]>
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 LT |
51 | |
52 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); | |
2a222ca9 | 53 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 54 | enum rw_hint hint, struct writeback_control *wbc); |
1da177e4 LT |
55 | |
56 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) | |
57 | ||
f0059afd TH |
58 | inline void touch_buffer(struct buffer_head *bh) |
59 | { | |
5305cb83 | 60 | trace_block_touch_buffer(bh); |
f0059afd TH |
61 | mark_page_accessed(bh->b_page); |
62 | } | |
63 | EXPORT_SYMBOL(touch_buffer); | |
64 | ||
fc9b52cd | 65 | void __lock_buffer(struct buffer_head *bh) |
1da177e4 | 66 | { |
74316201 | 67 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 LT |
68 | } |
69 | EXPORT_SYMBOL(__lock_buffer); | |
70 | ||
fc9b52cd | 71 | void unlock_buffer(struct buffer_head *bh) |
1da177e4 | 72 | { |
51b07fc3 | 73 | clear_bit_unlock(BH_Lock, &bh->b_state); |
4e857c58 | 74 | smp_mb__after_atomic(); |
1da177e4 LT |
75 | wake_up_bit(&bh->b_state, BH_Lock); |
76 | } | |
1fe72eaa | 77 | EXPORT_SYMBOL(unlock_buffer); |
1da177e4 | 78 | |
b4597226 MG |
79 | /* |
80 | * Returns if the page has dirty or writeback buffers. If all the buffers | |
81 | * are unlocked and clean then the PageDirty information is stale. If | |
82 | * any of the pages are locked, it is assumed they are locked for IO. | |
83 | */ | |
84 | void buffer_check_dirty_writeback(struct page *page, | |
85 | bool *dirty, bool *writeback) | |
86 | { | |
87 | struct buffer_head *head, *bh; | |
88 | *dirty = false; | |
89 | *writeback = false; | |
90 | ||
91 | BUG_ON(!PageLocked(page)); | |
92 | ||
93 | if (!page_has_buffers(page)) | |
94 | return; | |
95 | ||
96 | if (PageWriteback(page)) | |
97 | *writeback = true; | |
98 | ||
99 | head = page_buffers(page); | |
100 | bh = head; | |
101 | do { | |
102 | if (buffer_locked(bh)) | |
103 | *writeback = true; | |
104 | ||
105 | if (buffer_dirty(bh)) | |
106 | *dirty = true; | |
107 | ||
108 | bh = bh->b_this_page; | |
109 | } while (bh != head); | |
110 | } | |
111 | EXPORT_SYMBOL(buffer_check_dirty_writeback); | |
112 | ||
1da177e4 LT |
113 | /* |
114 | * Block until a buffer comes unlocked. This doesn't stop it | |
115 | * from becoming locked again - you have to lock it yourself | |
116 | * if you want to preserve its state. | |
117 | */ | |
118 | void __wait_on_buffer(struct buffer_head * bh) | |
119 | { | |
74316201 | 120 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 | 121 | } |
1fe72eaa | 122 | EXPORT_SYMBOL(__wait_on_buffer); |
1da177e4 LT |
123 | |
124 | static void | |
125 | __clear_page_buffers(struct page *page) | |
126 | { | |
127 | ClearPagePrivate(page); | |
4c21e2f2 | 128 | set_page_private(page, 0); |
09cbfeaf | 129 | put_page(page); |
1da177e4 LT |
130 | } |
131 | ||
b744c2ac | 132 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
1da177e4 | 133 | { |
432f16e6 RE |
134 | if (!test_bit(BH_Quiet, &bh->b_state)) |
135 | printk_ratelimited(KERN_ERR | |
a1c6f057 DM |
136 | "Buffer I/O error on dev %pg, logical block %llu%s\n", |
137 | bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); | |
1da177e4 LT |
138 | } |
139 | ||
140 | /* | |
68671f35 DM |
141 | * End-of-IO handler helper function which does not touch the bh after |
142 | * unlocking it. | |
143 | * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but | |
144 | * a race there is benign: unlock_buffer() only use the bh's address for | |
145 | * hashing after unlocking the buffer, so it doesn't actually touch the bh | |
146 | * itself. | |
1da177e4 | 147 | */ |
68671f35 | 148 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
1da177e4 LT |
149 | { |
150 | if (uptodate) { | |
151 | set_buffer_uptodate(bh); | |
152 | } else { | |
70246286 | 153 | /* This happens, due to failed read-ahead attempts. */ |
1da177e4 LT |
154 | clear_buffer_uptodate(bh); |
155 | } | |
156 | unlock_buffer(bh); | |
68671f35 DM |
157 | } |
158 | ||
159 | /* | |
160 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and | |
161 | * unlock the buffer. This is what ll_rw_block uses too. | |
162 | */ | |
163 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) | |
164 | { | |
165 | __end_buffer_read_notouch(bh, uptodate); | |
1da177e4 LT |
166 | put_bh(bh); |
167 | } | |
1fe72eaa | 168 | EXPORT_SYMBOL(end_buffer_read_sync); |
1da177e4 LT |
169 | |
170 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) | |
171 | { | |
1da177e4 LT |
172 | if (uptodate) { |
173 | set_buffer_uptodate(bh); | |
174 | } else { | |
432f16e6 | 175 | buffer_io_error(bh, ", lost sync page write"); |
87354e5d | 176 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
177 | clear_buffer_uptodate(bh); |
178 | } | |
179 | unlock_buffer(bh); | |
180 | put_bh(bh); | |
181 | } | |
1fe72eaa | 182 | EXPORT_SYMBOL(end_buffer_write_sync); |
1da177e4 | 183 | |
1da177e4 LT |
184 | /* |
185 | * Various filesystems appear to want __find_get_block to be non-blocking. | |
186 | * But it's the page lock which protects the buffers. To get around this, | |
187 | * we get exclusion from try_to_free_buffers with the blockdev mapping's | |
188 | * private_lock. | |
189 | * | |
b93b0163 | 190 | * Hack idea: for the blockdev mapping, private_lock contention |
1da177e4 | 191 | * may be quite high. This code could TryLock the page, and if that |
b93b0163 | 192 | * succeeds, there is no need to take private_lock. |
1da177e4 LT |
193 | */ |
194 | static struct buffer_head * | |
385fd4c5 | 195 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
1da177e4 LT |
196 | { |
197 | struct inode *bd_inode = bdev->bd_inode; | |
198 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
199 | struct buffer_head *ret = NULL; | |
200 | pgoff_t index; | |
201 | struct buffer_head *bh; | |
202 | struct buffer_head *head; | |
203 | struct page *page; | |
204 | int all_mapped = 1; | |
43636c80 | 205 | static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); |
1da177e4 | 206 | |
09cbfeaf | 207 | index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); |
2457aec6 | 208 | page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); |
1da177e4 LT |
209 | if (!page) |
210 | goto out; | |
211 | ||
212 | spin_lock(&bd_mapping->private_lock); | |
213 | if (!page_has_buffers(page)) | |
214 | goto out_unlock; | |
215 | head = page_buffers(page); | |
216 | bh = head; | |
217 | do { | |
97f76d3d NK |
218 | if (!buffer_mapped(bh)) |
219 | all_mapped = 0; | |
220 | else if (bh->b_blocknr == block) { | |
1da177e4 LT |
221 | ret = bh; |
222 | get_bh(bh); | |
223 | goto out_unlock; | |
224 | } | |
1da177e4 LT |
225 | bh = bh->b_this_page; |
226 | } while (bh != head); | |
227 | ||
228 | /* we might be here because some of the buffers on this page are | |
229 | * not mapped. This is due to various races between | |
230 | * file io on the block device and getblk. It gets dealt with | |
231 | * elsewhere, don't buffer_error if we had some unmapped buffers | |
232 | */ | |
43636c80 TH |
233 | ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); |
234 | if (all_mapped && __ratelimit(&last_warned)) { | |
235 | printk("__find_get_block_slow() failed. block=%llu, " | |
236 | "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " | |
237 | "device %pg blocksize: %d\n", | |
238 | (unsigned long long)block, | |
239 | (unsigned long long)bh->b_blocknr, | |
240 | bh->b_state, bh->b_size, bdev, | |
241 | 1 << bd_inode->i_blkbits); | |
1da177e4 LT |
242 | } |
243 | out_unlock: | |
244 | spin_unlock(&bd_mapping->private_lock); | |
09cbfeaf | 245 | put_page(page); |
1da177e4 LT |
246 | out: |
247 | return ret; | |
248 | } | |
249 | ||
1da177e4 LT |
250 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) |
251 | { | |
1da177e4 | 252 | unsigned long flags; |
a3972203 | 253 | struct buffer_head *first; |
1da177e4 LT |
254 | struct buffer_head *tmp; |
255 | struct page *page; | |
256 | int page_uptodate = 1; | |
257 | ||
258 | BUG_ON(!buffer_async_read(bh)); | |
259 | ||
260 | page = bh->b_page; | |
261 | if (uptodate) { | |
262 | set_buffer_uptodate(bh); | |
263 | } else { | |
264 | clear_buffer_uptodate(bh); | |
432f16e6 | 265 | buffer_io_error(bh, ", async page read"); |
1da177e4 LT |
266 | SetPageError(page); |
267 | } | |
268 | ||
269 | /* | |
270 | * Be _very_ careful from here on. Bad things can happen if | |
271 | * two buffer heads end IO at almost the same time and both | |
272 | * decide that the page is now completely done. | |
273 | */ | |
a3972203 NP |
274 | first = page_buffers(page); |
275 | local_irq_save(flags); | |
276 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
1da177e4 LT |
277 | clear_buffer_async_read(bh); |
278 | unlock_buffer(bh); | |
279 | tmp = bh; | |
280 | do { | |
281 | if (!buffer_uptodate(tmp)) | |
282 | page_uptodate = 0; | |
283 | if (buffer_async_read(tmp)) { | |
284 | BUG_ON(!buffer_locked(tmp)); | |
285 | goto still_busy; | |
286 | } | |
287 | tmp = tmp->b_this_page; | |
288 | } while (tmp != bh); | |
a3972203 NP |
289 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
290 | local_irq_restore(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: | |
a3972203 NP |
302 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
303 | local_irq_restore(flags); | |
1da177e4 LT |
304 | return; |
305 | } | |
306 | ||
31fb992c EB |
307 | struct decrypt_bh_ctx { |
308 | struct work_struct work; | |
309 | struct buffer_head *bh; | |
310 | }; | |
311 | ||
312 | static void decrypt_bh(struct work_struct *work) | |
313 | { | |
314 | struct decrypt_bh_ctx *ctx = | |
315 | container_of(work, struct decrypt_bh_ctx, work); | |
316 | struct buffer_head *bh = ctx->bh; | |
317 | int err; | |
318 | ||
319 | err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size, | |
320 | bh_offset(bh)); | |
321 | end_buffer_async_read(bh, err == 0); | |
322 | kfree(ctx); | |
323 | } | |
324 | ||
325 | /* | |
326 | * I/O completion handler for block_read_full_page() - pages | |
327 | * which come unlocked at the end of I/O. | |
328 | */ | |
329 | static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate) | |
330 | { | |
331 | /* Decrypt if needed */ | |
332 | if (uptodate && IS_ENABLED(CONFIG_FS_ENCRYPTION) && | |
333 | IS_ENCRYPTED(bh->b_page->mapping->host) && | |
334 | S_ISREG(bh->b_page->mapping->host->i_mode)) { | |
335 | struct decrypt_bh_ctx *ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC); | |
336 | ||
337 | if (ctx) { | |
338 | INIT_WORK(&ctx->work, decrypt_bh); | |
339 | ctx->bh = bh; | |
340 | fscrypt_enqueue_decrypt_work(&ctx->work); | |
341 | return; | |
342 | } | |
343 | uptodate = 0; | |
344 | } | |
345 | end_buffer_async_read(bh, uptodate); | |
346 | } | |
347 | ||
1da177e4 LT |
348 | /* |
349 | * Completion handler for block_write_full_page() - pages which are unlocked | |
350 | * during I/O, and which have PageWriteback cleared upon I/O completion. | |
351 | */ | |
35c80d5f | 352 | void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
1da177e4 | 353 | { |
1da177e4 | 354 | unsigned long flags; |
a3972203 | 355 | struct buffer_head *first; |
1da177e4 LT |
356 | struct buffer_head *tmp; |
357 | struct page *page; | |
358 | ||
359 | BUG_ON(!buffer_async_write(bh)); | |
360 | ||
361 | page = bh->b_page; | |
362 | if (uptodate) { | |
363 | set_buffer_uptodate(bh); | |
364 | } else { | |
432f16e6 | 365 | buffer_io_error(bh, ", lost async page write"); |
87354e5d | 366 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
367 | clear_buffer_uptodate(bh); |
368 | SetPageError(page); | |
369 | } | |
370 | ||
a3972203 NP |
371 | first = page_buffers(page); |
372 | local_irq_save(flags); | |
373 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
374 | ||
1da177e4 LT |
375 | clear_buffer_async_write(bh); |
376 | unlock_buffer(bh); | |
377 | tmp = bh->b_this_page; | |
378 | while (tmp != bh) { | |
379 | if (buffer_async_write(tmp)) { | |
380 | BUG_ON(!buffer_locked(tmp)); | |
381 | goto still_busy; | |
382 | } | |
383 | tmp = tmp->b_this_page; | |
384 | } | |
a3972203 NP |
385 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
386 | local_irq_restore(flags); | |
1da177e4 LT |
387 | end_page_writeback(page); |
388 | return; | |
389 | ||
390 | still_busy: | |
a3972203 NP |
391 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
392 | local_irq_restore(flags); | |
1da177e4 LT |
393 | return; |
394 | } | |
1fe72eaa | 395 | EXPORT_SYMBOL(end_buffer_async_write); |
1da177e4 LT |
396 | |
397 | /* | |
398 | * If a page's buffers are under async readin (end_buffer_async_read | |
399 | * completion) then there is a possibility that another thread of | |
400 | * control could lock one of the buffers after it has completed | |
401 | * but while some of the other buffers have not completed. This | |
402 | * locked buffer would confuse end_buffer_async_read() into not unlocking | |
403 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() | |
404 | * that this buffer is not under async I/O. | |
405 | * | |
406 | * The page comes unlocked when it has no locked buffer_async buffers | |
407 | * left. | |
408 | * | |
409 | * PageLocked prevents anyone starting new async I/O reads any of | |
410 | * the buffers. | |
411 | * | |
412 | * PageWriteback is used to prevent simultaneous writeout of the same | |
413 | * page. | |
414 | * | |
415 | * PageLocked prevents anyone from starting writeback of a page which is | |
416 | * under read I/O (PageWriteback is only ever set against a locked page). | |
417 | */ | |
418 | static void mark_buffer_async_read(struct buffer_head *bh) | |
419 | { | |
31fb992c | 420 | bh->b_end_io = end_buffer_async_read_io; |
1da177e4 LT |
421 | set_buffer_async_read(bh); |
422 | } | |
423 | ||
1fe72eaa HS |
424 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
425 | bh_end_io_t *handler) | |
1da177e4 | 426 | { |
35c80d5f | 427 | bh->b_end_io = handler; |
1da177e4 LT |
428 | set_buffer_async_write(bh); |
429 | } | |
35c80d5f CM |
430 | |
431 | void mark_buffer_async_write(struct buffer_head *bh) | |
432 | { | |
433 | mark_buffer_async_write_endio(bh, end_buffer_async_write); | |
434 | } | |
1da177e4 LT |
435 | EXPORT_SYMBOL(mark_buffer_async_write); |
436 | ||
437 | ||
438 | /* | |
439 | * fs/buffer.c contains helper functions for buffer-backed address space's | |
440 | * fsync functions. A common requirement for buffer-based filesystems is | |
441 | * that certain data from the backing blockdev needs to be written out for | |
442 | * a successful fsync(). For example, ext2 indirect blocks need to be | |
443 | * written back and waited upon before fsync() returns. | |
444 | * | |
445 | * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), | |
446 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the | |
447 | * management of a list of dependent buffers at ->i_mapping->private_list. | |
448 | * | |
449 | * Locking is a little subtle: try_to_free_buffers() will remove buffers | |
450 | * from their controlling inode's queue when they are being freed. But | |
451 | * try_to_free_buffers() will be operating against the *blockdev* mapping | |
452 | * at the time, not against the S_ISREG file which depends on those buffers. | |
453 | * So the locking for private_list is via the private_lock in the address_space | |
454 | * which backs the buffers. Which is different from the address_space | |
455 | * against which the buffers are listed. So for a particular address_space, | |
456 | * mapping->private_lock does *not* protect mapping->private_list! In fact, | |
457 | * mapping->private_list will always be protected by the backing blockdev's | |
458 | * ->private_lock. | |
459 | * | |
460 | * Which introduces a requirement: all buffers on an address_space's | |
461 | * ->private_list must be from the same address_space: the blockdev's. | |
462 | * | |
463 | * address_spaces which do not place buffers at ->private_list via these | |
464 | * utility functions are free to use private_lock and private_list for | |
465 | * whatever they want. The only requirement is that list_empty(private_list) | |
466 | * be true at clear_inode() time. | |
467 | * | |
468 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The | |
469 | * filesystems should do that. invalidate_inode_buffers() should just go | |
470 | * BUG_ON(!list_empty). | |
471 | * | |
472 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should | |
473 | * take an address_space, not an inode. And it should be called | |
474 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being | |
475 | * queued up. | |
476 | * | |
477 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the | |
478 | * list if it is already on a list. Because if the buffer is on a list, | |
