]>
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> |
1da177e4 LT |
50 | |
51 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); | |
2a222ca9 | 52 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 53 | enum rw_hint hint, struct writeback_control *wbc); |
1da177e4 LT |
54 | |
55 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) | |
56 | ||
f0059afd TH |
57 | inline void touch_buffer(struct buffer_head *bh) |
58 | { | |
5305cb83 | 59 | trace_block_touch_buffer(bh); |
f0059afd TH |
60 | mark_page_accessed(bh->b_page); |
61 | } | |
62 | EXPORT_SYMBOL(touch_buffer); | |
63 | ||
fc9b52cd | 64 | void __lock_buffer(struct buffer_head *bh) |
1da177e4 | 65 | { |
74316201 | 66 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 LT |
67 | } |
68 | EXPORT_SYMBOL(__lock_buffer); | |
69 | ||
fc9b52cd | 70 | void unlock_buffer(struct buffer_head *bh) |
1da177e4 | 71 | { |
51b07fc3 | 72 | clear_bit_unlock(BH_Lock, &bh->b_state); |
4e857c58 | 73 | smp_mb__after_atomic(); |
1da177e4 LT |
74 | wake_up_bit(&bh->b_state, BH_Lock); |
75 | } | |
1fe72eaa | 76 | EXPORT_SYMBOL(unlock_buffer); |
1da177e4 | 77 | |
b4597226 MG |
78 | /* |
79 | * Returns if the page has dirty or writeback buffers. If all the buffers | |
80 | * are unlocked and clean then the PageDirty information is stale. If | |
81 | * any of the pages are locked, it is assumed they are locked for IO. | |
82 | */ | |
83 | void buffer_check_dirty_writeback(struct page *page, | |
84 | bool *dirty, bool *writeback) | |
85 | { | |
86 | struct buffer_head *head, *bh; | |
87 | *dirty = false; | |
88 | *writeback = false; | |
89 | ||
90 | BUG_ON(!PageLocked(page)); | |
91 | ||
92 | if (!page_has_buffers(page)) | |
93 | return; | |
94 | ||
95 | if (PageWriteback(page)) | |
96 | *writeback = true; | |
97 | ||
98 | head = page_buffers(page); | |
99 | bh = head; | |
100 | do { | |
101 | if (buffer_locked(bh)) | |
102 | *writeback = true; | |
103 | ||
104 | if (buffer_dirty(bh)) | |
105 | *dirty = true; | |
106 | ||
107 | bh = bh->b_this_page; | |
108 | } while (bh != head); | |
109 | } | |
110 | EXPORT_SYMBOL(buffer_check_dirty_writeback); | |
111 | ||
1da177e4 LT |
112 | /* |
113 | * Block until a buffer comes unlocked. This doesn't stop it | |
114 | * from becoming locked again - you have to lock it yourself | |
115 | * if you want to preserve its state. | |
116 | */ | |
117 | void __wait_on_buffer(struct buffer_head * bh) | |
118 | { | |
74316201 | 119 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 | 120 | } |
1fe72eaa | 121 | EXPORT_SYMBOL(__wait_on_buffer); |
1da177e4 LT |
122 | |
123 | static void | |
124 | __clear_page_buffers(struct page *page) | |
125 | { | |
126 | ClearPagePrivate(page); | |
4c21e2f2 | 127 | set_page_private(page, 0); |
09cbfeaf | 128 | put_page(page); |
1da177e4 LT |
129 | } |
130 | ||
b744c2ac | 131 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
1da177e4 | 132 | { |
432f16e6 RE |
133 | if (!test_bit(BH_Quiet, &bh->b_state)) |
134 | printk_ratelimited(KERN_ERR | |
a1c6f057 DM |
135 | "Buffer I/O error on dev %pg, logical block %llu%s\n", |
136 | bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); | |
1da177e4 LT |
137 | } |
138 | ||
139 | /* | |
68671f35 DM |
140 | * End-of-IO handler helper function which does not touch the bh after |
141 | * unlocking it. | |
142 | * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but | |
143 | * a race there is benign: unlock_buffer() only use the bh's address for | |
144 | * hashing after unlocking the buffer, so it doesn't actually touch the bh | |
145 | * itself. | |
1da177e4 | 146 | */ |
68671f35 | 147 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
1da177e4 LT |
148 | { |
149 | if (uptodate) { | |
150 | set_buffer_uptodate(bh); | |
151 | } else { | |
70246286 | 152 | /* This happens, due to failed read-ahead attempts. */ |
1da177e4 LT |
153 | clear_buffer_uptodate(bh); |
154 | } | |
155 | unlock_buffer(bh); | |
68671f35 DM |
156 | } |
157 | ||
158 | /* | |
159 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and | |
160 | * unlock the buffer. This is what ll_rw_block uses too. | |
161 | */ | |
162 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) | |
163 | { | |
164 | __end_buffer_read_notouch(bh, uptodate); | |
1da177e4 LT |
165 | put_bh(bh); |
166 | } | |
1fe72eaa | 167 | EXPORT_SYMBOL(end_buffer_read_sync); |
1da177e4 LT |
168 | |
169 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) | |
170 | { | |
1da177e4 LT |
171 | if (uptodate) { |
172 | set_buffer_uptodate(bh); | |
173 | } else { | |
432f16e6 | 174 | buffer_io_error(bh, ", lost sync page write"); |
87354e5d | 175 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
176 | clear_buffer_uptodate(bh); |
177 | } | |
178 | unlock_buffer(bh); | |
179 | put_bh(bh); | |
180 | } | |
1fe72eaa | 181 | EXPORT_SYMBOL(end_buffer_write_sync); |
1da177e4 | 182 | |
1da177e4 LT |
183 | /* |
184 | * Various filesystems appear to want __find_get_block to be non-blocking. | |
185 | * But it's the page lock which protects the buffers. To get around this, | |
186 | * we get exclusion from try_to_free_buffers with the blockdev mapping's | |
187 | * private_lock. | |
188 | * | |
b93b0163 | 189 | * Hack idea: for the blockdev mapping, private_lock contention |
1da177e4 | 190 | * may be quite high. This code could TryLock the page, and if that |
b93b0163 | 191 | * succeeds, there is no need to take private_lock. |
1da177e4 LT |
192 | */ |
193 | static struct buffer_head * | |
385fd4c5 | 194 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
1da177e4 LT |
195 | { |
196 | struct inode *bd_inode = bdev->bd_inode; | |
197 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
198 | struct buffer_head *ret = NULL; | |
199 | pgoff_t index; | |
200 | struct buffer_head *bh; | |
201 | struct buffer_head *head; | |
202 | struct page *page; | |
203 | int all_mapped = 1; | |
43636c80 | 204 | static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); |
1da177e4 | 205 | |
09cbfeaf | 206 | index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); |
2457aec6 | 207 | page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); |
1da177e4 LT |
208 | if (!page) |
209 | goto out; | |
210 | ||
211 | spin_lock(&bd_mapping->private_lock); | |
212 | if (!page_has_buffers(page)) | |
213 | goto out_unlock; | |
214 | head = page_buffers(page); | |
215 | bh = head; | |
216 | do { | |
97f76d3d NK |
217 | if (!buffer_mapped(bh)) |
218 | all_mapped = 0; | |
219 | else if (bh->b_blocknr == block) { | |
1da177e4 LT |
220 | ret = bh; |
221 | get_bh(bh); | |
222 | goto out_unlock; | |
223 | } | |
1da177e4 LT |
224 | bh = bh->b_this_page; |
225 | } while (bh != head); | |
226 | ||
227 | /* we might be here because some of the buffers on this page are | |
228 | * not mapped. This is due to various races between | |
229 | * file io on the block device and getblk. It gets dealt with | |
230 | * elsewhere, don't buffer_error if we had some unmapped buffers | |
231 | */ | |
43636c80 TH |
232 | ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); |
233 | if (all_mapped && __ratelimit(&last_warned)) { | |
234 | printk("__find_get_block_slow() failed. block=%llu, " | |
235 | "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " | |
236 | "device %pg blocksize: %d\n", | |
237 | (unsigned long long)block, | |
238 | (unsigned long long)bh->b_blocknr, | |
239 | bh->b_state, bh->b_size, bdev, | |
240 | 1 << bd_inode->i_blkbits); | |
1da177e4 LT |
241 | } |
242 | out_unlock: | |
243 | spin_unlock(&bd_mapping->private_lock); | |
09cbfeaf | 244 | put_page(page); |
1da177e4 LT |
245 | out: |
246 | return ret; | |
247 | } | |
248 | ||
1da177e4 LT |
249 | /* |
250 | * I/O completion handler for block_read_full_page() - pages | |
251 | * which come unlocked at the end of I/O. | |
252 | */ | |
253 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) | |
254 | { | |
1da177e4 | 255 | unsigned long flags; |
a3972203 | 256 | struct buffer_head *first; |
1da177e4 LT |
257 | struct buffer_head *tmp; |
258 | struct page *page; | |
259 | int page_uptodate = 1; | |
260 | ||
261 | BUG_ON(!buffer_async_read(bh)); | |
262 | ||
263 | page = bh->b_page; | |
264 | if (uptodate) { | |
265 | set_buffer_uptodate(bh); | |
266 | } else { | |
267 | clear_buffer_uptodate(bh); | |
432f16e6 | 268 | buffer_io_error(bh, ", async page read"); |
1da177e4 LT |
269 | SetPageError(page); |
270 | } | |
271 | ||
272 | /* | |
273 | * Be _very_ careful from here on. Bad things can happen if | |
274 | * two buffer heads end IO at almost the same time and both | |
275 | * decide that the page is now completely done. | |
276 | */ | |
a3972203 NP |
277 | first = page_buffers(page); |
278 | local_irq_save(flags); | |
279 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
1da177e4 LT |
280 | clear_buffer_async_read(bh); |
281 | unlock_buffer(bh); | |
282 | tmp = bh; | |
283 | do { | |
284 | if (!buffer_uptodate(tmp)) | |
285 | page_uptodate = 0; | |
286 | if (buffer_async_read(tmp)) { | |
287 | BUG_ON(!buffer_locked(tmp)); | |
288 | goto still_busy; | |
289 | } | |
290 | tmp = tmp->b_this_page; | |
291 | } while (tmp != bh); | |
a3972203 NP |
292 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
293 | local_irq_restore(flags); | |
1da177e4 LT |
294 | |
295 | /* | |
296 | * If none of the buffers had errors and they are all | |
297 | * uptodate then we can set the page uptodate. | |
298 | */ | |
299 | if (page_uptodate && !PageError(page)) | |
300 | SetPageUptodate(page); | |
301 | unlock_page(page); | |
302 | return; | |
303 | ||
304 | still_busy: | |
a3972203 NP |
305 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
306 | local_irq_restore(flags); | |
1da177e4 LT |
307 | return; |
308 | } | |
309 | ||
310 | /* | |
311 | * Completion handler for block_write_full_page() - pages which are unlocked | |
312 | * during I/O, and which have PageWriteback cleared upon I/O completion. | |
313 | */ | |
35c80d5f | 314 | void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
1da177e4 | 315 | { |
1da177e4 | 316 | unsigned long flags; |
a3972203 | 317 | struct buffer_head *first; |
1da177e4 LT |
318 | struct buffer_head *tmp; |
319 | struct page *page; | |
320 | ||
321 | BUG_ON(!buffer_async_write(bh)); | |
322 | ||
323 | page = bh->b_page; | |
324 | if (uptodate) { | |
325 | set_buffer_uptodate(bh); | |
326 | } else { | |
432f16e6 | 327 | buffer_io_error(bh, ", lost async page write"); |
87354e5d | 328 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
329 | clear_buffer_uptodate(bh); |
330 | SetPageError(page); | |
331 | } | |
332 | ||
a3972203 NP |
333 | first = page_buffers(page); |
334 | local_irq_save(flags); | |
335 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
336 | ||
1da177e4 LT |
337 | clear_buffer_async_write(bh); |
338 | unlock_buffer(bh); | |
339 | tmp = bh->b_this_page; | |
340 | while (tmp != bh) { | |
341 | if (buffer_async_write(tmp)) { | |
342 | BUG_ON(!buffer_locked(tmp)); | |
343 | goto still_busy; | |
344 | } | |
345 | tmp = tmp->b_this_page; | |
346 | } | |
a3972203 NP |
347 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
348 | local_irq_restore(flags); | |
1da177e4 LT |
349 | end_page_writeback(page); |
350 | return; | |
351 | ||
352 | still_busy: | |
a3972203 NP |
353 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
354 | local_irq_restore(flags); | |
1da177e4 LT |
355 | return; |
356 | } | |
1fe72eaa | 357 | EXPORT_SYMBOL(end_buffer_async_write); |
1da177e4 LT |
358 | |
359 | /* | |
360 | * If a page's buffers are under async readin (end_buffer_async_read | |
361 | * completion) then there is a possibility that another thread of | |
362 | * control could lock one of the buffers after it has completed | |
363 | * but while some of the other buffers have not completed. This | |
364 | * locked buffer would confuse end_buffer_async_read() into not unlocking | |
365 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() | |
366 | * that this buffer is not under async I/O. | |
367 | * | |
368 | * The page comes unlocked when it has no locked buffer_async buffers | |
369 | * left. | |
370 | * | |
371 | * PageLocked prevents anyone starting new async I/O reads any of | |
372 | * the buffers. | |
373 | * | |
374 | * PageWriteback is used to prevent simultaneous writeout of the same | |
375 | * page. | |
376 | * | |
377 | * PageLocked prevents anyone from starting writeback of a page which is | |
378 | * under read I/O (PageWriteback is only ever set against a locked page). | |
379 | */ | |
380 | static void mark_buffer_async_read(struct buffer_head *bh) | |
381 | { | |
382 | bh->b_end_io = end_buffer_async_read; | |
383 | set_buffer_async_read(bh); | |
384 | } | |
385 | ||
1fe72eaa HS |
386 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
387 | bh_end_io_t *handler) | |
1da177e4 | 388 | { |
35c80d5f | 389 | bh->b_end_io = handler; |
1da177e4 LT |
390 | set_buffer_async_write(bh); |
391 | } | |
35c80d5f CM |
392 | |
393 | void mark_buffer_async_write(struct buffer_head *bh) | |
394 | { | |
395 | mark_buffer_async_write_endio(bh, end_buffer_async_write); | |
396 | } | |
1da177e4 LT |
397 | EXPORT_SYMBOL(mark_buffer_async_write); |
398 | ||
399 | ||
400 | /* | |
401 | * fs/buffer.c contains helper functions for buffer-backed address space's | |
402 | * fsync functions. A common requirement for buffer-based filesystems is | |
403 | * that certain data from the backing blockdev needs to be written out for | |
404 | * a successful fsync(). For example, ext2 indirect blocks need to be | |
405 | * written back and waited upon before fsync() returns. | |
406 | * | |
407 | * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), | |
408 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the | |
409 | * management of a list of dependent buffers at ->i_mapping->private_list. | |
410 | * | |
411 | * Locking is a little subtle: try_to_free_buffers() will remove buffers | |
412 | * from their controlling inode's queue when they are being freed. But | |
413 | * try_to_free_buffers() will be operating against the *blockdev* mapping | |
414 | * at the time, not against the S_ISREG file which depends on those buffers. | |
415 | * So the locking for private_list is via the private_lock in the address_space | |
416 | * which backs the buffers. Which is different from the address_space | |
417 | * against which the buffers are listed. So for a particular address_space, | |
418 | * mapping->private_lock does *not* protect mapping->private_list! In fact, | |
419 | * mapping->private_list will always be protected by the backing blockdev's | |
420 | * ->private_lock. | |
421 | * | |
422 | * Which introduces a requirement: all buffers on an address_space's | |
423 | * ->private_list must be from the same address_space: the blockdev's. | |
424 | * | |
425 | * address_spaces which do not place buffers at ->private_list via these | |
426 | * utility functions are free to use private_lock and private_list for | |
427 | * whatever they want. The only requirement is that list_empty(private_list) | |
428 | * be true at clear_inode() time. | |
429 | * | |
430 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The | |
431 | * filesystems should do that. invalidate_inode_buffers() should just go | |
432 | * BUG_ON(!list_empty). | |
433 | * | |
434 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should | |
435 | * take an address_space, not an inode. And it should be called | |
436 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being | |
437 | * queued up. | |
438 | * | |
439 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the | |
440 | * list if it is already on a list. Because if the buffer is on a list, | |
441 | * it *must* already be on the right one. If not, the filesystem is being | |
442 | * silly. This will save a ton of locking. But first we have to ensure | |
443 | * that buffers are taken *off* the old inode's list when they are freed | |
444 | * (presumably in truncate). That requires careful auditing of all | |
445 | * filesystems (do it inside bforget()). It could also be done by bringing | |
446 | * b_inode back. | |
447 | */ | |
448 | ||
449 | /* | |
450 | * The buffer's backing address_space's private_lock must be held | |
451 | */ | |
dbacefc9 | 452 | static void __remove_assoc_queue(struct buffer_head *bh) |
1da177e4 LT |
453 | { |
454 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 455 | WARN_ON(!bh->b_assoc_map); |
