]>
Commit | Line | Data |
---|---|---|
1da177e4 LT |
1 | /* |
2 | * linux/fs/buffer.c | |
3 | * | |
4 | * Copyright (C) 1991, 1992, 2002 Linus Torvalds | |
5 | */ | |
6 | ||
7 | /* | |
8 | * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95 | |
9 | * | |
10 | * Removed a lot of unnecessary code and simplified things now that | |
11 | * the buffer cache isn't our primary cache - Andrew Tridgell 12/96 | |
12 | * | |
13 | * Speed up hash, lru, and free list operations. Use gfp() for allocating | |
14 | * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM | |
15 | * | |
16 | * Added 32k buffer block sizes - these are required older ARM systems. - RMK | |
17 | * | |
18 | * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de> | |
19 | */ | |
20 | ||
1da177e4 | 21 | #include <linux/kernel.h> |
f361bf4a | 22 | #include <linux/sched/signal.h> |
1da177e4 LT |
23 | #include <linux/syscalls.h> |
24 | #include <linux/fs.h> | |
ae259a9c | 25 | #include <linux/iomap.h> |
1da177e4 LT |
26 | #include <linux/mm.h> |
27 | #include <linux/percpu.h> | |
28 | #include <linux/slab.h> | |
16f7e0fe | 29 | #include <linux/capability.h> |
1da177e4 LT |
30 | #include <linux/blkdev.h> |
31 | #include <linux/file.h> | |
32 | #include <linux/quotaops.h> | |
33 | #include <linux/highmem.h> | |
630d9c47 | 34 | #include <linux/export.h> |
bafc0dba | 35 | #include <linux/backing-dev.h> |
1da177e4 LT |
36 | #include <linux/writeback.h> |
37 | #include <linux/hash.h> | |
38 | #include <linux/suspend.h> | |
39 | #include <linux/buffer_head.h> | |
55e829af | 40 | #include <linux/task_io_accounting_ops.h> |
1da177e4 LT |
41 | #include <linux/bio.h> |
42 | #include <linux/notifier.h> | |
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> |
5305cb83 | 48 | #include <trace/events/block.h> |
1da177e4 LT |
49 | |
50 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); | |
2a222ca9 | 51 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 52 | enum rw_hint hint, struct writeback_control *wbc); |
1da177e4 LT |
53 | |
54 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) | |
55 | ||
a3f3c29c | 56 | void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private) |
1da177e4 LT |
57 | { |
58 | bh->b_end_io = handler; | |
59 | bh->b_private = private; | |
60 | } | |
1fe72eaa | 61 | EXPORT_SYMBOL(init_buffer); |
1da177e4 | 62 | |
f0059afd TH |
63 | inline void touch_buffer(struct buffer_head *bh) |
64 | { | |
5305cb83 | 65 | trace_block_touch_buffer(bh); |
f0059afd TH |
66 | mark_page_accessed(bh->b_page); |
67 | } | |
68 | EXPORT_SYMBOL(touch_buffer); | |
69 | ||
fc9b52cd | 70 | void __lock_buffer(struct buffer_head *bh) |
1da177e4 | 71 | { |
74316201 | 72 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 LT |
73 | } |
74 | EXPORT_SYMBOL(__lock_buffer); | |
75 | ||
fc9b52cd | 76 | void unlock_buffer(struct buffer_head *bh) |
1da177e4 | 77 | { |
51b07fc3 | 78 | clear_bit_unlock(BH_Lock, &bh->b_state); |
4e857c58 | 79 | smp_mb__after_atomic(); |
1da177e4 LT |
80 | wake_up_bit(&bh->b_state, BH_Lock); |
81 | } | |
1fe72eaa | 82 | EXPORT_SYMBOL(unlock_buffer); |
1da177e4 | 83 | |
b4597226 MG |
84 | /* |
85 | * Returns if the page has dirty or writeback buffers. If all the buffers | |
86 | * are unlocked and clean then the PageDirty information is stale. If | |
87 | * any of the pages are locked, it is assumed they are locked for IO. | |
88 | */ | |
89 | void buffer_check_dirty_writeback(struct page *page, | |
90 | bool *dirty, bool *writeback) | |
91 | { | |
92 | struct buffer_head *head, *bh; | |
93 | *dirty = false; | |
94 | *writeback = false; | |
95 | ||
96 | BUG_ON(!PageLocked(page)); | |
97 | ||
98 | if (!page_has_buffers(page)) | |
99 | return; | |
100 | ||
101 | if (PageWriteback(page)) | |
102 | *writeback = true; | |
103 | ||
104 | head = page_buffers(page); | |
105 | bh = head; | |
106 | do { | |
107 | if (buffer_locked(bh)) | |
108 | *writeback = true; | |
109 | ||
110 | if (buffer_dirty(bh)) | |
111 | *dirty = true; | |
112 | ||
113 | bh = bh->b_this_page; | |
114 | } while (bh != head); | |
115 | } | |
116 | EXPORT_SYMBOL(buffer_check_dirty_writeback); | |
117 | ||
1da177e4 LT |
118 | /* |
119 | * Block until a buffer comes unlocked. This doesn't stop it | |
120 | * from becoming locked again - you have to lock it yourself | |
121 | * if you want to preserve its state. | |
122 | */ | |
123 | void __wait_on_buffer(struct buffer_head * bh) | |
124 | { | |
74316201 | 125 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 | 126 | } |
1fe72eaa | 127 | EXPORT_SYMBOL(__wait_on_buffer); |
1da177e4 LT |
128 | |
129 | static void | |
130 | __clear_page_buffers(struct page *page) | |
131 | { | |
132 | ClearPagePrivate(page); | |
4c21e2f2 | 133 | set_page_private(page, 0); |
09cbfeaf | 134 | put_page(page); |
1da177e4 LT |
135 | } |
136 | ||
b744c2ac | 137 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
1da177e4 | 138 | { |
432f16e6 RE |
139 | if (!test_bit(BH_Quiet, &bh->b_state)) |
140 | printk_ratelimited(KERN_ERR | |
a1c6f057 DM |
141 | "Buffer I/O error on dev %pg, logical block %llu%s\n", |
142 | bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); | |
1da177e4 LT |
143 | } |
144 | ||
145 | /* | |
68671f35 DM |
146 | * End-of-IO handler helper function which does not touch the bh after |
147 | * unlocking it. | |
148 | * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but | |
149 | * a race there is benign: unlock_buffer() only use the bh's address for | |
150 | * hashing after unlocking the buffer, so it doesn't actually touch the bh | |
151 | * itself. | |
1da177e4 | 152 | */ |
68671f35 | 153 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
1da177e4 LT |
154 | { |
155 | if (uptodate) { | |
156 | set_buffer_uptodate(bh); | |
157 | } else { | |
70246286 | 158 | /* This happens, due to failed read-ahead attempts. */ |
1da177e4 LT |
159 | clear_buffer_uptodate(bh); |
160 | } | |
161 | unlock_buffer(bh); | |
68671f35 DM |
162 | } |
163 | ||
164 | /* | |
165 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and | |
166 | * unlock the buffer. This is what ll_rw_block uses too. | |
167 | */ | |
168 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) | |
169 | { | |
170 | __end_buffer_read_notouch(bh, uptodate); | |
1da177e4 LT |
171 | put_bh(bh); |
172 | } | |
1fe72eaa | 173 | EXPORT_SYMBOL(end_buffer_read_sync); |
1da177e4 LT |
174 | |
175 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) | |
176 | { | |
1da177e4 LT |
177 | if (uptodate) { |
178 | set_buffer_uptodate(bh); | |
179 | } else { | |
432f16e6 | 180 | buffer_io_error(bh, ", lost sync page write"); |
87354e5d | 181 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
182 | clear_buffer_uptodate(bh); |
183 | } | |
184 | unlock_buffer(bh); | |
185 | put_bh(bh); | |
186 | } | |
1fe72eaa | 187 | EXPORT_SYMBOL(end_buffer_write_sync); |
1da177e4 | 188 | |
1da177e4 LT |
189 | /* |
190 | * Various filesystems appear to want __find_get_block to be non-blocking. | |
191 | * But it's the page lock which protects the buffers. To get around this, | |
192 | * we get exclusion from try_to_free_buffers with the blockdev mapping's | |
193 | * private_lock. | |
194 | * | |
195 | * Hack idea: for the blockdev mapping, i_bufferlist_lock contention | |
196 | * may be quite high. This code could TryLock the page, and if that | |
197 | * succeeds, there is no need to take private_lock. (But if | |
198 | * private_lock is contended then so is mapping->tree_lock). | |
199 | */ | |
200 | static struct buffer_head * | |
385fd4c5 | 201 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
1da177e4 LT |
202 | { |
203 | struct inode *bd_inode = bdev->bd_inode; | |
204 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
205 | struct buffer_head *ret = NULL; | |
206 | pgoff_t index; | |
207 | struct buffer_head *bh; | |
208 | struct buffer_head *head; | |
209 | struct page *page; | |
210 | int all_mapped = 1; | |
5bdd3eb8 | 211 | static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); |
1da177e4 | 212 | |
09cbfeaf | 213 | index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); |
2457aec6 | 214 | page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED); |
1da177e4 LT |
215 | if (!page) |
216 | goto out; | |
217 | ||
218 | spin_lock(&bd_mapping->private_lock); | |
219 | if (!page_has_buffers(page)) | |
220 | goto out_unlock; | |
221 | head = page_buffers(page); | |
222 | bh = head; | |
223 | do { | |
97f76d3d NK |
224 | if (!buffer_mapped(bh)) |
225 | all_mapped = 0; | |
226 | else if (bh->b_blocknr == block) { | |
1da177e4 LT |
227 | ret = bh; |
228 | get_bh(bh); | |
229 | goto out_unlock; | |
230 | } | |
1da177e4 LT |
231 | bh = bh->b_this_page; |
232 | } while (bh != head); | |
233 | ||
234 | /* we might be here because some of the buffers on this page are | |
235 | * not mapped. This is due to various races between | |
236 | * file io on the block device and getblk. It gets dealt with | |
237 | * elsewhere, don't buffer_error if we had some unmapped buffers | |
238 | */ | |
5bdd3eb8 TH |
239 | ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE); |
240 | if (all_mapped && __ratelimit(&last_warned)) { | |
241 | printk("__find_get_block_slow() failed. block=%llu, " | |
242 | "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, " | |
243 | "device %pg blocksize: %d\n", | |
244 | (unsigned long long)block, | |
245 | (unsigned long long)bh->b_blocknr, | |
246 | bh->b_state, bh->b_size, bdev, | |
247 | 1 << bd_inode->i_blkbits); | |
1da177e4 LT |
248 | } |
249 | out_unlock: | |
250 | spin_unlock(&bd_mapping->private_lock); | |
09cbfeaf | 251 | put_page(page); |
1da177e4 LT |
252 | out: |
253 | return ret; | |
254 | } | |
255 | ||
1da177e4 LT |
256 | /* |
257 | * I/O completion handler for block_read_full_page() - pages | |
258 | * which come unlocked at the end of I/O. | |
259 | */ | |
260 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) | |
261 | { | |
1da177e4 | 262 | unsigned long flags; |
a3972203 | 263 | struct buffer_head *first; |
1da177e4 LT |
264 | struct buffer_head *tmp; |
265 | struct page *page; | |
266 | int page_uptodate = 1; | |
267 | ||
268 | BUG_ON(!buffer_async_read(bh)); | |
269 | ||
270 | page = bh->b_page; | |
271 | if (uptodate) { | |
272 | set_buffer_uptodate(bh); | |
273 | } else { | |
274 | clear_buffer_uptodate(bh); | |
432f16e6 | 275 | buffer_io_error(bh, ", async page read"); |
1da177e4 LT |
276 | SetPageError(page); |
277 | } | |
278 | ||
279 | /* | |
280 | * Be _very_ careful from here on. Bad things can happen if | |
281 | * two buffer heads end IO at almost the same time and both | |
282 | * decide that the page is now completely done. | |
283 | */ | |
a3972203 NP |
284 | first = page_buffers(page); |
285 | local_irq_save(flags); | |
286 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
1da177e4 LT |
287 | clear_buffer_async_read(bh); |
288 | unlock_buffer(bh); | |
289 | tmp = bh; | |
290 | do { | |
291 | if (!buffer_uptodate(tmp)) | |
292 | page_uptodate = 0; | |
293 | if (buffer_async_read(tmp)) { | |
294 | BUG_ON(!buffer_locked(tmp)); | |
295 | goto still_busy; | |
296 | } | |
297 | tmp = tmp->b_this_page; | |
298 | } while (tmp != bh); | |
a3972203 NP |
299 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
300 | local_irq_restore(flags); | |
1da177e4 LT |
301 | |
302 | /* | |
303 | * If none of the buffers had errors and they are all | |
304 | * uptodate then we can set the page uptodate. | |
305 | */ | |
306 | if (page_uptodate && !PageError(page)) | |
307 | SetPageUptodate(page); | |
308 | unlock_page(page); | |
309 | return; | |
310 | ||
311 | still_busy: | |
a3972203 NP |
312 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
313 | local_irq_restore(flags); | |
1da177e4 LT |
314 | return; |
315 | } | |
316 | ||
317 | /* | |
318 | * Completion handler for block_write_full_page() - pages which are unlocked | |
319 | * during I/O, and which have PageWriteback cleared upon I/O completion. | |
320 | */ | |
35c80d5f | 321 | void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
1da177e4 | 322 | { |
1da177e4 | 323 | unsigned long flags; |
a3972203 | 324 | struct buffer_head *first; |
1da177e4 LT |
325 | struct buffer_head *tmp; |
326 | struct page *page; | |
327 | ||
328 | BUG_ON(!buffer_async_write(bh)); | |
329 | ||
330 | page = bh->b_page; | |
331 | if (uptodate) { | |
332 | set_buffer_uptodate(bh); | |
333 | } else { | |
432f16e6 | 334 | buffer_io_error(bh, ", lost async page write"); |
87354e5d | 335 | mark_buffer_write_io_error(bh); |
1da177e4 LT |
336 | clear_buffer_uptodate(bh); |
337 | SetPageError(page); | |
338 | } | |
339 | ||
a3972203 NP |
340 | first = page_buffers(page); |
341 | local_irq_save(flags); | |
342 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
343 | ||
1da177e4 LT |
344 | clear_buffer_async_write(bh); |
345 | unlock_buffer(bh); | |
346 | tmp = bh->b_this_page; | |
347 | while (tmp != bh) { | |
348 | if (buffer_async_write(tmp)) { | |
349 | BUG_ON(!buffer_locked(tmp)); | |
350 | goto still_busy; | |
351 | } | |
352 | tmp = tmp->b_this_page; | |
353 | } | |
a3972203 NP |
354 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
355 | local_irq_restore(flags); | |
1da177e4 LT |
356 | end_page_writeback(page); |
357 | return; | |
358 | ||
359 | still_busy: | |
a3972203 NP |
360 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
361 | local_irq_restore(flags); | |
1da177e4 LT |
362 | return; |
363 | } | |
1fe72eaa | 364 | EXPORT_SYMBOL(end_buffer_async_write); |
1da177e4 LT |
365 | |
366 | /* | |
367 | * If a page's buffers are under async readin (end_buffer_async_read | |
368 | * completion) then there is a possibility that another thread of | |
369 | * control could lock one of the buffers after it has completed | |
370 | * but while some of the other buffers have not completed. This | |
371 | * locked buffer would confuse end_buffer_async_read() into not unlocking | |
372 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() | |
373 | * that this buffer is not under async I/O. | |
374 | * | |
375 | * The page comes unlocked when it has no locked buffer_async buffers | |
376 | * left. | |
377 | * | |
378 | * PageLocked prevents anyone starting new async I/O reads any of | |
379 | * the buffers. | |
380 | * | |
381 | * PageWriteback is used to prevent simultaneous writeout of the same | |
382 | * page. | |
383 | * | |
384 | * PageLocked prevents anyone from starting writeback of a page which is | |
385 | * under read I/O (PageWriteback is only ever set against a locked page). | |
386 | */ | |
387 | static void mark_buffer_async_read(struct buffer_head *bh) | |
388 | { | |
389 | bh->b_end_io = end_buffer_async_read; | |
390 | set_buffer_async_read(bh); | |
391 | } | |
392 | ||
1fe72eaa HS |
393 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
394 | bh_end_io_t *handler) | |
1da177e4 | 395 | { |
35c80d5f | 396 | bh->b_end_io = handler; |
1da177e4 LT |
397 | set_buffer_async_write(bh); |
398 | } | |
35c80d5f CM |
399 | |
400 | void mark_buffer_async_write(struct buffer_head *bh) | |
401 | { | |
402 | mark_buffer_async_write_endio(bh, end_buffer_async_write); | |
403 | } | |
1da177e4 LT |
404 | EXPORT_SYMBOL(mark_buffer_async_write); |
405 | ||
406 | ||
407 | /* | |
408 | * fs/buffer.c contains helper functions for buffer-backed address space's | |
409 | * fsync functions. A common requirement for buffer-based filesystems is | |
410 | * that certain data from the backing blockdev needs to be written out for | |
411 | * a successful fsync(). For example, ext2 indirect blocks need to be | |
412 | * written back and waited upon before fsync() returns. | |
413 | * | |
414 | * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), | |
415 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the | |
416 | * management of a list of dependent buffers at ->i_mapping->private_list. | |
417 | * | |
418 | * Locking is a little subtle: try_to_free_buffers() will remove buffers | |
419 | * from their controlling inode's queue when they are being freed. But | |
420 | * try_to_free_buffers() will be operating against the *blockdev* mapping | |
421 | * at the time, not against the S_ISREG file which depends on those buffers. | |
422 | * So the locking for private_list is via the private_lock in the address_space | |
423 | * which backs the buffers. Which is different from the address_space | |
424 | * against which the buffers are listed. So for a particular address_space, | |
425 | * mapping->private_lock does *not* protect mapping->private_list! In fact, | |
426 | * mapping->private_list will always be protected by the backing blockdev's | |
427 | * ->private_lock. | |
428 | * | |
429 | * Which introduces a requirement: all buffers on an address_space's | |
430 | * ->private_list must be from the same address_space: the blockdev's. | |
431 | * | |
432 | * address_spaces which do not place buffers at ->private_list via these | |
433 | * utility functions are free to use private_lock and private_list for | |
434 | * whatever they want. The only requirement is that list_empty(private_list) | |
435 | * be true at clear_inode() time. | |
436 | * | |
437 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The | |
438 | * filesystems should do that. invalidate_inode_buffers() should just go | |
439 | * BUG_ON(!list_empty). | |
