]>
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> |
1da177e4 | 51 | |
2b211dc0 BD |
52 | #include "internal.h" |
53 | ||
1da177e4 | 54 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); |
2a222ca9 | 55 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 56 | enum rw_hint hint, struct writeback_control *wbc); |
1da177e4 LT |
57 | |
58 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) | |
59 | ||
f0059afd TH |
60 | inline void touch_buffer(struct buffer_head *bh) |
61 | { | |
5305cb83 | 62 | trace_block_touch_buffer(bh); |
f0059afd TH |
63 | mark_page_accessed(bh->b_page); |
64 | } | |
65 | EXPORT_SYMBOL(touch_buffer); | |
66 | ||
fc9b52cd | 67 | void __lock_buffer(struct buffer_head *bh) |
1da177e4 | 68 | { |
74316201 | 69 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 LT |
70 | } |
71 | EXPORT_SYMBOL(__lock_buffer); | |
72 | ||
fc9b52cd | 73 | void unlock_buffer(struct buffer_head *bh) |
1da177e4 | 74 | { |
51b07fc3 | 75 | clear_bit_unlock(BH_Lock, &bh->b_state); |
4e857c58 | 76 | smp_mb__after_atomic(); |
1da177e4 LT |
77 | wake_up_bit(&bh->b_state, BH_Lock); |
78 | } | |
1fe72eaa | 79 | EXPORT_SYMBOL(unlock_buffer); |
1da177e4 | 80 | |
b4597226 MG |
81 | /* |
82 | * Returns if the page has dirty or writeback buffers. If all the buffers | |
83 | * are unlocked and clean then the PageDirty information is stale. If | |
84 | * any of the pages are locked, it is assumed they are locked for IO. | |
85 | */ | |
86 | void buffer_check_dirty_writeback(struct page *page, | |
87 | bool *dirty, bool *writeback) | |
88 | { | |
89 | struct buffer_head *head, *bh; | |
90 | *dirty = false; | |
91 | *writeback = false; | |
92 | ||
93 | BUG_ON(!PageLocked(page)); | |
94 | ||
95 | if (!page_has_buffers(page)) | |
96 | return; | |
97 | ||
98 | if (PageWriteback(page)) | |
99 | *writeback = true; | |
100 | ||
101 | head = page_buffers(page); | |
102 | bh = head; | |
103 | do { | |
104 | if (buffer_locked(bh)) | |
105 | *writeback = true; | |
106 | ||
107 | if (buffer_dirty(bh)) | |
108 | *dirty = true; | |
109 | ||
110 | bh = bh->b_this_page; | |
111 | } while (bh != head); | |
112 | } | |
113 | EXPORT_SYMBOL(buffer_check_dirty_writeback); | |
114 | ||
1da177e4 LT |
115 | /* |
116 | * Block until a buffer comes unlocked. This doesn't stop it | |
117 | * from becoming locked again - you have to lock it yourself | |
118 | * if you want to preserve its state. | |
119 | */ | |
120 | void __wait_on_buffer(struct buffer_head * bh) | |
121 | { | |
74316201 | 122 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 | 123 | } |
1fe72eaa | 124 | EXPORT_SYMBOL(__wait_on_buffer); |
1da177e4 LT |
125 | |
126 | static void | |
127 | __clear_page_buffers(struct page *page) | |
128 | { | |
129 | ClearPagePrivate(page); | |
4c21e2f2 | 130 | set_page_private(page, 0); |
09cbfeaf | 131 | put_page(page); |
1da177e4 LT |
132 | } |
133 | ||
b744c2ac | 134 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
1da177e4 | 135 | { |
432f16e6 RE |
136 | if (!test_bit(BH_Quiet, &bh->b_state)) |
137 | printk_ratelimited(KERN_ERR | |
a1c6f057 DM |
138 | "Buffer I/O error on dev %pg, logical block %llu%s\n", |
139 | bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); | |
1da177e4 LT |
140 | } |
141 | ||
142 | /* | |
68671f35 DM |
143 | * End-of-IO handler helper function which does not touch the bh after |
144 | * unlocking it. | |
145 | * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but | |
146 | * a race there is benign: unlock_buffer() only use the bh's address for | |
147 | * hashing after unlocking the buffer, so it doesn't actually touch the bh | |
148 | * itself. | |
1da177e4 | 149 | */ |
68671f35 | 150 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
1da177e4 LT |
151 | { |
152 | if (uptodate) { | |
153 | set_buffer_uptodate(bh); | |
154 | } else { | |
70246286 | 155 | /* This happens, due to failed read-ahead attempts. */ |
1da177e4 LT |
156 | clear_buffer_uptodate(bh); |
157 | } | |
158 | unlock_buffer(bh); | |
68671f35 DM |
159 | } |
160 | ||
161 | /* | |
162 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and | |
163 | * unlock the buffer. This is what ll_rw_block uses too. | |
164 | */ | |
165 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) | |
166 | { | |
167 | __end_buffer_read_notouch(bh, uptodate); | |
1da177e4 LT |
168 | put_bh(bh); |
169 | } | |
1fe72eaa | 170 | EXPORT_SYMBOL(end_buffer_read_sync); |
1da177e4 LT |
171 | |
172 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) | |
173 | { | |
1da177e4 LT |
174 | if (uptodate) { |
175 | set_buffer_uptodate(bh); | |
176 | } else { | |
432f16e6 | 177 | buffer_io_error(bh, ", lost sync page write"); |
87354e5d | 178 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
179 | clear_buffer_uptodate(bh); |
180 | } | |
181 | unlock_buffer(bh); | |
182 | put_bh(bh); | |
183 | } | |
1fe72eaa | 184 | EXPORT_SYMBOL(end_buffer_write_sync); |
1da177e4 | 185 | |
1da177e4 LT |
186 | /* |
187 | * Various filesystems appear to want __find_get_block to be non-blocking. | |
188 | * But it's the page lock which protects the buffers. To get around this, | |
189 | * we get exclusion from try_to_free_buffers with the blockdev mapping's | |
190 | * private_lock. | |
191 | * | |
b93b0163 | 192 | * Hack idea: for the blockdev mapping, private_lock contention |
1da177e4 | 193 | * may be quite high. This code could TryLock the page, and if that |
b93b0163 | 194 | * succeeds, there is no need to take private_lock. |
1da177e4 LT |
195 | */ |
196 | static struct buffer_head * | |
385fd4c5 | 197 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
1da177e4 LT |
198 | { |
199 | struct inode *bd_inode = bdev->bd_inode; | |
200 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
201 | struct buffer_head *ret = NULL; | |
202 | pgoff_t index; | |
203 | struct buffer_head *bh; | |
204 | struct buffer_head *head; | |
205 | struct page *page; | |
206 | int all_mapped = 1; | |
43636c80 | 207 | static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); |
1da177e4 | 208 | |
09cbfeaf | 209 | index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); |
2457aec6 | 210 | page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); |
1da177e4 LT |
211 | if (!page) |
212 | goto out; | |
213 | ||
214 | spin_lock(&bd_mapping->private_lock); | |
215 | if (!page_has_buffers(page)) | |
216 | goto out_unlock; | |
217 | head = page_buffers(page); | |
218 | bh = head; | |
219 | do { | |
97f76d3d NK |
220 | if (!buffer_mapped(bh)) |
221 | all_mapped = 0; | |
222 | else if (bh->b_blocknr == block) { | |
1da177e4 LT |
223 | ret = bh; |
224 | get_bh(bh); | |
225 | goto out_unlock; | |
226 | } | |
1da177e4 LT |
227 | bh = bh->b_this_page; |
228 | } while (bh != head); | |
229 | ||
230 | /* we might be here because some of the buffers on this page are | |
231 | * not mapped. This is due to various races between | |
232 | * file io on the block device and getblk. It gets dealt with | |
233 | * elsewhere, don't buffer_error if we had some unmapped buffers | |
234 | */ | |
43636c80 TH |
235 | ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); |
236 | if (all_mapped && __ratelimit(&last_warned)) { | |
237 | printk("__find_get_block_slow() failed. block=%llu, " | |
238 | "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " | |
239 | "device %pg blocksize: %d\n", | |
240 | (unsigned long long)block, | |
241 | (unsigned long long)bh->b_blocknr, | |
242 | bh->b_state, bh->b_size, bdev, | |
243 | 1 << bd_inode->i_blkbits); | |
1da177e4 LT |
244 | } |
245 | out_unlock: | |
246 | spin_unlock(&bd_mapping->private_lock); | |
09cbfeaf | 247 | put_page(page); |
1da177e4 LT |
248 | out: |
249 | return ret; | |
250 | } | |
251 | ||
1da177e4 LT |
252 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) |
253 | { | |
1da177e4 | 254 | unsigned long flags; |
a3972203 | 255 | struct buffer_head *first; |
1da177e4 LT |
256 | struct buffer_head *tmp; |
257 | struct page *page; | |
258 | int page_uptodate = 1; | |
259 | ||
260 | BUG_ON(!buffer_async_read(bh)); | |
261 | ||
262 | page = bh->b_page; | |
263 | if (uptodate) { | |
264 | set_buffer_uptodate(bh); | |
265 | } else { | |
266 | clear_buffer_uptodate(bh); | |
432f16e6 | 267 | buffer_io_error(bh, ", async page read"); |
1da177e4 LT |
268 | SetPageError(page); |
269 | } | |
270 | ||
271 | /* | |
272 | * Be _very_ careful from here on. Bad things can happen if | |
273 | * two buffer heads end IO at almost the same time and both | |
274 | * decide that the page is now completely done. | |
275 | */ | |
a3972203 | 276 | first = page_buffers(page); |
f1e67e35 | 277 | spin_lock_irqsave(&first->b_uptodate_lock, flags); |
1da177e4 LT |
278 | clear_buffer_async_read(bh); |
279 | unlock_buffer(bh); | |
280 | tmp = bh; | |
281 | do { | |
282 | if (!buffer_uptodate(tmp)) | |
283 | page_uptodate = 0; | |
284 | if (buffer_async_read(tmp)) { | |
285 | BUG_ON(!buffer_locked(tmp)); | |
286 | goto still_busy; | |
287 | } | |
288 | tmp = tmp->b_this_page; | |
289 | } while (tmp != bh); | |
f1e67e35 | 290 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
291 | |
292 | /* | |
293 | * If none of the buffers had errors and they are all | |
294 | * uptodate then we can set the page uptodate. | |
295 | */ | |
296 | if (page_uptodate && !PageError(page)) | |
297 | SetPageUptodate(page); | |
298 | unlock_page(page); | |
299 | return; | |
300 | ||
301 | still_busy: | |
f1e67e35 | 302 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
303 | return; |
304 | } | |
305 | ||
31fb992c EB |
306 | struct decrypt_bh_ctx { |
307 | struct work_struct work; | |
308 | struct buffer_head *bh; | |
309 | }; | |
310 | ||
311 | static void decrypt_bh(struct work_struct *work) | |
312 | { | |
313 | struct decrypt_bh_ctx *ctx = | |
314 | container_of(work, struct decrypt_bh_ctx, work); | |
315 | struct buffer_head *bh = ctx->bh; | |
316 | int err; | |
317 | ||
318 | err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size, | |
319 | bh_offset(bh)); | |
320 | end_buffer_async_read(bh, err == 0); | |
321 | kfree(ctx); | |
322 | } | |
323 | ||
324 | /* | |
325 | * I/O completion handler for block_read_full_page() - pages | |
326 | * which come unlocked at the end of I/O. | |
327 | */ | |
328 | static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate) | |
329 | { | |
330 | /* Decrypt if needed */ | |
331 | if (uptodate && IS_ENABLED(CONFIG_FS_ENCRYPTION) && | |
332 | IS_ENCRYPTED(bh->b_page->mapping->host) && | |
333 | S_ISREG(bh->b_page->mapping->host->i_mode)) { | |
334 | struct decrypt_bh_ctx *ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC); | |
335 | ||
336 | if (ctx) { | |
337 | INIT_WORK(&ctx->work, decrypt_bh); | |
338 | ctx->bh = bh; | |
339 | fscrypt_enqueue_decrypt_work(&ctx->work); | |
340 | return; | |
341 | } | |
342 | uptodate = 0; | |
343 | } | |
344 | end_buffer_async_read(bh, uptodate); | |
345 | } | |
346 | ||
1da177e4 LT |
347 | /* |
348 | * Completion handler for block_write_full_page() - pages which are unlocked | |
349 | * during I/O, and which have PageWriteback cleared upon I/O completion. | |
350 | */ | |
35c80d5f | 351 | void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
1da177e4 | 352 | { |
1da177e4 | 353 | unsigned long flags; |
a3972203 | 354 | struct buffer_head *first; |
1da177e4 LT |
355 | struct buffer_head *tmp; |
356 | struct page *page; | |
357 | ||
358 | BUG_ON(!buffer_async_write(bh)); | |
359 | ||
360 | page = bh->b_page; | |
361 | if (uptodate) { | |
362 | set_buffer_uptodate(bh); | |
363 | } else { | |
432f16e6 | 364 | buffer_io_error(bh, ", lost async page write"); |
87354e5d | 365 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
366 | clear_buffer_uptodate(bh); |
367 | SetPageError(page); | |
368 | } | |
369 | ||
a3972203 | 370 | first = page_buffers(page); |
f1e67e35 | 371 | spin_lock_irqsave(&first->b_uptodate_lock, flags); |
a3972203 | 372 | |
1da177e4 LT |
373 | clear_buffer_async_write(bh); |
374 | unlock_buffer(bh); | |
375 | tmp = bh->b_this_page; | |
376 | while (tmp != bh) { | |
377 | if (buffer_async_write(tmp)) { | |
378 | BUG_ON(!buffer_locked(tmp)); | |
379 | goto still_busy; | |
380 | } | |
381 | tmp = tmp->b_this_page; | |
382 | } | |
f1e67e35 | 383 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
384 | end_page_writeback(page); |
385 | return; | |
386 | ||
387 | still_busy: | |
f1e67e35 | 388 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
1da177e4 LT |
389 | return; |
390 | } | |
1fe72eaa | 391 | EXPORT_SYMBOL(end_buffer_async_write); |
1da177e4 LT |
392 | |
393 | /* | |
394 | * If a page's buffers are under async readin (end_buffer_async_read | |
395 | * completion) then there is a possibility that another thread of | |
396 | * control could lock one of the buffers after it has completed | |
397 | * but while some of the other buffers have not completed. This | |
398 | * locked buffer would confuse end_buffer_async_read() into not unlocking | |
399 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() | |
400 | * that this buffer is not under async I/O. | |
401 | * | |
402 | * The page comes unlocked when it has no locked buffer_async buffers | |
403 | * left. | |
404 | * | |
405 | * PageLocked prevents anyone starting new async I/O reads any of | |
406 | * the buffers. | |
407 | * | |
408 | * PageWriteback is used to prevent simultaneous writeout of the same | |
409 | * page. | |
410 | * | |
411 | * PageLocked prevents anyone from starting writeback of a page which is | |
412 | * under read I/O (PageWriteback is only ever set against a locked page). | |
413 | */ | |
414 | static void mark_buffer_async_read(struct buffer_head *bh) | |
415 | { | |
31fb992c | 416 | bh->b_end_io = end_buffer_async_read_io; |
1da177e4 LT |
417 | set_buffer_async_read(bh); |
418 | } | |
419 | ||
1fe72eaa HS |
420 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
421 | bh_end_io_t *handler) | |
1da177e4 | 422 | { |
35c80d5f | 423 | bh->b_end_io = handler; |
1da177e4 LT |
424 | set_buffer_async_write(bh); |
425 | } | |
35c80d5f CM |
426 | |
427 | void mark_buffer_async_write(struct buffer_head *bh) | |
428 | { | |
429 | mark_buffer_async_write_endio(bh, end_buffer_async_write); | |
430 | } | |
1da177e4 LT |
431 | EXPORT_SYMBOL(mark_buffer_async_write); |
432 | ||
433 | ||
434 | /* | |
435 | * fs/buffer.c contains helper functions for buffer-backed address space's | |
436 | * fsync functions. A common requirement for buffer-based filesystems is | |
437 | * that certain data from the backing blockdev needs to be written out for | |
438 | * a successful fsync(). For example, ext2 indirect blocks need to be | |
439 | * written back and waited upon before fsync() returns. | |
440 | * | |
441 | * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), | |
442 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the | |
443 | * management of a list of dependent buffers at ->i_mapping->private_list. | |
444 | * | |
445 | * Locking is a little subtle: try_to_free_buffers() will remove buffers | |
446 | * from their controlling inode's queue when they are being freed. But | |
447 | * try_to_free_buffers() will be operating against the *blockdev* mapping | |
448 | * at the time, not against the S_ISREG file which depends on those buffers. | |
449 | * So the locking for private_list is via the private_lock in the address_space | |
450 | * which backs the buffers. Which is different from the address_space | |
451 | * against which the buffers are listed. So for a particular address_space, | |
452 | * mapping->private_lock does *not* protect mapping->private_list! In fact, | |
453 | * mapping->private_list will always be protected by the backing blockdev's | |
454 | * ->private_lock. | |
455 | * | |
456 | * Which introduces a requirement: all buffers on an address_space's | |
457 | * ->private_list must be from the same address_space: the blockdev's. | |
458 | * | |
459 | * address_spaces which do not place buffers at ->private_list via these | |
460 | * utility functions are free to use private_lock and private_list for | |
461 | * whatever they want. The only requirement is that list_empty(private_list) | |
462 | * be true at clear_inode() time. | |
463 | * | |
464 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The | |
465 | * filesystems should do that. invalidate_inode_buffers() should just go | |