58ff407b | 456 | bh->b_assoc_map = NULL; |
1da177e4 LT |
457 | } |
458 | ||
459 | int inode_has_buffers(struct inode *inode) | |
460 | { | |
461 | return !list_empty(&inode->i_data.private_list); | |
462 | } | |
463 | ||
464 | /* | |
465 | * osync is designed to support O_SYNC io. It waits synchronously for | |
466 | * all already-submitted IO to complete, but does not queue any new | |
467 | * writes to the disk. | |
468 | * | |
469 | * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as | |
470 | * you dirty the buffers, and then use osync_inode_buffers to wait for | |
471 | * completion. Any other dirty buffers which are not yet queued for | |
472 | * write will not be flushed to disk by the osync. | |
473 | */ | |
474 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) | |
475 | { | |
476 | struct buffer_head *bh; | |
477 | struct list_head *p; | |
478 | int err = 0; | |
479 | ||
480 | spin_lock(lock); | |
481 | repeat: | |
482 | list_for_each_prev(p, list) { | |
483 | bh = BH_ENTRY(p); | |
484 | if (buffer_locked(bh)) { | |
485 | get_bh(bh); | |
486 | spin_unlock(lock); | |
487 | wait_on_buffer(bh); | |
488 | if (!buffer_uptodate(bh)) | |
489 | err = -EIO; | |
490 | brelse(bh); | |
491 | spin_lock(lock); | |
492 | goto repeat; | |
493 | } | |
494 | } | |
495 | spin_unlock(lock); | |
496 | return err; | |
497 | } | |
498 | ||
08fdc8a0 | 499 | void emergency_thaw_bdev(struct super_block *sb) |
c2d75438 | 500 | { |
01a05b33 | 501 | while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) |
a1c6f057 | 502 | printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); |
01a05b33 | 503 | } |
c2d75438 | 504 | |
1da177e4 | 505 | /** |
78a4a50a | 506 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
67be2dd1 | 507 | * @mapping: the mapping which wants those buffers written |
1da177e4 LT |
508 | * |
509 | * Starts I/O against the buffers at mapping->private_list, and waits upon | |
510 | * that I/O. | |
511 | * | |
67be2dd1 MW |
512 | * Basically, this is a convenience function for fsync(). |
513 | * @mapping is a file or directory which needs those buffers to be written for | |
514 | * a successful fsync(). | |
1da177e4 LT |
515 | */ |
516 | int sync_mapping_buffers(struct address_space *mapping) | |
517 | { | |
252aa6f5 | 518 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
519 | |
520 | if (buffer_mapping == NULL || list_empty(&mapping->private_list)) | |
521 | return 0; | |
522 | ||
523 | return fsync_buffers_list(&buffer_mapping->private_lock, | |
524 | &mapping->private_list); | |
525 | } | |
526 | EXPORT_SYMBOL(sync_mapping_buffers); | |
527 | ||
528 | /* | |
529 | * Called when we've recently written block `bblock', and it is known that | |
530 | * `bblock' was for a buffer_boundary() buffer. This means that the block at | |
531 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's | |
532 | * dirty, schedule it for IO. So that indirects merge nicely with their data. | |
533 | */ | |
534 | void write_boundary_block(struct block_device *bdev, | |
535 | sector_t bblock, unsigned blocksize) | |
536 | { | |
537 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); | |
538 | if (bh) { | |
539 | if (buffer_dirty(bh)) | |
dfec8a14 | 540 | ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); |
1da177e4 LT |
541 | put_bh(bh); |
542 | } | |
543 | } | |
544 | ||
545 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) | |
546 | { | |
547 | struct address_space *mapping = inode->i_mapping; | |
548 | struct address_space *buffer_mapping = bh->b_page->mapping; | |
549 | ||
550 | mark_buffer_dirty(bh); | |
252aa6f5 RA |
551 | if (!mapping->private_data) { |
552 | mapping->private_data = buffer_mapping; | |
1da177e4 | 553 | } else { |
252aa6f5 | 554 | BUG_ON(mapping->private_data != buffer_mapping); |
1da177e4 | 555 | } |
535ee2fb | 556 | if (!bh->b_assoc_map) { |
1da177e4 LT |
557 | spin_lock(&buffer_mapping->private_lock); |
558 | list_move_tail(&bh->b_assoc_buffers, | |
559 | &mapping->private_list); | |
58ff407b | 560 | bh->b_assoc_map = mapping; |
1da177e4 LT |
561 | spin_unlock(&buffer_mapping->private_lock); |
562 | } | |
563 | } | |
564 | EXPORT_SYMBOL(mark_buffer_dirty_inode); | |
565 | ||
787d2214 | 566 | /* |
ec82e1c1 | 567 | * Mark the page dirty, and set it dirty in the page cache, and mark the inode |
787d2214 NP |
568 | * dirty. |
569 | * | |
570 | * If warn is true, then emit a warning if the page is not uptodate and has | |
571 | * not been truncated. | |
c4843a75 | 572 | * |
81f8c3a4 | 573 | * The caller must hold lock_page_memcg(). |
787d2214 | 574 | */ |
f82b3764 | 575 | void __set_page_dirty(struct page *page, struct address_space *mapping, |
62cccb8c | 576 | int warn) |
787d2214 | 577 | { |
227d53b3 KM |
578 | unsigned long flags; |
579 | ||
b93b0163 | 580 | xa_lock_irqsave(&mapping->i_pages, flags); |
787d2214 NP |
581 | if (page->mapping) { /* Race with truncate? */ |
582 | WARN_ON_ONCE(warn && !PageUptodate(page)); | |
62cccb8c | 583 | account_page_dirtied(page, mapping); |
ec82e1c1 MW |
584 | __xa_set_mark(&mapping->i_pages, page_index(page), |
585 | PAGECACHE_TAG_DIRTY); | |
787d2214 | 586 | } |
b93b0163 | 587 | xa_unlock_irqrestore(&mapping->i_pages, flags); |
787d2214 | 588 | } |
f82b3764 | 589 | EXPORT_SYMBOL_GPL(__set_page_dirty); |
787d2214 | 590 | |
1da177e4 LT |
591 | /* |
592 | * Add a page to the dirty page list. | |
593 | * | |
594 | * It is a sad fact of life that this function is called from several places | |
595 | * deeply under spinlocking. It may not sleep. | |
596 | * | |
597 | * If the page has buffers, the uptodate buffers are set dirty, to preserve | |
598 | * dirty-state coherency between the page and the buffers. It the page does | |
599 | * not have buffers then when they are later attached they will all be set | |
600 | * dirty. | |
601 | * | |
602 | * The buffers are dirtied before the page is dirtied. There's a small race | |
603 | * window in which a writepage caller may see the page cleanness but not the | |
604 | * buffer dirtiness. That's fine. If this code were to set the page dirty | |
605 | * before the buffers, a concurrent writepage caller could clear the page dirty | |
606 | * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean | |
607 | * page on the dirty page list. | |
608 | * | |
609 | * We use private_lock to lock against try_to_free_buffers while using the | |
610 | * page's buffer list. Also use this to protect against clean buffers being | |
611 | * added to the page after it was set dirty. | |
612 | * | |
613 | * FIXME: may need to call ->reservepage here as well. That's rather up to the | |
614 | * address_space though. | |
615 | */ | |
616 | int __set_page_dirty_buffers(struct page *page) | |
617 | { | |
a8e7d49a | 618 | int newly_dirty; |
787d2214 | 619 | struct address_space *mapping = page_mapping(page); |
ebf7a227 NP |
620 | |
621 | if (unlikely(!mapping)) | |
622 | return !TestSetPageDirty(page); | |
1da177e4 LT |
623 | |
624 | spin_lock(&mapping->private_lock); | |
625 | if (page_has_buffers(page)) { | |
626 | struct buffer_head *head = page_buffers(page); | |
627 | struct buffer_head *bh = head; | |
628 | ||
629 | do { | |
630 | set_buffer_dirty(bh); | |
631 | bh = bh->b_this_page; | |
632 | } while (bh != head); | |
633 | } | |
c4843a75 | 634 | /* |
81f8c3a4 JW |
635 | * Lock out page->mem_cgroup migration to keep PageDirty |
636 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 637 | */ |
62cccb8c | 638 | lock_page_memcg(page); |
a8e7d49a | 639 | newly_dirty = !TestSetPageDirty(page); |
1da177e4 LT |
640 | spin_unlock(&mapping->private_lock); |
641 | ||
a8e7d49a | 642 | if (newly_dirty) |
62cccb8c | 643 | __set_page_dirty(page, mapping, 1); |
c4843a75 | 644 | |
62cccb8c | 645 | unlock_page_memcg(page); |
c4843a75 GT |
646 | |
647 | if (newly_dirty) | |
648 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
649 | ||
a8e7d49a | 650 | return newly_dirty; |
1da177e4 LT |
651 | } |
652 | EXPORT_SYMBOL(__set_page_dirty_buffers); | |
653 | ||
654 | /* | |
655 | * Write out and wait upon a list of buffers. | |
656 | * | |
657 | * We have conflicting pressures: we want to make sure that all | |
658 | * initially dirty buffers get waited on, but that any subsequently | |
659 | * dirtied buffers don't. After all, we don't want fsync to last | |
660 | * forever if somebody is actively writing to the file. | |
661 | * | |
662 | * Do this in two main stages: first we copy dirty buffers to a | |
663 | * temporary inode list, queueing the writes as we go. Then we clean | |
664 | * up, waiting for those writes to complete. | |
665 | * | |
666 | * During this second stage, any subsequent updates to the file may end | |
667 | * up refiling the buffer on the original inode's dirty list again, so | |
668 | * there is a chance we will end up with a buffer queued for write but | |
669 | * not yet completed on that list. So, as a final cleanup we go through | |
670 | * the osync code to catch these locked, dirty buffers without requeuing | |
671 | * any newly dirty buffers for write. | |
672 | */ | |
673 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) | |
674 | { | |
675 | struct buffer_head *bh; | |
676 | struct list_head tmp; | |
7eaceacc | 677 | struct address_space *mapping; |
1da177e4 | 678 | int err = 0, err2; |
4ee2491e | 679 | struct blk_plug plug; |
1da177e4 LT |
680 | |
681 | INIT_LIST_HEAD(&tmp); | |
4ee2491e | 682 | blk_start_plug(&plug); |
1da177e4 LT |
683 | |
684 | spin_lock(lock); | |
685 | while (!list_empty(list)) { | |
686 | bh = BH_ENTRY(list->next); | |
535ee2fb | 687 | mapping = bh->b_assoc_map; |
58ff407b | 688 | __remove_assoc_queue(bh); |
535ee2fb JK |
689 | /* Avoid race with mark_buffer_dirty_inode() which does |
690 | * a lockless check and we rely on seeing the dirty bit */ | |
691 | smp_mb(); | |
1da177e4 LT |
692 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
693 | list_add(&bh->b_assoc_buffers, &tmp); | |
535ee2fb | 694 | bh->b_assoc_map = mapping; |
1da177e4 LT |
695 | if (buffer_dirty(bh)) { |
696 | get_bh(bh); | |
697 | spin_unlock(lock); | |
698 | /* | |
699 | * Ensure any pending I/O completes so that | |
9cb569d6 CH |
700 | * write_dirty_buffer() actually writes the |
701 | * current contents - it is a noop if I/O is | |
702 | * still in flight on potentially older | |
703 | * contents. | |
1da177e4 | 704 | */ |
70fd7614 | 705 | write_dirty_buffer(bh, REQ_SYNC); |
9cf6b720 JA |
706 | |
707 | /* | |
708 | * Kick off IO for the previous mapping. Note | |
709 | * that we will not run the very last mapping, | |
710 | * wait_on_buffer() will do that for us | |
711 | * through sync_buffer(). | |
712 | */ | |
1da177e4 LT |
713 | brelse(bh); |
714 | spin_lock(lock); | |
715 | } | |
716 | } | |
717 | } | |
718 | ||
4ee2491e JA |
719 | spin_unlock(lock); |
720 | blk_finish_plug(&plug); | |
721 | spin_lock(lock); | |
722 | ||
1da177e4 LT |
723 | while (!list_empty(&tmp)) { |
724 | bh = BH_ENTRY(tmp.prev); | |
1da177e4 | 725 | get_bh(bh); |
535ee2fb JK |
726 | mapping = bh->b_assoc_map; |
727 | __remove_assoc_queue(bh); | |
728 | /* Avoid race with mark_buffer_dirty_inode() which does | |
729 | * a lockless check and we rely on seeing the dirty bit */ | |
730 | smp_mb(); | |
731 | if (buffer_dirty(bh)) { | |
732 | list_add(&bh->b_assoc_buffers, | |
e3892296 | 733 | &mapping->private_list); |
535ee2fb JK |
734 | bh->b_assoc_map = mapping; |
735 | } | |
1da177e4 LT |
736 | spin_unlock(lock); |
737 | wait_on_buffer(bh); | |
738 | if (!buffer_uptodate(bh)) | |
739 | err = -EIO; | |
740 | brelse(bh); | |
741 | spin_lock(lock); | |
742 | } | |
743 | ||
744 | spin_unlock(lock); | |
745 | err2 = osync_buffers_list(lock, list); | |
746 | if (err) | |
747 | return err; | |
748 | else | |
749 | return err2; | |
750 | } | |
751 | ||
752 | /* | |
753 | * Invalidate any and all dirty buffers on a given inode. We are | |
754 | * probably unmounting the fs, but that doesn't mean we have already | |
755 | * done a sync(). Just drop the buffers from the inode list. | |
756 | * | |
757 | * NOTE: we take the inode's blockdev's mapping's private_lock. Which | |
758 | * assumes that all the buffers are against the blockdev. Not true | |
759 | * for reiserfs. | |
760 | */ | |
761 | void invalidate_inode_buffers(struct inode *inode) | |
762 | { | |
763 | if (inode_has_buffers(inode)) { | |
764 | struct address_space *mapping = &inode->i_data; | |
765 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 766 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
767 | |
768 | spin_lock(&buffer_mapping->private_lock); | |
769 | while (!list_empty(list)) | |
770 | __remove_assoc_queue(BH_ENTRY(list->next)); | |
771 | spin_unlock(&buffer_mapping->private_lock); | |
772 | } | |
773 | } | |
52b19ac9 | 774 | EXPORT_SYMBOL(invalidate_inode_buffers); |
1da177e4 LT |
775 | |
776 | /* | |
777 | * Remove any clean buffers from the inode's buffer list. This is called | |
778 | * when we're trying to free the inode itself. Those buffers can pin it. | |
779 | * | |
780 | * Returns true if all buffers were removed. | |
781 | */ | |
782 | int remove_inode_buffers(struct inode *inode) | |
783 | { | |
784 | int ret = 1; | |
785 | ||
786 | if (inode_has_buffers(inode)) { | |
787 | struct address_space *mapping = &inode->i_data; | |
788 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 789 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
790 | |
791 | spin_lock(&buffer_mapping->private_lock); | |
792 | while (!list_empty(list)) { | |
793 | struct buffer_head *bh = BH_ENTRY(list->next); | |
794 | if (buffer_dirty(bh)) { | |
795 | ret = 0; | |
796 | break; | |
797 | } | |
798 | __remove_assoc_queue(bh); | |
799 | } | |
800 | spin_unlock(&buffer_mapping->private_lock); | |
801 | } | |
802 | return ret; | |
803 | } | |
804 | ||
805 | /* | |
806 | * Create the appropriate buffers when given a page for data area and | |
807 | * the size of each buffer.. Use the bh->b_this_page linked list to | |
808 | * follow the buffers created. Return NULL if unable to create more | |
809 | * buffers. | |
810 | * | |
811 | * The retry flag is used to differentiate async IO (paging, swapping) | |
812 | * which may not fail from ordinary buffer allocations. | |
813 | */ | |
814 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, | |
640ab98f | 815 | bool retry) |
1da177e4 LT |
816 | { |
817 | struct buffer_head *bh, *head; | |
f745c6f5 | 818 | gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; |
1da177e4 | 819 | long offset; |
f745c6f5 | 820 | struct mem_cgroup *memcg; |
1da177e4 | 821 | |
640ab98f JA |
822 | if (retry) |
823 | gfp |= __GFP_NOFAIL; | |
824 | ||
f745c6f5 SB |
825 | memcg = get_mem_cgroup_from_page(page); |
826 | memalloc_use_memcg(memcg); | |
827 | ||
1da177e4 LT |
828 | head = NULL; |
829 | offset = PAGE_SIZE; | |
830 | while ((offset -= size) >= 0) { | |
640ab98f | 831 | bh = alloc_buffer_head(gfp); |
1da177e4 LT |
832 | if (!bh) |
833 | goto no_grow; | |
834 | ||
1da177e4 LT |
835 | bh->b_this_page = head; |
836 | bh->b_blocknr = -1; | |
837 | head = bh; | |
838 | ||
1da177e4 LT |
839 | bh->b_size = size; |
840 | ||
841 | /* Link the buffer to its page */ | |
842 | set_bh_page(bh, page, offset); | |
1da177e4 | 843 | } |
f745c6f5 SB |
844 | out: |
845 | memalloc_unuse_memcg(); | |
846 | mem_cgroup_put(memcg); | |
1da177e4 LT |
847 | return head; |
848 | /* | |
849 | * In case anything failed, we just free everything we got. | |
850 | */ | |
851 | no_grow: | |
852 | if (head) { | |
853 | do { | |
854 | bh = head; | |
855 | head = head->b_this_page; | |
856 | free_buffer_head(bh); | |
857 | } while (head); | |
858 | } | |
859 | ||
f745c6f5 | 860 | goto out; |
1da177e4 LT |
861 | } |
862 | EXPORT_SYMBOL_GPL(alloc_page_buffers); | |
863 | ||
864 | static inline void | |
865 | link_dev_buffers(struct page *page, struct buffer_head *head) | |
866 | { | |
867 | struct buffer_head *bh, *tail; | |
868 | ||
869 | bh = head; | |
870 | do { | |
871 | tail = bh; | |
872 | bh = bh->b_this_page; | |
873 | } while (bh); | |
874 | tail->b_this_page = head; | |
875 | attach_page_buffers(page, head); | |
876 | } | |
877 | ||
bbec0270 LT |
878 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
879 | { | |
880 | sector_t retval = ~((sector_t)0); | |
881 | loff_t sz = i_size_read(bdev->bd_inode); | |
882 | ||
883 | if (sz) { | |
