |
1562 | ||
1563 | /* | |
7ba13abb | 1564 | * We release buffers only if the entire folio is being invalidated. |
1da177e4 LT |
1565 | * The get_block cached value has been unconditionally invalidated, |
1566 | * so real IO is not possible anymore. | |
1567 | */ | |
7ba13abb MWO |
1568 | if (length == folio_size(folio)) |
1569 | filemap_release_folio(folio, 0); | |
1da177e4 | 1570 | out: |
2ff28e22 | 1571 | return; |
1da177e4 | 1572 | } |
7ba13abb | 1573 | EXPORT_SYMBOL(block_invalidate_folio); |
1da177e4 | 1574 | |
d47992f8 | 1575 | |
1da177e4 LT |
1576 | /* |
1577 | * We attach and possibly dirty the buffers atomically wrt | |
e621900a | 1578 | * block_dirty_folio() via private_lock. try_to_free_buffers |
1da177e4 LT |
1579 | * is already excluded via the page lock. |
1580 | */ | |
1581 | void create_empty_buffers(struct page *page, | |
1582 | unsigned long blocksize, unsigned long b_state) | |
1583 | { | |
1584 | struct buffer_head *bh, *head, *tail; | |
1585 | ||
640ab98f | 1586 | head = alloc_page_buffers(page, blocksize, true); |
1da177e4 LT |
1587 | bh = head; |
1588 | do { | |
1589 | bh->b_state |= b_state; | |
1590 | tail = bh; | |
1591 | bh = bh->b_this_page; | |
1592 | } while (bh); | |
1593 | tail->b_this_page = head; | |
1594 | ||
1595 | spin_lock(&page->mapping->private_lock); | |
1596 | if (PageUptodate(page) || PageDirty(page)) { | |
1597 | bh = head; | |
1598 | do { | |
1599 | if (PageDirty(page)) | |
1600 | set_buffer_dirty(bh); | |
1601 | if (PageUptodate(page)) | |
1602 | set_buffer_uptodate(bh); | |
1603 | bh = bh->b_this_page; | |
1604 | } while (bh != head); | |
1605 | } | |
45dcfc27 | 1606 | attach_page_private(page, head); |
1da177e4 LT |
1607 | spin_unlock(&page->mapping->private_lock); |
1608 | } | |
1609 | EXPORT_SYMBOL(create_empty_buffers); | |
1610 | ||
29f3ad7d JK |
1611 | /** |
1612 | * clean_bdev_aliases: clean a range of buffers in block device | |
1613 | * @bdev: Block device to clean buffers in | |
1614 | * @block: Start of a range of blocks to clean | |
1615 | * @len: Number of blocks to clean | |
1da177e4 | 1616 | * |
29f3ad7d JK |
1617 | * We are taking a range of blocks for data and we don't want writeback of any |
1618 | * buffer-cache aliases starting from return from this function and until the | |
1619 | * moment when something will explicitly mark the buffer dirty (hopefully that | |
1620 | * will not happen until we will free that block ;-) We don't even need to mark | |
1621 | * it not-uptodate - nobody can expect anything from a newly allocated buffer | |
1622 | * anyway. We used to use unmap_buffer() for such invalidation, but that was | |
1623 | * wrong. We definitely don't want to mark the alias unmapped, for example - it | |
1624 | * would confuse anyone who might pick it with bread() afterwards... | |
1625 | * | |
1626 | * Also.. Note that bforget() doesn't lock the buffer. So there can be | |
1627 | * writeout I/O going on against recently-freed buffers. We don't wait on that | |
1628 | * I/O in bforget() - it's more efficient to wait on the I/O only if we really | |
1629 | * need to. That happens here. | |
1da177e4 | 1630 | */ |
29f3ad7d | 1631 | void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) |
1da177e4 | 1632 | { |
29f3ad7d JK |
1633 | struct inode *bd_inode = bdev->bd_inode; |
1634 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
9e0b6f31 | 1635 | struct folio_batch fbatch; |
29f3ad7d JK |
1636 | pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); |
1637 | pgoff_t end; | |
c10f778d | 1638 | int i, count; |
29f3ad7d JK |
1639 | struct buffer_head *bh; |
1640 | struct buffer_head *head; | |
1da177e4 | 1641 | |
29f3ad7d | 1642 | end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); |
9e0b6f31 MWO |
1643 | folio_batch_init(&fbatch); |
1644 | while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) { | |
1645 | count = folio_batch_count(&fbatch); | |
c10f778d | 1646 | for (i = 0; i < count; i++) { |
9e0b6f31 | 1647 | struct folio *folio = fbatch.folios[i]; |
1da177e4 | 1648 | |
9e0b6f31 | 1649 | if (!folio_buffers(folio)) |
29f3ad7d JK |
1650 | continue; |
1651 | /* | |
9e0b6f31 | 1652 | * We use folio lock instead of bd_mapping->private_lock |
29f3ad7d JK |
1653 | * to pin buffers here since we can afford to sleep and |
1654 | * it scales better than a global spinlock lock. | |
1655 | */ | |
9e0b6f31 MWO |
1656 | folio_lock(folio); |
1657 | /* Recheck when the folio is locked which pins bhs */ | |
1658 | head = folio_buffers(folio); | |
1659 | if (!head) | |
29f3ad7d | 1660 | goto unlock_page; |
29f3ad7d JK |
1661 | bh = head; |
1662 | do { | |
6c006a9d | 1663 | if (!buffer_mapped(bh) || (bh->b_blocknr < block)) |
29f3ad7d JK |
1664 | goto next; |
1665 | if (bh->b_blocknr >= block + len) | |
1666 | break; | |
1667 | clear_buffer_dirty(bh); | |
1668 | wait_on_buffer(bh); | |
1669 | clear_buffer_req(bh); | |
1670 | next: | |
1671 | bh = bh->b_this_page; | |
1672 | } while (bh != head); | |
1673 | unlock_page: | |
9e0b6f31 | 1674 | folio_unlock(folio); |
29f3ad7d | 1675 | } |
9e0b6f31 | 1676 | folio_batch_release(&fbatch); |
29f3ad7d | 1677 | cond_resched(); |
c10f778d JK |
1678 | /* End of range already reached? */ |
1679 | if (index > end || !index) | |
1680 | break; | |
1da177e4 LT |
1681 | } |
1682 | } | |
29f3ad7d | 1683 | EXPORT_SYMBOL(clean_bdev_aliases); |
1da177e4 | 1684 | |
45bce8f3 LT |
1685 | /* |
1686 | * Size is a power-of-two in the range 512..PAGE_SIZE, | |
1687 | * and the case we care about most is PAGE_SIZE. | |
1688 | * | |
1689 | * So this *could* possibly be written with those | |
1690 | * constraints in mind (relevant mostly if some | |
1691 | * architecture has a slow bit-scan instruction) | |
1692 | */ | |
1693 | static inline int block_size_bits(unsigned int blocksize) | |
1694 | { | |
1695 | return ilog2(blocksize); | |
1696 | } | |
1697 | ||
1698 | static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) | |
1699 | { | |
1700 | BUG_ON(!PageLocked(page)); | |
1701 | ||
1702 | if (!page_has_buffers(page)) | |
6aa7de05 MR |
1703 | create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits), |
1704 | b_state); | |
45bce8f3 LT |
1705 | return page_buffers(page); |
1706 | } | |
1707 | ||
1da177e4 LT |
1708 | /* |
1709 | * NOTE! All mapped/uptodate combinations are valid: | |
1710 | * | |
1711 | * Mapped Uptodate Meaning | |
1712 | * | |
1713 | * No No "unknown" - must do get_block() | |
1714 | * No Yes "hole" - zero-filled | |
1715 | * Yes No "allocated" - allocated on disk, not read in | |
1716 | * Yes Yes "valid" - allocated and up-to-date in memory. | |
1717 | * | |
1718 | * "Dirty" is valid only with the last case (mapped+uptodate). | |
1719 | */ | |
1720 | ||
1721 | /* | |
1722 | * While block_write_full_page is writing back the dirty buffers under | |
1723 | * the page lock, whoever dirtied the buffers may decide to clean them | |
1724 | * again at any time. We handle that by only looking at the buffer | |
1725 | * state inside lock_buffer(). | |
1726 | * | |
1727 | * If block_write_full_page() is called for regular writeback | |
1728 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a | |
1729 | * locked buffer. This only can happen if someone has written the buffer | |
1730 | * directly, with submit_bh(). At the address_space level PageWriteback | |
1731 | * prevents this contention from occurring. | |
6e34eedd TT |
1732 | * |
1733 | * If block_write_full_page() is called with wbc->sync_mode == | |