466 | * BUG_ON(!list_empty). | |
467 | * | |
468 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should | |
469 | * take an address_space, not an inode. And it should be called | |
470 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being | |
471 | * queued up. | |
472 | * | |
473 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the | |
474 | * list if it is already on a list. Because if the buffer is on a list, | |
475 | * it *must* already be on the right one. If not, the filesystem is being | |
476 | * silly. This will save a ton of locking. But first we have to ensure | |
477 | * that buffers are taken *off* the old inode's list when they are freed | |
478 | * (presumably in truncate). That requires careful auditing of all | |
479 | * filesystems (do it inside bforget()). It could also be done by bringing | |
480 | * b_inode back. | |
481 | */ | |
482 | ||
483 | /* | |
484 | * The buffer's backing address_space's private_lock must be held | |
485 | */ | |
dbacefc9 | 486 | static void __remove_assoc_queue(struct buffer_head *bh) |
1da177e4 LT |
487 | { |
488 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 489 | WARN_ON(!bh->b_assoc_map); |
58ff407b | 490 | bh->b_assoc_map = NULL; |
1da177e4 LT |
491 | } |
492 | ||
493 | int inode_has_buffers(struct inode *inode) | |
494 | { | |
495 | return !list_empty(&inode->i_data.private_list); | |
496 | } | |
497 | ||
498 | /* | |
499 | * osync is designed to support O_SYNC io. It waits synchronously for | |
500 | * all already-submitted IO to complete, but does not queue any new | |
501 | * writes to the disk. | |
502 | * | |
503 | * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as | |
504 | * you dirty the buffers, and then use osync_inode_buffers to wait for | |
505 | * completion. Any other dirty buffers which are not yet queued for | |
506 | * write will not be flushed to disk by the osync. | |
507 | */ | |
508 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) | |
509 | { | |
510 | struct buffer_head *bh; | |
511 | struct list_head *p; | |
512 | int err = 0; | |
513 | ||
514 | spin_lock(lock); | |
515 | repeat: | |
516 | list_for_each_prev(p, list) { | |
517 | bh = BH_ENTRY(p); | |
518 | if (buffer_locked(bh)) { | |
519 | get_bh(bh); | |
520 | spin_unlock(lock); | |
521 | wait_on_buffer(bh); | |
522 | if (!buffer_uptodate(bh)) | |
523 | err = -EIO; | |
524 | brelse(bh); | |
525 | spin_lock(lock); | |
526 | goto repeat; | |
527 | } | |
528 | } | |
529 | spin_unlock(lock); | |
530 | return err; | |
531 | } | |
532 | ||
08fdc8a0 | 533 | void emergency_thaw_bdev(struct super_block *sb) |
c2d75438 | 534 | { |
01a05b33 | 535 | while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) |
a1c6f057 | 536 | printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); |
01a05b33 | 537 | } |
c2d75438 | 538 | |
1da177e4 | 539 | /** |
78a4a50a | 540 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
67be2dd1 | 541 | * @mapping: the mapping which wants those buffers written |
1da177e4 LT |
542 | * |
543 | * Starts I/O against the buffers at mapping->private_list, and waits upon | |
544 | * that I/O. | |
545 | * | |
67be2dd1 MW |
546 | * Basically, this is a convenience function for fsync(). |
547 | * @mapping is a file or directory which needs those buffers to be written for | |
548 | * a successful fsync(). | |
1da177e4 LT |
549 | */ |
550 | int sync_mapping_buffers(struct address_space *mapping) | |
551 | { | |
252aa6f5 | 552 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
553 | |
554 | if (buffer_mapping == NULL || list_empty(&mapping->private_list)) | |
555 | return 0; | |
556 | ||
557 | return fsync_buffers_list(&buffer_mapping->private_lock, | |
558 | &mapping->private_list); | |
559 | } | |
560 | EXPORT_SYMBOL(sync_mapping_buffers); | |
561 | ||
562 | /* | |
563 | * Called when we've recently written block `bblock', and it is known that | |
564 | * `bblock' was for a buffer_boundary() buffer. This means that the block at | |
565 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's | |
566 | * dirty, schedule it for IO. So that indirects merge nicely with their data. | |
567 | */ | |
568 | void write_boundary_block(struct block_device *bdev, | |
569 | sector_t bblock, unsigned blocksize) | |
570 | { | |
571 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); | |
572 | if (bh) { | |
573 | if (buffer_dirty(bh)) | |
dfec8a14 | 574 | ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); |
1da177e4 LT |
575 | put_bh(bh); |
576 | } | |
577 | } | |
578 | ||
579 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) | |
580 | { | |
581 | struct address_space *mapping = inode->i_mapping; | |
582 | struct address_space *buffer_mapping = bh->b_page->mapping; | |
583 | ||
584 | mark_buffer_dirty(bh); | |
252aa6f5 RA |
585 | if (!mapping->private_data) { |
586 | mapping->private_data = buffer_mapping; | |
1da177e4 | 587 | } else { |
252aa6f5 | 588 | BUG_ON(mapping->private_data != buffer_mapping); |
1da177e4 | 589 | } |
535ee2fb | 590 | if (!bh->b_assoc_map) { |
1da177e4 LT |
591 | spin_lock(&buffer_mapping->private_lock); |
592 | list_move_tail(&bh->b_assoc_buffers, | |
593 | &mapping->private_list); | |
58ff407b | 594 | bh->b_assoc_map = mapping; |
1da177e4 LT |
595 | spin_unlock(&buffer_mapping->private_lock); |
596 | } | |
597 | } | |
598 | EXPORT_SYMBOL(mark_buffer_dirty_inode); | |
599 | ||
787d2214 | 600 | /* |
ec82e1c1 | 601 | * Mark the page dirty, and set it dirty in the page cache, and mark the inode |
787d2214 NP |
602 | * dirty. |
603 | * | |
604 | * If warn is true, then emit a warning if the page is not uptodate and has | |
605 | * not been truncated. | |
c4843a75 | 606 | * |
81f8c3a4 | 607 | * The caller must hold lock_page_memcg(). |
787d2214 | 608 | */ |
f82b3764 | 609 | void __set_page_dirty(struct page *page, struct address_space *mapping, |
62cccb8c | 610 | int warn) |
787d2214 | 611 | { |
227d53b3 KM |
612 | unsigned long flags; |
613 | ||
b93b0163 | 614 | xa_lock_irqsave(&mapping->i_pages, flags); |
787d2214 NP |
615 | if (page->mapping) { /* Race with truncate? */ |
616 | WARN_ON_ONCE(warn && !PageUptodate(page)); | |
62cccb8c | 617 | account_page_dirtied(page, mapping); |
ec82e1c1 MW |
618 | __xa_set_mark(&mapping->i_pages, page_index(page), |
619 | PAGECACHE_TAG_DIRTY); | |
787d2214 | 620 | } |
b93b0163 | 621 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
787d2214 | 622 | } |
f82b3764 | 623 | EXPORT_SYMBOL_GPL(__set_page_dirty); |
787d2214 | 624 | |
1da177e4 LT |
625 | /* |
626 | * Add a page to the dirty page list. | |
627 | * | |
628 | * It is a sad fact of life that this function is called from several places | |
629 | * deeply under spinlocking. It may not sleep. | |
630 | * | |
631 | * If the page has buffers, the uptodate buffers are set dirty, to preserve | |
632 | * dirty-state coherency between the page and the buffers. It the page does | |
633 | * not have buffers then when they are later attached they will all be set | |
634 | * dirty. | |
635 | * | |
636 | * The buffers are dirtied before the page is dirtied. There's a small race | |
637 | * window in which a writepage caller may see the page cleanness but not the | |
638 | * buffer dirtiness. That's fine. If this code were to set the page dirty | |
639 | * before the buffers, a concurrent writepage caller could clear the page dirty | |
640 | * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean | |
641 | * page on the dirty page list. | |
642 | * | |
643 | * We use private_lock to lock against try_to_free_buffers while using the | |
644 | * page's buffer list. Also use this to protect against clean buffers being | |
645 | * added to the page after it was set dirty. | |
646 | * | |
647 | * FIXME: may need to call ->reservepage here as well. That's rather up to the | |
648 | * address_space though. | |
649 | */ | |
650 | int __set_page_dirty_buffers(struct page *page) | |
651 | { | |
a8e7d49a | 652 | int newly_dirty; |
787d2214 | 653 | struct address_space *mapping = page_mapping(page); |
ebf7a227 NP |
654 | |
655 | if (unlikely(!mapping)) | |
656 | return !TestSetPageDirty(page); | |
1da177e4 LT |
657 | |
658 | spin_lock(&mapping->private_lock); | |
659 | if (page_has_buffers(page)) { | |
660 | struct buffer_head *head = page_buffers(page); | |
661 | struct buffer_head *bh = head; | |
662 | ||
663 | do { | |
664 | set_buffer_dirty(bh); | |
665 | bh = bh->b_this_page; | |
666 | } while (bh != head); | |
667 | } | |
c4843a75 | 668 | /* |
81f8c3a4 JW |
669 | * Lock out page->mem_cgroup migration to keep PageDirty |
670 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 671 | */ |
62cccb8c | 672 | lock_page_memcg(page); |
a8e7d49a | 673 | newly_dirty = !TestSetPageDirty(page); |
1da177e4 LT |
674 | spin_unlock(&mapping->private_lock); |
675 | ||
a8e7d49a | 676 | if (newly_dirty) |
62cccb8c | 677 | __set_page_dirty(page, mapping, 1); |
c4843a75 | 678 | |
62cccb8c | 679 | unlock_page_memcg(page); |
c4843a75 GT |
680 | |
681 | if (newly_dirty) | |
682 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
683 | ||
a8e7d49a | 684 | return newly_dirty; |
1da177e4 LT |
685 | } |
686 | EXPORT_SYMBOL(__set_page_dirty_buffers); | |
687 | ||
688 | /* | |
689 | * Write out and wait upon a list of buffers. | |
690 | * | |
691 | * We have conflicting pressures: we want to make sure that all | |
692 | * initially dirty buffers get waited on, but that any subsequently | |
693 | * dirtied buffers don't. After all, we don't want fsync to last | |
694 | * forever if somebody is actively writing to the file. | |
695 | * | |
696 | * Do this in two main stages: first we copy dirty buffers to a | |
697 | * temporary inode list, queueing the writes as we go. Then we clean | |
698 | * up, waiting for those writes to complete. | |
699 | * | |
700 | * During this second stage, any subsequent updates to the file may end | |
701 | * up refiling the buffer on the original inode's dirty list again, so | |
702 | * there is a chance we will end up with a buffer queued for write but | |
703 | * not yet completed on that list. So, as a final cleanup we go through | |
704 | * the osync code to catch these locked, dirty buffers without requeuing | |
705 | * any newly dirty buffers for write. | |
706 | */ | |
707 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) | |
708 | { | |
709 | struct buffer_head *bh; | |
710 | struct list_head tmp; | |
7eaceacc | 711 | struct address_space *mapping; |
1da177e4 | 712 | int err = 0, err2; |
4ee2491e | 713 | struct blk_plug plug; |
1da177e4 LT |
714 | |
715 | INIT_LIST_HEAD(&tmp); | |
4ee2491e | 716 | blk_start_plug(&plug); |
1da177e4 LT |
717 | |
718 | spin_lock(lock); | |
719 | while (!list_empty(list)) { | |
720 | bh = BH_ENTRY(list->next); | |
535ee2fb | 721 | mapping = bh->b_assoc_map; |
58ff407b | 722 | __remove_assoc_queue(bh); |
535ee2fb JK |
723 | /* Avoid race with mark_buffer_dirty_inode() which does |
724 | * a lockless check and we rely on seeing the dirty bit */ | |
725 | smp_mb(); | |
1da177e4 LT |
726 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
727 | list_add(&bh->b_assoc_buffers, &tmp); | |
535ee2fb | 728 | bh->b_assoc_map = mapping; |
1da177e4 LT |
729 | if (buffer_dirty(bh)) { |
730 | get_bh(bh); | |
731 | spin_unlock(lock); | |
732 | /* | |
733 | * Ensure any pending I/O completes so that | |
9cb569d6 CH |
734 | * write_dirty_buffer() actually writes the |
735 | * current contents - it is a noop if I/O is | |
736 | * still in flight on potentially older | |
737 | * contents. | |
1da177e4 | 738 | */ |
70fd7614 | 739 | write_dirty_buffer(bh, REQ_SYNC); |
9cf6b720 JA |
740 | |
741 | /* | |
742 | * Kick off IO for the previous mapping. Note | |
743 | * that we will not run the very last mapping, | |
744 | * wait_on_buffer() will do that for us | |
745 | * through sync_buffer(). | |
746 | */ | |
1da177e4 LT |
747 | brelse(bh); |
748 | spin_lock(lock); | |
749 | } | |
750 | } | |
751 | } | |
752 | ||
4ee2491e JA |
753 | spin_unlock(lock); |
754 | blk_finish_plug(&plug); | |
755 | spin_lock(lock); | |
756 | ||
1da177e4 LT |
757 | while (!list_empty(&tmp)) { |
758 | bh = BH_ENTRY(tmp.prev); | |
1da177e4 | 759 | get_bh(bh); |
535ee2fb JK |
760 | mapping = bh->b_assoc_map; |
761 | __remove_assoc_queue(bh); | |
762 | /* Avoid race with mark_buffer_dirty_inode() which does | |
763 | * a lockless check and we rely on seeing the dirty bit */ | |
764 | smp_mb(); | |
765 | if (buffer_dirty(bh)) { | |
766 | list_add(&bh->b_assoc_buffers, | |
e3892296 | 767 | &mapping->private_list); |
535ee2fb JK |
768 | bh->b_assoc_map = mapping; |
769 | } | |
1da177e4 LT |
770 | spin_unlock(lock); |
771 | wait_on_buffer(bh); | |
772 | if (!buffer_uptodate(bh)) | |
773 | err = -EIO; | |
774 | brelse(bh); | |
775 | spin_lock(lock); | |
776 | } | |
777 | ||
778 | spin_unlock(lock); | |
779 | err2 = osync_buffers_list(lock, list); | |
780 | if (err) | |
781 | return err; | |
782 | else | |
783 | return err2; | |
784 | } | |
785 | ||
786 | /* | |
787 | * Invalidate any and all dirty buffers on a given inode. We are | |
788 | * probably unmounting the fs, but that doesn't mean we have already | |
789 | * done a sync(). Just drop the buffers from the inode list. | |
790 | * | |
791 | * NOTE: we take the inode's blockdev's mapping's private_lock. Which | |
792 | * assumes that all the buffers are against the blockdev. Not true | |
793 | * for reiserfs. | |
794 | */ | |
795 | void invalidate_inode_buffers(struct inode *inode) | |
796 | { | |
797 | if (inode_has_buffers(inode)) { | |
798 | struct address_space *mapping = &inode->i_data; | |
799 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 800 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
801 | |
802 | spin_lock(&buffer_mapping->private_lock); | |
803 | while (!list_empty(list)) | |
804 | __remove_assoc_queue(BH_ENTRY(list->next)); | |
805 | spin_unlock(&buffer_mapping->private_lock); | |
806 | } | |
807 | } | |
52b19ac9 | 808 | EXPORT_SYMBOL(invalidate_inode_buffers); |
1da177e4 LT |
809 | |
810 | /* | |
811 | * Remove any clean buffers from the inode's buffer list. This is called | |
812 | * when we're trying to free the inode itself. Those buffers can pin it. | |
813 | * | |
814 | * Returns true if all buffers were removed. | |
815 | */ | |
816 | int remove_inode_buffers(struct inode *inode) | |
817 | { | |
818 | int ret = 1; | |
819 | ||
820 | if (inode_has_buffers(inode)) { | |
821 | struct address_space *mapping = &inode->i_data; | |
822 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 823 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
824 | |
825 | spin_lock(&buffer_mapping->private_lock); | |
826 | while (!list_empty(list)) { | |
827 | struct buffer_head *bh = BH_ENTRY(list->next); | |
828 | if (buffer_dirty(bh)) { | |
829 | ret = 0; | |
830 | break; | |
831 | } | |
832 | __remove_assoc_queue(bh); | |
833 | } | |
834 | spin_unlock(&buffer_mapping->private_lock); | |
835 | } | |
836 | return ret; | |
837 | } | |
838 | ||
839 | /* | |
840 | * Create the appropriate buffers when given a page for data area and | |
841 | * the size of each buffer.. Use the bh->b_this_page linked list to | |
842 | * follow the buffers created. Return NULL if unable to create more | |
843 | * buffers. | |
844 | * | |
845 | * The retry flag is used to differentiate async IO (paging, swapping) | |
846 | * which may not fail from ordinary buffer allocations. | |
847 | */ | |
848 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, | |
640ab98f | 849 | bool retry) |
1da177e4 LT |
850 | { |
851 | struct buffer_head *bh, *head; | |
f745c6f5 | 852 | gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; |
1da177e4 | 853 | long offset; |
f745c6f5 | 854 | struct mem_cgroup *memcg; |
1da177e4 | 855 | |
640ab98f JA |
856 | if (retry) |
857 | gfp |= __GFP_NOFAIL; | |
858 | ||
f745c6f5 SB |
859 | memcg = get_mem_cgroup_from_page(page); |
860 | memalloc_use_memcg(memcg); | |
861 | ||
1da177e4 LT |
862 | head = NULL; |
863 | offset = PAGE_SIZE; | |
864 | while ((offset -= size) >= 0) { | |
640ab98f | 865 | bh = alloc_buffer_head(gfp); |
1da177e4 LT |
866 | if (!bh) |
867 | goto no_grow; | |
868 | ||
1da177e4 LT |
869 | bh->b_this_page = head; |
870 | bh->b_blocknr = -1; | |
871 | head = bh; | |
872 | ||
1da177e4 LT |
873 | bh->b_size = size; |
874 | ||
875 | /* Link the buffer to its page */ | |
