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 LT |
21 | #include <linux/kernel.h> |
22 | #include <linux/syscalls.h> | |
23 | #include <linux/fs.h> | |
24 | #include <linux/mm.h> | |
25 | #include <linux/percpu.h> | |
26 | #include <linux/slab.h> | |
16f7e0fe | 27 | #include <linux/capability.h> |
1da177e4 LT |
28 | #include <linux/blkdev.h> |
29 | #include <linux/file.h> | |
30 | #include <linux/quotaops.h> | |
31 | #include <linux/highmem.h> | |
630d9c47 | 32 | #include <linux/export.h> |
bafc0dba | 33 | #include <linux/backing-dev.h> |
1da177e4 LT |
34 | #include <linux/writeback.h> |
35 | #include <linux/hash.h> | |
36 | #include <linux/suspend.h> | |
37 | #include <linux/buffer_head.h> | |
55e829af | 38 | #include <linux/task_io_accounting_ops.h> |
1da177e4 LT |
39 | #include <linux/bio.h> |
40 | #include <linux/notifier.h> | |
41 | #include <linux/cpu.h> | |
42 | #include <linux/bitops.h> | |
43 | #include <linux/mpage.h> | |
fb1c8f93 | 44 | #include <linux/bit_spinlock.h> |
5305cb83 | 45 | #include <trace/events/block.h> |
1da177e4 LT |
46 | |
47 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); | |
b16b1deb TH |
48 | static int submit_bh_wbc(int rw, struct buffer_head *bh, |
49 | unsigned long bio_flags, | |
50 | struct writeback_control *wbc); | |
1da177e4 LT |
51 | |
52 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) | |
53 | ||
a3f3c29c | 54 | void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private) |
1da177e4 LT |
55 | { |
56 | bh->b_end_io = handler; | |
57 | bh->b_private = private; | |
58 | } | |
1fe72eaa | 59 | EXPORT_SYMBOL(init_buffer); |
1da177e4 | 60 | |
f0059afd TH |
61 | inline void touch_buffer(struct buffer_head *bh) |
62 | { | |
5305cb83 | 63 | trace_block_touch_buffer(bh); |
f0059afd TH |
64 | mark_page_accessed(bh->b_page); |
65 | } | |
66 | EXPORT_SYMBOL(touch_buffer); | |
67 | ||
fc9b52cd | 68 | void __lock_buffer(struct buffer_head *bh) |
1da177e4 | 69 | { |
74316201 | 70 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 LT |
71 | } |
72 | EXPORT_SYMBOL(__lock_buffer); | |
73 | ||
fc9b52cd | 74 | void unlock_buffer(struct buffer_head *bh) |
1da177e4 | 75 | { |
51b07fc3 | 76 | clear_bit_unlock(BH_Lock, &bh->b_state); |
4e857c58 | 77 | smp_mb__after_atomic(); |
1da177e4 LT |
78 | wake_up_bit(&bh->b_state, BH_Lock); |
79 | } | |
1fe72eaa | 80 | EXPORT_SYMBOL(unlock_buffer); |
1da177e4 | 81 | |
b4597226 MG |
82 | /* |
83 | * Returns if the page has dirty or writeback buffers. If all the buffers | |
84 | * are unlocked and clean then the PageDirty information is stale. If | |
85 | * any of the pages are locked, it is assumed they are locked for IO. | |
86 | */ | |
87 | void buffer_check_dirty_writeback(struct page *page, | |
88 | bool *dirty, bool *writeback) | |
89 | { | |
90 | struct buffer_head *head, *bh; | |
91 | *dirty = false; | |
92 | *writeback = false; | |
93 | ||
94 | BUG_ON(!PageLocked(page)); | |
95 | ||
96 | if (!page_has_buffers(page)) | |
97 | return; | |
98 | ||
99 | if (PageWriteback(page)) | |
100 | *writeback = true; | |
101 | ||
102 | head = page_buffers(page); | |
103 | bh = head; | |
104 | do { | |
105 | if (buffer_locked(bh)) | |
106 | *writeback = true; | |
107 | ||
108 | if (buffer_dirty(bh)) | |
109 | *dirty = true; | |
110 | ||
111 | bh = bh->b_this_page; | |
112 | } while (bh != head); | |
113 | } | |
114 | EXPORT_SYMBOL(buffer_check_dirty_writeback); | |
115 | ||
1da177e4 LT |
116 | /* |
117 | * Block until a buffer comes unlocked. This doesn't stop it | |
118 | * from becoming locked again - you have to lock it yourself | |
119 | * if you want to preserve its state. | |
120 | */ | |
121 | void __wait_on_buffer(struct buffer_head * bh) | |
122 | { | |
74316201 | 123 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
1da177e4 | 124 | } |
1fe72eaa | 125 | EXPORT_SYMBOL(__wait_on_buffer); |
1da177e4 LT |
126 | |
127 | static void | |
128 | __clear_page_buffers(struct page *page) | |
129 | { | |
130 | ClearPagePrivate(page); | |
4c21e2f2 | 131 | set_page_private(page, 0); |
1da177e4 LT |
132 | page_cache_release(page); |
133 | } | |
134 | ||
b744c2ac | 135 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
1da177e4 LT |
136 | { |
137 | char b[BDEVNAME_SIZE]; | |
432f16e6 RE |
138 | |
139 | if (!test_bit(BH_Quiet, &bh->b_state)) | |
140 | printk_ratelimited(KERN_ERR | |
141 | "Buffer I/O error on dev %s, logical block %llu%s\n", | |
1da177e4 | 142 | bdevname(bh->b_bdev, b), |
b744c2ac | 143 | (unsigned long long)bh->b_blocknr, msg); |
1da177e4 LT |
144 | } |
145 | ||
146 | /* | |
68671f35 DM |
147 | * End-of-IO handler helper function which does not touch the bh after |
148 | * unlocking it. | |
149 | * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but | |
150 | * a race there is benign: unlock_buffer() only use the bh's address for | |
151 | * hashing after unlocking the buffer, so it doesn't actually touch the bh | |
152 | * itself. | |
1da177e4 | 153 | */ |
68671f35 | 154 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
1da177e4 LT |
155 | { |
156 | if (uptodate) { | |
157 | set_buffer_uptodate(bh); | |
158 | } else { | |
159 | /* This happens, due to failed READA attempts. */ | |
160 | clear_buffer_uptodate(bh); | |
161 | } | |
162 | unlock_buffer(bh); | |
68671f35 DM |
163 | } |
164 | ||
165 | /* | |
166 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and | |
167 | * unlock the buffer. This is what ll_rw_block uses too. | |
168 | */ | |
169 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) | |
170 | { | |
171 | __end_buffer_read_notouch(bh, uptodate); | |
1da177e4 LT |
172 | put_bh(bh); |
173 | } | |
1fe72eaa | 174 | EXPORT_SYMBOL(end_buffer_read_sync); |
1da177e4 LT |
175 | |
176 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) | |
177 | { | |
1da177e4 LT |
178 | if (uptodate) { |
179 | set_buffer_uptodate(bh); | |
180 | } else { | |
432f16e6 | 181 | buffer_io_error(bh, ", lost sync page write"); |
1da177e4 LT |
182 | set_buffer_write_io_error(bh); |
183 | clear_buffer_uptodate(bh); | |
184 | } | |
185 | unlock_buffer(bh); | |
186 | put_bh(bh); | |
187 | } | |
1fe72eaa | 188 | EXPORT_SYMBOL(end_buffer_write_sync); |
1da177e4 | 189 | |
1da177e4 LT |
190 | /* |
191 | * Various filesystems appear to want __find_get_block to be non-blocking. | |
192 | * But it's the page lock which protects the buffers. To get around this, | |
193 | * we get exclusion from try_to_free_buffers with the blockdev mapping's | |
194 | * private_lock. | |
195 | * | |
196 | * Hack idea: for the blockdev mapping, i_bufferlist_lock contention | |
197 | * may be quite high. This code could TryLock the page, and if that | |
198 | * succeeds, there is no need to take private_lock. (But if | |
199 | * private_lock is contended then so is mapping->tree_lock). | |
200 | */ | |
201 | static struct buffer_head * | |
385fd4c5 | 202 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
1da177e4 LT |
203 | { |
204 | struct inode *bd_inode = bdev->bd_inode; | |
205 | struct address_space *bd_mapping = bd_inode->i_mapping; | |
206 | struct buffer_head *ret = NULL; | |
207 | pgoff_t index; | |
208 | struct buffer_head *bh; | |
209 | struct buffer_head *head; | |
210 | struct page *page; | |
211 | int all_mapped = 1; | |
212 | ||
213 | index = block >> (PAGE_CACHE_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 | */ | |
239 | if (all_mapped) { | |
72a2ebd8 TM |
240 | char b[BDEVNAME_SIZE]; |
241 | ||
1da177e4 LT |
242 | printk("__find_get_block_slow() failed. " |
243 | "block=%llu, b_blocknr=%llu\n", | |
205f87f6 BP |
244 | (unsigned long long)block, |
245 | (unsigned long long)bh->b_blocknr); | |
246 | printk("b_state=0x%08lx, b_size=%zu\n", | |
247 | bh->b_state, bh->b_size); | |
72a2ebd8 TM |
248 | printk("device %s blocksize: %d\n", bdevname(bdev, b), |
249 | 1 << bd_inode->i_blkbits); | |
1da177e4 LT |
250 | } |
251 | out_unlock: | |
252 | spin_unlock(&bd_mapping->private_lock); | |
253 | page_cache_release(page); | |
254 | out: | |
255 | return ret; | |
256 | } | |
257 | ||
1da177e4 | 258 | /* |
5b0830cb | 259 | * Kick the writeback threads then try to free up some ZONE_NORMAL memory. |
1da177e4 LT |
260 | */ |
261 | static void free_more_memory(void) | |
262 | { | |
19770b32 | 263 | struct zone *zone; |
0e88460d | 264 | int nid; |
1da177e4 | 265 | |
0e175a18 | 266 | wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM); |
1da177e4 LT |
267 | yield(); |
268 | ||
0e88460d | 269 | for_each_online_node(nid) { |
19770b32 MG |
270 | (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS), |
271 | gfp_zone(GFP_NOFS), NULL, | |
272 | &zone); | |
273 | if (zone) | |
54a6eb5c | 274 | try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0, |
327c0e96 | 275 | GFP_NOFS, NULL); |
1da177e4 LT |
276 | } |
277 | } | |
278 | ||
279 | /* | |
280 | * I/O completion handler for block_read_full_page() - pages | |
281 | * which come unlocked at the end of I/O. | |
282 | */ | |
283 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) | |
284 | { | |
1da177e4 | 285 | unsigned long flags; |
a3972203 | 286 | struct buffer_head *first; |
1da177e4 LT |
287 | struct buffer_head *tmp; |
288 | struct page *page; | |
289 | int page_uptodate = 1; | |
290 | ||
291 | BUG_ON(!buffer_async_read(bh)); | |
292 | ||
293 | page = bh->b_page; | |
294 | if (uptodate) { | |
295 | set_buffer_uptodate(bh); | |
296 | } else { | |
297 | clear_buffer_uptodate(bh); | |
432f16e6 | 298 | buffer_io_error(bh, ", async page read"); |
1da177e4 LT |
299 | SetPageError(page); |
300 | } | |
301 | ||
302 | /* | |
303 | * Be _very_ careful from here on. Bad things can happen if | |
304 | * two buffer heads end IO at almost the same time and both | |
305 | * decide that the page is now completely done. | |
306 | */ | |
a3972203 NP |
307 | first = page_buffers(page); |
308 | local_irq_save(flags); | |
309 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
1da177e4 LT |
310 | clear_buffer_async_read(bh); |
311 | unlock_buffer(bh); | |
312 | tmp = bh; | |
313 | do { | |
314 | if (!buffer_uptodate(tmp)) | |
315 | page_uptodate = 0; | |
316 | if (buffer_async_read(tmp)) { | |
317 | BUG_ON(!buffer_locked(tmp)); | |
318 | goto still_busy; | |
319 | } | |
320 | tmp = tmp->b_this_page; | |
321 | } while (tmp != bh); | |
a3972203 NP |
322 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
323 | local_irq_restore(flags); | |
1da177e4 LT |
324 | |
325 | /* | |
326 | * If none of the buffers had errors and they are all | |
327 | * uptodate then we can set the page uptodate. | |
328 | */ | |
329 | if (page_uptodate && !PageError(page)) | |
330 | SetPageUptodate(page); | |
331 | unlock_page(page); | |
332 | return; | |
333 | ||
334 | still_busy: | |
a3972203 NP |
335 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
336 | local_irq_restore(flags); | |
1da177e4 LT |
337 | return; |
338 | } | |
339 | ||
340 | /* | |
341 | * Completion handler for block_write_full_page() - pages which are unlocked | |
342 | * during I/O, and which have PageWriteback cleared upon I/O completion. | |
343 | */ | |
35c80d5f | 344 | void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
1da177e4 | 345 | { |
1da177e4 | 346 | unsigned long flags; |
a3972203 | 347 | struct buffer_head *first; |
1da177e4 LT |
348 | struct buffer_head *tmp; |
349 | struct page *page; | |
350 | ||
351 | BUG_ON(!buffer_async_write(bh)); | |
352 | ||
353 | page = bh->b_page; | |
354 | if (uptodate) { | |
355 | set_buffer_uptodate(bh); | |
356 | } else { | |
432f16e6 | 357 | buffer_io_error(bh, ", lost async page write"); |
1da177e4 | 358 | set_bit(AS_EIO, &page->mapping->flags); |
58ff407b | 359 | set_buffer_write_io_error(bh); |
1da177e4 LT |
360 | clear_buffer_uptodate(bh); |
361 | SetPageError(page); | |
362 | } | |
363 | ||
a3972203 NP |
364 | first = page_buffers(page); |
365 | local_irq_save(flags); | |
366 | bit_spin_lock(BH_Uptodate_Lock, &first->b_state); | |
367 | ||
1da177e4 LT |
368 | clear_buffer_async_write(bh); |
369 | unlock_buffer(bh); | |
370 | tmp = bh->b_this_page; | |
371 | while (tmp != bh) { | |
372 | if (buffer_async_write(tmp)) { | |
373 | BUG_ON(!buffer_locked(tmp)); | |
374 | goto still_busy; | |
375 | } | |
376 | tmp = tmp->b_this_page; | |
377 | } | |
a3972203 NP |
378 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
379 | local_irq_restore(flags); | |
1da177e4 LT |
380 | end_page_writeback(page); |
381 | return; | |
382 | ||
383 | still_busy: | |
a3972203 NP |
384 | bit_spin_unlock(BH_Uptodate_Lock, &first->b_state); |
385 | local_irq_restore(flags); | |
1da177e4 LT |
386 | return; |
387 | } | |
1fe72eaa | 388 | EXPORT_SYMBOL(end_buffer_async_write); |
1da177e4 LT |
389 | |
390 | /* | |
391 | * If a page's buffers are under async readin (end_buffer_async_read | |
392 | * completion) then there is a possibility that another thread of | |
393 | * control could lock one of the buffers after it has completed | |
394 | * but while some of the other buffers have not completed. This | |
395 | * locked buffer would confuse end_buffer_async_read() into not unlocking | |
396 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() | |
397 | * that this buffer is not under async I/O. | |
398 | * | |
399 | * The page comes unlocked when it has no locked buffer_async buffers | |
400 | * left. | |
401 | * | |
402 | * PageLocked prevents anyone starting new async I/O reads any of | |
403 | * the buffers. | |
404 | * | |
405 | * PageWriteback is used to prevent simultaneous writeout of the same | |
406 | * page. | |
407 | * | |
408 | * PageLocked prevents anyone from starting writeback of a page which is | |
409 | * under read I/O (PageWriteback is only ever set against a locked page). | |
410 | */ | |
411 | static void mark_buffer_async_read(struct buffer_head *bh) | |
412 | { | |
413 | bh->b_end_io = end_buffer_async_read; | |
414 | set_buffer_async_read(bh); | |
415 | } | |
416 | ||
1fe72eaa HS |
417 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
418 | bh_end_io_t *handler) | |
1da177e4 | 419 | { |
35c80d5f | 420 | bh->b_end_io = handler; |
1da177e4 LT |
421 | set_buffer_async_write(bh); |
422 | } | |
35c80d5f CM |
423 | |
424 | void mark_buffer_async_write(struct buffer_head *bh) | |
425 | { | |
426 | mark_buffer_async_write_endio(bh, end_buffer_async_write); | |
427 | } | |
1da177e4 LT |
428 | EXPORT_SYMBOL(mark_buffer_async_write); |
429 | ||
430 | ||
431 | /* | |
432 | * fs/buffer.c contains helper functions for buffer-backed address space's | |
433 | * fsync functions. A common requirement for buffer-based filesystems is | |
434 | * that certain data from the backing blockdev needs to be written out for | |
435 | * a successful fsync(). For example, ext2 indirect blocks need to be | |
436 | * written back and waited upon before fsync() returns. | |
437 | * | |
438 | * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(), | |
439 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the | |
440 | * management of a list of dependent buffers at ->i_mapping->private_list. | |
441 | * | |
442 | * Locking is a little subtle: try_to_free_buffers() will remove buffers | |
443 | * from their controlling inode's queue when they are being freed. But | |
444 | * try_to_free_buffers() will be operating against the *blockdev* mapping | |
445 | * at the time, not against the S_ISREG file which depends on those buffers. | |
446 | * So the locking for private_list is via the private_lock in the address_space | |
447 | * which backs the buffers. Which is different from the address_space | |
448 | * against which the buffers are listed. So for a particular address_space, | |
449 | * mapping->private_lock does *not* protect mapping->private_list! In fact, | |
450 | * mapping->private_list will always be protected by the backing blockdev's | |
451 | * ->private_lock. | |
452 | * | |
453 | * Which introduces a requirement: all buffers on an address_space's | |
454 | * ->private_list must be from the same address_space: the blockdev's. | |
455 | * | |
456 | * address_spaces which do not place buffers at ->private_list via these | |
457 | * utility functions are free to use private_lock and private_list for | |
458 | * whatever they want. The only requirement is that list_empty(private_list) | |
459 | * be true at clear_inode() time. | |
460 | * | |
461 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The | |
462 | * filesystems should do that. invalidate_inode_buffers() should just go | |
463 | * BUG_ON(!list_empty). | |