479 | * it *must* already be on the right one. If not, the filesystem is being | |
480 | * silly. This will save a ton of locking. But first we have to ensure | |
481 | * that buffers are taken *off* the old inode's list when they are freed | |
482 | * (presumably in truncate). That requires careful auditing of all | |
483 | * filesystems (do it inside bforget()). It could also be done by bringing | |
484 | * b_inode back. | |
485 | */ | |
486 | ||
487 | /* | |
488 | * The buffer's backing address_space's private_lock must be held | |
489 | */ | |
dbacefc9 | 490 | static void __remove_assoc_queue(struct buffer_head *bh) |
1da177e4 LT |
491 | { |
492 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 493 | WARN_ON(!bh->b_assoc_map); |
58ff407b | 494 | bh->b_assoc_map = NULL; |
1da177e4 LT |
495 | } |
496 | ||
497 | int inode_has_buffers(struct inode *inode) | |
498 | { | |
499 | return !list_empty(&inode->i_data.private_list); | |
500 | } | |
501 | ||
502 | /* | |
503 | * osync is designed to support O_SYNC io. It waits synchronously for | |
504 | * all already-submitted IO to complete, but does not queue any new | |
505 | * writes to the disk. | |
506 | * | |
507 | * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as | |
508 | * you dirty the buffers, and then use osync_inode_buffers to wait for | |
509 | * completion. Any other dirty buffers which are not yet queued for | |
510 | * write will not be flushed to disk by the osync. | |
511 | */ | |
512 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) | |
513 | { | |
514 | struct buffer_head *bh; | |
515 | struct list_head *p; | |
516 | int err = 0; | |
517 | ||
518 | spin_lock(lock); | |
519 | repeat: | |
520 | list_for_each_prev(p, list) { | |
521 | bh = BH_ENTRY(p); | |
522 | if (buffer_locked(bh)) { | |
523 | get_bh(bh); | |
524 | spin_unlock(lock); | |
525 | wait_on_buffer(bh); | |
526 | if (!buffer_uptodate(bh)) | |
527 | err = -EIO; | |
528 | brelse(bh); | |
529 | spin_lock(lock); | |
530 | goto repeat; | |
531 | } | |
532 | } | |
533 | spin_unlock(lock); | |
534 | return err; | |
535 | } | |
536 | ||
08fdc8a0 | 537 | void emergency_thaw_bdev(struct super_block *sb) |
c2d75438 | 538 | { |
01a05b33 | 539 | while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) |
a1c6f057 | 540 | printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); |
01a05b33 | 541 | } |
c2d75438 | 542 | |
1da177e4 | 543 | /** |
78a4a50a | 544 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
67be2dd1 | 545 | * @mapping: the mapping which wants those buffers written |
1da177e4 LT |
546 | * |
547 | * Starts I/O against the buffers at mapping->private_list, and waits upon | |
548 | * that I/O. | |
549 | * | |
67be2dd1 MW |
550 | * Basically, this is a convenience function for fsync(). |
551 | * @mapping is a file or directory which needs those buffers to be written for | |
552 | * a successful fsync(). | |
1da177e4 LT |
553 | */ |
554 | int sync_mapping_buffers(struct address_space *mapping) | |
555 | { | |
252aa6f5 | 556 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
557 | |
558 | if (buffer_mapping == NULL || list_empty(&mapping->private_list)) | |
559 | return 0; | |
560 | ||
561 | return fsync_buffers_list(&buffer_mapping->private_lock, | |
562 | &mapping->private_list); | |
563 | } | |
564 | EXPORT_SYMBOL(sync_mapping_buffers); | |
565 | ||
566 | /* | |
567 | * Called when we've recently written block `bblock', and it is known that | |
568 | * `bblock' was for a buffer_boundary() buffer. This means that the block at | |
569 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's | |
570 | * dirty, schedule it for IO. So that indirects merge nicely with their data. | |
571 | */ | |
572 | void write_boundary_block(struct block_device *bdev, | |
573 | sector_t bblock, unsigned blocksize) | |
574 | { | |
575 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); | |
576 | if (bh) { | |
577 | if (buffer_dirty(bh)) | |
dfec8a14 | 578 | ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); |
1da177e4 LT |
579 | put_bh(bh); |
580 | } | |
581 | } | |
582 | ||
583 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) | |
584 | { | |
585 | struct address_space *mapping = inode->i_mapping; | |
586 | struct address_space *buffer_mapping = bh->b_page->mapping; | |
587 | ||
588 | mark_buffer_dirty(bh); | |
252aa6f5 RA |
589 | if (!mapping->private_data) { |
590 | mapping->private_data = buffer_mapping; | |
1da177e4 | 591 | } else { |
252aa6f5 | 592 | BUG_ON(mapping->private_data != buffer_mapping); |
1da177e4 | 593 | } |
535ee2fb | 594 | if (!bh->b_assoc_map) { |
1da177e4 LT |
595 | spin_lock(&buffer_mapping->private_lock); |
596 | list_move_tail(&bh->b_assoc_buffers, | |
597 | &mapping->private_list); | |
58ff407b | 598 | bh->b_assoc_map = mapping; |
1da177e4 LT |
599 | spin_unlock(&buffer_mapping->private_lock); |
600 | } | |
601 | } | |
602 | EXPORT_SYMBOL(mark_buffer_dirty_inode); | |
603 | ||
787d2214 | 604 | /* |
ec82e1c1 | 605 | * Mark the page dirty, and set it dirty in the page cache, and mark the inode |
787d2214 NP |
606 | * dirty. |
607 | * | |
608 | * If warn is true, then emit a warning if the page is not uptodate and has | |
609 | * not been truncated. | |
c4843a75 | 610 | * |
81f8c3a4 | 611 | * The caller must hold lock_page_memcg(). |
787d2214 | 612 | */ |
f82b3764 | 613 | void __set_page_dirty(struct page *page, struct address_space *mapping, |
62cccb8c | 614 | int warn) |
787d2214 | 615 | { |
227d53b3 KM |
616 | unsigned long flags; |
617 | ||
b93b0163 | 618 | xa_lock_irqsave(&mapping->i_pages, flags); |
787d2214 NP |
619 | if (page->mapping) { /* Race with truncate? */ |
620 | WARN_ON_ONCE(warn && !PageUptodate(page)); | |
62cccb8c | 621 | account_page_dirtied(page, mapping); |
ec82e1c1 MW |
622 | __xa_set_mark(&mapping->i_pages, page_index(page), |
623 | PAGECACHE_TAG_DIRTY); | |
787d2214 | 624 | } |
b93b0163 | 625 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
787d2214 | 626 | } |
f82b3764 | 627 | EXPORT_SYMBOL_GPL(__set_page_dirty); |
787d2214 | 628 | |
1da177e4 LT |
629 | /* |
630 | * Add a page to the dirty page list. | |
631 | * | |
632 | * It is a sad fact of life that this function is called from several places | |
633 | * deeply under spinlocking. It may not sleep. | |
634 | * | |
635 | * If the page has buffers, the uptodate buffers are set dirty, to preserve | |
636 | * dirty-state coherency between the page and the buffers. It the page does | |
637 | * not have buffers then when they are later attached they will all be set | |
638 | * dirty. | |
639 | * | |
640 | * The buffers are dirtied before the page is dirtied. There's a small race | |
641 | * window in which a writepage caller may see the page cleanness but not the | |
642 | * buffer dirtiness. That's fine. If this code were to set the page dirty | |
643 | * before the buffers, a concurrent writepage caller could clear the page dirty | |
644 | * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean | |
645 | * page on the dirty page list. | |
646 | * | |
647 | * We use private_lock to lock against try_to_free_buffers while using the | |
648 | * page's buffer list. Also use this to protect against clean buffers being | |
649 | * added to the page after it was set dirty. | |
650 | * | |
651 | * FIXME: may need to call ->reservepage here as well. That's rather up to the | |
652 | * address_space though. | |
653 | */ | |
654 | int __set_page_dirty_buffers(struct page *page) | |
655 | { | |
a8e7d49a | 656 | int newly_dirty; |
787d2214 | 657 | struct address_space *mapping = page_mapping(page); |
ebf7a227 NP |
658 | |
659 | if (unlikely(!mapping)) | |
660 | return !TestSetPageDirty(page); | |
1da177e4 LT |
661 | |
662 | spin_lock(&mapping->private_lock); | |
663 | if (page_has_buffers(page)) { | |
664 | struct buffer_head *head = page_buffers(page); | |
665 | struct buffer_head *bh = head; | |
666 | ||
667 | do { | |
668 | set_buffer_dirty(bh); | |
669 | bh = bh->b_this_page; | |
670 | } while (bh != head); | |
671 | } | |
c4843a75 | 672 | /* |
81f8c3a4 JW |
673 | * Lock out page->mem_cgroup migration to keep PageDirty |
674 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 675 | */ |
62cccb8c | 676 | lock_page_memcg(page); |
a8e7d49a | 677 | newly_dirty = !TestSetPageDirty(page); |
1da177e4 LT |
678 | spin_unlock(&mapping->private_lock); |
679 | ||
a8e7d49a | 680 | if (newly_dirty) |
62cccb8c | 681 | __set_page_dirty(page, mapping, 1); |
c4843a75 | 682 | |
62cccb8c | 683 | unlock_page_memcg(page); |
c4843a75 GT |
684 | |
685 | if (newly_dirty) | |
686 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
687 | ||
a8e7d49a | 688 | return newly_dirty; |
1da177e4 LT |
689 | } |
690 | EXPORT_SYMBOL(__set_page_dirty_buffers); | |
691 | ||
692 | /* | |
693 | * Write out and wait upon a list of buffers. | |
694 | * | |
695 | * We have conflicting pressures: we want to make sure that all | |
696 | * initially dirty buffers get waited on, but that any subsequently | |
697 | * dirtied buffers don't. After all, we don't want fsync to last | |
698 | * forever if somebody is actively writing to the file. | |
699 | * | |
700 | * Do this in two main stages: first we copy dirty buffers to a | |
701 | * temporary inode list, queueing the writes as we go. Then we clean | |
702 | * up, waiting for those writes to complete. | |
703 | * | |
704 | * During this second stage, any subsequent updates to the file may end | |
705 | * up refiling the buffer on the original inode's dirty list again, so | |
706 | * there is a chance we will end up with a buffer queued for write but | |
707 | * not yet completed on that list. So, as a final cleanup we go through | |
708 | * the osync code to catch these locked, dirty buffers without requeuing | |
709 | * any newly dirty buffers for write. | |
710 | */ | |
711 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) | |
712 | { | |
713 | struct buffer_head *bh; | |
714 | struct list_head tmp; | |
7eaceacc | 715 | struct address_space *mapping; |
1da177e4 | 716 | int err = 0, err2; |
4ee2491e | 717 | struct blk_plug plug; |
1da177e4 LT |
718 | |
719 | INIT_LIST_HEAD(&tmp); | |
4ee2491e | 720 | blk_start_plug(&plug); |
1da177e4 LT |
721 | |
722 | spin_lock(lock); | |
723 | while (!list_empty(list)) { | |
724 | bh = BH_ENTRY(list->next); | |
535ee2fb | 725 | mapping = bh->b_assoc_map; |
58ff407b | 726 | __remove_assoc_queue(bh); |
535ee2fb JK |
727 | /* Avoid race with mark_buffer_dirty_inode() which does |
728 | * a lockless check and we rely on seeing the dirty bit */ | |
729 | smp_mb(); | |
1da177e4 LT |
730 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
731 | list_add(&bh->b_assoc_buffers, &tmp); | |
535ee2fb | 732 | bh->b_assoc_map = mapping; |
1da177e4 LT |
733 | if (buffer_dirty(bh)) { |
734 | get_bh(bh); | |
735 | spin_unlock(lock); | |
736 | /* | |
737 | * Ensure any pending I/O completes so that | |
9cb569d6 CH |
738 | * write_dirty_buffer() actually writes the |
739 | * current contents - it is a noop if I/O is | |
740 | * still in flight on potentially older | |
741 | * contents. | |
1da177e4 | 742 | */ |
70fd7614 | 743 | write_dirty_buffer(bh, REQ_SYNC); |
9cf6b720 JA |
744 | |
745 | /* | |
746 | * Kick off IO for the previous mapping. Note | |
747 | * that we will not run the very last mapping, | |
748 | * wait_on_buffer() will do that for us | |
749 | * through sync_buffer(). | |
750 | */ | |
1da177e4 LT |
751 | brelse(bh); |
752 | spin_lock(lock); | |
753 | } | |
754 | } | |
755 | } | |
756 | ||
4ee2491e JA |
757 | spin_unlock(lock); |
758 | blk_finish_plug(&plug); | |
759 | spin_lock(lock); | |
760 | ||
1da177e4 LT |
761 | while (!list_empty(&tmp)) { |
762 | bh = BH_ENTRY(tmp.prev); | |
1da177e4 | 763 | get_bh(bh); |
535ee2fb JK |
764 | mapping = bh->b_assoc_map; |
765 | __remove_assoc_queue(bh); | |
766 | /* Avoid race with mark_buffer_dirty_inode() which does | |
767 | * a lockless check and we rely on seeing the dirty bit */ | |
768 | smp_mb(); | |
769 | if (buffer_dirty(bh)) { | |
770 | list_add(&bh->b_assoc_buffers, | |
e3892296 | 771 | &mapping->private_list); |
535ee2fb JK |
772 | bh->b_assoc_map = mapping; |
773 | } | |
1da177e4 LT |
774 | spin_unlock(lock); |
775 | wait_on_buffer(bh); | |
776 | if (!buffer_uptodate(bh)) | |
777 | err = -EIO; | |
778 | brelse(bh); | |
779 | spin_lock(lock); | |
780 | } | |
781 | ||
782 | spin_unlock(lock); | |
783 | err2 = osync_buffers_list(lock, list); | |
784 | if (err) | |
785 | return err; | |
786 | else | |
787 | return err2; | |
788 | } | |
789 | ||
790 | /* | |
791 | * Invalidate any and all dirty buffers on a given inode. We are | |
792 | * probably unmounting the fs, but that doesn't mean we have already | |
793 | * done a sync(). Just drop the buffers from the inode list. | |
794 | * | |
795 | * NOTE: we take the inode's blockdev's mapping's private_lock. Which | |
796 | * assumes that all the buffers are against the blockdev. Not true | |
797 | * for reiserfs. | |
798 | */ | |
799 | void invalidate_inode_buffers(struct inode *inode) | |
800 | { | |
801 | if (inode_has_buffers(inode)) { | |
802 | struct address_space *mapping = &inode->i_data; | |
803 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 804 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
805 | |
806 | spin_lock(&buffer_mapping->private_lock); | |
807 | while (!list_empty(list)) | |
808 | __remove_assoc_queue(BH_ENTRY(list->next)); | |
809 | spin_unlock(&buffer_mapping->private_lock); | |
810 | } | |
811 | } | |
52b19ac9 | 812 | EXPORT_SYMBOL(invalidate_inode_buffers); |
1da177e4 LT |
813 | |
814 | /* | |
815 | * Remove any clean buffers from the inode's buffer list. This is called | |
816 | * when we're trying to free the inode itself. Those buffers can pin it. | |
817 | * | |
818 | * Returns true if all buffers were removed. | |
819 | */ | |
820 | int remove_inode_buffers(struct inode *inode) | |
821 | { | |
822 | int ret = 1; | |
823 | ||
824 | if (inode_has_buffers(inode)) { | |
825 | struct address_space *mapping = &inode->i_data; | |
826 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 827 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
828 | |
829 | spin_lock(&buffer_mapping->private_lock); | |
830 | while (!list_empty(list)) { | |
831 | struct buffer_head *bh = BH_ENTRY(list->next); | |
832 | if (buffer_dirty(bh)) { | |
833 | ret = 0; | |
834 | break; | |
835 | } | |
836 | __remove_assoc_queue(bh); | |
837 | } | |
838 | spin_unlock(&buffer_mapping->private_lock); | |
839 | } | |
840 | return ret; | |
841 | } | |
842 | ||
843 | /* | |
844 | * Create the appropriate buffers when given a page for data area and | |
845 | * the size of each buffer.. Use the bh->b_this_page linked list to | |
846 | * follow the buffers created. Return NULL if unable to create more | |
847 | * buffers. | |
848 | * | |
849 | * The retry flag is used to differentiate async IO (paging, swapping) | |
850 | * which may not fail from ordinary buffer allocations. | |
851 | */ | |
852 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, | |
640ab98f | 853 | bool retry) |
1da177e4 LT |
854 | { |
855 | struct buffer_head *bh, *head; | |
f745c6f5 | 856 | gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; |
1da177e4 | 857 | long offset; |
f745c6f5 | 858 | struct mem_cgroup *memcg; |
1da177e4 | 859 | |
640ab98f JA |
860 | if (retry) |
861 | gfp |= __GFP_NOFAIL; | |
862 | ||
f745c6f5 SB |
863 | memcg = get_mem_cgroup_from_page(page); |
864 | memalloc_use_memcg(memcg); | |
865 | ||
1da177e4 LT |
866 | head = NULL; |
867 | offset = PAGE_SIZE; | |
868 | while ((offset -= size) >= 0) { | |
640ab98f | 869 | bh = alloc_buffer_head(gfp); |
1da177e4 LT |
870 | if (!bh) |
871 | goto no_grow; | |
872 | ||
1da177e4 LT |
873 | bh->b_this_page = head; |
874 | bh->b_blocknr = -1; | |
875 | head = bh; | |
876 | ||
1da177e4 LT |
877 | bh->b_size = size; |
878 | ||
879 | /* Link the buffer to its page */ | |
880 | set_bh_page(bh, page, offset); | |
1da177e4 | 881 | } |
f745c6f5 SB |
882 | out: |
883 | memalloc_unuse_memcg(); | |
884 | mem_cgroup_put(memcg); | |
1da177e4 LT |
885 | return head; |
886 | /* | |
887 | * In case anything failed, we just free everything we got. | |
888 | */ | |
889 | no_grow: | |
890 | if (head) { | |
891 | do { | |
892 | bh = head; | |
893 | head = head->b_this_page; | |
894 | free_buffer_head(bh); | |
895 | } while (head); | |
896 | } | |
897 | ||
f745c6f5 | 898 | goto out; |
1da177e4 LT |
899 | } |
900 | EXPORT_SYMBOL_GPL(alloc_page_buffers); | |
901 | ||
902 | static inline void | |
903 | link_dev_buffers(struct page *page, struct buffer_head *head) | |
904 | { | |