876 | set_bh_page(bh, page, offset); | |
1da177e4 | 877 | } |
f745c6f5 SB |
878 | out: |
879 | memalloc_unuse_memcg(); | |
880 | mem_cgroup_put(memcg); | |
1da177e4 LT |
881 | return head; |
882 | /* | |
883 | * In case anything failed, we just free everything we got. | |
884 | */ | |
885 | no_grow: | |
886 | if (head) { | |
887 | do { | |
888 | bh = head; | |
889 | head = head->b_this_page; | |
890 | free_buffer_head(bh); | |
891 | } while (head); | |
892 | } | |
893 | ||
f745c6f5 | 894 | goto out; |
1da177e4 LT |
895 | } |
896 | EXPORT_SYMBOL_GPL(alloc_page_buffers); | |
897 | ||
898 | static inline void | |
899 | link_dev_buffers(struct page *page, struct buffer_head *head) | |
900 | { | |
901 | struct buffer_head *bh, *tail; | |
902 | ||
903 | bh = head; | |
904 | do { | |
905 | tail = bh; | |
906 | bh = bh->b_this_page; | |
907 | } while (bh); | |
908 | tail->b_this_page = head; | |
909 | attach_page_buffers(page, head); | |
910 | } | |
911 | ||
bbec0270 LT |
912 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
913 | { | |
914 | sector_t retval = ~((sector_t)0); | |
915 | loff_t sz = i_size_read(bdev->bd_inode); | |
916 | ||
917 | if (sz) { | |
918 | unsigned int sizebits = blksize_bits(size); | |
919 | retval = (sz >> sizebits); | |
920 | } | |
921 | return retval; | |
922 | } | |
923 | ||
1da177e4 LT |
924 | /* |
925 | * Initialise the state of a blockdev page's buffers. | |
926 | */ | |
676ce6d5 | 927 | static sector_t |
1da177e4 LT |
928 | init_page_buffers(struct page *page, struct block_device *bdev, |
929 | sector_t block, int size) | |
930 | { | |
931 | struct buffer_head *head = page_buffers(page); | |
932 | struct buffer_head *bh = head; | |
933 | int uptodate = PageUptodate(page); | |
bbec0270 | 934 | sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); |
1da177e4 LT |
935 | |
936 | do { | |
937 | if (!buffer_mapped(bh)) { | |
01950a34 EB |
938 | bh->b_end_io = NULL; |
939 | bh->b_private = NULL; | |
1da177e4 LT |
940 | bh->b_bdev = bdev; |
941 | bh->b_blocknr = block; | |
942 | if (uptodate) | |
943 | set_buffer_uptodate(bh); | |
080399aa JM |
944 | if (block < end_block) |
945 | set_buffer_mapped(bh); | |
1da177e4 LT |
946 | } |
947 | block++; | |
948 | bh = bh->b_this_page; | |
949 | } while (bh != head); | |
676ce6d5 HD |
950 | |
951 | /* | |
952 | * Caller needs to validate requested block against end of device. | |
953 | */ | |
954 | return end_block; | |
1da177e4 LT |
955 | } |
956 | ||
957 | /* | |
958 | * Create the page-cache page that contains the requested block. | |
959 | * | |
676ce6d5 | 960 | * This is used purely for blockdev mappings. |
1da177e4 | 961 | */ |
676ce6d5 | 962 | static int |
1da177e4 | 963 | grow_dev_page(struct block_device *bdev, sector_t block, |
3b5e6454 | 964 | pgoff_t index, int size, int sizebits, gfp_t gfp) |
1da177e4 LT |
965 | { |
966 | struct inode *inode = bdev->bd_inode; | |
967 | struct page *page; | |
968 | struct buffer_head *bh; | |
676ce6d5 HD |
969 | sector_t end_block; |
970 | int ret = 0; /* Will call free_more_memory() */ | |
84235de3 | 971 | gfp_t gfp_mask; |
1da177e4 | 972 | |
c62d2555 | 973 | gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; |
3b5e6454 | 974 | |
84235de3 JW |
975 | /* |
976 | * XXX: __getblk_slow() can not really deal with failure and | |
977 | * will endlessly loop on improvised global reclaim. Prefer | |
978 | * looping in the allocator rather than here, at least that | |
979 | * code knows what it's doing. | |
980 | */ | |
981 | gfp_mask |= __GFP_NOFAIL; | |
982 | ||
983 | page = find_or_create_page(inode->i_mapping, index, gfp_mask); | |
1da177e4 | 984 | |
e827f923 | 985 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
986 | |
987 | if (page_has_buffers(page)) { | |
988 | bh = page_buffers(page); | |
989 | if (bh->b_size == size) { | |
676ce6d5 | 990 | end_block = init_page_buffers(page, bdev, |
f2d5a944 AA |
991 | (sector_t)index << sizebits, |
992 | size); | |
676ce6d5 | 993 | goto done; |
1da177e4 LT |
994 | } |
995 | if (!try_to_free_buffers(page)) | |
996 | goto failed; | |
997 | } | |
998 | ||
999 | /* | |
1000 | * Allocate some buffers for this page | |
1001 | */ | |
94dc24c0 | 1002 | bh = alloc_page_buffers(page, size, true); |
1da177e4 LT |
1003 | |
1004 | /* | |
1005 | * Link the page to the buffers and initialise them. Take the | |
1006 | * lock to be atomic wrt __find_get_block(), which does not | |
1007 | * run under the page lock. | |
1008 | */ | |
1009 | spin_lock(&inode->i_mapping->private_lock); | |
1010 | link_dev_buffers(page, bh); | |
f2d5a944 AA |
1011 | end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, |
1012 | size); | |
1da177e4 | 1013 | spin_unlock(&inode->i_mapping->private_lock); |
676ce6d5 HD |
1014 | done: |
1015 | ret = (block < end_block) ? 1 : -ENXIO; | |
1da177e4 | 1016 | failed: |
1da177e4 | 1017 | unlock_page(page); |
09cbfeaf | 1018 | put_page(page); |
676ce6d5 | 1019 | return ret; |
1da177e4 LT |
1020 | } |
1021 | ||
1022 | /* | |
1023 | * Create buffers for the specified block device block's page. If | |
1024 | * that page was dirty, the buffers are set dirty also. | |
1da177e4 | 1025 | */ |
858119e1 | 1026 | static int |
3b5e6454 | 1027 | grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) |
1da177e4 | 1028 | { |
1da177e4 LT |
1029 | pgoff_t index; |
1030 | int sizebits; | |
1031 | ||
1032 | sizebits = -1; | |
1033 | do { | |
1034 | sizebits++; | |
1035 | } while ((size << sizebits) < PAGE_SIZE); | |
1036 | ||
1037 | index = block >> sizebits; | |
1da177e4 | 1038 | |
e5657933 AM |
1039 | /* |
1040 | * Check for a block which wants to lie outside our maximum possible | |
1041 | * pagecache index. (this comparison is done using sector_t types). | |
1042 | */ | |
1043 | if (unlikely(index != block >> sizebits)) { | |
e5657933 | 1044 | printk(KERN_ERR "%s: requested out-of-range block %llu for " |
a1c6f057 | 1045 | "device %pg\n", |
8e24eea7 | 1046 | __func__, (unsigned long long)block, |
a1c6f057 | 1047 | bdev); |
e5657933 AM |
1048 | return -EIO; |
1049 | } | |
676ce6d5 | 1050 | |
1da177e4 | 1051 | /* Create a page with the proper size buffers.. */ |
3b5e6454 | 1052 | return grow_dev_page(bdev, block, index, size, sizebits, gfp); |
1da177e4 LT |
1053 | } |
1054 | ||
0026ba40 | 1055 | static struct buffer_head * |
3b5e6454 GK |
1056 | __getblk_slow(struct block_device *bdev, sector_t block, |
1057 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1058 | { |
1059 | /* Size must be multiple of hard sectorsize */ | |
e1defc4f | 1060 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
1da177e4 LT |
1061 | (size < 512 || size > PAGE_SIZE))) { |
1062 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", | |
1063 | size); | |
e1defc4f MP |
1064 | printk(KERN_ERR "logical block size: %d\n", |
1065 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1066 | |
1067 | dump_stack(); | |
1068 | return NULL; | |
1069 | } | |
1070 | ||
676ce6d5 HD |
1071 | for (;;) { |
1072 | struct buffer_head *bh; | |
1073 | int ret; | |
1da177e4 LT |
1074 | |
1075 | bh = __find_get_block(bdev, block, size); | |
1076 | if (bh) | |
1077 | return bh; | |
676ce6d5 | 1078 | |
3b5e6454 | 1079 | ret = grow_buffers(bdev, block, size, gfp); |
676ce6d5 HD |
1080 | if (ret < 0) |
1081 | return NULL; | |
1da177e4 LT |
1082 | } |
1083 | } | |
1084 | ||
1085 | /* | |
1086 | * The relationship between dirty buffers and dirty pages: | |
1087 | * | |
1088 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and | |
ec82e1c1 | 1089 | * the page is tagged dirty in the page cache. |
1da177e4 LT |
1090 | * |
1091 | * At all times, the dirtiness of the buffers represents the dirtiness of | |
1092 | * subsections of the page. If the page has buffers, the page dirty bit is | |
1093 | * merely a hint about the true dirty state. | |
1094 | * | |
1095 | * When a page is set dirty in its entirety, all its buffers are marked dirty | |
1096 | * (if the page has buffers). | |
1097 | * | |
1098 | * When a buffer is marked dirty, its page is dirtied, but the page's other | |
1099 | * buffers are not. | |
1100 | * | |
1101 | * Also. When blockdev buffers are explicitly read with bread(), they | |
1102 | * individually become uptodate. But their backing page remains not | |
1103 | * uptodate - even if all of its buffers are uptodate. A subsequent | |
1104 | * block_read_full_page() against that page will discover all the uptodate | |
1105 | * buffers, will set the page uptodate and will perform no I/O. | |
1106 | */ | |
1107 | ||
1108 | /** | |
1109 | * mark_buffer_dirty - mark a buffer_head as needing writeout | |
67be2dd1 | 1110 | * @bh: the buffer_head to mark dirty |
1da177e4 | 1111 | * |
ec82e1c1 MW |
1112 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set |
1113 | * its backing page dirty, then tag the page as dirty in the page cache | |
1114 | * and then attach the address_space's inode to its superblock's dirty | |
1da177e4 LT |
1115 | * inode list. |
1116 | * | |
1117 | * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, | |
b93b0163 | 1118 | * i_pages lock and mapping->host->i_lock. |
1da177e4 | 1119 | */ |
fc9b52cd | 1120 | void mark_buffer_dirty(struct buffer_head *bh) |
1da177e4 | 1121 | { |
787d2214 | 1122 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
1be62dc1 | 1123 | |
5305cb83 TH |
1124 | trace_block_dirty_buffer(bh); |
1125 | ||
1be62dc1 LT |
1126 | /* |
1127 | * Very *carefully* optimize the it-is-already-dirty case. | |
1128 | * | |
1129 | * Don't let the final "is it dirty" escape to before we | |
1130 | * perhaps modified the buffer. | |
1131 | */ | |
1132 | if (buffer_dirty(bh)) { | |
1133 | smp_mb(); | |
1134 | if (buffer_dirty(bh)) | |
1135 | return; | |
1136 | } | |
1137 | ||
a8e7d49a LT |
1138 | if (!test_set_buffer_dirty(bh)) { |
1139 | struct page *page = bh->b_page; | |
c4843a75 | 1140 | struct address_space *mapping = NULL; |
c4843a75 | 1141 | |
62cccb8c | 1142 | lock_page_memcg(page); |
8e9d78ed | 1143 | if (!TestSetPageDirty(page)) { |
c4843a75 | 1144 | mapping = page_mapping(page); |
8e9d78ed | 1145 | if (mapping) |
62cccb8c | 1146 | __set_page_dirty(page, mapping, 0); |
8e9d78ed | 1147 | } |
62cccb8c | 1148 | unlock_page_memcg(page); |
c4843a75 GT |
1149 | if (mapping) |
1150 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
a8e7d49a | 1151 | } |
1da177e4 | 1152 | } |
1fe72eaa | 1153 | EXPORT_SYMBOL(mark_buffer_dirty); |
1da177e4 | 1154 | |
87354e5d JL |
1155 | void mark_buffer_write_io_error(struct buffer_head *bh) |
1156 | { | |
1157 | set_buffer_write_io_error(bh); | |
1158 | /* FIXME: do we need to set this in both places? */ | |
1159 | if (bh->b_page && bh->b_page->mapping) | |
1160 | mapping_set_error(bh->b_page->mapping, -EIO); | |
1161 | if (bh->b_assoc_map) | |
1162 | mapping_set_error(bh->b_assoc_map, -EIO); | |
1163 | } | |
1164 | EXPORT_SYMBOL(mark_buffer_write_io_error); | |
1165 | ||
1da177e4 LT |
1166 | /* |
1167 | * Decrement a buffer_head's reference count. If all buffers against a page | |
1168 | * have zero reference count, are clean and unlocked, and if the page is clean | |
1169 | * and unlocked then try_to_free_buffers() may strip the buffers from the page | |
1170 | * in preparation for freeing it (sometimes, rarely, buffers are removed from | |
1171 | * a page but it ends up not being freed, and buffers may later be reattached). | |
1172 | */ | |
1173 | void __brelse(struct buffer_head * buf) | |
1174 | { | |
1175 | if (atomic_read(&buf->b_count)) { | |
1176 | put_bh(buf); | |
1177 | return; | |
1178 | } | |
5c752ad9 | 1179 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
1da177e4 | 1180 | } |
1fe72eaa | 1181 | EXPORT_SYMBOL(__brelse); |
1da177e4 LT |
1182 | |
1183 | /* | |
1184 | * bforget() is like brelse(), except it discards any | |
1185 | * potentially dirty data. | |
1186 | */ | |
1187 | void __bforget(struct buffer_head *bh) | |
1188 | { | |
1189 | clear_buffer_dirty(bh); | |
535ee2fb | 1190 | if (bh->b_assoc_map) { |
1da177e4 LT |
1191 | struct address_space *buffer_mapping = bh->b_page->mapping; |
1192 | ||
1193 | spin_lock(&buffer_mapping->private_lock); | |
1194 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 1195 | bh->b_assoc_map = NULL; |
1da177e4 LT |
1196 | spin_unlock(&buffer_mapping->private_lock); |
1197 | } | |
1198 | __brelse(bh); | |
1199 | } | |
1fe72eaa | 1200 | EXPORT_SYMBOL(__bforget); |
1da177e4 LT |
1201 | |
1202 | static struct buffer_head *__bread_slow(struct buffer_head *bh) | |
1203 | { | |
1204 | lock_buffer(bh); | |
1205 | if (buffer_uptodate(bh)) { | |
1206 | unlock_buffer(bh); | |
1207 | return bh; | |
1208 | } else { | |
1209 | get_bh(bh); | |
1210 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 1211 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
1212 | wait_on_buffer(bh); |
1213 | if (buffer_uptodate(bh)) | |
1214 | return bh; | |
1215 | } | |
1216 | brelse(bh); | |
1217 | return NULL; | |
1218 | } | |
1219 | ||
1220 | /* | |
1221 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). | |
1222 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their | |
1223 | * refcount elevated by one when they're in an LRU. A buffer can only appear | |
1224 | * once in a particular CPU's LRU. A single buffer can be present in multiple | |
1225 | * CPU's LRUs at the same time. | |
1226 | * | |
1227 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and | |
1228 | * sb_find_get_block(). | |
1229 | * | |
1230 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use | |
1231 | * a local interrupt disable for that. | |
1232 | */ | |
1233 | ||
86cf78d7 | 1234 | #define BH_LRU_SIZE 16 |
1da177e4 LT |
1235 | |
1236 | struct bh_lru { | |
1237 | struct buffer_head *bhs[BH_LRU_SIZE]; | |
1238 | }; | |
1239 | ||
1240 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; | |
1241 | ||
1242 | #ifdef CONFIG_SMP | |
1243 | #define bh_lru_lock() local_irq_disable() | |
1244 | #define bh_lru_unlock() local_irq_enable() | |
1245 | #else | |
1246 | #define bh_lru_lock() preempt_disable() | |
1247 | #define bh_lru_unlock() preempt_enable() | |
1248 | #endif | |
1249 | ||
1250 | static inline void check_irqs_on(void) | |
1251 | { | |
1252 | #ifdef irqs_disabled | |
1253 | BUG_ON(irqs_disabled()); | |
1254 | #endif | |
1255 | } | |
1256 | ||
1257 | /* | |
241f01fb EB |
1258 | * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is |
1259 | * inserted at the front, and the buffer_head at the back if any is evicted. | |
1260 | * Or, if already in the LRU it is moved to the front. | |
1da177e4 LT |
1261 | */ |
1262 | static void bh_lru_install(struct buffer_head *bh) | |
1263 | { | |
241f01fb EB |
1264 | struct buffer_head *evictee = bh; |
1265 | struct bh_lru *b; | |
1266 | int i; | |
1da177e4 LT |
1267 | |
1268 | check_irqs_on(); | |
1269 | bh_lru_lock(); | |
1da177e4 | 1270 | |
241f01fb EB |
1271 | b = this_cpu_ptr(&bh_lrus); |
1272 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1273 | swap(evictee, b->bhs[i]); | |
1274 | if (evictee == bh) { | |
1275 | bh_lru_unlock(); | |
1276 | return; | |
1da177e4 | 1277 | } |
1da177e4 | 1278 | } |
1da177e4 | 1279 | |
241f01fb EB |
1280 | get_bh(bh); |
1281 | bh_lru_unlock(); | |
1282 | brelse(evictee); | |
1da177e4 LT |
1283 | } |
1284 | ||
1285 | /* | |
1286 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. | |
1287 | */ | |
858119e1 | 1288 | static struct buffer_head * |
3991d3bd | 1289 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1290 | { |
1291 | struct buffer_head *ret = NULL; | |
3991d3bd | 1292 | unsigned int i; |
1da177e4 LT |
1293 | |
1294 | check_irqs_on(); | |
1295 | bh_lru_lock(); | |
1da177e4 | 1296 | for (i = 0; i < BH_LRU_SIZE; i++) { |
c7b92516 | 1297 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
1da177e4 | 1298 | |
9470dd5d ZB |
1299 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
1300 | bh->b_size == size) { | |
1da177e4 LT |
1301 | if (i) { |
1302 | while (i) { | |
c7b92516 CL |
1303 | __this_cpu_write(bh_lrus.bhs[i], |
1304 | __this_cpu_read(bh_lrus.bhs[i - 1])); | |
1da177e4 LT |
1305 | i--; |
1306 | } | |
c7b92516 | 1307 | __this_cpu_write(bh_lrus.bhs[0], bh); |
1da177e4 LT |
1308 | } |
1309 | get_bh(bh); | |
1310 | ret = bh; | |
1311 | break; | |
1312 | } | |
1313 | } | |
1314 | bh_lru_unlock(); | |
1315 | return ret; | |
1316 | } | |
1317 | ||
1318 | /* | |