884 | unsigned int sizebits = blksize_bits(size); | |
885 | retval = (sz >> sizebits); | |
886 | } | |
887 | return retval; | |
888 | } | |
889 | ||
1da177e4 LT |
890 | /* |
891 | * Initialise the state of a blockdev page's buffers. | |
892 | */ | |
676ce6d5 | 893 | static sector_t |
1da177e4 LT |
894 | init_page_buffers(struct page *page, struct block_device *bdev, |
895 | sector_t block, int size) | |
896 | { | |
897 | struct buffer_head *head = page_buffers(page); | |
898 | struct buffer_head *bh = head; | |
899 | int uptodate = PageUptodate(page); | |
bbec0270 | 900 | sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); |
1da177e4 LT |
901 | |
902 | do { | |
903 | if (!buffer_mapped(bh)) { | |
01950a34 EB |
904 | bh->b_end_io = NULL; |
905 | bh->b_private = NULL; | |
1da177e4 LT |
906 | bh->b_bdev = bdev; |
907 | bh->b_blocknr = block; | |
908 | if (uptodate) | |
909 | set_buffer_uptodate(bh); | |
080399aa JM |
910 | if (block < end_block) |
911 | set_buffer_mapped(bh); | |
1da177e4 LT |
912 | } |
913 | block++; | |
914 | bh = bh->b_this_page; | |
915 | } while (bh != head); | |
676ce6d5 HD |
916 | |
917 | /* | |
918 | * Caller needs to validate requested block against end of device. | |
919 | */ | |
920 | return end_block; | |
1da177e4 LT |
921 | } |
922 | ||
923 | /* | |
924 | * Create the page-cache page that contains the requested block. | |
925 | * | |
676ce6d5 | 926 | * This is used purely for blockdev mappings. |
1da177e4 | 927 | */ |
676ce6d5 | 928 | static int |
1da177e4 | 929 | grow_dev_page(struct block_device *bdev, sector_t block, |
3b5e6454 | 930 | pgoff_t index, int size, int sizebits, gfp_t gfp) |
1da177e4 LT |
931 | { |
932 | struct inode *inode = bdev->bd_inode; | |
933 | struct page *page; | |
934 | struct buffer_head *bh; | |
676ce6d5 HD |
935 | sector_t end_block; |
936 | int ret = 0; /* Will call free_more_memory() */ | |
84235de3 | 937 | gfp_t gfp_mask; |
1da177e4 | 938 | |
c62d2555 | 939 | gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; |
3b5e6454 | 940 | |
84235de3 JW |
941 | /* |
942 | * XXX: __getblk_slow() can not really deal with failure and | |
943 | * will endlessly loop on improvised global reclaim. Prefer | |
944 | * looping in the allocator rather than here, at least that | |
945 | * code knows what it's doing. | |
946 | */ | |
947 | gfp_mask |= __GFP_NOFAIL; | |
948 | ||
949 | page = find_or_create_page(inode->i_mapping, index, gfp_mask); | |
1da177e4 | 950 | |
e827f923 | 951 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
952 | |
953 | if (page_has_buffers(page)) { | |
954 | bh = page_buffers(page); | |
955 | if (bh->b_size == size) { | |
676ce6d5 | 956 | end_block = init_page_buffers(page, bdev, |
f2d5a944 AA |
957 | (sector_t)index << sizebits, |
958 | size); | |
676ce6d5 | 959 | goto done; |
1da177e4 LT |
960 | } |
961 | if (!try_to_free_buffers(page)) | |
962 | goto failed; | |
963 | } | |
964 | ||
965 | /* | |
966 | * Allocate some buffers for this page | |
967 | */ | |
94dc24c0 | 968 | bh = alloc_page_buffers(page, size, true); |
1da177e4 LT |
969 | |
970 | /* | |
971 | * Link the page to the buffers and initialise them. Take the | |
972 | * lock to be atomic wrt __find_get_block(), which does not | |
973 | * run under the page lock. | |
974 | */ | |
975 | spin_lock(&inode->i_mapping->private_lock); | |
976 | link_dev_buffers(page, bh); | |
f2d5a944 AA |
977 | end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, |
978 | size); | |
1da177e4 | 979 | spin_unlock(&inode->i_mapping->private_lock); |
676ce6d5 HD |
980 | done: |
981 | ret = (block < end_block) ? 1 : -ENXIO; | |
1da177e4 | 982 | failed: |
1da177e4 | 983 | unlock_page(page); |
09cbfeaf | 984 | put_page(page); |
676ce6d5 | 985 | return ret; |
1da177e4 LT |
986 | } |
987 | ||
988 | /* | |
989 | * Create buffers for the specified block device block's page. If | |
990 | * that page was dirty, the buffers are set dirty also. | |
1da177e4 | 991 | */ |
858119e1 | 992 | static int |
3b5e6454 | 993 | grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) |
1da177e4 | 994 | { |
1da177e4 LT |
995 | pgoff_t index; |
996 | int sizebits; | |
997 | ||
998 | sizebits = -1; | |
999 | do { | |
1000 | sizebits++; | |
1001 | } while ((size << sizebits) < PAGE_SIZE); | |
1002 | ||
1003 | index = block >> sizebits; | |
1da177e4 | 1004 | |
e5657933 AM |
1005 | /* |
1006 | * Check for a block which wants to lie outside our maximum possible | |
1007 | * pagecache index. (this comparison is done using sector_t types). | |
1008 | */ | |
1009 | if (unlikely(index != block >> sizebits)) { | |
e5657933 | 1010 | printk(KERN_ERR "%s: requested out-of-range block %llu for " |
a1c6f057 | 1011 | "device %pg\n", |
8e24eea7 | 1012 | __func__, (unsigned long long)block, |
a1c6f057 | 1013 | bdev); |
e5657933 AM |
1014 | return -EIO; |
1015 | } | |
676ce6d5 | 1016 | |
1da177e4 | 1017 | /* Create a page with the proper size buffers.. */ |
3b5e6454 | 1018 | return grow_dev_page(bdev, block, index, size, sizebits, gfp); |
1da177e4 LT |
1019 | } |
1020 | ||
0026ba40 | 1021 | static struct buffer_head * |
3b5e6454 GK |
1022 | __getblk_slow(struct block_device *bdev, sector_t block, |
1023 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1024 | { |
1025 | /* Size must be multiple of hard sectorsize */ | |
e1defc4f | 1026 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
1da177e4 LT |
1027 | (size < 512 || size > PAGE_SIZE))) { |
1028 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", | |
1029 | size); | |
e1defc4f MP |
1030 | printk(KERN_ERR "logical block size: %d\n", |
1031 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1032 | |
1033 | dump_stack(); | |
1034 | return NULL; | |
1035 | } | |
1036 | ||
676ce6d5 HD |
1037 | for (;;) { |
1038 | struct buffer_head *bh; | |
1039 | int ret; | |
1da177e4 LT |
1040 | |
1041 | bh = __find_get_block(bdev, block, size); | |
1042 | if (bh) | |
1043 | return bh; | |
676ce6d5 | 1044 | |
3b5e6454 | 1045 | ret = grow_buffers(bdev, block, size, gfp); |
676ce6d5 HD |
1046 | if (ret < 0) |
1047 | return NULL; | |
1da177e4 LT |
1048 | } |
1049 | } | |
1050 | ||
1051 | /* | |
1052 | * The relationship between dirty buffers and dirty pages: | |
1053 | * | |
1054 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and | |
ec82e1c1 | 1055 | * the page is tagged dirty in the page cache. |
1da177e4 LT |
1056 | * |
1057 | * At all times, the dirtiness of the buffers represents the dirtiness of | |
1058 | * subsections of the page. If the page has buffers, the page dirty bit is | |
1059 | * merely a hint about the true dirty state. | |
1060 | * | |
1061 | * When a page is set dirty in its entirety, all its buffers are marked dirty | |
1062 | * (if the page has buffers). | |
1063 | * | |
1064 | * When a buffer is marked dirty, its page is dirtied, but the page's other | |
1065 | * buffers are not. | |
1066 | * | |
1067 | * Also. When blockdev buffers are explicitly read with bread(), they | |
1068 | * individually become uptodate. But their backing page remains not | |
1069 | * uptodate - even if all of its buffers are uptodate. A subsequent | |
1070 | * block_read_full_page() against that page will discover all the uptodate | |
1071 | * buffers, will set the page uptodate and will perform no I/O. | |
1072 | */ | |
1073 | ||
1074 | /** | |
1075 | * mark_buffer_dirty - mark a buffer_head as needing writeout | |
67be2dd1 | 1076 | * @bh: the buffer_head to mark dirty |
1da177e4 | 1077 | * |
ec82e1c1 MW |
1078 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set |
1079 | * its backing page dirty, then tag the page as dirty in the page cache | |
1080 | * and then attach the address_space's inode to its superblock's dirty | |
1da177e4 LT |
1081 | * inode list. |
1082 | * | |
1083 | * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, | |
b93b0163 | 1084 | * i_pages lock and mapping->host->i_lock. |
1da177e4 | 1085 | */ |
fc9b52cd | 1086 | void mark_buffer_dirty(struct buffer_head *bh) |
1da177e4 | 1087 | { |
787d2214 | 1088 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
1be62dc1 | 1089 | |
5305cb83 TH |
1090 | trace_block_dirty_buffer(bh); |
1091 | ||
1be62dc1 LT |
1092 | /* |
1093 | * Very *carefully* optimize the it-is-already-dirty case. | |
1094 | * | |
1095 | * Don't let the final "is it dirty" escape to before we | |
1096 | * perhaps modified the buffer. | |
1097 | */ | |
1098 | if (buffer_dirty(bh)) { | |
1099 | smp_mb(); | |
1100 | if (buffer_dirty(bh)) | |
1101 | return; | |
1102 | } | |
1103 | ||
a8e7d49a LT |
1104 | if (!test_set_buffer_dirty(bh)) { |
1105 | struct page *page = bh->b_page; | |
c4843a75 | 1106 | struct address_space *mapping = NULL; |
c4843a75 | 1107 | |
62cccb8c | 1108 | lock_page_memcg(page); |
8e9d78ed | 1109 | if (!TestSetPageDirty(page)) { |
c4843a75 | 1110 | mapping = page_mapping(page); |
8e9d78ed | 1111 | if (mapping) |
62cccb8c | 1112 | __set_page_dirty(page, mapping, 0); |
8e9d78ed | 1113 | } |
62cccb8c | 1114 | unlock_page_memcg(page); |
c4843a75 GT |
1115 | if (mapping) |
1116 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
a8e7d49a | 1117 | } |
1da177e4 | 1118 | } |
1fe72eaa | 1119 | EXPORT_SYMBOL(mark_buffer_dirty); |
1da177e4 | 1120 | |
87354e5d JL |
1121 | void mark_buffer_write_io_error(struct buffer_head *bh) |
1122 | { | |
1123 | set_buffer_write_io_error(bh); | |
1124 | /* FIXME: do we need to set this in both places? */ | |
1125 | if (bh->b_page && bh->b_page->mapping) | |
1126 | mapping_set_error(bh->b_page->mapping, -EIO); | |
1127 | if (bh->b_assoc_map) | |
1128 | mapping_set_error(bh->b_assoc_map, -EIO); | |
1129 | } | |
1130 | EXPORT_SYMBOL(mark_buffer_write_io_error); | |
1131 | ||
1da177e4 LT |
1132 | /* |
1133 | * Decrement a buffer_head's reference count. If all buffers against a page | |
1134 | * have zero reference count, are clean and unlocked, and if the page is clean | |
1135 | * and unlocked then try_to_free_buffers() may strip the buffers from the page | |
1136 | * in preparation for freeing it (sometimes, rarely, buffers are removed from | |
1137 | * a page but it ends up not being freed, and buffers may later be reattached). | |
1138 | */ | |
1139 | void __brelse(struct buffer_head * buf) | |
1140 | { | |
1141 | if (atomic_read(&buf->b_count)) { | |
1142 | put_bh(buf); | |
1143 | return; | |
1144 | } | |
5c752ad9 | 1145 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
1da177e4 | 1146 | } |
1fe72eaa | 1147 | EXPORT_SYMBOL(__brelse); |
1da177e4 LT |
1148 | |
1149 | /* | |
1150 | * bforget() is like brelse(), except it discards any | |
1151 | * potentially dirty data. | |
1152 | */ | |
1153 | void __bforget(struct buffer_head *bh) | |
1154 | { | |
1155 | clear_buffer_dirty(bh); | |
535ee2fb | 1156 | if (bh->b_assoc_map) { |
1da177e4 LT |
1157 | struct address_space *buffer_mapping = bh->b_page->mapping; |
1158 | ||
1159 | spin_lock(&buffer_mapping->private_lock); | |
1160 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 1161 | bh->b_assoc_map = NULL; |
1da177e4 LT |
1162 | spin_unlock(&buffer_mapping->private_lock); |
1163 | } | |
1164 | __brelse(bh); | |
1165 | } | |
1fe72eaa | 1166 | EXPORT_SYMBOL(__bforget); |
1da177e4 LT |
1167 | |
1168 | static struct buffer_head *__bread_slow(struct buffer_head *bh) | |
1169 | { | |
1170 | lock_buffer(bh); | |
1171 | if (buffer_uptodate(bh)) { | |
1172 | unlock_buffer(bh); | |
1173 | return bh; | |
1174 | } else { | |
1175 | get_bh(bh); | |
1176 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 1177 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
1178 | wait_on_buffer(bh); |
1179 | if (buffer_uptodate(bh)) | |
1180 | return bh; | |
1181 | } | |
1182 | brelse(bh); | |
1183 | return NULL; | |
1184 | } | |
1185 | ||
1186 | /* | |
1187 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). | |
1188 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their | |
1189 | * refcount elevated by one when they're in an LRU. A buffer can only appear | |
1190 | * once in a particular CPU's LRU. A single buffer can be present in multiple | |
1191 | * CPU's LRUs at the same time. | |
1192 | * | |
1193 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and | |
1194 | * sb_find_get_block(). | |
1195 | * | |
1196 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use | |
1197 | * a local interrupt disable for that. | |
1198 | */ | |
1199 | ||
86cf78d7 | 1200 | #define BH_LRU_SIZE 16 |
1da177e4 LT |
1201 | |
1202 | struct bh_lru { | |
1203 | struct buffer_head *bhs[BH_LRU_SIZE]; | |
1204 | }; | |
1205 | ||
1206 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; | |
1207 | ||
1208 | #ifdef CONFIG_SMP | |
1209 | #define bh_lru_lock() local_irq_disable() | |
1210 | #define bh_lru_unlock() local_irq_enable() | |
1211 | #else | |
1212 | #define bh_lru_lock() preempt_disable() | |
1213 | #define bh_lru_unlock() preempt_enable() | |
1214 | #endif | |
1215 | ||
1216 | static inline void check_irqs_on(void) | |
1217 | { | |
1218 | #ifdef irqs_disabled | |
1219 | BUG_ON(irqs_disabled()); | |
1220 | #endif | |
1221 | } | |
1222 | ||
1223 | /* | |
241f01fb EB |
1224 | * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is |
1225 | * inserted at the front, and the buffer_head at the back if any is evicted. | |
1226 | * Or, if already in the LRU it is moved to the front. | |
1da177e4 LT |
1227 | */ |
1228 | static void bh_lru_install(struct buffer_head *bh) | |
1229 | { | |
241f01fb EB |
1230 | struct buffer_head *evictee = bh; |
1231 | struct bh_lru *b; | |
1232 | int i; | |
1da177e4 LT |
1233 | |
1234 | check_irqs_on(); | |
1235 | bh_lru_lock(); | |
1da177e4 | 1236 | |
241f01fb EB |
1237 | b = this_cpu_ptr(&bh_lrus); |
1238 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1239 | swap(evictee, b->bhs[i]); | |
1240 | if (evictee == bh) { | |
1241 | bh_lru_unlock(); | |
1242 | return; | |
1da177e4 | 1243 | } |
1da177e4 | 1244 | } |
1da177e4 | 1245 | |
241f01fb EB |
1246 | get_bh(bh); |
1247 | bh_lru_unlock(); | |
1248 | brelse(evictee); | |
1da177e4 LT |
1249 | } |
1250 | ||
1251 | /* | |
1252 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. | |
1253 | */ | |
858119e1 | 1254 | static struct buffer_head * |
3991d3bd | 1255 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1256 | { |
1257 | struct buffer_head *ret = NULL; | |
3991d3bd | 1258 | unsigned int i; |
1da177e4 LT |
1259 | |
1260 | check_irqs_on(); | |
1261 | bh_lru_lock(); | |
1da177e4 | 1262 | for (i = 0; i < BH_LRU_SIZE; i++) { |
c7b92516 | 1263 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
1da177e4 | 1264 | |
9470dd5d ZB |
1265 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
1266 | bh->b_size == size) { | |
1da177e4 LT |
1267 | if (i) { |
1268 | while (i) { | |
c7b92516 CL |
1269 | __this_cpu_write(bh_lrus.bhs[i], |
1270 | __this_cpu_read(bh_lrus.bhs[i - 1])); | |
1da177e4 LT |
1271 | i--; |
1272 | } | |
c7b92516 | 1273 | __this_cpu_write(bh_lrus.bhs[0], bh); |
1da177e4 LT |
1274 | } |
1275 | get_bh(bh); | |
1276 | ret = bh; | |
1277 | break; | |
1278 | } | |
1279 | } | |
1280 | bh_lru_unlock(); | |
1281 | return ret; | |
1282 | } | |
1283 | ||
1284 | /* | |
1285 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh | |
1286 | * it in the LRU and mark it as accessed. If it is not present then return | |
1287 | * NULL | |
1288 | */ | |
1289 | struct buffer_head * | |
3991d3bd | 1290 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1291 | { |
1292 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); | |
1293 | ||
1294 | if (bh == NULL) { | |
2457aec6 | 1295 | /* __find_get_block_slow will mark the page accessed */ |
385fd4c5 | 1296 | bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1297 | if (bh) |
1298 | bh_lru_install(bh); | |
2457aec6 | 1299 | } else |
1da177e4 | 1300 | touch_buffer(bh); |
2457aec6 | 1301 | |
1da177e4 LT |
1302 | return bh; |
1303 | } | |
1304 | EXPORT_SYMBOL(__find_get_block); | |
1305 | ||
1306 | /* | |
3b5e6454 | 1307 | * __getblk_gfp() will locate (and, if necessary, create) the buffer_head |
1da177e4 LT |
1308 | * which corresponds to the passed block_device, block and size. The |
1309 | * returned buffer has its reference count incremented. | |
1310 | * | |
3b5e6454 GK |
1311 | * __getblk_gfp() will lock up the machine if grow_dev_page's |
1312 | * try_to_free_buffers() attempt is failing. FIXME, perhaps? | |
1da177e4 LT |
1313 | */ |
1314 | struct buffer_head * | |
3b5e6454 GK |