905 | struct buffer_head *bh, *tail; | |
906 | ||
907 | bh = head; | |
908 | do { | |
909 | tail = bh; | |
910 | bh = bh->b_this_page; | |
911 | } while (bh); | |
912 | tail->b_this_page = head; | |
913 | attach_page_buffers(page, head); | |
914 | } | |
915 | ||
bbec0270 LT |
916 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
917 | { | |
918 | sector_t retval = ~((sector_t)0); | |
919 | loff_t sz = i_size_read(bdev->bd_inode); | |
920 | ||
921 | if (sz) { | |
922 | unsigned int sizebits = blksize_bits(size); | |
923 | retval = (sz >> sizebits); | |
924 | } | |
925 | return retval; | |
926 | } | |
927 | ||
1da177e4 LT |
928 | /* |
929 | * Initialise the state of a blockdev page's buffers. | |
930 | */ | |
676ce6d5 | 931 | static sector_t |
1da177e4 LT |
932 | init_page_buffers(struct page *page, struct block_device *bdev, |
933 | sector_t block, int size) | |
934 | { | |
935 | struct buffer_head *head = page_buffers(page); | |
936 | struct buffer_head *bh = head; | |
937 | int uptodate = PageUptodate(page); | |
bbec0270 | 938 | sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); |
1da177e4 LT |
939 | |
940 | do { | |
941 | if (!buffer_mapped(bh)) { | |
01950a34 EB |
942 | bh->b_end_io = NULL; |
943 | bh->b_private = NULL; | |
1da177e4 LT |
944 | bh->b_bdev = bdev; |
945 | bh->b_blocknr = block; | |
946 | if (uptodate) | |
947 | set_buffer_uptodate(bh); | |
080399aa JM |
948 | if (block < end_block) |
949 | set_buffer_mapped(bh); | |
1da177e4 LT |
950 | } |
951 | block++; | |
952 | bh = bh->b_this_page; | |
953 | } while (bh != head); | |
676ce6d5 HD |
954 | |
955 | /* | |
956 | * Caller needs to validate requested block against end of device. | |
957 | */ | |
958 | return end_block; | |
1da177e4 LT |
959 | } |
960 | ||
961 | /* | |
962 | * Create the page-cache page that contains the requested block. | |
963 | * | |
676ce6d5 | 964 | * This is used purely for blockdev mappings. |
1da177e4 | 965 | */ |
676ce6d5 | 966 | static int |
1da177e4 | 967 | grow_dev_page(struct block_device *bdev, sector_t block, |
3b5e6454 | 968 | pgoff_t index, int size, int sizebits, gfp_t gfp) |
1da177e4 LT |
969 | { |
970 | struct inode *inode = bdev->bd_inode; | |
971 | struct page *page; | |
972 | struct buffer_head *bh; | |
676ce6d5 HD |
973 | sector_t end_block; |
974 | int ret = 0; /* Will call free_more_memory() */ | |
84235de3 | 975 | gfp_t gfp_mask; |
1da177e4 | 976 | |
c62d2555 | 977 | gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; |
3b5e6454 | 978 | |
84235de3 JW |
979 | /* |
980 | * XXX: __getblk_slow() can not really deal with failure and | |
981 | * will endlessly loop on improvised global reclaim. Prefer | |
982 | * looping in the allocator rather than here, at least that | |
983 | * code knows what it's doing. | |
984 | */ | |
985 | gfp_mask |= __GFP_NOFAIL; | |
986 | ||
987 | page = find_or_create_page(inode->i_mapping, index, gfp_mask); | |
1da177e4 | 988 | |
e827f923 | 989 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
990 | |
991 | if (page_has_buffers(page)) { | |
992 | bh = page_buffers(page); | |
993 | if (bh->b_size == size) { | |
676ce6d5 | 994 | end_block = init_page_buffers(page, bdev, |
f2d5a944 AA |
995 | (sector_t)index << sizebits, |
996 | size); | |
676ce6d5 | 997 | goto done; |
1da177e4 LT |
998 | } |
999 | if (!try_to_free_buffers(page)) | |
1000 | goto failed; | |
1001 | } | |
1002 | ||
1003 | /* | |
1004 | * Allocate some buffers for this page | |
1005 | */ | |
94dc24c0 | 1006 | bh = alloc_page_buffers(page, size, true); |
1da177e4 LT |
1007 | |
1008 | /* | |
1009 | * Link the page to the buffers and initialise them. Take the | |
1010 | * lock to be atomic wrt __find_get_block(), which does not | |
1011 | * run under the page lock. | |
1012 | */ | |
1013 | spin_lock(&inode->i_mapping->private_lock); | |
1014 | link_dev_buffers(page, bh); | |
f2d5a944 AA |
1015 | end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, |
1016 | size); | |
1da177e4 | 1017 | spin_unlock(&inode->i_mapping->private_lock); |
676ce6d5 HD |
1018 | done: |
1019 | ret = (block < end_block) ? 1 : -ENXIO; | |
1da177e4 | 1020 | failed: |
1da177e4 | 1021 | unlock_page(page); |
09cbfeaf | 1022 | put_page(page); |
676ce6d5 | 1023 | return ret; |
1da177e4 LT |
1024 | } |
1025 | ||
1026 | /* | |
1027 | * Create buffers for the specified block device block's page. If | |
1028 | * that page was dirty, the buffers are set dirty also. | |
1da177e4 | 1029 | */ |
858119e1 | 1030 | static int |
3b5e6454 | 1031 | grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) |
1da177e4 | 1032 | { |
1da177e4 LT |
1033 | pgoff_t index; |
1034 | int sizebits; | |
1035 | ||
1036 | sizebits = -1; | |
1037 | do { | |
1038 | sizebits++; | |
1039 | } while ((size << sizebits) < PAGE_SIZE); | |
1040 | ||
1041 | index = block >> sizebits; | |
1da177e4 | 1042 | |
e5657933 AM |
1043 | /* |
1044 | * Check for a block which wants to lie outside our maximum possible | |
1045 | * pagecache index. (this comparison is done using sector_t types). | |
1046 | */ | |
1047 | if (unlikely(index != block >> sizebits)) { | |
e5657933 | 1048 | printk(KERN_ERR "%s: requested out-of-range block %llu for " |
a1c6f057 | 1049 | "device %pg\n", |
8e24eea7 | 1050 | __func__, (unsigned long long)block, |
a1c6f057 | 1051 | bdev); |
e5657933 AM |
1052 | return -EIO; |
1053 | } | |
676ce6d5 | 1054 | |
1da177e4 | 1055 | /* Create a page with the proper size buffers.. */ |
3b5e6454 | 1056 | return grow_dev_page(bdev, block, index, size, sizebits, gfp); |
1da177e4 LT |
1057 | } |
1058 | ||
0026ba40 | 1059 | static struct buffer_head * |
3b5e6454 GK |
1060 | __getblk_slow(struct block_device *bdev, sector_t block, |
1061 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1062 | { |
1063 | /* Size must be multiple of hard sectorsize */ | |
e1defc4f | 1064 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
1da177e4 LT |
1065 | (size < 512 || size > PAGE_SIZE))) { |
1066 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", | |
1067 | size); | |
e1defc4f MP |
1068 | printk(KERN_ERR "logical block size: %d\n", |
1069 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1070 | |
1071 | dump_stack(); | |
1072 | return NULL; | |
1073 | } | |
1074 | ||
676ce6d5 HD |
1075 | for (;;) { |
1076 | struct buffer_head *bh; | |
1077 | int ret; | |
1da177e4 LT |
1078 | |
1079 | bh = __find_get_block(bdev, block, size); | |
1080 | if (bh) | |
1081 | return bh; | |
676ce6d5 | 1082 | |
3b5e6454 | 1083 | ret = grow_buffers(bdev, block, size, gfp); |
676ce6d5 HD |
1084 | if (ret < 0) |
1085 | return NULL; | |
1da177e4 LT |
1086 | } |
1087 | } | |
1088 | ||
1089 | /* | |
1090 | * The relationship between dirty buffers and dirty pages: | |
1091 | * | |
1092 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and | |
ec82e1c1 | 1093 | * the page is tagged dirty in the page cache. |
1da177e4 LT |
1094 | * |
1095 | * At all times, the dirtiness of the buffers represents the dirtiness of | |
1096 | * subsections of the page. If the page has buffers, the page dirty bit is | |
1097 | * merely a hint about the true dirty state. | |
1098 | * | |
1099 | * When a page is set dirty in its entirety, all its buffers are marked dirty | |
1100 | * (if the page has buffers). | |
1101 | * | |
1102 | * When a buffer is marked dirty, its page is dirtied, but the page's other | |
1103 | * buffers are not. | |
1104 | * | |
1105 | * Also. When blockdev buffers are explicitly read with bread(), they | |
1106 | * individually become uptodate. But their backing page remains not | |
1107 | * uptodate - even if all of its buffers are uptodate. A subsequent | |
1108 | * block_read_full_page() against that page will discover all the uptodate | |
1109 | * buffers, will set the page uptodate and will perform no I/O. | |
1110 | */ | |
1111 | ||
1112 | /** | |
1113 | * mark_buffer_dirty - mark a buffer_head as needing writeout | |
67be2dd1 | 1114 | * @bh: the buffer_head to mark dirty |
1da177e4 | 1115 | * |
ec82e1c1 MW |
1116 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set |
1117 | * its backing page dirty, then tag the page as dirty in the page cache | |
1118 | * and then attach the address_space's inode to its superblock's dirty | |
1da177e4 LT |
1119 | * inode list. |
1120 | * | |
1121 | * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, | |
b93b0163 | 1122 | * i_pages lock and mapping->host->i_lock. |
1da177e4 | 1123 | */ |
fc9b52cd | 1124 | void mark_buffer_dirty(struct buffer_head *bh) |
1da177e4 | 1125 | { |
787d2214 | 1126 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
1be62dc1 | 1127 | |
5305cb83 TH |
1128 | trace_block_dirty_buffer(bh); |
1129 | ||
1be62dc1 LT |
1130 | /* |
1131 | * Very *carefully* optimize the it-is-already-dirty case. | |
1132 | * | |
1133 | * Don't let the final "is it dirty" escape to before we | |
1134 | * perhaps modified the buffer. | |
1135 | */ | |
1136 | if (buffer_dirty(bh)) { | |
1137 | smp_mb(); | |
1138 | if (buffer_dirty(bh)) | |
1139 | return; | |
1140 | } | |
1141 | ||
a8e7d49a LT |
1142 | if (!test_set_buffer_dirty(bh)) { |
1143 | struct page *page = bh->b_page; | |
c4843a75 | 1144 | struct address_space *mapping = NULL; |
c4843a75 | 1145 | |
62cccb8c | 1146 | lock_page_memcg(page); |
8e9d78ed | 1147 | if (!TestSetPageDirty(page)) { |
c4843a75 | 1148 | mapping = page_mapping(page); |
8e9d78ed | 1149 | if (mapping) |
62cccb8c | 1150 | __set_page_dirty(page, mapping, 0); |
8e9d78ed | 1151 | } |
62cccb8c | 1152 | unlock_page_memcg(page); |
c4843a75 GT |
1153 | if (mapping) |
1154 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
a8e7d49a | 1155 | } |
1da177e4 | 1156 | } |
1fe72eaa | 1157 | EXPORT_SYMBOL(mark_buffer_dirty); |
1da177e4 | 1158 | |
87354e5d JL |
1159 | void mark_buffer_write_io_error(struct buffer_head *bh) |
1160 | { | |
1161 | set_buffer_write_io_error(bh); | |
1162 | /* FIXME: do we need to set this in both places? */ | |
1163 | if (bh->b_page && bh->b_page->mapping) | |
1164 | mapping_set_error(bh->b_page->mapping, -EIO); | |
1165 | if (bh->b_assoc_map) | |
1166 | mapping_set_error(bh->b_assoc_map, -EIO); | |
1167 | } | |
1168 | EXPORT_SYMBOL(mark_buffer_write_io_error); | |
1169 | ||
1da177e4 LT |
1170 | /* |
1171 | * Decrement a buffer_head's reference count. If all buffers against a page | |
1172 | * have zero reference count, are clean and unlocked, and if the page is clean | |
1173 | * and unlocked then try_to_free_buffers() may strip the buffers from the page | |
1174 | * in preparation for freeing it (sometimes, rarely, buffers are removed from | |
1175 | * a page but it ends up not being freed, and buffers may later be reattached). | |
1176 | */ | |
1177 | void __brelse(struct buffer_head * buf) | |
1178 | { | |
1179 | if (atomic_read(&buf->b_count)) { | |
1180 | put_bh(buf); | |
1181 | return; | |
1182 | } | |
5c752ad9 | 1183 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
1da177e4 | 1184 | } |
1fe72eaa | 1185 | EXPORT_SYMBOL(__brelse); |
1da177e4 LT |
1186 | |
1187 | /* | |
1188 | * bforget() is like brelse(), except it discards any | |
1189 | * potentially dirty data. | |
1190 | */ | |
1191 | void __bforget(struct buffer_head *bh) | |
1192 | { | |
1193 | clear_buffer_dirty(bh); | |
535ee2fb | 1194 | if (bh->b_assoc_map) { |
1da177e4 LT |
1195 | struct address_space *buffer_mapping = bh->b_page->mapping; |
1196 | ||
1197 | spin_lock(&buffer_mapping->private_lock); | |
1198 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 1199 | bh->b_assoc_map = NULL; |
1da177e4 LT |
1200 | spin_unlock(&buffer_mapping->private_lock); |
1201 | } | |
1202 | __brelse(bh); | |
1203 | } | |
1fe72eaa | 1204 | EXPORT_SYMBOL(__bforget); |
1da177e4 LT |
1205 | |
1206 | static struct buffer_head *__bread_slow(struct buffer_head *bh) | |
1207 | { | |
1208 | lock_buffer(bh); | |
1209 | if (buffer_uptodate(bh)) { | |
1210 | unlock_buffer(bh); | |
1211 | return bh; | |
1212 | } else { | |
1213 | get_bh(bh); | |
1214 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 1215 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
1216 | wait_on_buffer(bh); |
1217 | if (buffer_uptodate(bh)) | |
1218 | return bh; | |
1219 | } | |
1220 | brelse(bh); | |
1221 | return NULL; | |
1222 | } | |
1223 | ||
1224 | /* | |
1225 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). | |
1226 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their | |
1227 | * refcount elevated by one when they're in an LRU. A buffer can only appear | |
1228 | * once in a particular CPU's LRU. A single buffer can be present in multiple | |
1229 | * CPU's LRUs at the same time. | |
1230 | * | |
1231 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and | |
1232 | * sb_find_get_block(). | |
1233 | * | |
1234 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use | |
1235 | * a local interrupt disable for that. | |
1236 | */ | |
1237 | ||
86cf78d7 | 1238 | #define BH_LRU_SIZE 16 |
1da177e4 LT |
1239 | |
1240 | struct bh_lru { | |
1241 | struct buffer_head *bhs[BH_LRU_SIZE]; | |
1242 | }; | |
1243 | ||
1244 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; | |
1245 | ||
1246 | #ifdef CONFIG_SMP | |
1247 | #define bh_lru_lock() local_irq_disable() | |
1248 | #define bh_lru_unlock() local_irq_enable() | |
1249 | #else | |
1250 | #define bh_lru_lock() preempt_disable() | |
1251 | #define bh_lru_unlock() preempt_enable() | |
1252 | #endif | |
1253 | ||
1254 | static inline void check_irqs_on(void) | |
1255 | { | |
1256 | #ifdef irqs_disabled | |
1257 | BUG_ON(irqs_disabled()); | |
1258 | #endif | |
1259 | } | |
1260 | ||
1261 | /* | |
241f01fb EB |
1262 | * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is |
1263 | * inserted at the front, and the buffer_head at the back if any is evicted. | |
1264 | * Or, if already in the LRU it is moved to the front. | |
1da177e4 LT |
1265 | */ |
1266 | static void bh_lru_install(struct buffer_head *bh) | |
1267 | { | |
241f01fb EB |
1268 | struct buffer_head *evictee = bh; |
1269 | struct bh_lru *b; | |
1270 | int i; | |
1da177e4 LT |
1271 | |
1272 | check_irqs_on(); | |
1273 | bh_lru_lock(); | |
1da177e4 | 1274 | |
241f01fb EB |
1275 | b = this_cpu_ptr(&bh_lrus); |
1276 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1277 | swap(evictee, b->bhs[i]); | |
1278 | if (evictee == bh) { | |
1279 | bh_lru_unlock(); | |
1280 | return; | |
1da177e4 | 1281 | } |
1da177e4 | 1282 | } |
1da177e4 | 1283 | |
241f01fb EB |
1284 | get_bh(bh); |
1285 | bh_lru_unlock(); | |
1286 | brelse(evictee); | |
1da177e4 LT |
1287 | } |
1288 | ||
1289 | /* | |
1290 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. | |
1291 | */ | |
858119e1 | 1292 | static struct buffer_head * |
3991d3bd | 1293 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1294 | { |
1295 | struct buffer_head *ret = NULL; | |
3991d3bd | 1296 | unsigned int i; |
1da177e4 LT |
1297 | |
1298 | check_irqs_on(); | |
1299 | bh_lru_lock(); | |
1da177e4 | 1300 | for (i = 0; i < BH_LRU_SIZE; i++) { |
c7b92516 | 1301 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
1da177e4 | 1302 | |
9470dd5d ZB |
1303 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
1304 | bh->b_size == size) { | |
1da177e4 LT |
1305 | if (i) { |
1306 | while (i) { | |
c7b92516 CL |
1307 | __this_cpu_write(bh_lrus.bhs[i], |
1308 | __this_cpu_read(bh_lrus.bhs[i - 1])); | |
1da177e4 LT |
1309 | i--; |
1310 | } | |
c7b92516 | 1311 | __this_cpu_write(bh_lrus.bhs[0], bh); |
1da177e4 LT |
1312 | } |
1313 | get_bh(bh); | |
1314 | ret = bh; | |
1315 | break; | |
1316 | } | |
1317 | } | |
1318 | bh_lru_unlock(); | |
1319 | return ret; | |
1320 | } | |
1321 | ||
1322 | /* | |
1323 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh | |
1324 | * it in the LRU and mark it as accessed. If it is not present then return | |
1325 | * NULL | |
1326 | */ | |
1327 | struct buffer_head * | |
3991d3bd | 1328 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1329 | { |
1330 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); | |
1331 | ||
1332 | if (bh == NULL) { | |
2457aec6 | 1333 | /* __find_get_block_slow will mark the page accessed */ |
385fd4c5 | 1334 | bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1335 | if (bh) |
1336 | bh_lru_install(bh); | |
2457aec6 | 1337 | } else |
1da177e4 | 1338 | touch_buffer(bh); |
2457aec6 | 1339 | |
1da177e4 LT |
1340 | return bh; |
1341 | } | |
1342 | EXPORT_SYMBOL(__find_get_block); | |
1343 | ||
1344 | /* | |
3b5e6454 | 1345 | * __getblk_gfp() will locate (and, if necessary, create) the buffer_head |
1da177e4 LT |
1346 | * which corresponds to the passed block_device, block and size. The |
1347 | * returned buffer has its reference count incremented. | |