70fd7614 | 1734 | * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this |
721a9602 | 1735 | * causes the writes to be flagged as synchronous writes. |
1da177e4 | 1736 | */ |
b4bba389 | 1737 | int __block_write_full_page(struct inode *inode, struct page *page, |
35c80d5f CM |
1738 | get_block_t *get_block, struct writeback_control *wbc, |
1739 | bh_end_io_t *handler) | |
1da177e4 LT |
1740 | { |
1741 | int err; | |
1742 | sector_t block; | |
1743 | sector_t last_block; | |
f0fbd5fc | 1744 | struct buffer_head *bh, *head; |
45bce8f3 | 1745 | unsigned int blocksize, bbits; |
1da177e4 | 1746 | int nr_underway = 0; |
3ae72869 | 1747 | blk_opf_t write_flags = wbc_to_write_flags(wbc); |
1da177e4 | 1748 | |
45bce8f3 | 1749 | head = create_page_buffers(page, inode, |
1da177e4 | 1750 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
1da177e4 LT |
1751 | |
1752 | /* | |
e621900a | 1753 | * Be very careful. We have no exclusion from block_dirty_folio |
1da177e4 LT |
1754 | * here, and the (potentially unmapped) buffers may become dirty at |
1755 | * any time. If a buffer becomes dirty here after we've inspected it | |
1756 | * then we just miss that fact, and the page stays dirty. | |
1757 | * | |
e621900a | 1758 | * Buffers outside i_size may be dirtied by block_dirty_folio; |
1da177e4 LT |
1759 | * handle that here by just cleaning them. |
1760 | */ | |
1761 | ||
1da177e4 | 1762 | bh = head; |
45bce8f3 LT |
1763 | blocksize = bh->b_size; |
1764 | bbits = block_size_bits(blocksize); | |
1765 | ||
09cbfeaf | 1766 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 1767 | last_block = (i_size_read(inode) - 1) >> bbits; |
1da177e4 LT |
1768 | |
1769 | /* | |
1770 | * Get all the dirty buffers mapped to disk addresses and | |
1771 | * handle any aliases from the underlying blockdev's mapping. | |
1772 | */ | |
1773 | do { | |
1774 | if (block > last_block) { | |
1775 | /* | |
1776 | * mapped buffers outside i_size will occur, because | |
1777 | * this page can be outside i_size when there is a | |
1778 | * truncate in progress. | |
1779 | */ | |
1780 | /* | |
1781 | * The buffer was zeroed by block_write_full_page() | |
1782 | */ | |
1783 | clear_buffer_dirty(bh); | |
1784 | set_buffer_uptodate(bh); | |
29a814d2 AT |
1785 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
1786 | buffer_dirty(bh)) { | |
b0cf2321 | 1787 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1788 | err = get_block(inode, block, bh, 1); |
1789 | if (err) | |
1790 | goto recover; | |
29a814d2 | 1791 | clear_buffer_delay(bh); |
1da177e4 LT |
1792 | if (buffer_new(bh)) { |
1793 | /* blockdev mappings never come here */ | |
1794 | clear_buffer_new(bh); | |
e64855c6 | 1795 | clean_bdev_bh_alias(bh); |
1da177e4 LT |
1796 | } |
1797 | } | |
1798 | bh = bh->b_this_page; | |
1799 | block++; | |
1800 | } while (bh != head); | |
1801 | ||
1802 | do { | |
1da177e4 LT |
1803 | if (!buffer_mapped(bh)) |
1804 | continue; | |
1805 | /* | |
1806 | * If it's a fully non-blocking write attempt and we cannot | |
1807 | * lock the buffer then redirty the page. Note that this can | |
5b0830cb JA |
1808 | * potentially cause a busy-wait loop from writeback threads |
1809 | * and kswapd activity, but those code paths have their own | |
1810 | * higher-level throttling. | |
1da177e4 | 1811 | */ |
1b430bee | 1812 | if (wbc->sync_mode != WB_SYNC_NONE) { |
1da177e4 | 1813 | lock_buffer(bh); |
ca5de404 | 1814 | } else if (!trylock_buffer(bh)) { |
1da177e4 LT |
1815 | redirty_page_for_writepage(wbc, page); |
1816 | continue; | |
1817 | } | |
1818 | if (test_clear_buffer_dirty(bh)) { | |
35c80d5f | 1819 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1820 | } else { |
1821 | unlock_buffer(bh); | |
1822 | } | |
1823 | } while ((bh = bh->b_this_page) != head); | |
1824 | ||
1825 | /* | |
1826 | * The page and its buffers are protected by PageWriteback(), so we can | |
1827 | * drop the bh refcounts early. | |
1828 | */ | |
1829 | BUG_ON(PageWriteback(page)); | |
1830 | set_page_writeback(page); | |
1da177e4 LT |
1831 | |
1832 | do { | |
1833 | struct buffer_head *next = bh->b_this_page; | |
1834 | if (buffer_async_write(bh)) { | |
1420c4a5 | 1835 | submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); |
1da177e4 LT |
1836 | nr_underway++; |
1837 | } | |
1da177e4 LT |
1838 | bh = next; |
1839 | } while (bh != head); | |
05937baa | 1840 | unlock_page(page); |
1da177e4 LT |
1841 | |
1842 | err = 0; | |
1843 | done: | |
1844 | if (nr_underway == 0) { | |
1845 | /* | |
1846 | * The page was marked dirty, but the buffers were | |
1847 | * clean. Someone wrote them back by hand with | |
79f59784 | 1848 | * write_dirty_buffer/submit_bh. A rare case. |
1da177e4 | 1849 | */ |
1da177e4 | 1850 | end_page_writeback(page); |
3d67f2d7 | 1851 | |
1da177e4 LT |
1852 | /* |
1853 | * The page and buffer_heads can be released at any time from | |
1854 | * here on. | |
1855 | */ | |
1da177e4 LT |
1856 | } |
1857 | return err; | |
1858 | ||
1859 | recover: | |
1860 | /* | |
1861 | * ENOSPC, or some other error. We may already have added some | |
1862 | * blocks to the file, so we need to write these out to avoid | |
1863 | * exposing stale data. | |
1864 | * The page is currently locked and not marked for writeback | |
1865 | */ | |
1866 | bh = head; | |
1867 | /* Recovery: lock and submit the mapped buffers */ | |
1868 | do { | |
29a814d2 AT |
1869 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
1870 | !buffer_delay(bh)) { | |
1da177e4 | 1871 | lock_buffer(bh); |
35c80d5f | 1872 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1873 | } else { |
1874 | /* | |
1875 | * The buffer may have been set dirty during | |
1876 | * attachment to a dirty page. | |
1877 | */ | |
1878 | clear_buffer_dirty(bh); | |
1879 | } | |
1880 | } while ((bh = bh->b_this_page) != head); | |
1881 | SetPageError(page); | |
1882 | BUG_ON(PageWriteback(page)); | |
7e4c3690 | 1883 | mapping_set_error(page->mapping, err); |
1da177e4 | 1884 | set_page_writeback(page); |
1da177e4 LT |
1885 | do { |
1886 | struct buffer_head *next = bh->b_this_page; | |
1887 | if (buffer_async_write(bh)) { | |
1888 | clear_buffer_dirty(bh); | |
1420c4a5 | 1889 | submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, wbc); |
1da177e4 LT |
1890 | nr_underway++; |
1891 | } | |
1da177e4 LT |
1892 | bh = next; |
1893 | } while (bh != head); | |
ffda9d30 | 1894 | unlock_page(page); |
1da177e4 LT |
1895 | goto done; |
1896 | } | |
b4bba389 | 1897 | EXPORT_SYMBOL(__block_write_full_page); |
1da177e4 | 1898 | |
afddba49 NP |
1899 | /* |
1900 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
1901 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1902 | * block data from leaking). And clear the new bit. | |
1903 | */ | |
1904 | void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1905 | { | |
1906 | unsigned int block_start, block_end; | |
1907 | struct buffer_head *head, *bh; | |
1908 | ||
1909 | BUG_ON(!PageLocked(page)); | |
1910 | if (!page_has_buffers(page)) | |
1911 | return; | |
1912 | ||
1913 | bh = head = page_buffers(page); | |
1914 | block_start = 0; | |
1915 | do { | |
1916 | block_end = block_start + bh->b_size; | |
1917 | ||
1918 | if (buffer_new(bh)) { | |
1919 | if (block_end > from && block_start < to) { | |
1920 | if (!PageUptodate(page)) { | |
1921 | unsigned start, size; | |
1922 | ||
1923 | start = max(from, block_start); | |
1924 | size = min(to, block_end) - start; | |
1925 | ||