1315 | __getblk_gfp(struct block_device *bdev, sector_t block, |
1316 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1317 | { |
1318 | struct buffer_head *bh = __find_get_block(bdev, block, size); | |
1319 | ||
1320 | might_sleep(); | |
1321 | if (bh == NULL) | |
3b5e6454 | 1322 | bh = __getblk_slow(bdev, block, size, gfp); |
1da177e4 LT |
1323 | return bh; |
1324 | } | |
3b5e6454 | 1325 | EXPORT_SYMBOL(__getblk_gfp); |
1da177e4 LT |
1326 | |
1327 | /* | |
1328 | * Do async read-ahead on a buffer.. | |
1329 | */ | |
3991d3bd | 1330 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1331 | { |
1332 | struct buffer_head *bh = __getblk(bdev, block, size); | |
a3e713b5 | 1333 | if (likely(bh)) { |
70246286 | 1334 | ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); |
a3e713b5 AM |
1335 | brelse(bh); |
1336 | } | |
1da177e4 LT |
1337 | } |
1338 | EXPORT_SYMBOL(__breadahead); | |
1339 | ||
dcb22991 RG |
1340 | void __breadahead_gfp(struct block_device *bdev, sector_t block, unsigned size, |
1341 | gfp_t gfp) | |
1342 | { | |
1343 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); | |
1344 | if (likely(bh)) { | |
1345 | ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); | |
1346 | brelse(bh); | |
1347 | } | |
1348 | } | |
1349 | EXPORT_SYMBOL(__breadahead_gfp); | |
1350 | ||
1da177e4 | 1351 | /** |
3b5e6454 | 1352 | * __bread_gfp() - reads a specified block and returns the bh |
67be2dd1 | 1353 | * @bdev: the block_device to read from |
1da177e4 LT |
1354 | * @block: number of block |
1355 | * @size: size (in bytes) to read | |
3b5e6454 GK |
1356 | * @gfp: page allocation flag |
1357 | * | |
1da177e4 | 1358 | * Reads a specified block, and returns buffer head that contains it. |
3b5e6454 GK |
1359 | * The page cache can be allocated from non-movable area |
1360 | * not to prevent page migration if you set gfp to zero. | |
1da177e4 LT |
1361 | * It returns NULL if the block was unreadable. |
1362 | */ | |
1363 | struct buffer_head * | |
3b5e6454 GK |
1364 | __bread_gfp(struct block_device *bdev, sector_t block, |
1365 | unsigned size, gfp_t gfp) | |
1da177e4 | 1366 | { |
3b5e6454 | 1367 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); |
1da177e4 | 1368 | |
a3e713b5 | 1369 | if (likely(bh) && !buffer_uptodate(bh)) |
1da177e4 LT |
1370 | bh = __bread_slow(bh); |
1371 | return bh; | |
1372 | } | |
3b5e6454 | 1373 | EXPORT_SYMBOL(__bread_gfp); |
1da177e4 LT |
1374 | |
1375 | /* | |
1376 | * invalidate_bh_lrus() is called rarely - but not only at unmount. | |
1377 | * This doesn't race because it runs in each cpu either in irq | |
1378 | * or with preempt disabled. | |
1379 | */ | |
1380 | static void invalidate_bh_lru(void *arg) | |
1381 | { | |
1382 | struct bh_lru *b = &get_cpu_var(bh_lrus); | |
1383 | int i; | |
1384 | ||
1385 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1386 | brelse(b->bhs[i]); | |
1387 | b->bhs[i] = NULL; | |
1388 | } | |
1389 | put_cpu_var(bh_lrus); | |
1390 | } | |
42be35d0 GBY |
1391 | |
1392 | static bool has_bh_in_lru(int cpu, void *dummy) | |
1393 | { | |
1394 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); | |
1395 | int i; | |
1da177e4 | 1396 | |
42be35d0 GBY |
1397 | for (i = 0; i < BH_LRU_SIZE; i++) { |
1398 | if (b->bhs[i]) | |
1399 | return 1; | |
1400 | } | |
1401 | ||
1402 | return 0; | |
1403 | } | |
1404 | ||
f9a14399 | 1405 | void invalidate_bh_lrus(void) |
1da177e4 | 1406 | { |
42be35d0 | 1407 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); |
1da177e4 | 1408 | } |
9db5579b | 1409 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
1da177e4 LT |
1410 | |
1411 | void set_bh_page(struct buffer_head *bh, | |
1412 | struct page *page, unsigned long offset) | |
1413 | { | |
1414 | bh->b_page = page; | |
e827f923 | 1415 | BUG_ON(offset >= PAGE_SIZE); |
1da177e4 LT |
1416 | if (PageHighMem(page)) |
1417 | /* | |
1418 | * This catches illegal uses and preserves the offset: | |
1419 | */ | |
1420 | bh->b_data = (char *)(0 + offset); | |
1421 | else | |
1422 | bh->b_data = page_address(page) + offset; | |
1423 | } | |
1424 | EXPORT_SYMBOL(set_bh_page); | |
1425 | ||
1426 | /* | |
1427 | * Called when truncating a buffer on a page completely. | |
1428 | */ | |
e7470ee8 MG |
1429 | |
1430 | /* Bits that are cleared during an invalidate */ | |
1431 | #define BUFFER_FLAGS_DISCARD \ | |
1432 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ | |
1433 | 1 << BH_Delay | 1 << BH_Unwritten) | |
1434 | ||
858119e1 | 1435 | static void discard_buffer(struct buffer_head * bh) |
1da177e4 | 1436 | { |
e7470ee8 MG |
1437 | unsigned long b_state, b_state_old; |
1438 | ||
1da177e4 LT |
1439 | lock_buffer(bh); |
1440 | clear_buffer_dirty(bh); | |
1441 | bh->b_bdev = NULL; | |
e7470ee8 MG |
1442 | b_state = bh->b_state; |
1443 | for (;;) { | |
1444 | b_state_old = cmpxchg(&bh->b_state, b_state, | |
1445 | (b_state & ~BUFFER_FLAGS_DISCARD)); | |
1446 | if (b_state_old == b_state) | |
1447 | break; | |
1448 | b_state = b_state_old; | |
1449 | } | |
1da177e4 LT |
1450 | unlock_buffer(bh); |
1451 | } | |
1452 | ||
1da177e4 | 1453 | /** |
814e1d25 | 1454 | * block_invalidatepage - invalidate part or all of a buffer-backed page |
1da177e4 LT |
1455 | * |
1456 | * @page: the page which is affected | |
d47992f8 LC |
1457 | * @offset: start of the range to invalidate |
1458 | * @length: length of the range to invalidate | |
1da177e4 LT |
1459 | * |
1460 | * block_invalidatepage() is called when all or part of the page has become | |
814e1d25 | 1461 | * invalidated by a truncate operation. |
1da177e4 LT |
1462 | * |
1463 | * block_invalidatepage() does not have to release all buffers, but it must | |
1464 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
1465 | * is underway against any of the blocks which are outside the truncation | |
1466 | * point. Because the caller is about to free (and possibly reuse) those | |
1467 | * blocks on-disk. | |
1468 | */ | |
d47992f8 LC |
1469 | void block_invalidatepage(struct page *page, unsigned int offset, |
1470 | unsigned int length) | |
1da177e4 LT |
1471 | { |
1472 | struct buffer_head *head, *bh, *next; | |
1473 | unsigned int curr_off = 0; | |
d47992f8 | 1474 | unsigned int stop = length + offset; |
1da177e4 LT |
1475 | |
1476 | BUG_ON(!PageLocked(page)); | |
1477 | if (!page_has_buffers(page)) | |
1478 | goto out; | |
1479 | ||
d47992f8 LC |
1480 | /* |
1481 | * Check for overflow | |
1482 | */ | |
09cbfeaf | 1483 | BUG_ON(stop > PAGE_SIZE || stop < length); |
d47992f8 | 1484 | |
1da177e4 LT |
1485 | head = page_buffers(page); |
1486 | bh = head; | |
1487 | do { | |
1488 | unsigned int next_off = curr_off + bh->b_size; | |
1489 | next = bh->b_this_page; | |
1490 | ||
d47992f8 LC |
1491 | /* |
1492 | * Are we still fully in range ? | |
1493 | */ | |
1494 | if (next_off > stop) | |
1495 | goto out; | |
1496 | ||
1da177e4 LT |
1497 | /* |
1498 | * is this block fully invalidated? | |
1499 | */ | |
1500 | if (offset <= curr_off) | |
1501 | discard_buffer(bh); | |
1502 | curr_off = next_off; | |
1503 | bh = next; | |
1504 | } while (bh != head); | |
1505 | ||
1506 | /* | |
1507 | * We release buffers only if the entire page is being invalidated. | |
1508 | * The get_block cached value has been unconditionally invalidated, | |
1509 | * so real IO is not possible anymore. | |
1510 | */ | |
3172485f | 1511 | if (length == PAGE_SIZE) |
2ff28e22 | 1512 | try_to_release_page(page, 0); |
1da177e4 | 1513 | out: |
2ff28e22 | 1514 | return; |
1da177e4 LT |
1515 | } |
1516 | EXPORT_SYMBOL(block_invalidatepage); | |
1517 | ||
d47992f8 | 1518 | |
1da177e4 LT |
1519 | /* |
1520 | * We attach and possibly dirty the buffers atomically wrt | |
1521 | * __set_page_dirty_buffers() via private_lock. try_to_free_buffers | |
1522 | * is already excluded via the page lock. | |
1523 | */ | |
1524 | void create_empty_buffers(struct page *page, | |
1525 | unsigned long blocksize, unsigned long b_state) | |
1526 | { | |
1527 | struct buffer_head *bh, *head, *tail; | |
1528 | ||
640ab98f | 1529 | head = alloc_page_buffers(page, blocksize, true); |
1da177e4 LT |
1530 | bh = head; |
1531 | do { | |
1532 | bh->b_state |= b_state; | |
1533 | tail = bh; | |
1534 | bh = bh->b_this_page; | |
1535 | } while (bh); | |
1536 | tail->b_this_page = head; | |
1537 | ||
1538 | spin_lock(&page->mapping->private_lock); | |
1539 | if (PageUptodate(page) || PageDirty(page)) { | |
1540 | bh = head; | |
1541 | do { | |
1542 | if (PageDirty(page)) | |
1543 | set_buffer_dirty(bh); | |
1544 | if (PageUptodate(page)) | |
1545 | set_buffer_uptodate(bh); | |
1546 | bh = bh->b_this_page; | |
1547 | } while (bh != head); | |
1548 | } | |
1549 | attach_page_buffers(page, head); | |
1550 | spin_unlock(&page->mapping->private_lock); | |
1551 | } | |
1552 | EXPORT_SYMBOL(create_empty_buffers); | |
1553 | ||
29f3ad7d JK |
1554 | /** |
1555 | * clean_bdev_aliases: clean a range of buffers in block device | |
1556 | * @bdev: Block device to clean buffers in | |
1557 | * @block: Start of a range of blocks to clean | |
1558 | * @len: Number of blocks to clean | |
1da177e4 | 1559 | * |
29f3ad7d JK |
1560 | * We are taking a range of blocks for data and we don't want writeback of any |
1561 | * buffer-cache aliases starting from return from this function and until the | |
1562 | * moment when something will explicitly mark the buffer dirty (hopefully that | |
1563 | * will not happen until we will free that block ;-) We don't even need to mark | |
1564 | * it not-uptodate - nobody can expect anything from a newly allocated buffer | |
1565 | * anyway. We used to use unmap_buffer() for such invalidation, but that was | |
1566 | * wrong. We definitely don't want to mark the alias unmapped, for example - it | |
1567 | * would confuse anyone who might pick it with bread() afterwards... | |
1568 | * | |
1569 | * Also.. Note that bforget() doesn't lock the buffer. So there can be | |
1570 | * writeout I/O going on against recently-freed buffers. We don't wait on that | |
1571 | * I/O in bforget() - it's more efficient to wait on the I/O only if we really | |
1572 | * need to. That happens here. | |
1da177e4 | 1573 | */ |
29f3ad7d | 1574 | void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) |
1da177e4 | 1575 | { |
29f3ad7d JK |
1576 | struct inode *bd_inode = bdev->bd_inode; |
1577 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
1578 | struct pagevec pvec; | |
1579 | pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); | |
1580 | pgoff_t end; | |
c10f778d | 1581 | int i, count; |
29f3ad7d JK |
1582 | struct buffer_head *bh; |
1583 | struct buffer_head *head; | |
1da177e4 | 1584 | |
29f3ad7d | 1585 | end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); |
86679820 | 1586 | pagevec_init(&pvec); |
397162ff | 1587 | while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) { |
c10f778d JK |
1588 | count = pagevec_count(&pvec); |
1589 | for (i = 0; i < count; i++) { | |
29f3ad7d | 1590 | struct page *page = pvec.pages[i]; |
1da177e4 | 1591 | |
29f3ad7d JK |
1592 | if (!page_has_buffers(page)) |
1593 | continue; | |
1594 | /* | |
1595 | * We use page lock instead of bd_mapping->private_lock | |
1596 | * to pin buffers here since we can afford to sleep and | |
1597 | * it scales better than a global spinlock lock. | |
1598 | */ | |
1599 | lock_page(page); | |
1600 | /* Recheck when the page is locked which pins bhs */ | |
1601 | if (!page_has_buffers(page)) | |
1602 | goto unlock_page; | |
1603 | head = page_buffers(page); | |
1604 | bh = head; | |
1605 | do { | |
6c006a9d | 1606 | if (!buffer_mapped(bh) || (bh->b_blocknr < block)) |
29f3ad7d JK |
1607 | goto next; |
1608 | if (bh->b_blocknr >= block + len) | |
1609 | break; | |
1610 | clear_buffer_dirty(bh); | |
1611 | wait_on_buffer(bh); | |
1612 | clear_buffer_req(bh); | |
1613 | next: | |
1614 | bh = bh->b_this_page; | |
1615 | } while (bh != head); | |
1616 | unlock_page: | |
1617 | unlock_page(page); | |
1618 | } | |
1619 | pagevec_release(&pvec); | |
1620 | cond_resched(); | |
c10f778d JK |
1621 | /* End of range already reached? */ |
1622 | if (index > end || !index) | |
1623 | break; | |
1da177e4 LT |
1624 | } |
1625 | } | |
29f3ad7d | 1626 | EXPORT_SYMBOL(clean_bdev_aliases); |
1da177e4 | 1627 | |
45bce8f3 LT |
1628 | /* |
1629 | * Size is a power-of-two in the range 512..PAGE_SIZE, | |
1630 | * and the case we care about most is PAGE_SIZE. | |
1631 | * | |
1632 | * So this *could* possibly be written with those | |
1633 | * constraints in mind (relevant mostly if some | |
1634 | * architecture has a slow bit-scan instruction) | |
1635 | */ | |
1636 | static inline int block_size_bits(unsigned int blocksize) | |
1637 | { | |
1638 | return ilog2(blocksize); | |
1639 | } | |
1640 | ||
1641 | static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) | |
1642 | { | |
1643 | BUG_ON(!PageLocked(page)); | |
1644 | ||
1645 | if (!page_has_buffers(page)) | |
6aa7de05 MR |
1646 | create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits), |
1647 | b_state); | |
45bce8f3 LT |
1648 | return page_buffers(page); |
1649 | } | |
1650 | ||
1da177e4 LT |
1651 | /* |
1652 | * NOTE! All mapped/uptodate combinations are valid: | |
1653 | * | |
1654 | * Mapped Uptodate Meaning | |
1655 | * | |
1656 | * No No "unknown" - must do get_block() | |
1657 | * No Yes "hole" - zero-filled | |
1658 | * Yes No "allocated" - allocated on disk, not read in | |
1659 | * Yes Yes "valid" - allocated and up-to-date in memory. | |
1660 | * | |
1661 | * "Dirty" is valid only with the last case (mapped+uptodate). | |
1662 | */ | |
1663 | ||
1664 | /* | |
1665 | * While block_write_full_page is writing back the dirty buffers under | |
1666 | * the page lock, whoever dirtied the buffers may decide to clean them | |
1667 | * again at any time. We handle that by only looking at the buffer | |
1668 | * state inside lock_buffer(). | |
1669 | * | |
1670 | * If block_write_full_page() is called for regular writeback | |
1671 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a | |
1672 | * locked buffer. This only can happen if someone has written the buffer | |
1673 | * directly, with submit_bh(). At the address_space level PageWriteback | |
1674 | * prevents this contention from occurring. | |
6e34eedd TT |
1675 | * |
1676 | * If block_write_full_page() is called with wbc->sync_mode == | |
70fd7614 | 1677 | * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this |
721a9602 | 1678 | * causes the writes to be flagged as synchronous writes. |
1da177e4 | 1679 | */ |
b4bba389 | 1680 | int __block_write_full_page(struct inode *inode, struct page *page, |
35c80d5f CM |
1681 | get_block_t *get_block, struct writeback_control *wbc, |
1682 | bh_end_io_t *handler) | |
1da177e4 LT |
1683 | { |
1684 | int err; | |
1685 | sector_t block; | |
1686 | sector_t last_block; | |
f0fbd5fc | 1687 | struct buffer_head *bh, *head; |
45bce8f3 | 1688 | unsigned int blocksize, bbits; |
1da177e4 | 1689 | int nr_underway = 0; |
7637241e | 1690 | int write_flags = wbc_to_write_flags(wbc); |
1da177e4 | 1691 | |
45bce8f3 | 1692 | head = create_page_buffers(page, inode, |
1da177e4 | 1693 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
1da177e4 LT |
1694 | |
1695 | /* | |
1696 | * Be very careful. We have no exclusion from __set_page_dirty_buffers | |
1697 | * here, and the (potentially unmapped) buffers may become dirty at | |
1698 | * any time. If a buffer becomes dirty here after we've inspected it | |
1699 | * then we just miss that fact, and the page stays dirty. | |
1700 | * | |
1701 | * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; | |
1702 | * handle that here by just cleaning them. | |
1703 | */ | |
1704 | ||
1da177e4 | 1705 | bh = head; |
45bce8f3 LT |
1706 | blocksize = bh->b_size; |
1707 | bbits = block_size_bits(blocksize); | |
1708 | ||
09cbfeaf | 1709 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 1710 | last_block = (i_size_read(inode) - 1) >> bbits; |
1da177e4 LT |
1711 | |
1712 | /* | |
1713 | * Get all the dirty buffers mapped to disk addresses and | |
1714 | * handle any aliases from the underlying blockdev's mapping. | |
1715 | */ | |
1716 | do { | |
1717 | if (block > last_block) { | |
1718 | /* | |
1719 | * mapped buffers outside i_size will occur, because | |
1720 | * this page can be outside i_size when there is a | |
1721 | * truncate in progress. | |
1722 | */ | |
1723 | /* | |