464 | * | |
465 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should | |
466 | * take an address_space, not an inode. And it should be called | |
467 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being | |
468 | * queued up. | |
469 | * | |
470 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the | |
471 | * list if it is already on a list. Because if the buffer is on a list, | |
472 | * it *must* already be on the right one. If not, the filesystem is being | |
473 | * silly. This will save a ton of locking. But first we have to ensure | |
474 | * that buffers are taken *off* the old inode's list when they are freed | |
475 | * (presumably in truncate). That requires careful auditing of all | |
476 | * filesystems (do it inside bforget()). It could also be done by bringing | |
477 | * b_inode back. | |
478 | */ | |
479 | ||
480 | /* | |
481 | * The buffer's backing address_space's private_lock must be held | |
482 | */ | |
dbacefc9 | 483 | static void __remove_assoc_queue(struct buffer_head *bh) |
1da177e4 LT |
484 | { |
485 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b JK |
486 | WARN_ON(!bh->b_assoc_map); |
487 | if (buffer_write_io_error(bh)) | |
488 | set_bit(AS_EIO, &bh->b_assoc_map->flags); | |
489 | bh->b_assoc_map = NULL; | |
1da177e4 LT |
490 | } |
491 | ||
492 | int inode_has_buffers(struct inode *inode) | |
493 | { | |
494 | return !list_empty(&inode->i_data.private_list); | |
495 | } | |
496 | ||
497 | /* | |
498 | * osync is designed to support O_SYNC io. It waits synchronously for | |
499 | * all already-submitted IO to complete, but does not queue any new | |
500 | * writes to the disk. | |
501 | * | |
502 | * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as | |
503 | * you dirty the buffers, and then use osync_inode_buffers to wait for | |
504 | * completion. Any other dirty buffers which are not yet queued for | |
505 | * write will not be flushed to disk by the osync. | |
506 | */ | |
507 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) | |
508 | { | |
509 | struct buffer_head *bh; | |
510 | struct list_head *p; | |
511 | int err = 0; | |
512 | ||
513 | spin_lock(lock); | |
514 | repeat: | |
515 | list_for_each_prev(p, list) { | |
516 | bh = BH_ENTRY(p); | |
517 | if (buffer_locked(bh)) { | |
518 | get_bh(bh); | |
519 | spin_unlock(lock); | |
520 | wait_on_buffer(bh); | |
521 | if (!buffer_uptodate(bh)) | |
522 | err = -EIO; | |
523 | brelse(bh); | |
524 | spin_lock(lock); | |
525 | goto repeat; | |
526 | } | |
527 | } | |
528 | spin_unlock(lock); | |
529 | return err; | |
530 | } | |
531 | ||
01a05b33 | 532 | static void do_thaw_one(struct super_block *sb, void *unused) |
c2d75438 | 533 | { |
c2d75438 | 534 | char b[BDEVNAME_SIZE]; |
01a05b33 AV |
535 | while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb)) |
536 | printk(KERN_WARNING "Emergency Thaw on %s\n", | |
537 | bdevname(sb->s_bdev, b)); | |
538 | } | |
c2d75438 | 539 | |
01a05b33 AV |
540 | static void do_thaw_all(struct work_struct *work) |
541 | { | |
542 | iterate_supers(do_thaw_one, NULL); | |
053c525f | 543 | kfree(work); |
c2d75438 ES |
544 | printk(KERN_WARNING "Emergency Thaw complete\n"); |
545 | } | |
546 | ||
547 | /** | |
548 | * emergency_thaw_all -- forcibly thaw every frozen filesystem | |
549 | * | |
550 | * Used for emergency unfreeze of all filesystems via SysRq | |
551 | */ | |
552 | void emergency_thaw_all(void) | |
553 | { | |
053c525f JA |
554 | struct work_struct *work; |
555 | ||
556 | work = kmalloc(sizeof(*work), GFP_ATOMIC); | |
557 | if (work) { | |
558 | INIT_WORK(work, do_thaw_all); | |
559 | schedule_work(work); | |
560 | } | |
c2d75438 ES |
561 | } |
562 | ||
1da177e4 | 563 | /** |
78a4a50a | 564 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
67be2dd1 | 565 | * @mapping: the mapping which wants those buffers written |
1da177e4 LT |
566 | * |
567 | * Starts I/O against the buffers at mapping->private_list, and waits upon | |
568 | * that I/O. | |
569 | * | |
67be2dd1 MW |
570 | * Basically, this is a convenience function for fsync(). |
571 | * @mapping is a file or directory which needs those buffers to be written for | |
572 | * a successful fsync(). | |
1da177e4 LT |
573 | */ |
574 | int sync_mapping_buffers(struct address_space *mapping) | |
575 | { | |
252aa6f5 | 576 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
577 | |
578 | if (buffer_mapping == NULL || list_empty(&mapping->private_list)) | |
579 | return 0; | |
580 | ||
581 | return fsync_buffers_list(&buffer_mapping->private_lock, | |
582 | &mapping->private_list); | |
583 | } | |
584 | EXPORT_SYMBOL(sync_mapping_buffers); | |
585 | ||
586 | /* | |
587 | * Called when we've recently written block `bblock', and it is known that | |
588 | * `bblock' was for a buffer_boundary() buffer. This means that the block at | |
589 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's | |
590 | * dirty, schedule it for IO. So that indirects merge nicely with their data. | |
591 | */ | |
592 | void write_boundary_block(struct block_device *bdev, | |
593 | sector_t bblock, unsigned blocksize) | |
594 | { | |
595 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); | |
596 | if (bh) { | |
597 | if (buffer_dirty(bh)) | |
598 | ll_rw_block(WRITE, 1, &bh); | |
599 | put_bh(bh); | |
600 | } | |
601 | } | |
602 | ||
603 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) | |
604 | { | |
605 | struct address_space *mapping = inode->i_mapping; | |
606 | struct address_space *buffer_mapping = bh->b_page->mapping; | |
607 | ||
608 | mark_buffer_dirty(bh); | |
252aa6f5 RA |
609 | if (!mapping->private_data) { |
610 | mapping->private_data = buffer_mapping; | |
1da177e4 | 611 | } else { |
252aa6f5 | 612 | BUG_ON(mapping->private_data != buffer_mapping); |
1da177e4 | 613 | } |
535ee2fb | 614 | if (!bh->b_assoc_map) { |
1da177e4 LT |
615 | spin_lock(&buffer_mapping->private_lock); |
616 | list_move_tail(&bh->b_assoc_buffers, | |
617 | &mapping->private_list); | |
58ff407b | 618 | bh->b_assoc_map = mapping; |
1da177e4 LT |
619 | spin_unlock(&buffer_mapping->private_lock); |
620 | } | |
621 | } | |
622 | EXPORT_SYMBOL(mark_buffer_dirty_inode); | |
623 | ||
1cac41cb MB |
624 | void mark_buffer_dirty_inode_sync(struct buffer_head *bh, struct inode *inode) |
625 | { | |
626 | set_buffer_sync_flush(bh); | |
627 | mark_buffer_dirty_inode(bh, inode); | |
628 | } | |
629 | EXPORT_SYMBOL(mark_buffer_dirty_inode_sync); | |
630 | ||
787d2214 NP |
631 | /* |
632 | * Mark the page dirty, and set it dirty in the radix tree, and mark the inode | |
633 | * dirty. | |
634 | * | |
635 | * If warn is true, then emit a warning if the page is not uptodate and has | |
636 | * not been truncated. | |
c4843a75 GT |
637 | * |
638 | * The caller must hold mem_cgroup_begin_page_stat() lock. | |
787d2214 | 639 | */ |
c4843a75 GT |
640 | static void __set_page_dirty(struct page *page, struct address_space *mapping, |
641 | struct mem_cgroup *memcg, int warn) | |
787d2214 | 642 | { |
227d53b3 KM |
643 | unsigned long flags; |
644 | ||
645 | spin_lock_irqsave(&mapping->tree_lock, flags); | |
787d2214 NP |
646 | if (page->mapping) { /* Race with truncate? */ |
647 | WARN_ON_ONCE(warn && !PageUptodate(page)); | |
c4843a75 | 648 | account_page_dirtied(page, mapping, memcg); |
787d2214 NP |
649 | radix_tree_tag_set(&mapping->page_tree, |
650 | page_index(page), PAGECACHE_TAG_DIRTY); | |
651 | } | |
227d53b3 | 652 | spin_unlock_irqrestore(&mapping->tree_lock, flags); |
787d2214 NP |
653 | } |
654 | ||
1da177e4 LT |
655 | /* |
656 | * Add a page to the dirty page list. | |
657 | * | |
658 | * It is a sad fact of life that this function is called from several places | |
659 | * deeply under spinlocking. It may not sleep. | |
660 | * | |
661 | * If the page has buffers, the uptodate buffers are set dirty, to preserve | |
662 | * dirty-state coherency between the page and the buffers. It the page does | |
663 | * not have buffers then when they are later attached they will all be set | |
664 | * dirty. | |
665 | * | |
666 | * The buffers are dirtied before the page is dirtied. There's a small race | |
667 | * window in which a writepage caller may see the page cleanness but not the | |
668 | * buffer dirtiness. That's fine. If this code were to set the page dirty | |
669 | * before the buffers, a concurrent writepage caller could clear the page dirty | |
670 | * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean | |
671 | * page on the dirty page list. | |
672 | * | |
673 | * We use private_lock to lock against try_to_free_buffers while using the | |
674 | * page's buffer list. Also use this to protect against clean buffers being | |
675 | * added to the page after it was set dirty. | |
676 | * | |
677 | * FIXME: may need to call ->reservepage here as well. That's rather up to the | |
678 | * address_space though. | |
679 | */ | |
680 | int __set_page_dirty_buffers(struct page *page) | |
681 | { | |
a8e7d49a | 682 | int newly_dirty; |
c4843a75 | 683 | struct mem_cgroup *memcg; |
787d2214 | 684 | struct address_space *mapping = page_mapping(page); |
ebf7a227 NP |
685 | |
686 | if (unlikely(!mapping)) | |
687 | return !TestSetPageDirty(page); | |
1da177e4 LT |
688 | |
689 | spin_lock(&mapping->private_lock); | |
690 | if (page_has_buffers(page)) { | |
691 | struct buffer_head *head = page_buffers(page); | |
692 | struct buffer_head *bh = head; | |
693 | ||
694 | do { | |
695 | set_buffer_dirty(bh); | |
696 | bh = bh->b_this_page; | |
697 | } while (bh != head); | |
698 | } | |
c4843a75 GT |
699 | /* |
700 | * Use mem_group_begin_page_stat() to keep PageDirty synchronized with | |
701 | * per-memcg dirty page counters. | |
702 | */ | |
703 | memcg = mem_cgroup_begin_page_stat(page); | |
a8e7d49a | 704 | newly_dirty = !TestSetPageDirty(page); |
1da177e4 LT |
705 | spin_unlock(&mapping->private_lock); |
706 | ||
a8e7d49a | 707 | if (newly_dirty) |
c4843a75 GT |
708 | __set_page_dirty(page, mapping, memcg, 1); |
709 | ||
710 | mem_cgroup_end_page_stat(memcg); | |
711 | ||
712 | if (newly_dirty) | |
713 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
714 | ||
a8e7d49a | 715 | return newly_dirty; |
1da177e4 LT |
716 | } |
717 | EXPORT_SYMBOL(__set_page_dirty_buffers); | |
718 | ||
719 | /* | |
720 | * Write out and wait upon a list of buffers. | |
721 | * | |
722 | * We have conflicting pressures: we want to make sure that all | |
723 | * initially dirty buffers get waited on, but that any subsequently | |
724 | * dirtied buffers don't. After all, we don't want fsync to last | |
725 | * forever if somebody is actively writing to the file. | |
726 | * | |
727 | * Do this in two main stages: first we copy dirty buffers to a | |
728 | * temporary inode list, queueing the writes as we go. Then we clean | |
729 | * up, waiting for those writes to complete. | |
730 | * | |
731 | * During this second stage, any subsequent updates to the file may end | |
732 | * up refiling the buffer on the original inode's dirty list again, so | |
733 | * there is a chance we will end up with a buffer queued for write but | |
734 | * not yet completed on that list. So, as a final cleanup we go through | |
735 | * the osync code to catch these locked, dirty buffers without requeuing | |
736 | * any newly dirty buffers for write. | |
737 | */ | |
738 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) | |
739 | { | |
740 | struct buffer_head *bh; | |
741 | struct list_head tmp; | |
7eaceacc | 742 | struct address_space *mapping; |
1da177e4 | 743 | int err = 0, err2; |
4ee2491e | 744 | struct blk_plug plug; |
1da177e4 LT |
745 | |
746 | INIT_LIST_HEAD(&tmp); | |
4ee2491e | 747 | blk_start_plug(&plug); |
1da177e4 LT |
748 | |
749 | spin_lock(lock); | |
750 | while (!list_empty(list)) { | |
751 | bh = BH_ENTRY(list->next); | |
535ee2fb | 752 | mapping = bh->b_assoc_map; |
58ff407b | 753 | __remove_assoc_queue(bh); |
535ee2fb JK |
754 | /* Avoid race with mark_buffer_dirty_inode() which does |
755 | * a lockless check and we rely on seeing the dirty bit */ | |
756 | smp_mb(); | |
1da177e4 LT |
757 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
758 | list_add(&bh->b_assoc_buffers, &tmp); | |
535ee2fb | 759 | bh->b_assoc_map = mapping; |
1da177e4 LT |
760 | if (buffer_dirty(bh)) { |
761 | get_bh(bh); | |
762 | spin_unlock(lock); | |
763 | /* | |
764 | * Ensure any pending I/O completes so that | |
9cb569d6 CH |
765 | * write_dirty_buffer() actually writes the |
766 | * current contents - it is a noop if I/O is | |
767 | * still in flight on potentially older | |
768 | * contents. | |
1da177e4 | 769 | */ |
721a9602 | 770 | write_dirty_buffer(bh, WRITE_SYNC); |
9cf6b720 JA |
771 | |
772 | /* | |
773 | * Kick off IO for the previous mapping. Note | |
774 | * that we will not run the very last mapping, | |
775 | * wait_on_buffer() will do that for us | |
776 | * through sync_buffer(). | |
777 | */ | |
1da177e4 LT |
778 | brelse(bh); |
779 | spin_lock(lock); | |
780 | } | |
781 | } | |
782 | } | |
783 | ||
4ee2491e JA |
784 | spin_unlock(lock); |
785 | blk_finish_plug(&plug); | |
786 | spin_lock(lock); | |
787 | ||
1da177e4 LT |
788 | while (!list_empty(&tmp)) { |
789 | bh = BH_ENTRY(tmp.prev); | |
1da177e4 | 790 | get_bh(bh); |
535ee2fb JK |
791 | mapping = bh->b_assoc_map; |
792 | __remove_assoc_queue(bh); | |
793 | /* Avoid race with mark_buffer_dirty_inode() which does | |
794 | * a lockless check and we rely on seeing the dirty bit */ | |
795 | smp_mb(); | |
796 | if (buffer_dirty(bh)) { | |
797 | list_add(&bh->b_assoc_buffers, | |
e3892296 | 798 | &mapping->private_list); |
535ee2fb JK |
799 | bh->b_assoc_map = mapping; |
800 | } | |
1da177e4 LT |
801 | spin_unlock(lock); |
802 | wait_on_buffer(bh); | |
803 | if (!buffer_uptodate(bh)) | |
804 | err = -EIO; | |
805 | brelse(bh); | |
806 | spin_lock(lock); | |
807 | } | |
808 | ||
809 | spin_unlock(lock); | |
810 | err2 = osync_buffers_list(lock, list); | |
811 | if (err) | |
812 | return err; | |
813 | else | |
814 | return err2; | |
815 | } | |
816 | ||
817 | /* | |
818 | * Invalidate any and all dirty buffers on a given inode. We are | |
819 | * probably unmounting the fs, but that doesn't mean we have already | |
820 | * done a sync(). Just drop the buffers from the inode list. | |
821 | * | |
822 | * NOTE: we take the inode's blockdev's mapping's private_lock. Which | |
823 | * assumes that all the buffers are against the blockdev. Not true | |
824 | * for reiserfs. | |
825 | */ | |
826 | void invalidate_inode_buffers(struct inode *inode) | |
827 | { | |
828 | if (inode_has_buffers(inode)) { | |
829 | struct address_space *mapping = &inode->i_data; | |
830 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 831 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
832 | |
833 | spin_lock(&buffer_mapping->private_lock); | |
834 | while (!list_empty(list)) | |
835 | __remove_assoc_queue(BH_ENTRY(list->next)); | |
836 | spin_unlock(&buffer_mapping->private_lock); | |
837 | } | |
838 | } | |
52b19ac9 | 839 | EXPORT_SYMBOL(invalidate_inode_buffers); |
1da177e4 LT |
840 | |
841 | /* | |
842 | * Remove any clean buffers from the inode's buffer list. This is called | |
843 | * when we're trying to free the inode itself. Those buffers can pin it. | |
844 | * | |
845 | * Returns true if all buffers were removed. | |
846 | */ | |
847 | int remove_inode_buffers(struct inode *inode) | |
848 | { | |
849 | int ret = 1; | |
850 | ||
851 | if (inode_has_buffers(inode)) { | |
852 | struct address_space *mapping = &inode->i_data; | |
853 | struct list_head *list = &mapping->private_list; | |
252aa6f5 | 854 | struct address_space *buffer_mapping = mapping->private_data; |
1da177e4 LT |
855 | |
856 | spin_lock(&buffer_mapping->private_lock); | |
857 | while (!list_empty(list)) { | |
858 | struct buffer_head *bh = BH_ENTRY(list->next); | |
859 | if (buffer_dirty(bh)) { | |
860 | ret = 0; | |
861 | break; | |
862 | } | |
863 | __remove_assoc_queue(bh); | |
864 | } | |
865 | spin_unlock(&buffer_mapping->private_lock); | |
866 | } | |
867 | return ret; | |
868 | } | |
869 | ||
870 | /* | |
871 | * Create the appropriate buffers when given a page for data area and | |
872 | * the size of each buffer.. Use the bh->b_this_page linked list to | |
873 | * follow the buffers created. Return NULL if unable to create more | |
874 | * buffers. | |
875 | * | |
876 | * The retry flag is used to differentiate async IO (paging, swapping) | |