eebd2aa3 | 1926 | zero_user(page, start, size); |
afddba49 NP |
1927 | set_buffer_uptodate(bh); |
1928 | } | |
1929 | ||
1930 | clear_buffer_new(bh); | |
1931 | mark_buffer_dirty(bh); | |
1932 | } | |
1933 | } | |
1934 | ||
1935 | block_start = block_end; | |
1936 | bh = bh->b_this_page; | |
1937 | } while (bh != head); | |
1938 | } | |
1939 | EXPORT_SYMBOL(page_zero_new_buffers); | |
1940 | ||
ae259a9c CH |
1941 | static void |
1942 | iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, | |
6d49cc85 | 1943 | const struct iomap *iomap) |
ae259a9c CH |
1944 | { |
1945 | loff_t offset = block << inode->i_blkbits; | |
1946 | ||
1947 | bh->b_bdev = iomap->bdev; | |
1948 | ||
1949 | /* | |
1950 | * Block points to offset in file we need to map, iomap contains | |
1951 | * the offset at which the map starts. If the map ends before the | |
1952 | * current block, then do not map the buffer and let the caller | |
1953 | * handle it. | |
1954 | */ | |
1955 | BUG_ON(offset >= iomap->offset + iomap->length); | |
1956 | ||
1957 | switch (iomap->type) { | |
1958 | case IOMAP_HOLE: | |
1959 | /* | |
1960 | * If the buffer is not up to date or beyond the current EOF, | |
1961 | * we need to mark it as new to ensure sub-block zeroing is | |
1962 | * executed if necessary. | |
1963 | */ | |
1964 | if (!buffer_uptodate(bh) || | |
1965 | (offset >= i_size_read(inode))) | |
1966 | set_buffer_new(bh); | |
1967 | break; | |
1968 | case IOMAP_DELALLOC: | |
1969 | if (!buffer_uptodate(bh) || | |
1970 | (offset >= i_size_read(inode))) | |
1971 | set_buffer_new(bh); | |
1972 | set_buffer_uptodate(bh); | |
1973 | set_buffer_mapped(bh); | |
1974 | set_buffer_delay(bh); | |
1975 | break; | |
1976 | case IOMAP_UNWRITTEN: | |
1977 | /* | |
3d7b6b21 AG |
1978 | * For unwritten regions, we always need to ensure that regions |
1979 | * in the block we are not writing to are zeroed. Mark the | |
1980 | * buffer as new to ensure this. | |
ae259a9c CH |
1981 | */ |
1982 | set_buffer_new(bh); | |
1983 | set_buffer_unwritten(bh); | |
df561f66 | 1984 | fallthrough; |
ae259a9c | 1985 | case IOMAP_MAPPED: |
3d7b6b21 AG |
1986 | if ((iomap->flags & IOMAP_F_NEW) || |
1987 | offset >= i_size_read(inode)) | |
ae259a9c | 1988 | set_buffer_new(bh); |
19fe5f64 AG |
1989 | bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> |
1990 | inode->i_blkbits; | |
ae259a9c CH |
1991 | set_buffer_mapped(bh); |
1992 | break; | |
1993 | } | |
1994 | } | |
1995 | ||
d1bd0b4e | 1996 | int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len, |
6d49cc85 | 1997 | get_block_t *get_block, const struct iomap *iomap) |
1da177e4 | 1998 | { |
09cbfeaf | 1999 | unsigned from = pos & (PAGE_SIZE - 1); |
ebdec241 | 2000 | unsigned to = from + len; |
d1bd0b4e | 2001 | struct inode *inode = folio->mapping->host; |
1da177e4 LT |
2002 | unsigned block_start, block_end; |
2003 | sector_t block; | |
2004 | int err = 0; | |
2005 | unsigned blocksize, bbits; | |
2006 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; | |
2007 | ||
d1bd0b4e | 2008 | BUG_ON(!folio_test_locked(folio)); |
09cbfeaf KS |
2009 | BUG_ON(from > PAGE_SIZE); |
2010 | BUG_ON(to > PAGE_SIZE); | |
1da177e4 LT |
2011 | BUG_ON(from > to); |
2012 | ||
d1bd0b4e | 2013 | head = create_page_buffers(&folio->page, inode, 0); |
45bce8f3 LT |
2014 | blocksize = head->b_size; |
2015 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2016 | |
d1bd0b4e | 2017 | block = (sector_t)folio->index << (PAGE_SHIFT - bbits); |
1da177e4 LT |
2018 | |
2019 | for(bh = head, block_start = 0; bh != head || !block_start; | |
2020 | block++, block_start=block_end, bh = bh->b_this_page) { | |
2021 | block_end = block_start + blocksize; | |
2022 | if (block_end <= from || block_start >= to) { | |
d1bd0b4e | 2023 | if (folio_test_uptodate(folio)) { |
1da177e4 LT |
2024 | if (!buffer_uptodate(bh)) |
2025 | set_buffer_uptodate(bh); | |
2026 | } | |
2027 | continue; | |
2028 | } | |
2029 | if (buffer_new(bh)) | |
2030 | clear_buffer_new(bh); | |
2031 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2032 | WARN_ON(bh->b_size != blocksize); |
ae259a9c CH |
2033 | if (get_block) { |
2034 | err = get_block(inode, block, bh, 1); | |
2035 | if (err) | |
2036 | break; | |
2037 | } else { | |
2038 | iomap_to_bh(inode, block, bh, iomap); | |
2039 | } | |
2040 | ||
1da177e4 | 2041 | if (buffer_new(bh)) { |
e64855c6 | 2042 | clean_bdev_bh_alias(bh); |
d1bd0b4e | 2043 | if (folio_test_uptodate(folio)) { |
637aff46 | 2044 | clear_buffer_new(bh); |
1da177e4 | 2045 | set_buffer_uptodate(bh); |
637aff46 | 2046 | mark_buffer_dirty(bh); |
1da177e4 LT |
2047 | continue; |
2048 | } | |
eebd2aa3 | 2049 | if (block_end > to || block_start < from) |
d1bd0b4e | 2050 | folio_zero_segments(folio, |
eebd2aa3 CL |
2051 | to, block_end, |
2052 | block_start, from); | |
1da177e4 LT |
2053 | continue; |
2054 | } | |
2055 | } | |
d1bd0b4e | 2056 | if (folio_test_uptodate(folio)) { |
1da177e4 LT |
2057 | if (!buffer_uptodate(bh)) |
2058 | set_buffer_uptodate(bh); | |
2059 | continue; | |
2060 | } | |
2061 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
33a266dd | 2062 | !buffer_unwritten(bh) && |
1da177e4 | 2063 | (block_start < from || block_end > to)) { |
e7ea1129 | 2064 | bh_read_nowait(bh, 0); |
1da177e4 LT |
2065 | *wait_bh++=bh; |
2066 | } | |
2067 | } | |
2068 | /* | |
2069 | * If we issued read requests - let them complete. | |
2070 | */ | |
2071 | while(wait_bh > wait) { | |
2072 | wait_on_buffer(*--wait_bh); | |
2073 | if (!buffer_uptodate(*wait_bh)) | |
f3ddbdc6 | 2074 | err = -EIO; |
1da177e4 | 2075 | } |
f9f07b6c | 2076 | if (unlikely(err)) |
d1bd0b4e | 2077 | page_zero_new_buffers(&folio->page, from, to); |
1da177e4 LT |
2078 | return err; |
2079 | } | |
ae259a9c CH |
2080 | |
2081 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, | |
2082 | get_block_t *get_block) | |
2083 | { | |
d1bd0b4e MWO |
2084 | return __block_write_begin_int(page_folio(page), pos, len, get_block, |
2085 | NULL); | |
ae259a9c | 2086 | } |
ebdec241 | 2087 | EXPORT_SYMBOL(__block_write_begin); |
1da177e4 LT |
2088 | |
2089 | static int __block_commit_write(struct inode *inode, struct page *page, | |
2090 | unsigned from, unsigned to) | |
2091 | { | |
2092 | unsigned block_start, block_end; | |
2093 | int partial = 0; | |
2094 | unsigned blocksize; | |
2095 | struct buffer_head *bh, *head; | |
2096 | ||
45bce8f3 LT |
2097 | bh = head = page_buffers(page); |
2098 | blocksize = bh->b_size; | |
1da177e4 | 2099 | |
45bce8f3 LT |
2100 | block_start = 0; |
2101 | do { | |
1da177e4 LT |
2102 | block_end = block_start + blocksize; |
2103 | if (block_end <= from || block_start >= to) { | |
2104 | if (!buffer_uptodate(bh)) | |
2105 | partial = 1; | |
2106 | } else { | |
2107 | set_buffer_uptodate(bh); | |
2108 | mark_buffer_dirty(bh); | |
2109 | } | |
4ebd3aec YG |
2110 | if (buffer_new(bh)) |
2111 | clear_buffer_new(bh); | |
45bce8f3 LT |
2112 | |
2113 | block_start = block_end; | |
2114 | bh = bh->b_this_page; | |
2115 | } while (bh != head); | |
1da177e4 LT |
2116 | |
2117 | /* | |
2118 | * If this is a partial write which happened to make all buffers | |
2c69e205 | 2119 | * uptodate then we can optimize away a bogus read_folio() for |
1da177e4 LT |
2120 | * the next read(). Here we 'discover' whether the page went |
2121 | * uptodate as a result of this (potentially partial) write. | |
2122 | */ | |
2123 | if (!partial) | |
2124 | SetPageUptodate(page); | |
2125 | return 0; | |
2126 | } | |
