blob: c017a2dfb9097e73837230d244ffb14f73dd0e1c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070021#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>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080027#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <linux/blkdev.h>
29#include <linux/file.h>
30#include <linux/quotaops.h>
31#include <linux/highmem.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050032#include <linux/export.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070033#include <linux/writeback.h>
34#include <linux/hash.h>
35#include <linux/suspend.h>
36#include <linux/buffer_head.h>
Andrew Morton55e829a2006-12-10 02:19:27 -080037#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038#include <linux/bio.h>
39#include <linux/notifier.h>
40#include <linux/cpu.h>
41#include <linux/bitops.h>
42#include <linux/mpage.h>
Ingo Molnarfb1c8f92005-09-10 00:25:56 -070043#include <linux/bit_spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
45static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
48
Yan Honga3f3c292012-12-12 13:52:15 -080049void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -070050{
51 bh->b_end_io = handler;
52 bh->b_private = private;
53}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070054EXPORT_SYMBOL(init_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070055
Jens Axboe7eaceac2011-03-10 08:52:07 +010056static int sleep_on_buffer(void *word)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057{
Linus Torvalds1da177e2005-04-16 15:20:36 -070058 io_schedule();
59 return 0;
60}
61
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080062void __lock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070063{
Jens Axboe7eaceac2011-03-10 08:52:07 +010064 wait_on_bit_lock(&bh->b_state, BH_Lock, sleep_on_buffer,
Linus Torvalds1da177e2005-04-16 15:20:36 -070065 TASK_UNINTERRUPTIBLE);
66}
67EXPORT_SYMBOL(__lock_buffer);
68
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080069void unlock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
Nick Piggin51b07fc2008-10-18 20:27:00 -070071 clear_bit_unlock(BH_Lock, &bh->b_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072 smp_mb__after_clear_bit();
73 wake_up_bit(&bh->b_state, BH_Lock);
74}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070075EXPORT_SYMBOL(unlock_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070076
77/*
78 * Block until a buffer comes unlocked. This doesn't stop it
79 * from becoming locked again - you have to lock it yourself
80 * if you want to preserve its state.
81 */
82void __wait_on_buffer(struct buffer_head * bh)
83{
Jens Axboe7eaceac2011-03-10 08:52:07 +010084 wait_on_bit(&bh->b_state, BH_Lock, sleep_on_buffer, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070085}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070086EXPORT_SYMBOL(__wait_on_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070087
88static void
89__clear_page_buffers(struct page *page)
90{
91 ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -070092 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -070093 page_cache_release(page);
94}
95
Keith Mannthey08bafc02008-11-25 10:24:35 +010096
97static int quiet_error(struct buffer_head *bh)
98{
99 if (!test_bit(BH_Quiet, &bh->b_state) && printk_ratelimit())
100 return 0;
101 return 1;
102}
103
104
Linus Torvalds1da177e2005-04-16 15:20:36 -0700105static void buffer_io_error(struct buffer_head *bh)
106{
107 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700108 printk(KERN_ERR "Buffer I/O error on device %s, logical block %Lu\n",
109 bdevname(bh->b_bdev, b),
110 (unsigned long long)bh->b_blocknr);
111}
112
113/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700114 * End-of-IO handler helper function which does not touch the bh after
115 * unlocking it.
116 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
117 * a race there is benign: unlock_buffer() only use the bh's address for
118 * hashing after unlocking the buffer, so it doesn't actually touch the bh
119 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700121static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700122{
123 if (uptodate) {
124 set_buffer_uptodate(bh);
125 } else {
126 /* This happens, due to failed READA attempts. */
127 clear_buffer_uptodate(bh);
128 }
129 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700130}
131
132/*
133 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
134 * unlock the buffer. This is what ll_rw_block uses too.
135 */
136void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
137{
138 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700139 put_bh(bh);
140}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700141EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142
143void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
144{
145 char b[BDEVNAME_SIZE];
146
147 if (uptodate) {
148 set_buffer_uptodate(bh);
149 } else {
Christoph Hellwig0edd55f2010-08-18 05:29:23 -0400150 if (!quiet_error(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 buffer_io_error(bh);
152 printk(KERN_WARNING "lost page write due to "
153 "I/O error on %s\n",
154 bdevname(bh->b_bdev, b));
155 }
156 set_buffer_write_io_error(bh);
157 clear_buffer_uptodate(bh);
158 }
159 unlock_buffer(bh);
160 put_bh(bh);
161}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700162EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700165 * Various filesystems appear to want __find_get_block to be non-blocking.
166 * But it's the page lock which protects the buffers. To get around this,
167 * we get exclusion from try_to_free_buffers with the blockdev mapping's
168 * private_lock.
169 *
170 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
171 * may be quite high. This code could TryLock the page, and if that
172 * succeeds, there is no need to take private_lock. (But if
173 * private_lock is contended then so is mapping->tree_lock).