1319 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh | |
1320 | * it in the LRU and mark it as accessed. If it is not present then return | |
1321 | * NULL | |
1322 | */ | |
1323 | struct buffer_head * | |
3991d3bd | 1324 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1325 | { |
1326 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); | |
1327 | ||
1328 | if (bh == NULL) { | |
2457aec6 | 1329 | /* __find_get_block_slow will mark the page accessed */ |
385fd4c5 | 1330 | bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1331 | if (bh) |
1332 | bh_lru_install(bh); | |
2457aec6 | 1333 | } else |
1da177e4 | 1334 | touch_buffer(bh); |
2457aec6 | 1335 | |
1da177e4 LT |
1336 | return bh; |
1337 | } | |
1338 | EXPORT_SYMBOL(__find_get_block); | |
1339 | ||
1340 | /* | |
3b5e6454 | 1341 | * __getblk_gfp() will locate (and, if necessary, create) the buffer_head |
1da177e4 LT |
1342 | * which corresponds to the passed block_device, block and size. The |
1343 | * returned buffer has its reference count incremented. | |
1344 | * | |
3b5e6454 GK |
1345 | * __getblk_gfp() will lock up the machine if grow_dev_page's |
1346 | * try_to_free_buffers() attempt is failing. FIXME, perhaps? | |
1da177e4 LT |
1347 | */ |
1348 | struct buffer_head * | |
3b5e6454 GK |
1349 | __getblk_gfp(struct block_device *bdev, sector_t block, |
1350 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1351 | { |
1352 | struct buffer_head *bh = __find_get_block(bdev, block, size); | |
1353 | ||
1354 | might_sleep(); | |
1355 | if (bh == NULL) | |
3b5e6454 | 1356 | bh = __getblk_slow(bdev, block, size, gfp); |
1da177e4 LT |
1357 | return bh; |
1358 | } | |
3b5e6454 | 1359 | EXPORT_SYMBOL(__getblk_gfp); |
1da177e4 LT |
1360 | |
1361 | /* | |
1362 | * Do async read-ahead on a buffer.. | |
1363 | */ | |
3991d3bd | 1364 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1365 | { |
1366 | struct buffer_head *bh = __getblk(bdev, block, size); | |
a3e713b5 | 1367 | if (likely(bh)) { |
70246286 | 1368 | ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); |
a3e713b5 AM |
1369 | brelse(bh); |
1370 | } | |
1da177e4 LT |
1371 | } |
1372 | EXPORT_SYMBOL(__breadahead); | |
1373 | ||
1374 | /** | |
3b5e6454 | 1375 | * __bread_gfp() - reads a specified block and returns the bh |
67be2dd1 | 1376 | * @bdev: the block_device to read from |
1da177e4 LT |
1377 | * @block: number of block |
1378 | * @size: size (in bytes) to read | |
3b5e6454 GK |
1379 | * @gfp: page allocation flag |
1380 | * | |
1da177e4 | 1381 | * Reads a specified block, and returns buffer head that contains it. |
3b5e6454 GK |
1382 | * The page cache can be allocated from non-movable area |
1383 | * not to prevent page migration if you set gfp to zero. | |
1da177e4 LT |
1384 | * It returns NULL if the block was unreadable. |
1385 | */ | |
1386 | struct buffer_head * | |
3b5e6454 GK |
1387 | __bread_gfp(struct block_device *bdev, sector_t block, |
1388 | unsigned size, gfp_t gfp) | |
1da177e4 | 1389 | { |
3b5e6454 | 1390 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); |
1da177e4 | 1391 | |
a3e713b5 | 1392 | if (likely(bh) && !buffer_uptodate(bh)) |
1da177e4 LT |
1393 | bh = __bread_slow(bh); |
1394 | return bh; | |
1395 | } | |
3b5e6454 | 1396 | EXPORT_SYMBOL(__bread_gfp); |
1da177e4 LT |
1397 | |
1398 | /* | |
1399 | * invalidate_bh_lrus() is called rarely - but not only at unmount. | |
1400 | * This doesn't race because it runs in each cpu either in irq | |
1401 | * or with preempt disabled. | |
1402 | */ | |
1403 | static void invalidate_bh_lru(void *arg) | |
1404 | { | |
1405 | struct bh_lru *b = &get_cpu_var(bh_lrus); | |
1406 | int i; | |
1407 | ||
1408 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1409 | brelse(b->bhs[i]); | |
1410 | b->bhs[i] = NULL; | |
1411 | } | |
1412 | put_cpu_var(bh_lrus); | |
1413 | } | |
42be35d0 GBY |
1414 | |
1415 | static bool has_bh_in_lru(int cpu, void *dummy) | |
1416 | { | |
1417 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); | |
1418 | int i; | |
1da177e4 | 1419 | |
42be35d0 GBY |
1420 | for (i = 0; i < BH_LRU_SIZE; i++) { |
1421 | if (b->bhs[i]) | |
1d706679 | 1422 | return true; |
42be35d0 GBY |
1423 | } |
1424 | ||
1d706679 | 1425 | return false; |
42be35d0 GBY |
1426 | } |
1427 | ||
f9a14399 | 1428 | void invalidate_bh_lrus(void) |
1da177e4 | 1429 | { |
cb923159 | 1430 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); |
1da177e4 | 1431 | } |
9db5579b | 1432 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
1da177e4 LT |
1433 | |
1434 | void set_bh_page(struct buffer_head *bh, | |
1435 | struct page *page, unsigned long offset) | |
1436 | { | |
1437 | bh->b_page = page; | |
e827f923 | 1438 | BUG_ON(offset >= PAGE_SIZE); |
1da177e4 LT |
1439 | if (PageHighMem(page)) |
1440 | /* | |
1441 | * This catches illegal uses and preserves the offset: | |
1442 | */ | |
1443 | bh->b_data = (char *)(0 + offset); | |
1444 | else | |
1445 | bh->b_data = page_address(page) + offset; | |
1446 | } | |
1447 | EXPORT_SYMBOL(set_bh_page); | |
1448 | ||
1449 | /* | |
1450 | * Called when truncating a buffer on a page completely. | |
1451 | */ | |
e7470ee8 MG |
1452 | |
1453 | /* Bits that are cleared during an invalidate */ | |
1454 | #define BUFFER_FLAGS_DISCARD \ | |
1455 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ | |
1456 | 1 << BH_Delay | 1 << BH_Unwritten) | |
1457 | ||
858119e1 | 1458 | static void discard_buffer(struct buffer_head * bh) |
1da177e4 | 1459 | { |
e7470ee8 MG |
1460 | unsigned long b_state, b_state_old; |
1461 | ||
1da177e4 LT |
1462 | lock_buffer(bh); |
1463 | clear_buffer_dirty(bh); | |
1464 | bh->b_bdev = NULL; | |
e7470ee8 MG |
1465 | b_state = bh->b_state; |
1466 | for (;;) { | |
1467 | b_state_old = cmpxchg(&bh->b_state, b_state, | |
1468 | (b_state & ~BUFFER_FLAGS_DISCARD)); | |
1469 | if (b_state_old == b_state) | |
1470 | break; | |
1471 | b_state = b_state_old; | |
1472 | } | |
1da177e4 LT |
1473 | unlock_buffer(bh); |
1474 | } | |
1475 | ||
1da177e4 | 1476 | /** |
814e1d25 | 1477 | * block_invalidatepage - invalidate part or all of a buffer-backed page |
1da177e4 LT |
1478 | * |
1479 | * @page: the page which is affected | |
d47992f8 LC |
1480 | * @offset: start of the range to invalidate |
1481 | * @length: length of the range to invalidate | |
1da177e4 LT |
1482 | * |
1483 | * block_invalidatepage() is called when all or part of the page has become | |
814e1d25 | 1484 | * invalidated by a truncate operation. |
1da177e4 LT |
1485 | * |
1486 | * block_invalidatepage() does not have to release all buffers, but it must | |
1487 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
1488 | * is underway against any of the blocks which are outside the truncation | |
1489 | * point. Because the caller is about to free (and possibly reuse) those | |
1490 | * blocks on-disk. | |
1491 | */ | |
d47992f8 LC |
1492 | void block_invalidatepage(struct page *page, unsigned int offset, |
1493 | unsigned int length) | |
1da177e4 LT |
1494 | { |
1495 | struct buffer_head *head, *bh, *next; | |
1496 | unsigned int curr_off = 0; | |
d47992f8 | 1497 | unsigned int stop = length + offset; |
1da177e4 LT |
1498 | |
1499 | BUG_ON(!PageLocked(page)); | |
1500 | if (!page_has_buffers(page)) | |
1501 | goto out; | |
1502 | ||
d47992f8 LC |
1503 | /* |
1504 | * Check for overflow | |
1505 | */ | |
09cbfeaf | 1506 | BUG_ON(stop > PAGE_SIZE || stop < length); |
d47992f8 | 1507 | |
1da177e4 LT |
1508 | head = page_buffers(page); |
1509 | bh = head; | |
1510 | do { | |
1511 | unsigned int next_off = curr_off + bh->b_size; | |
1512 | next = bh->b_this_page; | |
1513 | ||
d47992f8 LC |
1514 | /* |
1515 | * Are we still fully in range ? | |
1516 | */ | |
1517 | if (next_off > stop) | |
1518 | goto out; | |
1519 | ||
1da177e4 LT |
1520 | /* |
1521 | * is this block fully invalidated? | |
1522 | */ | |
1523 | if (offset <= curr_off) | |
1524 | discard_buffer(bh); | |
1525 | curr_off = next_off; | |
1526 | bh = next; | |
1527 | } while (bh != head); | |
1528 | ||
1529 | /* | |
1530 | * We release buffers only if the entire page is being invalidated. | |
1531 | * The get_block cached value has been unconditionally invalidated, | |
1532 | * so real IO is not possible anymore. | |
1533 | */ | |
3172485f | 1534 | if (length == PAGE_SIZE) |
2ff28e22 | 1535 | try_to_release_page(page, 0); |
1da177e4 | 1536 | out: |
2ff28e22 | 1537 | return; |
1da177e4 LT |
1538 | } |
1539 | EXPORT_SYMBOL(block_invalidatepage); | |
1540 | ||
d47992f8 | 1541 | |
1da177e4 LT |
1542 | /* |
1543 | * We attach and possibly dirty the buffers atomically wrt | |
1544 | * __set_page_dirty_buffers() via private_lock. try_to_free_buffers | |
1545 | * is already excluded via the page lock. | |
1546 | */ | |
1547 | void create_empty_buffers(struct page *page, | |
1548 | unsigned long blocksize, unsigned long b_state) | |
1549 | { | |
1550 | struct buffer_head *bh, *head, *tail; | |
1551 | ||
640ab98f | 1552 | head = alloc_page_buffers(page, blocksize, true); |
1da177e4 LT |
1553 | bh = head; |
1554 | do { | |
1555 | bh->b_state |= b_state; | |
1556 | tail = bh; | |
1557 | bh = bh->b_this_page; | |
1558 | } while (bh); | |
1559 | tail->b_this_page = head; | |
1560 | ||
1561 | spin_lock(&page->mapping->private_lock); | |
1562 | if (PageUptodate(page) || PageDirty(page)) { | |
1563 | bh = head; | |
1564 | do { | |
1565 | if (PageDirty(page)) | |
1566 | set_buffer_dirty(bh); | |
1567 | if (PageUptodate(page)) | |
1568 | set_buffer_uptodate(bh); | |
1569 | bh = bh->b_this_page; | |
1570 | } while (bh != head); | |
1571 | } | |
1572 | attach_page_buffers(page, head); | |
1573 | spin_unlock(&page->mapping->private_lock); | |
1574 | } | |
1575 | EXPORT_SYMBOL(create_empty_buffers); | |
1576 | ||
29f3ad7d JK |
1577 | /** |
1578 | * clean_bdev_aliases: clean a range of buffers in block device | |
1579 | * @bdev: Block device to clean buffers in | |
1580 | * @block: Start of a range of blocks to clean | |
1581 | * @len: Number of blocks to clean | |
1da177e4 | 1582 | * |
29f3ad7d JK |
1583 | * We are taking a range of blocks for data and we don't want writeback of any |
1584 | * buffer-cache aliases starting from return from this function and until the | |
1585 | * moment when something will explicitly mark the buffer dirty (hopefully that | |
1586 | * will not happen until we will free that block ;-) We don't even need to mark | |
1587 | * it not-uptodate - nobody can expect anything from a newly allocated buffer | |
1588 | * anyway. We used to use unmap_buffer() for such invalidation, but that was | |
1589 | * wrong. We definitely don't want to mark the alias unmapped, for example - it | |
1590 | * would confuse anyone who might pick it with bread() afterwards... | |
1591 | * | |
1592 | * Also.. Note that bforget() doesn't lock the buffer. So there can be | |
1593 | * writeout I/O going on against recently-freed buffers. We don't wait on that | |
1594 | * I/O in bforget() - it's more efficient to wait on the I/O only if we really | |
1595 | * need to. That happens here. | |
1da177e4 | 1596 | */ |
29f3ad7d | 1597 | void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) |
1da177e4 | 1598 | { |
29f3ad7d JK |
1599 | struct inode *bd_inode = bdev->bd_inode; |
1600 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
1601 | struct pagevec pvec; | |
1602 | pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); | |
1603 | pgoff_t end; | |
c10f778d | 1604 | int i, count; |
29f3ad7d JK |
1605 | struct buffer_head *bh; |
1606 | struct buffer_head *head; | |
1da177e4 | 1607 | |
29f3ad7d | 1608 | end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); |
86679820 | 1609 | pagevec_init(&pvec); |
397162ff | 1610 | while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) { |
c10f778d JK |
1611 | count = pagevec_count(&pvec); |
1612 | for (i = 0; i < count; i++) { | |
29f3ad7d | 1613 | struct page *page = pvec.pages[i]; |
1da177e4 | 1614 | |
29f3ad7d JK |
1615 | if (!page_has_buffers(page)) |
1616 | continue; | |
1617 | /* | |
1618 | * We use page lock instead of bd_mapping->private_lock | |
1619 | * to pin buffers here since we can afford to sleep and | |
1620 | * it scales better than a global spinlock lock. | |
1621 | */ | |
1622 | lock_page(page); | |
1623 | /* Recheck when the page is locked which pins bhs */ | |
1624 | if (!page_has_buffers(page)) | |
1625 | goto unlock_page; | |
1626 | head = page_buffers(page); | |
1627 | bh = head; | |
1628 | do { | |
6c006a9d | 1629 | if (!buffer_mapped(bh) || (bh->b_blocknr < block)) |
29f3ad7d JK |
1630 | goto next; |
1631 | if (bh->b_blocknr >= block + len) | |
1632 | break; | |
1633 | clear_buffer_dirty(bh); | |
1634 | wait_on_buffer(bh); | |
1635 | clear_buffer_req(bh); | |
1636 | next: | |
1637 | bh = bh->b_this_page; | |
1638 | } while (bh != head); | |
1639 | unlock_page: | |
1640 | unlock_page(page); | |
1641 | } | |
1642 | pagevec_release(&pvec); | |
1643 | cond_resched(); | |
c10f778d JK |
1644 | /* End of range already reached? */ |
1645 | if (index > end || !index) | |
1646 | break; | |
1da177e4 LT |
1647 | } |
1648 | } | |
29f3ad7d | 1649 | EXPORT_SYMBOL(clean_bdev_aliases); |
1da177e4 | 1650 | |
45bce8f3 LT |
1651 | /* |
1652 | * Size is a power-of-two in the range 512..PAGE_SIZE, | |
1653 | * and the case we care about most is PAGE_SIZE. | |
1654 | * | |
1655 | * So this *could* possibly be written with those | |
1656 | * constraints in mind (relevant mostly if some | |
1657 | * architecture has a slow bit-scan instruction) | |
1658 | */ | |
1659 | static inline int block_size_bits(unsigned int blocksize) | |
1660 | { | |
1661 | return ilog2(blocksize); | |
1662 | } | |
1663 | ||
1664 | static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) | |
1665 | { | |
1666 | BUG_ON(!PageLocked(page)); | |
1667 | ||
1668 | if (!page_has_buffers(page)) | |
6aa7de05 MR |
1669 | create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits), |
1670 | b_state); | |
45bce8f3 LT |
1671 | return page_buffers(page); |
1672 | } | |
1673 | ||
1da177e4 LT |
1674 | /* |
1675 | * NOTE! All mapped/uptodate combinations are valid: | |
1676 | * | |
1677 | * Mapped Uptodate Meaning | |
1678 | * | |
1679 | * No No "unknown" - must do get_block() | |
1680 | * No Yes "hole" - zero-filled | |
1681 | * Yes No "allocated" - allocated on disk, not read in | |
1682 | * Yes Yes "valid" - allocated and up-to-date in memory. | |
1683 | * | |
1684 | * "Dirty" is valid only with the last case (mapped+uptodate). | |
1685 | */ | |
1686 | ||
1687 | /* | |
1688 | * While block_write_full_page is writing back the dirty buffers under | |
1689 | * the page lock, whoever dirtied the buffers may decide to clean them | |
1690 | * again at any time. We handle that by only looking at the buffer | |
1691 | * state inside lock_buffer(). | |
1692 | * | |
1693 | * If block_write_full_page() is called for regular writeback | |
1694 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a | |
1695 | * locked buffer. This only can happen if someone has written the buffer | |
1696 | * directly, with submit_bh(). At the address_space level PageWriteback | |
1697 | * prevents this contention from occurring. | |
6e34eedd TT |
1698 | * |
1699 | * If block_write_full_page() is called with wbc->sync_mode == | |
70fd7614 | 1700 | * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this |
721a9602 | 1701 | * causes the writes to be flagged as synchronous writes. |
1da177e4 | 1702 | */ |
b4bba389 | 1703 | int __block_write_full_page(struct inode *inode, struct page *page, |
35c80d5f CM |
1704 | get_block_t *get_block, struct writeback_control *wbc, |
1705 | bh_end_io_t *handler) | |
1da177e4 LT |
1706 | { |
1707 | int err; | |
1708 | sector_t block; | |
1709 | sector_t last_block; | |
f0fbd5fc | 1710 | struct buffer_head *bh, *head; |
45bce8f3 | 1711 | unsigned int blocksize, bbits; |
1da177e4 | 1712 | int nr_underway = 0; |
7637241e | 1713 | int write_flags = wbc_to_write_flags(wbc); |
1da177e4 | 1714 | |
45bce8f3 | 1715 | head = create_page_buffers(page, inode, |
1da177e4 | 1716 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
1da177e4 LT |
1717 | |
1718 | /* | |
1719 | * Be very careful. We have no exclusion from __set_page_dirty_buffers | |
1720 | * here, and the (potentially unmapped) buffers may become dirty at | |