2127 | ||
afddba49 | 2128 | /* |
155130a4 CH |
2129 | * block_write_begin takes care of the basic task of block allocation and |
2130 | * bringing partial write blocks uptodate first. | |
2131 | * | |
7bb46a67 | 2132 | * The filesystem needs to handle block truncation upon failure. |
afddba49 | 2133 | */ |
155130a4 | 2134 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
b3992d1e | 2135 | struct page **pagep, get_block_t *get_block) |
afddba49 | 2136 | { |
09cbfeaf | 2137 | pgoff_t index = pos >> PAGE_SHIFT; |
afddba49 | 2138 | struct page *page; |
6e1db88d | 2139 | int status; |
afddba49 | 2140 | |
b7446e7c | 2141 | page = grab_cache_page_write_begin(mapping, index); |
6e1db88d CH |
2142 | if (!page) |
2143 | return -ENOMEM; | |
afddba49 | 2144 | |
6e1db88d | 2145 | status = __block_write_begin(page, pos, len, get_block); |
afddba49 | 2146 | if (unlikely(status)) { |
6e1db88d | 2147 | unlock_page(page); |
09cbfeaf | 2148 | put_page(page); |
6e1db88d | 2149 | page = NULL; |
afddba49 NP |
2150 | } |
2151 | ||
6e1db88d | 2152 | *pagep = page; |
afddba49 NP |
2153 | return status; |
2154 | } | |
2155 | EXPORT_SYMBOL(block_write_begin); | |
2156 | ||
2157 | int block_write_end(struct file *file, struct address_space *mapping, | |
2158 | loff_t pos, unsigned len, unsigned copied, | |
2159 | struct page *page, void *fsdata) | |
2160 | { | |
2161 | struct inode *inode = mapping->host; | |
2162 | unsigned start; | |
2163 | ||
09cbfeaf | 2164 | start = pos & (PAGE_SIZE - 1); |
afddba49 NP |
2165 | |
2166 | if (unlikely(copied < len)) { | |
2167 | /* | |
2c69e205 MWO |
2168 | * The buffers that were written will now be uptodate, so |
2169 | * we don't have to worry about a read_folio reading them | |
2170 | * and overwriting a partial write. However if we have | |
2171 | * encountered a short write and only partially written | |
2172 | * into a buffer, it will not be marked uptodate, so a | |
2173 | * read_folio might come in and destroy our partial write. | |
afddba49 NP |
2174 | * |
2175 | * Do the simplest thing, and just treat any short write to a | |
2176 | * non uptodate page as a zero-length write, and force the | |
2177 | * caller to redo the whole thing. | |
2178 | */ | |
2179 | if (!PageUptodate(page)) | |
2180 | copied = 0; | |
2181 | ||
2182 | page_zero_new_buffers(page, start+copied, start+len); | |
2183 | } | |
2184 | flush_dcache_page(page); | |
2185 | ||
2186 | /* This could be a short (even 0-length) commit */ | |
2187 | __block_commit_write(inode, page, start, start+copied); | |
2188 | ||
2189 | return copied; | |
2190 | } | |
2191 | EXPORT_SYMBOL(block_write_end); | |
2192 | ||
2193 | int generic_write_end(struct file *file, struct address_space *mapping, | |
2194 | loff_t pos, unsigned len, unsigned copied, | |
2195 | struct page *page, void *fsdata) | |
2196 | { | |
8af54f29 CH |
2197 | struct inode *inode = mapping->host; |
2198 | loff_t old_size = inode->i_size; | |
2199 | bool i_size_changed = false; | |
2200 | ||
afddba49 | 2201 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
8af54f29 CH |
2202 | |
2203 | /* | |
2204 | * No need to use i_size_read() here, the i_size cannot change under us | |
2205 | * because we hold i_rwsem. | |
2206 | * | |
2207 | * But it's important to update i_size while still holding page lock: | |
2208 | * page writeout could otherwise come in and zero beyond i_size. | |
2209 | */ | |
2210 | if (pos + copied > inode->i_size) { | |
2211 | i_size_write(inode, pos + copied); | |
2212 | i_size_changed = true; | |
2213 | } | |
2214 | ||
2215 | unlock_page(page); | |
7a77dad7 | 2216 | put_page(page); |
8af54f29 CH |
2217 | |
2218 | if (old_size < pos) | |
2219 | pagecache_isize_extended(inode, old_size, pos); | |
2220 | /* | |
2221 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
2222 | * makes the holding time of page lock longer. Second, it forces lock | |
2223 | * ordering of page lock and transaction start for journaling | |
2224 | * filesystems. | |
2225 | */ | |
2226 | if (i_size_changed) | |
2227 | mark_inode_dirty(inode); | |
26ddb1f4 | 2228 | return copied; |
afddba49 NP |
2229 | } |
2230 | EXPORT_SYMBOL(generic_write_end); | |
2231 | ||
8ab22b9a | 2232 | /* |
2e7e80f7 | 2233 | * block_is_partially_uptodate checks whether buffers within a folio are |
8ab22b9a HH |
2234 | * uptodate or not. |
2235 | * | |
2e7e80f7 MWO |
2236 | * Returns true if all buffers which correspond to the specified part |
2237 | * of the folio are uptodate. | |
8ab22b9a | 2238 | */ |
2e7e80f7 | 2239 | bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count) |
8ab22b9a | 2240 | { |
8ab22b9a HH |
2241 | unsigned block_start, block_end, blocksize; |
2242 | unsigned to; | |
2243 | struct buffer_head *bh, *head; | |
2e7e80f7 | 2244 | bool ret = true; |
8ab22b9a | 2245 | |
2e7e80f7 MWO |
2246 | head = folio_buffers(folio); |
2247 | if (!head) | |
2248 | return false; | |
45bce8f3 | 2249 | blocksize = head->b_size; |
2e7e80f7 | 2250 | to = min_t(unsigned, folio_size(folio) - from, count); |
8ab22b9a | 2251 | to = from + to; |
2e7e80f7 MWO |
2252 | if (from < blocksize && to > folio_size(folio) - blocksize) |
2253 | return false; | |
8ab22b9a | 2254 | |
8ab22b9a HH |
2255 | bh = head; |
2256 | block_start = 0; | |
2257 | do { | |
2258 | block_end = block_start + blocksize; | |
2259 | if (block_end > from && block_start < to) { | |
2260 | if (!buffer_uptodate(bh)) { | |
2e7e80f7 | 2261 | ret = false; |
8ab22b9a HH |
2262 | break; |
2263 | } | |
2264 | if (block_end >= to) | |
2265 | break; | |
2266 | } | |
2267 | block_start = block_end; | |
2268 | bh = bh->b_this_page; | |
2269 | } while (bh != head); | |
2270 | ||
2271 | return ret; | |
2272 | } | |
2273 | EXPORT_SYMBOL(block_is_partially_uptodate); | |
2274 | ||
1da177e4 | 2275 | /* |
2c69e205 | 2276 | * Generic "read_folio" function for block devices that have the normal |
1da177e4 | 2277 | * get_block functionality. This is most of the block device filesystems. |
2c69e205 | 2278 | * Reads the folio asynchronously --- the unlock_buffer() and |
1da177e4 | 2279 | * set/clear_buffer_uptodate() functions propagate buffer state into the |
2c69e205 | 2280 | * folio once IO has completed. |
1da177e4 | 2281 | */ |
2c69e205 | 2282 | int block_read_full_folio(struct folio *folio, get_block_t *get_block) |
1da177e4 | 2283 | { |
2c69e205 | 2284 | struct inode *inode = folio->mapping->host; |
1da177e4 LT |
2285 | sector_t iblock, lblock; |
2286 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; | |
45bce8f3 | 2287 | unsigned int blocksize, bbits; |
1da177e4 LT |
2288 | int nr, i; |
2289 | int fully_mapped = 1; | |
b7a6eb22 | 2290 | bool page_error = false; |
4fa512ce EB |
2291 | loff_t limit = i_size_read(inode); |
2292 | ||
2293 | /* This is needed for ext4. */ | |
2294 | if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) | |
2295 | limit = inode->i_sb->s_maxbytes; | |
1da177e4 | 2296 | |
2c69e205 MWO |
2297 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
2298 | ||
2299 | head = create_page_buffers(&folio->page, inode, 0); | |
45bce8f3 LT |
2300 | blocksize = head->b_size; |
2301 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2302 | |
2c69e205 | 2303 | iblock = (sector_t)folio->index << (PAGE_SHIFT - bbits); |
4fa512ce | 2304 | lblock = (limit+blocksize-1) >> bbits; |
1da177e4 LT |
2305 | bh = head; |
2306 | nr = 0; | |
2307 | i = 0; | |
2308 | ||
2309 | do { | |
2310 | if (buffer_uptodate(bh)) | |
2311 | continue; | |
2312 | ||