1724 | * The buffer was zeroed by block_write_full_page() | |
1725 | */ | |
1726 | clear_buffer_dirty(bh); | |
1727 | set_buffer_uptodate(bh); | |
29a814d2 AT |
1728 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
1729 | buffer_dirty(bh)) { | |
b0cf2321 | 1730 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1731 | err = get_block(inode, block, bh, 1); |
1732 | if (err) | |
1733 | goto recover; | |
29a814d2 | 1734 | clear_buffer_delay(bh); |
1da177e4 LT |
1735 | if (buffer_new(bh)) { |
1736 | /* blockdev mappings never come here */ | |
1737 | clear_buffer_new(bh); | |
e64855c6 | 1738 | clean_bdev_bh_alias(bh); |
1da177e4 LT |
1739 | } |
1740 | } | |
1741 | bh = bh->b_this_page; | |
1742 | block++; | |
1743 | } while (bh != head); | |
1744 | ||
1745 | do { | |
1da177e4 LT |
1746 | if (!buffer_mapped(bh)) |
1747 | continue; | |
1748 | /* | |
1749 | * If it's a fully non-blocking write attempt and we cannot | |
1750 | * lock the buffer then redirty the page. Note that this can | |
5b0830cb JA |
1751 | * potentially cause a busy-wait loop from writeback threads |
1752 | * and kswapd activity, but those code paths have their own | |
1753 | * higher-level throttling. | |
1da177e4 | 1754 | */ |
1b430bee | 1755 | if (wbc->sync_mode != WB_SYNC_NONE) { |
1da177e4 | 1756 | lock_buffer(bh); |
ca5de404 | 1757 | } else if (!trylock_buffer(bh)) { |
1da177e4 LT |
1758 | redirty_page_for_writepage(wbc, page); |
1759 | continue; | |
1760 | } | |
1761 | if (test_clear_buffer_dirty(bh)) { | |
35c80d5f | 1762 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1763 | } else { |
1764 | unlock_buffer(bh); | |
1765 | } | |
1766 | } while ((bh = bh->b_this_page) != head); | |
1767 | ||
1768 | /* | |
1769 | * The page and its buffers are protected by PageWriteback(), so we can | |
1770 | * drop the bh refcounts early. | |
1771 | */ | |
1772 | BUG_ON(PageWriteback(page)); | |
1773 | set_page_writeback(page); | |
1da177e4 LT |
1774 | |
1775 | do { | |
1776 | struct buffer_head *next = bh->b_this_page; | |
1777 | if (buffer_async_write(bh)) { | |
8e8f9298 JA |
1778 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1779 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1780 | nr_underway++; |
1781 | } | |
1da177e4 LT |
1782 | bh = next; |
1783 | } while (bh != head); | |
05937baa | 1784 | unlock_page(page); |
1da177e4 LT |
1785 | |
1786 | err = 0; | |
1787 | done: | |
1788 | if (nr_underway == 0) { | |
1789 | /* | |
1790 | * The page was marked dirty, but the buffers were | |
1791 | * clean. Someone wrote them back by hand with | |
1792 | * ll_rw_block/submit_bh. A rare case. | |
1793 | */ | |
1da177e4 | 1794 | end_page_writeback(page); |
3d67f2d7 | 1795 | |
1da177e4 LT |
1796 | /* |
1797 | * The page and buffer_heads can be released at any time from | |
1798 | * here on. | |
1799 | */ | |
1da177e4 LT |
1800 | } |
1801 | return err; | |
1802 | ||
1803 | recover: | |
1804 | /* | |
1805 | * ENOSPC, or some other error. We may already have added some | |
1806 | * blocks to the file, so we need to write these out to avoid | |
1807 | * exposing stale data. | |
1808 | * The page is currently locked and not marked for writeback | |
1809 | */ | |
1810 | bh = head; | |
1811 | /* Recovery: lock and submit the mapped buffers */ | |
1812 | do { | |
29a814d2 AT |
1813 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
1814 | !buffer_delay(bh)) { | |
1da177e4 | 1815 | lock_buffer(bh); |
35c80d5f | 1816 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1817 | } else { |
1818 | /* | |
1819 | * The buffer may have been set dirty during | |
1820 | * attachment to a dirty page. | |
1821 | */ | |
1822 | clear_buffer_dirty(bh); | |
1823 | } | |
1824 | } while ((bh = bh->b_this_page) != head); | |
1825 | SetPageError(page); | |
1826 | BUG_ON(PageWriteback(page)); | |
7e4c3690 | 1827 | mapping_set_error(page->mapping, err); |
1da177e4 | 1828 | set_page_writeback(page); |
1da177e4 LT |
1829 | do { |
1830 | struct buffer_head *next = bh->b_this_page; | |
1831 | if (buffer_async_write(bh)) { | |
1832 | clear_buffer_dirty(bh); | |
8e8f9298 JA |
1833 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1834 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1835 | nr_underway++; |
1836 | } | |
1da177e4 LT |
1837 | bh = next; |
1838 | } while (bh != head); | |
ffda9d30 | 1839 | unlock_page(page); |
1da177e4 LT |
1840 | goto done; |
1841 | } | |
b4bba389 | 1842 | EXPORT_SYMBOL(__block_write_full_page); |
1da177e4 | 1843 | |
afddba49 NP |
1844 | /* |
1845 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
1846 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1847 | * block data from leaking). And clear the new bit. | |
1848 | */ | |
1849 | void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1850 | { | |
1851 | unsigned int block_start, block_end; | |
1852 | struct buffer_head *head, *bh; | |
1853 | ||
1854 | BUG_ON(!PageLocked(page)); | |
1855 | if (!page_has_buffers(page)) | |
1856 | return; | |
1857 | ||
1858 | bh = head = page_buffers(page); | |
1859 | block_start = 0; | |
1860 | do { | |
1861 | block_end = block_start + bh->b_size; | |
1862 | ||
1863 | if (buffer_new(bh)) { | |
1864 | if (block_end > from && block_start < to) { | |
1865 | if (!PageUptodate(page)) { | |
1866 | unsigned start, size; | |
1867 | ||
1868 | start = max(from, block_start); | |
1869 | size = min(to, block_end) - start; | |
1870 | ||
eebd2aa3 | 1871 | zero_user(page, start, size); |
afddba49 NP |
1872 | set_buffer_uptodate(bh); |
1873 | } | |
1874 | ||
1875 | clear_buffer_new(bh); | |
1876 | mark_buffer_dirty(bh); | |
1877 | } | |
1878 | } | |
1879 | ||
1880 | block_start = block_end; | |
1881 | bh = bh->b_this_page; | |
1882 | } while (bh != head); | |
1883 | } | |
1884 | EXPORT_SYMBOL(page_zero_new_buffers); | |
1885 | ||
ae259a9c CH |
1886 | static void |
1887 | iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, | |
1888 | struct iomap *iomap) | |
1889 | { | |
1890 | loff_t offset = block << inode->i_blkbits; | |
1891 | ||
1892 | bh->b_bdev = iomap->bdev; | |
1893 | ||
1894 | /* | |
1895 | * Block points to offset in file we need to map, iomap contains | |
1896 | * the offset at which the map starts. If the map ends before the | |
1897 | * current block, then do not map the buffer and let the caller | |
1898 | * handle it. | |
1899 | */ | |
1900 | BUG_ON(offset >= iomap->offset + iomap->length); | |
1901 | ||
1902 | switch (iomap->type) { | |
1903 | case IOMAP_HOLE: | |
1904 | /* | |
1905 | * If the buffer is not up to date or beyond the current EOF, | |
1906 | * we need to mark it as new to ensure sub-block zeroing is | |
1907 | * executed if necessary. | |
1908 | */ | |
1909 | if (!buffer_uptodate(bh) || | |
1910 | (offset >= i_size_read(inode))) | |
1911 | set_buffer_new(bh); | |
1912 | break; | |
1913 | case IOMAP_DELALLOC: | |
1914 | if (!buffer_uptodate(bh) || | |
1915 | (offset >= i_size_read(inode))) | |
1916 | set_buffer_new(bh); | |
1917 | set_buffer_uptodate(bh); | |
1918 | set_buffer_mapped(bh); | |
1919 | set_buffer_delay(bh); | |
1920 | break; | |
1921 | case IOMAP_UNWRITTEN: | |
1922 | /* | |
3d7b6b21 AG |
1923 | * For unwritten regions, we always need to ensure that regions |
1924 | * in the block we are not writing to are zeroed. Mark the | |
1925 | * buffer as new to ensure this. | |
ae259a9c CH |
1926 | */ |
1927 | set_buffer_new(bh); | |
1928 | set_buffer_unwritten(bh); | |
1929 | /* FALLTHRU */ | |
1930 | case IOMAP_MAPPED: | |
3d7b6b21 AG |
1931 | if ((iomap->flags & IOMAP_F_NEW) || |
1932 | offset >= i_size_read(inode)) | |
ae259a9c | 1933 | set_buffer_new(bh); |
19fe5f64 AG |
1934 | bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> |
1935 | inode->i_blkbits; | |
ae259a9c CH |
1936 | set_buffer_mapped(bh); |
1937 | break; | |
1938 | } | |
1939 | } | |
1940 | ||
1941 | int __block_write_begin_int(struct page *page, loff_t pos, unsigned len, | |
1942 | get_block_t *get_block, struct iomap *iomap) | |
1da177e4 | 1943 | { |
09cbfeaf | 1944 | unsigned from = pos & (PAGE_SIZE - 1); |
ebdec241 | 1945 | unsigned to = from + len; |
6e1db88d | 1946 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
1947 | unsigned block_start, block_end; |
1948 | sector_t block; | |
1949 | int err = 0; | |
1950 | unsigned blocksize, bbits; | |
1951 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; | |
1952 | ||
1953 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1954 | BUG_ON(from > PAGE_SIZE); |
1955 | BUG_ON(to > PAGE_SIZE); | |
1da177e4 LT |
1956 | BUG_ON(from > to); |
1957 | ||
45bce8f3 LT |
1958 | head = create_page_buffers(page, inode, 0); |
1959 | blocksize = head->b_size; | |
1960 | bbits = block_size_bits(blocksize); | |
1da177e4 | 1961 | |
09cbfeaf | 1962 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
1da177e4 LT |
1963 | |
1964 | for(bh = head, block_start = 0; bh != head || !block_start; | |
1965 | block++, block_start=block_end, bh = bh->b_this_page) { | |
1966 | block_end = block_start + blocksize; | |
1967 | if (block_end <= from || block_start >= to) { | |
1968 | if (PageUptodate(page)) { | |
1969 | if (!buffer_uptodate(bh)) | |
1970 | set_buffer_uptodate(bh); | |
1971 | } | |
1972 | continue; | |
1973 | } | |
1974 | if (buffer_new(bh)) | |
1975 | clear_buffer_new(bh); | |
1976 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 1977 | WARN_ON(bh->b_size != blocksize); |
ae259a9c CH |
1978 | if (get_block) { |
1979 | err = get_block(inode, block, bh, 1); | |
1980 | if (err) | |
1981 | break; | |
1982 | } else { | |
1983 | iomap_to_bh(inode, block, bh, iomap); | |
1984 | } | |
1985 | ||
1da177e4 | 1986 | if (buffer_new(bh)) { |
e64855c6 | 1987 | clean_bdev_bh_alias(bh); |
1da177e4 | 1988 | if (PageUptodate(page)) { |
637aff46 | 1989 | clear_buffer_new(bh); |
1da177e4 | 1990 | set_buffer_uptodate(bh); |
637aff46 | 1991 | mark_buffer_dirty(bh); |
1da177e4 LT |
1992 | continue; |
1993 | } | |
eebd2aa3 CL |
1994 | if (block_end > to || block_start < from) |
1995 | zero_user_segments(page, | |
1996 | to, block_end, | |
1997 | block_start, from); | |
1da177e4 LT |
1998 | continue; |
1999 | } | |
2000 | } | |
2001 | if (PageUptodate(page)) { | |
2002 | if (!buffer_uptodate(bh)) | |
2003 | set_buffer_uptodate(bh); | |
2004 | continue; | |
2005 | } | |
2006 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
33a266dd | 2007 | !buffer_unwritten(bh) && |
1da177e4 | 2008 | (block_start < from || block_end > to)) { |
dfec8a14 | 2009 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2010 | *wait_bh++=bh; |
2011 | } | |
2012 | } | |
2013 | /* | |
2014 | * If we issued read requests - let them complete. | |
2015 | */ | |
2016 | while(wait_bh > wait) { | |
2017 | wait_on_buffer(*--wait_bh); | |
2018 | if (!buffer_uptodate(*wait_bh)) | |
f3ddbdc6 | 2019 | err = -EIO; |
1da177e4 | 2020 | } |
f9f07b6c | 2021 | if (unlikely(err)) |
afddba49 | 2022 | page_zero_new_buffers(page, from, to); |
1da177e4 LT |
2023 | return err; |
2024 | } | |
ae259a9c CH |
2025 | |
2026 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, | |
2027 | get_block_t *get_block) | |
2028 | { | |
2029 | return __block_write_begin_int(page, pos, len, get_block, NULL); | |
2030 | } | |
ebdec241 | 2031 | EXPORT_SYMBOL(__block_write_begin); |
1da177e4 LT |
2032 | |
2033 | static int __block_commit_write(struct inode *inode, struct page *page, | |
2034 | unsigned from, unsigned to) | |
2035 | { | |
2036 | unsigned block_start, block_end; | |
2037 | int partial = 0; | |
2038 | unsigned blocksize; | |
2039 | struct buffer_head *bh, *head; | |
2040 | ||
45bce8f3 LT |
2041 | bh = head = page_buffers(page); |
2042 | blocksize = bh->b_size; | |
1da177e4 | 2043 | |
45bce8f3 LT |
2044 | block_start = 0; |
2045 | do { | |
1da177e4 LT |
2046 | block_end = block_start + blocksize; |
2047 | if (block_end <= from || block_start >= to) { | |
2048 | if (!buffer_uptodate(bh)) | |
2049 | partial = 1; | |
2050 | } else { | |
2051 | set_buffer_uptodate(bh); | |
2052 | mark_buffer_dirty(bh); | |
2053 | } | |
afddba49 | 2054 | clear_buffer_new(bh); |
45bce8f3 LT |
2055 | |
2056 | block_start = block_end; | |
2057 | bh = bh->b_this_page; | |
2058 | } while (bh != head); | |
1da177e4 LT |
2059 | |
2060 | /* | |
2061 | * If this is a partial write which happened to make all buffers | |
2062 | * uptodate then we can optimize away a bogus readpage() for | |
2063 | * the next read(). Here we 'discover' whether the page went | |
2064 | * uptodate as a result of this (potentially partial) write. | |
2065 | */ | |
2066 | if (!partial) | |
2067 | SetPageUptodate(page); | |
2068 | return 0; | |
2069 | } | |
2070 | ||
afddba49 | 2071 | /* |
155130a4 CH |
2072 | * block_write_begin takes care of the basic task of block allocation and |
2073 | * bringing partial write blocks uptodate first. | |
2074 | * | |
7bb46a67 | 2075 | * The filesystem needs to handle block truncation upon failure. |
afddba49 | 2076 | */ |
155130a4 CH |
2077 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
2078 | unsigned flags, struct page **pagep, get_block_t *get_block) | |
afddba49 | 2079 | { |
09cbfeaf | 2080 | pgoff_t index = pos >> PAGE_SHIFT; |
afddba49 | 2081 | struct page *page; |
6e1db88d | 2082 | int status; |
afddba49 | 2083 | |
6e1db88d CH |
2084 | page = grab_cache_page_write_begin(mapping, index, flags); |
2085 | if (!page) | |
2086 | return -ENOMEM; | |
afddba49 | 2087 | |
6e1db88d | 2088 | status = __block_write_begin(page, pos, len, get_block); |
afddba49 | 2089 | if (unlikely(status)) { |
6e1db88d | 2090 | unlock_page(page); |
09cbfeaf | 2091 | put_page(page); |
6e1db88d | 2092 | page = NULL; |
afddba49 NP |
2093 | } |
2094 | ||
6e1db88d | 2095 | *pagep = page; |
afddba49 NP |
2096 | return status; |
2097 | } | |
2098 | EXPORT_SYMBOL(block_write_begin); | |
2099 | ||
2100 | int block_write_end(struct file *file, struct address_space *mapping, | |
2101 | loff_t pos, unsigned len, unsigned copied, | |
2102 | struct page *page, void *fsdata) | |
2103 | { | |
2104 | struct inode *inode = mapping->host; | |
2105 | unsigned start; | |
2106 | ||
09cbfeaf | 2107 | start = pos & (PAGE_SIZE - 1); |
afddba49 NP |
2108 | |
2109 | if (unlikely(copied < len)) { | |
2110 | /* | |
2111 | * The buffers that were written will now be uptodate, so we | |
2112 | * don't have to worry about a readpage reading them and | |
2113 | * overwriting a partial write. However if we have encountered | |
2114 | * a short write and only partially written into a buffer, it | |
2115 | * will not be marked uptodate, so a readpage might come in and | |
2116 | * destroy our partial write. | |
2117 | * | |
2118 | * Do the simplest thing, and just treat any short write to a | |
2119 | * non uptodate page as a zero-length write, and force the | |
2120 | * caller to redo the whole thing. | |
2121 | */ | |
2122 | if (!PageUptodate(page)) | |
2123 | copied = 0; | |
2124 | ||
2125 | page_zero_new_buffers(page, start+copied, start+len); | |
2126 | } | |
2127 | flush_dcache_page(page); | |
2128 | ||
2129 | /* This could be a short (even 0-length) commit */ | |
2130 | __block_commit_write(inode, page, start, start+copied); | |
2131 | ||
2132 | return copied; | |
2133 | } | |
2134 | EXPORT_SYMBOL(block_write_end); | |
2135 | ||
2136 | int generic_write_end(struct file *file, struct address_space *mapping, | |
2137 | loff_t pos, unsigned len, unsigned copied, | |
2138 | struct page *page, void *fsdata) | |
2139 | { | |
8af54f29 CH |
2140 | struct inode *inode = mapping->host; |
2141 | loff_t old_size = inode->i_size; | |
2142 | bool i_size_changed = false; | |
2143 | ||
afddba49 | 2144 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
8af54f29 CH |
2145 | |
2146 | /* | |
2147 | * No need to use i_size_read() here, the i_size cannot change under us | |
2148 | * because we hold i_rwsem. | |
2149 | * | |
2150 | * But it's important to update i_size while still holding page lock: | |
2151 | * page writeout could otherwise come in and zero beyond i_size. | |
2152 | */ | |
2153 | if (pos + copied > inode->i_size) { | |
2154 | i_size_write(inode, pos + copied); | |
2155 | i_size_changed = true; | |
2156 | } | |
2157 | ||
2158 | unlock_page(page); | |
7a77dad7 | 2159 | put_page(page); |
8af54f29 CH |
2160 | |
2161 | if (old_size < pos) | |
2162 | pagecache_isize_extended(inode, old_size, pos); | |
2163 | /* | |
2164 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