877 | * which may not fail from ordinary buffer allocations. | |
878 | */ | |
879 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, | |
880 | int retry) | |
881 | { | |
882 | struct buffer_head *bh, *head; | |
883 | long offset; | |
884 | ||
885 | try_again: | |
886 | head = NULL; | |
887 | offset = PAGE_SIZE; | |
888 | while ((offset -= size) >= 0) { | |
889 | bh = alloc_buffer_head(GFP_NOFS); | |
890 | if (!bh) | |
891 | goto no_grow; | |
892 | ||
1da177e4 LT |
893 | bh->b_this_page = head; |
894 | bh->b_blocknr = -1; | |
895 | head = bh; | |
896 | ||
1da177e4 LT |
897 | bh->b_size = size; |
898 | ||
899 | /* Link the buffer to its page */ | |
900 | set_bh_page(bh, page, offset); | |
1da177e4 LT |
901 | } |
902 | return head; | |
903 | /* | |
904 | * In case anything failed, we just free everything we got. | |
905 | */ | |
906 | no_grow: | |
907 | if (head) { | |
908 | do { | |
909 | bh = head; | |
910 | head = head->b_this_page; | |
911 | free_buffer_head(bh); | |
912 | } while (head); | |
913 | } | |
914 | ||
915 | /* | |
916 | * Return failure for non-async IO requests. Async IO requests | |
917 | * are not allowed to fail, so we have to wait until buffer heads | |
918 | * become available. But we don't want tasks sleeping with | |
919 | * partially complete buffers, so all were released above. | |
920 | */ | |
921 | if (!retry) | |
922 | return NULL; | |
923 | ||
924 | /* We're _really_ low on memory. Now we just | |
925 | * wait for old buffer heads to become free due to | |
926 | * finishing IO. Since this is an async request and | |
927 | * the reserve list is empty, we're sure there are | |
928 | * async buffer heads in use. | |
929 | */ | |
930 | free_more_memory(); | |
931 | goto try_again; | |
932 | } | |
933 | EXPORT_SYMBOL_GPL(alloc_page_buffers); | |
934 | ||
935 | static inline void | |
936 | link_dev_buffers(struct page *page, struct buffer_head *head) | |
937 | { | |
938 | struct buffer_head *bh, *tail; | |
939 | ||
940 | bh = head; | |
941 | do { | |
942 | tail = bh; | |
943 | bh = bh->b_this_page; | |
944 | } while (bh); | |
945 | tail->b_this_page = head; | |
946 | attach_page_buffers(page, head); | |
947 | } | |
948 | ||
bbec0270 LT |
949 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
950 | { | |
951 | sector_t retval = ~((sector_t)0); | |
952 | loff_t sz = i_size_read(bdev->bd_inode); | |
953 | ||
954 | if (sz) { | |
955 | unsigned int sizebits = blksize_bits(size); | |
956 | retval = (sz >> sizebits); | |
957 | } | |
958 | return retval; | |
959 | } | |
960 | ||
1da177e4 LT |
961 | /* |
962 | * Initialise the state of a blockdev page's buffers. | |
963 | */ | |
676ce6d5 | 964 | static sector_t |
1da177e4 LT |
965 | init_page_buffers(struct page *page, struct block_device *bdev, |
966 | sector_t block, int size) | |
967 | { | |
968 | struct buffer_head *head = page_buffers(page); | |
969 | struct buffer_head *bh = head; | |
970 | int uptodate = PageUptodate(page); | |
bbec0270 | 971 | sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size); |
1da177e4 LT |
972 | |
973 | do { | |
974 | if (!buffer_mapped(bh)) { | |
975 | init_buffer(bh, NULL, NULL); | |
976 | bh->b_bdev = bdev; | |
977 | bh->b_blocknr = block; | |
978 | if (uptodate) | |
979 | set_buffer_uptodate(bh); | |
080399aa JM |
980 | if (block < end_block) |
981 | set_buffer_mapped(bh); | |
1da177e4 LT |
982 | } |
983 | block++; | |
984 | bh = bh->b_this_page; | |
985 | } while (bh != head); | |
676ce6d5 HD |
986 | |
987 | /* | |
988 | * Caller needs to validate requested block against end of device. | |
989 | */ | |
990 | return end_block; | |
1da177e4 LT |
991 | } |
992 | ||
993 | /* | |
994 | * Create the page-cache page that contains the requested block. | |
995 | * | |
676ce6d5 | 996 | * This is used purely for blockdev mappings. |
1da177e4 | 997 | */ |
676ce6d5 | 998 | static int |
1da177e4 | 999 | grow_dev_page(struct block_device *bdev, sector_t block, |
3b5e6454 | 1000 | pgoff_t index, int size, int sizebits, gfp_t gfp) |
1da177e4 LT |
1001 | { |
1002 | struct inode *inode = bdev->bd_inode; | |
1003 | struct page *page; | |
1004 | struct buffer_head *bh; | |
676ce6d5 HD |
1005 | sector_t end_block; |
1006 | int ret = 0; /* Will call free_more_memory() */ | |
84235de3 | 1007 | gfp_t gfp_mask; |
1da177e4 | 1008 | |
c62d2555 | 1009 | gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp; |
3b5e6454 | 1010 | |
84235de3 JW |
1011 | /* |
1012 | * XXX: __getblk_slow() can not really deal with failure and | |
1013 | * will endlessly loop on improvised global reclaim. Prefer | |
1014 | * looping in the allocator rather than here, at least that | |
1015 | * code knows what it's doing. | |
1016 | */ | |
1017 | gfp_mask |= __GFP_NOFAIL; | |
1018 | ||
1019 | page = find_or_create_page(inode->i_mapping, index, gfp_mask); | |
1da177e4 | 1020 | if (!page) |
676ce6d5 | 1021 | return ret; |
1da177e4 | 1022 | |
e827f923 | 1023 | BUG_ON(!PageLocked(page)); |
1da177e4 LT |
1024 | |
1025 | if (page_has_buffers(page)) { | |
1026 | bh = page_buffers(page); | |
1027 | if (bh->b_size == size) { | |
676ce6d5 | 1028 | end_block = init_page_buffers(page, bdev, |
f2d5a944 AA |
1029 | (sector_t)index << sizebits, |
1030 | size); | |
676ce6d5 | 1031 | goto done; |
1da177e4 LT |
1032 | } |
1033 | if (!try_to_free_buffers(page)) | |
1034 | goto failed; | |
1035 | } | |
1036 | ||
1037 | /* | |
1038 | * Allocate some buffers for this page | |
1039 | */ | |
1040 | bh = alloc_page_buffers(page, size, 0); | |
1041 | if (!bh) | |
1042 | goto failed; | |
1043 | ||
1044 | /* | |
1045 | * Link the page to the buffers and initialise them. Take the | |
1046 | * lock to be atomic wrt __find_get_block(), which does not | |
1047 | * run under the page lock. | |
1048 | */ | |
1049 | spin_lock(&inode->i_mapping->private_lock); | |
1050 | link_dev_buffers(page, bh); | |
f2d5a944 AA |
1051 | end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits, |
1052 | size); | |
1da177e4 | 1053 | spin_unlock(&inode->i_mapping->private_lock); |
676ce6d5 HD |
1054 | done: |
1055 | ret = (block < end_block) ? 1 : -ENXIO; | |
1da177e4 | 1056 | failed: |
1da177e4 LT |
1057 | unlock_page(page); |
1058 | page_cache_release(page); | |
676ce6d5 | 1059 | return ret; |
1da177e4 LT |
1060 | } |
1061 | ||
1062 | /* | |
1063 | * Create buffers for the specified block device block's page. If | |
1064 | * that page was dirty, the buffers are set dirty also. | |
1da177e4 | 1065 | */ |
858119e1 | 1066 | static int |
3b5e6454 | 1067 | grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp) |
1da177e4 | 1068 | { |
1da177e4 LT |
1069 | pgoff_t index; |
1070 | int sizebits; | |
1071 | ||
1072 | sizebits = -1; | |
1073 | do { | |
1074 | sizebits++; | |
1075 | } while ((size << sizebits) < PAGE_SIZE); | |
1076 | ||
1077 | index = block >> sizebits; | |
1da177e4 | 1078 | |
e5657933 AM |
1079 | /* |
1080 | * Check for a block which wants to lie outside our maximum possible | |
1081 | * pagecache index. (this comparison is done using sector_t types). | |
1082 | */ | |
1083 | if (unlikely(index != block >> sizebits)) { | |
1084 | char b[BDEVNAME_SIZE]; | |
1085 | ||
1086 | printk(KERN_ERR "%s: requested out-of-range block %llu for " | |
1087 | "device %s\n", | |
8e24eea7 | 1088 | __func__, (unsigned long long)block, |
e5657933 AM |
1089 | bdevname(bdev, b)); |
1090 | return -EIO; | |
1091 | } | |
676ce6d5 | 1092 | |
1da177e4 | 1093 | /* Create a page with the proper size buffers.. */ |
3b5e6454 | 1094 | return grow_dev_page(bdev, block, index, size, sizebits, gfp); |
1da177e4 LT |
1095 | } |
1096 | ||
3b5e6454 GK |
1097 | struct buffer_head * |
1098 | __getblk_slow(struct block_device *bdev, sector_t block, | |
1099 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1100 | { |
1101 | /* Size must be multiple of hard sectorsize */ | |
e1defc4f | 1102 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
1da177e4 LT |
1103 | (size < 512 || size > PAGE_SIZE))) { |
1104 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", | |
1105 | size); | |
e1defc4f MP |
1106 | printk(KERN_ERR "logical block size: %d\n", |
1107 | bdev_logical_block_size(bdev)); | |
1da177e4 LT |
1108 | |
1109 | dump_stack(); | |
1110 | return NULL; | |
1111 | } | |
1112 | ||
676ce6d5 HD |
1113 | for (;;) { |
1114 | struct buffer_head *bh; | |
1115 | int ret; | |
1da177e4 LT |
1116 | |
1117 | bh = __find_get_block(bdev, block, size); | |
1118 | if (bh) | |
1119 | return bh; | |
676ce6d5 | 1120 | |
3b5e6454 | 1121 | ret = grow_buffers(bdev, block, size, gfp); |
676ce6d5 HD |
1122 | if (ret < 0) |
1123 | return NULL; | |
1124 | if (ret == 0) | |
1125 | free_more_memory(); | |
1da177e4 LT |
1126 | } |
1127 | } | |
3b5e6454 | 1128 | EXPORT_SYMBOL(__getblk_slow); |
1da177e4 LT |
1129 | |
1130 | /* | |
1131 | * The relationship between dirty buffers and dirty pages: | |
1132 | * | |
1133 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and | |
1134 | * the page is tagged dirty in its radix tree. | |
1135 | * | |
1136 | * At all times, the dirtiness of the buffers represents the dirtiness of | |
1137 | * subsections of the page. If the page has buffers, the page dirty bit is | |
1138 | * merely a hint about the true dirty state. | |
1139 | * | |
1140 | * When a page is set dirty in its entirety, all its buffers are marked dirty | |
1141 | * (if the page has buffers). | |
1142 | * | |
1143 | * When a buffer is marked dirty, its page is dirtied, but the page's other | |
1144 | * buffers are not. | |
1145 | * | |
1146 | * Also. When blockdev buffers are explicitly read with bread(), they | |
1147 | * individually become uptodate. But their backing page remains not | |
1148 | * uptodate - even if all of its buffers are uptodate. A subsequent | |
1149 | * block_read_full_page() against that page will discover all the uptodate | |
1150 | * buffers, will set the page uptodate and will perform no I/O. | |
1151 | */ | |
1152 | ||
1153 | /** | |
1154 | * mark_buffer_dirty - mark a buffer_head as needing writeout | |
67be2dd1 | 1155 | * @bh: the buffer_head to mark dirty |
1da177e4 LT |
1156 | * |
1157 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set its | |
1158 | * backing page dirty, then tag the page as dirty in its address_space's radix | |
1159 | * tree and then attach the address_space's inode to its superblock's dirty | |
1160 | * inode list. | |
1161 | * | |
1162 | * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock, | |
250df6ed | 1163 | * mapping->tree_lock and mapping->host->i_lock. |
1da177e4 | 1164 | */ |
fc9b52cd | 1165 | void mark_buffer_dirty(struct buffer_head *bh) |
1da177e4 | 1166 | { |
787d2214 | 1167 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
1be62dc1 | 1168 | |
5305cb83 TH |
1169 | trace_block_dirty_buffer(bh); |
1170 | ||
1be62dc1 LT |
1171 | /* |
1172 | * Very *carefully* optimize the it-is-already-dirty case. | |
1173 | * | |
1174 | * Don't let the final "is it dirty" escape to before we | |
1175 | * perhaps modified the buffer. | |
1176 | */ | |
1177 | if (buffer_dirty(bh)) { | |
1178 | smp_mb(); | |
1179 | if (buffer_dirty(bh)) | |
1180 | return; | |
1181 | } | |
1182 | ||
a8e7d49a LT |
1183 | if (!test_set_buffer_dirty(bh)) { |
1184 | struct page *page = bh->b_page; | |
c4843a75 GT |
1185 | struct address_space *mapping = NULL; |
1186 | struct mem_cgroup *memcg; | |
1187 | ||
1188 | memcg = mem_cgroup_begin_page_stat(page); | |
8e9d78ed | 1189 | if (!TestSetPageDirty(page)) { |
c4843a75 | 1190 | mapping = page_mapping(page); |
8e9d78ed | 1191 | if (mapping) |
c4843a75 | 1192 | __set_page_dirty(page, mapping, memcg, 0); |
8e9d78ed | 1193 | } |
c4843a75 GT |
1194 | mem_cgroup_end_page_stat(memcg); |
1195 | if (mapping) | |
1196 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
a8e7d49a | 1197 | } |
1da177e4 | 1198 | } |
1fe72eaa | 1199 | EXPORT_SYMBOL(mark_buffer_dirty); |
1da177e4 | 1200 | |
1cac41cb MB |
1201 | void mark_buffer_dirty_sync(struct buffer_head *bh) |
1202 | { | |
1203 | WARN_ON_ONCE(!buffer_uptodate(bh)); | |
1204 | ||
1205 | trace_block_dirty_buffer(bh); | |
1206 | ||
1207 | /* | |
1208 | * Very *carefully* optimize the it-is-already-dirty case. | |
1209 | * | |
1210 | * Don't let the final "is it dirty" escape to before we | |
1211 | * perhaps modified the buffer. | |
1212 | */ | |
1213 | if (buffer_dirty(bh)) { | |
1214 | smp_mb(); | |
1215 | if (buffer_dirty(bh)) | |
1216 | return; | |
1217 | } | |
1218 | ||
1219 | set_buffer_sync_flush(bh); | |
1220 | if (!test_set_buffer_dirty(bh)) { | |
1221 | struct page *page = bh->b_page; | |
1222 | struct address_space *mapping = NULL; | |
1223 | struct mem_cgroup *memcg; | |
1224 | ||
1225 | memcg = mem_cgroup_begin_page_stat(page); | |
1226 | if (!TestSetPageDirty(page)) { | |
1227 | mapping = page_mapping(page); | |
1228 | if (mapping) | |
1229 | __set_page_dirty(page, mapping, memcg, 0); | |
1230 | } | |
1231 | mem_cgroup_end_page_stat(memcg); | |
1232 | if (mapping) | |
1233 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); | |
1234 | } | |
1235 | } | |
1236 | EXPORT_SYMBOL(mark_buffer_dirty_sync); | |
1237 | ||
1da177e4 LT |
1238 | /* |
1239 | * Decrement a buffer_head's reference count. If all buffers against a page | |
1240 | * have zero reference count, are clean and unlocked, and if the page is clean | |
1241 | * and unlocked then try_to_free_buffers() may strip the buffers from the page | |
1242 | * in preparation for freeing it (sometimes, rarely, buffers are removed from | |
1243 | * a page but it ends up not being freed, and buffers may later be reattached). | |
1244 | */ | |
1245 | void __brelse(struct buffer_head * buf) | |
1246 | { | |
1247 | if (atomic_read(&buf->b_count)) { | |
1248 | put_bh(buf); | |
1249 | return; | |
1250 | } | |
5c752ad9 | 1251 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
1da177e4 | 1252 | } |
1fe72eaa | 1253 | EXPORT_SYMBOL(__brelse); |
1da177e4 LT |
1254 | |
1255 | /* | |
1256 | * bforget() is like brelse(), except it discards any | |
1257 | * potentially dirty data. | |
1258 | */ | |
1259 | void __bforget(struct buffer_head *bh) | |
1260 | { | |
1261 | clear_buffer_dirty(bh); | |
535ee2fb | 1262 | if (bh->b_assoc_map) { |
1da177e4 LT |
1263 | struct address_space *buffer_mapping = bh->b_page->mapping; |
1264 | ||
1265 | spin_lock(&buffer_mapping->private_lock); | |
1266 | list_del_init(&bh->b_assoc_buffers); | |
58ff407b | 1267 | bh->b_assoc_map = NULL; |
1da177e4 LT |
1268 | spin_unlock(&buffer_mapping->private_lock); |
1269 | } | |
1270 | __brelse(bh); | |
1271 | } | |
1fe72eaa | 1272 | EXPORT_SYMBOL(__bforget); |
1da177e4 LT |
1273 | |
1274 | static struct buffer_head *__bread_slow(struct buffer_head *bh) | |
1275 | { | |
1276 | lock_buffer(bh); | |
1277 | if (buffer_uptodate(bh)) { | |
1278 | unlock_buffer(bh); | |
1279 | return bh; | |
1280 | } else { | |
1281 | get_bh(bh); | |
1282 | bh->b_end_io = end_buffer_read_sync; | |
1283 | submit_bh(READ, bh); | |
1284 | wait_on_buffer(bh); | |
1285 | if (buffer_uptodate(bh)) | |
1286 | return bh; | |
1287 | } | |
1288 | brelse(bh); | |
1289 | return NULL; | |
1290 | } | |
1291 | ||
1292 | /* | |
1293 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). | |
1294 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their | |
1295 | * refcount elevated by one when they're in an LRU. A buffer can only appear | |
1296 | * once in a particular CPU's LRU. A single buffer can be present in multiple | |
1297 | * CPU's LRUs at the same time. | |
1298 | * | |
1299 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and | |
1300 | * sb_find_get_block(). | |
1301 | * | |
1302 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use | |
1303 | * a local interrupt disable for that. | |
1304 | */ | |
1305 | ||
86cf78d7 | 1306 | #define BH_LRU_SIZE 16 |
1da177e4 LT |
1307 | |
1308 | struct bh_lru { | |
1309 | struct buffer_head *bhs[BH_LRU_SIZE]; | |
1310 | }; | |
1311 | ||
1312 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; | |
1313 | ||
1314 | #ifdef CONFIG_SMP | |
1315 | #define bh_lru_lock() local_irq_disable() | |
1316 | #define bh_lru_unlock() local_irq_enable() | |
1317 | #else | |
1318 | #define bh_lru_lock() preempt_disable() | |
1319 | #define bh_lru_unlock() preempt_enable() | |
1320 | #endif | |
1321 | ||
1322 | static inline void check_irqs_on(void) | |
1323 | { | |
1324 | #ifdef irqs_disabled | |
1325 | BUG_ON(irqs_disabled()); | |
1326 | #endif | |
1327 | } | |
1328 | ||
1329 | /* | |
1330 | * The LRU management algorithm is dopey-but-simple. Sorry. | |