2313 | if (!buffer_mapped(bh)) { | |
c64610ba AM |
2314 | int err = 0; |
2315 | ||
1da177e4 LT |
2316 | fully_mapped = 0; |
2317 | if (iblock < lblock) { | |
b0cf2321 | 2318 | WARN_ON(bh->b_size != blocksize); |
c64610ba | 2319 | err = get_block(inode, iblock, bh, 0); |
b7a6eb22 | 2320 | if (err) { |
2c69e205 | 2321 | folio_set_error(folio); |
b7a6eb22 MWO |
2322 | page_error = true; |
2323 | } | |
1da177e4 LT |
2324 | } |
2325 | if (!buffer_mapped(bh)) { | |
2c69e205 MWO |
2326 | folio_zero_range(folio, i * blocksize, |
2327 | blocksize); | |
c64610ba AM |
2328 | if (!err) |
2329 | set_buffer_uptodate(bh); | |
1da177e4 LT |
2330 | continue; |
2331 | } | |
2332 | /* | |
2333 | * get_block() might have updated the buffer | |
2334 | * synchronously | |
2335 | */ | |
2336 | if (buffer_uptodate(bh)) | |
2337 | continue; | |
2338 | } | |
2339 | arr[nr++] = bh; | |
2340 | } while (i++, iblock++, (bh = bh->b_this_page) != head); | |
2341 | ||
2342 | if (fully_mapped) | |
2c69e205 | 2343 | folio_set_mappedtodisk(folio); |
1da177e4 LT |
2344 | |
2345 | if (!nr) { | |
2346 | /* | |
2c69e205 | 2347 | * All buffers are uptodate - we can set the folio uptodate |
1da177e4 LT |
2348 | * as well. But not if get_block() returned an error. |
2349 | */ | |
b7a6eb22 | 2350 | if (!page_error) |
2c69e205 MWO |
2351 | folio_mark_uptodate(folio); |
2352 | folio_unlock(folio); | |
1da177e4 LT |
2353 | return 0; |
2354 | } | |
2355 | ||
2356 | /* Stage two: lock the buffers */ | |
2357 | for (i = 0; i < nr; i++) { | |
2358 | bh = arr[i]; | |
2359 | lock_buffer(bh); | |
2360 | mark_buffer_async_read(bh); | |
2361 | } | |
2362 | ||
2363 | /* | |
2364 | * Stage 3: start the IO. Check for uptodateness | |
2365 | * inside the buffer lock in case another process reading | |
2366 | * the underlying blockdev brought it uptodate (the sct fix). | |
2367 | */ | |
2368 | for (i = 0; i < nr; i++) { | |
2369 | bh = arr[i]; | |
2370 | if (buffer_uptodate(bh)) | |
2371 | end_buffer_async_read(bh, 1); | |
2372 | else | |
1420c4a5 | 2373 | submit_bh(REQ_OP_READ, bh); |
1da177e4 LT |
2374 | } |
2375 | return 0; | |
2376 | } | |
2c69e205 | 2377 | EXPORT_SYMBOL(block_read_full_folio); |
1da177e4 LT |
2378 | |
2379 | /* utility function for filesystems that need to do work on expanding | |
89e10787 | 2380 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
1da177e4 LT |
2381 | * deal with the hole. |
2382 | */ | |
89e10787 | 2383 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
1da177e4 LT |
2384 | { |
2385 | struct address_space *mapping = inode->i_mapping; | |
53b524b8 | 2386 | const struct address_space_operations *aops = mapping->a_ops; |
1da177e4 | 2387 | struct page *page; |
1468c6f4 | 2388 | void *fsdata = NULL; |
1da177e4 LT |
2389 | int err; |
2390 | ||
c08d3b0e | 2391 | err = inode_newsize_ok(inode, size); |
2392 | if (err) | |
1da177e4 LT |
2393 | goto out; |
2394 | ||
53b524b8 | 2395 | err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata); |
89e10787 | 2396 | if (err) |
05eb0b51 | 2397 | goto out; |
05eb0b51 | 2398 | |
53b524b8 | 2399 | err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata); |
89e10787 | 2400 | BUG_ON(err > 0); |
05eb0b51 | 2401 | |
1da177e4 LT |
2402 | out: |
2403 | return err; | |
2404 | } | |
1fe72eaa | 2405 | EXPORT_SYMBOL(generic_cont_expand_simple); |
1da177e4 | 2406 | |
f1e3af72 AB |
2407 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
2408 | loff_t pos, loff_t *bytes) | |
1da177e4 | 2409 | { |
1da177e4 | 2410 | struct inode *inode = mapping->host; |
53b524b8 | 2411 | const struct address_space_operations *aops = mapping->a_ops; |
93407472 | 2412 | unsigned int blocksize = i_blocksize(inode); |
89e10787 | 2413 | struct page *page; |
1468c6f4 | 2414 | void *fsdata = NULL; |
89e10787 NP |
2415 | pgoff_t index, curidx; |
2416 | loff_t curpos; | |
2417 | unsigned zerofrom, offset, len; | |
2418 | int err = 0; | |
1da177e4 | 2419 | |
09cbfeaf KS |
2420 | index = pos >> PAGE_SHIFT; |
2421 | offset = pos & ~PAGE_MASK; | |
89e10787 | 2422 | |
09cbfeaf KS |
2423 | while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { |
2424 | zerofrom = curpos & ~PAGE_MASK; | |
1da177e4 LT |
2425 | if (zerofrom & (blocksize-1)) { |
2426 | *bytes |= (blocksize-1); | |
2427 | (*bytes)++; | |
2428 | } | |
09cbfeaf | 2429 | len = PAGE_SIZE - zerofrom; |
1da177e4 | 2430 | |
53b524b8 | 2431 | err = aops->write_begin(file, mapping, curpos, len, |
c718a975 | 2432 | &page, &fsdata); |
89e10787 NP |
2433 | if (err) |
2434 | goto out; | |
eebd2aa3 | 2435 | zero_user(page, zerofrom, len); |
53b524b8 | 2436 | err = aops->write_end(file, mapping, curpos, len, len, |
89e10787 NP |
2437 | page, fsdata); |
2438 | if (err < 0) | |
2439 | goto out; | |
2440 | BUG_ON(err != len); | |
2441 | err = 0; | |
061e9746 OH |
2442 | |
2443 | balance_dirty_pages_ratelimited(mapping); | |
c2ca0fcd | 2444 | |
08d405c8 | 2445 | if (fatal_signal_pending(current)) { |
c2ca0fcd MP |
2446 | err = -EINTR; |
2447 | goto out; | |
2448 | } | |
89e10787 | 2449 | } |
1da177e4 | 2450 | |
89e10787 NP |
2451 | /* page covers the boundary, find the boundary offset */ |
2452 | if (index == curidx) { | |
09cbfeaf | 2453 | zerofrom = curpos & ~PAGE_MASK; |
1da177e4 | 2454 | /* if we will expand the thing last block will be filled */ |
89e10787 NP |
2455 | if (offset <= zerofrom) { |
2456 | goto out; | |
2457 | } | |
2458 | if (zerofrom & (blocksize-1)) { | |
1da177e4 LT |
2459 | *bytes |= (blocksize-1); |
2460 | (*bytes)++; | |
2461 | } | |
89e10787 | 2462 | len = offset - zerofrom; |
1da177e4 | 2463 | |
53b524b8 | 2464 | err = aops->write_begin(file, mapping, curpos, len, |
c718a975 | 2465 | &page, &fsdata); |
89e10787 NP |
2466 | if (err) |
2467 | goto out; | |
eebd2aa3 | 2468 | zero_user(page, zerofrom, len); |
53b524b8 | 2469 | err = aops->write_end(file, mapping, curpos, len, len, |
89e10787 NP |
2470 | page, fsdata); |
2471 | if (err < 0) | |
2472 | goto out; | |
2473 | BUG_ON(err != len); | |
2474 | err = 0; | |
1da177e4 | 2475 | } |
89e10787 NP |
2476 | out: |
2477 | return err; | |
2478 | } | |
2479 | ||
2480 | /* | |
2481 | * For moronic filesystems that do not allow holes in file. | |
2482 | * We may have to extend the file. | |
2483 | */ | |
282dc178 | 2484 | int cont_write_begin(struct file *file, struct address_space *mapping, |
be3bbbc5 | 2485 | loff_t pos, unsigned len, |
89e10787 NP |
2486 | struct page **pagep, void **fsdata, |
2487 | get_block_t *get_block, loff_t *bytes) | |
2488 | { | |
2489 | struct inode *inode = mapping->host; | |
93407472 FF |
2490 | unsigned int blocksize = i_blocksize(inode); |
2491 | unsigned int zerofrom; | |
89e10787 NP |
2492 | int err; |
2493 | ||
2494 | err = cont_expand_zero(file, mapping, pos, bytes); | |
2495 | if (err) | |
155130a4 | 2496 | return err; |
89e10787 | 2497 | |
09cbfeaf | 2498 | zerofrom = *bytes & ~PAGE_MASK; |
89e10787 NP |
2499 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { |
2500 | *bytes |= (blocksize-1); | |
2501 | (*bytes)++; | |
1da177e4 | 2502 | } |
1da177e4 | 2503 | |
b3992d1e | 2504 | return block_write_begin(mapping, pos, len, pagep, get_block); |
1da177e4 | 2505 | } |
1fe72eaa | 2506 | EXPORT_SYMBOL(cont_write_begin); |
1da177e4 | 2507 | |
1da177e4 LT |
2508 | int block_commit_write(struct page *page, unsigned from, unsigned to) |
2509 | { | |
2510 | struct inode *inode = page->mapping->host; | |
2511 | __block_commit_write(inode,page,from,to); | |
2512 | return 0; | |
2513 | } | |