440 | * | |
441 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should | |
442 | * take an address_space, not an inode. And it should be called | |
443 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being | |
444 | * queued up. | |
445 | * | |
446 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the | |
447 | * list if it is already on a list. Because if the buffer is on a list, | |
448 | * it *must* already be on the right one. If not, the filesystem is being | |
449 | * silly. This will save a ton of locking. But first we have to ensure | |
450 | * that buffers are taken *off* the old inode's list when they are freed | |
451 | * (presumably in truncate). That requires careful auditing of all | |
452 | * filesystems (do it inside bforget()). It could also be done by bringing | |
453 | * b_inode back. | |
454 | */ | |
455 | ||
456 | /* | |
457 | * The buffer's backing address_space's private_lock must be held | |
458 | */ | |
dbacefc9 | 459 | static void __remove_assoc_queue(struct buffer_head *bh) |
1da177e4 LT |
460 | { |
461 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 462 | WARN_ON(!bh->b_assoc_map); |
58ff407b | 463 | bh->b_assoc_map = NULL; |
1da177e4 LT |
464 | } |
465 | ||
466 | int inode_has_buffers(struct inode *inode) | |
467 | { | |
468 | return !list_empty(&inode->i_data.private_list); | |
469 | } | |
470 | ||
471 | /* | |
472 | * osync is designed to support O_SYNC io. It waits synchronously for | |
473 | * all already-submitted IO to complete, but does not queue any new | |
474 | * writes to the disk. | |
475 | * | |
476 | * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as | |
477 | * you dirty the buffers, and then use osync_inode_buffers to wait for | |
478 | * completion. Any other dirty buffers which are not yet queued for | |
479 | * write will not be flushed to disk by the osync. | |
480 | */ | |
481 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) | |
482 | { | |
483 | struct buffer_head *bh; | |
484 | struct list_head *p; | |
485 | int err = 0; | |
486 | ||
487 | spin_lock(lock); | |
488 | repeat: | |
489 | list_for_each_prev(p, list) { | |
490 | bh = BH_ENTRY(p); | |
491 | if (buffer_locked(bh)) { | |
492 | get_bh(bh); | |
493 | spin_unlock(lock); | |
494 | wait_on_buffer(bh); | |
495 | if (!buffer_uptodate(bh)) | |
496 | err = -EIO; | |
497 | brelse(bh); | |
498 | spin_lock(lock); | |
499 | goto repeat; | |
500 | } | |
501 | } | |
502 | spin_unlock(lock); | |
503 | return err; | |
504 | } | |
505 | ||
01a05b33 | 506 | static void do_thaw_one(struct super_block *sb, void *unused) |
c2d75438 | 507 | { |
01a05b33 | 508 | while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) |
a1c6f057 | 509 | printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev); |
01a05b33 | 510 | } |
c2d75438 | 511 | |
01a05b33 AV |
512 | static void do_thaw_all(struct work_struct *work) |
513 | { | |
514 | iterate_supers(do_thaw_one, NULL); | |
053c525f | 515 | kfree(work); |
c2d75438 ES |
516 | printk(KERN_WARNING "Emergency Thaw complete\n"); |
517 | } | |
518 | ||
519 | /** | |
520 | * emergency_thaw_all -- forcibly thaw every frozen filesystem | |
521 | * | |
522 | * Used for emergency unfreeze of all filesystems via SysRq | |
523 | */ | |
524 | void emergency_thaw_all(void) | |
525 | { | |
053c525f JA |
526 | struct work_struct *work; |
527 | ||
528 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
529 | if (work) { | |
530 | INIT_WORK(work, do_thaw_all); | |
531 | schedule_work(work); | |
532 | } | |
c2d75438 ES |
533 | } |
534 | ||
1da177e4 | 535 | /** |
78a4a50a | 536 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
67be2dd1 | 537 | * @mapping: the mapping which wants those buffers written |
1da177e4 LT |
538 | * |
539 | * Starts I/O against the buffers at mapping->private_list, and waits upon | |
540 | * that I/O. | |
541 | * | |
67be2dd1 MW |
542 | * Basically, this is a convenience function for fsync(). |
543 | * @mapping is a file or directory which needs those buffers to be written for | |
544 | * a successful fsync(). | |
1da177e4 LT |
545 | */ |
546 | int sync_mapping_buffers(struct address_space *mapping) | |
547 | { | |
252aa6f5 | 548 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
549 | |
550 | if (buffer_mapping == NULL || list_empty(&mapping->private_list)) | |
551 | return 0; | |
552 | ||
553 | return fsync_buffers_list(&buffer_mapping->private_lock, | |
554 | &mapping->private_list); | |
555 | } | |
556 | EXPORT_SYMBOL(sync_mapping_buffers); | |
557 | ||
558 | /* | |
559 | * Called when we've recently written block `bblock', and it is known that | |
560 | * `bblock' was for a buffer_boundary() buffer. This means that the block at | |
561 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's | |
562 | * dirty, schedule it for IO. So that indirects merge nicely with their data. | |
563 | */ | |
564 | void write_boundary_block(struct block_device *bdev, | |
565 | sector_t bblock, unsigned blocksize) | |
566 | { | |
567 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); | |
568 | if (bh) { | |
569 | if (buffer_dirty(bh)) | |
dfec8a14 | 570 | ll_rw_block(REQ_OP_WRITE, 0, 1, &bh); |
1da177e4 LT |
571 | put_bh(bh); |
572 | } | |
573 | } | |
574 | ||
575 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) | |
576 | { | |
577 | struct address_space *mapping = inode->i_mapping; | |
578 | struct address_space *buffer_mapping = bh->b_page->mapping; | |
579 | ||
580 | mark_buffer_dirty(bh); | |
252aa6f5 RA |
581 | if (!mapping->private_data) { |
582 | mapping->private_data = buffer_mapping; | |
1da177e4 | 583 | } else { |
252aa6f5 | 584 | BUG_ON(mapping->private_data != buffer_mapping); |
1da177e4 | 585 | } |
535ee2fb | 586 | if (!bh->b_assoc_map) { |
1da177e4 LT |
587 | spin_lock(&buffer_mapping->private_lock); |
588 | list_move_tail(&bh->b_assoc_buffers, | |
589 | &mapping->private_list); | |
58ff407b | 590 | bh->b_assoc_map = mapping; |
1da177e4 LT |
591 | spin_unlock(&buffer_mapping->private_lock); |
592 | } | |
593 | } | |
594 | EXPORT_SYMBOL(mark_buffer_dirty_inode); | |
595 | ||
787d2214 NP |
596 | /* |
597 | * Mark the page dirty, and set it dirty in the radix tree, and mark the inode | |
598 | * dirty. | |
599 | * | |
600 | * If warn is true, then emit a warning if the page is not uptodate and has | |
601 | * not been truncated. | |
c4843a75 | 602 | * |
81f8c3a4 | 603 | * The caller must hold lock_page_memcg(). |
787d2214 | 604 | */ |
c4843a75 | 605 | static void __set_page_dirty(struct page *page, struct address_space *mapping, |
62cccb8c | 606 | int warn) |
787d2214 | 607 | { |
227d53b3 KM |
608 | unsigned long flags; |
609 | ||
610 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
787d2214 NP |
611 | if (page->mapping) { /* Race with truncate? */ |
612 | WARN_ON_ONCE(warn && !PageUptodate(page)); | |
62cccb8c | 613 | account_page_dirtied(page, mapping); |
787d2214 NP |
614 | radix_tree_tag_set(&mapping->page_tree, |
615 | page_index(page), PAGECACHE_TAG_DIRTY); | |
616 | } | |
227d53b3 | 617 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
787d2214 NP |
618 | } |
619 | ||
1da177e4 LT |
620 | /* |
621 | * Add a page to the dirty page list. | |
622 | * | |
623 | * It is a sad fact of life that this function is called from several places | |
624 | * deeply under spinlocking. It may not sleep. | |
625 | * | |
626 | * If the page has buffers, the uptodate buffers are set dirty, to preserve | |
627 | * dirty-state coherency between the page and the buffers. It the page does | |
628 | * not have buffers then when they are later attached they will all be set | |
629 | * dirty. | |
630 | * | |
631 | * The buffers are dirtied before the page is dirtied. There's a small race | |
632 | * window in which a writepage caller may see the page cleanness but not the | |
633 | * buffer dirtiness. That's fine. If this code were to set the page dirty | |
634 | * before the buffers, a concurrent writepage caller could clear the page dirty | |
635 | * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean | |
636 | * page on the dirty page list. | |
637 | * | |
638 | * We use private_lock to lock against try_to_free_buffers while using the | |
639 | * page's buffer list. Also use this to protect against clean buffers being | |
640 | * added to the page after it was set dirty. | |
641 | * | |
642 | * FIXME: may need to call ->reservepage here as well. That's rather up to the | |
643 | * address_space though. | |
644 | */ | |
645 | int __set_page_dirty_buffers(struct page *page) | |
646 | { | |
a8e7d49a | 647 | int newly_dirty; |
787d2214 | 648 | struct address_space *mapping = page_mapping(page); |
ebf7a227 NP |
649 | |
650 | if (unlikely(!mapping)) | |
651 | return !TestSetPageDirty(page); | |
1da177e4 LT |
652 | |
653 | spin_lock(&mapping->private_lock); | |
654 | if (page_has_buffers(page)) { | |
655 | struct buffer_head *head = page_buffers(page); | |
656 | struct buffer_head *bh = head; | |
657 | ||
658 | do { | |
659 | set_buffer_dirty(bh); | |
660 | bh = bh->b_this_page; | |
661 | } while (bh != head); | |
662 | } | |
c4843a75 | 663 | /* |
81f8c3a4 JW |
664 | * Lock out page->mem_cgroup migration to keep PageDirty |
665 | * synchronized with per-memcg dirty page counters. | |
c4843a75 | 666 | */ |
62cccb8c | 667 | lock_page_memcg(page); |
a8e7d49a | 668 | newly_dirty = !TestSetPageDirty(page); |
1da177e4 LT |
669 | spin_unlock(&mapping->private_lock); |
670 | ||
a8e7d49a | 671 | if (newly_dirty) |
62cccb8c | 672 | __set_page_dirty(page, mapping, 1); |
c4843a75 | 673 | |
62cccb8c | 674 | unlock_page_memcg(page); |
c4843a75 GT |
675 | |
676 | if (newly_dirty) | |
677 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
678 | ||
a8e7d49a | 679 | return newly_dirty; |
1da177e4 LT |
680 | } |
681 | EXPORT_SYMBOL(__set_page_dirty_buffers); | |
682 | ||
683 | /* | |
684 | * Write out and wait upon a list of buffers. | |
685 | * | |
686 | * We have conflicting pressures: we want to make sure that all | |
687 | * initially dirty buffers get waited on, but that any subsequently | |
688 | * dirtied buffers don't. After all, we don't want fsync to last | |
689 | * forever if somebody is actively writing to the file. | |
690 | * | |
691 | * Do this in two main stages: first we copy dirty buffers to a | |
692 | * temporary inode list, queueing the writes as we go. Then we clean | |
693 | * up, waiting for those writes to complete. | |
694 | * | |
695 | * During this second stage, any subsequent updates to the file may end | |
696 | * up refiling the buffer on the original inode's dirty list again, so | |
697 | * there is a chance we will end up with a buffer queued for write but | |
698 | * not yet completed on that list. So, as a final cleanup we go through | |
699 | * the osync code to catch these locked, dirty buffers without requeuing | |
700 | * any newly dirty buffers for write. | |
701 | */ | |
702 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) | |
703 | { | |
704 | struct buffer_head *bh; | |
705 | struct list_head tmp; | |
7eaceacc | 706 | struct address_space *mapping; |
1da177e4 | 707 | int err = 0, err2; |
4ee2491e | 708 | struct blk_plug plug; |
1da177e4 LT |
709 | |
710 | INIT_LIST_HEAD(&tmp); | |
4ee2491e | 711 | blk_start_plug(&plug); |
1da177e4 LT |
712 | |
713 | spin_lock(lock); | |
714 | while (!list_empty(list)) { | |
715 | bh = BH_ENTRY(list->next); | |
535ee2fb | 716 | mapping = bh->b_assoc_map; |
58ff407b | 717 | __remove_assoc_queue(bh); |
535ee2fb JK |
718 | /* Avoid race with mark_buffer_dirty_inode() which does |
719 | * a lockless check and we rely on seeing the dirty bit */ | |
720 | smp_mb(); | |
1da177e4 LT |
721 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
722 | list_add(&bh->b_assoc_buffers, &tmp); | |
535ee2fb | 723 | bh->b_assoc_map = mapping; |
1da177e4 LT |
724 | if (buffer_dirty(bh)) { |
725 | get_bh(bh); | |
726 | spin_unlock(lock); | |
727 | /* | |
728 | * Ensure any pending I/O completes so that | |
9cb569d6 CH |
729 | * write_dirty_buffer() actually writes the |
730 | * current contents - it is a noop if I/O is | |
731 | * still in flight on potentially older | |
732 | * contents. | |
1da177e4 | 733 | */ |
70fd7614 | 734 | write_dirty_buffer(bh, REQ_SYNC); |
9cf6b720 JA |
735 | |
736 | /* | |
737 | * Kick off IO for the previous mapping. Note | |
738 | * that we will not run the very last mapping, | |
739 | * wait_on_buffer() will do that for us | |
740 | * through sync_buffer(). | |
741 | */ | |
1da177e4 LT |
742 | brelse(bh); |
743 | spin_lock(lock); | |
744 | } | |
745 | } | |
746 | } | |
747 | ||
4ee2491e JA |
748 | spin_unlock(lock); |
749 | blk_finish_plug(&plug); | |
750 | spin_lock(lock); | |
751 | ||
1da177e4 LT |
752 | while (!list_empty(&tmp)) { |
753 | bh = BH_ENTRY(tmp.prev); | |
1da177e4 | 754 | get_bh(bh); |
535ee2fb JK |
755 | mapping = bh->b_assoc_map; |
756 | __remove_assoc_queue(bh); | |
757 | /* Avoid race with mark_buffer_dirty_inode() which does | |
758 | * a lockless check and we rely on seeing the dirty bit */ | |
759 | smp_mb(); | |
760 | if (buffer_dirty(bh)) { | |
761 | list_add(&bh->b_assoc_buffers, | |
e3892296 | 762 | &mapping->private_list); |
535ee2fb JK |
763 | bh->b_assoc_map = mapping; |
764 | } | |
1da177e4 LT |
765 | spin_unlock(lock); |
766 | wait_on_buffer(bh); | |
767 | if (!buffer_uptodate(bh)) | |
768 | err = -EIO; | |
769 | brelse(bh); | |
770 | spin_lock(lock); | |
771 | } | |
772 | ||
773 | spin_unlock(lock); | |
774 | err2 = osync_buffers_list(lock, list); | |
775 | if (err) | |
776 | return err; | |
777 | else | |
778 | return err2; | |
779 | } | |
780 | ||
781 | /* | |
782 | * Invalidate any and all dirty buffers on a given inode. We are | |
783 | * probably unmounting the fs, but that doesn't mean we have already | |
784 | * done a sync(). Just drop the buffers from the inode list. | |
785 | * | |
786 | * NOTE: we take the inode's blockdev's mapping's private_lock. Which | |
787 | * assumes that all the buffers are against the blockdev. Not true | |
788 | * for reiserfs. | |
789 | */ | |
790 | void invalidate_inode_buffers(struct inode *inode) | |
791 | { | |
792 | if (inode_has_buffers(inode)) { | |
793 | struct address_space *mapping = &inode->i_data; | |
794 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 795 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
796 | |
797 | spin_lock(&buffer_mapping->private_lock); | |
798 | while (!list_empty(list)) | |
799 | __remove_assoc_queue(BH_ENTRY(list->next)); | |
800 | spin_unlock(&buffer_mapping->private_lock); | |
801 | } | |
802 | } | |
52b19ac9 | 803 | EXPORT_SYMBOL(invalidate_inode_buffers); |
1da177e4 LT |
804 | |
805 | /* | |
806 | * Remove any clean buffers from the inode's buffer list. This is called | |
807 | * when we're trying to free the inode itself. Those buffers can pin it. | |
808 | * | |
809 | * Returns true if all buffers were removed. | |
810 | */ | |
811 | int remove_inode_buffers(struct inode *inode) | |
812 | { | |
813 | int ret = 1; | |
814 | ||
815 | if (inode_has_buffers(inode)) { | |
816 | struct address_space *mapping = &inode->i_data; | |
817 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 818 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
819 | |
820 | spin_lock(&buffer_mapping->private_lock); | |
821 | while (!list_empty(list)) { | |
822 | struct buffer_head *bh = BH_ENTRY(list->next); | |
823 | if (buffer_dirty(bh)) { | |
824 | ret = 0; | |
825 | break; | |
826 | } | |
827 | __remove_assoc_queue(bh); | |
828 | } | |
829 | spin_unlock(&buffer_mapping->private_lock); | |
830 | } | |
831 | return ret; | |
832 | } | |
833 | ||
834 | /* | |
835 | * Create the appropriate buffers when given a page for data area and | |
836 | * the size of each buffer.. Use the bh->b_this_page linked list to | |
837 | * follow the buffers created. Return NULL if unable to create more | |
838 | * buffers. | |
839 | * | |
840 | * The retry flag is used to differentiate async IO (paging, swapping) | |
841 | * which may not fail from ordinary buffer allocations. | |
842 | */ | |
843 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, | |
640ab98f | 844 | bool retry) |
1da177e4 LT |
845 | { |
846 | struct buffer_head *bh, *head; | |
640ab98f | 847 | gfp_t gfp = GFP_NOFS; |
1da177e4 LT |
848 | long offset; |
849 | ||
640ab98f JA |
850 | if (retry) |
851 | gfp |= __GFP_NOFAIL; | |
852 | ||
1da177e4 LT |
853 | head = NULL; |
854 | offset = PAGE_SIZE; | |
855 | while ((offset -= size) >= 0) { | |
640ab98f | 856 | bh = alloc_buffer_head(gfp); |
1da177e4 LT |
857 | if (!bh) |
858 | goto no_grow; | |
859 | ||
1da177e4 LT |
860 | bh->b_this_page = head; |
861 | bh->b_blocknr = -1; | |
862 | head = bh; | |
863 | ||
1da177e4 LT |
864 | bh->b_size = size; |
865 | ||
866 | /* Link the buffer to its page */ | |
867 | set_bh_page(bh, page, offset); | |
1da177e4 LT |
868 | } |
869 | return head; | |
870 | /* | |
871 | * In case anything failed, we just free everything we got. | |
872 | */ | |
873 | no_grow: | |
874 | if (head) { | |
875 | do { | |
876 | bh = head; | |
877 | head = head->b_this_page; | |
878 | free_buffer_head(bh); | |
879 | } while (head); | |
880 | } | |
881 | ||
640ab98f | 882 | return NULL; |
1da177e4 LT |
883 | } |
884 | EXPORT_SYMBOL_GPL(alloc_page_buffers); | |
885 | ||