1721 | * any time. If a buffer becomes dirty here after we've inspected it | |
1722 | * then we just miss that fact, and the page stays dirty. | |
1723 | * | |
1724 | * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; | |
1725 | * handle that here by just cleaning them. | |
1726 | */ | |
1727 | ||
1da177e4 | 1728 | bh = head; |
45bce8f3 LT |
1729 | blocksize = bh->b_size; |
1730 | bbits = block_size_bits(blocksize); | |
1731 | ||
09cbfeaf | 1732 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 1733 | last_block = (i_size_read(inode) - 1) >> bbits; |
1da177e4 LT |
1734 | |
1735 | /* | |
1736 | * Get all the dirty buffers mapped to disk addresses and | |
1737 | * handle any aliases from the underlying blockdev's mapping. | |
1738 | */ | |
1739 | do { | |
1740 | if (block > last_block) { | |
1741 | /* | |
1742 | * mapped buffers outside i_size will occur, because | |
1743 | * this page can be outside i_size when there is a | |
1744 | * truncate in progress. | |
1745 | */ | |
1746 | /* | |
1747 | * The buffer was zeroed by block_write_full_page() | |
1748 | */ | |
1749 | clear_buffer_dirty(bh); | |
1750 | set_buffer_uptodate(bh); | |
29a814d2 AT |
1751 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
1752 | buffer_dirty(bh)) { | |
b0cf2321 | 1753 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1754 | err = get_block(inode, block, bh, 1); |
1755 | if (err) | |
1756 | goto recover; | |
29a814d2 | 1757 | clear_buffer_delay(bh); |
1da177e4 LT |
1758 | if (buffer_new(bh)) { |
1759 | /* blockdev mappings never come here */ | |
1760 | clear_buffer_new(bh); | |
e64855c6 | 1761 | clean_bdev_bh_alias(bh); |
1da177e4 LT |
1762 | } |
1763 | } | |
1764 | bh = bh->b_this_page; | |
1765 | block++; | |
1766 | } while (bh != head); | |
1767 | ||
1768 | do { | |
1da177e4 LT |
1769 | if (!buffer_mapped(bh)) |
1770 | continue; | |
1771 | /* | |
1772 | * If it's a fully non-blocking write attempt and we cannot | |
1773 | * lock the buffer then redirty the page. Note that this can | |
5b0830cb JA |
1774 | * potentially cause a busy-wait loop from writeback threads |
1775 | * and kswapd activity, but those code paths have their own | |
1776 | * higher-level throttling. | |
1da177e4 | 1777 | */ |
1b430bee | 1778 | if (wbc->sync_mode != WB_SYNC_NONE) { |
1da177e4 | 1779 | lock_buffer(bh); |
ca5de404 | 1780 | } else if (!trylock_buffer(bh)) { |
1da177e4 LT |
1781 | redirty_page_for_writepage(wbc, page); |
1782 | continue; | |
1783 | } | |
1784 | if (test_clear_buffer_dirty(bh)) { | |
35c80d5f | 1785 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1786 | } else { |
1787 | unlock_buffer(bh); | |
1788 | } | |
1789 | } while ((bh = bh->b_this_page) != head); | |
1790 | ||
1791 | /* | |
1792 | * The page and its buffers are protected by PageWriteback(), so we can | |
1793 | * drop the bh refcounts early. | |
1794 | */ | |
1795 | BUG_ON(PageWriteback(page)); | |
1796 | set_page_writeback(page); | |
1da177e4 LT |
1797 | |
1798 | do { | |
1799 | struct buffer_head *next = bh->b_this_page; | |
1800 | if (buffer_async_write(bh)) { | |
8e8f9298 JA |
1801 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1802 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1803 | nr_underway++; |
1804 | } | |
1da177e4 LT |
1805 | bh = next; |
1806 | } while (bh != head); | |
05937baa | 1807 | unlock_page(page); |
1da177e4 LT |
1808 | |
1809 | err = 0; | |
1810 | done: | |
1811 | if (nr_underway == 0) { | |
1812 | /* | |
1813 | * The page was marked dirty, but the buffers were | |
1814 | * clean. Someone wrote them back by hand with | |
1815 | * ll_rw_block/submit_bh. A rare case. | |
1816 | */ | |
1da177e4 | 1817 | end_page_writeback(page); |
3d67f2d7 | 1818 | |
1da177e4 LT |
1819 | /* |
1820 | * The page and buffer_heads can be released at any time from | |
1821 | * here on. | |
1822 | */ | |
1da177e4 LT |
1823 | } |
1824 | return err; | |
1825 | ||
1826 | recover: | |
1827 | /* | |
1828 | * ENOSPC, or some other error. We may already have added some | |
1829 | * blocks to the file, so we need to write these out to avoid | |
1830 | * exposing stale data. | |
1831 | * The page is currently locked and not marked for writeback | |
1832 | */ | |
1833 | bh = head; | |
1834 | /* Recovery: lock and submit the mapped buffers */ | |
1835 | do { | |
29a814d2 AT |
1836 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
1837 | !buffer_delay(bh)) { | |
1da177e4 | 1838 | lock_buffer(bh); |
35c80d5f | 1839 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1840 | } else { |
1841 | /* | |
1842 | * The buffer may have been set dirty during | |
1843 | * attachment to a dirty page. | |
1844 | */ | |
1845 | clear_buffer_dirty(bh); | |
1846 | } | |
1847 | } while ((bh = bh->b_this_page) != head); | |
1848 | SetPageError(page); | |
1849 | BUG_ON(PageWriteback(page)); | |
7e4c3690 | 1850 | mapping_set_error(page->mapping, err); |
1da177e4 | 1851 | set_page_writeback(page); |
1da177e4 LT |
1852 | do { |
1853 | struct buffer_head *next = bh->b_this_page; | |
1854 | if (buffer_async_write(bh)) { | |
1855 | clear_buffer_dirty(bh); | |
8e8f9298 JA |
1856 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1857 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1858 | nr_underway++; |
1859 | } | |
1da177e4 LT |
1860 | bh = next; |
1861 | } while (bh != head); | |
ffda9d30 | 1862 | unlock_page(page); |
1da177e4 LT |
1863 | goto done; |
1864 | } | |
b4bba389 | 1865 | EXPORT_SYMBOL(__block_write_full_page); |
1da177e4 | 1866 | |
afddba49 NP |
1867 | /* |
1868 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
1869 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1870 | * block data from leaking). And clear the new bit. | |
1871 | */ | |
1872 | void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1873 | { | |
1874 | unsigned int block_start, block_end; | |
1875 | struct buffer_head *head, *bh; | |
1876 | ||
1877 | BUG_ON(!PageLocked(page)); | |
1878 | if (!page_has_buffers(page)) | |
1879 | return; | |
1880 | ||
1881 | bh = head = page_buffers(page); | |
1882 | block_start = 0; | |
1883 | do { | |
1884 | block_end = block_start + bh->b_size; | |
1885 | ||
1886 | if (buffer_new(bh)) { | |
1887 | if (block_end > from && block_start < to) { | |
1888 | if (!PageUptodate(page)) { | |
1889 | unsigned start, size; | |
1890 | ||
1891 | start = max(from, block_start); | |
1892 | size = min(to, block_end) - start; | |
1893 | ||
eebd2aa3 | 1894 | zero_user(page, start, size); |
afddba49 NP |
1895 | set_buffer_uptodate(bh); |
1896 | } | |
1897 | ||
1898 | clear_buffer_new(bh); | |
1899 | mark_buffer_dirty(bh); | |
1900 | } | |
1901 | } | |
1902 | ||
1903 | block_start = block_end; | |
1904 | bh = bh->b_this_page; | |
1905 | } while (bh != head); | |
1906 | } | |
1907 | EXPORT_SYMBOL(page_zero_new_buffers); | |
1908 | ||
ae259a9c CH |
1909 | static void |
1910 | iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, | |
1911 | struct iomap *iomap) | |
1912 | { | |
1913 | loff_t offset = block << inode->i_blkbits; | |
1914 | ||
1915 | bh->b_bdev = iomap->bdev; | |
1916 | ||
1917 | /* | |
1918 | * Block points to offset in file we need to map, iomap contains | |
1919 | * the offset at which the map starts. If the map ends before the | |
1920 | * current block, then do not map the buffer and let the caller | |
1921 | * handle it. | |
1922 | */ | |
1923 | BUG_ON(offset >= iomap->offset + iomap->length); | |
1924 | ||
1925 | switch (iomap->type) { | |
1926 | case IOMAP_HOLE: | |
1927 | /* | |
1928 | * If the buffer is not up to date or beyond the current EOF, | |
1929 | * we need to mark it as new to ensure sub-block zeroing is | |
1930 | * executed if necessary. | |
1931 | */ | |
1932 | if (!buffer_uptodate(bh) || | |
1933 | (offset >= i_size_read(inode))) | |
1934 | set_buffer_new(bh); | |
1935 | break; | |
1936 | case IOMAP_DELALLOC: | |
1937 | if (!buffer_uptodate(bh) || | |
1938 | (offset >= i_size_read(inode))) | |
1939 | set_buffer_new(bh); | |
1940 | set_buffer_uptodate(bh); | |
1941 | set_buffer_mapped(bh); | |
1942 | set_buffer_delay(bh); | |
1943 | break; | |
1944 | case IOMAP_UNWRITTEN: | |
1945 | /* | |
3d7b6b21 AG |
1946 | * For unwritten regions, we always need to ensure that regions |
1947 | * in the block we are not writing to are zeroed. Mark the | |
1948 | * buffer as new to ensure this. | |
ae259a9c CH |
1949 | */ |
1950 | set_buffer_new(bh); | |
1951 | set_buffer_unwritten(bh); | |
1952 | /* FALLTHRU */ | |
1953 | case IOMAP_MAPPED: | |
3d7b6b21 AG |
1954 | if ((iomap->flags & IOMAP_F_NEW) || |
1955 | offset >= i_size_read(inode)) | |
ae259a9c | 1956 | set_buffer_new(bh); |
19fe5f64 AG |
1957 | bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> |
1958 | inode->i_blkbits; | |
ae259a9c CH |
1959 | set_buffer_mapped(bh); |
1960 | break; | |
1961 | } | |
1962 | } | |
1963 | ||
1964 | int __block_write_begin_int(struct page *page, loff_t pos, unsigned len, | |
1965 | get_block_t *get_block, struct iomap *iomap) | |
1da177e4 | 1966 | { |
09cbfeaf | 1967 | unsigned from = pos & (PAGE_SIZE - 1); |
ebdec241 | 1968 | unsigned to = from + len; |
6e1db88d | 1969 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
1970 | unsigned block_start, block_end; |
1971 | sector_t block; | |
1972 | int err = 0; | |
1973 | unsigned blocksize, bbits; | |
1974 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; | |
1975 | ||
1976 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1977 | BUG_ON(from > PAGE_SIZE); |
1978 | BUG_ON(to > PAGE_SIZE); | |
1da177e4 LT |
1979 | BUG_ON(from > to); |
1980 | ||
45bce8f3 LT |
1981 | head = create_page_buffers(page, inode, 0); |
1982 | blocksize = head->b_size; | |
1983 | bbits = block_size_bits(blocksize); | |
1da177e4 | 1984 | |
09cbfeaf | 1985 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
1da177e4 LT |
1986 | |
1987 | for(bh = head, block_start = 0; bh != head || !block_start; | |
1988 | block++, block_start=block_end, bh = bh->b_this_page) { | |
1989 | block_end = block_start + blocksize; | |
1990 | if (block_end <= from || block_start >= to) { | |
1991 | if (PageUptodate(page)) { | |
1992 | if (!buffer_uptodate(bh)) | |
1993 | set_buffer_uptodate(bh); | |
1994 | } | |
1995 | continue; | |
1996 | } | |
1997 | if (buffer_new(bh)) | |
1998 | clear_buffer_new(bh); | |
1999 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2000 | WARN_ON(bh->b_size != blocksize); |
ae259a9c CH |
2001 | if (get_block) { |
2002 | err = get_block(inode, block, bh, 1); | |
2003 | if (err) | |
2004 | break; | |
2005 | } else { | |
2006 | iomap_to_bh(inode, block, bh, iomap); | |
2007 | } | |
2008 | ||
1da177e4 | 2009 | if (buffer_new(bh)) { |
e64855c6 | 2010 | clean_bdev_bh_alias(bh); |
1da177e4 | 2011 | if (PageUptodate(page)) { |
637aff46 | 2012 | clear_buffer_new(bh); |
1da177e4 | 2013 | set_buffer_uptodate(bh); |
637aff46 | 2014 | mark_buffer_dirty(bh); |
1da177e4 LT |
2015 | continue; |
2016 | } | |
eebd2aa3 CL |
2017 | if (block_end > to || block_start < from) |
2018 | zero_user_segments(page, | |
2019 | to, block_end, | |
2020 | block_start, from); | |
1da177e4 LT |
2021 | continue; |
2022 | } | |
2023 | } | |
2024 | if (PageUptodate(page)) { | |
2025 | if (!buffer_uptodate(bh)) | |
2026 | set_buffer_uptodate(bh); | |
2027 | continue; | |
2028 | } | |
2029 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
33a266dd | 2030 | !buffer_unwritten(bh) && |
1da177e4 | 2031 | (block_start < from || block_end > to)) { |
dfec8a14 | 2032 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2033 | *wait_bh++=bh; |
2034 | } | |
2035 | } | |
2036 | /* | |
2037 | * If we issued read requests - let them complete. | |
2038 | */ | |
2039 | while(wait_bh > wait) { | |
2040 | wait_on_buffer(*--wait_bh); | |
2041 | if (!buffer_uptodate(*wait_bh)) | |
f3ddbdc6 | 2042 | err = -EIO; |
1da177e4 | 2043 | } |
f9f07b6c | 2044 | if (unlikely(err)) |
afddba49 | 2045 | page_zero_new_buffers(page, from, to); |
1da177e4 LT |
2046 | return err; |
2047 | } | |
ae259a9c CH |
2048 | |
2049 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, | |
2050 | get_block_t *get_block) | |
2051 | { | |
2052 | return __block_write_begin_int(page, pos, len, get_block, NULL); | |
2053 | } | |
ebdec241 | 2054 | EXPORT_SYMBOL(__block_write_begin); |
1da177e4 LT |
2055 | |
2056 | static int __block_commit_write(struct inode *inode, struct page *page, | |
2057 | unsigned from, unsigned to) | |
2058 | { | |
2059 | unsigned block_start, block_end; | |
2060 | int partial = 0; | |
2061 | unsigned blocksize; | |
2062 | struct buffer_head *bh, *head; | |
2063 | ||
45bce8f3 LT |
2064 | bh = head = page_buffers(page); |
2065 | blocksize = bh->b_size; | |
1da177e4 | 2066 | |
45bce8f3 LT |
2067 | block_start = 0; |
2068 | do { | |
1da177e4 LT |
2069 | block_end = block_start + blocksize; |
2070 | if (block_end <= from || block_start >= to) { | |
2071 | if (!buffer_uptodate(bh)) | |
2072 | partial = 1; | |
2073 | } else { | |
2074 | set_buffer_uptodate(bh); | |
2075 | mark_buffer_dirty(bh); | |
2076 | } | |
afddba49 | 2077 | clear_buffer_new(bh); |
45bce8f3 LT |
2078 | |
2079 | block_start = block_end; | |
2080 | bh = bh->b_this_page; | |
2081 | } while (bh != head); | |
1da177e4 LT |
2082 | |
2083 | /* | |
2084 | * If this is a partial write which happened to make all buffers | |
2085 | * uptodate then we can optimize away a bogus readpage() for | |
2086 | * the next read(). Here we 'discover' whether the page went | |
2087 | * uptodate as a result of this (potentially partial) write. | |
2088 | */ | |
2089 | if (!partial) | |
2090 | SetPageUptodate(page); | |
2091 | return 0; | |
2092 | } | |
2093 | ||
afddba49 | 2094 | /* |
155130a4 CH |
2095 | * block_write_begin takes care of the basic task of block allocation and |
2096 | * bringing partial write blocks uptodate first. | |
2097 | * | |
7bb46a67 | 2098 | * The filesystem needs to handle block truncation upon failure. |
afddba49 | 2099 | */ |
155130a4 CH |
2100 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
2101 | unsigned flags, struct page **pagep, get_block_t *get_block) | |
afddba49 | 2102 | { |
09cbfeaf | 2103 | pgoff_t index = pos >> PAGE_SHIFT; |
afddba49 | 2104 | struct page *page; |
6e1db88d | 2105 | int status; |
afddba49 | 2106 | |
6e1db88d CH |
2107 | page = grab_cache_page_write_begin(mapping, index, flags); |
2108 | if (!page) | |
2109 | return -ENOMEM; | |
afddba49 | 2110 | |
6e1db88d | 2111 | status = __block_write_begin(page, pos, len, get_block); |
afddba49 | 2112 | if (unlikely(status)) { |
6e1db88d | 2113 | unlock_page(page); |
09cbfeaf | 2114 | put_page(page); |
6e1db88d | 2115 | page = NULL; |
afddba49 NP |
2116 | } |
2117 | ||
6e1db88d | 2118 | *pagep = page; |
afddba49 NP |
2119 | return status; |
2120 | } | |
2121 | EXPORT_SYMBOL(block_write_begin); | |
2122 | ||
2123 | int block_write_end(struct file *file, struct address_space *mapping, | |
2124 | loff_t pos, unsigned len, unsigned copied, | |
2125 | struct page *page, void *fsdata) | |
2126 | { | |
2127 | struct inode *inode = mapping->host; | |
2128 | unsigned start; | |
2129 | ||
09cbfeaf | 2130 | start = pos & (PAGE_SIZE - 1); |
afddba49 NP |
2131 | |
2132 | if (unlikely(copied < len)) { | |
2133 | /* | |
2134 | * The buffers that were written will now be uptodate, so we | |
2135 | * don't have to worry about a readpage reading them and | |
2136 | * overwriting a partial write. However if we have encountered | |
2137 | * a short write and only partially written into a buffer, it | |
2138 | * will not be marked uptodate, so a readpage might come in and | |
2139 | * destroy our partial write. | |
2140 | * | |
2141 | * Do the simplest thing, and just treat any short write to a | |
2142 | * non uptodate page as a zero-length write, and force the | |
2143 | * caller to redo the whole thing. | |
2144 | */ | |
2145 | if (!PageUptodate(page)) | |
2146 | copied = 0; | |
2147 | ||
2148 | page_zero_new_buffers(page, start+copied, start+len); | |
2149 | } | |
2150 | flush_dcache_page(page); | |
2151 | ||
2152 | /* This could be a short (even 0-length) commit */ | |
2153 | __block_commit_write(inode, page, start, start+copied); | |
2154 | ||
2155 | return copied; | |
2156 | } | |
2157 | EXPORT_SYMBOL(block_write_end); | |
2158 | ||
2159 | int generic_write_end(struct file *file, struct address_space *mapping, | |
2160 | loff_t pos, unsigned len, unsigned copied, | |