1348 | * | |
3b5e6454 GK |
1349 | * __getblk_gfp() will lock up the machine if grow_dev_page's |
1350 | * try_to_free_buffers() attempt is failing. FIXME, perhaps? | |
1da177e4 LT |
1351 | */ |
1352 | struct buffer_head * | |
3b5e6454 GK |
1353 | __getblk_gfp(struct block_device *bdev, sector_t block, |
1354 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1355 | { |
1356 | struct buffer_head *bh = __find_get_block(bdev, block, size); | |
1357 | ||
1358 | might_sleep(); | |
1359 | if (bh == NULL) | |
3b5e6454 | 1360 | bh = __getblk_slow(bdev, block, size, gfp); |
1da177e4 LT |
1361 | return bh; |
1362 | } | |
3b5e6454 | 1363 | EXPORT_SYMBOL(__getblk_gfp); |
1da177e4 LT |
1364 | |
1365 | /* | |
1366 | * Do async read-ahead on a buffer.. | |
1367 | */ | |
3991d3bd | 1368 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1369 | { |
1370 | struct buffer_head *bh = __getblk(bdev, block, size); | |
a3e713b5 | 1371 | if (likely(bh)) { |
70246286 | 1372 | ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); |
a3e713b5 AM |
1373 | brelse(bh); |
1374 | } | |
1da177e4 LT |
1375 | } |
1376 | EXPORT_SYMBOL(__breadahead); | |
1377 | ||
1378 | /** | |
3b5e6454 | 1379 | * __bread_gfp() - reads a specified block and returns the bh |
67be2dd1 | 1380 | * @bdev: the block_device to read from |
1da177e4 LT |
1381 | * @block: number of block |
1382 | * @size: size (in bytes) to read | |
3b5e6454 GK |
1383 | * @gfp: page allocation flag |
1384 | * | |
1da177e4 | 1385 | * Reads a specified block, and returns buffer head that contains it. |
3b5e6454 GK |
1386 | * The page cache can be allocated from non-movable area |
1387 | * not to prevent page migration if you set gfp to zero. | |
1da177e4 LT |
1388 | * It returns NULL if the block was unreadable. |
1389 | */ | |
1390 | struct buffer_head * | |
3b5e6454 GK |
1391 | __bread_gfp(struct block_device *bdev, sector_t block, |
1392 | unsigned size, gfp_t gfp) | |
1da177e4 | 1393 | { |
3b5e6454 | 1394 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); |
1da177e4 | 1395 | |
a3e713b5 | 1396 | if (likely(bh) && !buffer_uptodate(bh)) |
1da177e4 LT |
1397 | bh = __bread_slow(bh); |
1398 | return bh; | |
1399 | } | |
3b5e6454 | 1400 | EXPORT_SYMBOL(__bread_gfp); |
1da177e4 LT |
1401 | |
1402 | /* | |
1403 | * invalidate_bh_lrus() is called rarely - but not only at unmount. | |
1404 | * This doesn't race because it runs in each cpu either in irq | |
1405 | * or with preempt disabled. | |
1406 | */ | |
1407 | static void invalidate_bh_lru(void *arg) | |
1408 | { | |
1409 | struct bh_lru *b = &get_cpu_var(bh_lrus); | |
1410 | int i; | |
1411 | ||
1412 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1413 | brelse(b->bhs[i]); | |
1414 | b->bhs[i] = NULL; | |
1415 | } | |
1416 | put_cpu_var(bh_lrus); | |
1417 | } | |
42be35d0 GBY |
1418 | |
1419 | static bool has_bh_in_lru(int cpu, void *dummy) | |
1420 | { | |
1421 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); | |
1422 | int i; | |
1da177e4 | 1423 | |
42be35d0 GBY |
1424 | for (i = 0; i < BH_LRU_SIZE; i++) { |
1425 | if (b->bhs[i]) | |
1426 | return 1; | |
1427 | } | |
1428 | ||
1429 | return 0; | |
1430 | } | |
1431 | ||
f9a14399 | 1432 | void invalidate_bh_lrus(void) |
1da177e4 | 1433 | { |
42be35d0 | 1434 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); |
1da177e4 | 1435 | } |
9db5579b | 1436 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
1da177e4 LT |
1437 | |
1438 | void set_bh_page(struct buffer_head *bh, | |
1439 | struct page *page, unsigned long offset) | |
1440 | { | |
1441 | bh->b_page = page; | |
e827f923 | 1442 | BUG_ON(offset >= PAGE_SIZE); |
1da177e4 LT |
1443 | if (PageHighMem(page)) |
1444 | /* | |
1445 | * This catches illegal uses and preserves the offset: | |
1446 | */ | |
1447 | bh->b_data = (char *)(0 + offset); | |
1448 | else | |
1449 | bh->b_data = page_address(page) + offset; | |
1450 | } | |
1451 | EXPORT_SYMBOL(set_bh_page); | |
1452 | ||
1453 | /* | |
1454 | * Called when truncating a buffer on a page completely. | |
1455 | */ | |
e7470ee8 MG |
1456 | |
1457 | /* Bits that are cleared during an invalidate */ | |
1458 | #define BUFFER_FLAGS_DISCARD \ | |
1459 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ | |
1460 | 1 << BH_Delay | 1 << BH_Unwritten) | |
1461 | ||
858119e1 | 1462 | static void discard_buffer(struct buffer_head * bh) |
1da177e4 | 1463 | { |
e7470ee8 MG |
1464 | unsigned long b_state, b_state_old; |
1465 | ||
1da177e4 LT |
1466 | lock_buffer(bh); |
1467 | clear_buffer_dirty(bh); | |
1468 | bh->b_bdev = NULL; | |
e7470ee8 MG |
1469 | b_state = bh->b_state; |
1470 | for (;;) { | |
1471 | b_state_old = cmpxchg(&bh->b_state, b_state, | |
1472 | (b_state & ~BUFFER_FLAGS_DISCARD)); | |
1473 | if (b_state_old == b_state) | |
1474 | break; | |
1475 | b_state = b_state_old; | |
1476 | } | |
1da177e4 LT |
1477 | unlock_buffer(bh); |
1478 | } | |
1479 | ||
1da177e4 | 1480 | /** |
814e1d25 | 1481 | * block_invalidatepage - invalidate part or all of a buffer-backed page |
1da177e4 LT |
1482 | * |
1483 | * @page: the page which is affected | |
d47992f8 LC |
1484 | * @offset: start of the range to invalidate |
1485 | * @length: length of the range to invalidate | |
1da177e4 LT |
1486 | * |
1487 | * block_invalidatepage() is called when all or part of the page has become | |
814e1d25 | 1488 | * invalidated by a truncate operation. |
1da177e4 LT |
1489 | * |
1490 | * block_invalidatepage() does not have to release all buffers, but it must | |
1491 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
1492 | * is underway against any of the blocks which are outside the truncation | |
1493 | * point. Because the caller is about to free (and possibly reuse) those | |
1494 | * blocks on-disk. | |
1495 | */ | |
d47992f8 LC |
1496 | void block_invalidatepage(struct page *page, unsigned int offset, |
1497 | unsigned int length) | |
1da177e4 LT |
1498 | { |
1499 | struct buffer_head *head, *bh, *next; | |
1500 | unsigned int curr_off = 0; | |
d47992f8 | 1501 | unsigned int stop = length + offset; |
1da177e4 LT |
1502 | |
1503 | BUG_ON(!PageLocked(page)); | |
1504 | if (!page_has_buffers(page)) | |
1505 | goto out; | |
1506 | ||
d47992f8 LC |
1507 | /* |
1508 | * Check for overflow | |
1509 | */ | |
09cbfeaf | 1510 | BUG_ON(stop > PAGE_SIZE || stop < length); |
d47992f8 | 1511 | |
1da177e4 LT |
1512 | head = page_buffers(page); |
1513 | bh = head; | |
1514 | do { | |
1515 | unsigned int next_off = curr_off + bh->b_size; | |
1516 | next = bh->b_this_page; | |
1517 | ||
d47992f8 LC |
1518 | /* |
1519 | * Are we still fully in range ? | |
1520 | */ | |
1521 | if (next_off > stop) | |
1522 | goto out; | |
1523 | ||
1da177e4 LT |
1524 | /* |
1525 | * is this block fully invalidated? | |
1526 | */ | |
1527 | if (offset <= curr_off) | |
1528 | discard_buffer(bh); | |
1529 | curr_off = next_off; | |
1530 | bh = next; | |
1531 | } while (bh != head); | |
1532 | ||
1533 | /* | |
1534 | * We release buffers only if the entire page is being invalidated. | |
1535 | * The get_block cached value has been unconditionally invalidated, | |
1536 | * so real IO is not possible anymore. | |
1537 | */ | |
3172485f | 1538 | if (length == PAGE_SIZE) |
2ff28e22 | 1539 | try_to_release_page(page, 0); |
1da177e4 | 1540 | out: |
2ff28e22 | 1541 | return; |
1da177e4 LT |
1542 | } |
1543 | EXPORT_SYMBOL(block_invalidatepage); | |
1544 | ||
d47992f8 | 1545 | |
1da177e4 LT |
1546 | /* |
1547 | * We attach and possibly dirty the buffers atomically wrt | |
1548 | * __set_page_dirty_buffers() via private_lock. try_to_free_buffers | |
1549 | * is already excluded via the page lock. | |
1550 | */ | |
1551 | void create_empty_buffers(struct page *page, | |
1552 | unsigned long blocksize, unsigned long b_state) | |
1553 | { | |
1554 | struct buffer_head *bh, *head, *tail; | |
1555 | ||
640ab98f | 1556 | head = alloc_page_buffers(page, blocksize, true); |
1da177e4 LT |
1557 | bh = head; |
1558 | do { | |
1559 | bh->b_state |= b_state; | |
1560 | tail = bh; | |
1561 | bh = bh->b_this_page; | |
1562 | } while (bh); | |
1563 | tail->b_this_page = head; | |
1564 | ||
1565 | spin_lock(&page->mapping->private_lock); | |
1566 | if (PageUptodate(page) || PageDirty(page)) { | |
1567 | bh = head; | |
1568 | do { | |
1569 | if (PageDirty(page)) | |
1570 | set_buffer_dirty(bh); | |
1571 | if (PageUptodate(page)) | |
1572 | set_buffer_uptodate(bh); | |
1573 | bh = bh->b_this_page; | |
1574 | } while (bh != head); | |
1575 | } | |
1576 | attach_page_buffers(page, head); | |
1577 | spin_unlock(&page->mapping->private_lock); | |
1578 | } | |
1579 | EXPORT_SYMBOL(create_empty_buffers); | |
1580 | ||
29f3ad7d JK |
1581 | /** |
1582 | * clean_bdev_aliases: clean a range of buffers in block device | |
1583 | * @bdev: Block device to clean buffers in | |
1584 | * @block: Start of a range of blocks to clean | |
1585 | * @len: Number of blocks to clean | |
1da177e4 | 1586 | * |
29f3ad7d JK |
1587 | * We are taking a range of blocks for data and we don't want writeback of any |
1588 | * buffer-cache aliases starting from return from this function and until the | |
1589 | * moment when something will explicitly mark the buffer dirty (hopefully that | |
1590 | * will not happen until we will free that block ;-) We don't even need to mark | |
1591 | * it not-uptodate - nobody can expect anything from a newly allocated buffer | |
1592 | * anyway. We used to use unmap_buffer() for such invalidation, but that was | |
1593 | * wrong. We definitely don't want to mark the alias unmapped, for example - it | |
1594 | * would confuse anyone who might pick it with bread() afterwards... | |
1595 | * | |
1596 | * Also.. Note that bforget() doesn't lock the buffer. So there can be | |
1597 | * writeout I/O going on against recently-freed buffers. We don't wait on that | |
1598 | * I/O in bforget() - it's more efficient to wait on the I/O only if we really | |
1599 | * need to. That happens here. | |
1da177e4 | 1600 | */ |
29f3ad7d | 1601 | void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) |
1da177e4 | 1602 | { |
29f3ad7d JK |
1603 | struct inode *bd_inode = bdev->bd_inode; |
1604 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
1605 | struct pagevec pvec; | |
1606 | pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); | |
1607 | pgoff_t end; | |
c10f778d | 1608 | int i, count; |
29f3ad7d JK |
1609 | struct buffer_head *bh; |
1610 | struct buffer_head *head; | |
1da177e4 | 1611 | |
29f3ad7d | 1612 | end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); |
86679820 | 1613 | pagevec_init(&pvec); |
397162ff | 1614 | while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) { |
c10f778d JK |
1615 | count = pagevec_count(&pvec); |
1616 | for (i = 0; i < count; i++) { | |
29f3ad7d | 1617 | struct page *page = pvec.pages[i]; |
1da177e4 | 1618 | |
29f3ad7d JK |
1619 | if (!page_has_buffers(page)) |
1620 | continue; | |
1621 | /* | |
1622 | * We use page lock instead of bd_mapping->private_lock | |
1623 | * to pin buffers here since we can afford to sleep and | |
1624 | * it scales better than a global spinlock lock. | |
1625 | */ | |
1626 | lock_page(page); | |
1627 | /* Recheck when the page is locked which pins bhs */ | |
1628 | if (!page_has_buffers(page)) | |
1629 | goto unlock_page; | |
1630 | head = page_buffers(page); | |
1631 | bh = head; | |
1632 | do { | |
6c006a9d | 1633 | if (!buffer_mapped(bh) || (bh->b_blocknr < block)) |
29f3ad7d JK |
1634 | goto next; |
1635 | if (bh->b_blocknr >= block + len) | |
1636 | break; | |
1637 | clear_buffer_dirty(bh); | |
1638 | wait_on_buffer(bh); | |
1639 | clear_buffer_req(bh); | |
1640 | next: | |
1641 | bh = bh->b_this_page; | |
1642 | } while (bh != head); | |
1643 | unlock_page: | |
1644 | unlock_page(page); | |
1645 | } | |
1646 | pagevec_release(&pvec); | |
1647 | cond_resched(); | |
c10f778d JK |
1648 | /* End of range already reached? */ |
1649 | if (index > end || !index) | |
1650 | break; | |
1da177e4 LT |
1651 | } |
1652 | } | |
29f3ad7d | 1653 | EXPORT_SYMBOL(clean_bdev_aliases); |
1da177e4 | 1654 | |
45bce8f3 LT |
1655 | /* |
1656 | * Size is a power-of-two in the range 512..PAGE_SIZE, | |
1657 | * and the case we care about most is PAGE_SIZE. | |
1658 | * | |
1659 | * So this *could* possibly be written with those | |
1660 | * constraints in mind (relevant mostly if some | |
1661 | * architecture has a slow bit-scan instruction) | |
1662 | */ | |
1663 | static inline int block_size_bits(unsigned int blocksize) | |
1664 | { | |
1665 | return ilog2(blocksize); | |
1666 | } | |
1667 | ||
1668 | static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) | |
1669 | { | |
1670 | BUG_ON(!PageLocked(page)); | |
1671 | ||
1672 | if (!page_has_buffers(page)) | |
6aa7de05 MR |
1673 | create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits), |
1674 | b_state); | |
45bce8f3 LT |
1675 | return page_buffers(page); |
1676 | } | |
1677 | ||
1da177e4 LT |
1678 | /* |
1679 | * NOTE! All mapped/uptodate combinations are valid: | |
1680 | * | |
1681 | * Mapped Uptodate Meaning | |
1682 | * | |
1683 | * No No "unknown" - must do get_block() | |
1684 | * No Yes "hole" - zero-filled | |
1685 | * Yes No "allocated" - allocated on disk, not read in | |
1686 | * Yes Yes "valid" - allocated and up-to-date in memory. | |
1687 | * | |
1688 | * "Dirty" is valid only with the last case (mapped+uptodate). | |
1689 | */ | |
1690 | ||
1691 | /* | |
1692 | * While block_write_full_page is writing back the dirty buffers under | |
1693 | * the page lock, whoever dirtied the buffers may decide to clean them | |
1694 | * again at any time. We handle that by only looking at the buffer | |
1695 | * state inside lock_buffer(). | |
1696 | * | |
1697 | * If block_write_full_page() is called for regular writeback | |
1698 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a | |
1699 | * locked buffer. This only can happen if someone has written the buffer | |
1700 | * directly, with submit_bh(). At the address_space level PageWriteback | |
1701 | * prevents this contention from occurring. | |
6e34eedd TT |
1702 | * |
1703 | * If block_write_full_page() is called with wbc->sync_mode == | |
70fd7614 | 1704 | * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this |
721a9602 | 1705 | * causes the writes to be flagged as synchronous writes. |
1da177e4 | 1706 | */ |
b4bba389 | 1707 | int __block_write_full_page(struct inode *inode, struct page *page, |
35c80d5f CM |
1708 | get_block_t *get_block, struct writeback_control *wbc, |
1709 | bh_end_io_t *handler) | |
1da177e4 LT |
1710 | { |
1711 | int err; | |
1712 | sector_t block; | |
1713 | sector_t last_block; | |
f0fbd5fc | 1714 | struct buffer_head *bh, *head; |
45bce8f3 | 1715 | unsigned int blocksize, bbits; |
1da177e4 | 1716 | int nr_underway = 0; |
7637241e | 1717 | int write_flags = wbc_to_write_flags(wbc); |
1da177e4 | 1718 | |
45bce8f3 | 1719 | head = create_page_buffers(page, inode, |
1da177e4 | 1720 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
1da177e4 LT |
1721 | |
1722 | /* | |
1723 | * Be very careful. We have no exclusion from __set_page_dirty_buffers | |
1724 | * here, and the (potentially unmapped) buffers may become dirty at | |
1725 | * any time. If a buffer becomes dirty here after we've inspected it | |
1726 | * then we just miss that fact, and the page stays dirty. | |
1727 | * | |
1728 | * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; | |
1729 | * handle that here by just cleaning them. | |
1730 | */ | |
1731 | ||
1da177e4 | 1732 | bh = head; |
45bce8f3 LT |
1733 | blocksize = bh->b_size; |
1734 | bbits = block_size_bits(blocksize); | |
1735 | ||
09cbfeaf | 1736 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 1737 | last_block = (i_size_read(inode) - 1) >> bbits; |
1da177e4 LT |
1738 | |
1739 | /* | |
1740 | * Get all the dirty buffers mapped to disk addresses and | |
1741 | * handle any aliases from the underlying blockdev's mapping. | |
1742 | */ | |
1743 | do { | |
1744 | if (block > last_block) { | |
1745 | /* | |
1746 | * mapped buffers outside i_size will occur, because | |
1747 | * this page can be outside i_size when there is a | |
1748 | * truncate in progress. | |
1749 | */ | |
1750 | /* | |
1751 | * The buffer was zeroed by block_write_full_page() | |
1752 | */ | |
1753 | clear_buffer_dirty(bh); | |
1754 | set_buffer_uptodate(bh); | |
29a814d2 AT |
1755 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
1756 | buffer_dirty(bh)) { | |
b0cf2321 | 1757 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1758 | err = get_block(inode, block, bh, 1); |
1759 | if (err) | |