886 | static inline void | |
887 | link_dev_buffers(struct page *page, struct buffer_head *head) | |
888 | { | |
889 | struct buffer_head *bh, *tail; | |
890 | ||
891 | bh = head; | |
892 | do { | |
893 | tail = bh; | |
894 | bh = bh->b_this_page; | |
895 | } while (bh); | |
896 | tail->b_this_page = head; | |
897 | attach_page_buffers(page, head); | |
898 | } | |
899 | ||
bbec0270 LT |
900 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
901 | { | |
902 | sector_t retval = ~((sector_t)0); | |
903 | loff_t sz = i_size_read(bdev->bd_inode); | |
904 | ||
905 | if (sz) { | |
906 | unsigned int sizebits = blksize_bits(size); | |
907 | retval = (sz >> sizebits); | |
908 | } | |
909 | return retval; | |
910 | } | |
911 | ||
1da177e4 LT |
912 | /* |
913 | * Initialise the state of a blockdev page's buffers. | |
914 | */ | |
676ce6d5 | 915 | static sector_t |
1da177e4 LT |
916 | init_page_buffers(struct page *page, struct block_device *bdev, |
917 | sector_t block, int size) | |
918 | { | |
919 | struct buffer_head *head = page_buffers(page); | |
920 | struct buffer_head *bh = head; | |
921 | int uptodate = PageUptodate(page); | |
bbec0270 | 922 | sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); |
1da177e4 LT |
923 | |
924 | do { | |
925 | if (!buffer_mapped(bh)) { | |
926 | init_buffer(bh, NULL, NULL); | |
927 | bh->b_bdev = bdev; | |
928 | bh->b_blocknr = block; | |
929 | if (uptodate) | |
930 | set_buffer_uptodate(bh); | |
080399aa JM |
931 | if (block < end_block) |
932 | set_buffer_mapped(bh); | |
1da177e4 LT |
933 | } |
934 | block++; | |
935 | bh = bh->b_this_page; | |
936 | } while (bh != head); | |
676ce6d5 HD |
937 | |
938 | /* | |
939 | * Caller needs to validate requested block against end of device. | |
940 | */ | |
941 | return end_block; | |
1da177e4 LT |
942 | } |
943 | ||
944 | /* | |
945 | * Create the page-cache page that contains the requested block. | |
946 | * | |
676ce6d5 | 947 | * This is used purely for blockdev mappings. |
1da177e4 | 948 | */ |
676ce6d5 | 949 | static int |
1da177e4 | 950 | grow_dev_page(struct block_device *bdev, sector_t block, |
3b5e6454 | 951 | pgoff_t index, int size, int sizebits, gfp_t gfp) |
1da177e4 LT |
952 | { |
953 | struct inode *inode = bdev->bd_inode; | |
954 | struct page *page; | |
955 | struct buffer_head *bh; | |
676ce6d5 HD |
956 | sector_t end_block; |
957 | int ret = 0; /* Will call free_more_memory() */ | |
84235de3 | 958 | gfp_t gfp_mask; |
1da177e4 | 959 | |
c62d2555 | 960 | gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; |
3b5e6454 | 961 | |
84235de3 JW |
962 | /* |
963 | * XXX: __getblk_slow() can not really deal with failure and | |
964 | * will endlessly loop on improvised global reclaim. Prefer | |
965 | * looping in the allocator rather than here, at least that | |
966 | * code knows what it's doing. | |
967 | */ | |
968 | gfp_mask |= __GFP_NOFAIL; | |
969 | ||
970 | page = find_or_create_page(inode->i_mapping, index, gfp_mask); | |
1da177e4 | 971 | |
e827f923 | 972 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
973 | |
974 | if (page_has_buffers(page)) { | |
975 | bh = page_buffers(page); | |
976 | if (bh->b_size == size) { | |
676ce6d5 | 977 | end_block = init_page_buffers(page, bdev, |
f2d5a944 AA |
978 | (sector_t)index << sizebits, |
979 | size); | |
676ce6d5 | 980 | goto done; |
1da177e4 LT |
981 | } |
982 | if (!try_to_free_buffers(page)) | |
983 | goto failed; | |
984 | } | |
985 | ||
986 | /* | |
987 | * Allocate some buffers for this page | |
988 | */ | |
94dc24c0 | 989 | bh = alloc_page_buffers(page, size, true); |
1da177e4 LT |
990 | |
991 | /* | |
992 | * Link the page to the buffers and initialise them. Take the | |
993 | * lock to be atomic wrt __find_get_block(), which does not | |
994 | * run under the page lock. | |
995 | */ | |
996 | spin_lock(&inode->i_mapping->private_lock); | |
997 | link_dev_buffers(page, bh); | |
f2d5a944 AA |
998 | end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, |
999 | size); | |
1da177e4 | 1000 | spin_unlock(&inode->i_mapping->private_lock); |
676ce6d5 HD |
1001 | done: |
1002 | ret = (block < end_block) ? 1 : -ENXIO; | |
1da177e4 | 1003 | failed: |
1da177e4 | 1004 | unlock_page(page); |
09cbfeaf | 1005 | put_page(page); |
676ce6d5 | 1006 | return ret; |
1da177e4 LT |
1007 | } |
1008 | ||
1009 | /* | |
1010 | * Create buffers for the specified block device block's page. If | |
1011 | * that page was dirty, the buffers are set dirty also. | |
1da177e4 | 1012 | */ |
858119e1 | 1013 | static int |
3b5e6454 | 1014 | grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) |
1da177e4 | 1015 | { |
1da177e4 LT |
1016 | pgoff_t index; |
1017 | int sizebits; | |
1018 | ||
1019 | sizebits = -1; | |
1020 | do { | |
1021 | sizebits++; | |
1022 | } while ((size << sizebits) < PAGE_SIZE); | |
1023 | ||
1024 | index = block >> sizebits; | |
1da177e4 | 1025 | |
e5657933 AM |
1026 | /* |
1027 | * Check for a block which wants to lie outside our maximum possible | |
1028 | * pagecache index. (this comparison is done using sector_t types). | |
1029 | */ | |
1030 | if (unlikely(index != block >> sizebits)) { | |
e5657933 | 1031 | printk(KERN_ERR "%s: requested out-of-range block %llu for " |
a1c6f057 | 1032 | "device %pg\n", |
8e24eea7 | 1033 | __func__, (unsigned long long)block, |
a1c6f057 | 1034 | bdev); |
e5657933 AM |
1035 | return -EIO; |
1036 | } | |
676ce6d5 | 1037 | |
1da177e4 | 1038 | /* Create a page with the proper size buffers.. */ |
3b5e6454 | 1039 | return grow_dev_page(bdev, block, index, size, sizebits, gfp); |
1da177e4 LT |
1040 | } |
1041 | ||
0026ba40 | 1042 | static struct buffer_head * |
3b5e6454 GK |
1043 | __getblk_slow(struct block_device *bdev, sector_t block, |
1044 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1045 | { |
1046 | /* Size must be multiple of hard sectorsize */ | |
e1defc4f | 1047 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
1da177e4 LT |
1048 | (size < 512 || size > PAGE_SIZE))) { |
1049 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", | |
1050 | size); | |
e1defc4f MP |
1051 | printk(KERN_ERR "logical block size: %d\n", |
1052 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1053 | |
1054 | dump_stack(); | |
1055 | return NULL; | |
1056 | } | |
1057 | ||
676ce6d5 HD |
1058 | for (;;) { |
1059 | struct buffer_head *bh; | |
1060 | int ret; | |
1da177e4 LT |
1061 | |
1062 | bh = __find_get_block(bdev, block, size); | |
1063 | if (bh) | |
1064 | return bh; | |
676ce6d5 | 1065 | |
3b5e6454 | 1066 | ret = grow_buffers(bdev, block, size, gfp); |
676ce6d5 HD |
1067 | if (ret < 0) |
1068 | return NULL; | |
1da177e4 LT |
1069 | } |
1070 | } | |
1071 | ||
1072 | /* | |
1073 | * The relationship between dirty buffers and dirty pages: | |
1074 | * | |
1075 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and | |
1076 | * the page is tagged dirty in its radix tree. | |
1077 | * | |
1078 | * At all times, the dirtiness of the buffers represents the dirtiness of | |
1079 | * subsections of the page. If the page has buffers, the page dirty bit is | |
1080 | * merely a hint about the true dirty state. | |
1081 | * | |
1082 | * When a page is set dirty in its entirety, all its buffers are marked dirty | |
1083 | * (if the page has buffers). | |
1084 | * | |
1085 | * When a buffer is marked dirty, its page is dirtied, but the page's other | |
1086 | * buffers are not. | |
1087 | * | |
1088 | * Also. When blockdev buffers are explicitly read with bread(), they | |
1089 | * individually become uptodate. But their backing page remains not | |
1090 | * uptodate - even if all of its buffers are uptodate. A subsequent | |
1091 | * block_read_full_page() against that page will discover all the uptodate | |
1092 | * buffers, will set the page uptodate and will perform no I/O. | |
1093 | */ | |
1094 | ||
1095 | /** | |
1096 | * mark_buffer_dirty - mark a buffer_head as needing writeout | |
67be2dd1 | 1097 | * @bh: the buffer_head to mark dirty |
1da177e4 LT |
1098 | * |
1099 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set its | |
1100 | * backing page dirty, then tag the page as dirty in its address_space's radix | |
1101 | * tree and then attach the address_space's inode to its superblock's dirty | |
1102 | * inode list. | |
1103 | * | |
1104 | * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, | |
250df6ed | 1105 | * mapping->tree_lock and mapping->host->i_lock. |
1da177e4 | 1106 | */ |
fc9b52cd | 1107 | void mark_buffer_dirty(struct buffer_head *bh) |
1da177e4 | 1108 | { |
787d2214 | 1109 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
1be62dc1 | 1110 | |
5305cb83 TH |
1111 | trace_block_dirty_buffer(bh); |
1112 | ||
1be62dc1 LT |
1113 | /* |
1114 | * Very *carefully* optimize the it-is-already-dirty case. | |
1115 | * | |
1116 | * Don't let the final "is it dirty" escape to before we | |
1117 | * perhaps modified the buffer. | |
1118 | */ | |
1119 | if (buffer_dirty(bh)) { | |
1120 | smp_mb(); | |
1121 | if (buffer_dirty(bh)) | |
1122 | return; | |
1123 | } | |
1124 | ||
a8e7d49a LT |
1125 | if (!test_set_buffer_dirty(bh)) { |
1126 | struct page *page = bh->b_page; | |
c4843a75 | 1127 | struct address_space *mapping = NULL; |
c4843a75 | 1128 | |
62cccb8c | 1129 | lock_page_memcg(page); |
8e9d78ed | 1130 | if (!TestSetPageDirty(page)) { |
c4843a75 | 1131 | mapping = page_mapping(page); |
8e9d78ed | 1132 | if (mapping) |
62cccb8c | 1133 | __set_page_dirty(page, mapping, 0); |
8e9d78ed | 1134 | } |
62cccb8c | 1135 | unlock_page_memcg(page); |
c4843a75 GT |
1136 | if (mapping) |
1137 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
a8e7d49a | 1138 | } |
1da177e4 | 1139 | } |
1fe72eaa | 1140 | EXPORT_SYMBOL(mark_buffer_dirty); |
1da177e4 | 1141 | |
87354e5d JL |
1142 | void mark_buffer_write_io_error(struct buffer_head *bh) |
1143 | { | |
1144 | set_buffer_write_io_error(bh); | |
1145 | /* FIXME: do we need to set this in both places? */ | |
1146 | if (bh->b_page && bh->b_page->mapping) | |
1147 | mapping_set_error(bh->b_page->mapping, -EIO); | |
1148 | if (bh->b_assoc_map) | |
1149 | mapping_set_error(bh->b_assoc_map, -EIO); | |
1150 | } | |
1151 | EXPORT_SYMBOL(mark_buffer_write_io_error); | |
1152 | ||
1da177e4 LT |
1153 | /* |
1154 | * Decrement a buffer_head's reference count. If all buffers against a page | |
1155 | * have zero reference count, are clean and unlocked, and if the page is clean | |
1156 | * and unlocked then try_to_free_buffers() may strip the buffers from the page | |
1157 | * in preparation for freeing it (sometimes, rarely, buffers are removed from | |
1158 | * a page but it ends up not being freed, and buffers may later be reattached). | |
1159 | */ | |
1160 | void __brelse(struct buffer_head * buf) | |
1161 | { | |
1162 | if (atomic_read(&buf->b_count)) { | |
1163 | put_bh(buf); | |
1164 | return; | |
1165 | } | |
5c752ad9 | 1166 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
1da177e4 | 1167 | } |
1fe72eaa | 1168 | EXPORT_SYMBOL(__brelse); |
1da177e4 LT |
1169 | |
1170 | /* | |
1171 | * bforget() is like brelse(), except it discards any | |
1172 | * potentially dirty data. | |
1173 | */ | |
1174 | void __bforget(struct buffer_head *bh) | |
1175 | { | |
1176 | clear_buffer_dirty(bh); | |
535ee2fb | 1177 | if (bh->b_assoc_map) { |
1da177e4 LT |
1178 | struct address_space *buffer_mapping = bh->b_page->mapping; |
1179 | ||
1180 | spin_lock(&buffer_mapping->private_lock); | |
1181 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 1182 | bh->b_assoc_map = NULL; |
1da177e4 LT |
1183 | spin_unlock(&buffer_mapping->private_lock); |
1184 | } | |
1185 | __brelse(bh); | |
1186 | } | |
1fe72eaa | 1187 | EXPORT_SYMBOL(__bforget); |
1da177e4 LT |
1188 | |
1189 | static struct buffer_head *__bread_slow(struct buffer_head *bh) | |
1190 | { | |
1191 | lock_buffer(bh); | |
1192 | if (buffer_uptodate(bh)) { | |
1193 | unlock_buffer(bh); | |
1194 | return bh; | |
1195 | } else { | |
1196 | get_bh(bh); | |
1197 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 1198 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
1199 | wait_on_buffer(bh); |
1200 | if (buffer_uptodate(bh)) | |
1201 | return bh; | |
1202 | } | |
1203 | brelse(bh); | |
1204 | return NULL; | |
1205 | } | |
1206 | ||
1207 | /* | |
1208 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). | |
1209 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their | |
1210 | * refcount elevated by one when they're in an LRU. A buffer can only appear | |
1211 | * once in a particular CPU's LRU. A single buffer can be present in multiple | |
1212 | * CPU's LRUs at the same time. | |
1213 | * | |
1214 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and | |
1215 | * sb_find_get_block(). | |
1216 | * | |
1217 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use | |
1218 | * a local interrupt disable for that. | |
1219 | */ | |
1220 | ||
86cf78d7 | 1221 | #define BH_LRU_SIZE 16 |
1da177e4 LT |
1222 | |
1223 | struct bh_lru { | |
1224 | struct buffer_head *bhs[BH_LRU_SIZE]; | |
1225 | }; | |
1226 | ||
1227 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; | |
1228 | ||
1229 | #ifdef CONFIG_SMP | |
1230 | #define bh_lru_lock() local_irq_disable() | |
1231 | #define bh_lru_unlock() local_irq_enable() | |
1232 | #else | |
1233 | #define bh_lru_lock() preempt_disable() | |
1234 | #define bh_lru_unlock() preempt_enable() | |
1235 | #endif | |
1236 | ||
1237 | static inline void check_irqs_on(void) | |
1238 | { | |
1239 | #ifdef irqs_disabled | |
1240 | BUG_ON(irqs_disabled()); | |
1241 | #endif | |
1242 | } | |
1243 | ||
1244 | /* | |
241f01fb EB |
1245 | * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is |
1246 | * inserted at the front, and the buffer_head at the back if any is evicted. | |
1247 | * Or, if already in the LRU it is moved to the front. | |
1da177e4 LT |
1248 | */ |
1249 | static void bh_lru_install(struct buffer_head *bh) | |
1250 | { | |
241f01fb EB |
1251 | struct buffer_head *evictee = bh; |
1252 | struct bh_lru *b; | |
1253 | int i; | |
1da177e4 LT |
1254 | |
1255 | check_irqs_on(); | |
1256 | bh_lru_lock(); | |
1da177e4 | 1257 | |
241f01fb EB |
1258 | b = this_cpu_ptr(&bh_lrus); |
1259 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1260 | swap(evictee, b->bhs[i]); | |
1261 | if (evictee == bh) { | |
1262 | bh_lru_unlock(); | |
1263 | return; | |
1da177e4 | 1264 | } |
1da177e4 | 1265 | } |
1da177e4 | 1266 | |
241f01fb EB |
1267 | get_bh(bh); |
1268 | bh_lru_unlock(); | |
1269 | brelse(evictee); | |
1da177e4 LT |
1270 | } |
1271 | ||
1272 | /* | |
1273 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. | |
1274 | */ | |
858119e1 | 1275 | static struct buffer_head * |
3991d3bd | 1276 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1277 | { |
1278 | struct buffer_head *ret = NULL; | |
3991d3bd | 1279 | unsigned int i; |
1da177e4 LT |
1280 | |
1281 | check_irqs_on(); | |
1282 | bh_lru_lock(); | |
1da177e4 | 1283 | for (i = 0; i < BH_LRU_SIZE; i++) { |
c7b92516 | 1284 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
1da177e4 | 1285 | |
9470dd5d ZB |
1286 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
1287 | bh->b_size == size) { | |
1da177e4 LT |
1288 | if (i) { |
1289 | while (i) { | |
c7b92516 CL |
1290 | __this_cpu_write(bh_lrus.bhs[i], |
1291 | __this_cpu_read(bh_lrus.bhs[i - 1])); | |
1da177e4 LT |
1292 | i--; |
1293 | } | |
c7b92516 | 1294 | __this_cpu_write(bh_lrus.bhs[0], bh); |
1da177e4 LT |
1295 | } |
1296 | get_bh(bh); | |
1297 | ret = bh; | |
1298 | break; | |
1299 | } | |
1300 | } | |
1301 | bh_lru_unlock(); | |
1302 | return ret; | |
1303 | } | |
1304 | ||
1305 | /* | |
1306 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh | |
1307 | * it in the LRU and mark it as accessed. If it is not present then return | |
1308 | * NULL | |
1309 | */ | |
1310 | struct buffer_head * | |
3991d3bd | 1311 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1312 | { |
1313 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); | |
1314 | ||
1315 | if (bh == NULL) { | |
2457aec6 | 1316 | /* __find_get_block_slow will mark the page accessed */ |
385fd4c5 | 1317 | bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1318 | if (bh) |
1319 | bh_lru_install(bh); | |
2457aec6 | 1320 | } else |
1da177e4 | 1321 | touch_buffer(bh); |
2457aec6 | 1322 | |
1da177e4 LT |
1323 | return bh; |
1324 | } | |
1325 | EXPORT_SYMBOL(__find_get_block); | |
1326 | ||
1327 | /* | |
3b5e6454 | 1328 | * __getblk_gfp() will locate (and, if necessary, create) the buffer_head |
1da177e4 LT |
1329 | * which corresponds to the passed block_device, block and size. The |
1330 | * returned buffer has its reference count incremented. | |
1331 | * | |
3b5e6454 GK |
1332 | * __getblk_gfp() will lock up the machine if grow_dev_page's |
1333 | * try_to_free_buffers() attempt is failing. FIXME, perhaps? | |
1da177e4 LT |
1334 | */ |
1335 | struct buffer_head * | |
3b5e6454 GK |
1336 | __getblk_gfp(struct block_device *bdev, sector_t block, |
1337 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1338 | { |
1339 | struct buffer_head *bh = __find_get_block(bdev, block, size); | |
1340 | ||
1341 | might_sleep(); | |
1342 | if (bh == NULL) | |
3b5e6454 | 1343 | bh = __getblk_slow(bdev, block, size, gfp); |
1da177e4 LT |
1344 | return bh; |
1345 | } | |