2161 | struct page *page, void *fsdata) | |
2162 | { | |
8af54f29 CH |
2163 | struct inode *inode = mapping->host; |
2164 | loff_t old_size = inode->i_size; | |
2165 | bool i_size_changed = false; | |
2166 | ||
afddba49 | 2167 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
8af54f29 CH |
2168 | |
2169 | /* | |
2170 | * No need to use i_size_read() here, the i_size cannot change under us | |
2171 | * because we hold i_rwsem. | |
2172 | * | |
2173 | * But it's important to update i_size while still holding page lock: | |
2174 | * page writeout could otherwise come in and zero beyond i_size. | |
2175 | */ | |
2176 | if (pos + copied > inode->i_size) { | |
2177 | i_size_write(inode, pos + copied); | |
2178 | i_size_changed = true; | |
2179 | } | |
2180 | ||
2181 | unlock_page(page); | |
7a77dad7 | 2182 | put_page(page); |
8af54f29 CH |
2183 | |
2184 | if (old_size < pos) | |
2185 | pagecache_isize_extended(inode, old_size, pos); | |
2186 | /* | |
2187 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
2188 | * makes the holding time of page lock longer. Second, it forces lock | |
2189 | * ordering of page lock and transaction start for journaling | |
2190 | * filesystems. | |
2191 | */ | |
2192 | if (i_size_changed) | |
2193 | mark_inode_dirty(inode); | |
26ddb1f4 | 2194 | return copied; |
afddba49 NP |
2195 | } |
2196 | EXPORT_SYMBOL(generic_write_end); | |
2197 | ||
8ab22b9a HH |
2198 | /* |
2199 | * block_is_partially_uptodate checks whether buffers within a page are | |
2200 | * uptodate or not. | |
2201 | * | |
2202 | * Returns true if all buffers which correspond to a file portion | |
2203 | * we want to read are uptodate. | |
2204 | */ | |
c186afb4 AV |
2205 | int block_is_partially_uptodate(struct page *page, unsigned long from, |
2206 | unsigned long count) | |
8ab22b9a | 2207 | { |
8ab22b9a HH |
2208 | unsigned block_start, block_end, blocksize; |
2209 | unsigned to; | |
2210 | struct buffer_head *bh, *head; | |
2211 | int ret = 1; | |
2212 | ||
2213 | if (!page_has_buffers(page)) | |
2214 | return 0; | |
2215 | ||
45bce8f3 LT |
2216 | head = page_buffers(page); |
2217 | blocksize = head->b_size; | |
09cbfeaf | 2218 | to = min_t(unsigned, PAGE_SIZE - from, count); |
8ab22b9a | 2219 | to = from + to; |
09cbfeaf | 2220 | if (from < blocksize && to > PAGE_SIZE - blocksize) |
8ab22b9a HH |
2221 | return 0; |
2222 | ||
8ab22b9a HH |
2223 | bh = head; |
2224 | block_start = 0; | |
2225 | do { | |
2226 | block_end = block_start + blocksize; | |
2227 | if (block_end > from && block_start < to) { | |
2228 | if (!buffer_uptodate(bh)) { | |
2229 | ret = 0; | |
2230 | break; | |
2231 | } | |
2232 | if (block_end >= to) | |
2233 | break; | |
2234 | } | |
2235 | block_start = block_end; | |
2236 | bh = bh->b_this_page; | |
2237 | } while (bh != head); | |
2238 | ||
2239 | return ret; | |
2240 | } | |
2241 | EXPORT_SYMBOL(block_is_partially_uptodate); | |
2242 | ||
1da177e4 LT |
2243 | /* |
2244 | * Generic "read page" function for block devices that have the normal | |
2245 | * get_block functionality. This is most of the block device filesystems. | |
2246 | * Reads the page asynchronously --- the unlock_buffer() and | |
2247 | * set/clear_buffer_uptodate() functions propagate buffer state into the | |
2248 | * page struct once IO has completed. | |
2249 | */ | |
2250 | int block_read_full_page(struct page *page, get_block_t *get_block) | |
2251 | { | |
2252 | struct inode *inode = page->mapping->host; | |
2253 | sector_t iblock, lblock; | |
2254 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; | |
45bce8f3 | 2255 | unsigned int blocksize, bbits; |
1da177e4 LT |
2256 | int nr, i; |
2257 | int fully_mapped = 1; | |
2258 | ||
45bce8f3 LT |
2259 | head = create_page_buffers(page, inode, 0); |
2260 | blocksize = head->b_size; | |
2261 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2262 | |
09cbfeaf | 2263 | iblock = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 2264 | lblock = (i_size_read(inode)+blocksize-1) >> bbits; |
1da177e4 LT |
2265 | bh = head; |
2266 | nr = 0; | |
2267 | i = 0; | |
2268 | ||
2269 | do { | |
2270 | if (buffer_uptodate(bh)) | |
2271 | continue; | |
2272 | ||
2273 | if (!buffer_mapped(bh)) { | |
c64610ba AM |
2274 | int err = 0; |
2275 | ||
1da177e4 LT |
2276 | fully_mapped = 0; |
2277 | if (iblock < lblock) { | |
b0cf2321 | 2278 | WARN_ON(bh->b_size != blocksize); |
c64610ba AM |
2279 | err = get_block(inode, iblock, bh, 0); |
2280 | if (err) | |
1da177e4 LT |
2281 | SetPageError(page); |
2282 | } | |
2283 | if (!buffer_mapped(bh)) { | |
eebd2aa3 | 2284 | zero_user(page, i * blocksize, blocksize); |
c64610ba AM |
2285 | if (!err) |
2286 | set_buffer_uptodate(bh); | |
1da177e4 LT |
2287 | continue; |
2288 | } | |
2289 | /* | |
2290 | * get_block() might have updated the buffer | |
2291 | * synchronously | |
2292 | */ | |
2293 | if (buffer_uptodate(bh)) | |
2294 | continue; | |
2295 | } | |
2296 | arr[nr++] = bh; | |
2297 | } while (i++, iblock++, (bh = bh->b_this_page) != head); | |
2298 | ||
2299 | if (fully_mapped) | |
2300 | SetPageMappedToDisk(page); | |
2301 | ||
2302 | if (!nr) { | |
2303 | /* | |
2304 | * All buffers are uptodate - we can set the page uptodate | |
2305 | * as well. But not if get_block() returned an error. | |
2306 | */ | |
2307 | if (!PageError(page)) | |
2308 | SetPageUptodate(page); | |
2309 | unlock_page(page); | |
2310 | return 0; | |
2311 | } | |
2312 | ||
2313 | /* Stage two: lock the buffers */ | |
2314 | for (i = 0; i < nr; i++) { | |
2315 | bh = arr[i]; | |
2316 | lock_buffer(bh); | |
2317 | mark_buffer_async_read(bh); | |
2318 | } | |
2319 | ||
2320 | /* | |
2321 | * Stage 3: start the IO. Check for uptodateness | |
2322 | * inside the buffer lock in case another process reading | |
2323 | * the underlying blockdev brought it uptodate (the sct fix). | |
2324 | */ | |
2325 | for (i = 0; i < nr; i++) { | |
2326 | bh = arr[i]; | |
2327 | if (buffer_uptodate(bh)) | |
2328 | end_buffer_async_read(bh, 1); | |
2329 | else | |
2a222ca9 | 2330 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
2331 | } |
2332 | return 0; | |
2333 | } | |
1fe72eaa | 2334 | EXPORT_SYMBOL(block_read_full_page); |
1da177e4 LT |
2335 | |
2336 | /* utility function for filesystems that need to do work on expanding | |
89e10787 | 2337 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
1da177e4 LT |
2338 | * deal with the hole. |
2339 | */ | |
89e10787 | 2340 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
1da177e4 LT |
2341 | { |
2342 | struct address_space *mapping = inode->i_mapping; | |
2343 | struct page *page; | |
89e10787 | 2344 | void *fsdata; |
1da177e4 LT |
2345 | int err; |
2346 | ||
c08d3b0e NP |
2347 | err = inode_newsize_ok(inode, size); |
2348 | if (err) | |
1da177e4 LT |
2349 | goto out; |
2350 | ||
89e10787 | 2351 | err = pagecache_write_begin(NULL, mapping, size, 0, |
c718a975 | 2352 | AOP_FLAG_CONT_EXPAND, &page, &fsdata); |
89e10787 | 2353 | if (err) |
05eb0b51 | 2354 | goto out; |
05eb0b51 | 2355 | |
89e10787 NP |
2356 | err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); |
2357 | BUG_ON(err > 0); | |
05eb0b51 | 2358 | |
1da177e4 LT |
2359 | out: |
2360 | return err; | |
2361 | } | |
1fe72eaa | 2362 | EXPORT_SYMBOL(generic_cont_expand_simple); |
1da177e4 | 2363 | |
f1e3af72 AB |
2364 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
2365 | loff_t pos, loff_t *bytes) | |
1da177e4 | 2366 | { |
1da177e4 | 2367 | struct inode *inode = mapping->host; |
93407472 | 2368 | unsigned int blocksize = i_blocksize(inode); |
89e10787 NP |
2369 | struct page *page; |
2370 | void *fsdata; | |
2371 | pgoff_t index, curidx; | |
2372 | loff_t curpos; | |
2373 | unsigned zerofrom, offset, len; | |
2374 | int err = 0; | |
1da177e4 | 2375 | |
09cbfeaf KS |
2376 | index = pos >> PAGE_SHIFT; |
2377 | offset = pos & ~PAGE_MASK; | |
89e10787 | 2378 | |
09cbfeaf KS |
2379 | while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { |
2380 | zerofrom = curpos & ~PAGE_MASK; | |
1da177e4 LT |
2381 | if (zerofrom & (blocksize-1)) { |
2382 | *bytes |= (blocksize-1); | |
2383 | (*bytes)++; | |
2384 | } | |
09cbfeaf | 2385 | len = PAGE_SIZE - zerofrom; |
1da177e4 | 2386 | |
c718a975 TH |
2387 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2388 | &page, &fsdata); | |
89e10787 NP |
2389 | if (err) |
2390 | goto out; | |
eebd2aa3 | 2391 | zero_user(page, zerofrom, len); |
89e10787 NP |
2392 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2393 | page, fsdata); | |
2394 | if (err < 0) | |
2395 | goto out; | |
2396 | BUG_ON(err != len); | |
2397 | err = 0; | |
061e9746 OH |
2398 | |
2399 | balance_dirty_pages_ratelimited(mapping); | |
c2ca0fcd | 2400 | |
08d405c8 | 2401 | if (fatal_signal_pending(current)) { |
c2ca0fcd MP |
2402 | err = -EINTR; |
2403 | goto out; | |
2404 | } | |
89e10787 | 2405 | } |
1da177e4 | 2406 | |
89e10787 NP |
2407 | /* page covers the boundary, find the boundary offset */ |
2408 | if (index == curidx) { | |
09cbfeaf | 2409 | zerofrom = curpos & ~PAGE_MASK; |
1da177e4 | 2410 | /* if we will expand the thing last block will be filled */ |
89e10787 NP |
2411 | if (offset <= zerofrom) { |
2412 | goto out; | |
2413 | } | |
2414 | if (zerofrom & (blocksize-1)) { | |
1da177e4 LT |
2415 | *bytes |= (blocksize-1); |
2416 | (*bytes)++; | |
2417 | } | |
89e10787 | 2418 | len = offset - zerofrom; |
1da177e4 | 2419 | |
c718a975 TH |
2420 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2421 | &page, &fsdata); | |
89e10787 NP |
2422 | if (err) |
2423 | goto out; | |
eebd2aa3 | 2424 | zero_user(page, zerofrom, len); |
89e10787 NP |
2425 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2426 | page, fsdata); | |
2427 | if (err < 0) | |
2428 | goto out; | |
2429 | BUG_ON(err != len); | |
2430 | err = 0; | |
1da177e4 | 2431 | } |
89e10787 NP |
2432 | out: |
2433 | return err; | |
2434 | } | |
2435 | ||
2436 | /* | |
2437 | * For moronic filesystems that do not allow holes in file. | |
2438 | * We may have to extend the file. | |
2439 | */ | |
282dc178 | 2440 | int cont_write_begin(struct file *file, struct address_space *mapping, |
89e10787 NP |
2441 | loff_t pos, unsigned len, unsigned flags, |
2442 | struct page **pagep, void **fsdata, | |
2443 | get_block_t *get_block, loff_t *bytes) | |
2444 | { | |
2445 | struct inode *inode = mapping->host; | |
93407472 FF |
2446 | unsigned int blocksize = i_blocksize(inode); |
2447 | unsigned int zerofrom; | |
89e10787 NP |
2448 | int err; |
2449 | ||
2450 | err = cont_expand_zero(file, mapping, pos, bytes); | |
2451 | if (err) | |
155130a4 | 2452 | return err; |
89e10787 | 2453 | |
09cbfeaf | 2454 | zerofrom = *bytes & ~PAGE_MASK; |
89e10787 NP |
2455 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { |
2456 | *bytes |= (blocksize-1); | |
2457 | (*bytes)++; | |
1da177e4 | 2458 | } |
1da177e4 | 2459 | |
155130a4 | 2460 | return block_write_begin(mapping, pos, len, flags, pagep, get_block); |
1da177e4 | 2461 | } |
1fe72eaa | 2462 | EXPORT_SYMBOL(cont_write_begin); |
1da177e4 | 2463 | |
1da177e4 LT |
2464 | int block_commit_write(struct page *page, unsigned from, unsigned to) |
2465 | { | |
2466 | struct inode *inode = page->mapping->host; | |
2467 | __block_commit_write(inode,page,from,to); | |
2468 | return 0; | |
2469 | } | |
1fe72eaa | 2470 | EXPORT_SYMBOL(block_commit_write); |
1da177e4 | 2471 | |
54171690 DC |
2472 | /* |
2473 | * block_page_mkwrite() is not allowed to change the file size as it gets | |
2474 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2475 | * be careful to check for EOF conditions here. We set the page up correctly | |
2476 | * for a written page which means we get ENOSPC checking when writing into | |
2477 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2478 | * support these features. | |
2479 | * | |
2480 | * We are not allowed to take the i_mutex here so we have to play games to | |
2481 | * protect against truncate races as the page could now be beyond EOF. Because | |
7bb46a67 | 2482 | * truncate writes the inode size before removing pages, once we have the |
54171690 DC |
2483 | * page lock we can determine safely if the page is beyond EOF. If it is not |
2484 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2485 | * unlock the page. | |
ea13a864 | 2486 | * |
14da9200 | 2487 | * Direct callers of this function should protect against filesystem freezing |
5c500029 | 2488 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
54171690 | 2489 | */ |
5c500029 | 2490 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
24da4fab | 2491 | get_block_t get_block) |
54171690 | 2492 | { |
c2ec175c | 2493 | struct page *page = vmf->page; |
496ad9aa | 2494 | struct inode *inode = file_inode(vma->vm_file); |
54171690 DC |
2495 | unsigned long end; |
2496 | loff_t size; | |
24da4fab | 2497 | int ret; |
54171690 DC |
2498 | |
2499 | lock_page(page); | |
2500 | size = i_size_read(inode); | |
2501 | if ((page->mapping != inode->i_mapping) || | |
18336338 | 2502 | (page_offset(page) > size)) { |
24da4fab JK |
2503 | /* We overload EFAULT to mean page got truncated */ |
2504 | ret = -EFAULT; | |
2505 | goto out_unlock; | |
54171690 DC |
2506 | } |
2507 | ||
2508 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
2509 | if (((page->index + 1) << PAGE_SHIFT) > size) |
2510 | end = size & ~PAGE_MASK; | |
54171690 | 2511 | else |
09cbfeaf | 2512 | end = PAGE_SIZE; |
54171690 | 2513 | |
ebdec241 | 2514 | ret = __block_write_begin(page, 0, end, get_block); |
54171690 DC |
2515 | if (!ret) |
2516 | ret = block_commit_write(page, 0, end); | |
2517 | ||
24da4fab JK |
2518 | if (unlikely(ret < 0)) |
2519 | goto out_unlock; | |
ea13a864 | 2520 | set_page_dirty(page); |
1d1d1a76 | 2521 | wait_for_stable_page(page); |
24da4fab JK |
2522 | return 0; |
2523 | out_unlock: | |
2524 | unlock_page(page); | |
54171690 | 2525 | return ret; |
24da4fab | 2526 | } |
1fe72eaa | 2527 | EXPORT_SYMBOL(block_page_mkwrite); |
1da177e4 LT |
2528 | |
2529 | /* | |
03158cd7 | 2530 | * nobh_write_begin()'s prereads are special: the buffer_heads are freed |
1da177e4 LT |
2531 | * immediately, while under the page lock. So it needs a special end_io |
2532 | * handler which does not touch the bh after unlocking it. | |
1da177e4 LT |
2533 | */ |
2534 | static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) | |
2535 | { | |
68671f35 | 2536 | __end_buffer_read_notouch(bh, uptodate); |
1da177e4 LT |
2537 | } |
2538 | ||
03158cd7 NP |
2539 | /* |
2540 | * Attach the singly-linked list of buffers created by nobh_write_begin, to | |
2541 | * the page (converting it to circular linked list and taking care of page | |
2542 | * dirty races). | |
2543 | */ | |
2544 | static void attach_nobh_buffers(struct page *page, struct buffer_head *head) | |
2545 | { | |
2546 | struct buffer_head *bh; | |
2547 | ||
2548 | BUG_ON(!PageLocked(page)); | |
2549 | ||
2550 | spin_lock(&page->mapping->private_lock); | |
2551 | bh = head; | |
2552 | do { | |
2553 | if (PageDirty(page)) | |
2554 | set_buffer_dirty(bh); | |
2555 | if (!bh->b_this_page) | |
2556 | bh->b_this_page = head; | |
2557 | bh = bh->b_this_page; | |
2558 | } while (bh != head); | |
2559 | attach_page_buffers(page, head); | |
2560 | spin_unlock(&page->mapping->private_lock); | |
2561 | } | |
2562 | ||
1da177e4 | 2563 | /* |
ea0f04e5 CH |
2564 | * On entry, the page is fully not uptodate. |
2565 | * On exit the page is fully uptodate in the areas outside (from,to) | |
7bb46a67 | 2566 | * The filesystem needs to handle block truncation upon failure. |
1da177e4 | 2567 | */ |
ea0f04e5 | 2568 | int nobh_write_begin(struct address_space *mapping, |
03158cd7 NP |
2569 | loff_t pos, unsigned len, unsigned flags, |
2570 | struct page **pagep, void **fsdata, | |
1da177e4 LT |
2571 | get_block_t *get_block) |
2572 | { | |
03158cd7 | 2573 | struct inode *inode = mapping->host; |
1da177e4 LT |
2574 | const unsigned blkbits = inode->i_blkbits; |
2575 | const unsigned blocksize = 1 << blkbits; | |
a4b0672d | 2576 | struct buffer_head *head, *bh; |
03158cd7 NP |
2577 | struct page *page; |
2578 | pgoff_t index; | |
2579 | unsigned from, to; | |
1da177e4 | 2580 | unsigned block_in_page; |
a4b0672d | 2581 | unsigned block_start, block_end; |
1da177e4 | 2582 | sector_t block_in_file; |
1da177e4 | 2583 | int nr_reads = 0; |
1da177e4 LT |
2584 | int ret = 0; |