1760 | goto recover; | |
29a814d2 | 1761 | clear_buffer_delay(bh); |
1da177e4 LT |
1762 | if (buffer_new(bh)) { |
1763 | /* blockdev mappings never come here */ | |
1764 | clear_buffer_new(bh); | |
e64855c6 | 1765 | clean_bdev_bh_alias(bh); |
1da177e4 LT |
1766 | } |
1767 | } | |
1768 | bh = bh->b_this_page; | |
1769 | block++; | |
1770 | } while (bh != head); | |
1771 | ||
1772 | do { | |
1da177e4 LT |
1773 | if (!buffer_mapped(bh)) |
1774 | continue; | |
1775 | /* | |
1776 | * If it's a fully non-blocking write attempt and we cannot | |
1777 | * lock the buffer then redirty the page. Note that this can | |
5b0830cb JA |
1778 | * potentially cause a busy-wait loop from writeback threads |
1779 | * and kswapd activity, but those code paths have their own | |
1780 | * higher-level throttling. | |
1da177e4 | 1781 | */ |
1b430bee | 1782 | if (wbc->sync_mode != WB_SYNC_NONE) { |
1da177e4 | 1783 | lock_buffer(bh); |
ca5de404 | 1784 | } else if (!trylock_buffer(bh)) { |
1da177e4 LT |
1785 | redirty_page_for_writepage(wbc, page); |
1786 | continue; | |
1787 | } | |
1788 | if (test_clear_buffer_dirty(bh)) { | |
35c80d5f | 1789 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1790 | } else { |
1791 | unlock_buffer(bh); | |
1792 | } | |
1793 | } while ((bh = bh->b_this_page) != head); | |
1794 | ||
1795 | /* | |
1796 | * The page and its buffers are protected by PageWriteback(), so we can | |
1797 | * drop the bh refcounts early. | |
1798 | */ | |
1799 | BUG_ON(PageWriteback(page)); | |
1800 | set_page_writeback(page); | |
1da177e4 LT |
1801 | |
1802 | do { | |
1803 | struct buffer_head *next = bh->b_this_page; | |
1804 | if (buffer_async_write(bh)) { | |
8e8f9298 JA |
1805 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1806 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1807 | nr_underway++; |
1808 | } | |
1da177e4 LT |
1809 | bh = next; |
1810 | } while (bh != head); | |
05937baa | 1811 | unlock_page(page); |
1da177e4 LT |
1812 | |
1813 | err = 0; | |
1814 | done: | |
1815 | if (nr_underway == 0) { | |
1816 | /* | |
1817 | * The page was marked dirty, but the buffers were | |
1818 | * clean. Someone wrote them back by hand with | |
1819 | * ll_rw_block/submit_bh. A rare case. | |
1820 | */ | |
1da177e4 | 1821 | end_page_writeback(page); |
3d67f2d7 | 1822 | |
1da177e4 LT |
1823 | /* |
1824 | * The page and buffer_heads can be released at any time from | |
1825 | * here on. | |
1826 | */ | |
1da177e4 LT |
1827 | } |
1828 | return err; | |
1829 | ||
1830 | recover: | |
1831 | /* | |
1832 | * ENOSPC, or some other error. We may already have added some | |
1833 | * blocks to the file, so we need to write these out to avoid | |
1834 | * exposing stale data. | |
1835 | * The page is currently locked and not marked for writeback | |
1836 | */ | |
1837 | bh = head; | |
1838 | /* Recovery: lock and submit the mapped buffers */ | |
1839 | do { | |
29a814d2 AT |
1840 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
1841 | !buffer_delay(bh)) { | |
1da177e4 | 1842 | lock_buffer(bh); |
35c80d5f | 1843 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1844 | } else { |
1845 | /* | |
1846 | * The buffer may have been set dirty during | |
1847 | * attachment to a dirty page. | |
1848 | */ | |
1849 | clear_buffer_dirty(bh); | |
1850 | } | |
1851 | } while ((bh = bh->b_this_page) != head); | |
1852 | SetPageError(page); | |
1853 | BUG_ON(PageWriteback(page)); | |
7e4c3690 | 1854 | mapping_set_error(page->mapping, err); |
1da177e4 | 1855 | set_page_writeback(page); |
1da177e4 LT |
1856 | do { |
1857 | struct buffer_head *next = bh->b_this_page; | |
1858 | if (buffer_async_write(bh)) { | |
1859 | clear_buffer_dirty(bh); | |
8e8f9298 JA |
1860 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1861 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1862 | nr_underway++; |
1863 | } | |
1da177e4 LT |
1864 | bh = next; |
1865 | } while (bh != head); | |
ffda9d30 | 1866 | unlock_page(page); |
1da177e4 LT |
1867 | goto done; |
1868 | } | |
b4bba389 | 1869 | EXPORT_SYMBOL(__block_write_full_page); |
1da177e4 | 1870 | |
afddba49 NP |
1871 | /* |
1872 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
1873 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1874 | * block data from leaking). And clear the new bit. | |
1875 | */ | |
1876 | void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1877 | { | |
1878 | unsigned int block_start, block_end; | |
1879 | struct buffer_head *head, *bh; | |
1880 | ||
1881 | BUG_ON(!PageLocked(page)); | |
1882 | if (!page_has_buffers(page)) | |
1883 | return; | |
1884 | ||
1885 | bh = head = page_buffers(page); | |
1886 | block_start = 0; | |
1887 | do { | |
1888 | block_end = block_start + bh->b_size; | |
1889 | ||
1890 | if (buffer_new(bh)) { | |
1891 | if (block_end > from && block_start < to) { | |
1892 | if (!PageUptodate(page)) { | |
1893 | unsigned start, size; | |
1894 | ||
1895 | start = max(from, block_start); | |
1896 | size = min(to, block_end) - start; | |
1897 | ||
eebd2aa3 | 1898 | zero_user(page, start, size); |
afddba49 NP |
1899 | set_buffer_uptodate(bh); |
1900 | } | |
1901 | ||
1902 | clear_buffer_new(bh); | |
1903 | mark_buffer_dirty(bh); | |
1904 | } | |
1905 | } | |
1906 | ||
1907 | block_start = block_end; | |
1908 | bh = bh->b_this_page; | |
1909 | } while (bh != head); | |
1910 | } | |
1911 | EXPORT_SYMBOL(page_zero_new_buffers); | |
1912 | ||
ae259a9c CH |
1913 | static void |
1914 | iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, | |
1915 | struct iomap *iomap) | |
1916 | { | |
1917 | loff_t offset = block << inode->i_blkbits; | |
1918 | ||
1919 | bh->b_bdev = iomap->bdev; | |
1920 | ||
1921 | /* | |
1922 | * Block points to offset in file we need to map, iomap contains | |
1923 | * the offset at which the map starts. If the map ends before the | |
1924 | * current block, then do not map the buffer and let the caller | |
1925 | * handle it. | |
1926 | */ | |
1927 | BUG_ON(offset >= iomap->offset + iomap->length); | |
1928 | ||
1929 | switch (iomap->type) { | |
1930 | case IOMAP_HOLE: | |
1931 | /* | |
1932 | * If the buffer is not up to date or beyond the current EOF, | |
1933 | * we need to mark it as new to ensure sub-block zeroing is | |
1934 | * executed if necessary. | |
1935 | */ | |
1936 | if (!buffer_uptodate(bh) || | |
1937 | (offset >= i_size_read(inode))) | |
1938 | set_buffer_new(bh); | |
1939 | break; | |
1940 | case IOMAP_DELALLOC: | |
1941 | if (!buffer_uptodate(bh) || | |
1942 | (offset >= i_size_read(inode))) | |
1943 | set_buffer_new(bh); | |
1944 | set_buffer_uptodate(bh); | |
1945 | set_buffer_mapped(bh); | |
1946 | set_buffer_delay(bh); | |
1947 | break; | |
1948 | case IOMAP_UNWRITTEN: | |
1949 | /* | |
3d7b6b21 AG |
1950 | * For unwritten regions, we always need to ensure that regions |
1951 | * in the block we are not writing to are zeroed. Mark the | |
1952 | * buffer as new to ensure this. | |
ae259a9c CH |
1953 | */ |
1954 | set_buffer_new(bh); | |
1955 | set_buffer_unwritten(bh); | |
1956 | /* FALLTHRU */ | |
1957 | case IOMAP_MAPPED: | |
3d7b6b21 AG |
1958 | if ((iomap->flags & IOMAP_F_NEW) || |
1959 | offset >= i_size_read(inode)) | |
ae259a9c | 1960 | set_buffer_new(bh); |
19fe5f64 AG |
1961 | bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> |
1962 | inode->i_blkbits; | |
ae259a9c CH |
1963 | set_buffer_mapped(bh); |
1964 | break; | |
1965 | } | |
1966 | } | |
1967 | ||
1968 | int __block_write_begin_int(struct page *page, loff_t pos, unsigned len, | |
1969 | get_block_t *get_block, struct iomap *iomap) | |
1da177e4 | 1970 | { |
09cbfeaf | 1971 | unsigned from = pos & (PAGE_SIZE - 1); |
ebdec241 | 1972 | unsigned to = from + len; |
6e1db88d | 1973 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
1974 | unsigned block_start, block_end; |
1975 | sector_t block; | |
1976 | int err = 0; | |
1977 | unsigned blocksize, bbits; | |
1978 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; | |
1979 | ||
1980 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1981 | BUG_ON(from > PAGE_SIZE); |
1982 | BUG_ON(to > PAGE_SIZE); | |
1da177e4 LT |
1983 | BUG_ON(from > to); |
1984 | ||
45bce8f3 LT |
1985 | head = create_page_buffers(page, inode, 0); |
1986 | blocksize = head->b_size; | |
1987 | bbits = block_size_bits(blocksize); | |
1da177e4 | 1988 | |
09cbfeaf | 1989 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
1da177e4 LT |
1990 | |
1991 | for(bh = head, block_start = 0; bh != head || !block_start; | |
1992 | block++, block_start=block_end, bh = bh->b_this_page) { | |
1993 | block_end = block_start + blocksize; | |
1994 | if (block_end <= from || block_start >= to) { | |
1995 | if (PageUptodate(page)) { | |
1996 | if (!buffer_uptodate(bh)) | |
1997 | set_buffer_uptodate(bh); | |
1998 | } | |
1999 | continue; | |
2000 | } | |
2001 | if (buffer_new(bh)) | |
2002 | clear_buffer_new(bh); | |
2003 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2004 | WARN_ON(bh->b_size != blocksize); |
ae259a9c CH |
2005 | if (get_block) { |
2006 | err = get_block(inode, block, bh, 1); | |
2007 | if (err) | |
2008 | break; | |
2009 | } else { | |
2010 | iomap_to_bh(inode, block, bh, iomap); | |
2011 | } | |
2012 | ||
1da177e4 | 2013 | if (buffer_new(bh)) { |
e64855c6 | 2014 | clean_bdev_bh_alias(bh); |
1da177e4 | 2015 | if (PageUptodate(page)) { |
637aff46 | 2016 | clear_buffer_new(bh); |
1da177e4 | 2017 | set_buffer_uptodate(bh); |
637aff46 | 2018 | mark_buffer_dirty(bh); |
1da177e4 LT |
2019 | continue; |
2020 | } | |
eebd2aa3 CL |
2021 | if (block_end > to || block_start < from) |
2022 | zero_user_segments(page, | |
2023 | to, block_end, | |
2024 | block_start, from); | |
1da177e4 LT |
2025 | continue; |
2026 | } | |
2027 | } | |
2028 | if (PageUptodate(page)) { | |
2029 | if (!buffer_uptodate(bh)) | |
2030 | set_buffer_uptodate(bh); | |
2031 | continue; | |
2032 | } | |
2033 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
33a266dd | 2034 | !buffer_unwritten(bh) && |
1da177e4 | 2035 | (block_start < from || block_end > to)) { |
dfec8a14 | 2036 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2037 | *wait_bh++=bh; |
2038 | } | |
2039 | } | |
2040 | /* | |
2041 | * If we issued read requests - let them complete. | |
2042 | */ | |
2043 | while(wait_bh > wait) { | |
2044 | wait_on_buffer(*--wait_bh); | |
2045 | if (!buffer_uptodate(*wait_bh)) | |
f3ddbdc6 | 2046 | err = -EIO; |
1da177e4 | 2047 | } |
f9f07b6c | 2048 | if (unlikely(err)) |
afddba49 | 2049 | page_zero_new_buffers(page, from, to); |
1da177e4 LT |
2050 | return err; |
2051 | } | |
ae259a9c CH |
2052 | |
2053 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, | |
2054 | get_block_t *get_block) | |
2055 | { | |
2056 | return __block_write_begin_int(page, pos, len, get_block, NULL); | |
2057 | } | |
ebdec241 | 2058 | EXPORT_SYMBOL(__block_write_begin); |
1da177e4 LT |
2059 | |
2060 | static int __block_commit_write(struct inode *inode, struct page *page, | |
2061 | unsigned from, unsigned to) | |
2062 | { | |
2063 | unsigned block_start, block_end; | |
2064 | int partial = 0; | |
2065 | unsigned blocksize; | |
2066 | struct buffer_head *bh, *head; | |
2067 | ||
45bce8f3 LT |
2068 | bh = head = page_buffers(page); |
2069 | blocksize = bh->b_size; | |
1da177e4 | 2070 | |
45bce8f3 LT |
2071 | block_start = 0; |
2072 | do { | |
1da177e4 LT |
2073 | block_end = block_start + blocksize; |
2074 | if (block_end <= from || block_start >= to) { | |
2075 | if (!buffer_uptodate(bh)) | |
2076 | partial = 1; | |
2077 | } else { | |
2078 | set_buffer_uptodate(bh); | |
2079 | mark_buffer_dirty(bh); | |
2080 | } | |
afddba49 | 2081 | clear_buffer_new(bh); |
45bce8f3 LT |
2082 | |
2083 | block_start = block_end; | |
2084 | bh = bh->b_this_page; | |
2085 | } while (bh != head); | |
1da177e4 LT |
2086 | |
2087 | /* | |
2088 | * If this is a partial write which happened to make all buffers | |
2089 | * uptodate then we can optimize away a bogus readpage() for | |
2090 | * the next read(). Here we 'discover' whether the page went | |
2091 | * uptodate as a result of this (potentially partial) write. | |
2092 | */ | |
2093 | if (!partial) | |
2094 | SetPageUptodate(page); | |
2095 | return 0; | |
2096 | } | |
2097 | ||
afddba49 | 2098 | /* |
155130a4 CH |
2099 | * block_write_begin takes care of the basic task of block allocation and |
2100 | * bringing partial write blocks uptodate first. | |
2101 | * | |
7bb46a67 | 2102 | * The filesystem needs to handle block truncation upon failure. |
afddba49 | 2103 | */ |
155130a4 CH |
2104 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
2105 | unsigned flags, struct page **pagep, get_block_t *get_block) | |
afddba49 | 2106 | { |
09cbfeaf | 2107 | pgoff_t index = pos >> PAGE_SHIFT; |
afddba49 | 2108 | struct page *page; |
6e1db88d | 2109 | int status; |
afddba49 | 2110 | |
6e1db88d CH |
2111 | page = grab_cache_page_write_begin(mapping, index, flags); |
2112 | if (!page) | |
2113 | return -ENOMEM; | |
afddba49 | 2114 | |
6e1db88d | 2115 | status = __block_write_begin(page, pos, len, get_block); |
afddba49 | 2116 | if (unlikely(status)) { |
6e1db88d | 2117 | unlock_page(page); |
09cbfeaf | 2118 | put_page(page); |
6e1db88d | 2119 | page = NULL; |
afddba49 NP |
2120 | } |
2121 | ||
6e1db88d | 2122 | *pagep = page; |
afddba49 NP |
2123 | return status; |
2124 | } | |
2125 | EXPORT_SYMBOL(block_write_begin); | |
2126 | ||
2127 | int block_write_end(struct file *file, struct address_space *mapping, | |
2128 | loff_t pos, unsigned len, unsigned copied, | |
2129 | struct page *page, void *fsdata) | |
2130 | { | |
2131 | struct inode *inode = mapping->host; | |
2132 | unsigned start; | |
2133 | ||
09cbfeaf | 2134 | start = pos & (PAGE_SIZE - 1); |
afddba49 NP |
2135 | |
2136 | if (unlikely(copied < len)) { | |
2137 | /* | |
2138 | * The buffers that were written will now be uptodate, so we | |
2139 | * don't have to worry about a readpage reading them and | |
2140 | * overwriting a partial write. However if we have encountered | |
2141 | * a short write and only partially written into a buffer, it | |
2142 | * will not be marked uptodate, so a readpage might come in and | |
2143 | * destroy our partial write. | |
2144 | * | |
2145 | * Do the simplest thing, and just treat any short write to a | |
2146 | * non uptodate page as a zero-length write, and force the | |
2147 | * caller to redo the whole thing. | |
2148 | */ | |
2149 | if (!PageUptodate(page)) | |
2150 | copied = 0; | |
2151 | ||
2152 | page_zero_new_buffers(page, start+copied, start+len); | |
2153 | } | |
2154 | flush_dcache_page(page); | |
2155 | ||
2156 | /* This could be a short (even 0-length) commit */ | |
2157 | __block_commit_write(inode, page, start, start+copied); | |
2158 | ||
2159 | return copied; | |
2160 | } | |
2161 | EXPORT_SYMBOL(block_write_end); | |
2162 | ||
2163 | int generic_write_end(struct file *file, struct address_space *mapping, | |
2164 | loff_t pos, unsigned len, unsigned copied, | |
2165 | struct page *page, void *fsdata) | |
2166 | { | |
8af54f29 CH |
2167 | struct inode *inode = mapping->host; |
2168 | loff_t old_size = inode->i_size; | |
2169 | bool i_size_changed = false; | |
2170 | ||
afddba49 | 2171 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
8af54f29 CH |
2172 | |
2173 | /* | |
2174 | * No need to use i_size_read() here, the i_size cannot change under us | |
2175 | * because we hold i_rwsem. | |
2176 | * | |
2177 | * But it's important to update i_size while still holding page lock: | |
2178 | * page writeout could otherwise come in and zero beyond i_size. | |
2179 | */ | |
2180 | if (pos + copied > inode->i_size) { | |
2181 | i_size_write(inode, pos + copied); | |
2182 | i_size_changed = true; | |
2183 | } | |
2184 | ||
2185 | unlock_page(page); | |
7a77dad7 | 2186 | put_page(page); |
8af54f29 CH |
2187 | |
2188 | if (old_size < pos) | |
2189 | pagecache_isize_extended(inode, old_size, pos); | |
2190 | /* | |
2191 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
2192 | * makes the holding time of page lock longer. Second, it forces lock | |
2193 | * ordering of page lock and transaction start for journaling | |
2194 | * filesystems. | |
2195 | */ | |
2196 | if (i_size_changed) | |
2197 | mark_inode_dirty(inode); | |
26ddb1f4 | 2198 | return copied; |
afddba49 NP |
2199 | } |
2200 | EXPORT_SYMBOL(generic_write_end); | |
2201 | ||
8ab22b9a HH |
2202 | /* |
2203 | * block_is_partially_uptodate checks whether buffers within a page are | |
2204 | * uptodate or not. | |
2205 | * | |
2206 | * Returns true if all buffers which correspond to a file portion | |
2207 | * we want to read are uptodate. | |
2208 | */ | |