3b5e6454 | 1346 | EXPORT_SYMBOL(__getblk_gfp); |
1da177e4 LT |
1347 | |
1348 | /* | |
1349 | * Do async read-ahead on a buffer.. | |
1350 | */ | |
3991d3bd | 1351 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1352 | { |
1353 | struct buffer_head *bh = __getblk(bdev, block, size); | |
a3e713b5 | 1354 | if (likely(bh)) { |
70246286 | 1355 | ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh); |
a3e713b5 AM |
1356 | brelse(bh); |
1357 | } | |
1da177e4 LT |
1358 | } |
1359 | EXPORT_SYMBOL(__breadahead); | |
1360 | ||
1361 | /** | |
3b5e6454 | 1362 | * __bread_gfp() - reads a specified block and returns the bh |
67be2dd1 | 1363 | * @bdev: the block_device to read from |
1da177e4 LT |
1364 | * @block: number of block |
1365 | * @size: size (in bytes) to read | |
3b5e6454 GK |
1366 | * @gfp: page allocation flag |
1367 | * | |
1da177e4 | 1368 | * Reads a specified block, and returns buffer head that contains it. |
3b5e6454 GK |
1369 | * The page cache can be allocated from non-movable area |
1370 | * not to prevent page migration if you set gfp to zero. | |
1da177e4 LT |
1371 | * It returns NULL if the block was unreadable. |
1372 | */ | |
1373 | struct buffer_head * | |
3b5e6454 GK |
1374 | __bread_gfp(struct block_device *bdev, sector_t block, |
1375 | unsigned size, gfp_t gfp) | |
1da177e4 | 1376 | { |
3b5e6454 | 1377 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); |
1da177e4 | 1378 | |
a3e713b5 | 1379 | if (likely(bh) && !buffer_uptodate(bh)) |
1da177e4 LT |
1380 | bh = __bread_slow(bh); |
1381 | return bh; | |
1382 | } | |
3b5e6454 | 1383 | EXPORT_SYMBOL(__bread_gfp); |
1da177e4 LT |
1384 | |
1385 | /* | |
1386 | * invalidate_bh_lrus() is called rarely - but not only at unmount. | |
1387 | * This doesn't race because it runs in each cpu either in irq | |
1388 | * or with preempt disabled. | |
1389 | */ | |
1390 | static void invalidate_bh_lru(void *arg) | |
1391 | { | |
1392 | struct bh_lru *b = &get_cpu_var(bh_lrus); | |
1393 | int i; | |
1394 | ||
1395 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1396 | brelse(b->bhs[i]); | |
1397 | b->bhs[i] = NULL; | |
1398 | } | |
1399 | put_cpu_var(bh_lrus); | |
1400 | } | |
42be35d0 GBY |
1401 | |
1402 | static bool has_bh_in_lru(int cpu, void *dummy) | |
1403 | { | |
1404 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); | |
1405 | int i; | |
1da177e4 | 1406 | |
42be35d0 GBY |
1407 | for (i = 0; i < BH_LRU_SIZE; i++) { |
1408 | if (b->bhs[i]) | |
1409 | return 1; | |
1410 | } | |
1411 | ||
1412 | return 0; | |
1413 | } | |
1414 | ||
f9a14399 | 1415 | void invalidate_bh_lrus(void) |
1da177e4 | 1416 | { |
42be35d0 | 1417 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); |
1da177e4 | 1418 | } |
9db5579b | 1419 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
1da177e4 LT |
1420 | |
1421 | void set_bh_page(struct buffer_head *bh, | |
1422 | struct page *page, unsigned long offset) | |
1423 | { | |
1424 | bh->b_page = page; | |
e827f923 | 1425 | BUG_ON(offset >= PAGE_SIZE); |
1da177e4 LT |
1426 | if (PageHighMem(page)) |
1427 | /* | |
1428 | * This catches illegal uses and preserves the offset: | |
1429 | */ | |
1430 | bh->b_data = (char *)(0 + offset); | |
1431 | else | |
1432 | bh->b_data = page_address(page) + offset; | |
1433 | } | |
1434 | EXPORT_SYMBOL(set_bh_page); | |
1435 | ||
1436 | /* | |
1437 | * Called when truncating a buffer on a page completely. | |
1438 | */ | |
e7470ee8 MG |
1439 | |
1440 | /* Bits that are cleared during an invalidate */ | |
1441 | #define BUFFER_FLAGS_DISCARD \ | |
1442 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ | |
1443 | 1 << BH_Delay | 1 << BH_Unwritten) | |
1444 | ||
858119e1 | 1445 | static void discard_buffer(struct buffer_head * bh) |
1da177e4 | 1446 | { |
e7470ee8 MG |
1447 | unsigned long b_state, b_state_old; |
1448 | ||
1da177e4 LT |
1449 | lock_buffer(bh); |
1450 | clear_buffer_dirty(bh); | |
1451 | bh->b_bdev = NULL; | |
e7470ee8 MG |
1452 | b_state = bh->b_state; |
1453 | for (;;) { | |
1454 | b_state_old = cmpxchg(&bh->b_state, b_state, | |
1455 | (b_state & ~BUFFER_FLAGS_DISCARD)); | |
1456 | if (b_state_old == b_state) | |
1457 | break; | |
1458 | b_state = b_state_old; | |
1459 | } | |
1da177e4 LT |
1460 | unlock_buffer(bh); |
1461 | } | |
1462 | ||
1da177e4 | 1463 | /** |
814e1d25 | 1464 | * block_invalidatepage - invalidate part or all of a buffer-backed page |
1da177e4 LT |
1465 | * |
1466 | * @page: the page which is affected | |
d47992f8 LC |
1467 | * @offset: start of the range to invalidate |
1468 | * @length: length of the range to invalidate | |
1da177e4 LT |
1469 | * |
1470 | * block_invalidatepage() is called when all or part of the page has become | |
814e1d25 | 1471 | * invalidated by a truncate operation. |
1da177e4 LT |
1472 | * |
1473 | * block_invalidatepage() does not have to release all buffers, but it must | |
1474 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
1475 | * is underway against any of the blocks which are outside the truncation | |
1476 | * point. Because the caller is about to free (and possibly reuse) those | |
1477 | * blocks on-disk. | |
1478 | */ | |
d47992f8 LC |
1479 | void block_invalidatepage(struct page *page, unsigned int offset, |
1480 | unsigned int length) | |
1da177e4 LT |
1481 | { |
1482 | struct buffer_head *head, *bh, *next; | |
1483 | unsigned int curr_off = 0; | |
d47992f8 | 1484 | unsigned int stop = length + offset; |
1da177e4 LT |
1485 | |
1486 | BUG_ON(!PageLocked(page)); | |
1487 | if (!page_has_buffers(page)) | |
1488 | goto out; | |
1489 | ||
d47992f8 LC |
1490 | /* |
1491 | * Check for overflow | |
1492 | */ | |
09cbfeaf | 1493 | BUG_ON(stop > PAGE_SIZE || stop < length); |
d47992f8 | 1494 | |
1da177e4 LT |
1495 | head = page_buffers(page); |
1496 | bh = head; | |
1497 | do { | |
1498 | unsigned int next_off = curr_off + bh->b_size; | |
1499 | next = bh->b_this_page; | |
1500 | ||
d47992f8 LC |
1501 | /* |
1502 | * Are we still fully in range ? | |
1503 | */ | |
1504 | if (next_off > stop) | |
1505 | goto out; | |
1506 | ||
1da177e4 LT |
1507 | /* |
1508 | * is this block fully invalidated? | |
1509 | */ | |
1510 | if (offset <= curr_off) | |
1511 | discard_buffer(bh); | |
1512 | curr_off = next_off; | |
1513 | bh = next; | |
1514 | } while (bh != head); | |
1515 | ||
1516 | /* | |
1517 | * We release buffers only if the entire page is being invalidated. | |
1518 | * The get_block cached value has been unconditionally invalidated, | |
1519 | * so real IO is not possible anymore. | |
1520 | */ | |
1521 | if (offset == 0) | |
2ff28e22 | 1522 | try_to_release_page(page, 0); |
1da177e4 | 1523 | out: |
2ff28e22 | 1524 | return; |
1da177e4 LT |
1525 | } |
1526 | EXPORT_SYMBOL(block_invalidatepage); | |
1527 | ||
d47992f8 | 1528 | |
1da177e4 LT |
1529 | /* |
1530 | * We attach and possibly dirty the buffers atomically wrt | |
1531 | * __set_page_dirty_buffers() via private_lock. try_to_free_buffers | |
1532 | * is already excluded via the page lock. | |
1533 | */ | |
1534 | void create_empty_buffers(struct page *page, | |
1535 | unsigned long blocksize, unsigned long b_state) | |
1536 | { | |
1537 | struct buffer_head *bh, *head, *tail; | |
1538 | ||
640ab98f | 1539 | head = alloc_page_buffers(page, blocksize, true); |
1da177e4 LT |
1540 | bh = head; |
1541 | do { | |
1542 | bh->b_state |= b_state; | |
1543 | tail = bh; | |
1544 | bh = bh->b_this_page; | |
1545 | } while (bh); | |
1546 | tail->b_this_page = head; | |
1547 | ||
1548 | spin_lock(&page->mapping->private_lock); | |
1549 | if (PageUptodate(page) || PageDirty(page)) { | |
1550 | bh = head; | |
1551 | do { | |
1552 | if (PageDirty(page)) | |
1553 | set_buffer_dirty(bh); | |
1554 | if (PageUptodate(page)) | |
1555 | set_buffer_uptodate(bh); | |
1556 | bh = bh->b_this_page; | |
1557 | } while (bh != head); | |
1558 | } | |
1559 | attach_page_buffers(page, head); | |
1560 | spin_unlock(&page->mapping->private_lock); | |
1561 | } | |
1562 | EXPORT_SYMBOL(create_empty_buffers); | |
1563 | ||
29f3ad7d JK |
1564 | /** |
1565 | * clean_bdev_aliases: clean a range of buffers in block device | |
1566 | * @bdev: Block device to clean buffers in | |
1567 | * @block: Start of a range of blocks to clean | |
1568 | * @len: Number of blocks to clean | |
1da177e4 | 1569 | * |
29f3ad7d JK |
1570 | * We are taking a range of blocks for data and we don't want writeback of any |
1571 | * buffer-cache aliases starting from return from this function and until the | |
1572 | * moment when something will explicitly mark the buffer dirty (hopefully that | |
1573 | * will not happen until we will free that block ;-) We don't even need to mark | |
1574 | * it not-uptodate - nobody can expect anything from a newly allocated buffer | |
1575 | * anyway. We used to use unmap_buffer() for such invalidation, but that was | |
1576 | * wrong. We definitely don't want to mark the alias unmapped, for example - it | |
1577 | * would confuse anyone who might pick it with bread() afterwards... | |
1578 | * | |
1579 | * Also.. Note that bforget() doesn't lock the buffer. So there can be | |
1580 | * writeout I/O going on against recently-freed buffers. We don't wait on that | |
1581 | * I/O in bforget() - it's more efficient to wait on the I/O only if we really | |
1582 | * need to. That happens here. | |
1da177e4 | 1583 | */ |
29f3ad7d | 1584 | void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) |
1da177e4 | 1585 | { |
29f3ad7d JK |
1586 | struct inode *bd_inode = bdev->bd_inode; |
1587 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
1588 | struct pagevec pvec; | |
1589 | pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits); | |
1590 | pgoff_t end; | |
c10f778d | 1591 | int i, count; |
29f3ad7d JK |
1592 | struct buffer_head *bh; |
1593 | struct buffer_head *head; | |
1da177e4 | 1594 | |
29f3ad7d | 1595 | end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits); |
86679820 | 1596 | pagevec_init(&pvec); |
397162ff | 1597 | while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) { |
c10f778d JK |
1598 | count = pagevec_count(&pvec); |
1599 | for (i = 0; i < count; i++) { | |
29f3ad7d | 1600 | struct page *page = pvec.pages[i]; |
1da177e4 | 1601 | |
29f3ad7d JK |
1602 | if (!page_has_buffers(page)) |
1603 | continue; | |
1604 | /* | |
1605 | * We use page lock instead of bd_mapping->private_lock | |
1606 | * to pin buffers here since we can afford to sleep and | |
1607 | * it scales better than a global spinlock lock. | |
1608 | */ | |
1609 | lock_page(page); | |
1610 | /* Recheck when the page is locked which pins bhs */ | |
1611 | if (!page_has_buffers(page)) | |
1612 | goto unlock_page; | |
1613 | head = page_buffers(page); | |
1614 | bh = head; | |
1615 | do { | |
6c006a9d | 1616 | if (!buffer_mapped(bh) || (bh->b_blocknr < block)) |
29f3ad7d JK |
1617 | goto next; |
1618 | if (bh->b_blocknr >= block + len) | |
1619 | break; | |
1620 | clear_buffer_dirty(bh); | |
1621 | wait_on_buffer(bh); | |
1622 | clear_buffer_req(bh); | |
1623 | next: | |
1624 | bh = bh->b_this_page; | |
1625 | } while (bh != head); | |
1626 | unlock_page: | |
1627 | unlock_page(page); | |
1628 | } | |
1629 | pagevec_release(&pvec); | |
1630 | cond_resched(); | |
c10f778d JK |
1631 | /* End of range already reached? */ |
1632 | if (index > end || !index) | |
1633 | break; | |
1da177e4 LT |
1634 | } |
1635 | } | |
29f3ad7d | 1636 | EXPORT_SYMBOL(clean_bdev_aliases); |
1da177e4 | 1637 | |
45bce8f3 LT |
1638 | /* |
1639 | * Size is a power-of-two in the range 512..PAGE_SIZE, | |
1640 | * and the case we care about most is PAGE_SIZE. | |
1641 | * | |
1642 | * So this *could* possibly be written with those | |
1643 | * constraints in mind (relevant mostly if some | |
1644 | * architecture has a slow bit-scan instruction) | |
1645 | */ | |
1646 | static inline int block_size_bits(unsigned int blocksize) | |
1647 | { | |
1648 | return ilog2(blocksize); | |
1649 | } | |
1650 | ||
1651 | static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) | |
1652 | { | |
1653 | BUG_ON(!PageLocked(page)); | |
1654 | ||
1655 | if (!page_has_buffers(page)) | |
6aa7de05 MR |
1656 | create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits), |
1657 | b_state); | |
45bce8f3 LT |
1658 | return page_buffers(page); |
1659 | } | |
1660 | ||
1da177e4 LT |
1661 | /* |
1662 | * NOTE! All mapped/uptodate combinations are valid: | |
1663 | * | |
1664 | * Mapped Uptodate Meaning | |
1665 | * | |
1666 | * No No "unknown" - must do get_block() | |
1667 | * No Yes "hole" - zero-filled | |
1668 | * Yes No "allocated" - allocated on disk, not read in | |
1669 | * Yes Yes "valid" - allocated and up-to-date in memory. | |
1670 | * | |
1671 | * "Dirty" is valid only with the last case (mapped+uptodate). | |
1672 | */ | |
1673 | ||
1674 | /* | |
1675 | * While block_write_full_page is writing back the dirty buffers under | |
1676 | * the page lock, whoever dirtied the buffers may decide to clean them | |
1677 | * again at any time. We handle that by only looking at the buffer | |
1678 | * state inside lock_buffer(). | |
1679 | * | |
1680 | * If block_write_full_page() is called for regular writeback | |
1681 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a | |
1682 | * locked buffer. This only can happen if someone has written the buffer | |
1683 | * directly, with submit_bh(). At the address_space level PageWriteback | |
1684 | * prevents this contention from occurring. | |
6e34eedd TT |
1685 | * |
1686 | * If block_write_full_page() is called with wbc->sync_mode == | |
70fd7614 | 1687 | * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this |
721a9602 | 1688 | * causes the writes to be flagged as synchronous writes. |
1da177e4 | 1689 | */ |
b4bba389 | 1690 | int __block_write_full_page(struct inode *inode, struct page *page, |
35c80d5f CM |
1691 | get_block_t *get_block, struct writeback_control *wbc, |
1692 | bh_end_io_t *handler) | |
1da177e4 LT |
1693 | { |
1694 | int err; | |
1695 | sector_t block; | |
1696 | sector_t last_block; | |
f0fbd5fc | 1697 | struct buffer_head *bh, *head; |
45bce8f3 | 1698 | unsigned int blocksize, bbits; |
1da177e4 | 1699 | int nr_underway = 0; |
7637241e | 1700 | int write_flags = wbc_to_write_flags(wbc); |
1da177e4 | 1701 | |
45bce8f3 | 1702 | head = create_page_buffers(page, inode, |
1da177e4 | 1703 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
1da177e4 LT |
1704 | |
1705 | /* | |
1706 | * Be very careful. We have no exclusion from __set_page_dirty_buffers | |
1707 | * here, and the (potentially unmapped) buffers may become dirty at | |
1708 | * any time. If a buffer becomes dirty here after we've inspected it | |
1709 | * then we just miss that fact, and the page stays dirty. | |
1710 | * | |
1711 | * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; | |
1712 | * handle that here by just cleaning them. | |
1713 | */ | |
1714 | ||
1da177e4 | 1715 | bh = head; |
45bce8f3 LT |
1716 | blocksize = bh->b_size; |
1717 | bbits = block_size_bits(blocksize); | |
1718 | ||
09cbfeaf | 1719 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 1720 | last_block = (i_size_read(inode) - 1) >> bbits; |
1da177e4 LT |
1721 | |
1722 | /* | |
1723 | * Get all the dirty buffers mapped to disk addresses and | |
1724 | * handle any aliases from the underlying blockdev's mapping. | |
1725 | */ | |
1726 | do { | |
1727 | if (block > last_block) { | |
1728 | /* | |
1729 | * mapped buffers outside i_size will occur, because | |
1730 | * this page can be outside i_size when there is a | |
1731 | * truncate in progress. | |
1732 | */ | |
1733 | /* | |
1734 | * The buffer was zeroed by block_write_full_page() | |
1735 | */ | |
1736 | clear_buffer_dirty(bh); | |
1737 | set_buffer_uptodate(bh); | |
29a814d2 AT |
1738 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
1739 | buffer_dirty(bh)) { | |
b0cf2321 | 1740 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1741 | err = get_block(inode, block, bh, 1); |
1742 | if (err) | |
1743 | goto recover; | |
29a814d2 | 1744 | clear_buffer_delay(bh); |
1da177e4 LT |
1745 | if (buffer_new(bh)) { |
1746 | /* blockdev mappings never come here */ | |
1747 | clear_buffer_new(bh); | |
e64855c6 | 1748 | clean_bdev_bh_alias(bh); |
1da177e4 LT |
1749 | } |
1750 | } | |
1751 | bh = bh->b_this_page; | |
1752 | block++; | |
1753 | } while (bh != head); | |
1754 | ||
1755 | do { | |
1da177e4 LT |
1756 | if (!buffer_mapped(bh)) |
1757 | continue; | |
1758 | /* | |
1759 | * If it's a fully non-blocking write attempt and we cannot | |
1760 | * lock the buffer then redirty the page. Note that this can | |
5b0830cb JA |
1761 | * potentially cause a busy-wait loop from writeback threads |
1762 | * and kswapd activity, but those code paths have their own | |
1763 | * higher-level throttling. | |
1da177e4 | 1764 | */ |
1b430bee | 1765 | if (wbc->sync_mode != WB_SYNC_NONE) { |
1da177e4 | 1766 | lock_buffer(bh); |
ca5de404 | 1767 | } else if (!trylock_buffer(bh)) { |
1da177e4 LT |
1768 | redirty_page_for_writepage(wbc, page); |
1769 | continue; | |
1770 | } | |
1771 | if (test_clear_buffer_dirty(bh)) { | |
35c80d5f | 1772 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1773 | } else { |
1774 | unlock_buffer(bh); | |
1775 | } | |
1776 | } while ((bh = bh->b_this_page) != head); | |
1777 | ||
1778 | /* | |