2585 | int is_mapped_to_disk = 1; | |
1da177e4 | 2586 | |
09cbfeaf KS |
2587 | index = pos >> PAGE_SHIFT; |
2588 | from = pos & (PAGE_SIZE - 1); | |
03158cd7 NP |
2589 | to = from + len; |
2590 | ||
54566b2c | 2591 | page = grab_cache_page_write_begin(mapping, index, flags); |
03158cd7 NP |
2592 | if (!page) |
2593 | return -ENOMEM; | |
2594 | *pagep = page; | |
2595 | *fsdata = NULL; | |
2596 | ||
2597 | if (page_has_buffers(page)) { | |
309f77ad NK |
2598 | ret = __block_write_begin(page, pos, len, get_block); |
2599 | if (unlikely(ret)) | |
2600 | goto out_release; | |
2601 | return ret; | |
03158cd7 | 2602 | } |
a4b0672d | 2603 | |
1da177e4 LT |
2604 | if (PageMappedToDisk(page)) |
2605 | return 0; | |
2606 | ||
a4b0672d NP |
2607 | /* |
2608 | * Allocate buffers so that we can keep track of state, and potentially | |
2609 | * attach them to the page if an error occurs. In the common case of | |
2610 | * no error, they will just be freed again without ever being attached | |
2611 | * to the page (which is all OK, because we're under the page lock). | |
2612 | * | |
2613 | * Be careful: the buffer linked list is a NULL terminated one, rather | |
2614 | * than the circular one we're used to. | |
2615 | */ | |
640ab98f | 2616 | head = alloc_page_buffers(page, blocksize, false); |
03158cd7 NP |
2617 | if (!head) { |
2618 | ret = -ENOMEM; | |
2619 | goto out_release; | |
2620 | } | |
a4b0672d | 2621 | |
09cbfeaf | 2622 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
1da177e4 LT |
2623 | |
2624 | /* | |
2625 | * We loop across all blocks in the page, whether or not they are | |
2626 | * part of the affected region. This is so we can discover if the | |
2627 | * page is fully mapped-to-disk. | |
2628 | */ | |
a4b0672d | 2629 | for (block_start = 0, block_in_page = 0, bh = head; |
09cbfeaf | 2630 | block_start < PAGE_SIZE; |
a4b0672d | 2631 | block_in_page++, block_start += blocksize, bh = bh->b_this_page) { |
1da177e4 LT |
2632 | int create; |
2633 | ||
a4b0672d NP |
2634 | block_end = block_start + blocksize; |
2635 | bh->b_state = 0; | |
1da177e4 LT |
2636 | create = 1; |
2637 | if (block_start >= to) | |
2638 | create = 0; | |
2639 | ret = get_block(inode, block_in_file + block_in_page, | |
a4b0672d | 2640 | bh, create); |
1da177e4 LT |
2641 | if (ret) |
2642 | goto failed; | |
a4b0672d | 2643 | if (!buffer_mapped(bh)) |
1da177e4 | 2644 | is_mapped_to_disk = 0; |
a4b0672d | 2645 | if (buffer_new(bh)) |
e64855c6 | 2646 | clean_bdev_bh_alias(bh); |
a4b0672d NP |
2647 | if (PageUptodate(page)) { |
2648 | set_buffer_uptodate(bh); | |
1da177e4 | 2649 | continue; |
a4b0672d NP |
2650 | } |
2651 | if (buffer_new(bh) || !buffer_mapped(bh)) { | |
eebd2aa3 CL |
2652 | zero_user_segments(page, block_start, from, |
2653 | to, block_end); | |
1da177e4 LT |
2654 | continue; |
2655 | } | |
a4b0672d | 2656 | if (buffer_uptodate(bh)) |
1da177e4 LT |
2657 | continue; /* reiserfs does this */ |
2658 | if (block_start < from || block_end > to) { | |
a4b0672d NP |
2659 | lock_buffer(bh); |
2660 | bh->b_end_io = end_buffer_read_nobh; | |
2a222ca9 | 2661 | submit_bh(REQ_OP_READ, 0, bh); |
a4b0672d | 2662 | nr_reads++; |
1da177e4 LT |
2663 | } |
2664 | } | |
2665 | ||
2666 | if (nr_reads) { | |
1da177e4 LT |
2667 | /* |
2668 | * The page is locked, so these buffers are protected from | |
2669 | * any VM or truncate activity. Hence we don't need to care | |
2670 | * for the buffer_head refcounts. | |
2671 | */ | |
a4b0672d | 2672 | for (bh = head; bh; bh = bh->b_this_page) { |
1da177e4 LT |
2673 | wait_on_buffer(bh); |
2674 | if (!buffer_uptodate(bh)) | |
2675 | ret = -EIO; | |
1da177e4 LT |
2676 | } |
2677 | if (ret) | |
2678 | goto failed; | |
2679 | } | |
2680 | ||
2681 | if (is_mapped_to_disk) | |
2682 | SetPageMappedToDisk(page); | |
1da177e4 | 2683 | |
03158cd7 | 2684 | *fsdata = head; /* to be released by nobh_write_end */ |
a4b0672d | 2685 | |
1da177e4 LT |
2686 | return 0; |
2687 | ||
2688 | failed: | |
03158cd7 | 2689 | BUG_ON(!ret); |
1da177e4 | 2690 | /* |
a4b0672d NP |
2691 | * Error recovery is a bit difficult. We need to zero out blocks that |
2692 | * were newly allocated, and dirty them to ensure they get written out. | |
2693 | * Buffers need to be attached to the page at this point, otherwise | |
2694 | * the handling of potential IO errors during writeout would be hard | |
2695 | * (could try doing synchronous writeout, but what if that fails too?) | |
1da177e4 | 2696 | */ |
03158cd7 NP |
2697 | attach_nobh_buffers(page, head); |
2698 | page_zero_new_buffers(page, from, to); | |
a4b0672d | 2699 | |
03158cd7 NP |
2700 | out_release: |
2701 | unlock_page(page); | |
09cbfeaf | 2702 | put_page(page); |
03158cd7 | 2703 | *pagep = NULL; |
a4b0672d | 2704 | |
7bb46a67 NP |
2705 | return ret; |
2706 | } | |
03158cd7 | 2707 | EXPORT_SYMBOL(nobh_write_begin); |
1da177e4 | 2708 | |
03158cd7 NP |
2709 | int nobh_write_end(struct file *file, struct address_space *mapping, |
2710 | loff_t pos, unsigned len, unsigned copied, | |
2711 | struct page *page, void *fsdata) | |
1da177e4 LT |
2712 | { |
2713 | struct inode *inode = page->mapping->host; | |
efdc3131 | 2714 | struct buffer_head *head = fsdata; |
03158cd7 | 2715 | struct buffer_head *bh; |
5b41e74a | 2716 | BUG_ON(fsdata != NULL && page_has_buffers(page)); |
1da177e4 | 2717 | |
d4cf109f | 2718 | if (unlikely(copied < len) && head) |
5b41e74a DM |
2719 | attach_nobh_buffers(page, head); |
2720 | if (page_has_buffers(page)) | |
2721 | return generic_write_end(file, mapping, pos, len, | |
2722 | copied, page, fsdata); | |
a4b0672d | 2723 | |
22c8ca78 | 2724 | SetPageUptodate(page); |
1da177e4 | 2725 | set_page_dirty(page); |
03158cd7 NP |
2726 | if (pos+copied > inode->i_size) { |
2727 | i_size_write(inode, pos+copied); | |
1da177e4 LT |
2728 | mark_inode_dirty(inode); |
2729 | } | |
03158cd7 NP |
2730 | |
2731 | unlock_page(page); | |
09cbfeaf | 2732 | put_page(page); |
03158cd7 | 2733 | |
03158cd7 NP |
2734 | while (head) { |
2735 | bh = head; | |
2736 | head = head->b_this_page; | |
2737 | free_buffer_head(bh); | |
2738 | } | |
2739 | ||
2740 | return copied; | |
1da177e4 | 2741 | } |
03158cd7 | 2742 | EXPORT_SYMBOL(nobh_write_end); |
1da177e4 LT |
2743 | |
2744 | /* | |
2745 | * nobh_writepage() - based on block_full_write_page() except | |
2746 | * that it tries to operate without attaching bufferheads to | |
2747 | * the page. | |
2748 | */ | |
2749 | int nobh_writepage(struct page *page, get_block_t *get_block, | |
2750 | struct writeback_control *wbc) | |
2751 | { | |
2752 | struct inode * const inode = page->mapping->host; | |
2753 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2754 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2755 | unsigned offset; |
1da177e4 LT |
2756 | int ret; |
2757 | ||
2758 | /* Is the page fully inside i_size? */ | |
2759 | if (page->index < end_index) | |
2760 | goto out; | |
2761 | ||
2762 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2763 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 LT |
2764 | if (page->index >= end_index+1 || !offset) { |
2765 | /* | |
2766 | * The page may have dirty, unmapped buffers. For example, | |
2767 | * they may have been added in ext3_writepage(). Make them | |
2768 | * freeable here, so the page does not leak. | |
2769 | */ | |
2770 | #if 0 | |
2771 | /* Not really sure about this - do we need this ? */ | |
2772 | if (page->mapping->a_ops->invalidatepage) | |
2773 | page->mapping->a_ops->invalidatepage(page, offset); | |
2774 | #endif | |
2775 | unlock_page(page); | |
2776 | return 0; /* don't care */ | |
2777 | } | |
2778 | ||
2779 | /* | |
2780 | * The page straddles i_size. It must be zeroed out on each and every | |
2781 | * writepage invocation because it may be mmapped. "A file is mapped | |
2782 | * in multiples of the page size. For a file that is not a multiple of | |
2783 | * the page size, the remaining memory is zeroed when mapped, and | |
2784 | * writes to that region are not written out to the file." | |
2785 | */ | |
09cbfeaf | 2786 | zero_user_segment(page, offset, PAGE_SIZE); |
1da177e4 LT |
2787 | out: |
2788 | ret = mpage_writepage(page, get_block, wbc); | |
2789 | if (ret == -EAGAIN) | |
35c80d5f CM |
2790 | ret = __block_write_full_page(inode, page, get_block, wbc, |
2791 | end_buffer_async_write); | |
1da177e4 LT |
2792 | return ret; |
2793 | } | |
2794 | EXPORT_SYMBOL(nobh_writepage); | |
2795 | ||
03158cd7 NP |
2796 | int nobh_truncate_page(struct address_space *mapping, |
2797 | loff_t from, get_block_t *get_block) | |
1da177e4 | 2798 | { |
09cbfeaf KS |
2799 | pgoff_t index = from >> PAGE_SHIFT; |
2800 | unsigned offset = from & (PAGE_SIZE-1); | |
03158cd7 NP |
2801 | unsigned blocksize; |
2802 | sector_t iblock; | |
2803 | unsigned length, pos; | |
2804 | struct inode *inode = mapping->host; | |
1da177e4 | 2805 | struct page *page; |
03158cd7 NP |
2806 | struct buffer_head map_bh; |
2807 | int err; | |
1da177e4 | 2808 | |
93407472 | 2809 | blocksize = i_blocksize(inode); |
03158cd7 NP |
2810 | length = offset & (blocksize - 1); |
2811 | ||
2812 | /* Block boundary? Nothing to do */ | |
2813 | if (!length) | |
2814 | return 0; | |
2815 | ||
2816 | length = blocksize - length; | |
09cbfeaf | 2817 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 | 2818 | |
1da177e4 | 2819 | page = grab_cache_page(mapping, index); |
03158cd7 | 2820 | err = -ENOMEM; |
1da177e4 LT |
2821 | if (!page) |
2822 | goto out; | |
2823 | ||
03158cd7 NP |
2824 | if (page_has_buffers(page)) { |
2825 | has_buffers: | |
2826 | unlock_page(page); | |
09cbfeaf | 2827 | put_page(page); |
03158cd7 NP |
2828 | return block_truncate_page(mapping, from, get_block); |
2829 | } | |
2830 | ||
2831 | /* Find the buffer that contains "offset" */ | |
2832 | pos = blocksize; | |
2833 | while (offset >= pos) { | |
2834 | iblock++; | |
2835 | pos += blocksize; | |
2836 | } | |
2837 | ||
460bcf57 TT |
2838 | map_bh.b_size = blocksize; |
2839 | map_bh.b_state = 0; | |
03158cd7 NP |
2840 | err = get_block(inode, iblock, &map_bh, 0); |
2841 | if (err) | |
2842 | goto unlock; | |
2843 | /* unmapped? It's a hole - nothing to do */ | |
2844 | if (!buffer_mapped(&map_bh)) | |
2845 | goto unlock; | |
2846 | ||
2847 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2848 | if (!PageUptodate(page)) { | |
2849 | err = mapping->a_ops->readpage(NULL, page); | |
2850 | if (err) { | |
09cbfeaf | 2851 | put_page(page); |
03158cd7 NP |
2852 | goto out; |
2853 | } | |
2854 | lock_page(page); | |
2855 | if (!PageUptodate(page)) { | |
2856 | err = -EIO; | |
2857 | goto unlock; | |
2858 | } | |
2859 | if (page_has_buffers(page)) | |
2860 | goto has_buffers; | |
1da177e4 | 2861 | } |
eebd2aa3 | 2862 | zero_user(page, offset, length); |
03158cd7 NP |
2863 | set_page_dirty(page); |
2864 | err = 0; | |
2865 | ||
2866 | unlock: | |
1da177e4 | 2867 | unlock_page(page); |
09cbfeaf | 2868 | put_page(page); |
1da177e4 | 2869 | out: |
03158cd7 | 2870 | return err; |
1da177e4 LT |
2871 | } |
2872 | EXPORT_SYMBOL(nobh_truncate_page); | |
2873 | ||
2874 | int block_truncate_page(struct address_space *mapping, | |
2875 | loff_t from, get_block_t *get_block) | |
2876 | { | |
09cbfeaf KS |
2877 | pgoff_t index = from >> PAGE_SHIFT; |
2878 | unsigned offset = from & (PAGE_SIZE-1); | |
1da177e4 | 2879 | unsigned blocksize; |
54b21a79 | 2880 | sector_t iblock; |
1da177e4 LT |
2881 | unsigned length, pos; |
2882 | struct inode *inode = mapping->host; | |
2883 | struct page *page; | |
2884 | struct buffer_head *bh; | |
1da177e4 LT |
2885 | int err; |
2886 | ||
93407472 | 2887 | blocksize = i_blocksize(inode); |
1da177e4 LT |
2888 | length = offset & (blocksize - 1); |
2889 | ||
2890 | /* Block boundary? Nothing to do */ | |
2891 | if (!length) | |
2892 | return 0; | |
2893 | ||
2894 | length = blocksize - length; | |
09cbfeaf | 2895 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 LT |
2896 | |
2897 | page = grab_cache_page(mapping, index); | |
2898 | err = -ENOMEM; | |
2899 | if (!page) | |
2900 | goto out; | |
2901 | ||
2902 | if (!page_has_buffers(page)) | |
2903 | create_empty_buffers(page, blocksize, 0); | |
2904 | ||
2905 | /* Find the buffer that contains "offset" */ | |
2906 | bh = page_buffers(page); | |
2907 | pos = blocksize; | |
2908 | while (offset >= pos) { | |
2909 | bh = bh->b_this_page; | |
2910 | iblock++; | |
2911 | pos += blocksize; | |
2912 | } | |
2913 | ||
2914 | err = 0; | |
2915 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2916 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
2917 | err = get_block(inode, iblock, bh, 0); |
2918 | if (err) | |
2919 | goto unlock; | |
2920 | /* unmapped? It's a hole - nothing to do */ | |
2921 | if (!buffer_mapped(bh)) | |
2922 | goto unlock; | |
2923 | } | |
2924 | ||
2925 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2926 | if (PageUptodate(page)) | |
2927 | set_buffer_uptodate(bh); | |
2928 | ||
33a266dd | 2929 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
1da177e4 | 2930 | err = -EIO; |
dfec8a14 | 2931 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2932 | wait_on_buffer(bh); |
2933 | /* Uhhuh. Read error. Complain and punt. */ | |
2934 | if (!buffer_uptodate(bh)) | |
2935 | goto unlock; | |
2936 | } | |
2937 | ||
eebd2aa3 | 2938 | zero_user(page, offset, length); |
1da177e4 LT |
2939 | mark_buffer_dirty(bh); |
2940 | err = 0; | |
2941 | ||
2942 | unlock: | |
2943 | unlock_page(page); | |
09cbfeaf | 2944 | put_page(page); |
1da177e4 LT |
2945 | out: |
2946 | return err; | |
2947 | } | |
1fe72eaa | 2948 | EXPORT_SYMBOL(block_truncate_page); |
1da177e4 LT |
2949 | |
2950 | /* | |
2951 | * The generic ->writepage function for buffer-backed address_spaces | |
2952 | */ | |
1b938c08 MW |
2953 | int block_write_full_page(struct page *page, get_block_t *get_block, |
2954 | struct writeback_control *wbc) | |
1da177e4 LT |
2955 | { |
2956 | struct inode * const inode = page->mapping->host; | |
2957 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2958 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2959 | unsigned offset; |
1da177e4 LT |
2960 | |
2961 | /* Is the page fully inside i_size? */ | |
2962 | if (page->index < end_index) | |
35c80d5f | 2963 | return __block_write_full_page(inode, page, get_block, wbc, |
1b938c08 | 2964 | end_buffer_async_write); |
1da177e4 LT |
2965 | |
2966 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2967 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 LT |
2968 | if (page->index >= end_index+1 || !offset) { |
2969 | /* | |
2970 | * The page may have dirty, unmapped buffers. For example, | |
2971 | * they may have been added in ext3_writepage(). Make them | |
2972 | * freeable here, so the page does not leak. | |
2973 | */ | |
09cbfeaf | 2974 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 LT |
2975 | unlock_page(page); |
2976 | return 0; /* don't care */ | |
2977 | } | |
2978 | ||
2979 | /* | |
2980 | * The page straddles i_size. It must be zeroed out on each and every | |
2a61aa40 | 2981 | * writepage invocation because it may be mmapped. "A file is mapped |
1da177e4 LT |
2982 | * in multiples of the page size. For a file that is not a multiple of |
2983 | * the page size, the remaining memory is zeroed when mapped, and | |
2984 | * writes to that region are not written out to the file." | |
2985 | */ | |
09cbfeaf | 2986 | zero_user_segment(page, offset, PAGE_SIZE); |
1b938c08 MW |
2987 | return __block_write_full_page(inode, page, get_block, wbc, |
2988 | end_buffer_async_write); | |
35c80d5f | 2989 | } |
1fe72eaa | 2990 | EXPORT_SYMBOL(block_write_full_page); |
35c80d5f | 2991 | |
1da177e4 LT |
2992 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
2993 | get_block_t *get_block) | |
2994 | { | |
1da177e4 | 2995 | struct inode *inode = mapping->host; |
2a527d68 AP |
2996 | struct buffer_head tmp = { |
2997 | .b_size = i_blocksize(inode), | |
2998 | }; | |
2999 | ||
1da177e4 LT |
3000 | get_block(inode, block, &tmp, 0); |
3001 | return tmp.b_blocknr; | |
3002 | } | |
1fe72eaa | 3003 | EXPORT_SYMBOL(generic_block_bmap); |
1da177e4 | 3004 | |
4246a0b6 | 3005 | static void end_bio_bh_io_sync(struct bio *bio) |
1da177e4 LT |
3006 | { |
3007 | struct buffer_head *bh = bio->bi_private; | |
3008 | ||
b7c44ed9 | 3009 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
08bafc03 KM |