1331 | */ | |
1332 | static void bh_lru_install(struct buffer_head *bh) | |
1333 | { | |
1334 | struct buffer_head *evictee = NULL; | |
1da177e4 LT |
1335 | |
1336 | check_irqs_on(); | |
1337 | bh_lru_lock(); | |
c7b92516 | 1338 | if (__this_cpu_read(bh_lrus.bhs[0]) != bh) { |
1da177e4 LT |
1339 | struct buffer_head *bhs[BH_LRU_SIZE]; |
1340 | int in; | |
1341 | int out = 0; | |
1342 | ||
1343 | get_bh(bh); | |
1344 | bhs[out++] = bh; | |
1345 | for (in = 0; in < BH_LRU_SIZE; in++) { | |
c7b92516 CL |
1346 | struct buffer_head *bh2 = |
1347 | __this_cpu_read(bh_lrus.bhs[in]); | |
1da177e4 LT |
1348 | |
1349 | if (bh2 == bh) { | |
1350 | __brelse(bh2); | |
1351 | } else { | |
1352 | if (out >= BH_LRU_SIZE) { | |
1353 | BUG_ON(evictee != NULL); | |
1354 | evictee = bh2; | |
1355 | } else { | |
1356 | bhs[out++] = bh2; | |
1357 | } | |
1358 | } | |
1359 | } | |
1360 | while (out < BH_LRU_SIZE) | |
1361 | bhs[out++] = NULL; | |
ca6673b0 | 1362 | memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs)); |
1da177e4 LT |
1363 | } |
1364 | bh_lru_unlock(); | |
1365 | ||
1366 | if (evictee) | |
1367 | __brelse(evictee); | |
1368 | } | |
1369 | ||
1370 | /* | |
1371 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. | |
1372 | */ | |
858119e1 | 1373 | static struct buffer_head * |
3991d3bd | 1374 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1375 | { |
1376 | struct buffer_head *ret = NULL; | |
3991d3bd | 1377 | unsigned int i; |
1da177e4 LT |
1378 | |
1379 | check_irqs_on(); | |
1380 | bh_lru_lock(); | |
1da177e4 | 1381 | for (i = 0; i < BH_LRU_SIZE; i++) { |
c7b92516 | 1382 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
1da177e4 | 1383 | |
9470dd5d ZB |
1384 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
1385 | bh->b_size == size) { | |
1da177e4 LT |
1386 | if (i) { |
1387 | while (i) { | |
c7b92516 CL |
1388 | __this_cpu_write(bh_lrus.bhs[i], |
1389 | __this_cpu_read(bh_lrus.bhs[i - 1])); | |
1da177e4 LT |
1390 | i--; |
1391 | } | |
c7b92516 | 1392 | __this_cpu_write(bh_lrus.bhs[0], bh); |
1da177e4 LT |
1393 | } |
1394 | get_bh(bh); | |
1395 | ret = bh; | |
1396 | break; | |
1397 | } | |
1398 | } | |
1399 | bh_lru_unlock(); | |
1400 | return ret; | |
1401 | } | |
1402 | ||
1403 | /* | |
1404 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh | |
1405 | * it in the LRU and mark it as accessed. If it is not present then return | |
1406 | * NULL | |
1407 | */ | |
1408 | struct buffer_head * | |
3991d3bd | 1409 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1410 | { |
1411 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); | |
1412 | ||
1413 | if (bh == NULL) { | |
2457aec6 | 1414 | /* __find_get_block_slow will mark the page accessed */ |
385fd4c5 | 1415 | bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1416 | if (bh) |
1417 | bh_lru_install(bh); | |
2457aec6 | 1418 | } else |
1da177e4 | 1419 | touch_buffer(bh); |
2457aec6 | 1420 | |
1da177e4 LT |
1421 | return bh; |
1422 | } | |
1423 | EXPORT_SYMBOL(__find_get_block); | |
1424 | ||
1425 | /* | |
3b5e6454 | 1426 | * __getblk_gfp() will locate (and, if necessary, create) the buffer_head |
1da177e4 LT |
1427 | * which corresponds to the passed block_device, block and size. The |
1428 | * returned buffer has its reference count incremented. | |
1429 | * | |
3b5e6454 GK |
1430 | * __getblk_gfp() will lock up the machine if grow_dev_page's |
1431 | * try_to_free_buffers() attempt is failing. FIXME, perhaps? | |
1da177e4 LT |
1432 | */ |
1433 | struct buffer_head * | |
3b5e6454 GK |
1434 | __getblk_gfp(struct block_device *bdev, sector_t block, |
1435 | unsigned size, gfp_t gfp) | |
1da177e4 LT |
1436 | { |
1437 | struct buffer_head *bh = __find_get_block(bdev, block, size); | |
1438 | ||
1439 | might_sleep(); | |
1440 | if (bh == NULL) | |
3b5e6454 | 1441 | bh = __getblk_slow(bdev, block, size, gfp); |
1da177e4 LT |
1442 | return bh; |
1443 | } | |
3b5e6454 | 1444 | EXPORT_SYMBOL(__getblk_gfp); |
1da177e4 LT |
1445 | |
1446 | /* | |
1447 | * Do async read-ahead on a buffer.. | |
1448 | */ | |
3991d3bd | 1449 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
1da177e4 LT |
1450 | { |
1451 | struct buffer_head *bh = __getblk(bdev, block, size); | |
a3e713b5 AM |
1452 | if (likely(bh)) { |
1453 | ll_rw_block(READA, 1, &bh); | |
1454 | brelse(bh); | |
1455 | } | |
1da177e4 LT |
1456 | } |
1457 | EXPORT_SYMBOL(__breadahead); | |
1458 | ||
1459 | /** | |
3b5e6454 | 1460 | * __bread_gfp() - reads a specified block and returns the bh |
67be2dd1 | 1461 | * @bdev: the block_device to read from |
1da177e4 LT |
1462 | * @block: number of block |
1463 | * @size: size (in bytes) to read | |
3b5e6454 GK |
1464 | * @gfp: page allocation flag |
1465 | * | |
1da177e4 | 1466 | * Reads a specified block, and returns buffer head that contains it. |
3b5e6454 GK |
1467 | * The page cache can be allocated from non-movable area |
1468 | * not to prevent page migration if you set gfp to zero. | |
1da177e4 LT |
1469 | * It returns NULL if the block was unreadable. |
1470 | */ | |
1471 | struct buffer_head * | |
3b5e6454 GK |
1472 | __bread_gfp(struct block_device *bdev, sector_t block, |
1473 | unsigned size, gfp_t gfp) | |
1da177e4 | 1474 | { |
3b5e6454 | 1475 | struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp); |
1da177e4 | 1476 | |
a3e713b5 | 1477 | if (likely(bh) && !buffer_uptodate(bh)) |
1da177e4 LT |
1478 | bh = __bread_slow(bh); |
1479 | return bh; | |
1480 | } | |
3b5e6454 | 1481 | EXPORT_SYMBOL(__bread_gfp); |
1da177e4 LT |
1482 | |
1483 | /* | |
1484 | * invalidate_bh_lrus() is called rarely - but not only at unmount. | |
1485 | * This doesn't race because it runs in each cpu either in irq | |
1486 | * or with preempt disabled. | |
1487 | */ | |
1488 | static void invalidate_bh_lru(void *arg) | |
1489 | { | |
1490 | struct bh_lru *b = &get_cpu_var(bh_lrus); | |
1491 | int i; | |
1492 | ||
1493 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
1494 | brelse(b->bhs[i]); | |
1495 | b->bhs[i] = NULL; | |
1496 | } | |
1497 | put_cpu_var(bh_lrus); | |
1498 | } | |
42be35d0 GBY |
1499 | |
1500 | static bool has_bh_in_lru(int cpu, void *dummy) | |
1501 | { | |
1502 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); | |
1503 | int i; | |
1da177e4 | 1504 | |
42be35d0 GBY |
1505 | for (i = 0; i < BH_LRU_SIZE; i++) { |
1506 | if (b->bhs[i]) | |
1507 | return 1; | |
1508 | } | |
1509 | ||
1510 | return 0; | |
1511 | } | |
1512 | ||
f9a14399 | 1513 | void invalidate_bh_lrus(void) |
1da177e4 | 1514 | { |
42be35d0 | 1515 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL); |
1da177e4 | 1516 | } |
9db5579b | 1517 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
1da177e4 LT |
1518 | |
1519 | void set_bh_page(struct buffer_head *bh, | |
1520 | struct page *page, unsigned long offset) | |
1521 | { | |
1522 | bh->b_page = page; | |
e827f923 | 1523 | BUG_ON(offset >= PAGE_SIZE); |
1da177e4 LT |
1524 | if (PageHighMem(page)) |
1525 | /* | |
1526 | * This catches illegal uses and preserves the offset: | |
1527 | */ | |
1528 | bh->b_data = (char *)(0 + offset); | |
1529 | else | |
1530 | bh->b_data = page_address(page) + offset; | |
1531 | } | |
1532 | EXPORT_SYMBOL(set_bh_page); | |
1533 | ||
1534 | /* | |
1535 | * Called when truncating a buffer on a page completely. | |
1536 | */ | |
e7470ee8 MG |
1537 | |
1538 | /* Bits that are cleared during an invalidate */ | |
1539 | #define BUFFER_FLAGS_DISCARD \ | |
1540 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ | |
1541 | 1 << BH_Delay | 1 << BH_Unwritten) | |
1542 | ||
858119e1 | 1543 | static void discard_buffer(struct buffer_head * bh) |
1da177e4 | 1544 | { |
e7470ee8 MG |
1545 | unsigned long b_state, b_state_old; |
1546 | ||
1da177e4 LT |
1547 | lock_buffer(bh); |
1548 | clear_buffer_dirty(bh); | |
1549 | bh->b_bdev = NULL; | |
e7470ee8 MG |
1550 | b_state = bh->b_state; |
1551 | for (;;) { | |
1552 | b_state_old = cmpxchg(&bh->b_state, b_state, | |
1553 | (b_state & ~BUFFER_FLAGS_DISCARD)); | |
1554 | if (b_state_old == b_state) | |
1555 | break; | |
1556 | b_state = b_state_old; | |
1557 | } | |
1da177e4 LT |
1558 | unlock_buffer(bh); |
1559 | } | |
1560 | ||
1da177e4 | 1561 | /** |
814e1d25 | 1562 | * block_invalidatepage - invalidate part or all of a buffer-backed page |
1da177e4 LT |
1563 | * |
1564 | * @page: the page which is affected | |
d47992f8 LC |
1565 | * @offset: start of the range to invalidate |
1566 | * @length: length of the range to invalidate | |
1da177e4 LT |
1567 | * |
1568 | * block_invalidatepage() is called when all or part of the page has become | |
814e1d25 | 1569 | * invalidated by a truncate operation. |
1da177e4 LT |
1570 | * |
1571 | * block_invalidatepage() does not have to release all buffers, but it must | |
1572 | * ensure that no dirty buffer is left outside @offset and that no I/O | |
1573 | * is underway against any of the blocks which are outside the truncation | |
1574 | * point. Because the caller is about to free (and possibly reuse) those | |
1575 | * blocks on-disk. | |
1576 | */ | |
d47992f8 LC |
1577 | void block_invalidatepage(struct page *page, unsigned int offset, |
1578 | unsigned int length) | |
1da177e4 LT |
1579 | { |
1580 | struct buffer_head *head, *bh, *next; | |
1581 | unsigned int curr_off = 0; | |
d47992f8 | 1582 | unsigned int stop = length + offset; |
1da177e4 LT |
1583 | |
1584 | BUG_ON(!PageLocked(page)); | |
1585 | if (!page_has_buffers(page)) | |
1586 | goto out; | |
1587 | ||
d47992f8 LC |
1588 | /* |
1589 | * Check for overflow | |
1590 | */ | |
1591 | BUG_ON(stop > PAGE_CACHE_SIZE || stop < length); | |
1592 | ||
1da177e4 LT |
1593 | head = page_buffers(page); |
1594 | bh = head; | |
1595 | do { | |
1596 | unsigned int next_off = curr_off + bh->b_size; | |
1597 | next = bh->b_this_page; | |
1598 | ||
d47992f8 LC |
1599 | /* |
1600 | * Are we still fully in range ? | |
1601 | */ | |
1602 | if (next_off > stop) | |
1603 | goto out; | |
1604 | ||
1da177e4 LT |
1605 | /* |
1606 | * is this block fully invalidated? | |
1607 | */ | |
1608 | if (offset <= curr_off) | |
1609 | discard_buffer(bh); | |
1610 | curr_off = next_off; | |
1611 | bh = next; | |
1612 | } while (bh != head); | |
1613 | ||
1614 | /* | |
1615 | * We release buffers only if the entire page is being invalidated. | |
1616 | * The get_block cached value has been unconditionally invalidated, | |
1617 | * so real IO is not possible anymore. | |
1618 | */ | |
1619 | if (offset == 0) | |
2ff28e22 | 1620 | try_to_release_page(page, 0); |
1da177e4 | 1621 | out: |
2ff28e22 | 1622 | return; |
1da177e4 LT |
1623 | } |
1624 | EXPORT_SYMBOL(block_invalidatepage); | |
1625 | ||
d47992f8 | 1626 | |
1da177e4 LT |
1627 | /* |
1628 | * We attach and possibly dirty the buffers atomically wrt | |
1629 | * __set_page_dirty_buffers() via private_lock. try_to_free_buffers | |
1630 | * is already excluded via the page lock. | |
1631 | */ | |
1632 | void create_empty_buffers(struct page *page, | |
1633 | unsigned long blocksize, unsigned long b_state) | |
1634 | { | |
1635 | struct buffer_head *bh, *head, *tail; | |
1636 | ||
1637 | head = alloc_page_buffers(page, blocksize, 1); | |
1638 | bh = head; | |
1639 | do { | |
1640 | bh->b_state |= b_state; | |
1641 | tail = bh; | |
1642 | bh = bh->b_this_page; | |
1643 | } while (bh); | |
1644 | tail->b_this_page = head; | |
1645 | ||
1646 | spin_lock(&page->mapping->private_lock); | |
1647 | if (PageUptodate(page) || PageDirty(page)) { | |
1648 | bh = head; | |
1649 | do { | |
1650 | if (PageDirty(page)) | |
1651 | set_buffer_dirty(bh); | |
1652 | if (PageUptodate(page)) | |
1653 | set_buffer_uptodate(bh); | |
1654 | bh = bh->b_this_page; | |
1655 | } while (bh != head); | |
1656 | } | |
1657 | attach_page_buffers(page, head); | |
1658 | spin_unlock(&page->mapping->private_lock); | |
1659 | } | |
1660 | EXPORT_SYMBOL(create_empty_buffers); | |
1661 | ||
1662 | /* | |
1663 | * We are taking a block for data and we don't want any output from any | |
1664 | * buffer-cache aliases starting from return from that function and | |
1665 | * until the moment when something will explicitly mark the buffer | |
1666 | * dirty (hopefully that will not happen until we will free that block ;-) | |
1667 | * We don't even need to mark it not-uptodate - nobody can expect | |
1668 | * anything from a newly allocated buffer anyway. We used to used | |
1669 | * unmap_buffer() for such invalidation, but that was wrong. We definitely | |
1670 | * don't want to mark the alias unmapped, for example - it would confuse | |
1671 | * anyone who might pick it with bread() afterwards... | |
1672 | * | |
1673 | * Also.. Note that bforget() doesn't lock the buffer. So there can | |
1674 | * be writeout I/O going on against recently-freed buffers. We don't | |
1675 | * wait on that I/O in bforget() - it's more efficient to wait on the I/O | |
1676 | * only if we really need to. That happens here. | |
1677 | */ | |
1678 | void unmap_underlying_metadata(struct block_device *bdev, sector_t block) | |
1679 | { | |
1680 | struct buffer_head *old_bh; | |
1681 | ||
1682 | might_sleep(); | |
1683 | ||
385fd4c5 | 1684 | old_bh = __find_get_block_slow(bdev, block); |
1da177e4 LT |
1685 | if (old_bh) { |
1686 | clear_buffer_dirty(old_bh); | |
1687 | wait_on_buffer(old_bh); | |
1688 | clear_buffer_req(old_bh); | |
1689 | __brelse(old_bh); | |
1690 | } | |
1691 | } | |
1692 | EXPORT_SYMBOL(unmap_underlying_metadata); | |
1693 | ||
45bce8f3 LT |
1694 | /* |
1695 | * Size is a power-of-two in the range 512..PAGE_SIZE, | |
1696 | * and the case we care about most is PAGE_SIZE. | |
1697 | * | |
1698 | * So this *could* possibly be written with those | |
1699 | * constraints in mind (relevant mostly if some | |
1700 | * architecture has a slow bit-scan instruction) | |
1701 | */ | |
1702 | static inline int block_size_bits(unsigned int blocksize) | |
1703 | { | |
1704 | return ilog2(blocksize); | |
1705 | } | |
1706 | ||
1707 | static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state) | |
1708 | { | |
1709 | BUG_ON(!PageLocked(page)); | |
1710 | ||
1711 | if (!page_has_buffers(page)) | |
1712 | create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state); | |
1713 | return page_buffers(page); | |
1714 | } | |
1715 | ||
1da177e4 LT |
1716 | /* |
1717 | * NOTE! All mapped/uptodate combinations are valid: | |
1718 | * | |
1719 | * Mapped Uptodate Meaning | |
1720 | * | |
1721 | * No No "unknown" - must do get_block() | |
1722 | * No Yes "hole" - zero-filled | |
1723 | * Yes No "allocated" - allocated on disk, not read in | |
1724 | * Yes Yes "valid" - allocated and up-to-date in memory. | |
1725 | * | |
1726 | * "Dirty" is valid only with the last case (mapped+uptodate). | |
1727 | */ | |
1728 | ||
1729 | /* | |
1730 | * While block_write_full_page is writing back the dirty buffers under | |
1731 | * the page lock, whoever dirtied the buffers may decide to clean them | |
1732 | * again at any time. We handle that by only looking at the buffer | |
1733 | * state inside lock_buffer(). | |
1734 | * | |
1735 | * If block_write_full_page() is called for regular writeback | |
1736 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a | |
1737 | * locked buffer. This only can happen if someone has written the buffer | |
1738 | * directly, with submit_bh(). At the address_space level PageWriteback | |
1739 | * prevents this contention from occurring. | |
6e34eedd TT |
1740 | * |
1741 | * If block_write_full_page() is called with wbc->sync_mode == | |
721a9602 JA |
1742 | * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this |
1743 | * causes the writes to be flagged as synchronous writes. | |
1da177e4 LT |
1744 | */ |
1745 | static int __block_write_full_page(struct inode *inode, struct page *page, | |
35c80d5f CM |
1746 | get_block_t *get_block, struct writeback_control *wbc, |
1747 | bh_end_io_t *handler) | |
1da177e4 LT |
1748 | { |
1749 | int err; | |
1750 | sector_t block; | |
1751 | sector_t last_block; | |
f0fbd5fc | 1752 | struct buffer_head *bh, *head; |
45bce8f3 | 1753 | unsigned int blocksize, bbits; |
1da177e4 | 1754 | int nr_underway = 0; |
bafc0dba | 1755 | int write_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE); |
1da177e4 | 1756 | |
45bce8f3 | 1757 | head = create_page_buffers(page, inode, |
1da177e4 | 1758 | (1 << BH_Dirty)|(1 << BH_Uptodate)); |
1da177e4 LT |
1759 | |
1760 | /* | |