2165 | * makes the holding time of page lock longer. Second, it forces lock | |
2166 | * ordering of page lock and transaction start for journaling | |
2167 | * filesystems. | |
2168 | */ | |
2169 | if (i_size_changed) | |
2170 | mark_inode_dirty(inode); | |
26ddb1f4 | 2171 | return copied; |
afddba49 NP |
2172 | } |
2173 | EXPORT_SYMBOL(generic_write_end); | |
2174 | ||
8ab22b9a HH |
2175 | /* |
2176 | * block_is_partially_uptodate checks whether buffers within a page are | |
2177 | * uptodate or not. | |
2178 | * | |
2179 | * Returns true if all buffers which correspond to a file portion | |
2180 | * we want to read are uptodate. | |
2181 | */ | |
c186afb4 AV |
2182 | int block_is_partially_uptodate(struct page *page, unsigned long from, |
2183 | unsigned long count) | |
8ab22b9a | 2184 | { |
8ab22b9a HH |
2185 | unsigned block_start, block_end, blocksize; |
2186 | unsigned to; | |
2187 | struct buffer_head *bh, *head; | |
2188 | int ret = 1; | |
2189 | ||
2190 | if (!page_has_buffers(page)) | |
2191 | return 0; | |
2192 | ||
45bce8f3 LT |
2193 | head = page_buffers(page); |
2194 | blocksize = head->b_size; | |
09cbfeaf | 2195 | to = min_t(unsigned, PAGE_SIZE - from, count); |
8ab22b9a | 2196 | to = from + to; |
09cbfeaf | 2197 | if (from < blocksize && to > PAGE_SIZE - blocksize) |
8ab22b9a HH |
2198 | return 0; |
2199 | ||
8ab22b9a HH |
2200 | bh = head; |
2201 | block_start = 0; | |
2202 | do { | |
2203 | block_end = block_start + blocksize; | |
2204 | if (block_end > from && block_start < to) { | |
2205 | if (!buffer_uptodate(bh)) { | |
2206 | ret = 0; | |
2207 | break; | |
2208 | } | |
2209 | if (block_end >= to) | |
2210 | break; | |
2211 | } | |
2212 | block_start = block_end; | |
2213 | bh = bh->b_this_page; | |
2214 | } while (bh != head); | |
2215 | ||
2216 | return ret; | |
2217 | } | |
2218 | EXPORT_SYMBOL(block_is_partially_uptodate); | |
2219 | ||
1da177e4 LT |
2220 | /* |
2221 | * Generic "read page" function for block devices that have the normal | |
2222 | * get_block functionality. This is most of the block device filesystems. | |
2223 | * Reads the page asynchronously --- the unlock_buffer() and | |
2224 | * set/clear_buffer_uptodate() functions propagate buffer state into the | |
2225 | * page struct once IO has completed. | |
2226 | */ | |
2227 | int block_read_full_page(struct page *page, get_block_t *get_block) | |
2228 | { | |
2229 | struct inode *inode = page->mapping->host; | |
2230 | sector_t iblock, lblock; | |
2231 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; | |
45bce8f3 | 2232 | unsigned int blocksize, bbits; |
1da177e4 LT |
2233 | int nr, i; |
2234 | int fully_mapped = 1; | |
2235 | ||
45bce8f3 LT |
2236 | head = create_page_buffers(page, inode, 0); |
2237 | blocksize = head->b_size; | |
2238 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2239 | |
09cbfeaf | 2240 | iblock = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 2241 | lblock = (i_size_read(inode)+blocksize-1) >> bbits; |
1da177e4 LT |
2242 | bh = head; |
2243 | nr = 0; | |
2244 | i = 0; | |
2245 | ||
2246 | do { | |
2247 | if (buffer_uptodate(bh)) | |
2248 | continue; | |
2249 | ||
2250 | if (!buffer_mapped(bh)) { | |
c64610ba AM |
2251 | int err = 0; |
2252 | ||
1da177e4 LT |
2253 | fully_mapped = 0; |
2254 | if (iblock < lblock) { | |
b0cf2321 | 2255 | WARN_ON(bh->b_size != blocksize); |
c64610ba AM |
2256 | err = get_block(inode, iblock, bh, 0); |
2257 | if (err) | |
1da177e4 LT |
2258 | SetPageError(page); |
2259 | } | |
2260 | if (!buffer_mapped(bh)) { | |
eebd2aa3 | 2261 | zero_user(page, i * blocksize, blocksize); |
c64610ba AM |
2262 | if (!err) |
2263 | set_buffer_uptodate(bh); | |
1da177e4 LT |
2264 | continue; |
2265 | } | |
2266 | /* | |
2267 | * get_block() might have updated the buffer | |
2268 | * synchronously | |
2269 | */ | |
2270 | if (buffer_uptodate(bh)) | |
2271 | continue; | |
2272 | } | |
2273 | arr[nr++] = bh; | |
2274 | } while (i++, iblock++, (bh = bh->b_this_page) != head); | |
2275 | ||
2276 | if (fully_mapped) | |
2277 | SetPageMappedToDisk(page); | |
2278 | ||
2279 | if (!nr) { | |
2280 | /* | |
2281 | * All buffers are uptodate - we can set the page uptodate | |
2282 | * as well. But not if get_block() returned an error. | |
2283 | */ | |
2284 | if (!PageError(page)) | |
2285 | SetPageUptodate(page); | |
2286 | unlock_page(page); | |
2287 | return 0; | |
2288 | } | |
2289 | ||
2290 | /* Stage two: lock the buffers */ | |
2291 | for (i = 0; i < nr; i++) { | |
2292 | bh = arr[i]; | |
2293 | lock_buffer(bh); | |
2294 | mark_buffer_async_read(bh); | |
2295 | } | |
2296 | ||
2297 | /* | |
2298 | * Stage 3: start the IO. Check for uptodateness | |
2299 | * inside the buffer lock in case another process reading | |
2300 | * the underlying blockdev brought it uptodate (the sct fix). | |
2301 | */ | |
2302 | for (i = 0; i < nr; i++) { | |
2303 | bh = arr[i]; | |
2304 | if (buffer_uptodate(bh)) | |
2305 | end_buffer_async_read(bh, 1); | |
2306 | else | |
2a222ca9 | 2307 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
2308 | } |
2309 | return 0; | |
2310 | } | |
1fe72eaa | 2311 | EXPORT_SYMBOL(block_read_full_page); |
1da177e4 LT |
2312 | |
2313 | /* utility function for filesystems that need to do work on expanding | |
89e10787 | 2314 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
1da177e4 LT |
2315 | * deal with the hole. |
2316 | */ | |
89e10787 | 2317 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
1da177e4 LT |
2318 | { |
2319 | struct address_space *mapping = inode->i_mapping; | |
2320 | struct page *page; | |
89e10787 | 2321 | void *fsdata; |
1da177e4 LT |
2322 | int err; |
2323 | ||
c08d3b0e | 2324 | err = inode_newsize_ok(inode, size); |
2325 | if (err) | |
1da177e4 LT |
2326 | goto out; |
2327 | ||
89e10787 | 2328 | err = pagecache_write_begin(NULL, mapping, size, 0, |
c718a975 | 2329 | AOP_FLAG_CONT_EXPAND, &page, &fsdata); |
89e10787 | 2330 | if (err) |
05eb0b51 | 2331 | goto out; |
05eb0b51 | 2332 | |
89e10787 NP |
2333 | err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); |
2334 | BUG_ON(err > 0); | |
05eb0b51 | 2335 | |
1da177e4 LT |
2336 | out: |
2337 | return err; | |
2338 | } | |
1fe72eaa | 2339 | EXPORT_SYMBOL(generic_cont_expand_simple); |
1da177e4 | 2340 | |
f1e3af72 AB |
2341 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
2342 | loff_t pos, loff_t *bytes) | |
1da177e4 | 2343 | { |
1da177e4 | 2344 | struct inode *inode = mapping->host; |
93407472 | 2345 | unsigned int blocksize = i_blocksize(inode); |
89e10787 NP |
2346 | struct page *page; |
2347 | void *fsdata; | |
2348 | pgoff_t index, curidx; | |
2349 | loff_t curpos; | |
2350 | unsigned zerofrom, offset, len; | |
2351 | int err = 0; | |
1da177e4 | 2352 | |
09cbfeaf KS |
2353 | index = pos >> PAGE_SHIFT; |
2354 | offset = pos & ~PAGE_MASK; | |
89e10787 | 2355 | |
09cbfeaf KS |
2356 | while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { |
2357 | zerofrom = curpos & ~PAGE_MASK; | |
1da177e4 LT |
2358 | if (zerofrom & (blocksize-1)) { |
2359 | *bytes |= (blocksize-1); | |
2360 | (*bytes)++; | |
2361 | } | |
09cbfeaf | 2362 | len = PAGE_SIZE - zerofrom; |
1da177e4 | 2363 | |
c718a975 TH |
2364 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2365 | &page, &fsdata); | |
89e10787 NP |
2366 | if (err) |
2367 | goto out; | |
eebd2aa3 | 2368 | zero_user(page, zerofrom, len); |
89e10787 NP |
2369 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2370 | page, fsdata); | |
2371 | if (err < 0) | |
2372 | goto out; | |
2373 | BUG_ON(err != len); | |
2374 | err = 0; | |
061e9746 OH |
2375 | |
2376 | balance_dirty_pages_ratelimited(mapping); | |
c2ca0fcd | 2377 | |
08d405c8 | 2378 | if (fatal_signal_pending(current)) { |
c2ca0fcd MP |
2379 | err = -EINTR; |
2380 | goto out; | |
2381 | } | |
89e10787 | 2382 | } |
1da177e4 | 2383 | |
89e10787 NP |
2384 | /* page covers the boundary, find the boundary offset */ |
2385 | if (index == curidx) { | |
09cbfeaf | 2386 | zerofrom = curpos & ~PAGE_MASK; |
1da177e4 | 2387 | /* if we will expand the thing last block will be filled */ |
89e10787 NP |
2388 | if (offset <= zerofrom) { |
2389 | goto out; | |
2390 | } | |
2391 | if (zerofrom & (blocksize-1)) { | |
1da177e4 LT |
2392 | *bytes |= (blocksize-1); |
2393 | (*bytes)++; | |
2394 | } | |
89e10787 | 2395 | len = offset - zerofrom; |
1da177e4 | 2396 | |
c718a975 TH |
2397 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2398 | &page, &fsdata); | |
89e10787 NP |
2399 | if (err) |
2400 | goto out; | |
eebd2aa3 | 2401 | zero_user(page, zerofrom, len); |
89e10787 NP |
2402 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2403 | page, fsdata); | |
2404 | if (err < 0) | |
2405 | goto out; | |
2406 | BUG_ON(err != len); | |
2407 | err = 0; | |
1da177e4 | 2408 | } |
89e10787 NP |
2409 | out: |
2410 | return err; | |
2411 | } | |
2412 | ||
2413 | /* | |
2414 | * For moronic filesystems that do not allow holes in file. | |
2415 | * We may have to extend the file. | |
2416 | */ | |
282dc178 | 2417 | int cont_write_begin(struct file *file, struct address_space *mapping, |
89e10787 NP |
2418 | loff_t pos, unsigned len, unsigned flags, |
2419 | struct page **pagep, void **fsdata, | |
2420 | get_block_t *get_block, loff_t *bytes) | |
2421 | { | |
2422 | struct inode *inode = mapping->host; | |
93407472 FF |
2423 | unsigned int blocksize = i_blocksize(inode); |
2424 | unsigned int zerofrom; | |
89e10787 NP |
2425 | int err; |
2426 | ||
2427 | err = cont_expand_zero(file, mapping, pos, bytes); | |
2428 | if (err) | |
155130a4 | 2429 | return err; |
89e10787 | 2430 | |
09cbfeaf | 2431 | zerofrom = *bytes & ~PAGE_MASK; |
89e10787 NP |
2432 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { |
2433 | *bytes |= (blocksize-1); | |
2434 | (*bytes)++; | |
1da177e4 | 2435 | } |
1da177e4 | 2436 | |
155130a4 | 2437 | return block_write_begin(mapping, pos, len, flags, pagep, get_block); |
1da177e4 | 2438 | } |
1fe72eaa | 2439 | EXPORT_SYMBOL(cont_write_begin); |
1da177e4 | 2440 | |
1da177e4 LT |
2441 | int block_commit_write(struct page *page, unsigned from, unsigned to) |
2442 | { | |
2443 | struct inode *inode = page->mapping->host; | |
2444 | __block_commit_write(inode,page,from,to); | |
2445 | return 0; | |
2446 | } | |
1fe72eaa | 2447 | EXPORT_SYMBOL(block_commit_write); |
1da177e4 | 2448 | |
54171690 DC |
2449 | /* |
2450 | * block_page_mkwrite() is not allowed to change the file size as it gets | |
2451 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2452 | * be careful to check for EOF conditions here. We set the page up correctly | |
2453 | * for a written page which means we get ENOSPC checking when writing into | |
2454 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2455 | * support these features. | |
2456 | * | |
2457 | * We are not allowed to take the i_mutex here so we have to play games to | |
2458 | * protect against truncate races as the page could now be beyond EOF. Because | |
7bb46a67 | 2459 | * truncate writes the inode size before removing pages, once we have the |
54171690 DC |
2460 | * page lock we can determine safely if the page is beyond EOF. If it is not |
2461 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2462 | * unlock the page. | |
ea13a864 | 2463 | * |
14da9200 | 2464 | * Direct callers of this function should protect against filesystem freezing |
5c500029 | 2465 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
54171690 | 2466 | */ |
5c500029 | 2467 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
24da4fab | 2468 | get_block_t get_block) |
54171690 | 2469 | { |
c2ec175c | 2470 | struct page *page = vmf->page; |
496ad9aa | 2471 | struct inode *inode = file_inode(vma->vm_file); |
54171690 DC |
2472 | unsigned long end; |
2473 | loff_t size; | |
24da4fab | 2474 | int ret; |
54171690 DC |
2475 | |
2476 | lock_page(page); | |
2477 | size = i_size_read(inode); | |
2478 | if ((page->mapping != inode->i_mapping) || | |
18336338 | 2479 | (page_offset(page) > size)) { |
24da4fab JK |
2480 | /* We overload EFAULT to mean page got truncated */ |
2481 | ret = -EFAULT; | |
2482 | goto out_unlock; | |
54171690 DC |
2483 | } |
2484 | ||
2485 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
2486 | if (((page->index + 1) << PAGE_SHIFT) > size) |
2487 | end = size & ~PAGE_MASK; | |
54171690 | 2488 | else |
09cbfeaf | 2489 | end = PAGE_SIZE; |
54171690 | 2490 | |
ebdec241 | 2491 | ret = __block_write_begin(page, 0, end, get_block); |
54171690 DC |
2492 | if (!ret) |
2493 | ret = block_commit_write(page, 0, end); | |
2494 | ||
24da4fab JK |
2495 | if (unlikely(ret < 0)) |
2496 | goto out_unlock; | |
ea13a864 | 2497 | set_page_dirty(page); |
1d1d1a76 | 2498 | wait_for_stable_page(page); |
24da4fab JK |
2499 | return 0; |
2500 | out_unlock: | |
2501 | unlock_page(page); | |
54171690 | 2502 | return ret; |
24da4fab | 2503 | } |
1fe72eaa | 2504 | EXPORT_SYMBOL(block_page_mkwrite); |
1da177e4 LT |
2505 | |
2506 | /* | |
03158cd7 | 2507 | * nobh_write_begin()'s prereads are special: the buffer_heads are freed |
1da177e4 LT |
2508 | * immediately, while under the page lock. So it needs a special end_io |
2509 | * handler which does not touch the bh after unlocking it. | |
1da177e4 LT |
2510 | */ |
2511 | static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) | |
2512 | { | |
68671f35 | 2513 | __end_buffer_read_notouch(bh, uptodate); |
1da177e4 LT |
2514 | } |
2515 | ||
03158cd7 NP |
2516 | /* |
2517 | * Attach the singly-linked list of buffers created by nobh_write_begin, to | |
2518 | * the page (converting it to circular linked list and taking care of page | |
2519 | * dirty races). | |
2520 | */ | |
2521 | static void attach_nobh_buffers(struct page *page, struct buffer_head *head) | |
2522 | { | |
2523 | struct buffer_head *bh; | |
2524 | ||
2525 | BUG_ON(!PageLocked(page)); | |
2526 | ||
2527 | spin_lock(&page->mapping->private_lock); | |
2528 | bh = head; | |
2529 | do { | |
2530 | if (PageDirty(page)) | |
2531 | set_buffer_dirty(bh); | |
2532 | if (!bh->b_this_page) | |
2533 | bh->b_this_page = head; | |
2534 | bh = bh->b_this_page; | |
2535 | } while (bh != head); | |
2536 | attach_page_buffers(page, head); | |
2537 | spin_unlock(&page->mapping->private_lock); | |
2538 | } | |
2539 | ||
1da177e4 | 2540 | /* |
ea0f04e5 CH |
2541 | * On entry, the page is fully not uptodate. |
2542 | * On exit the page is fully uptodate in the areas outside (from,to) | |
7bb46a67 | 2543 | * The filesystem needs to handle block truncation upon failure. |
1da177e4 | 2544 | */ |
ea0f04e5 | 2545 | int nobh_write_begin(struct address_space *mapping, |
03158cd7 NP |
2546 | loff_t pos, unsigned len, unsigned flags, |
2547 | struct page **pagep, void **fsdata, | |
1da177e4 LT |
2548 | get_block_t *get_block) |
2549 | { | |
03158cd7 | 2550 | struct inode *inode = mapping->host; |
1da177e4 LT |
2551 | const unsigned blkbits = inode->i_blkbits; |
2552 | const unsigned blocksize = 1 << blkbits; | |
a4b0672d | 2553 | struct buffer_head *head, *bh; |
03158cd7 NP |
2554 | struct page *page; |
2555 | pgoff_t index; | |
2556 | unsigned from, to; | |
1da177e4 | 2557 | unsigned block_in_page; |
a4b0672d | 2558 | unsigned block_start, block_end; |
1da177e4 | 2559 | sector_t block_in_file; |
1da177e4 | 2560 | int nr_reads = 0; |
1da177e4 LT |
2561 | int ret = 0; |
2562 | int is_mapped_to_disk = 1; | |
1da177e4 | 2563 | |
09cbfeaf KS |
2564 | index = pos >> PAGE_SHIFT; |
2565 | from = pos & (PAGE_SIZE - 1); | |
03158cd7 NP |
2566 | to = from + len; |
2567 | ||
54566b2c | 2568 | page = grab_cache_page_write_begin(mapping, index, flags); |
03158cd7 NP |
2569 | if (!page) |
2570 | return -ENOMEM; | |
2571 | *pagep = page; | |
2572 | *fsdata = NULL; | |
2573 | ||
2574 | if (page_has_buffers(page)) { | |
309f77ad NK |
2575 | ret = __block_write_begin(page, pos, len, get_block); |
2576 | if (unlikely(ret)) | |
2577 | goto out_release; | |
2578 | return ret; | |
03158cd7 | 2579 | } |
a4b0672d | 2580 | |
1da177e4 LT |
2581 | if (PageMappedToDisk(page)) |
2582 | return 0; | |
2583 | ||
a4b0672d NP |
2584 | /* |
2585 | * Allocate buffers so that we can keep track of state, and potentially | |
2586 | * attach them to the page if an error occurs. In the common case of | |
2587 | * no error, they will just be freed again without ever being attached | |
2588 | * to the page (which is all OK, because we're under the page lock). | |