3010 | set_bit(BH_Quiet, &bh->b_state); |
3011 | ||
4e4cbee9 | 3012 | bh->b_end_io(bh, !bio->bi_status); |
1da177e4 | 3013 | bio_put(bio); |
1da177e4 LT |
3014 | } |
3015 | ||
2a222ca9 | 3016 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 3017 | enum rw_hint write_hint, struct writeback_control *wbc) |
1da177e4 LT |
3018 | { |
3019 | struct bio *bio; | |
1da177e4 LT |
3020 | |
3021 | BUG_ON(!buffer_locked(bh)); | |
3022 | BUG_ON(!buffer_mapped(bh)); | |
3023 | BUG_ON(!bh->b_end_io); | |
8fb0e342 AK |
3024 | BUG_ON(buffer_delay(bh)); |
3025 | BUG_ON(buffer_unwritten(bh)); | |
1da177e4 | 3026 | |
1da177e4 | 3027 | /* |
48fd4f93 | 3028 | * Only clear out a write error when rewriting |
1da177e4 | 3029 | */ |
2a222ca9 | 3030 | if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) |
1da177e4 LT |
3031 | clear_buffer_write_io_error(bh); |
3032 | ||
3033 | /* | |
3034 | * from here on down, it's all bio -- do the initial mapping, | |
3035 | * submit_bio -> generic_make_request may further map this bio around | |
3036 | */ | |
3037 | bio = bio_alloc(GFP_NOIO, 1); | |
3038 | ||
4f024f37 | 3039 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
74d46992 | 3040 | bio_set_dev(bio, bh->b_bdev); |
8e8f9298 | 3041 | bio->bi_write_hint = write_hint; |
1da177e4 | 3042 | |
6cf66b4c KO |
3043 | bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
3044 | BUG_ON(bio->bi_iter.bi_size != bh->b_size); | |
1da177e4 LT |
3045 | |
3046 | bio->bi_end_io = end_bio_bh_io_sync; | |
3047 | bio->bi_private = bh; | |
3048 | ||
877f962c | 3049 | if (buffer_meta(bh)) |
2a222ca9 | 3050 | op_flags |= REQ_META; |
877f962c | 3051 | if (buffer_prio(bh)) |
2a222ca9 MC |
3052 | op_flags |= REQ_PRIO; |
3053 | bio_set_op_attrs(bio, op, op_flags); | |
877f962c | 3054 | |
83c9c547 ML |
3055 | /* Take care of bh's that straddle the end of the device */ |
3056 | guard_bio_eod(bio); | |
3057 | ||
fd42df30 DZ |
3058 | if (wbc) { |
3059 | wbc_init_bio(wbc, bio); | |
34e51a5e | 3060 | wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); |
fd42df30 DZ |
3061 | } |
3062 | ||
4e49ea4a | 3063 | submit_bio(bio); |
f6454b04 | 3064 | return 0; |
1da177e4 | 3065 | } |
bafc0dba | 3066 | |
020c2833 | 3067 | int submit_bh(int op, int op_flags, struct buffer_head *bh) |
bafc0dba | 3068 | { |
8e8f9298 | 3069 | return submit_bh_wbc(op, op_flags, bh, 0, NULL); |
71368511 | 3070 | } |
1fe72eaa | 3071 | EXPORT_SYMBOL(submit_bh); |
1da177e4 LT |
3072 | |
3073 | /** | |
3074 | * ll_rw_block: low-level access to block devices (DEPRECATED) | |
dfec8a14 | 3075 | * @op: whether to %READ or %WRITE |
ef295ecf | 3076 | * @op_flags: req_flag_bits |
1da177e4 LT |
3077 | * @nr: number of &struct buffer_heads in the array |
3078 | * @bhs: array of pointers to &struct buffer_head | |
3079 | * | |
a7662236 | 3080 | * ll_rw_block() takes an array of pointers to &struct buffer_heads, and |
70246286 CH |
3081 | * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE. |
3082 | * @op_flags contains flags modifying the detailed I/O behavior, most notably | |
3083 | * %REQ_RAHEAD. | |
1da177e4 LT |
3084 | * |
3085 | * This function drops any buffer that it cannot get a lock on (with the | |
9cb569d6 CH |
3086 | * BH_Lock state bit), any buffer that appears to be clean when doing a write |
3087 | * request, and any buffer that appears to be up-to-date when doing read | |
3088 | * request. Further it marks as clean buffers that are processed for | |
3089 | * writing (the buffer cache won't assume that they are actually clean | |
3090 | * until the buffer gets unlocked). | |
1da177e4 LT |
3091 | * |
3092 | * ll_rw_block sets b_end_io to simple completion handler that marks | |
e227867f | 3093 | * the buffer up-to-date (if appropriate), unlocks the buffer and wakes |
1da177e4 LT |
3094 | * any waiters. |
3095 | * | |
3096 | * All of the buffers must be for the same device, and must also be a | |
3097 | * multiple of the current approved size for the device. | |
3098 | */ | |
dfec8a14 | 3099 | void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[]) |
1da177e4 LT |
3100 | { |
3101 | int i; | |
3102 | ||
3103 | for (i = 0; i < nr; i++) { | |
3104 | struct buffer_head *bh = bhs[i]; | |
3105 | ||
9cb569d6 | 3106 | if (!trylock_buffer(bh)) |
1da177e4 | 3107 | continue; |
dfec8a14 | 3108 | if (op == WRITE) { |
1da177e4 | 3109 | if (test_clear_buffer_dirty(bh)) { |
76c3073a | 3110 | bh->b_end_io = end_buffer_write_sync; |
e60e5c50 | 3111 | get_bh(bh); |
dfec8a14 | 3112 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3113 | continue; |
3114 | } | |
3115 | } else { | |
1da177e4 | 3116 | if (!buffer_uptodate(bh)) { |
76c3073a | 3117 | bh->b_end_io = end_buffer_read_sync; |
e60e5c50 | 3118 | get_bh(bh); |
dfec8a14 | 3119 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3120 | continue; |
3121 | } | |
3122 | } | |
3123 | unlock_buffer(bh); | |
1da177e4 LT |
3124 | } |
3125 | } | |
1fe72eaa | 3126 | EXPORT_SYMBOL(ll_rw_block); |
1da177e4 | 3127 | |
2a222ca9 | 3128 | void write_dirty_buffer(struct buffer_head *bh, int op_flags) |
9cb569d6 CH |
3129 | { |
3130 | lock_buffer(bh); | |
3131 | if (!test_clear_buffer_dirty(bh)) { | |
3132 | unlock_buffer(bh); | |
3133 | return; | |
3134 | } | |
3135 | bh->b_end_io = end_buffer_write_sync; | |
3136 | get_bh(bh); | |
2a222ca9 | 3137 | submit_bh(REQ_OP_WRITE, op_flags, bh); |
9cb569d6 CH |
3138 | } |
3139 | EXPORT_SYMBOL(write_dirty_buffer); | |
3140 | ||
1da177e4 LT |
3141 | /* |
3142 | * For a data-integrity writeout, we need to wait upon any in-progress I/O | |
3143 | * and then start new I/O and then wait upon it. The caller must have a ref on | |
3144 | * the buffer_head. | |
3145 | */ | |
2a222ca9 | 3146 | int __sync_dirty_buffer(struct buffer_head *bh, int op_flags) |
1da177e4 LT |
3147 | { |
3148 | int ret = 0; | |
3149 | ||
3150 | WARN_ON(atomic_read(&bh->b_count) < 1); | |
3151 | lock_buffer(bh); | |
3152 | if (test_clear_buffer_dirty(bh)) { | |
3153 | get_bh(bh); | |
3154 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 3155 | ret = submit_bh(REQ_OP_WRITE, op_flags, bh); |
1da177e4 | 3156 | wait_on_buffer(bh); |
1da177e4 LT |
3157 | if (!ret && !buffer_uptodate(bh)) |
3158 | ret = -EIO; | |
3159 | } else { | |
3160 | unlock_buffer(bh); | |
3161 | } | |
3162 | return ret; | |
3163 | } | |
87e99511 CH |
3164 | EXPORT_SYMBOL(__sync_dirty_buffer); |
3165 | ||
3166 | int sync_dirty_buffer(struct buffer_head *bh) | |
3167 | { | |
70fd7614 | 3168 | return __sync_dirty_buffer(bh, REQ_SYNC); |
87e99511 | 3169 | } |
1fe72eaa | 3170 | EXPORT_SYMBOL(sync_dirty_buffer); |
1da177e4 LT |
3171 | |
3172 | /* | |
3173 | * try_to_free_buffers() checks if all the buffers on this particular page | |
3174 | * are unused, and releases them if so. | |
3175 | * | |
3176 | * Exclusion against try_to_free_buffers may be obtained by either | |
3177 | * locking the page or by holding its mapping's private_lock. | |
3178 | * | |
3179 | * If the page is dirty but all the buffers are clean then we need to | |
3180 | * be sure to mark the page clean as well. This is because the page | |
3181 | * may be against a block device, and a later reattachment of buffers | |
3182 | * to a dirty page will set *all* buffers dirty. Which would corrupt | |
3183 | * filesystem data on the same device. | |
3184 | * | |
3185 | * The same applies to regular filesystem pages: if all the buffers are | |
3186 | * clean then we set the page clean and proceed. To do that, we require | |
3187 | * total exclusion from __set_page_dirty_buffers(). That is obtained with | |
3188 | * private_lock. | |
3189 | * | |
3190 | * try_to_free_buffers() is non-blocking. | |
3191 | */ | |
3192 | static inline int buffer_busy(struct buffer_head *bh) | |
3193 | { | |
3194 | return atomic_read(&bh->b_count) | | |
3195 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); | |
3196 | } | |
3197 | ||
3198 | static int | |
3199 | drop_buffers(struct page *page, struct buffer_head **buffers_to_free) | |
3200 | { | |
3201 | struct buffer_head *head = page_buffers(page); | |
3202 | struct buffer_head *bh; | |
3203 | ||
3204 | bh = head; | |
3205 | do { | |
1da177e4 LT |
3206 | if (buffer_busy(bh)) |
3207 | goto failed; | |
3208 | bh = bh->b_this_page; | |
3209 | } while (bh != head); | |
3210 | ||
3211 | do { | |
3212 | struct buffer_head *next = bh->b_this_page; | |
3213 | ||
535ee2fb | 3214 | if (bh->b_assoc_map) |
1da177e4 LT |
3215 | __remove_assoc_queue(bh); |
3216 | bh = next; | |
3217 | } while (bh != head); | |
3218 | *buffers_to_free = head; | |
3219 | __clear_page_buffers(page); | |
3220 | return 1; | |
3221 | failed: | |
3222 | return 0; | |
3223 | } | |
3224 | ||
3225 | int try_to_free_buffers(struct page *page) | |
3226 | { | |
3227 | struct address_space * const mapping = page->mapping; | |
3228 | struct buffer_head *buffers_to_free = NULL; | |
3229 | int ret = 0; | |
3230 | ||
3231 | BUG_ON(!PageLocked(page)); | |
ecdfc978 | 3232 | if (PageWriteback(page)) |
1da177e4 LT |
3233 | return 0; |
3234 | ||
3235 | if (mapping == NULL) { /* can this still happen? */ | |
3236 | ret = drop_buffers(page, &buffers_to_free); | |
3237 | goto out; | |
3238 | } | |
3239 | ||
3240 | spin_lock(&mapping->private_lock); | |
3241 | ret = drop_buffers(page, &buffers_to_free); | |
ecdfc978 LT |
3242 | |
3243 | /* | |
3244 | * If the filesystem writes its buffers by hand (eg ext3) | |
3245 | * then we can have clean buffers against a dirty page. We | |
3246 | * clean the page here; otherwise the VM will never notice | |
3247 | * that the filesystem did any IO at all. | |
3248 | * | |
3249 | * Also, during truncate, discard_buffer will have marked all | |
3250 | * the page's buffers clean. We discover that here and clean | |
3251 | * the page also. | |
87df7241 NP |
3252 | * |
3253 | * private_lock must be held over this entire operation in order | |
3254 | * to synchronise against __set_page_dirty_buffers and prevent the | |
3255 | * dirty bit from being lost. | |
ecdfc978 | 3256 | */ |
11f81bec TH |
3257 | if (ret) |
3258 | cancel_dirty_page(page); | |
87df7241 | 3259 | spin_unlock(&mapping->private_lock); |
1da177e4 LT |
3260 | out: |
3261 | if (buffers_to_free) { | |
3262 | struct buffer_head *bh = buffers_to_free; | |
3263 | ||
3264 | do { | |
3265 | struct buffer_head *next = bh->b_this_page; | |
3266 | free_buffer_head(bh); | |
3267 | bh = next; | |
3268 | } while (bh != buffers_to_free); | |
3269 | } | |
3270 | return ret; | |
3271 | } | |
3272 | EXPORT_SYMBOL(try_to_free_buffers); | |
3273 | ||
1da177e4 LT |
3274 | /* |
3275 | * There are no bdflush tunables left. But distributions are | |
3276 | * still running obsolete flush daemons, so we terminate them here. | |
3277 | * | |
3278 | * Use of bdflush() is deprecated and will be removed in a future kernel. | |
5b0830cb | 3279 | * The `flush-X' kernel threads fully replace bdflush daemons and this call. |
1da177e4 | 3280 | */ |
bdc480e3 | 3281 | SYSCALL_DEFINE2(bdflush, int, func, long, data) |
1da177e4 LT |
3282 | { |
3283 | static int msg_count; | |
3284 | ||
3285 | if (!capable(CAP_SYS_ADMIN)) | |
3286 | return -EPERM; | |
3287 | ||
3288 | if (msg_count < 5) { | |
3289 | msg_count++; | |
3290 | printk(KERN_INFO | |
3291 | "warning: process `%s' used the obsolete bdflush" | |
3292 | " system call\n", current->comm); | |
3293 | printk(KERN_INFO "Fix your initscripts?\n"); | |
3294 | } | |
3295 | ||
3296 | if (func == 1) | |
3297 | do_exit(0); | |
3298 | return 0; | |
3299 | } | |
3300 | ||
3301 | /* | |
3302 | * Buffer-head allocation | |
3303 | */ | |
a0a9b043 | 3304 | static struct kmem_cache *bh_cachep __read_mostly; |
1da177e4 LT |
3305 | |
3306 | /* | |
3307 | * Once the number of bh's in the machine exceeds this level, we start | |
3308 | * stripping them in writeback. | |
3309 | */ | |
43be594a | 3310 | static unsigned long max_buffer_heads; |
1da177e4 LT |
3311 | |
3312 | int buffer_heads_over_limit; | |
3313 | ||
3314 | struct bh_accounting { | |
3315 | int nr; /* Number of live bh's */ | |
3316 | int ratelimit; /* Limit cacheline bouncing */ | |
3317 | }; | |
3318 | ||
3319 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; | |
3320 | ||
3321 | static void recalc_bh_state(void) | |
3322 | { | |
3323 | int i; | |
3324 | int tot = 0; | |
3325 | ||
ee1be862 | 3326 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
1da177e4 | 3327 | return; |
c7b92516 | 3328 | __this_cpu_write(bh_accounting.ratelimit, 0); |
8a143426 | 3329 | for_each_online_cpu(i) |
1da177e4 LT |
3330 | tot += per_cpu(bh_accounting, i).nr; |
3331 | buffer_heads_over_limit = (tot > max_buffer_heads); | |
3332 | } | |
c7b92516 | 3333 | |
dd0fc66f | 3334 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
1da177e4 | 3335 | { |
019b4d12 | 3336 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
1da177e4 | 3337 | if (ret) { |
a35afb83 | 3338 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
f1e67e35 | 3339 | spin_lock_init(&ret->b_uptodate_lock); |
c7b92516 CL |
3340 | preempt_disable(); |
3341 | __this_cpu_inc(bh_accounting.nr); | |
1da177e4 | 3342 | recalc_bh_state(); |
c7b92516 | 3343 | preempt_enable(); |
1da177e4 LT |
3344 | } |
3345 | return ret; | |
3346 | } | |
3347 | EXPORT_SYMBOL(alloc_buffer_head); | |
3348 | ||
3349 | void free_buffer_head(struct buffer_head *bh) | |
3350 | { | |
3351 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); | |
3352 | kmem_cache_free(bh_cachep, bh); | |
c7b92516 CL |
3353 | preempt_disable(); |
3354 | __this_cpu_dec(bh_accounting.nr); | |
1da177e4 | 3355 | recalc_bh_state(); |
c7b92516 | 3356 | preempt_enable(); |
1da177e4 LT |
3357 | } |
3358 | EXPORT_SYMBOL(free_buffer_head); | |
3359 | ||
fc4d24c9 | 3360 | static int buffer_exit_cpu_dead(unsigned int cpu) |
1da177e4 LT |
3361 | { |
3362 | int i; | |
3363 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); | |
3364 | ||
3365 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
3366 | brelse(b->bhs[i]); | |
3367 | b->bhs[i] = NULL; | |
3368 | } | |
c7b92516 | 3369 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
8a143426 | 3370 | per_cpu(bh_accounting, cpu).nr = 0; |
fc4d24c9 | 3371 | return 0; |
1da177e4 | 3372 | } |
1da177e4 | 3373 | |
389d1b08 | 3374 | /** |
a6b91919 | 3375 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
389d1b08 AK |
3376 | * @bh: struct buffer_head |
3377 | * | |
3378 | * Return true if the buffer is up-to-date and false, | |
3379 | * with the buffer locked, if not. | |
3380 | */ | |
3381 | int bh_uptodate_or_lock(struct buffer_head *bh) | |
3382 | { | |
3383 | if (!buffer_uptodate(bh)) { | |
3384 | lock_buffer(bh); | |
3385 | if (!buffer_uptodate(bh)) | |
3386 | return 0; | |
3387 | unlock_buffer(bh); | |
3388 | } | |
3389 | return 1; | |
3390 | } | |
3391 | EXPORT_SYMBOL(bh_uptodate_or_lock); | |
3392 | ||
3393 | /** | |
a6b91919 | 3394 | * bh_submit_read - Submit a locked buffer for reading |
389d1b08 AK |
3395 | * @bh: struct buffer_head |
3396 | * | |
3397 | * Returns zero on success and -EIO on error. | |
3398 | */ | |
3399 | int bh_submit_read(struct buffer_head *bh) | |
3400 | { | |
3401 | BUG_ON(!buffer_locked(bh)); | |
3402 | ||
3403 | if (buffer_uptodate(bh)) { | |
3404 | unlock_buffer(bh); | |
3405 | return 0; | |
3406 | } | |
3407 | ||
3408 | get_bh(bh); | |
3409 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 3410 | submit_bh(REQ_OP_READ, 0, bh); |
389d1b08 AK |
3411 | wait_on_buffer(bh); |
3412 | if (buffer_uptodate(bh)) | |
3413 | return 0; | |
3414 | return -EIO; | |
3415 | } | |
3416 | EXPORT_SYMBOL(bh_submit_read); | |
3417 | ||
1da177e4 LT |
3418 | void __init buffer_init(void) |
3419 | { | |
43be594a | 3420 | unsigned long nrpages; |
fc4d24c9 | 3421 | int ret; |
1da177e4 | 3422 | |
b98938c3 CL |
3423 | bh_cachep = kmem_cache_create("buffer_head", |
3424 | sizeof(struct buffer_head), 0, | |
3425 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
3426 | SLAB_MEM_SPREAD), | |
019b4d12 | 3427 | NULL); |
1da177e4 LT |
3428 | |
3429 | /* | |
3430 | * Limit the bh occupancy to 10% of ZONE_NORMAL | |
3431 | */ | |
3432 | nrpages = (nr_free_buffer_pages() * 10) / 100; | |
3433 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); | |
fc4d24c9 SAS |
3434 | ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", |
3435 | NULL, buffer_exit_cpu_dead); | |
3436 | WARN_ON(ret < 0); | |
1da177e4 | 3437 | } |