1761 | * Be very careful. We have no exclusion from __set_page_dirty_buffers | |
1762 | * here, and the (potentially unmapped) buffers may become dirty at | |
1763 | * any time. If a buffer becomes dirty here after we've inspected it | |
1764 | * then we just miss that fact, and the page stays dirty. | |
1765 | * | |
1766 | * Buffers outside i_size may be dirtied by __set_page_dirty_buffers; | |
1767 | * handle that here by just cleaning them. | |
1768 | */ | |
1769 | ||
1da177e4 | 1770 | bh = head; |
45bce8f3 LT |
1771 | blocksize = bh->b_size; |
1772 | bbits = block_size_bits(blocksize); | |
1773 | ||
1774 | block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); | |
1775 | last_block = (i_size_read(inode) - 1) >> bbits; | |
1da177e4 LT |
1776 | |
1777 | /* | |
1778 | * Get all the dirty buffers mapped to disk addresses and | |
1779 | * handle any aliases from the underlying blockdev's mapping. | |
1780 | */ | |
1781 | do { | |
1782 | if (block > last_block) { | |
1783 | /* | |
1784 | * mapped buffers outside i_size will occur, because | |
1785 | * this page can be outside i_size when there is a | |
1786 | * truncate in progress. | |
1787 | */ | |
1788 | /* | |
1789 | * The buffer was zeroed by block_write_full_page() | |
1790 | */ | |
1791 | clear_buffer_dirty(bh); | |
1792 | set_buffer_uptodate(bh); | |
29a814d2 AT |
1793 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
1794 | buffer_dirty(bh)) { | |
b0cf2321 | 1795 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1796 | err = get_block(inode, block, bh, 1); |
1797 | if (err) | |
1798 | goto recover; | |
29a814d2 | 1799 | clear_buffer_delay(bh); |
1da177e4 LT |
1800 | if (buffer_new(bh)) { |
1801 | /* blockdev mappings never come here */ | |
1802 | clear_buffer_new(bh); | |
1803 | unmap_underlying_metadata(bh->b_bdev, | |
1804 | bh->b_blocknr); | |
1805 | } | |
1806 | } | |
1807 | bh = bh->b_this_page; | |
1808 | block++; | |
1809 | } while (bh != head); | |
1810 | ||
1811 | do { | |
1da177e4 LT |
1812 | if (!buffer_mapped(bh)) |
1813 | continue; | |
1814 | /* | |
1815 | * If it's a fully non-blocking write attempt and we cannot | |
1816 | * lock the buffer then redirty the page. Note that this can | |
5b0830cb JA |
1817 | * potentially cause a busy-wait loop from writeback threads |
1818 | * and kswapd activity, but those code paths have their own | |
1819 | * higher-level throttling. | |
1da177e4 | 1820 | */ |
1b430bee | 1821 | if (wbc->sync_mode != WB_SYNC_NONE) { |
1da177e4 | 1822 | lock_buffer(bh); |
ca5de404 | 1823 | } else if (!trylock_buffer(bh)) { |
1da177e4 LT |
1824 | redirty_page_for_writepage(wbc, page); |
1825 | continue; | |
1826 | } | |
1827 | if (test_clear_buffer_dirty(bh)) { | |
35c80d5f | 1828 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1829 | } else { |
1830 | unlock_buffer(bh); | |
1831 | } | |
1832 | } while ((bh = bh->b_this_page) != head); | |
1833 | ||
1834 | /* | |
1835 | * The page and its buffers are protected by PageWriteback(), so we can | |
1836 | * drop the bh refcounts early. | |
1837 | */ | |
1838 | BUG_ON(PageWriteback(page)); | |
1839 | set_page_writeback(page); | |
1da177e4 LT |
1840 | |
1841 | do { | |
1842 | struct buffer_head *next = bh->b_this_page; | |
1843 | if (buffer_async_write(bh)) { | |
b16b1deb | 1844 | submit_bh_wbc(write_op, bh, 0, wbc); |
1da177e4 LT |
1845 | nr_underway++; |
1846 | } | |
1da177e4 LT |
1847 | bh = next; |
1848 | } while (bh != head); | |
05937baa | 1849 | unlock_page(page); |
1da177e4 LT |
1850 | |
1851 | err = 0; | |
1852 | done: | |
1853 | if (nr_underway == 0) { | |
1854 | /* | |
1855 | * The page was marked dirty, but the buffers were | |
1856 | * clean. Someone wrote them back by hand with | |
1857 | * ll_rw_block/submit_bh. A rare case. | |
1858 | */ | |
1da177e4 | 1859 | end_page_writeback(page); |
3d67f2d7 | 1860 | |
1da177e4 LT |
1861 | /* |
1862 | * The page and buffer_heads can be released at any time from | |
1863 | * here on. | |
1864 | */ | |
1da177e4 LT |
1865 | } |
1866 | return err; | |
1867 | ||
1868 | recover: | |
1869 | /* | |
1870 | * ENOSPC, or some other error. We may already have added some | |
1871 | * blocks to the file, so we need to write these out to avoid | |
1872 | * exposing stale data. | |
1873 | * The page is currently locked and not marked for writeback | |
1874 | */ | |
1875 | bh = head; | |
1876 | /* Recovery: lock and submit the mapped buffers */ | |
1877 | do { | |
29a814d2 AT |
1878 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
1879 | !buffer_delay(bh)) { | |
1da177e4 | 1880 | lock_buffer(bh); |
35c80d5f | 1881 | mark_buffer_async_write_endio(bh, handler); |
1da177e4 LT |
1882 | } else { |
1883 | /* | |
1884 | * The buffer may have been set dirty during | |
1885 | * attachment to a dirty page. | |
1886 | */ | |
1887 | clear_buffer_dirty(bh); | |
1888 | } | |
1889 | } while ((bh = bh->b_this_page) != head); | |
1890 | SetPageError(page); | |
1891 | BUG_ON(PageWriteback(page)); | |
7e4c3690 | 1892 | mapping_set_error(page->mapping, err); |
1da177e4 | 1893 | set_page_writeback(page); |
1da177e4 LT |
1894 | do { |
1895 | struct buffer_head *next = bh->b_this_page; | |
1896 | if (buffer_async_write(bh)) { | |
1897 | clear_buffer_dirty(bh); | |
b16b1deb | 1898 | submit_bh_wbc(write_op, bh, 0, wbc); |
1da177e4 LT |
1899 | nr_underway++; |
1900 | } | |
1da177e4 LT |
1901 | bh = next; |
1902 | } while (bh != head); | |
ffda9d30 | 1903 | unlock_page(page); |
1da177e4 LT |
1904 | goto done; |
1905 | } | |
1906 | ||
afddba49 NP |
1907 | /* |
1908 | * If a page has any new buffers, zero them out here, and mark them uptodate | |
1909 | * and dirty so they'll be written out (in order to prevent uninitialised | |
1910 | * block data from leaking). And clear the new bit. | |
1911 | */ | |
1912 | void page_zero_new_buffers(struct page *page, unsigned from, unsigned to) | |
1913 | { | |
1914 | unsigned int block_start, block_end; | |
1915 | struct buffer_head *head, *bh; | |
1916 | ||
1917 | BUG_ON(!PageLocked(page)); | |
1918 | if (!page_has_buffers(page)) | |
1919 | return; | |
1920 | ||
1921 | bh = head = page_buffers(page); | |
1922 | block_start = 0; | |
1923 | do { | |
1924 | block_end = block_start + bh->b_size; | |
1925 | ||
1926 | if (buffer_new(bh)) { | |
1927 | if (block_end > from && block_start < to) { | |
1928 | if (!PageUptodate(page)) { | |
1929 | unsigned start, size; | |
1930 | ||
1931 | start = max(from, block_start); | |
1932 | size = min(to, block_end) - start; | |
1933 | ||
eebd2aa3 | 1934 | zero_user(page, start, size); |
afddba49 NP |
1935 | set_buffer_uptodate(bh); |
1936 | } | |
1937 | ||
1938 | clear_buffer_new(bh); | |
1939 | mark_buffer_dirty(bh); | |
1940 | } | |
1941 | } | |
1942 | ||
1943 | block_start = block_end; | |
1944 | bh = bh->b_this_page; | |
1945 | } while (bh != head); | |
1946 | } | |
1947 | EXPORT_SYMBOL(page_zero_new_buffers); | |
1948 | ||
ebdec241 | 1949 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, |
6e1db88d | 1950 | get_block_t *get_block) |
1da177e4 | 1951 | { |
ebdec241 CH |
1952 | unsigned from = pos & (PAGE_CACHE_SIZE - 1); |
1953 | unsigned to = from + len; | |
6e1db88d | 1954 | struct inode *inode = page->mapping->host; |
1da177e4 LT |
1955 | unsigned block_start, block_end; |
1956 | sector_t block; | |
1957 | int err = 0; | |
1958 | unsigned blocksize, bbits; | |
1959 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; | |
1960 | ||
1961 | BUG_ON(!PageLocked(page)); | |
1962 | BUG_ON(from > PAGE_CACHE_SIZE); | |
1963 | BUG_ON(to > PAGE_CACHE_SIZE); | |
1964 | BUG_ON(from > to); | |
1965 | ||
45bce8f3 LT |
1966 | head = create_page_buffers(page, inode, 0); |
1967 | blocksize = head->b_size; | |
1968 | bbits = block_size_bits(blocksize); | |
1da177e4 | 1969 | |
1da177e4 LT |
1970 | block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); |
1971 | ||
1972 | for(bh = head, block_start = 0; bh != head || !block_start; | |
1973 | block++, block_start=block_end, bh = bh->b_this_page) { | |
1974 | block_end = block_start + blocksize; | |
1975 | if (block_end <= from || block_start >= to) { | |
1976 | if (PageUptodate(page)) { | |
1977 | if (!buffer_uptodate(bh)) | |
1978 | set_buffer_uptodate(bh); | |
1979 | } | |
1980 | continue; | |
1981 | } | |
1982 | if (buffer_new(bh)) | |
1983 | clear_buffer_new(bh); | |
1984 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 1985 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
1986 | err = get_block(inode, block, bh, 1); |
1987 | if (err) | |
f3ddbdc6 | 1988 | break; |
1da177e4 | 1989 | if (buffer_new(bh)) { |
1da177e4 LT |
1990 | unmap_underlying_metadata(bh->b_bdev, |
1991 | bh->b_blocknr); | |
1992 | if (PageUptodate(page)) { | |
637aff46 | 1993 | clear_buffer_new(bh); |
1da177e4 | 1994 | set_buffer_uptodate(bh); |
637aff46 | 1995 | mark_buffer_dirty(bh); |
1da177e4 LT |
1996 | continue; |
1997 | } | |
eebd2aa3 CL |
1998 | if (block_end > to || block_start < from) |
1999 | zero_user_segments(page, | |
2000 | to, block_end, | |
2001 | block_start, from); | |
1da177e4 LT |
2002 | continue; |
2003 | } | |
2004 | } | |
2005 | if (PageUptodate(page)) { | |
2006 | if (!buffer_uptodate(bh)) | |
2007 | set_buffer_uptodate(bh); | |
2008 | continue; | |
2009 | } | |
2010 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && | |
33a266dd | 2011 | !buffer_unwritten(bh) && |
1da177e4 LT |
2012 | (block_start < from || block_end > to)) { |
2013 | ll_rw_block(READ, 1, &bh); | |
2014 | *wait_bh++=bh; | |
2015 | } | |
2016 | } | |
2017 | /* | |
2018 | * If we issued read requests - let them complete. | |
2019 | */ | |
2020 | while(wait_bh > wait) { | |
2021 | wait_on_buffer(*--wait_bh); | |
2022 | if (!buffer_uptodate(*wait_bh)) | |
f3ddbdc6 | 2023 | err = -EIO; |
1da177e4 | 2024 | } |
f9f07b6c | 2025 | if (unlikely(err)) |
afddba49 | 2026 | page_zero_new_buffers(page, from, to); |
1da177e4 LT |
2027 | return err; |
2028 | } | |
ebdec241 | 2029 | EXPORT_SYMBOL(__block_write_begin); |
1da177e4 LT |
2030 | |
2031 | static int __block_commit_write(struct inode *inode, struct page *page, | |
2032 | unsigned from, unsigned to) | |
2033 | { | |
2034 | unsigned block_start, block_end; | |
2035 | int partial = 0; | |
2036 | unsigned blocksize; | |
2037 | struct buffer_head *bh, *head; | |
2038 | ||
45bce8f3 LT |
2039 | bh = head = page_buffers(page); |
2040 | blocksize = bh->b_size; | |
1da177e4 | 2041 | |
45bce8f3 LT |
2042 | block_start = 0; |
2043 | do { | |
1da177e4 LT |
2044 | block_end = block_start + blocksize; |
2045 | if (block_end <= from || block_start >= to) { | |
2046 | if (!buffer_uptodate(bh)) | |
2047 | partial = 1; | |
2048 | } else { | |
2049 | set_buffer_uptodate(bh); | |
2050 | mark_buffer_dirty(bh); | |
2051 | } | |
afddba49 | 2052 | clear_buffer_new(bh); |
45bce8f3 LT |
2053 | |
2054 | block_start = block_end; | |
2055 | bh = bh->b_this_page; | |
2056 | } while (bh != head); | |
1da177e4 LT |
2057 | |
2058 | /* | |
2059 | * If this is a partial write which happened to make all buffers | |
2060 | * uptodate then we can optimize away a bogus readpage() for | |
2061 | * the next read(). Here we 'discover' whether the page went | |
2062 | * uptodate as a result of this (potentially partial) write. | |
2063 | */ | |
2064 | if (!partial) | |
2065 | SetPageUptodate(page); | |
2066 | return 0; | |
2067 | } | |
2068 | ||
afddba49 | 2069 | /* |
155130a4 CH |
2070 | * block_write_begin takes care of the basic task of block allocation and |
2071 | * bringing partial write blocks uptodate first. | |
2072 | * | |
7bb46a67 | 2073 | * The filesystem needs to handle block truncation upon failure. |
afddba49 | 2074 | */ |
155130a4 CH |
2075 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
2076 | unsigned flags, struct page **pagep, get_block_t *get_block) | |
afddba49 | 2077 | { |
6e1db88d | 2078 | pgoff_t index = pos >> PAGE_CACHE_SHIFT; |
afddba49 | 2079 | struct page *page; |
6e1db88d | 2080 | int status; |
afddba49 | 2081 | |
6e1db88d CH |
2082 | page = grab_cache_page_write_begin(mapping, index, flags); |
2083 | if (!page) | |
2084 | return -ENOMEM; | |
afddba49 | 2085 | |
6e1db88d | 2086 | status = __block_write_begin(page, pos, len, get_block); |
afddba49 | 2087 | if (unlikely(status)) { |
6e1db88d CH |
2088 | unlock_page(page); |
2089 | page_cache_release(page); | |
2090 | page = NULL; | |
afddba49 NP |
2091 | } |
2092 | ||
6e1db88d | 2093 | *pagep = page; |
afddba49 NP |
2094 | return status; |
2095 | } | |
2096 | EXPORT_SYMBOL(block_write_begin); | |
2097 | ||
2098 | int block_write_end(struct file *file, struct address_space *mapping, | |
2099 | loff_t pos, unsigned len, unsigned copied, | |
2100 | struct page *page, void *fsdata) | |
2101 | { | |
2102 | struct inode *inode = mapping->host; | |
2103 | unsigned start; | |
2104 | ||
2105 | start = pos & (PAGE_CACHE_SIZE - 1); | |
2106 | ||
2107 | if (unlikely(copied < len)) { | |
2108 | /* | |
2109 | * The buffers that were written will now be uptodate, so we | |
2110 | * don't have to worry about a readpage reading them and | |
2111 | * overwriting a partial write. However if we have encountered | |
2112 | * a short write and only partially written into a buffer, it | |
2113 | * will not be marked uptodate, so a readpage might come in and | |
2114 | * destroy our partial write. | |
2115 | * | |
2116 | * Do the simplest thing, and just treat any short write to a | |
2117 | * non uptodate page as a zero-length write, and force the | |
2118 | * caller to redo the whole thing. | |
2119 | */ | |
2120 | if (!PageUptodate(page)) | |
2121 | copied = 0; | |
2122 | ||
2123 | page_zero_new_buffers(page, start+copied, start+len); | |
2124 | } | |
2125 | flush_dcache_page(page); | |
2126 | ||
2127 | /* This could be a short (even 0-length) commit */ | |
2128 | __block_commit_write(inode, page, start, start+copied); | |
2129 | ||
2130 | return copied; | |
2131 | } | |
2132 | EXPORT_SYMBOL(block_write_end); | |
2133 | ||
2134 | int generic_write_end(struct file *file, struct address_space *mapping, | |
2135 | loff_t pos, unsigned len, unsigned copied, | |
2136 | struct page *page, void *fsdata) | |
2137 | { | |
2138 | struct inode *inode = mapping->host; | |
90a80202 | 2139 | loff_t old_size = inode->i_size; |
c7d206b3 | 2140 | int i_size_changed = 0; |
afddba49 NP |
2141 | |
2142 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); | |
2143 | ||
2144 | /* | |
2145 | * No need to use i_size_read() here, the i_size | |
2146 | * cannot change under us because we hold i_mutex. | |
2147 | * | |
2148 | * But it's important to update i_size while still holding page lock: | |
2149 | * page writeout could otherwise come in and zero beyond i_size. | |
2150 | */ | |
2151 | if (pos+copied > inode->i_size) { | |
2152 | i_size_write(inode, pos+copied); | |
c7d206b3 | 2153 | i_size_changed = 1; |
afddba49 NP |
2154 | } |
2155 | ||
2156 | unlock_page(page); | |
2157 | page_cache_release(page); | |
2158 | ||
90a80202 JK |
2159 | if (old_size < pos) |
2160 | pagecache_isize_extended(inode, old_size, pos); | |
c7d206b3 JK |
2161 | /* |
2162 | * Don't mark the inode dirty under page lock. First, it unnecessarily | |
2163 | * makes the holding time of page lock longer. Second, it forces lock | |
2164 | * ordering of page lock and transaction start for journaling | |
2165 | * filesystems. | |
2166 | */ | |
2167 | if (i_size_changed) | |
2168 | mark_inode_dirty(inode); | |
2169 | ||
afddba49 NP |
2170 | return copied; |
2171 | } | |
2172 | EXPORT_SYMBOL(generic_write_end); | |
2173 | ||
8ab22b9a HH |
2174 | /* |
2175 | * block_is_partially_uptodate checks whether buffers within a page are | |
2176 | * uptodate or not. | |
2177 | * | |
2178 | * Returns true if all buffers which correspond to a file portion | |
2179 | * we want to read are uptodate. | |
2180 | */ | |
c186afb4 AV |
2181 | int block_is_partially_uptodate(struct page *page, unsigned long from, |
2182 | unsigned long count) | |
8ab22b9a | 2183 | { |
8ab22b9a HH |
2184 | unsigned block_start, block_end, blocksize; |
2185 | unsigned to; | |
2186 | struct buffer_head *bh, *head; | |
2187 | int ret = 1; | |
2188 | ||
2189 | if (!page_has_buffers(page)) | |
2190 | return 0; | |
2191 | ||
45bce8f3 LT |
2192 | head = page_buffers(page); |
2193 | blocksize = head->b_size; | |
c186afb4 | 2194 | to = min_t(unsigned, PAGE_CACHE_SIZE - from, count); |
8ab22b9a HH |
2195 | to = from + to; |
2196 | if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize) | |
2197 | return 0; | |
2198 | ||
8ab22b9a HH |