1779 | * The page and its buffers are protected by PageWriteback(), so we can | |
1780 | * drop the bh refcounts early. | |
1781 | */ | |
1782 | BUG_ON(PageWriteback(page)); | |
1783 | set_page_writeback(page); | |
1da177e4 LT |
1784 | |
1785 | do { | |
1786 | struct buffer_head *next = bh->b_this_page; | |
1787 | if (buffer_async_write(bh)) { | |
8e8f9298 JA |
1788 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1789 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1790 | nr_underway++; |
1791 | } | |
1da177e4 LT |
1792 | bh = next; |
1793 | } while (bh != head); | |
05937baa | 1794 | unlock_page(page); |
1da177e4 LT |
1795 | |
1796 | err = 0; | |
1797 | done: | |
1798 | if (nr_underway == 0) { | |
1799 | /* | |
1800 | * The page was marked dirty, but the buffers were | |
1801 | * clean. Someone wrote them back by hand with | |
1802 | * ll_rw_block/submit_bh. A rare case. | |
1803 | */ | |
1da177e4 | 1804 | end_page_writeback(page); |
3d67f2d7 | 1805 | |
1da177e4 LT |
1806 | /* |
1807 | * The page and buffer_heads can be released at any time from | |
1808 | * here on. | |
1809 | */ | |
1da177e4 LT |
1810 | } |
1811 | return err; | |
1812 | ||
1813 | recover: | |
1814 | /* | |
1815 | * ENOSPC, or some other error. We may already have added some | |
1816 | * blocks to the file, so we need to write these out to avoid | |
1817 | * exposing stale data. | |
1818 | * The page is currently locked and not marked for writeback | |
1819 | */ | |
1820 | bh = head; | |
1821 | /* Recovery: lock and submit the mapped buffers */ | |
1822 | do { | |
29a814d2 AT |
1823 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
1824 | !buffer_delay(bh)) { | |
1da177e4 | 1825 | lock_buffer(bh); |
35c80d5f | 1826 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1827 | } else { |
1828 | /* | |
1829 | * The buffer may have been set dirty during | |
1830 | * attachment to a dirty page. | |
1831 | */ | |
1832 | clear_buffer_dirty(bh); | |
1833 | } | |
1834 | } while ((bh = bh->b_this_page) != head); | |
1835 | SetPageError(page); | |
1836 | BUG_ON(PageWriteback(page)); | |
7e4c3690 | 1837 | mapping_set_error(page->mapping, err); |
1da177e4 | 1838 | set_page_writeback(page); |
1da177e4 LT |
1839 | do { |
1840 | struct buffer_head *next = bh->b_this_page; | |
1841 | if (buffer_async_write(bh)) { | |
1842 | clear_buffer_dirty(bh); | |
8e8f9298 JA |
1843 | submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, |
1844 | inode->i_write_hint, wbc); | |
1da177e4 LT |
1845 | nr_underway++; |
1846 | } | |
1da177e4 LT |
1847 | bh = next; |
1848 | } while (bh != head); | |
ffda9d30 | 1849 | unlock_page(page); |
1da177e4 LT |
1850 | goto done; |
1851 | } | |
b4bba389 | 1852 | EXPORT_SYMBOL(__block_write_full_page); |
1da177e4 | 1853 | |
afddba49 NP |
1854 | /* |
1855 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
1856 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1857 | * block data from leaking). And clear the new bit. | |
1858 | */ | |
1859 | void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1860 | { | |
1861 | unsigned int block_start, block_end; | |
1862 | struct buffer_head *head, *bh; | |
1863 | ||
1864 | BUG_ON(!PageLocked(page)); | |
1865 | if (!page_has_buffers(page)) | |
1866 | return; | |
1867 | ||
1868 | bh = head = page_buffers(page); | |
1869 | block_start = 0; | |
1870 | do { | |
1871 | block_end = block_start + bh->b_size; | |
1872 | ||
1873 | if (buffer_new(bh)) { | |
1874 | if (block_end > from && block_start < to) { | |
1875 | if (!PageUptodate(page)) { | |
1876 | unsigned start, size; | |
1877 | ||
1878 | start = max(from, block_start); | |
1879 | size = min(to, block_end) - start; | |
1880 | ||
eebd2aa3 | 1881 | zero_user(page, start, size); |
afddba49 NP |
1882 | set_buffer_uptodate(bh); |
1883 | } | |
1884 | ||
1885 | clear_buffer_new(bh); | |
1886 | mark_buffer_dirty(bh); | |
1887 | } | |
1888 | } | |
1889 | ||
1890 | block_start = block_end; | |
1891 | bh = bh->b_this_page; | |
1892 | } while (bh != head); | |
1893 | } | |
1894 | EXPORT_SYMBOL(page_zero_new_buffers); | |
1895 | ||
ae259a9c CH |
1896 | static void |
1897 | iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, | |
1898 | struct iomap *iomap) | |
1899 | { | |
1900 | loff_t offset = block << inode->i_blkbits; | |
1901 | ||
1902 | bh->b_bdev = iomap->bdev; | |
1903 | ||
1904 | /* | |
1905 | * Block points to offset in file we need to map, iomap contains | |
1906 | * the offset at which the map starts. If the map ends before the | |
1907 | * current block, then do not map the buffer and let the caller | |
1908 | * handle it. | |
1909 | */ | |
1910 | BUG_ON(offset >= iomap->offset + iomap->length); | |
1911 | ||
1912 | switch (iomap->type) { | |
1913 | case IOMAP_HOLE: | |
1914 | /* | |
1915 | * If the buffer is not up to date or beyond the current EOF, | |
1916 | * we need to mark it as new to ensure sub-block zeroing is | |
1917 | * executed if necessary. | |
1918 | */ | |
1919 | if (!buffer_uptodate(bh) || | |
1920 | (offset >= i_size_read(inode))) | |
1921 | set_buffer_new(bh); | |
1922 | break; | |
1923 | case IOMAP_DELALLOC: | |
1924 | if (!buffer_uptodate(bh) || | |
1925 | (offset >= i_size_read(inode))) | |
1926 | set_buffer_new(bh); | |
1927 | set_buffer_uptodate(bh); | |
1928 | set_buffer_mapped(bh); | |
1929 | set_buffer_delay(bh); | |
1930 | break; | |
1931 | case IOMAP_UNWRITTEN: | |
1932 | /* | |
1933 | * For unwritten regions, we always need to ensure that | |
1934 | * sub-block writes cause the regions in the block we are not | |
1935 | * writing to are zeroed. Set the buffer as new to ensure this. | |
1936 | */ | |
1937 | set_buffer_new(bh); | |
1938 | set_buffer_unwritten(bh); | |
1939 | /* FALLTHRU */ | |
1940 | case IOMAP_MAPPED: | |
1941 | if (offset >= i_size_read(inode)) | |
1942 | set_buffer_new(bh); | |
19fe5f64 AG |
1943 | bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> |
1944 | inode->i_blkbits; | |
ae259a9c CH |
1945 | set_buffer_mapped(bh); |
1946 | break; | |
1947 | } | |
1948 | } | |
1949 | ||
1950 | int __block_write_begin_int(struct page *page, loff_t pos, unsigned len, | |
1951 | get_block_t *get_block, struct iomap *iomap) | |
1da177e4 | 1952 | { |
09cbfeaf | 1953 | unsigned from = pos & (PAGE_SIZE - 1); |
ebdec241 | 1954 | unsigned to = from + len; |
6e1db88d | 1955 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
1956 | unsigned block_start, block_end; |
1957 | sector_t block; | |
1958 | int err = 0; | |
1959 | unsigned blocksize, bbits; | |
1960 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; | |
1961 | ||
1962 | BUG_ON(!PageLocked(page)); | |
09cbfeaf KS |
1963 | BUG_ON(from > PAGE_SIZE); |
1964 | BUG_ON(to > PAGE_SIZE); | |
1da177e4 LT |
1965 | BUG_ON(from > to); |
1966 | ||
45bce8f3 LT |
1967 | head = create_page_buffers(page, inode, 0); |
1968 | blocksize = head->b_size; | |
1969 | bbits = block_size_bits(blocksize); | |
1da177e4 | 1970 | |
09cbfeaf | 1971 | block = (sector_t)page->index << (PAGE_SHIFT - bbits); |
1da177e4 LT |
1972 | |
1973 | for(bh = head, block_start = 0; bh != head || !block_start; | |
1974 | block++, block_start=block_end, bh = bh->b_this_page) { | |
1975 | block_end = block_start + blocksize; | |
1976 | if (block_end <= from || block_start >= to) { | |
1977 | if (PageUptodate(page)) { | |
1978 | if (!buffer_uptodate(bh)) | |
1979 | set_buffer_uptodate(bh); | |
1980 | } | |
1981 | continue; | |
1982 | } | |
1983 | if (buffer_new(bh)) | |
1984 | clear_buffer_new(bh); | |
1985 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 1986 | WARN_ON(bh->b_size != blocksize); |
ae259a9c CH |
1987 | if (get_block) { |
1988 | err = get_block(inode, block, bh, 1); | |
1989 | if (err) | |
1990 | break; | |
1991 | } else { | |
1992 | iomap_to_bh(inode, block, bh, iomap); | |
1993 | } | |
1994 | ||
1da177e4 | 1995 | if (buffer_new(bh)) { |
e64855c6 | 1996 | clean_bdev_bh_alias(bh); |
1da177e4 | 1997 | if (PageUptodate(page)) { |
637aff46 | 1998 | clear_buffer_new(bh); |
1da177e4 | 1999 | set_buffer_uptodate(bh); |
637aff46 | 2000 | mark_buffer_dirty(bh); |
1da177e4 LT |
2001 | continue; |
2002 | } | |
eebd2aa3 CL |
2003 | if (block_end > to || block_start < from) |
2004 | zero_user_segments(page, | |
2005 | to, block_end, | |
2006 | block_start, from); | |
1da177e4 LT |
2007 | continue; |
2008 | } | |
2009 | } | |
2010 | if (PageUptodate(page)) { | |
2011 | if (!buffer_uptodate(bh)) | |
2012 | set_buffer_uptodate(bh); | |
2013 | continue; | |
2014 | } | |
2015 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
33a266dd | 2016 | !buffer_unwritten(bh) && |
1da177e4 | 2017 | (block_start < from || block_end > to)) { |
dfec8a14 | 2018 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2019 | *wait_bh++=bh; |
2020 | } | |
2021 | } | |
2022 | /* | |
2023 | * If we issued read requests - let them complete. | |
2024 | */ | |
2025 | while(wait_bh > wait) { | |
2026 | wait_on_buffer(*--wait_bh); | |
2027 | if (!buffer_uptodate(*wait_bh)) | |
f3ddbdc6 | 2028 | err = -EIO; |
1da177e4 | 2029 | } |
f9f07b6c | 2030 | if (unlikely(err)) |
afddba49 | 2031 | page_zero_new_buffers(page, from, to); |
1da177e4 LT |
2032 | return err; |
2033 | } | |
ae259a9c CH |
2034 | |
2035 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, | |
2036 | get_block_t *get_block) | |
2037 | { | |
2038 | return __block_write_begin_int(page, pos, len, get_block, NULL); | |
2039 | } | |
ebdec241 | 2040 | EXPORT_SYMBOL(__block_write_begin); |
1da177e4 LT |
2041 | |
2042 | static int __block_commit_write(struct inode *inode, struct page *page, | |
2043 | unsigned from, unsigned to) | |
2044 | { | |
2045 | unsigned block_start, block_end; | |
2046 | int partial = 0; | |
2047 | unsigned blocksize; | |
2048 | struct buffer_head *bh, *head; | |
2049 | ||
45bce8f3 LT |
2050 | bh = head = page_buffers(page); |
2051 | blocksize = bh->b_size; | |
1da177e4 | 2052 | |
45bce8f3 LT |
2053 | block_start = 0; |
2054 | do { | |
1da177e4 LT |
2055 | block_end = block_start + blocksize; |
2056 | if (block_end <= from || block_start >= to) { | |
2057 | if (!buffer_uptodate(bh)) | |
2058 | partial = 1; | |
2059 | } else { | |
2060 | set_buffer_uptodate(bh); | |
2061 | mark_buffer_dirty(bh); | |
2062 | } | |
afddba49 | 2063 | clear_buffer_new(bh); |
45bce8f3 LT |
2064 | |
2065 | block_start = block_end; | |
2066 | bh = bh->b_this_page; | |
2067 | } while (bh != head); | |
1da177e4 LT |
2068 | |
2069 | /* | |
2070 | * If this is a partial write which happened to make all buffers | |
2071 | * uptodate then we can optimize away a bogus readpage() for | |
2072 | * the next read(). Here we 'discover' whether the page went | |
2073 | * uptodate as a result of this (potentially partial) write. | |
2074 | */ | |
2075 | if (!partial) | |
2076 | SetPageUptodate(page); | |
2077 | return 0; | |
2078 | } | |
2079 | ||
afddba49 | 2080 | /* |
155130a4 CH |
2081 | * block_write_begin takes care of the basic task of block allocation and |
2082 | * bringing partial write blocks uptodate first. | |
2083 | * | |
7bb46a67 | 2084 | * The filesystem needs to handle block truncation upon failure. |
afddba49 | 2085 | */ |
155130a4 CH |
2086 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
2087 | unsigned flags, struct page **pagep, get_block_t *get_block) | |
afddba49 | 2088 | { |
09cbfeaf | 2089 | pgoff_t index = pos >> PAGE_SHIFT; |
afddba49 | 2090 | struct page *page; |
6e1db88d | 2091 | int status; |
afddba49 | 2092 | |
6e1db88d CH |
2093 | page = grab_cache_page_write_begin(mapping, index, flags); |
2094 | if (!page) | |
2095 | return -ENOMEM; | |
afddba49 | 2096 | |
6e1db88d | 2097 | status = __block_write_begin(page, pos, len, get_block); |
afddba49 | 2098 | if (unlikely(status)) { |
6e1db88d | 2099 | unlock_page(page); |
09cbfeaf | 2100 | put_page(page); |
6e1db88d | 2101 | page = NULL; |
afddba49 NP |
2102 | } |
2103 | ||
6e1db88d | 2104 | *pagep = page; |
afddba49 NP |
2105 | return status; |
2106 | } | |
2107 | EXPORT_SYMBOL(block_write_begin); | |
2108 | ||
2109 | int block_write_end(struct file *file, struct address_space *mapping, | |
2110 | loff_t pos, unsigned len, unsigned copied, | |
2111 | struct page *page, void *fsdata) | |
2112 | { | |
2113 | struct inode *inode = mapping->host; | |
2114 | unsigned start; | |
2115 | ||
09cbfeaf | 2116 | start = pos & (PAGE_SIZE - 1); |
afddba49 NP |
2117 | |
2118 | if (unlikely(copied < len)) { | |
2119 | /* | |
2120 | * The buffers that were written will now be uptodate, so we | |
2121 | * don't have to worry about a readpage reading them and | |
2122 | * overwriting a partial write. However if we have encountered | |
2123 | * a short write and only partially written into a buffer, it | |
2124 | * will not be marked uptodate, so a readpage might come in and | |
2125 | * destroy our partial write. | |
2126 | * | |
2127 | * Do the simplest thing, and just treat any short write to a | |
2128 | * non uptodate page as a zero-length write, and force the | |
2129 | * caller to redo the whole thing. | |
2130 | */ | |
2131 | if (!PageUptodate(page)) | |
2132 | copied = 0; | |
2133 | ||
2134 | page_zero_new_buffers(page, start+copied, start+len); | |
2135 | } | |
2136 | flush_dcache_page(page); | |
2137 | ||
2138 | /* This could be a short (even 0-length) commit */ | |
2139 | __block_commit_write(inode, page, start, start+copied); | |
2140 | ||
2141 | return copied; | |
2142 | } | |
2143 | EXPORT_SYMBOL(block_write_end); | |
2144 | ||
2145 | int generic_write_end(struct file *file, struct address_space *mapping, | |
2146 | loff_t pos, unsigned len, unsigned copied, | |
2147 | struct page *page, void *fsdata) | |
2148 | { | |
2149 | struct inode *inode = mapping->host; | |
90a80202 | 2150 | loff_t old_size = inode->i_size; |
c7d206b3 | 2151 | int i_size_changed = 0; |
afddba49 NP |
2152 | |
2153 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
2154 | ||
2155 | /* | |
2156 | * No need to use i_size_read() here, the i_size | |
2157 | * cannot change under us because we hold i_mutex. | |
2158 | * | |
2159 | * But it's important to update i_size while still holding page lock: | |
2160 | * page writeout could otherwise come in and zero beyond i_size. | |
2161 | */ | |
2162 | if (pos+copied > inode->i_size) { | |
2163 | i_size_write(inode, pos+copied); | |
c7d206b3 | 2164 | i_size_changed = 1; |
afddba49 NP |
2165 | } |
2166 | ||
2167 | unlock_page(page); | |
09cbfeaf | 2168 | put_page(page); |
afddba49 | 2169 | |
90a80202 JK |
2170 | if (old_size < pos) |
2171 | pagecache_isize_extended(inode, old_size, pos); | |
c7d206b3 JK |
2172 | /* |
2173 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
2174 | * makes the holding time of page lock longer. Second, it forces lock | |
2175 | * ordering of page lock and transaction start for journaling | |
2176 | * filesystems. | |
2177 | */ | |
2178 | if (i_size_changed) | |
2179 | mark_inode_dirty(inode); | |
2180 | ||
afddba49 NP |
2181 | return copied; |
2182 | } | |
2183 | EXPORT_SYMBOL(generic_write_end); | |
2184 | ||
8ab22b9a HH |
2185 | /* |
2186 | * block_is_partially_uptodate checks whether buffers within a page are | |
2187 | * uptodate or not. | |
2188 | * | |
2189 | * Returns true if all buffers which correspond to a file portion | |
2190 | * we want to read are uptodate. | |
2191 | */ | |
c186afb4 AV |
2192 | int block_is_partially_uptodate(struct page *page, unsigned long from, |
2193 | unsigned long count) | |
8ab22b9a | 2194 | { |
8ab22b9a HH |
2195 | unsigned block_start, block_end, blocksize; |
2196 | unsigned to; | |
2197 | struct buffer_head *bh, *head; | |
2198 | int ret = 1; | |
2199 | ||
2200 | if (!page_has_buffers(page)) | |
2201 | return 0; | |
2202 | ||
45bce8f3 LT |
2203 | head = page_buffers(page); |
2204 | blocksize = head->b_size; | |
09cbfeaf | 2205 | to = min_t(unsigned, PAGE_SIZE - from, count); |
8ab22b9a | 2206 | to = from + to; |
09cbfeaf | 2207 | if (from < blocksize && to > PAGE_SIZE - blocksize) |
8ab22b9a HH |
2208 | return 0; |
2209 | ||
8ab22b9a HH |
2210 | bh = head; |
2211 | block_start = 0; | |
2212 | do { | |
2213 | block_end = block_start + blocksize; | |
2214 | if (block_end > from && block_start < to) { | |
2215 | if (!buffer_uptodate(bh)) { | |
2216 | ret = 0; | |
2217 | break; | |
2218 | } | |
2219 | if (block_end >= to) | |
2220 | break; | |
2221 | } | |
2222 | block_start = block_end; | |
2223 | bh = bh->b_this_page; | |
2224 | } while (bh != head); | |
2225 | ||
2226 | return ret; | |
2227 | } | |
2228 | EXPORT_SYMBOL(block_is_partially_uptodate); | |
2229 | ||
1da177e4 LT |
2230 | /* |
2231 | * Generic "read page" function for block devices that have the normal | |
2232 | * get_block functionality. This is most of the block device filesystems. | |
2233 | * Reads the page asynchronously --- the unlock_buffer() and | |
2234 | * set/clear_buffer_uptodate() functions propagate buffer state into the | |
2235 | * page struct once IO has completed. | |
2236 | */ | |
2237 | int block_read_full_page(struct page *page, get_block_t *get_block) | |
2238 | { | |
2239 | struct inode *inode = page->mapping->host; | |
2240 | sector_t iblock, lblock; | |
2241 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; | |
45bce8f3 | 2242 | unsigned int blocksize, bbits; |
1da177e4 LT |
2243 | int nr, i; |
2244 | int fully_mapped = 1; | |
2245 | ||