1fe72eaa | 2514 | EXPORT_SYMBOL(block_commit_write); |
1da177e4 | 2515 | |
54171690 DC |
2516 | /* |
2517 | * block_page_mkwrite() is not allowed to change the file size as it gets | |
2518 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2519 | * be careful to check for EOF conditions here. We set the page up correctly | |
2520 | * for a written page which means we get ENOSPC checking when writing into | |
2521 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2522 | * support these features. | |
2523 | * | |
2524 | * We are not allowed to take the i_mutex here so we have to play games to | |
2525 | * protect against truncate races as the page could now be beyond EOF. Because | |
7bb46a67 | 2526 | * truncate writes the inode size before removing pages, once we have the |
54171690 DC |
2527 | * page lock we can determine safely if the page is beyond EOF. If it is not |
2528 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2529 | * unlock the page. | |
ea13a864 | 2530 | * |
14da9200 | 2531 | * Direct callers of this function should protect against filesystem freezing |
5c500029 | 2532 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
54171690 | 2533 | */ |
5c500029 | 2534 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
24da4fab | 2535 | get_block_t get_block) |
54171690 | 2536 | { |
c2ec175c | 2537 | struct page *page = vmf->page; |
496ad9aa | 2538 | struct inode *inode = file_inode(vma->vm_file); |
54171690 DC |
2539 | unsigned long end; |
2540 | loff_t size; | |
24da4fab | 2541 | int ret; |
54171690 DC |
2542 | |
2543 | lock_page(page); | |
2544 | size = i_size_read(inode); | |
2545 | if ((page->mapping != inode->i_mapping) || | |
18336338 | 2546 | (page_offset(page) > size)) { |
24da4fab JK |
2547 | /* We overload EFAULT to mean page got truncated */ |
2548 | ret = -EFAULT; | |
2549 | goto out_unlock; | |
54171690 DC |
2550 | } |
2551 | ||
2552 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
2553 | if (((page->index + 1) << PAGE_SHIFT) > size) |
2554 | end = size & ~PAGE_MASK; | |
54171690 | 2555 | else |
09cbfeaf | 2556 | end = PAGE_SIZE; |
54171690 | 2557 | |
ebdec241 | 2558 | ret = __block_write_begin(page, 0, end, get_block); |
54171690 DC |
2559 | if (!ret) |
2560 | ret = block_commit_write(page, 0, end); | |
2561 | ||
24da4fab JK |
2562 | if (unlikely(ret < 0)) |
2563 | goto out_unlock; | |
ea13a864 | 2564 | set_page_dirty(page); |
1d1d1a76 | 2565 | wait_for_stable_page(page); |
24da4fab JK |
2566 | return 0; |
2567 | out_unlock: | |
2568 | unlock_page(page); | |
54171690 | 2569 | return ret; |
24da4fab | 2570 | } |
1fe72eaa | 2571 | EXPORT_SYMBOL(block_page_mkwrite); |
1da177e4 | 2572 | |
1da177e4 LT |
2573 | int block_truncate_page(struct address_space *mapping, |
2574 | loff_t from, get_block_t *get_block) | |
2575 | { | |
09cbfeaf KS |
2576 | pgoff_t index = from >> PAGE_SHIFT; |
2577 | unsigned offset = from & (PAGE_SIZE-1); | |
1da177e4 | 2578 | unsigned blocksize; |
54b21a79 | 2579 | sector_t iblock; |
1da177e4 LT |
2580 | unsigned length, pos; |
2581 | struct inode *inode = mapping->host; | |
2582 | struct page *page; | |
2583 | struct buffer_head *bh; | |
1da177e4 LT |
2584 | int err; |
2585 | ||
93407472 | 2586 | blocksize = i_blocksize(inode); |
1da177e4 LT |
2587 | length = offset & (blocksize - 1); |
2588 | ||
2589 | /* Block boundary? Nothing to do */ | |
2590 | if (!length) | |
2591 | return 0; | |
2592 | ||
2593 | length = blocksize - length; | |
09cbfeaf | 2594 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 LT |
2595 | |
2596 | page = grab_cache_page(mapping, index); | |
2597 | err = -ENOMEM; | |
2598 | if (!page) | |
2599 | goto out; | |
2600 | ||
2601 | if (!page_has_buffers(page)) | |
2602 | create_empty_buffers(page, blocksize, 0); | |
2603 | ||
2604 | /* Find the buffer that contains "offset" */ | |
2605 | bh = page_buffers(page); | |
2606 | pos = blocksize; | |
2607 | while (offset >= pos) { | |
2608 | bh = bh->b_this_page; | |
2609 | iblock++; | |
2610 | pos += blocksize; | |
2611 | } | |
2612 | ||
2613 | err = 0; | |
2614 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2615 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
2616 | err = get_block(inode, iblock, bh, 0); |
2617 | if (err) | |
2618 | goto unlock; | |
2619 | /* unmapped? It's a hole - nothing to do */ | |
2620 | if (!buffer_mapped(bh)) | |
2621 | goto unlock; | |
2622 | } | |
2623 | ||
2624 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2625 | if (PageUptodate(page)) | |
2626 | set_buffer_uptodate(bh); | |
2627 | ||
33a266dd | 2628 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
e7ea1129 | 2629 | err = bh_read(bh, 0); |
1da177e4 | 2630 | /* Uhhuh. Read error. Complain and punt. */ |
e7ea1129 | 2631 | if (err < 0) |
1da177e4 LT |
2632 | goto unlock; |
2633 | } | |
2634 | ||
eebd2aa3 | 2635 | zero_user(page, offset, length); |
1da177e4 LT |
2636 | mark_buffer_dirty(bh); |
2637 | err = 0; | |
2638 | ||
2639 | unlock: | |
2640 | unlock_page(page); | |
09cbfeaf | 2641 | put_page(page); |
1da177e4 LT |
2642 | out: |
2643 | return err; | |
2644 | } | |
1fe72eaa | 2645 | EXPORT_SYMBOL(block_truncate_page); |
1da177e4 LT |
2646 | |
2647 | /* | |
2648 | * The generic ->writepage function for buffer-backed address_spaces | |
2649 | */ | |
1b938c08 MW |
2650 | int block_write_full_page(struct page *page, get_block_t *get_block, |
2651 | struct writeback_control *wbc) | |
1da177e4 LT |
2652 | { |
2653 | struct inode * const inode = page->mapping->host; | |
2654 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2655 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2656 | unsigned offset; |
1da177e4 LT |
2657 | |
2658 | /* Is the page fully inside i_size? */ | |
2659 | if (page->index < end_index) | |
35c80d5f | 2660 | return __block_write_full_page(inode, page, get_block, wbc, |
1b938c08 | 2661 | end_buffer_async_write); |
1da177e4 LT |
2662 | |
2663 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2664 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 | 2665 | if (page->index >= end_index+1 || !offset) { |
1da177e4 LT |
2666 | unlock_page(page); |
2667 | return 0; /* don't care */ | |
2668 | } | |
2669 | ||
2670 | /* | |
2671 | * The page straddles i_size. It must be zeroed out on each and every | |
2a61aa40 | 2672 | * writepage invocation because it may be mmapped. "A file is mapped |
1da177e4 LT |
2673 | * in multiples of the page size. For a file that is not a multiple of |
2674 | * the page size, the remaining memory is zeroed when mapped, and | |
2675 | * writes to that region are not written out to the file." | |
2676 | */ | |
09cbfeaf | 2677 | zero_user_segment(page, offset, PAGE_SIZE); |
1b938c08 MW |
2678 | return __block_write_full_page(inode, page, get_block, wbc, |
2679 | end_buffer_async_write); | |
35c80d5f | 2680 | } |
1fe72eaa | 2681 | EXPORT_SYMBOL(block_write_full_page); |
35c80d5f | 2682 | |
1da177e4 LT |
2683 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
2684 | get_block_t *get_block) | |
2685 | { | |
1da177e4 | 2686 | struct inode *inode = mapping->host; |
2a527d68 AP |
2687 | struct buffer_head tmp = { |
2688 | .b_size = i_blocksize(inode), | |
2689 | }; | |
2690 | ||
1da177e4 LT |
2691 | get_block(inode, block, &tmp, 0); |
2692 | return tmp.b_blocknr; | |
2693 | } | |
1fe72eaa | 2694 | EXPORT_SYMBOL(generic_block_bmap); |
1da177e4 | 2695 | |
4246a0b6 | 2696 | static void end_bio_bh_io_sync(struct bio *bio) |
1da177e4 LT |
2697 | { |