2199 | bh = head; |
2200 | block_start = 0; | |
2201 | do { | |
2202 | block_end = block_start + blocksize; | |
2203 | if (block_end > from && block_start < to) { | |
2204 | if (!buffer_uptodate(bh)) { | |
2205 | ret = 0; | |
2206 | break; | |
2207 | } | |
2208 | if (block_end >= to) | |
2209 | break; | |
2210 | } | |
2211 | block_start = block_end; | |
2212 | bh = bh->b_this_page; | |
2213 | } while (bh != head); | |
2214 | ||
2215 | return ret; | |
2216 | } | |
2217 | EXPORT_SYMBOL(block_is_partially_uptodate); | |
2218 | ||
1da177e4 LT |
2219 | /* |
2220 | * Generic "read page" function for block devices that have the normal | |
2221 | * get_block functionality. This is most of the block device filesystems. | |
2222 | * Reads the page asynchronously --- the unlock_buffer() and | |
2223 | * set/clear_buffer_uptodate() functions propagate buffer state into the | |
2224 | * page struct once IO has completed. | |
2225 | */ | |
2226 | int block_read_full_page(struct page *page, get_block_t *get_block) | |
2227 | { | |
2228 | struct inode *inode = page->mapping->host; | |
2229 | sector_t iblock, lblock; | |
2230 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; | |
45bce8f3 | 2231 | unsigned int blocksize, bbits; |
1da177e4 LT |
2232 | int nr, i; |
2233 | int fully_mapped = 1; | |
2234 | ||
45bce8f3 LT |
2235 | head = create_page_buffers(page, inode, 0); |
2236 | blocksize = head->b_size; | |
2237 | bbits = block_size_bits(blocksize); | |
1da177e4 | 2238 | |
45bce8f3 LT |
2239 | iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits); |
2240 | lblock = (i_size_read(inode)+blocksize-1) >> bbits; | |
1da177e4 LT |
2241 | bh = head; |
2242 | nr = 0; | |
2243 | i = 0; | |
2244 | ||
2245 | do { | |
2246 | if (buffer_uptodate(bh)) | |
2247 | continue; | |
2248 | ||
2249 | if (!buffer_mapped(bh)) { | |
c64610ba AM |
2250 | int err = 0; |
2251 | ||
1da177e4 LT |
2252 | fully_mapped = 0; |
2253 | if (iblock < lblock) { | |
b0cf2321 | 2254 | WARN_ON(bh->b_size != blocksize); |
c64610ba AM |
2255 | err = get_block(inode, iblock, bh, 0); |
2256 | if (err) | |
1da177e4 LT |
2257 | SetPageError(page); |
2258 | } | |
2259 | if (!buffer_mapped(bh)) { | |
eebd2aa3 | 2260 | zero_user(page, i * blocksize, blocksize); |
c64610ba AM |
2261 | if (!err) |
2262 | set_buffer_uptodate(bh); | |
1da177e4 LT |
2263 | continue; |
2264 | } | |
2265 | /* | |
2266 | * get_block() might have updated the buffer | |
2267 | * synchronously | |
2268 | */ | |
2269 | if (buffer_uptodate(bh)) | |
2270 | continue; | |
2271 | } | |
2272 | arr[nr++] = bh; | |
2273 | } while (i++, iblock++, (bh = bh->b_this_page) != head); | |
2274 | ||
2275 | if (fully_mapped) | |
2276 | SetPageMappedToDisk(page); | |
2277 | ||
2278 | if (!nr) { | |
2279 | /* | |
2280 | * All buffers are uptodate - we can set the page uptodate | |
2281 | * as well. But not if get_block() returned an error. | |
2282 | */ | |
2283 | if (!PageError(page)) | |
2284 | SetPageUptodate(page); | |
2285 | unlock_page(page); | |
2286 | return 0; | |
2287 | } | |
2288 | ||
2289 | /* Stage two: lock the buffers */ | |
2290 | for (i = 0; i < nr; i++) { | |
2291 | bh = arr[i]; | |
2292 | lock_buffer(bh); | |
2293 | mark_buffer_async_read(bh); | |
2294 | } | |
2295 | ||
2296 | /* | |
2297 | * Stage 3: start the IO. Check for uptodateness | |
2298 | * inside the buffer lock in case another process reading | |
2299 | * the underlying blockdev brought it uptodate (the sct fix). | |
2300 | */ | |
2301 | for (i = 0; i < nr; i++) { | |
2302 | bh = arr[i]; | |
2303 | if (buffer_uptodate(bh)) | |
2304 | end_buffer_async_read(bh, 1); | |
2305 | else | |
2306 | submit_bh(READ, bh); | |
2307 | } | |
2308 | return 0; | |
2309 | } | |
1fe72eaa | 2310 | EXPORT_SYMBOL(block_read_full_page); |
1da177e4 LT |
2311 | |
2312 | /* utility function for filesystems that need to do work on expanding | |
89e10787 | 2313 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
1da177e4 LT |
2314 | * deal with the hole. |
2315 | */ | |
89e10787 | 2316 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
1da177e4 LT |
2317 | { |
2318 | struct address_space *mapping = inode->i_mapping; | |
2319 | struct page *page; | |
89e10787 | 2320 | void *fsdata; |
1da177e4 LT |
2321 | int err; |
2322 | ||
c08d3b0e | 2323 | err = inode_newsize_ok(inode, size); |
2324 | if (err) | |
1da177e4 LT |
2325 | goto out; |
2326 | ||
89e10787 NP |
2327 | err = pagecache_write_begin(NULL, mapping, size, 0, |
2328 | AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND, | |
2329 | &page, &fsdata); | |
2330 | if (err) | |
05eb0b51 | 2331 | goto out; |
05eb0b51 | 2332 | |
89e10787 NP |
2333 | err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata); |
2334 | BUG_ON(err > 0); | |
05eb0b51 | 2335 | |
1da177e4 LT |
2336 | out: |
2337 | return err; | |
2338 | } | |
1fe72eaa | 2339 | EXPORT_SYMBOL(generic_cont_expand_simple); |
1da177e4 | 2340 | |
f1e3af72 AB |
2341 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
2342 | loff_t pos, loff_t *bytes) | |
1da177e4 | 2343 | { |
1da177e4 | 2344 | struct inode *inode = mapping->host; |
04447026 | 2345 | unsigned int blocksize = i_blocksize(inode); |
89e10787 NP |
2346 | struct page *page; |
2347 | void *fsdata; | |
2348 | pgoff_t index, curidx; | |
2349 | loff_t curpos; | |
2350 | unsigned zerofrom, offset, len; | |
2351 | int err = 0; | |
1da177e4 | 2352 | |
89e10787 NP |
2353 | index = pos >> PAGE_CACHE_SHIFT; |
2354 | offset = pos & ~PAGE_CACHE_MASK; | |
2355 | ||
2356 | while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) { | |
2357 | zerofrom = curpos & ~PAGE_CACHE_MASK; | |
1da177e4 LT |
2358 | if (zerofrom & (blocksize-1)) { |
2359 | *bytes |= (blocksize-1); | |
2360 | (*bytes)++; | |
2361 | } | |
89e10787 | 2362 | len = PAGE_CACHE_SIZE - zerofrom; |
1da177e4 | 2363 | |
89e10787 NP |
2364 | err = pagecache_write_begin(file, mapping, curpos, len, |
2365 | AOP_FLAG_UNINTERRUPTIBLE, | |
2366 | &page, &fsdata); | |
2367 | if (err) | |
2368 | goto out; | |
eebd2aa3 | 2369 | zero_user(page, zerofrom, len); |
89e10787 NP |
2370 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2371 | page, fsdata); | |
2372 | if (err < 0) | |
2373 | goto out; | |
2374 | BUG_ON(err != len); | |
2375 | err = 0; | |
061e9746 OH |
2376 | |
2377 | balance_dirty_pages_ratelimited(mapping); | |
c2ca0fcd MP |
2378 | |
2379 | if (unlikely(fatal_signal_pending(current))) { | |
2380 | err = -EINTR; | |
2381 | goto out; | |
2382 | } | |
89e10787 | 2383 | } |
1da177e4 | 2384 | |
89e10787 NP |
2385 | /* page covers the boundary, find the boundary offset */ |
2386 | if (index == curidx) { | |
2387 | zerofrom = curpos & ~PAGE_CACHE_MASK; | |
1da177e4 | 2388 | /* if we will expand the thing last block will be filled */ |
89e10787 NP |
2389 | if (offset <= zerofrom) { |
2390 | goto out; | |
2391 | } | |
2392 | if (zerofrom & (blocksize-1)) { | |
1da177e4 LT |
2393 | *bytes |= (blocksize-1); |
2394 | (*bytes)++; | |
2395 | } | |
89e10787 | 2396 | len = offset - zerofrom; |
1da177e4 | 2397 | |
89e10787 NP |
2398 | err = pagecache_write_begin(file, mapping, curpos, len, |
2399 | AOP_FLAG_UNINTERRUPTIBLE, | |
2400 | &page, &fsdata); | |
2401 | if (err) | |
2402 | goto out; | |
eebd2aa3 | 2403 | zero_user(page, zerofrom, len); |
89e10787 NP |
2404 | err = pagecache_write_end(file, mapping, curpos, len, len, |
2405 | page, fsdata); | |
2406 | if (err < 0) | |
2407 | goto out; | |
2408 | BUG_ON(err != len); | |
2409 | err = 0; | |
1da177e4 | 2410 | } |
89e10787 NP |
2411 | out: |
2412 | return err; | |
2413 | } | |
2414 | ||
2415 | /* | |
2416 | * For moronic filesystems that do not allow holes in file. | |
2417 | * We may have to extend the file. | |
2418 | */ | |
282dc178 | 2419 | int cont_write_begin(struct file *file, struct address_space *mapping, |
89e10787 NP |
2420 | loff_t pos, unsigned len, unsigned flags, |
2421 | struct page **pagep, void **fsdata, | |
2422 | get_block_t *get_block, loff_t *bytes) | |
2423 | { | |
2424 | struct inode *inode = mapping->host; | |
04447026 FF |
2425 | unsigned int blocksize = i_blocksize(inode); |
2426 | unsigned int zerofrom; | |
89e10787 NP |
2427 | int err; |
2428 | ||
2429 | err = cont_expand_zero(file, mapping, pos, bytes); | |
2430 | if (err) | |
155130a4 | 2431 | return err; |
89e10787 NP |
2432 | |
2433 | zerofrom = *bytes & ~PAGE_CACHE_MASK; | |
2434 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { | |
2435 | *bytes |= (blocksize-1); | |
2436 | (*bytes)++; | |
1da177e4 | 2437 | } |
1da177e4 | 2438 | |
155130a4 | 2439 | return block_write_begin(mapping, pos, len, flags, pagep, get_block); |
1da177e4 | 2440 | } |
1fe72eaa | 2441 | EXPORT_SYMBOL(cont_write_begin); |
1da177e4 | 2442 | |
1da177e4 LT |
2443 | int block_commit_write(struct page *page, unsigned from, unsigned to) |
2444 | { | |
2445 | struct inode *inode = page->mapping->host; | |
2446 | __block_commit_write(inode,page,from,to); | |
2447 | return 0; | |
2448 | } | |
1fe72eaa | 2449 | EXPORT_SYMBOL(block_commit_write); |
1da177e4 | 2450 | |
54171690 DC |
2451 | /* |
2452 | * block_page_mkwrite() is not allowed to change the file size as it gets | |
2453 | * called from a page fault handler when a page is first dirtied. Hence we must | |
2454 | * be careful to check for EOF conditions here. We set the page up correctly | |
2455 | * for a written page which means we get ENOSPC checking when writing into | |
2456 | * holes and correct delalloc and unwritten extent mapping on filesystems that | |
2457 | * support these features. | |
2458 | * | |
2459 | * We are not allowed to take the i_mutex here so we have to play games to | |
2460 | * protect against truncate races as the page could now be beyond EOF. Because | |
7bb46a67 | 2461 | * truncate writes the inode size before removing pages, once we have the |
54171690 DC |
2462 | * page lock we can determine safely if the page is beyond EOF. If it is not |
2463 | * beyond EOF, then the page is guaranteed safe against truncation until we | |
2464 | * unlock the page. | |
ea13a864 | 2465 | * |
14da9200 | 2466 | * Direct callers of this function should protect against filesystem freezing |
5c500029 | 2467 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
54171690 | 2468 | */ |
5c500029 | 2469 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
24da4fab | 2470 | get_block_t get_block) |
54171690 | 2471 | { |
c2ec175c | 2472 | struct page *page = vmf->page; |
496ad9aa | 2473 | struct inode *inode = file_inode(vma->vm_file); |
54171690 DC |
2474 | unsigned long end; |
2475 | loff_t size; | |
24da4fab | 2476 | int ret; |
54171690 DC |
2477 | |
2478 | lock_page(page); | |
2479 | size = i_size_read(inode); | |
2480 | if ((page->mapping != inode->i_mapping) || | |
18336338 | 2481 | (page_offset(page) > size)) { |
24da4fab JK |
2482 | /* We overload EFAULT to mean page got truncated */ |
2483 | ret = -EFAULT; | |
2484 | goto out_unlock; | |
54171690 DC |
2485 | } |
2486 | ||
2487 | /* page is wholly or partially inside EOF */ | |
2488 | if (((page->index + 1) << PAGE_CACHE_SHIFT) > size) | |
2489 | end = size & ~PAGE_CACHE_MASK; | |
2490 | else | |
2491 | end = PAGE_CACHE_SIZE; | |
2492 | ||
ebdec241 | 2493 | ret = __block_write_begin(page, 0, end, get_block); |
54171690 DC |
2494 | if (!ret) |
2495 | ret = block_commit_write(page, 0, end); | |
2496 | ||
24da4fab JK |
2497 | if (unlikely(ret < 0)) |
2498 | goto out_unlock; | |
ea13a864 | 2499 | set_page_dirty(page); |
1d1d1a76 | 2500 | wait_for_stable_page(page); |
24da4fab JK |
2501 | return 0; |
2502 | out_unlock: | |
2503 | unlock_page(page); | |
54171690 | 2504 | return ret; |
24da4fab | 2505 | } |
1fe72eaa | 2506 | EXPORT_SYMBOL(block_page_mkwrite); |
1da177e4 LT |
2507 | |
2508 | /* | |
03158cd7 | 2509 | * nobh_write_begin()'s prereads are special: the buffer_heads are freed |
1da177e4 LT |
2510 | * immediately, while under the page lock. So it needs a special end_io |
2511 | * handler which does not touch the bh after unlocking it. | |
1da177e4 LT |
2512 | */ |
2513 | static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate) | |
2514 | { | |
68671f35 | 2515 | __end_buffer_read_notouch(bh, uptodate); |
1da177e4 LT |
2516 | } |
2517 | ||
03158cd7 NP |
2518 | /* |
2519 | * Attach the singly-linked list of buffers created by nobh_write_begin, to | |
2520 | * the page (converting it to circular linked list and taking care of page | |
2521 | * dirty races). | |
2522 | */ | |
2523 | static void attach_nobh_buffers(struct page *page, struct buffer_head *head) | |
2524 | { | |
2525 | struct buffer_head *bh; | |
2526 | ||
2527 | BUG_ON(!PageLocked(page)); | |
2528 | ||
2529 | spin_lock(&page->mapping->private_lock); | |
2530 | bh = head; | |
2531 | do { | |
2532 | if (PageDirty(page)) | |
2533 | set_buffer_dirty(bh); | |
2534 | if (!bh->b_this_page) | |
2535 | bh->b_this_page = head; | |
2536 | bh = bh->b_this_page; | |
2537 | } while (bh != head); | |
2538 | attach_page_buffers(page, head); | |
2539 | spin_unlock(&page->mapping->private_lock); | |
2540 | } | |
2541 | ||
1da177e4 | 2542 | /* |
ea0f04e5 CH |
2543 | * On entry, the page is fully not uptodate. |
2544 | * On exit the page is fully uptodate in the areas outside (from,to) | |
7bb46a67 | 2545 | * The filesystem needs to handle block truncation upon failure. |
1da177e4 | 2546 | */ |
ea0f04e5 | 2547 | int nobh_write_begin(struct address_space *mapping, |
03158cd7 NP |
2548 | loff_t pos, unsigned len, unsigned flags, |
2549 | struct page **pagep, void **fsdata, | |
1da177e4 LT |
2550 | get_block_t *get_block) |
2551 | { | |
03158cd7 | 2552 | struct inode *inode = mapping->host; |
1da177e4 LT |
2553 | const unsigned blkbits = inode->i_blkbits; |
2554 | const unsigned blocksize = 1 << blkbits; | |
a4b0672d | 2555 | struct buffer_head *head, *bh; |
03158cd7 NP |
2556 | struct page *page; |
2557 | pgoff_t index; | |
2558 | unsigned from, to; | |
1da177e4 | 2559 | unsigned block_in_page; |
a4b0672d | 2560 | unsigned block_start, block_end; |
1da177e4 | 2561 | sector_t block_in_file; |
1da177e4 | 2562 | int nr_reads = 0; |
1da177e4 LT |
2563 | int ret = 0; |
2564 | int is_mapped_to_disk = 1; | |
1da177e4 | 2565 | |
03158cd7 NP |
2566 | index = pos >> PAGE_CACHE_SHIFT; |
2567 | from = pos & (PAGE_CACHE_SIZE - 1); | |
2568 | to = from + len; | |
2569 | ||
54566b2c | 2570 | page = grab_cache_page_write_begin(mapping, index, flags); |
03158cd7 NP |
2571 | if (!page) |
2572 | return -ENOMEM; | |
2573 | *pagep = page; | |
2574 | *fsdata = NULL; | |
2575 | ||
2576 | if (page_has_buffers(page)) { | |
309f77ad NK |
2577 | ret = __block_write_begin(page, pos, len, get_block); |
2578 | if (unlikely(ret)) | |
2579 | goto out_release; | |
2580 | return ret; | |
03158cd7 | 2581 | } |
a4b0672d | 2582 | |
1da177e4 LT |
2583 | if (PageMappedToDisk(page)) |
2584 | return 0; | |
2585 | ||
a4b0672d NP |
2586 | /* |
2587 | * Allocate buffers so that we can keep track of state, and potentially | |
2588 | * attach them to the page if an error occurs. In the common case of | |
2589 | * no error, they will just be freed again without ever being attached | |
2590 | * to the page (which is all OK, because we're under the page lock). | |
2591 | * | |
2592 | * Be careful: the buffer linked list is a NULL terminated one, rather | |
2593 | * than the circular one we're used to. | |
2594 | */ | |
2595 | head = alloc_page_buffers(page, blocksize, 0); | |
03158cd7 NP |
2596 | if (!head) { |
2597 | ret = -ENOMEM; | |
2598 | goto out_release; | |
2599 | } | |
a4b0672d | 2600 | |
1da177e4 | 2601 | block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits); |
1da177e4 LT |
2602 | |
2603 | /* | |
2604 | * We loop across all blocks in the page, whether or not they are | |
2605 | * part of the affected region. This is so we can discover if the | |
2606 | * page is fully mapped-to-disk. | |
2607 | */ | |
a4b0672d | 2608 | for (block_start = 0, block_in_page = 0, bh = head; |
1da177e4 | 2609 | block_start < PAGE_CACHE_SIZE; |
a4b0672d | 2610 | block_in_page++, block_start += blocksize, bh = bh->b_this_page) { |
1da177e4 LT |
2611 | int create; |
2612 | ||
a4b0672d NP |
2613 | block_end = block_start + blocksize; |
2614 | bh->b_state = 0; | |
1da177e4 LT |
2615 | create = 1; |
2616 | if (block_start >= to) | |
2617 | create = 0; | |
2618 | ret = get_block(inode, block_in_file + block_in_page, | |
a4b0672d | 2619 | bh, create); |
1da177e4 LT |
2620 | if (ret) |
2621 | goto failed; | |
a4b0672d | 2622 | if (!buffer_mapped(bh)) |