45bce8f3 LT |
2246 | head = create_page_buffers(page, inode, 0); |
2247 | blocksize = head->b_size; | |
2248 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2249 | |
09cbfeaf | 2250 | iblock = (sector_t)page->index << (PAGE_SHIFT - bbits); |
45bce8f3 | 2251 | lblock = (i_size_read(inode)+blocksize-1) >> bbits; |
1da177e4 LT |
2252 | bh = head; |
2253 | nr = 0; | |
2254 | i = 0; | |
2255 | ||
2256 | do { | |
2257 | if (buffer_uptodate(bh)) | |
2258 | continue; | |
2259 | ||
2260 | if (!buffer_mapped(bh)) { | |
c64610ba AM |
2261 | int err = 0; |
2262 | ||
1da177e4 LT |
2263 | fully_mapped = 0; |
2264 | if (iblock < lblock) { | |
b0cf2321 | 2265 | WARN_ON(bh->b_size != blocksize); |
c64610ba AM |
2266 | err = get_block(inode, iblock, bh, 0); |
2267 | if (err) | |
1da177e4 LT |
2268 | SetPageError(page); |
2269 | } | |
2270 | if (!buffer_mapped(bh)) { | |
eebd2aa3 | 2271 | zero_user(page, i * blocksize, blocksize); |
c64610ba AM |
2272 | if (!err) |
2273 | set_buffer_uptodate(bh); | |
1da177e4 LT |
2274 | continue; |
2275 | } | |
2276 | /* | |
2277 | * get_block() might have updated the buffer | |
2278 | * synchronously | |
2279 | */ | |
2280 | if (buffer_uptodate(bh)) | |
2281 | continue; | |
2282 | } | |
2283 | arr[nr++] = bh; | |
2284 | } while (i++, iblock++, (bh = bh->b_this_page) != head); | |
2285 | ||
2286 | if (fully_mapped) | |
2287 | SetPageMappedToDisk(page); | |
2288 | ||
2289 | if (!nr) { | |
2290 | /* | |
2291 | * All buffers are uptodate - we can set the page uptodate | |
2292 | * as well. But not if get_block() returned an error. | |
2293 | */ | |
2294 | if (!PageError(page)) | |
2295 | SetPageUptodate(page); | |
2296 | unlock_page(page); | |
2297 | return 0; | |
2298 | } | |
2299 | ||
2300 | /* Stage two: lock the buffers */ | |
2301 | for (i = 0; i < nr; i++) { | |
2302 | bh = arr[i]; | |
2303 | lock_buffer(bh); | |
2304 | mark_buffer_async_read(bh); | |
2305 | } | |
2306 | ||
2307 | /* | |
2308 | * Stage 3: start the IO. Check for uptodateness | |
2309 | * inside the buffer lock in case another process reading | |
2310 | * the underlying blockdev brought it uptodate (the sct fix). | |
2311 | */ | |
2312 | for (i = 0; i < nr; i++) { | |
2313 | bh = arr[i]; | |
2314 | if (buffer_uptodate(bh)) | |
2315 | end_buffer_async_read(bh, 1); | |
2316 | else | |
2a222ca9 | 2317 | submit_bh(REQ_OP_READ, 0, bh); |
1da177e4 LT |
2318 | } |
2319 | return 0; | |
2320 | } | |
1fe72eaa | 2321 | EXPORT_SYMBOL(block_read_full_page); |
1da177e4 LT |
2322 | |
2323 | /* utility function for filesystems that need to do work on expanding | |
89e10787 | 2324 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
1da177e4 LT |
2325 | * deal with the hole. |
2326 | */ | |
89e10787 | 2327 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
1da177e4 LT |
2328 | { |
2329 | struct address_space *mapping = inode->i_mapping; | |
2330 | struct page *page; | |
89e10787 | 2331 | void *fsdata; |
1da177e4 LT |
2332 | int err; |
2333 | ||
c08d3b0e | 2334 | err = inode_newsize_ok(inode, size); |
2335 | if (err) | |
1da177e4 LT |
2336 | goto out; |
2337 | ||
89e10787 | 2338 | err = pagecache_write_begin(NULL, mapping, size, 0, |
c718a975 | 2339 | AOP_FLAG_CONT_EXPAND, &page, &fsdata); |
89e10787 | 2340 | if (err) |
05eb0b51 | 2341 | goto out; |
05eb0b51 | 2342 | |
89e10787 NP |
2343 | err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); |
2344 | BUG_ON(err > 0); | |
05eb0b51 | 2345 | |
1da177e4 LT |
2346 | out: |
2347 | return err; | |
2348 | } | |
1fe72eaa | 2349 | EXPORT_SYMBOL(generic_cont_expand_simple); |
1da177e4 | 2350 | |
f1e3af72 AB |
2351 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
2352 | loff_t pos, loff_t *bytes) | |
1da177e4 | 2353 | { |
1da177e4 | 2354 | struct inode *inode = mapping->host; |
93407472 | 2355 | unsigned int blocksize = i_blocksize(inode); |
89e10787 NP |
2356 | struct page *page; |
2357 | void *fsdata; | |
2358 | pgoff_t index, curidx; | |
2359 | loff_t curpos; | |
2360 | unsigned zerofrom, offset, len; | |
2361 | int err = 0; | |
1da177e4 | 2362 | |
09cbfeaf KS |
2363 | index = pos >> PAGE_SHIFT; |
2364 | offset = pos & ~PAGE_MASK; | |
89e10787 | 2365 | |
09cbfeaf KS |
2366 | while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { |
2367 | zerofrom = curpos & ~PAGE_MASK; | |
1da177e4 LT |
2368 | if (zerofrom & (blocksize-1)) { |
2369 | *bytes |= (blocksize-1); | |
2370 | (*bytes)++; | |
2371 | } | |
09cbfeaf | 2372 | len = PAGE_SIZE - zerofrom; |
1da177e4 | 2373 | |
c718a975 TH |
2374 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2375 | &page, &fsdata); | |
89e10787 NP |
2376 | if (err) |
2377 | goto out; | |
eebd2aa3 | 2378 | zero_user(page, zerofrom, len); |
89e10787 NP |
2379 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2380 | page, fsdata); | |
2381 | if (err < 0) | |
2382 | goto out; | |
2383 | BUG_ON(err != len); | |
2384 | err = 0; | |
061e9746 OH |
2385 | |
2386 | balance_dirty_pages_ratelimited(mapping); | |
c2ca0fcd MP |
2387 | |
2388 | if (unlikely(fatal_signal_pending(current))) { | |
2389 | err = -EINTR; | |
2390 | goto out; | |
2391 | } | |
89e10787 | 2392 | } |
1da177e4 | 2393 | |
89e10787 NP |
2394 | /* page covers the boundary, find the boundary offset */ |
2395 | if (index == curidx) { | |
09cbfeaf | 2396 | zerofrom = curpos & ~PAGE_MASK; |
1da177e4 | 2397 | /* if we will expand the thing last block will be filled */ |
89e10787 NP |
2398 | if (offset <= zerofrom) { |
2399 | goto out; | |
2400 | } | |
2401 | if (zerofrom & (blocksize-1)) { | |
1da177e4 LT |
2402 | *bytes |= (blocksize-1); |
2403 | (*bytes)++; | |
2404 | } | |
89e10787 | 2405 | len = offset - zerofrom; |
1da177e4 | 2406 | |
c718a975 TH |
2407 | err = pagecache_write_begin(file, mapping, curpos, len, 0, |
2408 | &page, &fsdata); | |
89e10787 NP |
2409 | if (err) |
2410 | goto out; | |
eebd2aa3 | 2411 | zero_user(page, zerofrom, len); |
89e10787 NP |
2412 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2413 | page, fsdata); | |
2414 | if (err < 0) | |
2415 | goto out; | |
2416 | BUG_ON(err != len); | |
2417 | err = 0; | |
1da177e4 | 2418 | } |
89e10787 NP |
2419 | out: |
2420 | return err; | |
2421 | } | |
2422 | ||
2423 | /* | |
2424 | * For moronic filesystems that do not allow holes in file. | |
2425 | * We may have to extend the file. | |
2426 | */ | |
282dc178 | 2427 | int cont_write_begin(struct file *file, struct address_space *mapping, |
89e10787 NP |
2428 | loff_t pos, unsigned len, unsigned flags, |
2429 | struct page **pagep, void **fsdata, | |
2430 | get_block_t *get_block, loff_t *bytes) | |
2431 | { | |
2432 | struct inode *inode = mapping->host; | |
93407472 FF |
2433 | unsigned int blocksize = i_blocksize(inode); |
2434 | unsigned int zerofrom; | |
89e10787 NP |
2435 | int err; |
2436 | ||
2437 | err = cont_expand_zero(file, mapping, pos, bytes); | |
2438 | if (err) | |
155130a4 | 2439 | return err; |
89e10787 | 2440 | |
09cbfeaf | 2441 | zerofrom = *bytes & ~PAGE_MASK; |
89e10787 NP |
2442 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { |
2443 | *bytes |= (blocksize-1); | |
2444 | (*bytes)++; | |
1da177e4 | 2445 | } |
1da177e4 | 2446 | |
155130a4 | 2447 | return block_write_begin(mapping, pos, len, flags, pagep, get_block); |
1da177e4 | 2448 | } |
1fe72eaa | 2449 | EXPORT_SYMBOL(cont_write_begin); |
1da177e4 | 2450 | |
1da177e4 LT |
2451 | int block_commit_write(struct page *page, unsigned from, unsigned to) |
2452 | { | |
2453 | struct inode *inode = page->mapping->host; | |
2454 | __block_commit_write(inode,page,from,to); | |
2455 | return 0; | |
2456 | } | |
1fe72eaa | 2457 | EXPORT_SYMBOL(block_commit_write); |
1da177e4 | 2458 | |
54171690 DC |
2459 | /* |
2460 | * block_page_mkwrite() is not allowed to change the file size as it gets | |
2461 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2462 | * be careful to check for EOF conditions here. We set the page up correctly | |
2463 | * for a written page which means we get ENOSPC checking when writing into | |
2464 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2465 | * support these features. | |
2466 | * | |
2467 | * We are not allowed to take the i_mutex here so we have to play games to | |
2468 | * protect against truncate races as the page could now be beyond EOF. Because | |
7bb46a67 | 2469 | * truncate writes the inode size before removing pages, once we have the |
54171690 DC |
2470 | * page lock we can determine safely if the page is beyond EOF. If it is not |
2471 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2472 | * unlock the page. | |
ea13a864 | 2473 | * |
14da9200 | 2474 | * Direct callers of this function should protect against filesystem freezing |
5c500029 | 2475 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
54171690 | 2476 | */ |
5c500029 | 2477 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
24da4fab | 2478 | get_block_t get_block) |
54171690 | 2479 | { |
c2ec175c | 2480 | struct page *page = vmf->page; |
496ad9aa | 2481 | struct inode *inode = file_inode(vma->vm_file); |
54171690 DC |
2482 | unsigned long end; |
2483 | loff_t size; | |
24da4fab | 2484 | int ret; |
54171690 DC |
2485 | |
2486 | lock_page(page); | |
2487 | size = i_size_read(inode); | |
2488 | if ((page->mapping != inode->i_mapping) || | |
18336338 | 2489 | (page_offset(page) > size)) { |
24da4fab JK |
2490 | /* We overload EFAULT to mean page got truncated */ |
2491 | ret = -EFAULT; | |
2492 | goto out_unlock; | |
54171690 DC |
2493 | } |
2494 | ||
2495 | /* page is wholly or partially inside EOF */ | |
09cbfeaf KS |
2496 | if (((page->index + 1) << PAGE_SHIFT) > size) |
2497 | end = size & ~PAGE_MASK; | |
54171690 | 2498 | else |
09cbfeaf | 2499 | end = PAGE_SIZE; |
54171690 | 2500 | |
ebdec241 | 2501 | ret = __block_write_begin(page, 0, end, get_block); |
54171690 DC |
2502 | if (!ret) |
2503 | ret = block_commit_write(page, 0, end); | |
2504 | ||
24da4fab JK |
2505 | if (unlikely(ret < 0)) |
2506 | goto out_unlock; | |
ea13a864 | 2507 | set_page_dirty(page); |
1d1d1a76 | 2508 | wait_for_stable_page(page); |
24da4fab JK |
2509 | return 0; |
2510 | out_unlock: | |
2511 | unlock_page(page); | |
54171690 | 2512 | return ret; |
24da4fab | 2513 | } |
1fe72eaa | 2514 | EXPORT_SYMBOL(block_page_mkwrite); |
1da177e4 LT |
2515 | |
2516 | /* | |
03158cd7 | 2517 | * nobh_write_begin()'s prereads are special: the buffer_heads are freed |
1da177e4 LT |
2518 | * immediately, while under the page lock. So it needs a special end_io |
2519 | * handler which does not touch the bh after unlocking it. | |
1da177e4 LT |
2520 | */ |
2521 | static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) | |
2522 | { | |
68671f35 | 2523 | __end_buffer_read_notouch(bh, uptodate); |
1da177e4 LT |
2524 | } |
2525 | ||
03158cd7 NP |
2526 | /* |
2527 | * Attach the singly-linked list of buffers created by nobh_write_begin, to | |
2528 | * the page (converting it to circular linked list and taking care of page | |
2529 | * dirty races). | |
2530 | */ | |
2531 | static void attach_nobh_buffers(struct page *page, struct buffer_head *head) | |
2532 | { | |
2533 | struct buffer_head *bh; | |
2534 | ||
2535 | BUG_ON(!PageLocked(page)); | |
2536 | ||
2537 | spin_lock(&page->mapping->private_lock); | |
2538 | bh = head; | |
2539 | do { | |
2540 | if (PageDirty(page)) | |
2541 | set_buffer_dirty(bh); | |
2542 | if (!bh->b_this_page) | |
2543 | bh->b_this_page = head; | |
2544 | bh = bh->b_this_page; | |
2545 | } while (bh != head); | |
2546 | attach_page_buffers(page, head); | |
2547 | spin_unlock(&page->mapping->private_lock); | |
2548 | } | |
2549 | ||
1da177e4 | 2550 | /* |
ea0f04e5 CH |
2551 | * On entry, the page is fully not uptodate. |
2552 | * On exit the page is fully uptodate in the areas outside (from,to) | |
7bb46a67 | 2553 | * The filesystem needs to handle block truncation upon failure. |
1da177e4 | 2554 | */ |
ea0f04e5 | 2555 | int nobh_write_begin(struct address_space *mapping, |
03158cd7 NP |
2556 | loff_t pos, unsigned len, unsigned flags, |
2557 | struct page **pagep, void **fsdata, | |
1da177e4 LT |
2558 | get_block_t *get_block) |
2559 | { | |
03158cd7 | 2560 | struct inode *inode = mapping->host; |
1da177e4 LT |
2561 | const unsigned blkbits = inode->i_blkbits; |
2562 | const unsigned blocksize = 1 << blkbits; | |
a4b0672d | 2563 | struct buffer_head *head, *bh; |
03158cd7 NP |
2564 | struct page *page; |
2565 | pgoff_t index; | |
2566 | unsigned from, to; | |
1da177e4 | 2567 | unsigned block_in_page; |
a4b0672d | 2568 | unsigned block_start, block_end; |
1da177e4 | 2569 | sector_t block_in_file; |
1da177e4 | 2570 | int nr_reads = 0; |
1da177e4 LT |
2571 | int ret = 0; |
2572 | int is_mapped_to_disk = 1; | |
1da177e4 | 2573 | |
09cbfeaf KS |
2574 | index = pos >> PAGE_SHIFT; |
2575 | from = pos & (PAGE_SIZE - 1); | |
03158cd7 NP |
2576 | to = from + len; |
2577 | ||
54566b2c | 2578 | page = grab_cache_page_write_begin(mapping, index, flags); |
03158cd7 NP |
2579 | if (!page) |
2580 | return -ENOMEM; | |
2581 | *pagep = page; | |
2582 | *fsdata = NULL; | |
2583 | ||
2584 | if (page_has_buffers(page)) { | |
309f77ad NK |
2585 | ret = __block_write_begin(page, pos, len, get_block); |
2586 | if (unlikely(ret)) | |
2587 | goto out_release; | |
2588 | return ret; | |
03158cd7 | 2589 | } |
a4b0672d | 2590 | |
1da177e4 LT |
2591 | if (PageMappedToDisk(page)) |
2592 | return 0; | |
2593 | ||
a4b0672d NP |
2594 | /* |
2595 | * Allocate buffers so that we can keep track of state, and potentially | |
2596 | * attach them to the page if an error occurs. In the common case of | |
2597 | * no error, they will just be freed again without ever being attached | |
2598 | * to the page (which is all OK, because we're under the page lock). | |
2599 | * | |
2600 | * Be careful: the buffer linked list is a NULL terminated one, rather | |
2601 | * than the circular one we're used to. | |
2602 | */ | |
640ab98f | 2603 | head = alloc_page_buffers(page, blocksize, false); |
03158cd7 NP |
2604 | if (!head) { |
2605 | ret = -ENOMEM; | |
2606 | goto out_release; | |
2607 | } | |
a4b0672d | 2608 | |
09cbfeaf | 2609 | block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits); |
1da177e4 LT |
2610 | |
2611 | /* | |
2612 | * We loop across all blocks in the page, whether or not they are | |
2613 | * part of the affected region. This is so we can discover if the | |
2614 | * page is fully mapped-to-disk. | |
2615 | */ | |
a4b0672d | 2616 | for (block_start = 0, block_in_page = 0, bh = head; |
09cbfeaf | 2617 | block_start < PAGE_SIZE; |
a4b0672d | 2618 | block_in_page++, block_start += blocksize, bh = bh->b_this_page) { |
1da177e4 LT |
2619 | int create; |
2620 | ||
a4b0672d NP |
2621 | block_end = block_start + blocksize; |
2622 | bh->b_state = 0; | |
1da177e4 LT |
2623 | create = 1; |
2624 | if (block_start >= to) | |
2625 | create = 0; | |
2626 | ret = get_block(inode, block_in_file + block_in_page, | |
a4b0672d | 2627 | bh, create); |
1da177e4 LT |
2628 | if (ret) |
2629 | goto failed; | |
a4b0672d | 2630 | if (!buffer_mapped(bh)) |
1da177e4 | 2631 | is_mapped_to_disk = 0; |
a4b0672d | 2632 | if (buffer_new(bh)) |
e64855c6 | 2633 | clean_bdev_bh_alias(bh); |
a4b0672d NP |
2634 | if (PageUptodate(page)) { |
2635 | set_buffer_uptodate(bh); | |
1da177e4 | 2636 | continue; |
a4b0672d NP |
2637 | } |
2638 | if (buffer_new(bh) || !buffer_mapped(bh)) { | |
eebd2aa3 CL |
2639 | zero_user_segments(page, block_start, from, |
2640 | to, block_end); | |
1da177e4 LT |
2641 | continue; |
2642 | } | |
a4b0672d | 2643 | if (buffer_uptodate(bh)) |
1da177e4 LT |
2644 | continue; /* reiserfs does this */ |
2645 | if (block_start < from || block_end > to) { | |
a4b0672d NP |
2646 | lock_buffer(bh); |
2647 | bh->b_end_io = end_buffer_read_nobh; | |
2a222ca9 | 2648 | submit_bh(REQ_OP_READ, 0, bh); |
a4b0672d | 2649 | nr_reads++; |
1da177e4 LT |
2650 | } |
2651 | } | |
2652 | ||
2653 | if (nr_reads) { | |
1da177e4 LT |
2654 | /* |
2655 | * The page is locked, so these buffers are protected from | |
2656 | * any VM or truncate activity. Hence we don't need to care | |
2657 | * for the buffer_head refcounts. | |
2658 | */ | |
a4b0672d | 2659 | for (bh = head; bh; bh = bh->b_this_page) { |
1da177e4 LT |
2660 | wait_on_buffer(bh); |
2661 | if (!buffer_uptodate(bh)) | |
2662 | ret = -EIO; | |
1da177e4 LT |
2663 | } |
2664 | if (ret) | |
2665 | goto failed; | |
2666 | } | |
2667 | ||
2668 | if (is_mapped_to_disk) | |
2669 | SetPageMappedToDisk(page); | |
1da177e4 | 2670 | |
03158cd7 | 2671 | *fsdata = head; /* to be released by nobh_write_end */ |
a4b0672d | 2672 | |
1da177e4 LT |
2673 | return 0; |
2674 | ||
2675 | failed: | |
03158cd7 | 2676 | BUG_ON(!ret); |
1da177e4 | 2677 | /* |
a4b0672d NP |
2678 | * Error recovery is a bit difficult. We need to zero out blocks that |
2679 | * were newly allocated, and dirty them to ensure they get written out. | |
2680 | * Buffers need to be attached to the page at this point, otherwise | |
2681 | * the handling of potential IO errors during writeout would be hard | |
2682 | * (could try doing synchronous writeout, but what if that fails too?) | |
1da177e4 | 2683 | */ |
03158cd7 NP |
2684 | attach_nobh_buffers(page, head); |
2685 | page_zero_new_buffers(page, from, to); | |
a4b0672d | 2686 | |
03158cd7 NP |
2687 | out_release: |
2688 | unlock_page(page); | |