2698 | struct buffer_head *bh = bio->bi_private; | |
2699 | ||
b7c44ed9 | 2700 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
08bafc03 KM |
2701 | set_bit(BH_Quiet, &bh->b_state); |
2702 | ||
4e4cbee9 | 2703 | bh->b_end_io(bh, !bio->bi_status); |
1da177e4 | 2704 | bio_put(bio); |
1da177e4 LT |
2705 | } |
2706 | ||
5bdf402a RHI |
2707 | static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, |
2708 | struct writeback_control *wbc) | |
1da177e4 | 2709 | { |
1420c4a5 | 2710 | const enum req_op op = opf & REQ_OP_MASK; |
1da177e4 | 2711 | struct bio *bio; |
1da177e4 LT |
2712 | |
2713 | BUG_ON(!buffer_locked(bh)); | |
2714 | BUG_ON(!buffer_mapped(bh)); | |
2715 | BUG_ON(!bh->b_end_io); | |
8fb0e342 AK |
2716 | BUG_ON(buffer_delay(bh)); |
2717 | BUG_ON(buffer_unwritten(bh)); | |
1da177e4 | 2718 | |
1da177e4 | 2719 | /* |
48fd4f93 | 2720 | * Only clear out a write error when rewriting |
1da177e4 | 2721 | */ |
2a222ca9 | 2722 | if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) |
1da177e4 LT |
2723 | clear_buffer_write_io_error(bh); |
2724 | ||
07888c66 | 2725 | if (buffer_meta(bh)) |
1420c4a5 | 2726 | opf |= REQ_META; |
07888c66 | 2727 | if (buffer_prio(bh)) |
1420c4a5 | 2728 | opf |= REQ_PRIO; |
07888c66 | 2729 | |
1420c4a5 | 2730 | bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO); |
1da177e4 | 2731 | |
4f74d15f EB |
2732 | fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); |
2733 | ||
4f024f37 | 2734 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
1da177e4 | 2735 | |
6cf66b4c KO |
2736 | bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
2737 | BUG_ON(bio->bi_iter.bi_size != bh->b_size); | |
1da177e4 LT |
2738 | |
2739 | bio->bi_end_io = end_bio_bh_io_sync; | |
2740 | bio->bi_private = bh; | |
2741 | ||
83c9c547 ML |
2742 | /* Take care of bh's that straddle the end of the device */ |
2743 | guard_bio_eod(bio); | |
2744 | ||
fd42df30 DZ |
2745 | if (wbc) { |
2746 | wbc_init_bio(wbc, bio); | |
34e51a5e | 2747 | wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); |
fd42df30 DZ |
2748 | } |
2749 | ||
4e49ea4a | 2750 | submit_bio(bio); |
1da177e4 | 2751 | } |
bafc0dba | 2752 | |
5bdf402a | 2753 | void submit_bh(blk_opf_t opf, struct buffer_head *bh) |
bafc0dba | 2754 | { |
5bdf402a | 2755 | submit_bh_wbc(opf, bh, NULL); |
71368511 | 2756 | } |
1fe72eaa | 2757 | EXPORT_SYMBOL(submit_bh); |
1da177e4 | 2758 | |
3ae72869 | 2759 | void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) |
9cb569d6 CH |
2760 | { |
2761 | lock_buffer(bh); | |
2762 | if (!test_clear_buffer_dirty(bh)) { | |
2763 | unlock_buffer(bh); | |
2764 | return; | |
2765 | } | |
2766 | bh->b_end_io = end_buffer_write_sync; | |
2767 | get_bh(bh); | |
1420c4a5 | 2768 | submit_bh(REQ_OP_WRITE | op_flags, bh); |
9cb569d6 CH |
2769 | } |
2770 | EXPORT_SYMBOL(write_dirty_buffer); | |
2771 | ||
1da177e4 LT |
2772 | /* |
2773 | * For a data-integrity writeout, we need to wait upon any in-progress I/O | |
2774 | * and then start new I/O and then wait upon it. The caller must have a ref on | |
2775 | * the buffer_head. | |
2776 | */ | |
3ae72869 | 2777 | int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) |
1da177e4 | 2778 | { |
1da177e4 LT |
2779 | WARN_ON(atomic_read(&bh->b_count) < 1); |
2780 | lock_buffer(bh); | |
2781 | if (test_clear_buffer_dirty(bh)) { | |
377254b2 XT |
2782 | /* |
2783 | * The bh should be mapped, but it might not be if the | |
2784 | * device was hot-removed. Not much we can do but fail the I/O. | |
2785 | */ | |
2786 | if (!buffer_mapped(bh)) { | |
2787 | unlock_buffer(bh); | |
2788 | return -EIO; | |
2789 | } | |
2790 | ||
1da177e4 LT |
2791 | get_bh(bh); |
2792 | bh->b_end_io = end_buffer_write_sync; | |
ab620620 | 2793 | submit_bh(REQ_OP_WRITE | op_flags, bh); |
1da177e4 | 2794 | wait_on_buffer(bh); |
ab620620 RHI |
2795 | if (!buffer_uptodate(bh)) |
2796 | return -EIO; | |
1da177e4 LT |
2797 | } else { |
2798 | unlock_buffer(bh); | |
2799 | } | |
ab620620 | 2800 | return 0; |
1da177e4 | 2801 | } |
87e99511 CH |
2802 | EXPORT_SYMBOL(__sync_dirty_buffer); |
2803 | ||
2804 | int sync_dirty_buffer(struct buffer_head *bh) | |
2805 | { | |
70fd7614 | 2806 | return __sync_dirty_buffer(bh, REQ_SYNC); |
87e99511 | 2807 | } |
1fe72eaa | 2808 | EXPORT_SYMBOL(sync_dirty_buffer); |
1da177e4 LT |
2809 | |
2810 | /* | |
68189fef | 2811 | * try_to_free_buffers() checks if all the buffers on this particular folio |
1da177e4 LT |
2812 | * are unused, and releases them if so. |
2813 | * | |
2814 | * Exclusion against try_to_free_buffers may be obtained by either | |
68189fef | 2815 | * locking the folio or by holding its mapping's private_lock. |
1da177e4 | 2816 | * |
68189fef MWO |
2817 | * If the folio is dirty but all the buffers are clean then we need to |
2818 | * be sure to mark the folio clean as well. This is because the folio | |
1da177e4 | 2819 | * may be against a block device, and a later reattachment of buffers |
68189fef | 2820 | * to a dirty folio will set *all* buffers dirty. Which would corrupt |
1da177e4 LT |
2821 | * filesystem data on the same device. |
2822 | * | |
68189fef MWO |
2823 | * The same applies to regular filesystem folios: if all the buffers are |
2824 | * clean then we set the folio clean and proceed. To do that, we require | |
e621900a | 2825 | * total exclusion from block_dirty_folio(). That is obtained with |
1da177e4 LT |
2826 | * private_lock. |
2827 | * | |
2828 | * try_to_free_buffers() is non-blocking. | |
2829 | */ | |
2830 | static inline int buffer_busy(struct buffer_head *bh) | |
2831 | { | |
2832 | return atomic_read(&bh->b_count) | | |
2833 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); | |
2834 | } | |
2835 | ||
64394763 MWO |
2836 | static bool |
2837 | drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free) | |
1da177e4 | 2838 | { |
64394763 | 2839 | struct buffer_head *head = folio_buffers(folio); |
1da177e4 LT |
2840 | struct buffer_head *bh; |
2841 | ||
2842 | bh = head; | |
2843 | do { | |
1da177e4 LT |
2844 | if (buffer_busy(bh)) |
2845 | goto failed; | |
2846 | bh = bh->b_this_page; | |
2847 | } while (bh != head); | |
2848 | ||
2849 | do { | |
2850 | struct buffer_head *next = bh->b_this_page; | |
2851 | ||
535ee2fb | 2852 | if (bh->b_assoc_map) |
1da177e4 LT |
2853 | __remove_assoc_queue(bh); |
2854 | bh = next; | |
2855 | } while (bh != head); | |
2856 | *buffers_to_free = head; | |
64394763 MWO |
2857 | folio_detach_private(folio); |
2858 | return true; | |
1da177e4 | 2859 | failed: |
64394763 | 2860 | return false; |
1da177e4 LT |
2861 | } |
2862 | ||
68189fef | 2863 | bool try_to_free_buffers(struct folio *folio) |
1da177e4 | 2864 | { |
68189fef | 2865 | struct address_space * const mapping = folio->mapping; |
1da177e4 | 2866 | struct buffer_head *buffers_to_free = NULL; |
68189fef | 2867 | bool ret = 0; |
1da177e4 | 2868 | |
68189fef MWO |
2869 | BUG_ON(!folio_test_locked(folio)); |
2870 | if (folio_test_writeback(folio)) | |
2871 | return false; | |
1da177e4 LT |
2872 | |
2873 | if (mapping == NULL) { /* can this still happen? */ | |
64394763 | 2874 | ret = drop_buffers(folio, &buffers_to_free); |
1da177e4 LT |
2875 | goto out; |
2876 | } | |
2877 | ||
2878 | spin_lock(&mapping->private_lock); | |
64394763 | 2879 | ret = drop_buffers(folio, &buffers_to_free); |
ecdfc978 LT |
2880 | |
2881 | /* | |
2882 | * If the filesystem writes its buffers by hand (eg ext3) | |
68189fef MWO |
2883 | * then we can have clean buffers against a dirty folio. We |