1da177e4 | 2623 | is_mapped_to_disk = 0; |
a4b0672d NP |
2624 | if (buffer_new(bh)) |
2625 | unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr); | |
2626 | if (PageUptodate(page)) { | |
2627 | set_buffer_uptodate(bh); | |
1da177e4 | 2628 | continue; |
a4b0672d NP |
2629 | } |
2630 | if (buffer_new(bh) || !buffer_mapped(bh)) { | |
eebd2aa3 CL |
2631 | zero_user_segments(page, block_start, from, |
2632 | to, block_end); | |
1da177e4 LT |
2633 | continue; |
2634 | } | |
a4b0672d | 2635 | if (buffer_uptodate(bh)) |
1da177e4 LT |
2636 | continue; /* reiserfs does this */ |
2637 | if (block_start < from || block_end > to) { | |
a4b0672d NP |
2638 | lock_buffer(bh); |
2639 | bh->b_end_io = end_buffer_read_nobh; | |
2640 | submit_bh(READ, bh); | |
2641 | nr_reads++; | |
1da177e4 LT |
2642 | } |
2643 | } | |
2644 | ||
2645 | if (nr_reads) { | |
1da177e4 LT |
2646 | /* |
2647 | * The page is locked, so these buffers are protected from | |
2648 | * any VM or truncate activity. Hence we don't need to care | |
2649 | * for the buffer_head refcounts. | |
2650 | */ | |
a4b0672d | 2651 | for (bh = head; bh; bh = bh->b_this_page) { |
1da177e4 LT |
2652 | wait_on_buffer(bh); |
2653 | if (!buffer_uptodate(bh)) | |
2654 | ret = -EIO; | |
1da177e4 LT |
2655 | } |
2656 | if (ret) | |
2657 | goto failed; | |
2658 | } | |
2659 | ||
2660 | if (is_mapped_to_disk) | |
2661 | SetPageMappedToDisk(page); | |
1da177e4 | 2662 | |
03158cd7 | 2663 | *fsdata = head; /* to be released by nobh_write_end */ |
a4b0672d | 2664 | |
1da177e4 LT |
2665 | return 0; |
2666 | ||
2667 | failed: | |
03158cd7 | 2668 | BUG_ON(!ret); |
1da177e4 | 2669 | /* |
a4b0672d NP |
2670 | * Error recovery is a bit difficult. We need to zero out blocks that |
2671 | * were newly allocated, and dirty them to ensure they get written out. | |
2672 | * Buffers need to be attached to the page at this point, otherwise | |
2673 | * the handling of potential IO errors during writeout would be hard | |
2674 | * (could try doing synchronous writeout, but what if that fails too?) | |
1da177e4 | 2675 | */ |
03158cd7 NP |
2676 | attach_nobh_buffers(page, head); |
2677 | page_zero_new_buffers(page, from, to); | |
a4b0672d | 2678 | |
03158cd7 NP |
2679 | out_release: |
2680 | unlock_page(page); | |
2681 | page_cache_release(page); | |
2682 | *pagep = NULL; | |
a4b0672d | 2683 | |
7bb46a67 | 2684 | return ret; |
2685 | } | |
03158cd7 | 2686 | EXPORT_SYMBOL(nobh_write_begin); |
1da177e4 | 2687 | |
03158cd7 NP |
2688 | int nobh_write_end(struct file *file, struct address_space *mapping, |
2689 | loff_t pos, unsigned len, unsigned copied, | |
2690 | struct page *page, void *fsdata) | |
1da177e4 LT |
2691 | { |
2692 | struct inode *inode = page->mapping->host; | |
efdc3131 | 2693 | struct buffer_head *head = fsdata; |
03158cd7 | 2694 | struct buffer_head *bh; |
5b41e74a | 2695 | BUG_ON(fsdata != NULL && page_has_buffers(page)); |
1da177e4 | 2696 | |
d4cf109f | 2697 | if (unlikely(copied < len) && head) |
5b41e74a DM |
2698 | attach_nobh_buffers(page, head); |
2699 | if (page_has_buffers(page)) | |
2700 | return generic_write_end(file, mapping, pos, len, | |
2701 | copied, page, fsdata); | |
a4b0672d | 2702 | |
22c8ca78 | 2703 | SetPageUptodate(page); |
1da177e4 | 2704 | set_page_dirty(page); |
03158cd7 NP |
2705 | if (pos+copied > inode->i_size) { |
2706 | i_size_write(inode, pos+copied); | |
1da177e4 LT |
2707 | mark_inode_dirty(inode); |
2708 | } | |
03158cd7 NP |
2709 | |
2710 | unlock_page(page); | |
2711 | page_cache_release(page); | |
2712 | ||
03158cd7 NP |
2713 | while (head) { |
2714 | bh = head; | |
2715 | head = head->b_this_page; | |
2716 | free_buffer_head(bh); | |
2717 | } | |
2718 | ||
2719 | return copied; | |
1da177e4 | 2720 | } |
03158cd7 | 2721 | EXPORT_SYMBOL(nobh_write_end); |
1da177e4 LT |
2722 | |
2723 | /* | |
2724 | * nobh_writepage() - based on block_full_write_page() except | |
2725 | * that it tries to operate without attaching bufferheads to | |
2726 | * the page. | |
2727 | */ | |
2728 | int nobh_writepage(struct page *page, get_block_t *get_block, | |
2729 | struct writeback_control *wbc) | |
2730 | { | |
2731 | struct inode * const inode = page->mapping->host; | |
2732 | loff_t i_size = i_size_read(inode); | |
2733 | const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
2734 | unsigned offset; | |
1da177e4 LT |
2735 | int ret; |
2736 | ||
2737 | /* Is the page fully inside i_size? */ | |
2738 | if (page->index < end_index) | |
2739 | goto out; | |
2740 | ||
2741 | /* Is the page fully outside i_size? (truncate in progress) */ | |
2742 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
2743 | if (page->index >= end_index+1 || !offset) { | |
2744 | /* | |
2745 | * The page may have dirty, unmapped buffers. For example, | |
2746 | * they may have been added in ext3_writepage(). Make them | |
2747 | * freeable here, so the page does not leak. | |
2748 | */ | |
2749 | #if 0 | |
2750 | /* Not really sure about this - do we need this ? */ | |
2751 | if (page->mapping->a_ops->invalidatepage) | |
2752 | page->mapping->a_ops->invalidatepage(page, offset); | |
2753 | #endif | |
2754 | unlock_page(page); | |
2755 | return 0; /* don't care */ | |
2756 | } | |
2757 | ||
2758 | /* | |
2759 | * The page straddles i_size. It must be zeroed out on each and every | |
2760 | * writepage invocation because it may be mmapped. "A file is mapped | |
2761 | * in multiples of the page size. For a file that is not a multiple of | |
2762 | * the page size, the remaining memory is zeroed when mapped, and | |
2763 | * writes to that region are not written out to the file." | |
2764 | */ | |
eebd2aa3 | 2765 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); |
1da177e4 LT |
2766 | out: |
2767 | ret = mpage_writepage(page, get_block, wbc); | |
2768 | if (ret == -EAGAIN) | |
35c80d5f CM |
2769 | ret = __block_write_full_page(inode, page, get_block, wbc, |
2770 | end_buffer_async_write); | |
1da177e4 LT |
2771 | return ret; |
2772 | } | |
2773 | EXPORT_SYMBOL(nobh_writepage); | |
2774 | ||
03158cd7 NP |
2775 | int nobh_truncate_page(struct address_space *mapping, |
2776 | loff_t from, get_block_t *get_block) | |
1da177e4 | 2777 | { |
1da177e4 LT |
2778 | pgoff_t index = from >> PAGE_CACHE_SHIFT; |
2779 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
03158cd7 NP |
2780 | unsigned blocksize; |
2781 | sector_t iblock; | |
2782 | unsigned length, pos; | |
2783 | struct inode *inode = mapping->host; | |
1da177e4 | 2784 | struct page *page; |
03158cd7 NP |
2785 | struct buffer_head map_bh; |
2786 | int err; | |
1da177e4 | 2787 | |
04447026 | 2788 | blocksize = i_blocksize(inode); |
03158cd7 NP |
2789 | length = offset & (blocksize - 1); |
2790 | ||
2791 | /* Block boundary? Nothing to do */ | |
2792 | if (!length) | |
2793 | return 0; | |
2794 | ||
2795 | length = blocksize - length; | |
2796 | iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); | |
1da177e4 | 2797 | |
1da177e4 | 2798 | page = grab_cache_page(mapping, index); |
03158cd7 | 2799 | err = -ENOMEM; |
1da177e4 LT |
2800 | if (!page) |
2801 | goto out; | |
2802 | ||
03158cd7 NP |
2803 | if (page_has_buffers(page)) { |
2804 | has_buffers: | |
2805 | unlock_page(page); | |
2806 | page_cache_release(page); | |
2807 | return block_truncate_page(mapping, from, get_block); | |
2808 | } | |
2809 | ||
2810 | /* Find the buffer that contains "offset" */ | |
2811 | pos = blocksize; | |
2812 | while (offset >= pos) { | |
2813 | iblock++; | |
2814 | pos += blocksize; | |
2815 | } | |
2816 | ||
460bcf57 TT |
2817 | map_bh.b_size = blocksize; |
2818 | map_bh.b_state = 0; | |
03158cd7 NP |
2819 | err = get_block(inode, iblock, &map_bh, 0); |
2820 | if (err) | |
2821 | goto unlock; | |
2822 | /* unmapped? It's a hole - nothing to do */ | |
2823 | if (!buffer_mapped(&map_bh)) | |
2824 | goto unlock; | |
2825 | ||
2826 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2827 | if (!PageUptodate(page)) { | |
2828 | err = mapping->a_ops->readpage(NULL, page); | |
2829 | if (err) { | |
2830 | page_cache_release(page); | |
2831 | goto out; | |
2832 | } | |
2833 | lock_page(page); | |
2834 | if (!PageUptodate(page)) { | |
2835 | err = -EIO; | |
2836 | goto unlock; | |
2837 | } | |
2838 | if (page_has_buffers(page)) | |
2839 | goto has_buffers; | |
1da177e4 | 2840 | } |
eebd2aa3 | 2841 | zero_user(page, offset, length); |
03158cd7 NP |
2842 | set_page_dirty(page); |
2843 | err = 0; | |
2844 | ||
2845 | unlock: | |
1da177e4 LT |
2846 | unlock_page(page); |
2847 | page_cache_release(page); | |
2848 | out: | |
03158cd7 | 2849 | return err; |
1da177e4 LT |
2850 | } |
2851 | EXPORT_SYMBOL(nobh_truncate_page); | |
2852 | ||
2853 | int block_truncate_page(struct address_space *mapping, | |
2854 | loff_t from, get_block_t *get_block) | |
2855 | { | |
2856 | pgoff_t index = from >> PAGE_CACHE_SHIFT; | |
2857 | unsigned offset = from & (PAGE_CACHE_SIZE-1); | |
2858 | unsigned blocksize; | |
54b21a79 | 2859 | sector_t iblock; |
1da177e4 LT |
2860 | unsigned length, pos; |
2861 | struct inode *inode = mapping->host; | |
2862 | struct page *page; | |
2863 | struct buffer_head *bh; | |
1da177e4 LT |
2864 | int err; |
2865 | ||
04447026 | 2866 | blocksize = i_blocksize(inode); |
1da177e4 LT |
2867 | length = offset & (blocksize - 1); |
2868 | ||
2869 | /* Block boundary? Nothing to do */ | |
2870 | if (!length) | |
2871 | return 0; | |
2872 | ||
2873 | length = blocksize - length; | |
54b21a79 | 2874 | iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits); |
1da177e4 LT |
2875 | |
2876 | page = grab_cache_page(mapping, index); | |
2877 | err = -ENOMEM; | |
2878 | if (!page) | |
2879 | goto out; | |
2880 | ||
2881 | if (!page_has_buffers(page)) | |
2882 | create_empty_buffers(page, blocksize, 0); | |
2883 | ||
2884 | /* Find the buffer that contains "offset" */ | |
2885 | bh = page_buffers(page); | |
2886 | pos = blocksize; | |
2887 | while (offset >= pos) { | |
2888 | bh = bh->b_this_page; | |
2889 | iblock++; | |
2890 | pos += blocksize; | |
2891 | } | |
2892 | ||
2893 | err = 0; | |
2894 | if (!buffer_mapped(bh)) { | |
b0cf2321 | 2895 | WARN_ON(bh->b_size != blocksize); |
1da177e4 LT |
2896 | err = get_block(inode, iblock, bh, 0); |
2897 | if (err) | |
2898 | goto unlock; | |
2899 | /* unmapped? It's a hole - nothing to do */ | |
2900 | if (!buffer_mapped(bh)) | |
2901 | goto unlock; | |
2902 | } | |
2903 | ||
2904 | /* Ok, it's mapped. Make sure it's up-to-date */ | |
2905 | if (PageUptodate(page)) | |
2906 | set_buffer_uptodate(bh); | |
2907 | ||
33a266dd | 2908 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
1da177e4 LT |
2909 | err = -EIO; |
2910 | ll_rw_block(READ, 1, &bh); | |
2911 | wait_on_buffer(bh); | |
2912 | /* Uhhuh. Read error. Complain and punt. */ | |
2913 | if (!buffer_uptodate(bh)) | |
2914 | goto unlock; | |
2915 | } | |
2916 | ||
eebd2aa3 | 2917 | zero_user(page, offset, length); |
1da177e4 LT |
2918 | mark_buffer_dirty(bh); |
2919 | err = 0; | |
2920 | ||
2921 | unlock: | |
2922 | unlock_page(page); | |
2923 | page_cache_release(page); | |
2924 | out: | |
2925 | return err; | |
2926 | } | |
1fe72eaa | 2927 | EXPORT_SYMBOL(block_truncate_page); |
1da177e4 LT |
2928 | |
2929 | /* | |
2930 | * The generic ->writepage function for buffer-backed address_spaces | |
2931 | */ | |
1b938c08 MW |
2932 | int block_write_full_page(struct page *page, get_block_t *get_block, |
2933 | struct writeback_control *wbc) | |
1da177e4 LT |
2934 | { |
2935 | struct inode * const inode = page->mapping->host; | |
2936 | loff_t i_size = i_size_read(inode); | |
2937 | const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT; | |
2938 | unsigned offset; | |
1da177e4 LT |
2939 | |
2940 | /* Is the page fully inside i_size? */ | |
2941 | if (page->index < end_index) | |
35c80d5f | 2942 | return __block_write_full_page(inode, page, get_block, wbc, |
1b938c08 | 2943 | end_buffer_async_write); |
1da177e4 LT |
2944 | |
2945 | /* Is the page fully outside i_size? (truncate in progress) */ | |
2946 | offset = i_size & (PAGE_CACHE_SIZE-1); | |
2947 | if (page->index >= end_index+1 || !offset) { | |
2948 | /* | |
2949 | * The page may have dirty, unmapped buffers. For example, | |
2950 | * they may have been added in ext3_writepage(). Make them | |
2951 | * freeable here, so the page does not leak. | |
2952 | */ | |
d47992f8 | 2953 | do_invalidatepage(page, 0, PAGE_CACHE_SIZE); |
1da177e4 LT |
2954 | unlock_page(page); |
2955 | return 0; /* don't care */ | |
2956 | } | |
2957 | ||
2958 | /* | |
2959 | * The page straddles i_size. It must be zeroed out on each and every | |
2a61aa40 | 2960 | * writepage invocation because it may be mmapped. "A file is mapped |
1da177e4 LT |
2961 | * in multiples of the page size. For a file that is not a multiple of |
2962 | * the page size, the remaining memory is zeroed when mapped, and | |
2963 | * writes to that region are not written out to the file." | |
2964 | */ | |
eebd2aa3 | 2965 | zero_user_segment(page, offset, PAGE_CACHE_SIZE); |
1b938c08 MW |
2966 | return __block_write_full_page(inode, page, get_block, wbc, |
2967 | end_buffer_async_write); | |
35c80d5f | 2968 | } |
1fe72eaa | 2969 | EXPORT_SYMBOL(block_write_full_page); |
35c80d5f | 2970 | |
1da177e4 LT |
2971 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
2972 | get_block_t *get_block) | |
2973 | { | |
2974 | struct buffer_head tmp; | |
2975 | struct inode *inode = mapping->host; | |
2976 | tmp.b_state = 0; | |
2977 | tmp.b_blocknr = 0; | |
04447026 | 2978 | tmp.b_size = i_blocksize(inode); |
1da177e4 LT |
2979 | get_block(inode, block, &tmp, 0); |
2980 | return tmp.b_blocknr; | |
2981 | } | |
1fe72eaa | 2982 | EXPORT_SYMBOL(generic_block_bmap); |
1da177e4 | 2983 | |
4246a0b6 | 2984 | static void end_bio_bh_io_sync(struct bio *bio) |
1da177e4 LT |
2985 | { |
2986 | struct buffer_head *bh = bio->bi_private; | |
2987 | ||
b7c44ed9 | 2988 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
08bafc03 KM |
2989 | set_bit(BH_Quiet, &bh->b_state); |
2990 | ||
4246a0b6 | 2991 | bh->b_end_io(bh, !bio->bi_error); |
1da177e4 | 2992 | bio_put(bio); |
1da177e4 LT |
2993 | } |
2994 | ||
57302e0d LT |
2995 | /* |
2996 | * This allows us to do IO even on the odd last sectors | |
59d43914 | 2997 | * of a device, even if the block size is some multiple |
57302e0d LT |
2998 | * of the physical sector size. |
2999 | * | |
3000 | * We'll just truncate the bio to the size of the device, | |
3001 | * and clear the end of the buffer head manually. | |
3002 | * | |
3003 | * Truly out-of-range accesses will turn into actual IO | |
3004 | * errors, this only handles the "we need to be able to | |
3005 | * do IO at the final sector" case. | |
3006 | */ | |
4db96b71 | 3007 | void guard_bio_eod(int rw, struct bio *bio) |
57302e0d LT |
3008 | { |
3009 | sector_t maxsector; | |
59d43914 AM |
3010 | struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1]; |
3011 | unsigned truncated_bytes; | |
57302e0d LT |
3012 | |
3013 | maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9; | |
3014 | if (!maxsector) | |
3015 | return; | |
3016 | ||
3017 | /* | |
3018 | * If the *whole* IO is past the end of the device, | |
3019 | * let it through, and the IO layer will turn it into | |
3020 | * an EIO. | |
3021 | */ | |
4f024f37 | 3022 | if (unlikely(bio->bi_iter.bi_sector >= maxsector)) |
57302e0d LT |
3023 | return; |
3024 | ||
4f024f37 | 3025 | maxsector -= bio->bi_iter.bi_sector; |
59d43914 | 3026 | if (likely((bio->bi_iter.bi_size >> 9) <= maxsector)) |
57302e0d LT |
3027 | return; |
3028 | ||
59d43914 AM |
3029 | /* Uhhuh. We've got a bio that straddles the device size! */ |
3030 | truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9); | |
57302e0d LT |
3031 | |
3032 | /* Truncate the bio.. */ | |
59d43914 AM |
3033 | bio->bi_iter.bi_size -= truncated_bytes; |
3034 | bvec->bv_len -= truncated_bytes; | |
57302e0d LT |
3035 | |
3036 | /* ..and clear the end of the buffer for reads */ | |
27d7c2a0 | 3037 | if ((rw & RW_MASK) == READ) { |
59d43914 AM |
3038 | zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len, |
3039 | truncated_bytes); | |
57302e0d LT |
3040 | } |
3041 | } | |
3042 | ||
b16b1deb TH |
3043 | static int submit_bh_wbc(int rw, struct buffer_head *bh, |
3044 | unsigned long bio_flags, struct writeback_control *wbc) | |
1da177e4 LT |
3045 | { |
3046 | struct bio *bio; | |
1da177e4 LT |
3047 | |
3048 | BUG_ON(!buffer_locked(bh)); | |
3049 | BUG_ON(!buffer_mapped(bh)); | |