09cbfeaf | 2689 | put_page(page); |
03158cd7 | 2690 | *pagep = NULL; |
a4b0672d | 2691 | |
7bb46a67 | 2692 | return ret; |
2693 | } | |
03158cd7 | 2694 | EXPORT_SYMBOL(nobh_write_begin); |
1da177e4 | 2695 | |
03158cd7 NP |
2696 | int nobh_write_end(struct file *file, struct address_space *mapping, |
2697 | loff_t pos, unsigned len, unsigned copied, | |
2698 | struct page *page, void *fsdata) | |
1da177e4 LT |
2699 | { |
2700 | struct inode *inode = page->mapping->host; | |
efdc3131 | 2701 | struct buffer_head *head = fsdata; |
03158cd7 | 2702 | struct buffer_head *bh; |
5b41e74a | 2703 | BUG_ON(fsdata != NULL && page_has_buffers(page)); |
1da177e4 | 2704 | |
d4cf109f | 2705 | if (unlikely(copied < len) && head) |
5b41e74a DM |
2706 | attach_nobh_buffers(page, head); |
2707 | if (page_has_buffers(page)) | |
2708 | return generic_write_end(file, mapping, pos, len, | |
2709 | copied, page, fsdata); | |
a4b0672d | 2710 | |
22c8ca78 | 2711 | SetPageUptodate(page); |
1da177e4 | 2712 | set_page_dirty(page); |
03158cd7 NP |
2713 | if (pos+copied > inode->i_size) { |
2714 | i_size_write(inode, pos+copied); | |
1da177e4 LT |
2715 | mark_inode_dirty(inode); |
2716 | } | |
03158cd7 NP |
2717 | |
2718 | unlock_page(page); | |
09cbfeaf | 2719 | put_page(page); |
03158cd7 | 2720 | |
03158cd7 NP |
2721 | while (head) { |
2722 | bh = head; | |
2723 | head = head->b_this_page; | |
2724 | free_buffer_head(bh); | |
2725 | } | |
2726 | ||
2727 | return copied; | |
1da177e4 | 2728 | } |
03158cd7 | 2729 | EXPORT_SYMBOL(nobh_write_end); |
1da177e4 LT |
2730 | |
2731 | /* | |
2732 | * nobh_writepage() - based on block_full_write_page() except | |
2733 | * that it tries to operate without attaching bufferheads to | |
2734 | * the page. | |
2735 | */ | |
2736 | int nobh_writepage(struct page *page, get_block_t *get_block, | |
2737 | struct writeback_control *wbc) | |
2738 | { | |
2739 | struct inode * const inode = page->mapping->host; | |
2740 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2741 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2742 | unsigned offset; |
1da177e4 LT |
2743 | int ret; |
2744 | ||
2745 | /* Is the page fully inside i_size? */ | |
2746 | if (page->index < end_index) | |
2747 | goto out; | |
2748 | ||
2749 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2750 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 LT |
2751 | if (page->index >= end_index+1 || !offset) { |
2752 | /* | |
2753 | * The page may have dirty, unmapped buffers. For example, | |
2754 | * they may have been added in ext3_writepage(). Make them | |
2755 | * freeable here, so the page does not leak. | |
2756 | */ | |
2757 | #if 0 | |
2758 | /* Not really sure about this - do we need this ? */ | |
2759 | if (page->mapping->a_ops->invalidatepage) | |
2760 | page->mapping->a_ops->invalidatepage(page, offset); | |
2761 | #endif | |
2762 | unlock_page(page); | |
2763 | return 0; /* don't care */ | |
2764 | } | |
2765 | ||
2766 | /* | |
2767 | * The page straddles i_size. It must be zeroed out on each and every | |
2768 | * writepage invocation because it may be mmapped. "A file is mapped | |
2769 | * in multiples of the page size. For a file that is not a multiple of | |
2770 | * the page size, the remaining memory is zeroed when mapped, and | |
2771 | * writes to that region are not written out to the file." | |
2772 | */ | |
09cbfeaf | 2773 | zero_user_segment(page, offset, PAGE_SIZE); |
1da177e4 LT |
2774 | out: |
2775 | ret = mpage_writepage(page, get_block, wbc); | |
2776 | if (ret == -EAGAIN) | |
35c80d5f CM |
2777 | ret = __block_write_full_page(inode, page, get_block, wbc, |
2778 | end_buffer_async_write); | |
1da177e4 LT |
2779 | return ret; |
2780 | } | |
2781 | EXPORT_SYMBOL(nobh_writepage); | |
2782 | ||
03158cd7 NP |
2783 | int nobh_truncate_page(struct address_space *mapping, |
2784 | loff_t from, get_block_t *get_block) | |
1da177e4 | 2785 | { |
09cbfeaf KS |
2786 | pgoff_t index = from >> PAGE_SHIFT; |
2787 | unsigned offset = from & (PAGE_SIZE-1); | |
03158cd7 NP |
2788 | unsigned blocksize; |
2789 | sector_t iblock; | |
2790 | unsigned length, pos; | |
2791 | struct inode *inode = mapping->host; | |
1da177e4 | 2792 | struct page *page; |
03158cd7 NP |
2793 | struct buffer_head map_bh; |
2794 | int err; | |
1da177e4 | 2795 | |
93407472 | 2796 | blocksize = i_blocksize(inode); |
03158cd7 NP |
2797 | length = offset & (blocksize - 1); |
2798 | ||
2799 | /* Block boundary? Nothing to do */ | |
2800 | if (!length) | |
2801 | return 0; | |
2802 | ||
2803 | length = blocksize - length; | |
09cbfeaf | 2804 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 | 2805 | |
1da177e4 | 2806 | page = grab_cache_page(mapping, index); |
03158cd7 | 2807 | err = -ENOMEM; |
1da177e4 LT |
2808 | if (!page) |
2809 | goto out; | |
2810 | ||
03158cd7 NP |
2811 | if (page_has_buffers(page)) { |
2812 | has_buffers: | |
2813 | unlock_page(page); | |
09cbfeaf | 2814 | put_page(page); |
03158cd7 NP |
2815 | return block_truncate_page(mapping, from, get_block); |
2816 | } | |
2817 | ||
2818 | /* Find the buffer that contains "offset" */ | |
2819 | pos = blocksize; | |
2820 | while (offset >= pos) { | |
2821 | iblock++; | |
2822 | pos += blocksize; | |
2823 | } | |
2824 | ||
460bcf57 TT |
2825 | map_bh.b_size = blocksize; |
2826 | map_bh.b_state = 0; | |
03158cd7 NP |
2827 | err = get_block(inode, iblock, &map_bh, 0); |
2828 | if (err) | |
2829 | goto unlock; | |
2830 | /* unmapped? It's a hole - nothing to do */ | |
2831 | if (!buffer_mapped(&map_bh)) | |
2832 | goto unlock; | |
2833 | ||
2834 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2835 | if (!PageUptodate(page)) { | |
2836 | err = mapping->a_ops->readpage(NULL, page); | |
2837 | if (err) { | |
09cbfeaf | 2838 | put_page(page); |
03158cd7 NP |
2839 | goto out; |
2840 | } | |
2841 | lock_page(page); | |
2842 | if (!PageUptodate(page)) { | |
2843 | err = -EIO; | |
2844 | goto unlock; | |
2845 | } | |
2846 | if (page_has_buffers(page)) | |
2847 | goto has_buffers; | |
1da177e4 | 2848 | } |
eebd2aa3 | 2849 | zero_user(page, offset, length); |
03158cd7 NP |
2850 | set_page_dirty(page); |
2851 | err = 0; | |
2852 | ||
2853 | unlock: | |
1da177e4 | 2854 | unlock_page(page); |
09cbfeaf | 2855 | put_page(page); |
1da177e4 | 2856 | out: |
03158cd7 | 2857 | return err; |
1da177e4 LT |
2858 | } |
2859 | EXPORT_SYMBOL(nobh_truncate_page); | |
2860 | ||
2861 | int block_truncate_page(struct address_space *mapping, | |
2862 | loff_t from, get_block_t *get_block) | |
2863 | { | |
09cbfeaf KS |
2864 | pgoff_t index = from >> PAGE_SHIFT; |
2865 | unsigned offset = from & (PAGE_SIZE-1); | |
1da177e4 | 2866 | unsigned blocksize; |
54b21a79 | 2867 | sector_t iblock; |
1da177e4 LT |
2868 | unsigned length, pos; |
2869 | struct inode *inode = mapping->host; | |
2870 | struct page *page; | |
2871 | struct buffer_head *bh; | |
1da177e4 LT |
2872 | int err; |
2873 | ||
93407472 | 2874 | blocksize = i_blocksize(inode); |
1da177e4 LT |
2875 | length = offset & (blocksize - 1); |
2876 | ||
2877 | /* Block boundary? Nothing to do */ | |
2878 | if (!length) | |
2879 | return 0; | |
2880 | ||
2881 | length = blocksize - length; | |
09cbfeaf | 2882 | iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits); |
1da177e4 LT |
2883 | |
2884 | page = grab_cache_page(mapping, index); | |
2885 | err = -ENOMEM; | |
2886 | if (!page) | |
2887 | goto out; | |
2888 | ||
2889 | if (!page_has_buffers(page)) | |
2890 | create_empty_buffers(page, blocksize, 0); | |
2891 | ||
2892 | /* Find the buffer that contains "offset" */ | |
2893 | bh = page_buffers(page); | |
2894 | pos = blocksize; | |
2895 | while (offset >= pos) { | |
2896 | bh = bh->b_this_page; | |
2897 | iblock++; | |
2898 | pos += blocksize; | |
2899 | } | |
2900 | ||
2901 | err = 0; | |
2902 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2903 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
2904 | err = get_block(inode, iblock, bh, 0); |
2905 | if (err) | |
2906 | goto unlock; | |
2907 | /* unmapped? It's a hole - nothing to do */ | |
2908 | if (!buffer_mapped(bh)) | |
2909 | goto unlock; | |
2910 | } | |
2911 | ||
2912 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2913 | if (PageUptodate(page)) | |
2914 | set_buffer_uptodate(bh); | |
2915 | ||
33a266dd | 2916 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
1da177e4 | 2917 | err = -EIO; |
dfec8a14 | 2918 | ll_rw_block(REQ_OP_READ, 0, 1, &bh); |
1da177e4 LT |
2919 | wait_on_buffer(bh); |
2920 | /* Uhhuh. Read error. Complain and punt. */ | |
2921 | if (!buffer_uptodate(bh)) | |
2922 | goto unlock; | |
2923 | } | |
2924 | ||
eebd2aa3 | 2925 | zero_user(page, offset, length); |
1da177e4 LT |
2926 | mark_buffer_dirty(bh); |
2927 | err = 0; | |
2928 | ||
2929 | unlock: | |
2930 | unlock_page(page); | |
09cbfeaf | 2931 | put_page(page); |
1da177e4 LT |
2932 | out: |
2933 | return err; | |
2934 | } | |
1fe72eaa | 2935 | EXPORT_SYMBOL(block_truncate_page); |
1da177e4 LT |
2936 | |
2937 | /* | |
2938 | * The generic ->writepage function for buffer-backed address_spaces | |
2939 | */ | |
1b938c08 MW |
2940 | int block_write_full_page(struct page *page, get_block_t *get_block, |
2941 | struct writeback_control *wbc) | |
1da177e4 LT |
2942 | { |
2943 | struct inode * const inode = page->mapping->host; | |
2944 | loff_t i_size = i_size_read(inode); | |
09cbfeaf | 2945 | const pgoff_t end_index = i_size >> PAGE_SHIFT; |
1da177e4 | 2946 | unsigned offset; |
1da177e4 LT |
2947 | |
2948 | /* Is the page fully inside i_size? */ | |
2949 | if (page->index < end_index) | |
35c80d5f | 2950 | return __block_write_full_page(inode, page, get_block, wbc, |
1b938c08 | 2951 | end_buffer_async_write); |
1da177e4 LT |
2952 | |
2953 | /* Is the page fully outside i_size? (truncate in progress) */ | |
09cbfeaf | 2954 | offset = i_size & (PAGE_SIZE-1); |
1da177e4 LT |
2955 | if (page->index >= end_index+1 || !offset) { |
2956 | /* | |
2957 | * The page may have dirty, unmapped buffers. For example, | |
2958 | * they may have been added in ext3_writepage(). Make them | |
2959 | * freeable here, so the page does not leak. | |
2960 | */ | |
09cbfeaf | 2961 | do_invalidatepage(page, 0, PAGE_SIZE); |
1da177e4 LT |
2962 | unlock_page(page); |
2963 | return 0; /* don't care */ | |
2964 | } | |
2965 | ||
2966 | /* | |
2967 | * The page straddles i_size. It must be zeroed out on each and every | |
2a61aa40 | 2968 | * writepage invocation because it may be mmapped. "A file is mapped |
1da177e4 LT |
2969 | * in multiples of the page size. For a file that is not a multiple of |
2970 | * the page size, the remaining memory is zeroed when mapped, and | |
2971 | * writes to that region are not written out to the file." | |
2972 | */ | |
09cbfeaf | 2973 | zero_user_segment(page, offset, PAGE_SIZE); |
1b938c08 MW |
2974 | return __block_write_full_page(inode, page, get_block, wbc, |
2975 | end_buffer_async_write); | |
35c80d5f | 2976 | } |
1fe72eaa | 2977 | EXPORT_SYMBOL(block_write_full_page); |
35c80d5f | 2978 | |
1da177e4 LT |
2979 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
2980 | get_block_t *get_block) | |
2981 | { | |
1da177e4 | 2982 | struct inode *inode = mapping->host; |
2a527d68 AP |
2983 | struct buffer_head tmp = { |
2984 | .b_size = i_blocksize(inode), | |
2985 | }; | |
2986 | ||
1da177e4 LT |
2987 | get_block(inode, block, &tmp, 0); |
2988 | return tmp.b_blocknr; | |
2989 | } | |
1fe72eaa | 2990 | EXPORT_SYMBOL(generic_block_bmap); |
1da177e4 | 2991 | |
4246a0b6 | 2992 | static void end_bio_bh_io_sync(struct bio *bio) |
1da177e4 LT |
2993 | { |
2994 | struct buffer_head *bh = bio->bi_private; | |
2995 | ||
b7c44ed9 | 2996 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
08bafc03 KM |
2997 | set_bit(BH_Quiet, &bh->b_state); |
2998 | ||
4e4cbee9 | 2999 | bh->b_end_io(bh, !bio->bi_status); |
1da177e4 | 3000 | bio_put(bio); |
1da177e4 LT |
3001 | } |
3002 | ||
57302e0d LT |
3003 | /* |
3004 | * This allows us to do IO even on the odd last sectors | |
59d43914 | 3005 | * of a device, even if the block size is some multiple |
57302e0d LT |
3006 | * of the physical sector size. |
3007 | * | |
3008 | * We'll just truncate the bio to the size of the device, | |
3009 | * and clear the end of the buffer head manually. | |
3010 | * | |
3011 | * Truly out-of-range accesses will turn into actual IO | |
3012 | * errors, this only handles the "we need to be able to | |
3013 | * do IO at the final sector" case. | |
3014 | */ | |
2a222ca9 | 3015 | void guard_bio_eod(int op, struct bio *bio) |
57302e0d LT |
3016 | { |
3017 | sector_t maxsector; | |
59d43914 AM |
3018 | struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
3019 | unsigned truncated_bytes; | |
67f2519f GE |
3020 | struct hd_struct *part; |
3021 | ||
3022 | rcu_read_lock(); | |
3023 | part = __disk_get_part(bio->bi_disk, bio->bi_partno); | |
3024 | if (part) | |
3025 | maxsector = part_nr_sects_read(part); | |
3026 | else | |
3027 | maxsector = get_capacity(bio->bi_disk); | |
3028 | rcu_read_unlock(); | |
57302e0d | 3029 | |
57302e0d LT |
3030 | if (!maxsector) |
3031 | return; | |
3032 | ||
3033 | /* | |
3034 | * If the *whole* IO is past the end of the device, | |
3035 | * let it through, and the IO layer will turn it into | |
3036 | * an EIO. | |
3037 | */ | |
4f024f37 | 3038 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) |
57302e0d LT |
3039 | return; |
3040 | ||
4f024f37 | 3041 | maxsector -= bio->bi_iter.bi_sector; |
59d43914 | 3042 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) |
57302e0d LT |
3043 | return; |
3044 | ||
59d43914 AM |
3045 | /* Uhhuh. We've got a bio that straddles the device size! */ |
3046 | truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9); | |
57302e0d | 3047 | |
13ea0fc8 CM |
3048 | /* |
3049 | * The bio contains more than one segment which spans EOD, just return | |
3050 | * and let IO layer turn it into an EIO | |
3051 | */ | |
3052 | if (truncated_bytes > bvec->bv_len) | |
3053 | return; | |
3054 | ||
57302e0d | 3055 | /* Truncate the bio.. */ |
59d43914 AM |
3056 | bio->bi_iter.bi_size -= truncated_bytes; |
3057 | bvec->bv_len -= truncated_bytes; | |
57302e0d LT |
3058 | |
3059 | /* ..and clear the end of the buffer for reads */ | |
2a222ca9 | 3060 | if (op == REQ_OP_READ) { |
59d43914 AM |
3061 | zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len, |
3062 | truncated_bytes); | |
57302e0d LT |
3063 | } |
3064 | } | |
3065 | ||
2a222ca9 | 3066 | static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh, |
8e8f9298 | 3067 | enum rw_hint write_hint, struct writeback_control *wbc) |
1da177e4 LT |
3068 | { |
3069 | struct bio *bio; | |
1da177e4 LT |
3070 | |
3071 | BUG_ON(!buffer_locked(bh)); | |
3072 | BUG_ON(!buffer_mapped(bh)); | |
3073 | BUG_ON(!bh->b_end_io); | |
8fb0e342 AK |
3074 | BUG_ON(buffer_delay(bh)); |
3075 | BUG_ON(buffer_unwritten(bh)); | |
1da177e4 | 3076 | |
1da177e4 | 3077 | /* |
48fd4f93 | 3078 | * Only clear out a write error when rewriting |
1da177e4 | 3079 | */ |
2a222ca9 | 3080 | if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) |
1da177e4 LT |
3081 | clear_buffer_write_io_error(bh); |
3082 | ||
3083 | /* | |
3084 | * from here on down, it's all bio -- do the initial mapping, | |
3085 | * submit_bio -> generic_make_request may further map this bio around | |
3086 | */ | |
3087 | bio = bio_alloc(GFP_NOIO, 1); | |
3088 | ||
2a814908 | 3089 | if (wbc) { |
b16b1deb | 3090 | wbc_init_bio(wbc, bio); |
2a814908 TH |
3091 | wbc_account_io(wbc, bh->b_page, bh->b_size); |
3092 | } | |
bafc0dba | 3093 | |
4f024f37 | 3094 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
74d46992 | 3095 | bio_set_dev(bio, bh->b_bdev); |
8e8f9298 | 3096 | bio->bi_write_hint = write_hint; |
1da177e4 | 3097 | |
6cf66b4c KO |
3098 | bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
3099 | BUG_ON(bio->bi_iter.bi_size != bh->b_size); | |
1da177e4 LT |
3100 | |
3101 | bio->bi_end_io = end_bio_bh_io_sync; | |
3102 | bio->bi_private = bh; | |
3103 | ||
57302e0d | 3104 | /* Take care of bh's that straddle the end of the device */ |
2a222ca9 | 3105 | guard_bio_eod(op, bio); |
57302e0d | 3106 | |
877f962c | 3107 | if (buffer_meta(bh)) |
2a222ca9 | 3108 | op_flags |= REQ_META; |
877f962c | 3109 | if (buffer_prio(bh)) |
2a222ca9 MC |
3110 | op_flags |= REQ_PRIO; |
3111 | bio_set_op_attrs(bio, op, op_flags); | |
877f962c | 3112 | |
4e49ea4a | 3113 | submit_bio(bio); |
f6454b04 | 3114 | return 0; |
1da177e4 | 3115 | } |
bafc0dba | 3116 | |
020c2833 | 3117 | int submit_bh(int op, int op_flags, struct buffer_head *bh) |
bafc0dba | 3118 | { |
8e8f9298 | 3119 | return submit_bh_wbc(op, op_flags, bh, 0, NULL); |
71368511 | 3120 | } |
1fe72eaa | 3121 | EXPORT_SYMBOL(submit_bh); |
1da177e4 LT |
3122 | |
3123 | /** | |
3124 | * ll_rw_block: low-level access to block devices (DEPRECATED) | |
dfec8a14 | 3125 | * @op: whether to %READ or %WRITE |
ef295ecf | 3126 | * @op_flags: req_flag_bits |
1da177e4 LT |
3127 | * @nr: number of &struct buffer_heads in the array |
3128 | * @bhs: array of pointers to &struct buffer_head | |
3129 | * | |
a7662236 | 3130 | * ll_rw_block() takes an array of pointers to &struct buffer_heads, and |
70246286 CH |
3131 | * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE. |