2589 | * | |
2590 | * Be careful: the buffer linked list is a NULL terminated one, rather | |
2591 | * than the circular one we're used to. | |
2592 | */ | |
640ab98f | 2593 | head = alloc_page_buffers(page, blocksize, false); |
03158cd7 NP |
2594 | if (!head) { |
2595 | ret = -ENOMEM; | |
2596 | goto out_release; | |
2597 | } | |
a4b0672d | 2598 | |
09cbfeaf | 2599 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
1da177e4 LT |
2600 | |
2601 | /* | |
2602 | * We loop across all blocks in the page, whether or not they are | |
2603 | * part of the affected region. This is so we can discover if the | |
2604 | * page is fully mapped-to-disk. | |
2605 | */ | |
a4b0672d | 2606 | for (block_start = 0, block_in_page = 0, bh = head; |
09cbfeaf | 2607 | block_start < PAGE_SIZE; |
a4b0672d | 2608 | block_in_page++, block_start += blocksize, bh = bh->b_this_page) { |
1da177e4 LT |
2609 | int create; |
2610 | ||
a4b0672d NP |
2611 | block_end = block_start + blocksize; |
2612 | bh->b_state = 0; | |
1da177e4 LT |
2613 | create = 1; |
2614 | if (block_start >= to) | |
2615 | create = 0; | |
2616 | ret = get_block(inode, block_in_file + block_in_page, | |
a4b0672d | 2617 | bh, create); |
1da177e4 LT |
2618 | if (ret) |
2619 | goto failed; | |
a4b0672d | 2620 | if (!buffer_mapped(bh)) |
1da177e4 | 2621 | is_mapped_to_disk = 0; |
a4b0672d | 2622 | if (buffer_new(bh)) |
e64855c6 | 2623 | clean_bdev_bh_alias(bh); |
a4b0672d NP |
2624 | if (PageUptodate(page)) { |
2625 | set_buffer_uptodate(bh); | |
1da177e4 | 2626 | continue; |
a4b0672d NP |
2627 | } |
2628 | if (buffer_new(bh) || !buffer_mapped(bh)) { | |
eebd2aa3 CL |
2629 | zero_user_segments(page, block_start, from, |
2630 | to, block_end); | |
1da177e4 LT |
2631 | continue; |
2632 | } | |
a4b0672d | 2633 | if (buffer_uptodate(bh)) |
1da177e4 LT |
2634 | continue; /* reiserfs does this */ |
2635 | if (block_start < from || block_end > to) { | |
a4b0672d NP |
2636 | lock_buffer(bh); |
2637 | bh->b_end_io = end_buffer_read_nobh; | |
2a222ca9 | 2638 | submit_bh(REQ_OP_READ, 0, bh); |
a4b0672d | 2639 | nr_reads++; |
1da177e4 LT |
2640 | } |
2641 | } | |
2642 | ||
2643 | if (nr_reads) { | |
1da177e4 LT |
2644 | /* |
2645 | * The page is locked, so these buffers are protected from | |
2646 | * any VM or truncate activity. Hence we don't need to care | |
2647 | * for the buffer_head refcounts. | |
2648 | */ | |
a4b0672d | 2649 | for (bh = head; bh; bh = bh->b_this_page) { |
1da177e4 LT |
2650 | wait_on_buffer(bh); |
2651 | if (!buffer_uptodate(bh)) | |
2652 | ret = -EIO; | |
1da177e4 LT |
2653 | } |
2654 | if (ret) | |
2655 | goto failed; | |
2656 | } | |
2657 | ||
2658 | if (is_mapped_to_disk) | |
2659 | SetPageMappedToDisk(page); | |
1da177e4 | 2660 | |
03158cd7 | 2661 | *fsdata = head; /* to be released by nobh_write_end */ |
a4b0672d | 2662 | |
1da177e4 LT |
2663 | return 0; |
2664 | ||
2665 | failed: | |
03158cd7 | 2666 | BUG_ON(!ret); |
1da177e4 | 2667 | /* |
a4b0672d NP |
2668 | * Error recovery is a bit difficult. We need to zero out blocks that |
2669 | * were newly allocated, and dirty them to ensure they get written out. | |
2670 | * Buffers need to be attached to the page at this point, otherwise | |
2671 | * the handling of potential IO errors during writeout would be hard | |
2672 | * (could try doing synchronous writeout, but what if that fails too?) | |
1da177e4 | 2673 | */ |
03158cd7 NP |
2674 | attach_nobh_buffers(page, head); |
2675 | page_zero_new_buffers(page, from, to); | |
a4b0672d | 2676 | |
03158cd7 NP |
2677 | out_release: |
2678 | unlock_page(page); | |
09cbfeaf | 2679 | put_page(page); |
03158cd7 | 2680 | *pagep = NULL; |
a4b0672d | 2681 | |
7bb46a67 | 2682 | return ret; |
2683 | } | |
03158cd7 | 2684 | EXPORT_SYMBOL(nobh_write_begin); |
1da177e4 | 2685 | |
03158cd7 NP |
2686 | int nobh_write_end(struct file *file, struct address_space *mapping, |
2687 | loff_t pos, unsigned len, unsigned copied, | |
2688 | struct page *page, void *fsdata) | |
1da177e4 LT |
2689 | { |
2690 | struct inode *inode = page->mapping->host; | |
efdc3131 | 2691 | struct buffer_head *head = fsdata; |
03158cd7 | 2692 | struct buffer_head *bh; |
5b41e74a | 2693 | BUG_ON(fsdata != NULL && page_has_buffers(page)); |
1da177e4 | 2694 | |
d4cf109f | 2695 | if (unlikely(copied < len) && head) |
5b41e74a DM |
2696 | attach_nobh_buffers(page, head); |
2697 | if (page_has_buffers(page)) | |
2698 | return generic_write_end(file, mapping, pos, len, | |
2699 | copied, page, fsdata); | |
a4b0672d | 2700 | |
22c8ca78 | 2701 | SetPageUptodate(page); |
1da177e4 | 2702 | set_page_dirty(page); |
03158cd7 NP |
2703 | if (pos+copied > inode->i_size) { |
2704 | i_size_write(inode, pos+copied); | |
1da177e4 LT |
2705 | mark_inode_dirty(inode); |
2706 | } | |
03158cd7 NP |
2707 | |
2708 | unlock_page(page); | |
09cbfeaf | 2709 | put_page(page); |
03158cd7 | 2710 | |
03158cd7 NP |
2711 | while (head) { |
2712 | bh = head; | |
2713 | head = head->b_this_page; | |
2714 | free_buffer_head(bh); | |
2715 | } | |
2716 | ||
2717 | return copied; | |
1da177e4 | 2718 | } |
03158cd7 | 2719 | EXPORT_SYMBOL(nobh_write_end); |
1da177e4 LT |
2720 | |
2721 | /* | |
2722 | * nobh_writepage() - based on block_full_write_page() except | |
2723 | * that it tries to operate without attaching bufferheads to | |
2724 | * the page. | |
2725 | */ | |
2726 | int nobh_writepage(struct page *page, get_block_t *get_block, | |
2727 | struct writeback_control *wbc) | |
2728 | { | |
2729 | struct inode * const inode = page->mapping->host; | |
2730 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2731 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2732 | unsigned offset; |
1da177e4 LT |
2733 | int ret; |
2734 | ||
2735 | /* Is the page fully inside i_size? */ | |
2736 | if (page->index < end_index) | |
2737 | goto out; | |
2738 | ||
2739 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2740 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 | 2741 | if (page->index >= end_index+1 || !offset) { |
1da177e4 LT |
2742 | unlock_page(page); |
2743 | return 0; /* don't care */ | |
2744 | } | |
2745 | ||
2746 | /* | |
2747 | * The page straddles i_size. It must be zeroed out on each and every | |
2748 | * writepage invocation because it may be mmapped. "A file is mapped | |
2749 | * in multiples of the page size. For a file that is not a multiple of | |
2750 | * the page size, the remaining memory is zeroed when mapped, and | |
2751 | * writes to that region are not written out to the file." | |
2752 | */ | |
09cbfeaf | 2753 | zero_user_segment(page, offset, PAGE_SIZE); |
1da177e4 LT |
2754 | out: |
2755 | ret = mpage_writepage(page, get_block, wbc); | |
2756 | if (ret == -EAGAIN) | |
35c80d5f CM |
2757 | ret = __block_write_full_page(inode, page, get_block, wbc, |
2758 | end_buffer_async_write); | |
1da177e4 LT |
2759 | return ret; |
2760 | } | |
2761 | EXPORT_SYMBOL(nobh_writepage); | |
2762 | ||
03158cd7 NP |
2763 | int nobh_truncate_page(struct address_space *mapping, |
2764 | loff_t from, get_block_t *get_block) | |
1da177e4 | 2765 | { |
09cbfeaf KS |
2766 | pgoff_t index = from >> PAGE_SHIFT; |
2767 | unsigned offset = from & (PAGE_SIZE-1); | |
03158cd7 NP |
2768 | unsigned blocksize; |
2769 | sector_t iblock; | |
2770 | unsigned length, pos; | |
2771 | struct inode *inode = mapping->host; | |
1da177e4 | 2772 | struct page *page; |
03158cd7 NP |
2773 | struct buffer_head map_bh; |
2774 | int err; | |
1da177e4 | 2775 | |
93407472 | 2776 | blocksize = i_blocksize(inode); |
03158cd7 NP |
2777 | length = offset & (blocksize - 1); |
2778 | ||
2779 | /* Block boundary? Nothing to do */ | |
2780 | if (!length) | |
2781 | return 0; | |
2782 | ||
2783 | length = blocksize - length; | |
09cbfeaf | 2784 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 | 2785 | |
1da177e4 | 2786 | page = grab_cache_page(mapping, index); |
03158cd7 | 2787 | err = -ENOMEM; |
1da177e4 LT |
2788 | if (!page) |
2789 | goto out; | |
2790 | ||
03158cd7 NP |
2791 | if (page_has_buffers(page)) { |
2792 | has_buffers: | |
2793 | unlock_page(page); | |
09cbfeaf | 2794 | put_page(page); |
03158cd7 NP |
2795 | return block_truncate_page(mapping, from, get_block); |
2796 | } | |
2797 | ||
2798 | /* Find the buffer that contains "offset" */ | |
2799 | pos = blocksize; | |
2800 | while (offset >= pos) { | |
2801 | iblock++; | |
2802 | pos += blocksize; | |
2803 | } | |
2804 | ||
460bcf57 TT |
2805 | map_bh.b_size = blocksize; |
2806 | map_bh.b_state = 0; | |
03158cd7 NP |
2807 | err = get_block(inode, iblock, &map_bh, 0); |
2808 | if (err) | |
2809 | goto unlock; | |
2810 | /* unmapped? It's a hole - nothing to do */ | |
2811 | if (!buffer_mapped(&map_bh)) | |
2812 | goto unlock; | |
2813 | ||
2814 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2815 | if (!PageUptodate(page)) { | |
2816 | err = mapping->a_ops->readpage(NULL, page); | |
2817 | if (err) { | |
09cbfeaf | 2818 | put_page(page); |
03158cd7 NP |
2819 | goto out; |
2820 | } | |
2821 | lock_page(page); | |
2822 | if (!PageUptodate(page)) { | |
2823 | err = -EIO; | |
2824 | goto unlock; | |
2825 | } | |
2826 | if (page_has_buffers(page)) | |
2827 | goto has_buffers; | |
1da177e4 | 2828 | } |
eebd2aa3 | 2829 | zero_user(page, offset, length); |
03158cd7 NP |
2830 | set_page_dirty(page); |
2831 | err = 0; | |
2832 | ||
2833 | unlock: | |
1da177e4 | 2834 | unlock_page(page); |
09cbfeaf | 2835 | put_page(page); |
1da177e4 | 2836 | out: |
03158cd7 | 2837 | return err; |
1da177e4 LT |
2838 | } |
2839 | EXPORT_SYMBOL(nobh_truncate_page); | |
2840 | ||
2841 | int block_truncate_page(struct address_space *mapping, | |
2842 | loff_t from, get_block_t *get_block) | |
2843 | { | |
09cbfeaf KS |
2844 | pgoff_t index = from >> PAGE_SHIFT; |
2845 | unsigned offset = from & (PAGE_SIZE-1); | |
1da177e4 | 2846 | unsigned blocksize; |
54b21a79 | 2847 | sector_t iblock; |
1da177e4 LT |
2848 | unsigned length, pos; |
2849 | struct inode *inode = mapping->host; | |
2850 | struct page *page; | |
2851 | struct buffer_head *bh; | |
1da177e4 LT |
2852 | int err; |
2853 | ||
93407472 | 2854 | blocksize = i_blocksize(inode); |
1da177e4 LT |
2855 | length = offset & (blocksize - 1); |
2856 | ||
2857 | /* Block boundary? Nothing to do */ | |
2858 | if (!length) | |
2859 | return 0; | |
2860 | ||
2861 | length = blocksize - length; | |
09cbfeaf | 2862 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 LT |
2863 | |
2864 | page = grab_cache_page(mapping, index); | |
2865 | err = -ENOMEM; | |
2866 | if (!page) | |
2867 | goto out; | |
2868 | ||
2869 | if (!page_has_buffers(page)) | |
2870 | create_empty_buffers(page, blocksize, 0); | |
2871 | ||
2872 | /* Find the buffer that contains "offset" */ | |
2873 | bh = page_buffers(page); | |
2874 | pos = blocksize; | |
2875 | while (offset >= pos) { | |
2876 | bh = bh->b_this_page; | |
2877 | iblock++; | |
2878 | pos += blocksize; | |
2879 | } | |
2880 | ||
2881 | err = 0; | |
2882 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2883 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
2884 | err = get_block(inode, iblock, bh, 0); |
2885 | if (err) | |
2886 | goto unlock; | |
2887 | /* unmapped? It's a hole - nothing to do */ | |
2888 | if (!buffer_mapped(bh)) | |
2889 | goto unlock; | |
2890 | } | |
2891 | ||
2892 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2893 | if (PageUptodate(page)) | |
2894 | set_buffer_uptodate(bh); | |
2895 | ||
33a266dd | 2896 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
1da177e4 | 2897 | err = -EIO; |
dfec8a14 | 2898 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2899 | wait_on_buffer(bh); |
2900 | /* Uhhuh. Read error. Complain and punt. */ | |
2901 | if (!buffer_uptodate(bh)) | |
2902 | goto unlock; | |
2903 | } | |
2904 | ||
eebd2aa3 | 2905 | zero_user(page, offset, length); |
1da177e4 LT |
2906 | mark_buffer_dirty(bh); |
2907 | err = 0; | |
2908 | ||
2909 | unlock: | |
2910 | unlock_page(page); | |
09cbfeaf | 2911 | put_page(page); |
1da177e4 LT |
2912 | out: |
2913 | return err; | |
2914 | } | |
1fe72eaa | 2915 | EXPORT_SYMBOL(block_truncate_page); |
1da177e4 LT |
2916 | |
2917 | /* | |
2918 | * The generic ->writepage function for buffer-backed address_spaces | |
2919 | */ | |
1b938c08 MW |
2920 | int block_write_full_page(struct page *page, get_block_t *get_block, |
2921 | struct writeback_control *wbc) | |
1da177e4 LT |
2922 | { |
2923 | struct inode * const inode = page->mapping->host; | |
2924 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2925 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2926 | unsigned offset; |
1da177e4 LT |
2927 | |
2928 | /* Is the page fully inside i_size? */ | |
2929 | if (page->index < end_index) | |
35c80d5f | 2930 | return __block_write_full_page(inode, page, get_block, wbc, |
1b938c08 | 2931 | end_buffer_async_write); |
1da177e4 LT |
2932 | |
2933 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2934 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 | 2935 | if (page->index >= end_index+1 || !offset) { |
1da177e4 LT |
2936 | unlock_page(page); |
2937 | return 0; /* don't care */ | |
2938 | } | |
2939 | ||
2940 | /* | |
2941 | * The page straddles i_size. It must be zeroed out on each and every | |
2a61aa40 | 2942 | * writepage invocation because it may be mmapped. "A file is mapped |
1da177e4 LT |
2943 | * in multiples of the page size. For a file that is not a multiple of |
2944 | * the page size, the remaining memory is zeroed when mapped, and | |
2945 | * writes to that region are not written out to the file." | |
2946 | */ | |
09cbfeaf | 2947 | zero_user_segment(page, offset, PAGE_SIZE); |
1b938c08 MW |
2948 | return __block_write_full_page(inode, page, get_block, wbc, |
2949 | end_buffer_async_write); | |
35c80d5f | 2950 | } |
1fe72eaa | 2951 | EXPORT_SYMBOL(block_write_full_page); |
35c80d5f | 2952 | |
1da177e4 LT |
2953 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
2954 | get_block_t *get_block) | |
2955 | { | |
1da177e4 | 2956 | struct inode *inode = mapping->host; |
2a527d68 AP |
2957 | struct buffer_head tmp = { |
2958 | .b_size = i_blocksize(inode), | |
2959 | }; | |
2960 | ||
1da177e4 LT |
2961 | get_block(inode, block, &tmp, 0); |
2962 | return tmp.b_blocknr; | |
2963 | } | |
1fe72eaa | 2964 | EXPORT_SYMBOL(generic_block_bmap); |
1da177e4 | 2965 | |
4246a0b6 | 2966 | static void end_bio_bh_io_sync(struct bio *bio) |
1da177e4 LT |
2967 | { |
2968 | struct buffer_head *bh = bio->bi_private; | |
2969 | ||
b7c44ed9 | 2970 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
08bafc03 KM |
2971 | set_bit(BH_Quiet, &bh->b_state); |
2972 | ||
4e4cbee9 | 2973 | bh->b_end_io(bh, !bio->bi_status); |
1da177e4 | 2974 | bio_put(bio); |
1da177e4 LT |
2975 | } |
2976 | ||
57302e0d LT |
2977 | /* |
2978 | * This allows us to do IO even on the odd last sectors | |
59d43914 | 2979 | * of a device, even if the block size is some multiple |
57302e0d LT |
2980 | * of the physical sector size. |
2981 | * | |
2982 | * We'll just truncate the bio to the size of the device, | |
2983 | * and clear the end of the buffer head manually. | |
2984 | * | |
2985 | * Truly out-of-range accesses will turn into actual IO | |
2986 | * errors, this only handles the "we need to be able to | |
2987 | * do IO at the final sector" case. | |
2988 | */ | |
d43dbe13 | 2989 | void guard_bio_eod(struct bio *bio) |
57302e0d LT |
2990 | { |
2991 | sector_t maxsector; | |
67f2519f GE |
2992 | struct hd_struct *part; |
2993 | ||
2994 | rcu_read_lock(); | |
2995 | part = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
2996 | if (part) | |
2997 | maxsector = part_nr_sects_read(part); | |
2998 | else | |
2999 | maxsector = get_capacity(bio->bi_disk); | |
3000 | rcu_read_unlock(); | |
57302e0d | 3001 | |
57302e0d LT |
3002 | if (!maxsector) |
3003 | return; | |
3004 | ||
3005 | /* | |
3006 | * If the *whole* IO is past the end of the device, | |
3007 | * let it through, and the IO layer will turn it into | |
3008 | * an EIO. | |
3009 | */ | |
4f024f37 | 3010 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) |
57302e0d LT |
3011 | return; |
3012 | ||
4f024f37 | 3013 | maxsector -= bio->bi_iter.bi_sector; |
59d43914 | 3014 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) |
57302e0d LT |
3015 | return; |
3016 | ||
d8a54693 | 3017 | bio_truncate(bio, maxsector << 9); |
57302e0d LT |
3018 | } |
3019 | ||