2884 | * clean the folio here; otherwise the VM will never notice | |
ecdfc978 LT |
2885 | * that the filesystem did any IO at all. |
2886 | * | |
2887 | * Also, during truncate, discard_buffer will have marked all | |
68189fef MWO |
2888 | * the folio's buffers clean. We discover that here and clean |
2889 | * the folio also. | |
87df7241 NP |
2890 | * |
2891 | * private_lock must be held over this entire operation in order | |
e621900a | 2892 | * to synchronise against block_dirty_folio and prevent the |
87df7241 | 2893 | * dirty bit from being lost. |
ecdfc978 | 2894 | */ |
11f81bec | 2895 | if (ret) |
68189fef | 2896 | folio_cancel_dirty(folio); |
87df7241 | 2897 | spin_unlock(&mapping->private_lock); |
1da177e4 LT |
2898 | out: |
2899 | if (buffers_to_free) { | |
2900 | struct buffer_head *bh = buffers_to_free; | |
2901 | ||
2902 | do { | |
2903 | struct buffer_head *next = bh->b_this_page; | |
2904 | free_buffer_head(bh); | |
2905 | bh = next; | |
2906 | } while (bh != buffers_to_free); | |
2907 | } | |
2908 | return ret; | |
2909 | } | |
2910 | EXPORT_SYMBOL(try_to_free_buffers); | |
2911 | ||
1da177e4 LT |
2912 | /* |
2913 | * Buffer-head allocation | |
2914 | */ | |
a0a9b043 | 2915 | static struct kmem_cache *bh_cachep __read_mostly; |
1da177e4 LT |
2916 | |
2917 | /* | |
2918 | * Once the number of bh's in the machine exceeds this level, we start | |
2919 | * stripping them in writeback. | |
2920 | */ | |
43be594a | 2921 | static unsigned long max_buffer_heads; |
1da177e4 LT |
2922 | |
2923 | int buffer_heads_over_limit; | |
2924 | ||
2925 | struct bh_accounting { | |
2926 | int nr; /* Number of live bh's */ | |
2927 | int ratelimit; /* Limit cacheline bouncing */ | |
2928 | }; | |
2929 | ||
2930 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; | |
2931 | ||
2932 | static void recalc_bh_state(void) | |
2933 | { | |
2934 | int i; | |
2935 | int tot = 0; | |
2936 | ||
ee1be862 | 2937 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
1da177e4 | 2938 | return; |
c7b92516 | 2939 | __this_cpu_write(bh_accounting.ratelimit, 0); |
8a143426 | 2940 | for_each_online_cpu(i) |
1da177e4 LT |
2941 | tot += per_cpu(bh_accounting, i).nr; |
2942 | buffer_heads_over_limit = (tot > max_buffer_heads); | |
2943 | } | |
c7b92516 | 2944 | |
dd0fc66f | 2945 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
1da177e4 | 2946 | { |
019b4d12 | 2947 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
1da177e4 | 2948 | if (ret) { |
a35afb83 | 2949 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
f1e67e35 | 2950 | spin_lock_init(&ret->b_uptodate_lock); |
c7b92516 CL |
2951 | preempt_disable(); |
2952 | __this_cpu_inc(bh_accounting.nr); | |
1da177e4 | 2953 | recalc_bh_state(); |
c7b92516 | 2954 | preempt_enable(); |
1da177e4 LT |
2955 | } |
2956 | return ret; | |
2957 | } | |
2958 | EXPORT_SYMBOL(alloc_buffer_head); | |
2959 | ||
2960 | void free_buffer_head(struct buffer_head *bh) | |
2961 | { | |
2962 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); | |
2963 | kmem_cache_free(bh_cachep, bh); | |
c7b92516 CL |
2964 | preempt_disable(); |
2965 | __this_cpu_dec(bh_accounting.nr); | |
1da177e4 | 2966 | recalc_bh_state(); |
c7b92516 | 2967 | preempt_enable(); |
1da177e4 LT |
2968 | } |
2969 | EXPORT_SYMBOL(free_buffer_head); | |
2970 | ||
fc4d24c9 | 2971 | static int buffer_exit_cpu_dead(unsigned int cpu) |
1da177e4 LT |
2972 | { |
2973 | int i; | |
2974 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); | |
2975 | ||
2976 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
2977 | brelse(b->bhs[i]); | |
2978 | b->bhs[i] = NULL; | |
2979 | } | |
c7b92516 | 2980 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
8a143426 | 2981 | per_cpu(bh_accounting, cpu).nr = 0; |
fc4d24c9 | 2982 | return 0; |
1da177e4 | 2983 | } |
1da177e4 | 2984 | |
389d1b08 | 2985 | /** |
a6b91919 | 2986 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
389d1b08 AK |
2987 | * @bh: struct buffer_head |
2988 | * | |
2989 | * Return true if the buffer is up-to-date and false, | |
2990 | * with the buffer locked, if not. | |
2991 | */ | |
2992 | int bh_uptodate_or_lock(struct buffer_head *bh) | |
2993 | { | |
2994 | if (!buffer_uptodate(bh)) { | |
2995 | lock_buffer(bh); | |
2996 | if (!buffer_uptodate(bh)) | |
2997 | return 0; | |
2998 | unlock_buffer(bh); | |
2999 | } | |
3000 | return 1; | |
3001 | } | |
3002 | EXPORT_SYMBOL(bh_uptodate_or_lock); | |
3003 | ||
3004 | /** | |
fdee117e | 3005 | * __bh_read - Submit read for a locked buffer |
389d1b08 | 3006 | * @bh: struct buffer_head |
fdee117e ZY |
3007 | * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ |
3008 | * @wait: wait until reading finish | |
389d1b08 | 3009 | * |
fdee117e | 3010 | * Returns zero on success or don't wait, and -EIO on error. |
389d1b08 | 3011 | */ |
fdee117e | 3012 | int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait) |
389d1b08 | 3013 | { |
fdee117e | 3014 | int ret = 0; |
389d1b08 | 3015 | |
fdee117e | 3016 | BUG_ON(!buffer_locked(bh)); |
389d1b08 AK |
3017 | |
3018 | get_bh(bh); | |
3019 | bh->b_end_io = end_buffer_read_sync; | |
fdee117e ZY |
3020 | submit_bh(REQ_OP_READ | op_flags, bh); |
3021 | if (wait) { | |
3022 | wait_on_buffer(bh); | |
3023 | if (!buffer_uptodate(bh)) | |
3024 | ret = -EIO; | |
3025 | } | |
3026 | return ret; | |
3027 | } | |
3028 | EXPORT_SYMBOL(__bh_read); | |
3029 | ||
3030 | /** | |
3031 | * __bh_read_batch - Submit read for a batch of unlocked buffers | |
3032 | * @nr: entry number of the buffer batch | |
3033 | * @bhs: a batch of struct buffer_head | |
3034 | * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ | |
3035 | * @force_lock: force to get a lock on the buffer if set, otherwise drops any | |
3036 | * buffer that cannot lock. | |
3037 | * | |
3038 | * Returns zero on success or don't wait, and -EIO on error. | |
3039 | */ | |
3040 | void __bh_read_batch(int nr, struct buffer_head *bhs[], | |
3041 | blk_opf_t op_flags, bool force_lock) | |
3042 | { | |
3043 | int i; | |
3044 | ||
3045 | for (i = 0; i < nr; i++) { | |
3046 | struct buffer_head *bh = bhs[i]; | |
3047 | ||
3048 | if (buffer_uptodate(bh)) | |
3049 | continue; | |
3050 | ||
3051 | if (force_lock) | |
3052 | lock_buffer(bh); | |
3053 | else | |
3054 | if (!trylock_buffer(bh)) | |
3055 | continue; | |
3056 | ||
3057 | if (buffer_uptodate(bh)) { | |
3058 | unlock_buffer(bh); | |
3059 | continue; | |
3060 | } | |
3061 | ||
3062 | bh->b_end_io = end_buffer_read_sync; | |
3063 | get_bh(bh); | |
3064 | submit_bh(REQ_OP_READ | op_flags, bh); | |
3065 | } | |
389d1b08 | 3066 | } |
fdee117e | 3067 | EXPORT_SYMBOL(__bh_read_batch); |
389d1b08 | 3068 | |
1da177e4 LT |
3069 | void __init buffer_init(void) |
3070 | { | |
43be594a | 3071 | unsigned long nrpages; |
fc4d24c9 | 3072 | int ret; |
1da177e4 | 3073 | |
b98938c3 CL |
3074 | bh_cachep = kmem_cache_create("buffer_head", |
3075 | sizeof(struct buffer_head), 0, | |
3076 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
3077 | SLAB_MEM_SPREAD), | |
019b4d12 | 3078 | NULL); |
1da177e4 LT |
3079 | |
3080 | /* | |
3081 | * Limit the bh occupancy to 10% of ZONE_NORMAL | |
3082 | */ | |
3083 | nrpages = (nr_free_buffer_pages() * 10) / 100; | |
3084 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); | |
fc4d24c9 SAS |
3085 | ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", |
3086 | NULL, buffer_exit_cpu_dead); | |
3087 | WARN_ON(ret < 0); | |
1da177e4 | 3088 | } |