3132 | * @op_flags contains flags modifying the detailed I/O behavior, most notably | |
3133 | * %REQ_RAHEAD. | |
1da177e4 LT |
3134 | * |
3135 | * This function drops any buffer that it cannot get a lock on (with the | |
9cb569d6 CH |
3136 | * BH_Lock state bit), any buffer that appears to be clean when doing a write |
3137 | * request, and any buffer that appears to be up-to-date when doing read | |
3138 | * request. Further it marks as clean buffers that are processed for | |
3139 | * writing (the buffer cache won't assume that they are actually clean | |
3140 | * until the buffer gets unlocked). | |
1da177e4 LT |
3141 | * |
3142 | * ll_rw_block sets b_end_io to simple completion handler that marks | |
e227867f | 3143 | * the buffer up-to-date (if appropriate), unlocks the buffer and wakes |
1da177e4 LT |
3144 | * any waiters. |
3145 | * | |
3146 | * All of the buffers must be for the same device, and must also be a | |
3147 | * multiple of the current approved size for the device. | |
3148 | */ | |
dfec8a14 | 3149 | void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[]) |
1da177e4 LT |
3150 | { |
3151 | int i; | |
3152 | ||
3153 | for (i = 0; i < nr; i++) { | |
3154 | struct buffer_head *bh = bhs[i]; | |
3155 | ||
9cb569d6 | 3156 | if (!trylock_buffer(bh)) |
1da177e4 | 3157 | continue; |
dfec8a14 | 3158 | if (op == WRITE) { |
1da177e4 | 3159 | if (test_clear_buffer_dirty(bh)) { |
76c3073a | 3160 | bh->b_end_io = end_buffer_write_sync; |
e60e5c50 | 3161 | get_bh(bh); |
dfec8a14 | 3162 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3163 | continue; |
3164 | } | |
3165 | } else { | |
1da177e4 | 3166 | if (!buffer_uptodate(bh)) { |
76c3073a | 3167 | bh->b_end_io = end_buffer_read_sync; |
e60e5c50 | 3168 | get_bh(bh); |
dfec8a14 | 3169 | submit_bh(op, op_flags, bh); |
1da177e4 LT |
3170 | continue; |
3171 | } | |
3172 | } | |
3173 | unlock_buffer(bh); | |
1da177e4 LT |
3174 | } |
3175 | } | |
1fe72eaa | 3176 | EXPORT_SYMBOL(ll_rw_block); |
1da177e4 | 3177 | |
2a222ca9 | 3178 | void write_dirty_buffer(struct buffer_head *bh, int op_flags) |
9cb569d6 CH |
3179 | { |
3180 | lock_buffer(bh); | |
3181 | if (!test_clear_buffer_dirty(bh)) { | |
3182 | unlock_buffer(bh); | |
3183 | return; | |
3184 | } | |
3185 | bh->b_end_io = end_buffer_write_sync; | |
3186 | get_bh(bh); | |
2a222ca9 | 3187 | submit_bh(REQ_OP_WRITE, op_flags, bh); |
9cb569d6 CH |
3188 | } |
3189 | EXPORT_SYMBOL(write_dirty_buffer); | |
3190 | ||
1da177e4 LT |
3191 | /* |
3192 | * For a data-integrity writeout, we need to wait upon any in-progress I/O | |
3193 | * and then start new I/O and then wait upon it. The caller must have a ref on | |
3194 | * the buffer_head. | |
3195 | */ | |
2a222ca9 | 3196 | int __sync_dirty_buffer(struct buffer_head *bh, int op_flags) |
1da177e4 LT |
3197 | { |
3198 | int ret = 0; | |
3199 | ||
3200 | WARN_ON(atomic_read(&bh->b_count) < 1); | |
3201 | lock_buffer(bh); | |
3202 | if (test_clear_buffer_dirty(bh)) { | |
3203 | get_bh(bh); | |
3204 | bh->b_end_io = end_buffer_write_sync; | |
2a222ca9 | 3205 | ret = submit_bh(REQ_OP_WRITE, op_flags, bh); |
1da177e4 | 3206 | wait_on_buffer(bh); |
1da177e4 LT |
3207 | if (!ret && !buffer_uptodate(bh)) |
3208 | ret = -EIO; | |
3209 | } else { | |
3210 | unlock_buffer(bh); | |
3211 | } | |
3212 | return ret; | |
3213 | } | |
87e99511 CH |
3214 | EXPORT_SYMBOL(__sync_dirty_buffer); |
3215 | ||
3216 | int sync_dirty_buffer(struct buffer_head *bh) | |
3217 | { | |
70fd7614 | 3218 | return __sync_dirty_buffer(bh, REQ_SYNC); |
87e99511 | 3219 | } |
1fe72eaa | 3220 | EXPORT_SYMBOL(sync_dirty_buffer); |
1da177e4 LT |
3221 | |
3222 | /* | |
3223 | * try_to_free_buffers() checks if all the buffers on this particular page | |
3224 | * are unused, and releases them if so. | |
3225 | * | |
3226 | * Exclusion against try_to_free_buffers may be obtained by either | |
3227 | * locking the page or by holding its mapping's private_lock. | |
3228 | * | |
3229 | * If the page is dirty but all the buffers are clean then we need to | |
3230 | * be sure to mark the page clean as well. This is because the page | |
3231 | * may be against a block device, and a later reattachment of buffers | |
3232 | * to a dirty page will set *all* buffers dirty. Which would corrupt | |
3233 | * filesystem data on the same device. | |
3234 | * | |
3235 | * The same applies to regular filesystem pages: if all the buffers are | |
3236 | * clean then we set the page clean and proceed. To do that, we require | |
3237 | * total exclusion from __set_page_dirty_buffers(). That is obtained with | |
3238 | * private_lock. | |
3239 | * | |
3240 | * try_to_free_buffers() is non-blocking. | |
3241 | */ | |
3242 | static inline int buffer_busy(struct buffer_head *bh) | |
3243 | { | |
3244 | return atomic_read(&bh->b_count) | | |
3245 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); | |
3246 | } | |
3247 | ||
3248 | static int | |
3249 | drop_buffers(struct page *page, struct buffer_head **buffers_to_free) | |
3250 | { | |
3251 | struct buffer_head *head = page_buffers(page); | |
3252 | struct buffer_head *bh; | |
3253 | ||
3254 | bh = head; | |
3255 | do { | |
1da177e4 LT |
3256 | if (buffer_busy(bh)) |
3257 | goto failed; | |
3258 | bh = bh->b_this_page; | |
3259 | } while (bh != head); | |
3260 | ||
3261 | do { | |
3262 | struct buffer_head *next = bh->b_this_page; | |
3263 | ||
535ee2fb | 3264 | if (bh->b_assoc_map) |
1da177e4 LT |
3265 | __remove_assoc_queue(bh); |
3266 | bh = next; | |
3267 | } while (bh != head); | |
3268 | *buffers_to_free = head; | |
3269 | __clear_page_buffers(page); | |
3270 | return 1; | |
3271 | failed: | |
3272 | return 0; | |
3273 | } | |
3274 | ||
3275 | int try_to_free_buffers(struct page *page) | |
3276 | { | |
3277 | struct address_space * const mapping = page->mapping; | |
3278 | struct buffer_head *buffers_to_free = NULL; | |
3279 | int ret = 0; | |
3280 | ||
3281 | BUG_ON(!PageLocked(page)); | |
ecdfc978 | 3282 | if (PageWriteback(page)) |
1da177e4 LT |
3283 | return 0; |
3284 | ||
3285 | if (mapping == NULL) { /* can this still happen? */ | |
3286 | ret = drop_buffers(page, &buffers_to_free); | |
3287 | goto out; | |
3288 | } | |
3289 | ||
3290 | spin_lock(&mapping->private_lock); | |
3291 | ret = drop_buffers(page, &buffers_to_free); | |
ecdfc978 LT |
3292 | |
3293 | /* | |
3294 | * If the filesystem writes its buffers by hand (eg ext3) | |
3295 | * then we can have clean buffers against a dirty page. We | |
3296 | * clean the page here; otherwise the VM will never notice | |
3297 | * that the filesystem did any IO at all. | |
3298 | * | |
3299 | * Also, during truncate, discard_buffer will have marked all | |
3300 | * the page's buffers clean. We discover that here and clean | |
3301 | * the page also. | |
87df7241 NP |
3302 | * |
3303 | * private_lock must be held over this entire operation in order | |
3304 | * to synchronise against __set_page_dirty_buffers and prevent the | |
3305 | * dirty bit from being lost. | |
ecdfc978 | 3306 | */ |
11f81bec TH |
3307 | if (ret) |
3308 | cancel_dirty_page(page); | |
87df7241 | 3309 | spin_unlock(&mapping->private_lock); |
1da177e4 LT |
3310 | out: |
3311 | if (buffers_to_free) { | |
3312 | struct buffer_head *bh = buffers_to_free; | |
3313 | ||
3314 | do { | |
3315 | struct buffer_head *next = bh->b_this_page; | |
3316 | free_buffer_head(bh); | |
3317 | bh = next; | |
3318 | } while (bh != buffers_to_free); | |
3319 | } | |
3320 | return ret; | |
3321 | } | |
3322 | EXPORT_SYMBOL(try_to_free_buffers); | |
3323 | ||
1da177e4 LT |
3324 | /* |
3325 | * There are no bdflush tunables left. But distributions are | |
3326 | * still running obsolete flush daemons, so we terminate them here. | |
3327 | * | |
3328 | * Use of bdflush() is deprecated and will be removed in a future kernel. | |
5b0830cb | 3329 | * The `flush-X' kernel threads fully replace bdflush daemons and this call. |
1da177e4 | 3330 | */ |
bdc480e3 | 3331 | SYSCALL_DEFINE2(bdflush, int, func, long, data) |
1da177e4 LT |
3332 | { |
3333 | static int msg_count; | |
3334 | ||
3335 | if (!capable(CAP_SYS_ADMIN)) | |
3336 | return -EPERM; | |
3337 | ||
3338 | if (msg_count < 5) { | |
3339 | msg_count++; | |
3340 | printk(KERN_INFO | |
3341 | "warning: process `%s' used the obsolete bdflush" | |
3342 | " system call\n", current->comm); | |
3343 | printk(KERN_INFO "Fix your initscripts?\n"); | |
3344 | } | |
3345 | ||
3346 | if (func == 1) | |
3347 | do_exit(0); | |
3348 | return 0; | |
3349 | } | |
3350 | ||
3351 | /* | |
3352 | * Buffer-head allocation | |
3353 | */ | |
a0a9b043 | 3354 | static struct kmem_cache *bh_cachep __read_mostly; |
1da177e4 LT |
3355 | |
3356 | /* | |
3357 | * Once the number of bh's in the machine exceeds this level, we start | |
3358 | * stripping them in writeback. | |
3359 | */ | |
43be594a | 3360 | static unsigned long max_buffer_heads; |
1da177e4 LT |
3361 | |
3362 | int buffer_heads_over_limit; | |
3363 | ||
3364 | struct bh_accounting { | |
3365 | int nr; /* Number of live bh's */ | |
3366 | int ratelimit; /* Limit cacheline bouncing */ | |
3367 | }; | |
3368 | ||
3369 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; | |
3370 | ||
3371 | static void recalc_bh_state(void) | |
3372 | { | |
3373 | int i; | |
3374 | int tot = 0; | |
3375 | ||
ee1be862 | 3376 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
1da177e4 | 3377 | return; |
c7b92516 | 3378 | __this_cpu_write(bh_accounting.ratelimit, 0); |
8a143426 | 3379 | for_each_online_cpu(i) |
1da177e4 LT |
3380 | tot += per_cpu(bh_accounting, i).nr; |
3381 | buffer_heads_over_limit = (tot > max_buffer_heads); | |
3382 | } | |
c7b92516 | 3383 | |
dd0fc66f | 3384 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
1da177e4 | 3385 | { |
019b4d12 | 3386 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
1da177e4 | 3387 | if (ret) { |
a35afb83 | 3388 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
c7b92516 CL |
3389 | preempt_disable(); |
3390 | __this_cpu_inc(bh_accounting.nr); | |
1da177e4 | 3391 | recalc_bh_state(); |
c7b92516 | 3392 | preempt_enable(); |
1da177e4 LT |
3393 | } |
3394 | return ret; | |
3395 | } | |
3396 | EXPORT_SYMBOL(alloc_buffer_head); | |
3397 | ||
3398 | void free_buffer_head(struct buffer_head *bh) | |
3399 | { | |
3400 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); | |
3401 | kmem_cache_free(bh_cachep, bh); | |
c7b92516 CL |
3402 | preempt_disable(); |
3403 | __this_cpu_dec(bh_accounting.nr); | |
1da177e4 | 3404 | recalc_bh_state(); |
c7b92516 | 3405 | preempt_enable(); |
1da177e4 LT |
3406 | } |
3407 | EXPORT_SYMBOL(free_buffer_head); | |
3408 | ||
fc4d24c9 | 3409 | static int buffer_exit_cpu_dead(unsigned int cpu) |
1da177e4 LT |
3410 | { |
3411 | int i; | |
3412 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); | |
3413 | ||
3414 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
3415 | brelse(b->bhs[i]); | |
3416 | b->bhs[i] = NULL; | |
3417 | } | |
c7b92516 | 3418 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
8a143426 | 3419 | per_cpu(bh_accounting, cpu).nr = 0; |
fc4d24c9 | 3420 | return 0; |
1da177e4 | 3421 | } |
1da177e4 | 3422 | |
389d1b08 | 3423 | /** |
a6b91919 | 3424 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
389d1b08 AK |
3425 | * @bh: struct buffer_head |
3426 | * | |
3427 | * Return true if the buffer is up-to-date and false, | |
3428 | * with the buffer locked, if not. | |
3429 | */ | |
3430 | int bh_uptodate_or_lock(struct buffer_head *bh) | |
3431 | { | |
3432 | if (!buffer_uptodate(bh)) { | |
3433 | lock_buffer(bh); | |
3434 | if (!buffer_uptodate(bh)) | |
3435 | return 0; | |
3436 | unlock_buffer(bh); | |
3437 | } | |
3438 | return 1; | |
3439 | } | |
3440 | EXPORT_SYMBOL(bh_uptodate_or_lock); | |
3441 | ||
3442 | /** | |
a6b91919 | 3443 | * bh_submit_read - Submit a locked buffer for reading |
389d1b08 AK |
3444 | * @bh: struct buffer_head |
3445 | * | |
3446 | * Returns zero on success and -EIO on error. | |
3447 | */ | |
3448 | int bh_submit_read(struct buffer_head *bh) | |
3449 | { | |
3450 | BUG_ON(!buffer_locked(bh)); | |
3451 | ||
3452 | if (buffer_uptodate(bh)) { | |
3453 | unlock_buffer(bh); | |
3454 | return 0; | |
3455 | } | |
3456 | ||
3457 | get_bh(bh); | |
3458 | bh->b_end_io = end_buffer_read_sync; | |
2a222ca9 | 3459 | submit_bh(REQ_OP_READ, 0, bh); |
389d1b08 AK |
3460 | wait_on_buffer(bh); |
3461 | if (buffer_uptodate(bh)) | |
3462 | return 0; | |
3463 | return -EIO; | |
3464 | } | |
3465 | EXPORT_SYMBOL(bh_submit_read); | |
3466 | ||
334fd34d AG |
3467 | /* |
3468 | * Seek for SEEK_DATA / SEEK_HOLE within @page, starting at @lastoff. | |
3469 | * | |
3470 | * Returns the offset within the file on success, and -ENOENT otherwise. | |
3471 | */ | |
3472 | static loff_t | |
3473 | page_seek_hole_data(struct page *page, loff_t lastoff, int whence) | |
3474 | { | |
3475 | loff_t offset = page_offset(page); | |
3476 | struct buffer_head *bh, *head; | |
3477 | bool seek_data = whence == SEEK_DATA; | |
3478 | ||
3479 | if (lastoff < offset) | |
3480 | lastoff = offset; | |
3481 | ||
3482 | bh = head = page_buffers(page); | |
3483 | do { | |
3484 | offset += bh->b_size; | |
3485 | if (lastoff >= offset) | |
3486 | continue; | |
3487 | ||
3488 | /* | |
3489 | * Unwritten extents that have data in the page cache covering | |
3490 | * them can be identified by the BH_Unwritten state flag. | |
3491 | * Pages with multiple buffers might have a mix of holes, data | |
3492 | * and unwritten extents - any buffer with valid data in it | |
3493 | * should have BH_Uptodate flag set on it. | |
3494 | */ | |
3495 | ||
3496 | if ((buffer_unwritten(bh) || buffer_uptodate(bh)) == seek_data) | |
3497 | return lastoff; | |
3498 | ||
3499 | lastoff = offset; | |
3500 | } while ((bh = bh->b_this_page) != head); | |
3501 | return -ENOENT; | |
3502 | } | |
3503 | ||
3504 | /* | |
3505 | * Seek for SEEK_DATA / SEEK_HOLE in the page cache. | |
3506 | * | |
3507 | * Within unwritten extents, the page cache determines which parts are holes | |
3508 | * and which are data: unwritten and uptodate buffer heads count as data; | |
3509 | * everything else counts as a hole. | |
3510 | * | |
3511 | * Returns the resulting offset on successs, and -ENOENT otherwise. | |
3512 | */ | |
3513 | loff_t | |
3514 | page_cache_seek_hole_data(struct inode *inode, loff_t offset, loff_t length, | |
3515 | int whence) | |
3516 | { | |
3517 | pgoff_t index = offset >> PAGE_SHIFT; | |
3518 | pgoff_t end = DIV_ROUND_UP(offset + length, PAGE_SIZE); | |
3519 | loff_t lastoff = offset; | |
3520 | struct pagevec pvec; | |
3521 | ||
3522 | if (length <= 0) | |
3523 | return -ENOENT; | |
3524 | ||
86679820 | 3525 | pagevec_init(&pvec); |
334fd34d AG |
3526 | |
3527 | do { | |
8338141f | 3528 | unsigned nr_pages, i; |
334fd34d | 3529 | |
8338141f | 3530 | nr_pages = pagevec_lookup_range(&pvec, inode->i_mapping, &index, |
397162ff | 3531 | end - 1); |
334fd34d AG |
3532 | if (nr_pages == 0) |
3533 | break; | |
3534 | ||
3535 | for (i = 0; i < nr_pages; i++) { | |
3536 | struct page *page = pvec.pages[i]; | |
3537 | ||
3538 | /* | |
3539 | * At this point, the page may be truncated or | |
3540 | * invalidated (changing page->mapping to NULL), or | |
3541 | * even swizzled back from swapper_space to tmpfs file | |
3542 | * mapping. However, page->index will not change | |
3543 | * because we have a reference on the page. | |
3544 | * | |
3545 | * If current page offset is beyond where we've ended, | |
3546 | * we've found a hole. | |
3547 | */ | |
3548 | if (whence == SEEK_HOLE && | |
3549 | lastoff < page_offset(page)) | |
3550 | goto check_range; | |
3551 | ||
334fd34d AG |
3552 | lock_page(page); |
3553 | if (likely(page->mapping == inode->i_mapping) && | |
3554 | page_has_buffers(page)) { | |
3555 | lastoff = page_seek_hole_data(page, lastoff, whence); | |
3556 | if (lastoff >= 0) { | |
3557 | unlock_page(page); | |
3558 | goto check_range; | |
3559 | } | |
3560 | } | |
3561 | unlock_page(page); | |
3562 | lastoff = page_offset(page) + PAGE_SIZE; | |
3563 | } | |
334fd34d AG |
3564 | pagevec_release(&pvec); |
3565 | } while (index < end); | |
3566 | ||
3567 | /* When no page at lastoff and we are not done, we found a hole. */ | |
3568 | if (whence != SEEK_HOLE) | |
3569 | goto not_found; | |
3570 | ||
3571 | check_range: | |
3572 | if (lastoff < offset + length) | |
3573 | goto out; | |
3574 | not_found: | |
3575 | lastoff = -ENOENT; | |
3576 | out: | |
3577 | pagevec_release(&pvec); | |
3578 | return lastoff; | |
3579 | } | |
3580 | ||
1da177e4 LT |
3581 | void __init buffer_init(void) |
3582 | { | |
43be594a | 3583 | unsigned long nrpages; |
fc4d24c9 | 3584 | int ret; |
1da177e4 | 3585 | |
b98938c3 CL |
3586 | bh_cachep = kmem_cache_create("buffer_head", |
3587 | sizeof(struct buffer_head), 0, | |
3588 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
3589 | SLAB_MEM_SPREAD), | |
019b4d12 | 3590 | NULL); |
1da177e4 LT |
3591 | |
3592 | /* | |
3593 | * Limit the bh occupancy to 10% of ZONE_NORMAL | |
3594 | */ | |
3595 | nrpages = (nr_free_buffer_pages() * 10) / 100; | |
3596 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); | |
fc4d24c9 SAS |
3597 | ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", |
3598 | NULL, buffer_exit_cpu_dead); | |
3599 | WARN_ON(ret < 0); | |
1da177e4 | 3600 | } |