c186afb4 AV |
2209 | int block_is_partially_uptodate(struct page *page, unsigned long from, |
2210 | unsigned long count) | |
8ab22b9a | 2211 | { |
8ab22b9a HH |
2212 | unsigned block_start, block_end, blocksize; |
2213 | unsigned to; | |
2214 | struct buffer_head *bh, *head; | |
2215 | int ret = 1; | |
2216 | ||
2217 | if (!page_has_buffers(page)) | |
2218 | return 0; | |
2219 | ||
45bce8f3 LT |
2220 | head = page_buffers(page); |
2221 | blocksize = head->b_size; | |
09cbfeaf | 2222 | to = min_t(unsigned, PAGE_SIZE - from, count); |
8ab22b9a | 2223 | to = from + to; |
09cbfeaf | 2224 | if (from < blocksize && to > PAGE_SIZE - blocksize) |
8ab22b9a HH |
2225 | return 0; |
2226 | ||
8ab22b9a HH |
2227 | bh = head; |
2228 | block_start = 0; | |
2229 | do { | |
2230 | block_end = block_start + blocksize; | |
2231 | if (block_end > from && block_start < to) { | |
2232 | if (!buffer_uptodate(bh)) { | |
2233 | ret = 0; | |
2234 | break; | |
2235 | } | |
2236 | if (block_end >= to) | |
2237 | break; | |
2238 | } | |
2239 | block_start = block_end; | |
2240 | bh = bh->b_this_page; | |
2241 | } while (bh != head); | |
2242 | ||
2243 | return ret; | |
2244 | } | |
2245 | EXPORT_SYMBOL(block_is_partially_uptodate); | |
2246 | ||
1da177e4 LT |
2247 | /* |
2248 | * Generic "read page" function for block devices that have the normal | |
2249 | * get_block functionality. This is most of the block device filesystems. | |
2250 | * Reads the page asynchronously --- the unlock_buffer() and | |
2251 | * set/clear_buffer_uptodate() functions propagate buffer state into the | |
2252 | * page struct once IO has completed. | |
2253 | */ | |
2254 | int block_read_full_page(struct page *page, get_block_t *get_block) | |
2255 | { | |
2256 | struct inode *inode = page->mapping->host; | |
2257 | sector_t iblock, lblock; | |
2258 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; | |
45bce8f3 | 2259 | unsigned int blocksize, bbits; |
1da177e4 LT |
2260 | int nr, i; |
2261 | int fully_mapped = 1; | |
2262 | ||
45bce8f3 LT |
2263 | head = create_page_buffers(page, inode, 0); |
2264 | blocksize = head->b_size; | |
2265 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2266 | |
09cbfeaf | 2267 | iblock = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 2268 | lblock = (i_size_read(inode)+blocksize-1) >> bbits; |
1da177e4 LT |
2269 | bh = head; |
2270 | nr = 0; | |
2271 | i = 0; | |
2272 | ||
2273 | do { | |
2274 | if (buffer_uptodate(bh)) | |
2275 | continue; | |
2276 | ||
2277 | if (!buffer_mapped(bh)) { | |
c64610ba AM |
2278 | int err = 0; |
2279 | ||
1da177e4 LT |
2280 | fully_mapped = 0; |
2281 | if (iblock < lblock) { | |
b0cf2321 | 2282 | WARN_ON(bh->b_size != blocksize); |
c64610ba AM |
2283 | err = get_block(inode, iblock, bh, 0); |
2284 | if (err) | |
1da177e4 LT |
2285 | SetPageError(page); |
2286 | } | |
2287 | if (!buffer_mapped(bh)) { | |
eebd2aa3 | 2288 | zero_user(page, i * blocksize, blocksize); |
c64610ba AM |
2289 | if (!err) |
2290 | set_buffer_uptodate(bh); | |
1da177e4 LT |
2291 | continue; |
2292 | } | |
2293 | /* | |
2294 | * get_block() might have updated the buffer | |
2295 | * synchronously | |
2296 | */ | |
2297 | if (buffer_uptodate(bh)) | |
2298 | continue; | |
2299 | } | |
2300 | arr[nr++] = bh; | |
2301 | } while (i++, iblock++, (bh = bh->b_this_page) != head); | |
2302 | ||
2303 | if (fully_mapped) | |
2304 | SetPageMappedToDisk(page); | |
2305 | ||
2306 | if (!nr) { | |
2307 | /* | |
2308 | * All buffers are uptodate - we can set the page uptodate | |
2309 | * as well. But not if get_block() returned an error. | |
2310 | */ | |
2311 | if (!PageError(page)) | |
2312 | SetPageUptodate(page); | |
2313 | unlock_page(page); | |
2314 | return 0; | |
2315 | } | |
2316 | ||
2317 | /* Stage two: lock the buffers */ | |
2318 | for (i = 0; i < nr; i++) { | |
2319 | bh = arr[i]; | |
2320 | lock_buffer(bh); | |
2321 | mark_buffer_async_read(bh); | |
2322 | } | |
2323 | ||
2324 | /* | |
2325 | * Stage 3: start the IO. Check for uptodateness | |
2326 | * inside the buffer lock in case another process reading | |
2327 | * the underlying blockdev brought it uptodate (the sct fix). | |
2328 | */ | |
2329 | for (i = 0; i < nr; i++) { | |
2330 | bh = arr[i]; | |
2331 | if (buffer_uptodate(bh)) | |
2332 | end_buffer_async_read(bh, 1); | |
2333 | else | |
2a222ca9 | 2334 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
2335 | } |
2336 | return 0; | |
2337 | } | |
1fe72eaa | 2338 | EXPORT_SYMBOL(block_read_full_page); |
1da177e4 LT |
2339 | |
2340 | /* utility function for filesystems that need to do work on expanding | |
89e10787 | 2341 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
1da177e4 LT |
2342 | * deal with the hole. |
2343 | */ | |
89e10787 | 2344 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
1da177e4 LT |
2345 | { |
2346 | struct address_space *mapping = inode->i_mapping; | |
2347 | struct page *page; | |
89e10787 | 2348 | void *fsdata; |
1da177e4 LT |
2349 | int err; |
2350 | ||
c08d3b0e NP |
2351 | err = inode_newsize_ok(inode, size); |
2352 | if (err) | |
1da177e4 LT |
2353 | goto out; |
2354 | ||
89e10787 | 2355 | err = pagecache_write_begin(NULL, mapping, size, 0, |
c718a975 | 2356 | AOP_FLAG_CONT_EXPAND, &page, &fsdata); |
89e10787 | 2357 | if (err) |
05eb0b51 | 2358 | goto out; |
05eb0b51 | 2359 | |
89e10787 NP |
2360 | err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); |
2361 | BUG_ON(err > 0); | |
05eb0b51 | 2362 | |
1da177e4 LT |
2363 | out: |
2364 | return err; | |
2365 | } | |
1fe72eaa | 2366 | EXPORT_SYMBOL(generic_cont_expand_simple); |
1da177e4 | 2367 | |
f1e3af72 AB |
2368 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
2369 | loff_t pos, loff_t *bytes) | |
1da177e4 | 2370 | { |
1da177e4 | 2371 | struct inode *inode = mapping->host; |
93407472 | 2372 | unsigned int blocksize = i_blocksize(inode); |
89e10787 NP |
2373 | struct page *page; |
2374 | void *fsdata; | |
2375 | pgoff_t index, curidx; | |
2376 | loff_t curpos; | |
2377 | unsigned zerofrom, offset, len; | |
2378 | int err = 0; | |
1da177e4 | 2379 | |
09cbfeaf KS |
2380 | index = pos >> PAGE_SHIFT; |
2381 | offset = pos & ~PAGE_MASK; | |
89e10787 | 2382 | |
09cbfeaf KS |
2383 | while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { |
2384 | zerofrom = curpos & ~PAGE_MASK; | |
1da177e4 LT |
2385 | if (zerofrom & (blocksize-1)) { |
2386 | *bytes |= (blocksize-1); | |
2387 | (*bytes)++; | |
2388 | } | |
09cbfeaf | 2389 | len = PAGE_SIZE - zerofrom; |
1da177e4 | 2390 | |
c718a975 TH |
2391 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2392 | &page, &fsdata); | |
89e10787 NP |
2393 | if (err) |
2394 | goto out; | |
eebd2aa3 | 2395 | zero_user(page, zerofrom, len); |
89e10787 NP |
2396 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2397 | page, fsdata); | |
2398 | if (err < 0) | |
2399 | goto out; | |
2400 | BUG_ON(err != len); | |
2401 | err = 0; | |
061e9746 OH |
2402 | |
2403 | balance_dirty_pages_ratelimited(mapping); | |
c2ca0fcd | 2404 | |
08d405c8 | 2405 | if (fatal_signal_pending(current)) { |
c2ca0fcd MP |
2406 | err = -EINTR; |
2407 | goto out; | |
2408 | } | |
89e10787 | 2409 | } |
1da177e4 | 2410 | |
89e10787 NP |
2411 | /* page covers the boundary, find the boundary offset */ |
2412 | if (index == curidx) { | |
09cbfeaf | 2413 | zerofrom = curpos & ~PAGE_MASK; |
1da177e4 | 2414 | /* if we will expand the thing last block will be filled */ |
89e10787 NP |
2415 | if (offset <= zerofrom) { |
2416 | goto out; | |
2417 | } | |
2418 | if (zerofrom & (blocksize-1)) { | |
1da177e4 LT |
2419 | *bytes |= (blocksize-1); |
2420 | (*bytes)++; | |
2421 | } | |
89e10787 | 2422 | len = offset - zerofrom; |
1da177e4 | 2423 | |
c718a975 TH |
2424 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2425 | &page, &fsdata); | |
89e10787 NP |
2426 | if (err) |
2427 | goto out; | |
eebd2aa3 | 2428 | zero_user(page, zerofrom, len); |
89e10787 NP |
2429 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2430 | page, fsdata); | |
2431 | if (err < 0) | |
2432 | goto out; | |
2433 | BUG_ON(err != len); | |
2434 | err = 0; | |
1da177e4 | 2435 | } |
89e10787 NP |
2436 | out: |
2437 | return err; | |
2438 | } | |
2439 | ||
2440 | /* | |
2441 | * For moronic filesystems that do not allow holes in file. | |
2442 | * We may have to extend the file. | |
2443 | */ | |
282dc178 | 2444 | int cont_write_begin(struct file *file, struct address_space *mapping, |
89e10787 NP |
2445 | loff_t pos, unsigned len, unsigned flags, |
2446 | struct page **pagep, void **fsdata, | |
2447 | get_block_t *get_block, loff_t *bytes) | |
2448 | { | |
2449 | struct inode *inode = mapping->host; | |
93407472 FF |
2450 | unsigned int blocksize = i_blocksize(inode); |
2451 | unsigned int zerofrom; | |
89e10787 NP |
2452 | int err; |
2453 | ||
2454 | err = cont_expand_zero(file, mapping, pos, bytes); | |
2455 | if (err) | |
155130a4 | 2456 | return err; |
89e10787 | 2457 | |
09cbfeaf | 2458 | zerofrom = *bytes & ~PAGE_MASK; |
89e10787 NP |
2459 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { |
2460 | *bytes |= (blocksize-1); | |
2461 | (*bytes)++; | |
1da177e4 | 2462 | } |
1da177e4 | 2463 | |
155130a4 | 2464 | return block_write_begin(mapping, pos, len, flags, pagep, get_block); |
1da177e4 | 2465 | } |
1fe72eaa | 2466 | EXPORT_SYMBOL(cont_write_begin); |
1da177e4 | 2467 | |
1da177e4 LT |
2468 | int block_commit_write(struct page *page, unsigned from, unsigned to) |
2469 | { | |
2470 | struct inode *inode = page->mapping->host; | |
2471 | __block_commit_write(inode,page,from,to); | |
2472 | return 0; | |
2473 | } | |
1fe72eaa | 2474 | EXPORT_SYMBOL(block_commit_write); |
1da177e4 | 2475 | |
54171690 DC |
2476 | /* |
2477 | * block_page_mkwrite() is not allowed to change the file size as it gets | |
2478 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2479 | * be careful to check for EOF conditions here. We set the page up correctly | |
2480 | * for a written page which means we get ENOSPC checking when writing into | |
2481 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2482 | * support these features. | |
2483 | * | |
2484 | * We are not allowed to take the i_mutex here so we have to play games to | |
2485 | * protect against truncate races as the page could now be beyond EOF. Because | |
7bb46a67 | 2486 | * truncate writes the inode size before removing pages, once we have the |
54171690 DC |
2487 | * page lock we can determine safely if the page is beyond EOF. If it is not |
2488 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2489 | * unlock the page. | |
ea13a864 | 2490 | * |
14da9200 | 2491 | * Direct callers of this function should protect against filesystem freezing |
5c500029 | 2492 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
54171690 | 2493 | */ |
5c500029 | 2494 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
24da4fab | 2495 | get_block_t get_block) |
54171690 | 2496 | { |
c2ec175c | 2497 | struct page *page = vmf->page; |
496ad9aa | 2498 | struct inode *inode = file_inode(vma->vm_file); |
54171690 DC |
2499 | unsigned long end; |
2500 | loff_t size; | |
24da4fab | 2501 | int ret; |
54171690 DC |
2502 | |
2503 | lock_page(page); | |
2504 | size = i_size_read(inode); | |
2505 | if ((page->mapping != inode->i_mapping) || | |
18336338 | 2506 | (page_offset(page) > size)) { |
24da4fab JK |
2507 | /* We overload EFAULT to mean page got truncated */ |
2508 | ret = -EFAULT; | |
2509 | goto out_unlock; | |
54171690 DC |
2510 | } |
2511 | ||
2512 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
2513 | if (((page->index + 1) << PAGE_SHIFT) > size) |
2514 | end = size & ~PAGE_MASK; | |
54171690 | 2515 | else |
09cbfeaf | 2516 | end = PAGE_SIZE; |
54171690 | 2517 | |
ebdec241 | 2518 | ret = __block_write_begin(page, 0, end, get_block); |
54171690 DC |
2519 | if (!ret) |
2520 | ret = block_commit_write(page, 0, end); | |
2521 | ||
24da4fab JK |
2522 | if (unlikely(ret < 0)) |
2523 | goto out_unlock; | |
ea13a864 | 2524 | set_page_dirty(page); |
1d1d1a76 | 2525 | wait_for_stable_page(page); |
24da4fab JK |
2526 | return 0; |
2527 | out_unlock: | |
2528 | unlock_page(page); | |
54171690 | 2529 | return ret; |
24da4fab | 2530 | } |
1fe72eaa | 2531 | EXPORT_SYMBOL(block_page_mkwrite); |
1da177e4 LT |
2532 | |
2533 | /* | |
03158cd7 | 2534 | * nobh_write_begin()'s prereads are special: the buffer_heads are freed |
1da177e4 LT |
2535 | * immediately, while under the page lock. So it needs a special end_io |
2536 | * handler which does not touch the bh after unlocking it. | |
1da177e4 LT |
2537 | */ |
2538 | static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) | |
2539 | { | |
68671f35 | 2540 | __end_buffer_read_notouch(bh, uptodate); |
1da177e4 LT |
2541 | } |
2542 | ||
03158cd7 NP |
2543 | /* |
2544 | * Attach the singly-linked list of buffers created by nobh_write_begin, to | |
2545 | * the page (converting it to circular linked list and taking care of page | |
2546 | * dirty races). | |
2547 | */ | |
2548 | static void attach_nobh_buffers(struct page *page, struct buffer_head *head) | |
2549 | { | |
2550 | struct buffer_head *bh; | |
2551 | ||
2552 | BUG_ON(!PageLocked(page)); | |
2553 | ||
2554 | spin_lock(&page->mapping->private_lock); | |
2555 | bh = head; | |
2556 | do { | |
2557 | if (PageDirty(page)) | |
2558 | set_buffer_dirty(bh); | |
2559 | if (!bh->b_this_page) | |
2560 | bh->b_this_page = head; | |
2561 | bh = bh->b_this_page; | |
2562 | } while (bh != head); | |
2563 | attach_page_buffers(page, head); | |
2564 | spin_unlock(&page->mapping->private_lock); | |
2565 | } | |
2566 | ||
1da177e4 | 2567 | /* |
ea0f04e5 CH |
2568 | * On entry, the page is fully not uptodate. |
2569 | * On exit the page is fully uptodate in the areas outside (from,to) | |
7bb46a67 | 2570 | * The filesystem needs to handle block truncation upon failure. |
1da177e4 | 2571 | */ |
ea0f04e5 | 2572 | int nobh_write_begin(struct address_space *mapping, |
03158cd7 NP |
2573 | loff_t pos, unsigned len, unsigned flags, |
2574 | struct page **pagep, void **fsdata, | |
1da177e4 LT |
2575 | get_block_t *get_block) |
2576 | { | |
03158cd7 | 2577 | struct inode *inode = mapping->host; |
1da177e4 LT |
2578 | const unsigned blkbits = inode->i_blkbits; |
2579 | const unsigned blocksize = 1 << blkbits; | |
a4b0672d | 2580 | struct buffer_head *head, *bh; |
03158cd7 NP |
2581 | struct page *page; |
2582 | pgoff_t index; | |
2583 | unsigned from, to; | |
1da177e4 | 2584 | unsigned block_in_page; |
a4b0672d | 2585 | unsigned block_start, block_end; |
1da177e4 | 2586 | sector_t block_in_file; |
1da177e4 | 2587 | int nr_reads = 0; |
1da177e4 LT |
2588 | int ret = 0; |
2589 | int is_mapped_to_disk = 1; | |
1da177e4 | 2590 | |
09cbfeaf KS |
2591 | index = pos >> PAGE_SHIFT; |
2592 | from = pos & (PAGE_SIZE - 1); | |
03158cd7 NP |
2593 | to = from + len; |
2594 | ||
54566b2c | 2595 | page = grab_cache_page_write_begin(mapping, index, flags); |
03158cd7 NP |
2596 | if (!page) |
2597 | return -ENOMEM; | |
2598 | *pagep = page; | |
2599 | *fsdata = NULL; | |
2600 | ||
2601 | if (page_has_buffers(page)) { | |
309f77ad NK |
2602 | ret = __block_write_begin(page, pos, len, get_block); |
2603 | if (unlikely(ret)) | |
2604 | goto out_release; | |
2605 | return ret; | |
03158cd7 | 2606 | } |
a4b0672d | 2607 | |
1da177e4 LT |
2608 | if (PageMappedToDisk(page)) |
2609 | return 0; | |
2610 | ||
a4b0672d NP |
2611 | /* |
2612 | * Allocate buffers so that we can keep track of state, and potentially | |
2613 | * attach them to the page if an error occurs. In the common case of | |
2614 | * no error, they will just be freed again without ever being attached | |
2615 | * to the page (which is all OK, because we're under the page lock). | |
2616 | * | |
2617 | * Be careful: the buffer linked list is a NULL terminated one, rather | |
2618 | * than the circular one we're used to. | |
2619 | */ | |
640ab98f | 2620 | head = alloc_page_buffers(page, blocksize, false); |
03158cd7 NP |
2621 | if (!head) { |
2622 | ret = -ENOMEM; | |
2623 | goto out_release; | |
2624 | } | |
a4b0672d | 2625 | |
09cbfeaf | 2626 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
1da177e4 LT |
2627 | |
2628 | /* | |
2629 | * We loop across all blocks in the page, whether or not they are | |
2630 | * part of the affected region. This is so we can discover if the | |
2631 | * page is fully mapped-to-disk. | |
2632 | */ | |
a4b0672d | 2633 | for (block_start = 0, block_in_page = 0, bh = head; |
09cbfeaf | 2634 | block_start < PAGE_SIZE; |
a4b0672d | 2635 | block_in_page++, block_start += blocksize, bh = bh->b_this_page) { |
1da177e4 LT |
2636 | int create; |