2a222ca9 | 3020 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 3021 | enum rw_hint write_hint, struct writeback_control *wbc) |
1da177e4 LT |
3022 | { |
3023 | struct bio *bio; | |
1da177e4 LT |
3024 | |
3025 | BUG_ON(!buffer_locked(bh)); | |
3026 | BUG_ON(!buffer_mapped(bh)); | |
3027 | BUG_ON(!bh->b_end_io); | |
8fb0e342 AK |
3028 | BUG_ON(buffer_delay(bh)); |
3029 | BUG_ON(buffer_unwritten(bh)); | |
1da177e4 | 3030 | |
1da177e4 | 3031 | /* |
48fd4f93 | 3032 | * Only clear out a write error when rewriting |
1da177e4 | 3033 | */ |
2a222ca9 | 3034 | if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) |
1da177e4 LT |
3035 | clear_buffer_write_io_error(bh); |
3036 | ||
3037 | /* | |
3038 | * from here on down, it's all bio -- do the initial mapping, | |
3039 | * submit_bio -> generic_make_request may further map this bio around | |
3040 | */ | |
3041 | bio = bio_alloc(GFP_NOIO, 1); | |
3042 | ||
4f024f37 | 3043 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
74d46992 | 3044 | bio_set_dev(bio, bh->b_bdev); |
8e8f9298 | 3045 | bio->bi_write_hint = write_hint; |
1da177e4 | 3046 | |
6cf66b4c KO |
3047 | bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
3048 | BUG_ON(bio->bi_iter.bi_size != bh->b_size); | |
1da177e4 LT |
3049 | |
3050 | bio->bi_end_io = end_bio_bh_io_sync; | |
3051 | bio->bi_private = bh; | |
3052 | ||
877f962c | 3053 | if (buffer_meta(bh)) |
2a222ca9 | 3054 | op_flags |= REQ_META; |
877f962c | 3055 | if (buffer_prio(bh)) |
2a222ca9 MC |
3056 | op_flags |= REQ_PRIO; |
3057 | bio_set_op_attrs(bio, op, op_flags); | |
877f962c | 3058 | |
d43dbe13 ML |
3059 | /* Take care of bh's that straddle the end of the device */ |
3060 | guard_bio_eod(bio); | |
3061 | ||
fd42df30 DZ |
3062 | if (wbc) { |
3063 | wbc_init_bio(wbc, bio); | |
34e51a5e | 3064 | wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); |
fd42df30 DZ |
3065 | } |
3066 | ||
4e49ea4a | 3067 | submit_bio(bio); |
f6454b04 | 3068 | return 0; |
1da177e4 | 3069 | } |
bafc0dba | 3070 | |
020c2833 | 3071 | int submit_bh(int op, int op_flags, struct buffer_head *bh) |
bafc0dba | 3072 | { |
8e8f9298 | 3073 | return submit_bh_wbc(op, op_flags, bh, 0, NULL); |
71368511 | 3074 | } |
1fe72eaa | 3075 | EXPORT_SYMBOL(submit_bh); |
1da177e4 LT |
3076 | |
3077 | /** | |
3078 | * ll_rw_block: low-level access to block devices (DEPRECATED) | |
dfec8a14 | 3079 | * @op: whether to %READ or %WRITE |
ef295ecf | 3080 | * @op_flags: req_flag_bits |
1da177e4 LT |
3081 | * @nr: number of &struct buffer_heads in the array |
3082 | * @bhs: array of pointers to &struct buffer_head | |
3083 | * | |
a7662236 | 3084 | * ll_rw_block() takes an array of pointers to &struct buffer_heads, and |
70246286 CH |
3085 | * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE. |
3086 | * @op_flags contains flags modifying the detailed I/O behavior, most notably | |
3087 | * %REQ_RAHEAD. | |
1da177e4 LT |
3088 | * |
3089 | * This function drops any buffer that it cannot get a lock on (with the | |
9cb569d6 CH |
3090 | * BH_Lock state bit), any buffer that appears to be clean when doing a write |
3091 | * request, and any buffer that appears to be up-to-date when doing read | |
3092 | * request. Further it marks as clean buffers that are processed for | |
3093 | * writing (the buffer cache won't assume that they are actually clean | |
3094 | * until the buffer gets unlocked). | |
1da177e4 LT |
3095 | * |
3096 | * ll_rw_block sets b_end_io to simple completion handler that marks | |
e227867f | 3097 | * the buffer up-to-date (if appropriate), unlocks the buffer and wakes |
1da177e4 LT |
3098 | * any waiters. |
3099 | * | |
3100 | * All of the buffers must be for the same device, and must also be a | |
3101 | * multiple of the current approved size for the device. | |
3102 | */ | |
dfec8a14 | 3103 | void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[]) |
1da177e4 LT |
3104 | { |
3105 | int i; | |
3106 | ||
3107 | for (i = 0; i < nr; i++) { | |
3108 | struct buffer_head *bh = bhs[i]; | |
3109 | ||
9cb569d6 | 3110 | if (!trylock_buffer(bh)) |
1da177e4 | 3111 | continue; |
dfec8a14 | 3112 | if (op == WRITE) { |
1da177e4 | 3113 | if (test_clear_buffer_dirty(bh)) { |
76c3073a | 3114 | bh->b_end_io = end_buffer_write_sync; |
e60e5c50 | 3115 | get_bh(bh); |
dfec8a14 | 3116 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3117 | continue; |
3118 | } | |
3119 | } else { | |
1da177e4 | 3120 | if (!buffer_uptodate(bh)) { |
76c3073a | 3121 | bh->b_end_io = end_buffer_read_sync; |
e60e5c50 | 3122 | get_bh(bh); |
dfec8a14 | 3123 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3124 | continue; |
3125 | } | |
3126 | } | |
3127 | unlock_buffer(bh); | |
1da177e4 LT |
3128 | } |
3129 | } | |
1fe72eaa | 3130 | EXPORT_SYMBOL(ll_rw_block); |
1da177e4 | 3131 | |
2a222ca9 | 3132 | void write_dirty_buffer(struct buffer_head *bh, int op_flags) |
9cb569d6 CH |
3133 | { |
3134 | lock_buffer(bh); | |
3135 | if (!test_clear_buffer_dirty(bh)) { | |
3136 | unlock_buffer(bh); | |
3137 | return; | |
3138 | } | |
3139 | bh->b_end_io = end_buffer_write_sync; | |
3140 | get_bh(bh); | |
2a222ca9 | 3141 | submit_bh(REQ_OP_WRITE, op_flags, bh); |
9cb569d6 CH |
3142 | } |
3143 | EXPORT_SYMBOL(write_dirty_buffer); | |
3144 | ||
1da177e4 LT |
3145 | /* |
3146 | * For a data-integrity writeout, we need to wait upon any in-progress I/O | |
3147 | * and then start new I/O and then wait upon it. The caller must have a ref on | |
3148 | * the buffer_head. | |
3149 | */ | |
2a222ca9 | 3150 | int __sync_dirty_buffer(struct buffer_head *bh, int op_flags) |
1da177e4 LT |
3151 | { |
3152 | int ret = 0; | |
3153 | ||
3154 | WARN_ON(atomic_read(&bh->b_count) < 1); | |
3155 | lock_buffer(bh); | |
3156 | if (test_clear_buffer_dirty(bh)) { | |
a099c645 XT |
3157 | /* |
3158 | * The bh should be mapped, but it might not be if the | |
3159 | * device was hot-removed. Not much we can do but fail the I/O. | |
3160 | */ | |
3161 | if (!buffer_mapped(bh)) { | |
3162 | unlock_buffer(bh); | |
3163 | return -EIO; | |
3164 | } | |
3165 | ||
1da177e4 LT |
3166 | get_bh(bh); |
3167 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 3168 | ret = submit_bh(REQ_OP_WRITE, op_flags, bh); |
1da177e4 | 3169 | wait_on_buffer(bh); |
1da177e4 LT |
3170 | if (!ret && !buffer_uptodate(bh)) |
3171 | ret = -EIO; | |
3172 | } else { | |
3173 | unlock_buffer(bh); | |
3174 | } | |
3175 | return ret; | |
3176 | } | |
87e99511 CH |
3177 | EXPORT_SYMBOL(__sync_dirty_buffer); |
3178 | ||
3179 | int sync_dirty_buffer(struct buffer_head *bh) | |
3180 | { | |
70fd7614 | 3181 | return __sync_dirty_buffer(bh, REQ_SYNC); |
87e99511 | 3182 | } |
1fe72eaa | 3183 | EXPORT_SYMBOL(sync_dirty_buffer); |
1da177e4 LT |
3184 | |
3185 | /* | |
3186 | * try_to_free_buffers() checks if all the buffers on this particular page | |
3187 | * are unused, and releases them if so. | |
3188 | * | |
3189 | * Exclusion against try_to_free_buffers may be obtained by either | |
3190 | * locking the page or by holding its mapping's private_lock. | |
3191 | * | |
3192 | * If the page is dirty but all the buffers are clean then we need to | |
3193 | * be sure to mark the page clean as well. This is because the page | |
3194 | * may be against a block device, and a later reattachment of buffers | |
3195 | * to a dirty page will set *all* buffers dirty. Which would corrupt | |
3196 | * filesystem data on the same device. | |
3197 | * | |
3198 | * The same applies to regular filesystem pages: if all the buffers are | |
3199 | * clean then we set the page clean and proceed. To do that, we require | |
3200 | * total exclusion from __set_page_dirty_buffers(). That is obtained with | |
3201 | * private_lock. | |
3202 | * | |
3203 | * try_to_free_buffers() is non-blocking. | |
3204 | */ | |
3205 | static inline int buffer_busy(struct buffer_head *bh) | |
3206 | { | |
3207 | return atomic_read(&bh->b_count) | | |
3208 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); | |
3209 | } | |
3210 | ||
3211 | static int | |
3212 | drop_buffers(struct page *page, struct buffer_head **buffers_to_free) | |
3213 | { | |
3214 | struct buffer_head *head = page_buffers(page); | |
3215 | struct buffer_head *bh; | |
3216 | ||
3217 | bh = head; | |
3218 | do { | |
1da177e4 LT |
3219 | if (buffer_busy(bh)) |
3220 | goto failed; | |
3221 | bh = bh->b_this_page; | |
3222 | } while (bh != head); | |
3223 | ||
3224 | do { | |
3225 | struct buffer_head *next = bh->b_this_page; | |
3226 | ||
535ee2fb | 3227 | if (bh->b_assoc_map) |
1da177e4 LT |
3228 | __remove_assoc_queue(bh); |
3229 | bh = next; | |
3230 | } while (bh != head); | |
3231 | *buffers_to_free = head; | |
3232 | __clear_page_buffers(page); | |
3233 | return 1; | |
3234 | failed: | |
3235 | return 0; | |
3236 | } | |
3237 | ||
3238 | int try_to_free_buffers(struct page *page) | |
3239 | { | |
3240 | struct address_space * const mapping = page->mapping; | |
3241 | struct buffer_head *buffers_to_free = NULL; | |
3242 | int ret = 0; | |
3243 | ||
3244 | BUG_ON(!PageLocked(page)); | |
ecdfc978 | 3245 | if (PageWriteback(page)) |
1da177e4 LT |
3246 | return 0; |
3247 | ||
3248 | if (mapping == NULL) { /* can this still happen? */ | |
3249 | ret = drop_buffers(page, &buffers_to_free); | |
3250 | goto out; | |
3251 | } | |
3252 | ||
3253 | spin_lock(&mapping->private_lock); | |
3254 | ret = drop_buffers(page, &buffers_to_free); | |
ecdfc978 LT |
3255 | |
3256 | /* | |
3257 | * If the filesystem writes its buffers by hand (eg ext3) | |
3258 | * then we can have clean buffers against a dirty page. We | |
3259 | * clean the page here; otherwise the VM will never notice | |
3260 | * that the filesystem did any IO at all. | |
3261 | * | |
3262 | * Also, during truncate, discard_buffer will have marked all | |
3263 | * the page's buffers clean. We discover that here and clean | |
3264 | * the page also. | |
87df7241 NP |
3265 | * |
3266 | * private_lock must be held over this entire operation in order | |
3267 | * to synchronise against __set_page_dirty_buffers and prevent the | |
3268 | * dirty bit from being lost. | |
ecdfc978 | 3269 | */ |
11f81bec TH |
3270 | if (ret) |
3271 | cancel_dirty_page(page); | |
87df7241 | 3272 | spin_unlock(&mapping->private_lock); |
1da177e4 LT |
3273 | out: |
3274 | if (buffers_to_free) { | |
3275 | struct buffer_head *bh = buffers_to_free; | |
3276 | ||
3277 | do { | |
3278 | struct buffer_head *next = bh->b_this_page; | |
3279 | free_buffer_head(bh); | |
3280 | bh = next; | |
3281 | } while (bh != buffers_to_free); | |
3282 | } | |
3283 | return ret; | |
3284 | } | |
3285 | EXPORT_SYMBOL(try_to_free_buffers); | |
3286 | ||
1da177e4 LT |
3287 | /* |
3288 | * There are no bdflush tunables left. But distributions are | |
3289 | * still running obsolete flush daemons, so we terminate them here. | |
3290 | * | |
3291 | * Use of bdflush() is deprecated and will be removed in a future kernel. | |
5b0830cb | 3292 | * The `flush-X' kernel threads fully replace bdflush daemons and this call. |
1da177e4 | 3293 | */ |
bdc480e3 | 3294 | SYSCALL_DEFINE2(bdflush, int, func, long, data) |
1da177e4 LT |
3295 | { |
3296 | static int msg_count; | |
3297 | ||
3298 | if (!capable(CAP_SYS_ADMIN)) | |
3299 | return -EPERM; | |
3300 | ||
3301 | if (msg_count < 5) { | |
3302 | msg_count++; | |
3303 | printk(KERN_INFO | |
3304 | "warning: process `%s' used the obsolete bdflush" | |
3305 | " system call\n", current->comm); | |
3306 | printk(KERN_INFO "Fix your initscripts?\n"); | |
3307 | } | |
3308 | ||
3309 | if (func == 1) | |
3310 | do_exit(0); | |
3311 | return 0; | |
3312 | } | |
3313 | ||
3314 | /* | |
3315 | * Buffer-head allocation | |
3316 | */ | |
a0a9b043 | 3317 | static struct kmem_cache *bh_cachep __read_mostly; |
1da177e4 LT |
3318 | |
3319 | /* | |
3320 | * Once the number of bh's in the machine exceeds this level, we start | |
3321 | * stripping them in writeback. | |
3322 | */ | |
43be594a | 3323 | static unsigned long max_buffer_heads; |
1da177e4 LT |
3324 | |
3325 | int buffer_heads_over_limit; | |
3326 | ||
3327 | struct bh_accounting { | |
3328 | int nr; /* Number of live bh's */ | |
3329 | int ratelimit; /* Limit cacheline bouncing */ | |
3330 | }; | |
3331 | ||
3332 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; | |
3333 | ||
3334 | static void recalc_bh_state(void) | |
3335 | { | |
3336 | int i; | |
3337 | int tot = 0; | |
3338 | ||
ee1be862 | 3339 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
1da177e4 | 3340 | return; |
c7b92516 | 3341 | __this_cpu_write(bh_accounting.ratelimit, 0); |
8a143426 | 3342 | for_each_online_cpu(i) |
1da177e4 LT |
3343 | tot += per_cpu(bh_accounting, i).nr; |
3344 | buffer_heads_over_limit = (tot > max_buffer_heads); | |
3345 | } | |
c7b92516 | 3346 | |
dd0fc66f | 3347 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
1da177e4 | 3348 | { |
019b4d12 | 3349 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
1da177e4 | 3350 | if (ret) { |
a35afb83 | 3351 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
c7b92516 CL |
3352 | preempt_disable(); |
3353 | __this_cpu_inc(bh_accounting.nr); | |
1da177e4 | 3354 | recalc_bh_state(); |
c7b92516 | 3355 | preempt_enable(); |
1da177e4 LT |
3356 | } |
3357 | return ret; | |
3358 | } | |
3359 | EXPORT_SYMBOL(alloc_buffer_head); | |
3360 | ||
3361 | void free_buffer_head(struct buffer_head *bh) | |
3362 | { | |
3363 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); | |
3364 | kmem_cache_free(bh_cachep, bh); | |
c7b92516 CL |
3365 | preempt_disable(); |
3366 | __this_cpu_dec(bh_accounting.nr); | |
1da177e4 | 3367 | recalc_bh_state(); |
c7b92516 | 3368 | preempt_enable(); |
1da177e4 LT |
3369 | } |
3370 | EXPORT_SYMBOL(free_buffer_head); | |
3371 | ||
fc4d24c9 | 3372 | static int buffer_exit_cpu_dead(unsigned int cpu) |
1da177e4 LT |
3373 | { |
3374 | int i; | |
3375 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); | |
3376 | ||
3377 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
3378 | brelse(b->bhs[i]); | |
3379 | b->bhs[i] = NULL; | |
3380 | } | |
c7b92516 | 3381 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
8a143426 | 3382 | per_cpu(bh_accounting, cpu).nr = 0; |
fc4d24c9 | 3383 | return 0; |
1da177e4 | 3384 | } |
1da177e4 | 3385 | |
389d1b08 | 3386 | /** |
a6b91919 | 3387 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
389d1b08 AK |
3388 | * @bh: struct buffer_head |
3389 | * | |
3390 | * Return true if the buffer is up-to-date and false, | |
3391 | * with the buffer locked, if not. | |
3392 | */ | |
3393 | int bh_uptodate_or_lock(struct buffer_head *bh) | |
3394 | { | |
3395 | if (!buffer_uptodate(bh)) { | |
3396 | lock_buffer(bh); | |
3397 | if (!buffer_uptodate(bh)) | |
3398 | return 0; | |
3399 | unlock_buffer(bh); | |
3400 | } | |
3401 | return 1; | |
3402 | } | |
3403 | EXPORT_SYMBOL(bh_uptodate_or_lock); | |
3404 | ||
3405 | /** | |
a6b91919 | 3406 | * bh_submit_read - Submit a locked buffer for reading |
389d1b08 AK |
3407 | * @bh: struct buffer_head |
3408 | * | |
3409 | * Returns zero on success and -EIO on error. | |
3410 | */ | |
3411 | int bh_submit_read(struct buffer_head *bh) | |
3412 | { | |
3413 | BUG_ON(!buffer_locked(bh)); | |
3414 | ||
3415 | if (buffer_uptodate(bh)) { | |
3416 | unlock_buffer(bh); | |
3417 | return 0; | |
3418 | } | |
3419 | ||
3420 | get_bh(bh); | |
3421 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 3422 | submit_bh(REQ_OP_READ, 0, bh); |
389d1b08 AK |
3423 | wait_on_buffer(bh); |
3424 | if (buffer_uptodate(bh)) | |
3425 | return 0; | |
3426 | return -EIO; | |
3427 | } | |
3428 | EXPORT_SYMBOL(bh_submit_read); | |
3429 | ||
1da177e4 LT |
3430 | void __init buffer_init(void) |
3431 | { | |
43be594a | 3432 | unsigned long nrpages; |
fc4d24c9 | 3433 | int ret; |
1da177e4 | 3434 | |
b98938c3 CL |
3435 | bh_cachep = kmem_cache_create("buffer_head", |
3436 | sizeof(struct buffer_head), 0, | |
3437 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
3438 | SLAB_MEM_SPREAD), | |
019b4d12 | 3439 | NULL); |
1da177e4 LT |
3440 | |
3441 | /* | |
3442 | * Limit the bh occupancy to 10% of ZONE_NORMAL | |
3443 | */ | |
3444 | nrpages = (nr_free_buffer_pages() * 10) / 100; | |
3445 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); | |
fc4d24c9 SAS |
3446 | ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", |
3447 | NULL, buffer_exit_cpu_dead); | |
3448 | WARN_ON(ret < 0); | |
1da177e4 | 3449 | } |