3050 | BUG_ON(!bh->b_end_io); | |
8fb0e342 AK |
3051 | BUG_ON(buffer_delay(bh)); |
3052 | BUG_ON(buffer_unwritten(bh)); | |
1da177e4 | 3053 | |
1da177e4 | 3054 | /* |
48fd4f93 | 3055 | * Only clear out a write error when rewriting |
1da177e4 | 3056 | */ |
48fd4f93 | 3057 | if (test_set_buffer_req(bh) && (rw & WRITE)) |
1da177e4 LT |
3058 | clear_buffer_write_io_error(bh); |
3059 | ||
3060 | /* | |
3061 | * from here on down, it's all bio -- do the initial mapping, | |
3062 | * submit_bio -> generic_make_request may further map this bio around | |
3063 | */ | |
3064 | bio = bio_alloc(GFP_NOIO, 1); | |
3065 | ||
2a814908 | 3066 | if (wbc) { |
b16b1deb | 3067 | wbc_init_bio(wbc, bio); |
2a814908 TH |
3068 | wbc_account_io(wbc, bh->b_page, bh->b_size); |
3069 | } | |
bafc0dba | 3070 | |
4f024f37 | 3071 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
1da177e4 | 3072 | bio->bi_bdev = bh->b_bdev; |
1da177e4 | 3073 | |
6cf66b4c KO |
3074 | bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
3075 | BUG_ON(bio->bi_iter.bi_size != bh->b_size); | |
1da177e4 LT |
3076 | |
3077 | bio->bi_end_io = end_bio_bh_io_sync; | |
3078 | bio->bi_private = bh; | |
71368511 | 3079 | bio->bi_flags |= bio_flags; |
1da177e4 | 3080 | |
57302e0d | 3081 | /* Take care of bh's that straddle the end of the device */ |
59d43914 | 3082 | guard_bio_eod(rw, bio); |
57302e0d | 3083 | |
877f962c TT |
3084 | if (buffer_meta(bh)) |
3085 | rw |= REQ_META; | |
3086 | if (buffer_prio(bh)) | |
3087 | rw |= REQ_PRIO; | |
3088 | ||
1cac41cb MB |
3089 | if(buffer_sync_flush(bh)) { |
3090 | rw |= REQ_SYNC; | |
3091 | clear_buffer_sync_flush(bh); | |
3092 | } | |
3093 | #ifdef CONFIG_JOURNAL_DATA_TAG | |
3094 | if(buffer_journal(bh)) { | |
3095 | bio->bi_flags |= (1 << BIO_JOURNAL); | |
3096 | clear_buffer_journal(bh); | |
3097 | } | |
3098 | if(buffer_jmeta(bh)) { | |
3099 | //bio->bi_flags |= (1 << BIO_JMETA); | |
3100 | clear_buffer_jmeta(bh); | |
3101 | } | |
3102 | #endif | |
3103 | ||
1da177e4 | 3104 | submit_bio(rw, bio); |
f6454b04 | 3105 | return 0; |
1da177e4 | 3106 | } |
bafc0dba TH |
3107 | |
3108 | int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags) | |
3109 | { | |
b16b1deb | 3110 | return submit_bh_wbc(rw, bh, bio_flags, NULL); |
bafc0dba | 3111 | } |
71368511 DW |
3112 | EXPORT_SYMBOL_GPL(_submit_bh); |
3113 | ||
3114 | int submit_bh(int rw, struct buffer_head *bh) | |
3115 | { | |
b16b1deb | 3116 | return submit_bh_wbc(rw, bh, 0, NULL); |
71368511 | 3117 | } |
1fe72eaa | 3118 | EXPORT_SYMBOL(submit_bh); |
1da177e4 LT |
3119 | |
3120 | /** | |
3121 | * ll_rw_block: low-level access to block devices (DEPRECATED) | |
9cb569d6 | 3122 | * @rw: whether to %READ or %WRITE or maybe %READA (readahead) |
1da177e4 LT |
3123 | * @nr: number of &struct buffer_heads in the array |
3124 | * @bhs: array of pointers to &struct buffer_head | |
3125 | * | |
a7662236 JK |
3126 | * ll_rw_block() takes an array of pointers to &struct buffer_heads, and |
3127 | * requests an I/O operation on them, either a %READ or a %WRITE. The third | |
9cb569d6 CH |
3128 | * %READA option is described in the documentation for generic_make_request() |
3129 | * which ll_rw_block() calls. | |
1da177e4 LT |
3130 | * |
3131 | * This function drops any buffer that it cannot get a lock on (with the | |
9cb569d6 CH |
3132 | * BH_Lock state bit), any buffer that appears to be clean when doing a write |
3133 | * request, and any buffer that appears to be up-to-date when doing read | |
3134 | * request. Further it marks as clean buffers that are processed for | |
3135 | * writing (the buffer cache won't assume that they are actually clean | |
3136 | * until the buffer gets unlocked). | |
1da177e4 LT |
3137 | * |
3138 | * ll_rw_block sets b_end_io to simple completion handler that marks | |
e227867f | 3139 | * the buffer up-to-date (if appropriate), unlocks the buffer and wakes |
1da177e4 LT |
3140 | * any waiters. |
3141 | * | |
3142 | * All of the buffers must be for the same device, and must also be a | |
3143 | * multiple of the current approved size for the device. | |
3144 | */ | |
3145 | void ll_rw_block(int rw, int nr, struct buffer_head *bhs[]) | |
3146 | { | |
3147 | int i; | |
3148 | ||
3149 | for (i = 0; i < nr; i++) { | |
3150 | struct buffer_head *bh = bhs[i]; | |
3151 | ||
9cb569d6 | 3152 | if (!trylock_buffer(bh)) |
1da177e4 | 3153 | continue; |
9cb569d6 | 3154 | if (rw == WRITE) { |
1da177e4 | 3155 | if (test_clear_buffer_dirty(bh)) { |
76c3073a | 3156 | bh->b_end_io = end_buffer_write_sync; |
e60e5c50 | 3157 | get_bh(bh); |
9cb569d6 | 3158 | submit_bh(WRITE, bh); |
1da177e4 LT |
3159 | continue; |
3160 | } | |
3161 | } else { | |
1da177e4 | 3162 | if (!buffer_uptodate(bh)) { |
76c3073a | 3163 | bh->b_end_io = end_buffer_read_sync; |
e60e5c50 | 3164 | get_bh(bh); |
1da177e4 LT |
3165 | submit_bh(rw, bh); |
3166 | continue; | |
3167 | } | |
3168 | } | |
3169 | unlock_buffer(bh); | |
1da177e4 LT |
3170 | } |
3171 | } | |
1fe72eaa | 3172 | EXPORT_SYMBOL(ll_rw_block); |
1da177e4 | 3173 | |
9cb569d6 CH |
3174 | void write_dirty_buffer(struct buffer_head *bh, int rw) |
3175 | { | |
3176 | lock_buffer(bh); | |
3177 | if (!test_clear_buffer_dirty(bh)) { | |
3178 | unlock_buffer(bh); | |
3179 | return; | |
3180 | } | |
3181 | bh->b_end_io = end_buffer_write_sync; | |
3182 | get_bh(bh); | |
3183 | submit_bh(rw, bh); | |
3184 | } | |
3185 | EXPORT_SYMBOL(write_dirty_buffer); | |
3186 | ||
1da177e4 LT |
3187 | /* |
3188 | * For a data-integrity writeout, we need to wait upon any in-progress I/O | |
3189 | * and then start new I/O and then wait upon it. The caller must have a ref on | |
3190 | * the buffer_head. | |
3191 | */ | |
87e99511 | 3192 | int __sync_dirty_buffer(struct buffer_head *bh, int rw) |
1da177e4 LT |
3193 | { |
3194 | int ret = 0; | |
3195 | ||
3196 | WARN_ON(atomic_read(&bh->b_count) < 1); | |
3197 | lock_buffer(bh); | |
3198 | if (test_clear_buffer_dirty(bh)) { | |
3199 | get_bh(bh); | |
3200 | bh->b_end_io = end_buffer_write_sync; | |
87e99511 | 3201 | ret = submit_bh(rw, bh); |
1da177e4 | 3202 | wait_on_buffer(bh); |
1da177e4 LT |
3203 | if (!ret && !buffer_uptodate(bh)) |
3204 | ret = -EIO; | |
3205 | } else { | |
3206 | unlock_buffer(bh); | |
3207 | } | |
3208 | return ret; | |
3209 | } | |
87e99511 CH |
3210 | EXPORT_SYMBOL(__sync_dirty_buffer); |
3211 | ||
3212 | int sync_dirty_buffer(struct buffer_head *bh) | |
3213 | { | |
3214 | return __sync_dirty_buffer(bh, WRITE_SYNC); | |
3215 | } | |
1fe72eaa | 3216 | EXPORT_SYMBOL(sync_dirty_buffer); |
1da177e4 LT |
3217 | |
3218 | /* | |
3219 | * try_to_free_buffers() checks if all the buffers on this particular page | |
3220 | * are unused, and releases them if so. | |
3221 | * | |
3222 | * Exclusion against try_to_free_buffers may be obtained by either | |
3223 | * locking the page or by holding its mapping's private_lock. | |
3224 | * | |
3225 | * If the page is dirty but all the buffers are clean then we need to | |
3226 | * be sure to mark the page clean as well. This is because the page | |
3227 | * may be against a block device, and a later reattachment of buffers | |
3228 | * to a dirty page will set *all* buffers dirty. Which would corrupt | |
3229 | * filesystem data on the same device. | |
3230 | * | |
3231 | * The same applies to regular filesystem pages: if all the buffers are | |
3232 | * clean then we set the page clean and proceed. To do that, we require | |
3233 | * total exclusion from __set_page_dirty_buffers(). That is obtained with | |
3234 | * private_lock. | |
3235 | * | |
3236 | * try_to_free_buffers() is non-blocking. | |
3237 | */ | |
3238 | static inline int buffer_busy(struct buffer_head *bh) | |
3239 | { | |
3240 | return atomic_read(&bh->b_count) | | |
3241 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); | |
3242 | } | |
3243 | ||
3244 | static int | |
3245 | drop_buffers(struct page *page, struct buffer_head **buffers_to_free) | |
3246 | { | |
3247 | struct buffer_head *head = page_buffers(page); | |
3248 | struct buffer_head *bh; | |
3249 | ||
3250 | bh = head; | |
3251 | do { | |
de7d5a3b | 3252 | if (buffer_write_io_error(bh) && page->mapping) |
1da177e4 LT |
3253 | set_bit(AS_EIO, &page->mapping->flags); |
3254 | if (buffer_busy(bh)) | |
3255 | goto failed; | |
3256 | bh = bh->b_this_page; | |
3257 | } while (bh != head); | |
3258 | ||
3259 | do { | |
3260 | struct buffer_head *next = bh->b_this_page; | |
3261 | ||
535ee2fb | 3262 | if (bh->b_assoc_map) |
1da177e4 LT |
3263 | __remove_assoc_queue(bh); |
3264 | bh = next; | |
3265 | } while (bh != head); | |
3266 | *buffers_to_free = head; | |
3267 | __clear_page_buffers(page); | |
3268 | return 1; | |
3269 | failed: | |
3270 | return 0; | |
3271 | } | |
3272 | ||
3273 | int try_to_free_buffers(struct page *page) | |
3274 | { | |
3275 | struct address_space * const mapping = page->mapping; | |
3276 | struct buffer_head *buffers_to_free = NULL; | |
3277 | int ret = 0; | |
3278 | ||
3279 | BUG_ON(!PageLocked(page)); | |
ecdfc978 | 3280 | if (PageWriteback(page)) |
1da177e4 LT |
3281 | return 0; |
3282 | ||
3283 | if (mapping == NULL) { /* can this still happen? */ | |
3284 | ret = drop_buffers(page, &buffers_to_free); | |
3285 | goto out; | |
3286 | } | |
3287 | ||
3288 | spin_lock(&mapping->private_lock); | |
3289 | ret = drop_buffers(page, &buffers_to_free); | |
ecdfc978 LT |
3290 | |
3291 | /* | |
3292 | * If the filesystem writes its buffers by hand (eg ext3) | |
3293 | * then we can have clean buffers against a dirty page. We | |
3294 | * clean the page here; otherwise the VM will never notice | |
3295 | * that the filesystem did any IO at all. | |
3296 | * | |
3297 | * Also, during truncate, discard_buffer will have marked all | |
3298 | * the page's buffers clean. We discover that here and clean | |
3299 | * the page also. | |
87df7241 NP |
3300 | * |
3301 | * private_lock must be held over this entire operation in order | |
3302 | * to synchronise against __set_page_dirty_buffers and prevent the | |
3303 | * dirty bit from being lost. | |
ecdfc978 | 3304 | */ |
11f81bec TH |
3305 | if (ret) |
3306 | cancel_dirty_page(page); | |
87df7241 | 3307 | spin_unlock(&mapping->private_lock); |
1da177e4 LT |
3308 | out: |
3309 | if (buffers_to_free) { | |
3310 | struct buffer_head *bh = buffers_to_free; | |
3311 | ||
3312 | do { | |
3313 | struct buffer_head *next = bh->b_this_page; | |
3314 | free_buffer_head(bh); | |
3315 | bh = next; | |
3316 | } while (bh != buffers_to_free); | |
3317 | } | |
3318 | return ret; | |
3319 | } | |
3320 | EXPORT_SYMBOL(try_to_free_buffers); | |
3321 | ||
1da177e4 LT |
3322 | /* |
3323 | * There are no bdflush tunables left. But distributions are | |
3324 | * still running obsolete flush daemons, so we terminate them here. | |
3325 | * | |
3326 | * Use of bdflush() is deprecated and will be removed in a future kernel. | |
5b0830cb | 3327 | * The `flush-X' kernel threads fully replace bdflush daemons and this call. |
1da177e4 | 3328 | */ |
bdc480e3 | 3329 | SYSCALL_DEFINE2(bdflush, int, func, long, data) |
1da177e4 LT |
3330 | { |
3331 | static int msg_count; | |
3332 | ||
3333 | if (!capable(CAP_SYS_ADMIN)) | |
3334 | return -EPERM; | |
3335 | ||
3336 | if (msg_count < 5) { | |
3337 | msg_count++; | |
3338 | printk(KERN_INFO | |
3339 | "warning: process `%s' used the obsolete bdflush" | |
3340 | " system call\n", current->comm); | |
3341 | printk(KERN_INFO "Fix your initscripts?\n"); | |
3342 | } | |
3343 | ||
3344 | if (func == 1) | |
3345 | do_exit(0); | |
3346 | return 0; | |
3347 | } | |
3348 | ||
3349 | /* | |
3350 | * Buffer-head allocation | |
3351 | */ | |
a0a9b043 | 3352 | static struct kmem_cache *bh_cachep __read_mostly; |
1da177e4 LT |
3353 | |
3354 | /* | |
3355 | * Once the number of bh's in the machine exceeds this level, we start | |
3356 | * stripping them in writeback. | |
3357 | */ | |
43be594a | 3358 | static unsigned long max_buffer_heads; |
1da177e4 LT |
3359 | |
3360 | int buffer_heads_over_limit; | |
3361 | ||
3362 | struct bh_accounting { | |
3363 | int nr; /* Number of live bh's */ | |
3364 | int ratelimit; /* Limit cacheline bouncing */ | |
3365 | }; | |
3366 | ||
3367 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; | |
3368 | ||
3369 | static void recalc_bh_state(void) | |
3370 | { | |
3371 | int i; | |
3372 | int tot = 0; | |
3373 | ||
ee1be862 | 3374 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
1da177e4 | 3375 | return; |
c7b92516 | 3376 | __this_cpu_write(bh_accounting.ratelimit, 0); |
8a143426 | 3377 | for_each_online_cpu(i) |
1da177e4 LT |
3378 | tot += per_cpu(bh_accounting, i).nr; |
3379 | buffer_heads_over_limit = (tot > max_buffer_heads); | |
3380 | } | |
c7b92516 | 3381 | |
dd0fc66f | 3382 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
1da177e4 | 3383 | { |
019b4d12 | 3384 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
1da177e4 | 3385 | if (ret) { |
a35afb83 | 3386 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
c7b92516 CL |
3387 | preempt_disable(); |
3388 | __this_cpu_inc(bh_accounting.nr); | |
1da177e4 | 3389 | recalc_bh_state(); |
c7b92516 | 3390 | preempt_enable(); |
1da177e4 LT |
3391 | } |
3392 | return ret; | |
3393 | } | |
3394 | EXPORT_SYMBOL(alloc_buffer_head); | |
3395 | ||
3396 | void free_buffer_head(struct buffer_head *bh) | |
3397 | { | |
3398 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); | |
3399 | kmem_cache_free(bh_cachep, bh); | |
c7b92516 CL |
3400 | preempt_disable(); |
3401 | __this_cpu_dec(bh_accounting.nr); | |
1da177e4 | 3402 | recalc_bh_state(); |
c7b92516 | 3403 | preempt_enable(); |
1da177e4 LT |
3404 | } |
3405 | EXPORT_SYMBOL(free_buffer_head); | |
3406 | ||
1da177e4 LT |
3407 | static void buffer_exit_cpu(int cpu) |
3408 | { | |
3409 | int i; | |
3410 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); | |
3411 | ||
3412 | for (i = 0; i < BH_LRU_SIZE; i++) { | |
3413 | brelse(b->bhs[i]); | |
3414 | b->bhs[i] = NULL; | |
3415 | } | |
c7b92516 | 3416 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
8a143426 | 3417 | per_cpu(bh_accounting, cpu).nr = 0; |
1da177e4 LT |
3418 | } |
3419 | ||
3420 | static int buffer_cpu_notify(struct notifier_block *self, | |
3421 | unsigned long action, void *hcpu) | |
3422 | { | |
8bb78442 | 3423 | if (action == CPU_DEAD || action == CPU_DEAD_FROZEN) |
1da177e4 LT |
3424 | buffer_exit_cpu((unsigned long)hcpu); |
3425 | return NOTIFY_OK; | |
3426 | } | |
1da177e4 | 3427 | |
389d1b08 | 3428 | /** |
a6b91919 | 3429 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
389d1b08 AK |
3430 | * @bh: struct buffer_head |
3431 | * | |
3432 | * Return true if the buffer is up-to-date and false, | |
3433 | * with the buffer locked, if not. | |
3434 | */ | |
3435 | int bh_uptodate_or_lock(struct buffer_head *bh) | |
3436 | { | |
3437 | if (!buffer_uptodate(bh)) { | |
3438 | lock_buffer(bh); | |
3439 | if (!buffer_uptodate(bh)) | |
3440 | return 0; | |
3441 | unlock_buffer(bh); | |
3442 | } | |
3443 | return 1; | |
3444 | } | |
3445 | EXPORT_SYMBOL(bh_uptodate_or_lock); | |
3446 | ||
3447 | /** | |
a6b91919 | 3448 | * bh_submit_read - Submit a locked buffer for reading |
389d1b08 AK |
3449 | * @bh: struct buffer_head |
3450 | * | |
3451 | * Returns zero on success and -EIO on error. | |
3452 | */ | |
3453 | int bh_submit_read(struct buffer_head *bh) | |
3454 | { | |
3455 | BUG_ON(!buffer_locked(bh)); | |
3456 | ||
3457 | if (buffer_uptodate(bh)) { | |
3458 | unlock_buffer(bh); | |
3459 | return 0; | |
3460 | } | |
3461 | ||
3462 | get_bh(bh); | |
3463 | bh->b_end_io = end_buffer_read_sync; | |
3464 | submit_bh(READ, bh); | |
3465 | wait_on_buffer(bh); | |
3466 | if (buffer_uptodate(bh)) | |
3467 | return 0; | |
3468 | return -EIO; | |
3469 | } | |
3470 | EXPORT_SYMBOL(bh_submit_read); | |
3471 | ||
1da177e4 LT |
3472 | void __init buffer_init(void) |
3473 | { | |
43be594a | 3474 | unsigned long nrpages; |
1da177e4 | 3475 | |
b98938c3 CL |
3476 | bh_cachep = kmem_cache_create("buffer_head", |
3477 | sizeof(struct buffer_head), 0, | |
3478 | (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC| | |
3479 | SLAB_MEM_SPREAD), | |
019b4d12 | 3480 | NULL); |
1da177e4 LT |
3481 | |
3482 | /* | |
3483 | * Limit the bh occupancy to 10% of ZONE_NORMAL | |
3484 | */ | |
3485 | nrpages = (nr_free_buffer_pages() * 10) / 100; | |
3486 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); | |
3487 | hotcpu_notifier(buffer_cpu_notify, 0); | |
3488 | } |