2637 | ||
a4b0672d NP |
2638 | block_end = block_start + blocksize; |
2639 | bh->b_state = 0; | |
1da177e4 LT |
2640 | create = 1; |
2641 | if (block_start >= to) | |
2642 | create = 0; | |
2643 | ret = get_block(inode, block_in_file + block_in_page, | |
a4b0672d | 2644 | bh, create); |
1da177e4 LT |
2645 | if (ret) |
2646 | goto failed; | |
a4b0672d | 2647 | if (!buffer_mapped(bh)) |
1da177e4 | 2648 | is_mapped_to_disk = 0; |
a4b0672d | 2649 | if (buffer_new(bh)) |
e64855c6 | 2650 | clean_bdev_bh_alias(bh); |
a4b0672d NP |
2651 | if (PageUptodate(page)) { |
2652 | set_buffer_uptodate(bh); | |
1da177e4 | 2653 | continue; |
a4b0672d NP |
2654 | } |
2655 | if (buffer_new(bh) || !buffer_mapped(bh)) { | |
eebd2aa3 CL |
2656 | zero_user_segments(page, block_start, from, |
2657 | to, block_end); | |
1da177e4 LT |
2658 | continue; |
2659 | } | |
a4b0672d | 2660 | if (buffer_uptodate(bh)) |
1da177e4 LT |
2661 | continue; /* reiserfs does this */ |
2662 | if (block_start < from || block_end > to) { | |
a4b0672d NP |
2663 | lock_buffer(bh); |
2664 | bh->b_end_io = end_buffer_read_nobh; | |
2a222ca9 | 2665 | submit_bh(REQ_OP_READ, 0, bh); |
a4b0672d | 2666 | nr_reads++; |
1da177e4 LT |
2667 | } |
2668 | } | |
2669 | ||
2670 | if (nr_reads) { | |
1da177e4 LT |
2671 | /* |
2672 | * The page is locked, so these buffers are protected from | |
2673 | * any VM or truncate activity. Hence we don't need to care | |
2674 | * for the buffer_head refcounts. | |
2675 | */ | |
a4b0672d | 2676 | for (bh = head; bh; bh = bh->b_this_page) { |
1da177e4 LT |
2677 | wait_on_buffer(bh); |
2678 | if (!buffer_uptodate(bh)) | |
2679 | ret = -EIO; | |
1da177e4 LT |
2680 | } |
2681 | if (ret) | |
2682 | goto failed; | |
2683 | } | |
2684 | ||
2685 | if (is_mapped_to_disk) | |
2686 | SetPageMappedToDisk(page); | |
1da177e4 | 2687 | |
03158cd7 | 2688 | *fsdata = head; /* to be released by nobh_write_end */ |
a4b0672d | 2689 | |
1da177e4 LT |
2690 | return 0; |
2691 | ||
2692 | failed: | |
03158cd7 | 2693 | BUG_ON(!ret); |
1da177e4 | 2694 | /* |
a4b0672d NP |
2695 | * Error recovery is a bit difficult. We need to zero out blocks that |
2696 | * were newly allocated, and dirty them to ensure they get written out. | |
2697 | * Buffers need to be attached to the page at this point, otherwise | |
2698 | * the handling of potential IO errors during writeout would be hard | |
2699 | * (could try doing synchronous writeout, but what if that fails too?) | |
1da177e4 | 2700 | */ |
03158cd7 NP |
2701 | attach_nobh_buffers(page, head); |
2702 | page_zero_new_buffers(page, from, to); | |
a4b0672d | 2703 | |
03158cd7 NP |
2704 | out_release: |
2705 | unlock_page(page); | |
09cbfeaf | 2706 | put_page(page); |
03158cd7 | 2707 | *pagep = NULL; |
a4b0672d | 2708 | |
7bb46a67 NP |
2709 | return ret; |
2710 | } | |
03158cd7 | 2711 | EXPORT_SYMBOL(nobh_write_begin); |
1da177e4 | 2712 | |
03158cd7 NP |
2713 | int nobh_write_end(struct file *file, struct address_space *mapping, |
2714 | loff_t pos, unsigned len, unsigned copied, | |
2715 | struct page *page, void *fsdata) | |
1da177e4 LT |
2716 | { |
2717 | struct inode *inode = page->mapping->host; | |
efdc3131 | 2718 | struct buffer_head *head = fsdata; |
03158cd7 | 2719 | struct buffer_head *bh; |
5b41e74a | 2720 | BUG_ON(fsdata != NULL && page_has_buffers(page)); |
1da177e4 | 2721 | |
d4cf109f | 2722 | if (unlikely(copied < len) && head) |
5b41e74a DM |
2723 | attach_nobh_buffers(page, head); |
2724 | if (page_has_buffers(page)) | |
2725 | return generic_write_end(file, mapping, pos, len, | |
2726 | copied, page, fsdata); | |
a4b0672d | 2727 | |
22c8ca78 | 2728 | SetPageUptodate(page); |
1da177e4 | 2729 | set_page_dirty(page); |
03158cd7 NP |
2730 | if (pos+copied > inode->i_size) { |
2731 | i_size_write(inode, pos+copied); | |
1da177e4 LT |
2732 | mark_inode_dirty(inode); |
2733 | } | |
03158cd7 NP |
2734 | |
2735 | unlock_page(page); | |
09cbfeaf | 2736 | put_page(page); |
03158cd7 | 2737 | |
03158cd7 NP |
2738 | while (head) { |
2739 | bh = head; | |
2740 | head = head->b_this_page; | |
2741 | free_buffer_head(bh); | |
2742 | } | |
2743 | ||
2744 | return copied; | |
1da177e4 | 2745 | } |
03158cd7 | 2746 | EXPORT_SYMBOL(nobh_write_end); |
1da177e4 LT |
2747 | |
2748 | /* | |
2749 | * nobh_writepage() - based on block_full_write_page() except | |
2750 | * that it tries to operate without attaching bufferheads to | |
2751 | * the page. | |
2752 | */ | |
2753 | int nobh_writepage(struct page *page, get_block_t *get_block, | |
2754 | struct writeback_control *wbc) | |
2755 | { | |
2756 | struct inode * const inode = page->mapping->host; | |
2757 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2758 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2759 | unsigned offset; |
1da177e4 LT |
2760 | int ret; |
2761 | ||
2762 | /* Is the page fully inside i_size? */ | |
2763 | if (page->index < end_index) | |
2764 | goto out; | |
2765 | ||
2766 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2767 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 LT |
2768 | if (page->index >= end_index+1 || !offset) { |
2769 | /* | |
2770 | * The page may have dirty, unmapped buffers. For example, | |
2771 | * they may have been added in ext3_writepage(). Make them | |
2772 | * freeable here, so the page does not leak. | |
2773 | */ | |
2774 | #if 0 | |
2775 | /* Not really sure about this - do we need this ? */ | |
2776 | if (page->mapping->a_ops->invalidatepage) | |
2777 | page->mapping->a_ops->invalidatepage(page, offset); | |
2778 | #endif | |
2779 | unlock_page(page); | |
2780 | return 0; /* don't care */ | |
2781 | } | |
2782 | ||
2783 | /* | |
2784 | * The page straddles i_size. It must be zeroed out on each and every | |
2785 | * writepage invocation because it may be mmapped. "A file is mapped | |
2786 | * in multiples of the page size. For a file that is not a multiple of | |
2787 | * the page size, the remaining memory is zeroed when mapped, and | |
2788 | * writes to that region are not written out to the file." | |
2789 | */ | |
09cbfeaf | 2790 | zero_user_segment(page, offset, PAGE_SIZE); |
1da177e4 LT |
2791 | out: |
2792 | ret = mpage_writepage(page, get_block, wbc); | |
2793 | if (ret == -EAGAIN) | |
35c80d5f CM |
2794 | ret = __block_write_full_page(inode, page, get_block, wbc, |
2795 | end_buffer_async_write); | |
1da177e4 LT |
2796 | return ret; |
2797 | } | |
2798 | EXPORT_SYMBOL(nobh_writepage); | |
2799 | ||
03158cd7 NP |
2800 | int nobh_truncate_page(struct address_space *mapping, |
2801 | loff_t from, get_block_t *get_block) | |
1da177e4 | 2802 | { |
09cbfeaf KS |
2803 | pgoff_t index = from >> PAGE_SHIFT; |
2804 | unsigned offset = from & (PAGE_SIZE-1); | |
03158cd7 NP |
2805 | unsigned blocksize; |
2806 | sector_t iblock; | |
2807 | unsigned length, pos; | |
2808 | struct inode *inode = mapping->host; | |
1da177e4 | 2809 | struct page *page; |
03158cd7 NP |
2810 | struct buffer_head map_bh; |
2811 | int err; | |
1da177e4 | 2812 | |
93407472 | 2813 | blocksize = i_blocksize(inode); |
03158cd7 NP |
2814 | length = offset & (blocksize - 1); |
2815 | ||
2816 | /* Block boundary? Nothing to do */ | |
2817 | if (!length) | |
2818 | return 0; | |
2819 | ||
2820 | length = blocksize - length; | |
09cbfeaf | 2821 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 | 2822 | |
1da177e4 | 2823 | page = grab_cache_page(mapping, index); |
03158cd7 | 2824 | err = -ENOMEM; |
1da177e4 LT |
2825 | if (!page) |
2826 | goto out; | |
2827 | ||
03158cd7 NP |
2828 | if (page_has_buffers(page)) { |
2829 | has_buffers: | |
2830 | unlock_page(page); | |
09cbfeaf | 2831 | put_page(page); |
03158cd7 NP |
2832 | return block_truncate_page(mapping, from, get_block); |
2833 | } | |
2834 | ||
2835 | /* Find the buffer that contains "offset" */ | |
2836 | pos = blocksize; | |
2837 | while (offset >= pos) { | |
2838 | iblock++; | |
2839 | pos += blocksize; | |
2840 | } | |
2841 | ||
460bcf57 TT |
2842 | map_bh.b_size = blocksize; |
2843 | map_bh.b_state = 0; | |
03158cd7 NP |
2844 | err = get_block(inode, iblock, &map_bh, 0); |
2845 | if (err) | |
2846 | goto unlock; | |
2847 | /* unmapped? It's a hole - nothing to do */ | |
2848 | if (!buffer_mapped(&map_bh)) | |
2849 | goto unlock; | |
2850 | ||
2851 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2852 | if (!PageUptodate(page)) { | |
2853 | err = mapping->a_ops->readpage(NULL, page); | |
2854 | if (err) { | |
09cbfeaf | 2855 | put_page(page); |
03158cd7 NP |
2856 | goto out; |
2857 | } | |
2858 | lock_page(page); | |
2859 | if (!PageUptodate(page)) { | |
2860 | err = -EIO; | |
2861 | goto unlock; | |
2862 | } | |
2863 | if (page_has_buffers(page)) | |
2864 | goto has_buffers; | |
1da177e4 | 2865 | } |
eebd2aa3 | 2866 | zero_user(page, offset, length); |
03158cd7 NP |
2867 | set_page_dirty(page); |
2868 | err = 0; | |
2869 | ||
2870 | unlock: | |
1da177e4 | 2871 | unlock_page(page); |
09cbfeaf | 2872 | put_page(page); |
1da177e4 | 2873 | out: |
03158cd7 | 2874 | return err; |
1da177e4 LT |
2875 | } |
2876 | EXPORT_SYMBOL(nobh_truncate_page); | |
2877 | ||
2878 | int block_truncate_page(struct address_space *mapping, | |
2879 | loff_t from, get_block_t *get_block) | |
2880 | { | |
09cbfeaf KS |
2881 | pgoff_t index = from >> PAGE_SHIFT; |
2882 | unsigned offset = from & (PAGE_SIZE-1); | |
1da177e4 | 2883 | unsigned blocksize; |
54b21a79 | 2884 | sector_t iblock; |
1da177e4 LT |
2885 | unsigned length, pos; |
2886 | struct inode *inode = mapping->host; | |
2887 | struct page *page; | |
2888 | struct buffer_head *bh; | |
1da177e4 LT |
2889 | int err; |
2890 | ||
93407472 | 2891 | blocksize = i_blocksize(inode); |
1da177e4 LT |
2892 | length = offset & (blocksize - 1); |
2893 | ||
2894 | /* Block boundary? Nothing to do */ | |
2895 | if (!length) | |
2896 | return 0; | |
2897 | ||
2898 | length = blocksize - length; | |
09cbfeaf | 2899 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 LT |
2900 | |
2901 | page = grab_cache_page(mapping, index); | |
2902 | err = -ENOMEM; | |
2903 | if (!page) | |
2904 | goto out; | |
2905 | ||
2906 | if (!page_has_buffers(page)) | |
2907 | create_empty_buffers(page, blocksize, 0); | |
2908 | ||
2909 | /* Find the buffer that contains "offset" */ | |
2910 | bh = page_buffers(page); | |
2911 | pos = blocksize; | |
2912 | while (offset >= pos) { | |
2913 | bh = bh->b_this_page; | |
2914 | iblock++; | |
2915 | pos += blocksize; | |
2916 | } | |
2917 | ||
2918 | err = 0; | |
2919 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2920 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
2921 | err = get_block(inode, iblock, bh, 0); |
2922 | if (err) | |
2923 | goto unlock; | |
2924 | /* unmapped? It's a hole - nothing to do */ | |
2925 | if (!buffer_mapped(bh)) | |
2926 | goto unlock; | |
2927 | } | |
2928 | ||
2929 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2930 | if (PageUptodate(page)) | |
2931 | set_buffer_uptodate(bh); | |
2932 | ||
33a266dd | 2933 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
1da177e4 | 2934 | err = -EIO; |
dfec8a14 | 2935 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2936 | wait_on_buffer(bh); |
2937 | /* Uhhuh. Read error. Complain and punt. */ | |
2938 | if (!buffer_uptodate(bh)) | |
2939 | goto unlock; | |
2940 | } | |
2941 | ||
eebd2aa3 | 2942 | zero_user(page, offset, length); |
1da177e4 LT |
2943 | mark_buffer_dirty(bh); |
2944 | err = 0; | |
2945 | ||
2946 | unlock: | |
2947 | unlock_page(page); | |
09cbfeaf | 2948 | put_page(page); |
1da177e4 LT |
2949 | out: |
2950 | return err; | |
2951 | } | |
1fe72eaa | 2952 | EXPORT_SYMBOL(block_truncate_page); |
1da177e4 LT |
2953 | |
2954 | /* | |
2955 | * The generic ->writepage function for buffer-backed address_spaces | |
2956 | */ | |
1b938c08 MW |
2957 | int block_write_full_page(struct page *page, get_block_t *get_block, |
2958 | struct writeback_control *wbc) | |
1da177e4 LT |
2959 | { |
2960 | struct inode * const inode = page->mapping->host; | |
2961 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2962 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2963 | unsigned offset; |
1da177e4 LT |
2964 | |
2965 | /* Is the page fully inside i_size? */ | |
2966 | if (page->index < end_index) | |
35c80d5f | 2967 | return __block_write_full_page(inode, page, get_block, wbc, |
1b938c08 | 2968 | end_buffer_async_write); |
1da177e4 LT |
2969 | |
2970 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2971 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 LT |
2972 | if (page->index >= end_index+1 || !offset) { |
2973 | /* | |
2974 | * The page may have dirty, unmapped buffers. For example, | |
2975 | * they may have been added in ext3_writepage(). Make them | |
2976 | * freeable here, so the page does not leak. | |
2977 | */ | |
09cbfeaf | 2978 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 LT |
2979 | unlock_page(page); |
2980 | return 0; /* don't care */ | |
2981 | } | |
2982 | ||
2983 | /* | |
2984 | * The page straddles i_size. It must be zeroed out on each and every | |
2a61aa40 | 2985 | * writepage invocation because it may be mmapped. "A file is mapped |
1da177e4 LT |
2986 | * in multiples of the page size. For a file that is not a multiple of |
2987 | * the page size, the remaining memory is zeroed when mapped, and | |
2988 | * writes to that region are not written out to the file." | |
2989 | */ | |
09cbfeaf | 2990 | zero_user_segment(page, offset, PAGE_SIZE); |
1b938c08 MW |
2991 | return __block_write_full_page(inode, page, get_block, wbc, |
2992 | end_buffer_async_write); | |
35c80d5f | 2993 | } |
1fe72eaa | 2994 | EXPORT_SYMBOL(block_write_full_page); |
35c80d5f | 2995 | |
1da177e4 LT |
2996 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
2997 | get_block_t *get_block) | |
2998 | { | |
1da177e4 | 2999 | struct inode *inode = mapping->host; |
2a527d68 AP |
3000 | struct buffer_head tmp = { |
3001 | .b_size = i_blocksize(inode), | |
3002 | }; | |
3003 | ||
1da177e4 LT |
3004 | get_block(inode, block, &tmp, 0); |
3005 | return tmp.b_blocknr; | |
3006 | } | |
1fe72eaa | 3007 | EXPORT_SYMBOL(generic_block_bmap); |
1da177e4 | 3008 | |
4246a0b6 | 3009 | static void end_bio_bh_io_sync(struct bio *bio) |
1da177e4 LT |
3010 | { |
3011 | struct buffer_head *bh = bio->bi_private; | |
3012 | ||
b7c44ed9 | 3013 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
08bafc03 KM |
3014 | set_bit(BH_Quiet, &bh->b_state); |
3015 | ||
4e4cbee9 | 3016 | bh->b_end_io(bh, !bio->bi_status); |
1da177e4 | 3017 | bio_put(bio); |
1da177e4 LT |
3018 | } |
3019 | ||
57302e0d LT |
3020 | /* |
3021 | * This allows us to do IO even on the odd last sectors | |
59d43914 | 3022 | * of a device, even if the block size is some multiple |
57302e0d LT |
3023 | * of the physical sector size. |
3024 | * | |
3025 | * We'll just truncate the bio to the size of the device, | |
3026 | * and clear the end of the buffer head manually. | |
3027 | * | |
3028 | * Truly out-of-range accesses will turn into actual IO | |
3029 | * errors, this only handles the "we need to be able to | |
3030 | * do IO at the final sector" case. | |
3031 | */ | |
2a222ca9 | 3032 | void guard_bio_eod(int op, struct bio *bio) |
57302e0d LT |
3033 | { |
3034 | sector_t maxsector; | |
c45a8f2d | 3035 | struct bio_vec *bvec = bio_last_bvec_all(bio); |
59d43914 | 3036 | unsigned truncated_bytes; |
67f2519f GE |
3037 | struct hd_struct *part; |
3038 | ||
3039 | rcu_read_lock(); | |
3040 | part = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
3041 | if (part) | |
3042 | maxsector = part_nr_sects_read(part); | |
3043 | else | |
3044 | maxsector = get_capacity(bio->bi_disk); | |
3045 | rcu_read_unlock(); | |
57302e0d | 3046 | |
57302e0d LT |
3047 | if (!maxsector) |
3048 | return; | |
3049 | ||
3050 | /* | |
3051 | * If the *whole* IO is past the end of the device, | |
3052 | * let it through, and the IO layer will turn it into | |
3053 | * an EIO. | |
3054 | */ | |
4f024f37 | 3055 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) |
57302e0d LT |
3056 | return; |
3057 | ||
4f024f37 | 3058 | maxsector -= bio->bi_iter.bi_sector; |
59d43914 | 3059 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) |
57302e0d LT |
3060 | return; |
3061 | ||
59d43914 AM |
3062 | /* Uhhuh. We've got a bio that straddles the device size! */ |
3063 | truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9); | |
57302e0d | 3064 | |
dce30ca9 CM |
3065 | /* |
3066 | * The bio contains more than one segment which spans EOD, just return | |
3067 | * and let IO layer turn it into an EIO | |
3068 | */ | |
3069 | if (truncated_bytes > bvec->bv_len) | |
3070 | return; | |
3071 | ||
57302e0d | 3072 | /* Truncate the bio.. */ |
59d43914 AM |
3073 | bio->bi_iter.bi_size -= truncated_bytes; |
3074 | bvec->bv_len -= truncated_bytes; | |
57302e0d LT |
3075 | |
3076 | /* ..and clear the end of the buffer for reads */ | |
2a222ca9 | 3077 | if (op == REQ_OP_READ) { |
f70f4464 ML |
3078 | struct bio_vec bv; |
3079 | ||
3080 | mp_bvec_last_segment(bvec, &bv); | |
3081 | zero_user(bv.bv_page, bv.bv_offset + bv.bv_len, | |
59d43914 | 3082 | truncated_bytes); |
57302e0d LT |
3083 | } |
3084 | } | |
3085 | ||
2a222ca9 | 3086 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 3087 | enum rw_hint write_hint, struct writeback_control *wbc) |
1da177e4 LT |
3088 | { |
3089 | struct bio *bio; | |
1da177e4 LT |
3090 | |
3091 | BUG_ON(!buffer_locked(bh)); | |
3092 | BUG_ON(!buffer_mapped(bh)); | |
3093 | BUG_ON(!bh->b_end_io); | |
8fb0e342 AK |
3094 | BUG_ON(buffer_delay(bh)); |
3095 | BUG_ON(buffer_unwritten(bh)); | |
1da177e4 | 3096 | |
1da177e4 | 3097 | /* |
48fd4f93 | 3098 | * Only clear out a write error when rewriting |
1da177e4 | 3099 | */ |
2a222ca9 | 3100 | if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) |
1da177e4 LT |
3101 | clear_buffer_write_io_error(bh); |
3102 | ||
3103 | /* | |
3104 | * from here on down, it's all bio -- do the initial mapping, | |
3105 | * submit_bio -> generic_make_request may further map this bio around | |
3106 | */ | |
3107 | bio = bio_alloc(GFP_NOIO, 1); | |
3108 | ||
4f024f37 | 3109 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
74d46992 | 3110 | bio_set_dev(bio, bh->b_bdev); |
8e8f9298 | 3111 | bio->bi_write_hint = write_hint; |
1da177e4 | 3112 | |
6cf66b4c KO |
3113 | bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
3114 | BUG_ON(bio->bi_iter.bi_size != bh->b_size); | |
1da177e4 LT |
3115 | |
3116 | bio->bi_end_io = end_bio_bh_io_sync; | |
3117 | bio->bi_private = bh; | |
3118 | ||
57302e0d | 3119 | /* Take care of bh's that straddle the end of the device */ |
2a222ca9 | 3120 | guard_bio_eod(op, bio); |
57302e0d | 3121 | |
877f962c | 3122 | if (buffer_meta(bh)) |
2a222ca9 | 3123 | op_flags |= REQ_META; |
877f962c | 3124 | if (buffer_prio(bh)) |
2a222ca9 MC |
3125 | op_flags |= REQ_PRIO; |
3126 | bio_set_op_attrs(bio, op, op_flags); | |
877f962c | 3127 | |
fd42df30 DZ |
3128 | if (wbc) { |
3129 | wbc_init_bio(wbc, bio); | |
34e51a5e | 3130 | wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); |
fd42df30 DZ |
3131 | } |
3132 | ||
4e49ea4a | 3133 | submit_bio(bio); |
f6454b04 | 3134 | return 0; |
1da177e4 | 3135 | } |
bafc0dba | 3136 | |
020c2833 | 3137 | int submit_bh(int op, int op_flags, struct buffer_head *bh) |
bafc0dba | 3138 | { |
8e8f9298 | 3139 | return submit_bh_wbc(op, op_flags, bh, 0, NULL); |
71368511 | 3140 | } |
1fe72eaa | 3141 | EXPORT_SYMBOL(submit_bh); |
1da177e4 LT |
3142 | |
3143 | /** | |
3144 | * ll_rw_block: low-level access to block devices (DEPRECATED) | |
dfec8a14 | 3145 | * @op: whether to %READ or %WRITE |
ef295ecf | 3146 | * @op_flags: req_flag_bits |
1da177e4 LT |
3147 | * @nr: number of &struct buffer_heads in the array |
3148 | * @bhs: array of pointers to &struct buffer_head | |
3149 | * | |
a7662236 | 3150 | * ll_rw_block() takes an array of pointers to &struct buffer_heads, and |
70246286 CH |
3151 | * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE. |
3152 | * @op_flags contains flags modifying the detailed I/O behavior, most notably | |
3153 | * %REQ_RAHEAD. | |
1da177e4 LT |
3154 | * |
3155 | * This function drops any buffer that it cannot get a lock on (with the | |
9cb569d6 CH |
3156 | * BH_Lock state bit), any buffer that appears to be clean when doing a write |
3157 | * request, and any buffer that appears to be up-to-date when doing read | |
3158 | * request. Further it marks as clean buffers that are processed for | |
3159 | * writing (the buffer cache won't assume that they are actually clean | |
3160 | * until the buffer gets unlocked). | |
1da177e4 LT |
3161 | * |
3162 | * ll_rw_block sets b_end_io to simple completion handler that marks | |
e227867f | 3163 | * the buffer up-to-date (if appropriate), unlocks the buffer and wakes |
1da177e4 LT |
3164 | * any waiters. |
3165 | * | |
3166 | * All of the buffers must be for the same device, and must also be a | |
3167 | * multiple of the current approved size for the device. | |
3168 | */ | |
dfec8a14 | 3169 | void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[]) |
1da177e4 LT |
3170 | { |
3171 | int i; | |
3172 | ||
3173 | for (i = 0; i < nr; i++) { | |
3174 | struct buffer_head *bh = bhs[i]; | |
3175 | ||
9cb569d6 | 3176 | if (!trylock_buffer(bh)) |
1da177e4 | 3177 | continue; |
dfec8a14 | 3178 | if (op == WRITE) { |
1da177e4 | 3179 | if (test_clear_buffer_dirty(bh)) { |
76c3073a | 3180 | bh->b_end_io = end_buffer_write_sync; |
e60e5c50 | 3181 | get_bh(bh); |
dfec8a14 | 3182 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3183 | continue; |
3184 | } | |
3185 | } else { | |
1da177e4 | 3186 | if (!buffer_uptodate(bh)) { |
76c3073a | 3187 | bh->b_end_io = end_buffer_read_sync; |
e60e5c50 | 3188 | get_bh(bh); |
dfec8a14 | 3189 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3190 | continue; |
3191 | } | |
3192 | } | |
3193 | unlock_buffer(bh); | |
1da177e4 LT |
3194 | } |
3195 | } | |
1fe72eaa | 3196 | EXPORT_SYMBOL(ll_rw_block); |
1da177e4 | 3197 | |
2a222ca9 | 3198 | void write_dirty_buffer(struct buffer_head *bh, int op_flags) |
9cb569d6 CH |
3199 | { |
3200 | lock_buffer(bh); | |
3201 | if (!test_clear_buffer_dirty(bh)) { | |
3202 | unlock_buffer(bh); | |
3203 | return; | |
3204 | } | |
3205 | bh->b_end_io = end_buffer_write_sync; | |
3206 | get_bh(bh); | |
2a222ca9 | 3207 | submit_bh(REQ_OP_WRITE, op_flags, bh); |
9cb569d6 CH |
3208 | } |
3209 | EXPORT_SYMBOL(write_dirty_buffer); | |
3210 | ||
1da177e4 LT |
3211 | /* |
3212 | * For a data-integrity writeout, we need to wait upon any in-progress I/O | |
3213 | * and then start new I/O and then wait upon it. The caller must have a ref on | |
3214 | * the buffer_head. | |
3215 | */ | |
2a222ca9 | 3216 | int __sync_dirty_buffer(struct buffer_head *bh, int op_flags) |
1da177e4 LT |
3217 | { |
3218 | int ret = 0; | |
3219 | ||
3220 | WARN_ON(atomic_read(&bh->b_count) < 1); | |
3221 | lock_buffer(bh); | |
3222 | if (test_clear_buffer_dirty(bh)) { | |
3223 | get_bh(bh); | |
3224 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 3225 | ret = submit_bh(REQ_OP_WRITE, op_flags, bh); |
1da177e4 | 3226 | wait_on_buffer(bh); |
1da177e4 LT |
3227 | if (!ret && !buffer_uptodate(bh)) |
3228 | ret = -EIO; | |
3229 | } else { | |
3230 | unlock_buffer(bh); | |
3231 | } | |
3232 | return ret; | |
3233 | } | |
87e99511 CH |
3234 | EXPORT_SYMBOL(__sync_dirty_buffer); |
3235 | ||
3236 | int sync_dirty_buffer(struct buffer_head *bh) | |
3237 | { | |
70fd7614 | 3238 | return __sync_dirty_buffer(bh, REQ_SYNC); |
87e99511 | 3239 | } |
1fe72eaa | 3240 | EXPORT_SYMBOL(sync_dirty_buffer); |
1da177e4 LT |
3241 | |
3242 | /* | |
3243 | * try_to_free_buffers() checks if all the buffers on this particular page | |
3244 | * are unused, and releases them if so. | |
3245 | * | |
3246 | * Exclusion against try_to_free_buffers may be obtained by either | |
3247 | * locking the page or by holding its mapping's private_lock. | |
3248 | * | |
3249 | * If the page is dirty but all the buffers are clean then we need to | |
3250 | * be sure to mark the page clean as well. This is because the page | |
3251 | * may be against a block device, and a later reattachment of buffers | |
3252 | * to a dirty page will set *all* buffers dirty. Which would corrupt | |
3253 | * filesystem data on the same device. | |
3254 | * | |
3255 | * The same applies to regular filesystem pages: if all the buffers are | |
3256 | * clean then we set the page clean and proceed. To do that, we require | |
3257 | * total exclusion from __set_page_dirty_buffers(). That is obtained with | |
3258 | * private_lock. | |
3259 | * | |
3260 | * try_to_free_buffers() is non-blocking. | |
3261 | */ | |
3262 | static inline int buffer_busy(struct buffer_head *bh) | |
3263 | { | |
3264 | return atomic_read(&bh->b_count) | | |
3265 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); | |
3266 | } | |
3267 | ||
3268 | static int | |
3269 | drop_buffers(struct page *page, struct buffer_head **buffers_to_free) | |
3270 | { | |
3271 | struct buffer_head *head = page_buffers(page); | |
3272 | struct buffer_head *bh; | |
3273 | ||
3274 | bh = head; | |
3275 | do { | |
1da177e4 LT |
3276 | if (buffer_busy(bh)) |
3277 | goto failed; | |
3278 | bh = bh->b_this_page; | |
3279 | } while (bh != head); | |
3280 | ||
3281 | do { | |
3282 | struct buffer_head *next = bh->b_this_page; | |
3283 | ||
535ee2fb | 3284 | if (bh->b_assoc_map) |
1da177e4 LT |
3285 | __remove_assoc_queue(bh); |
3286 | bh = next; | |
3287 | } while (bh != head); | |
3288 | *buffers_to_free = head; | |
3289 | __clear_page_buffers(page); | |
3290 | return 1; | |
3291 | failed: | |
3292 | return 0; | |
3293 | } | |
3294 | ||
3295 | int try_to_free_buffers(struct page *page) | |
3296 | { | |
3297 | struct address_space * const mapping = page->mapping; | |
3298 | struct buffer_head *buffers_to_free = NULL; | |
3299 | int ret = 0; | |
3300 | ||
3301 | BUG_ON(!PageLocked(page)); | |
ecdfc978 | 3302 | if (PageWriteback(page)) |
1da177e4 LT |
3303 | return 0; |
3304 | ||
3305 | if (mapping == NULL) { /* can this still happen? */ | |
3306 | ret = drop_buffers(page, &buffers_to_free); | |
3307 | goto out; | |
3308 | } | |
3309 | ||
3310 | spin_lock(&mapping->private_lock); | |
3311 | ret = drop_buffers(page, &buffers_to_free); | |
ecdfc978 LT |
3312 | |
3313 | /* | |
3314 | * If the filesystem writes its buffers by hand (eg ext3) | |
3315 | * then we can have clean buffers against a dirty page. We | |
3316 | * clean the page here; otherwise the VM will never notice | |
3317 | * that the filesystem did any IO at all. | |
3318 | * | |
3319 | * Also, during truncate, discard_buffer will have marked all | |
3320 | * the page's buffers clean. We discover that here and clean | |
3321 | * the page also. | |
87df7241 NP |
3322 | * |
3323 | * private_lock must be held over this entire operation in order | |
3324 | * to synchronise against __set_page_dirty_buffers and prevent the | |
3325 | * dirty bit from being lost. | |
ecdfc978 | 3326 | */ |
11f81bec TH |
3327 | if (ret) |
3328 | cancel_dirty_page(page); | |
87df7241 | 3329 | spin_unlock(&mapping->private_lock); |
1da177e4 LT |
3330 | out: |
3331 | if (buffers_to_free) { | |
3332 | struct buffer_head *bh = buffers_to_free; | |
3333 | ||
3334 | do { | |
3335 | struct buffer_head *next = bh->b_this_page; | |
3336 | free_buffer_head(bh); | |
3337 | bh = next; | |
3338 | } while (bh != buffers_to_free); | |
3339 | } | |
3340 | return ret; | |
3341 | } | |
3342 | EXPORT_SYMBOL(try_to_free_buffers); | |
3343 | ||
1da177e4 LT |
3344 | /* |
3345 | * There are no bdflush tunables left. But distributions are | |
3346 | * still running obsolete flush daemons, so we terminate them here. | |
3347 | * | |
3348 | * Use of bdflush() is deprecated and will be removed in a future kernel. | |
5b0830cb | 3349 | * The `flush-X' kernel threads fully replace bdflush daemons and this call. |
1da177e4 | 3350 | */ |
bdc480e3 | 3351 | SYSCALL_DEFINE2(bdflush, int, func, long, data) |
1da177e4 LT |
3352 | { |
3353 | static int msg_count; | |
3354 | ||
3355 | if (!capable(CAP_SYS_ADMIN)) | |
3356 | return -EPERM; | |
3357 | ||
3358 | if (msg_count < 5) { | |
3359 | msg_count++; | |
3360 | printk(KERN_INFO | |
3361 | "warning: process `%s' used the obsolete bdflush" | |
3362 | " system call\n", current->comm); | |
3363 | printk(KERN_INFO "Fix your initscripts?\n"); | |
3364 | } | |
3365 | ||
3366 | if (func == 1) | |
3367 | do_exit(0); | |
3368 | return 0; | |
3369 | } | |
3370 | ||
3371 | /* | |
3372 | * Buffer-head allocation | |
3373 | */ | |
a0a9b043 | 3374 | static struct kmem_cache *bh_cachep __read_mostly; |
1da177e4 LT |
3375 | |
3376 | /* | |
3377 | * Once the number of bh's in the machine exceeds this level, we start | |
3378 | * stripping them in writeback. | |
3379 | */ | |
43be594a | 3380 | static unsigned long max_buffer_heads; |
1da177e4 LT |
3381 | |
3382 | int buffer_heads_over_limit; | |
3383 | ||
3384 | struct bh_accounting { | |
3385 | int nr; /* Number of live bh's */ | |
3386 | int ratelimit; /* Limit cacheline bouncing */ | |
3387 | }; | |
3388 | ||
3389 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; | |
3390 | ||
3391 | static void recalc_bh_state(void) | |
3392 | { | |
3393 | int i; | |
3394 | int tot = 0; | |
3395 | ||
ee1be862 | 3396 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
1da177e4 | 3397 | return; |
c7b92516 | 3398 | __this_cpu_write(bh_accounting.ratelimit, 0); |
8a143426 | 3399 | for_each_online_cpu(i) |
1da177e4 LT |
3400 | tot += per_cpu(bh_accounting, i).nr; |
3401 | buffer_heads_over_limit = (tot > max_buffer_heads); | |
3402 | } | |
c7b92516 | 3403 | |
dd0fc66f | 3404 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
1da177e4 | 3405 | { |
019b4d12 | 3406 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
1da177e4 | 3407 | if (ret) { |
a35afb83 | 3408 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
c7b92516 CL |
3409 | preempt_disable(); |
3410 | __this_cpu_inc(bh_accounting.nr); | |
1da177e4 | 3411 | recalc_bh_state(); |
c7b92516 | 3412 | preempt_enable(); |
1da177e4 LT |
3413 | } |
3414 | return ret; | |
3415 | } | |
3416 | EXPORT_SYMBOL(alloc_buffer_head); | |
3417 | ||
3418 | void free_buffer_head(struct buffer_head *bh) | |
3419 | { | |
3420 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); | |
3421 | kmem_cache_free(bh_cachep, bh); | |
c7b92516 CL |
3422 | preempt_disable(); |
3423 | __this_cpu_dec(bh_accounting.nr); | |
1da177e4 | 3424 | recalc_bh_state(); |
c7b92516 | 3425 | preempt_enable(); |
1da177e4 LT |
3426 | } |
3427 | EXPORT_SYMBOL(free_buffer_head); | |
3428 | ||
fc4d24c9 | 3429 | static int buffer_exit_cpu_dead(unsigned int cpu) |
1da177e4 LT |
3430 | { |
3431 | int i; | |
3432 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); | |
3433 | ||
3434 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
3435 | brelse(b->bhs[i]); | |
3436 | b->bhs[i] = NULL; | |
3437 | } | |
c7b92516 | 3438 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
8a143426 | 3439 | per_cpu(bh_accounting, cpu).nr = 0; |
fc4d24c9 | 3440 | return 0; |
1da177e4 | 3441 | } |
1da177e4 | 3442 | |
389d1b08 | 3443 | /** |
a6b91919 | 3444 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
389d1b08 AK |
3445 | * @bh: struct buffer_head |
3446 | * | |
3447 | * Return true if the buffer is up-to-date and false, | |
3448 | * with the buffer locked, if not. | |
3449 | */ | |
3450 | int bh_uptodate_or_lock(struct buffer_head *bh) | |
3451 | { | |
3452 | if (!buffer_uptodate(bh)) { | |
3453 | lock_buffer(bh); | |
3454 | if (!buffer_uptodate(bh)) | |
3455 | return 0; | |
3456 | unlock_buffer(bh); | |
3457 | } | |
3458 | return 1; | |
3459 | } | |
3460 | EXPORT_SYMBOL(bh_uptodate_or_lock); | |
3461 | ||
3462 | /** | |
a6b91919 | 3463 | * bh_submit_read - Submit a locked buffer for reading |
389d1b08 AK |
3464 | * @bh: struct buffer_head |
3465 | * | |
3466 | * Returns zero on success and -EIO on error. | |
3467 | */ | |
3468 | int bh_submit_read(struct buffer_head *bh) | |
3469 | { | |
3470 | BUG_ON(!buffer_locked(bh)); | |
3471 | ||
3472 | if (buffer_uptodate(bh)) { | |
3473 | unlock_buffer(bh); | |
3474 | return 0; | |
3475 | } | |
3476 | ||
3477 | get_bh(bh); | |
3478 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 3479 | submit_bh(REQ_OP_READ, 0, bh); |
389d1b08 AK |
3480 | wait_on_buffer(bh); |
3481 | if (buffer_uptodate(bh)) | |
3482 | return 0; | |
3483 | return -EIO; | |
3484 | } | |
3485 | EXPORT_SYMBOL(bh_submit_read); | |
3486 | ||
1da177e4 LT |
3487 | void __init buffer_init(void) |
3488 | { | |
43be594a | 3489 | unsigned long nrpages; |
fc4d24c9 | 3490 | int ret; |
1da177e4 | 3491 | |
b98938c3 CL |
3492 | bh_cachep = kmem_cache_create("buffer_head", |
3493 | sizeof(struct buffer_head), 0, | |
3494 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
3495 | SLAB_MEM_SPREAD), | |
019b4d12 | 3496 | NULL); |
1da177e4 LT |
3497 | |
3498 | /* | |
3499 | * Limit the bh occupancy to 10% of ZONE_NORMAL | |
3500 | */ | |
3501 | nrpages = (nr_free_buffer_pages() * 10) / 100; | |
3502 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); | |
fc4d24c9 SAS |
3503 | ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", |
3504 | NULL, buffer_exit_cpu_dead); | |
3505 | WARN_ON(ret < 0); | |
1da177e4 | 3506 | } |