174 */
175static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800176__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177{
178 struct inode *bd_inode = bdev->bd_inode;
179 struct address_space *bd_mapping = bd_inode->i_mapping;
180 struct buffer_head *ret = NULL;
181 pgoff_t index;
182 struct buffer_head *bh;
183 struct buffer_head *head;
184 struct page *page;
185 int all_mapped = 1;
186
187 index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits);
188 page = find_get_page(bd_mapping, index);
189 if (!page)
190 goto out;
191
192 spin_lock(&bd_mapping->private_lock);
193 if (!page_has_buffers(page))
194 goto out_unlock;
195 head = page_buffers(page);
196 bh = head;
197 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700198 if (!buffer_mapped(bh))
199 all_mapped = 0;
200 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201 ret = bh;
202 get_bh(bh);
203 goto out_unlock;
204 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700205 bh = bh->b_this_page;
206 } while (bh != head);
207
208 /* we might be here because some of the buffers on this page are
209 * not mapped. This is due to various races between
210 * file io on the block device and getblk. It gets dealt with
211 * elsewhere, don't buffer_error if we had some unmapped buffers
212 */
213 if (all_mapped) {
Tao Ma72a2ebd2011-10-31 17:09:00 -0700214 char b[BDEVNAME_SIZE];
215
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 printk("__find_get_block_slow() failed. "
217 "block=%llu, b_blocknr=%llu\n",
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800218 (unsigned long long)block,
219 (unsigned long long)bh->b_blocknr);
220 printk("b_state=0x%08lx, b_size=%zu\n",
221 bh->b_state, bh->b_size);
Tao Ma72a2ebd2011-10-31 17:09:00 -0700222 printk("device %s blocksize: %d\n", bdevname(bdev, b),
223 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 }
225out_unlock:
226 spin_unlock(&bd_mapping->private_lock);
227 page_cache_release(page);
228out:
229 return ret;
230}
231
Linus Torvalds1da177e2005-04-16 15:20:36 -0700232/*
Jens Axboe5b0830c2009-09-23 19:37:09 +0200233 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234 */
235static void free_more_memory(void)
236{
Mel Gorman19770b32008-04-28 02:12:18 -0700237 struct zone *zone;
Mel Gorman0e884602008-04-28 02:12:14 -0700238 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700239
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600240 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241 yield();
242
Mel Gorman0e884602008-04-28 02:12:14 -0700243 for_each_online_node(nid) {
Mel Gorman19770b32008-04-28 02:12:18 -0700244 (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
245 gfp_zone(GFP_NOFS), NULL,
246 &zone);
247 if (zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700248 try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -0700249 GFP_NOFS, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
251}
252
253/*
254 * I/O completion handler for block_read_full_page() - pages
255 * which come unlocked at the end of I/O.
256 */
257static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
258{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700259 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700260 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 struct buffer_head *tmp;
262 struct page *page;
263 int page_uptodate = 1;
264
265 BUG_ON(!buffer_async_read(bh));
266
267 page = bh->b_page;
268 if (uptodate) {
269 set_buffer_uptodate(bh);
270 } else {
271 clear_buffer_uptodate(bh);
Keith Mannthey08bafc02008-11-25 10:24:35 +0100272 if (!quiet_error(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 buffer_io_error(bh);
274 SetPageError(page);
275 }
276
277 /*
278 * Be _very_ careful from here on. Bad things can happen if
279 * two buffer heads end IO at almost the same time and both
280 * decide that the page is now completely done.
281 */
Nick Piggina3972202005-07-07 17:56:56 -0700282 first = page_buffers(page);
283 local_irq_save(flags);
284 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 clear_buffer_async_read(bh);
286 unlock_buffer(bh);
287 tmp = bh;
288 do {
289 if (!buffer_uptodate(tmp))
290 page_uptodate = 0;
291 if (buffer_async_read(tmp)) {
292 BUG_ON(!buffer_locked(tmp));
293 goto still_busy;
294 }
295 tmp = tmp->b_this_page;
296 } while (tmp != bh);
Nick Piggina3972202005-07-07 17:56:56 -0700297 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
298 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299
300 /*
301 * If none of the buffers had errors and they are all
302 * uptodate then we can set the page uptodate.
303 */
304 if (page_uptodate && !PageError(page))
305 SetPageUptodate(page);
306 unlock_page(page);
307 return;
308
309still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700310 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
311 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700312 return;
313}
314
315/*
316 * Completion handler for block_write_full_page() - pages which are unlocked
317 * during I/O, and which have PageWriteback cleared upon I/O completion.
318 */
Chris Mason35c80d52009-04-15 13:22:38 -0400319void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320{
321 char b[BDEVNAME_SIZE];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700323 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324 struct buffer_head *tmp;
325 struct page *page;
326
327 BUG_ON(!buffer_async_write(bh));
328
329 page = bh->b_page;
330 if (uptodate) {
331 set_buffer_uptodate(bh);
332 } else {
Keith Mannthey08bafc02008-11-25 10:24:35 +0100333 if (!quiet_error(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 buffer_io_error(bh);
335 printk(KERN_WARNING "lost page write due to "
336 "I/O error on %s\n",
337 bdevname(bh->b_bdev, b));
338 }
339 set_bit(AS_EIO, &page->mapping->flags);
Jan Kara58ff4072006-10-17 00:10:19 -0700340 set_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 clear_buffer_uptodate(bh);
342 SetPageError(page);
343 }
344
Nick Piggina3972202005-07-07 17:56:56 -0700345 first = page_buffers(page);
346 local_irq_save(flags);
347 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
348
Linus Torvalds1da177e2005-04-16 15:20:36 -0700349 clear_buffer_async_write(bh);
350 unlock_buffer(bh);
351 tmp = bh->b_this_page;
352 while (tmp != bh) {
353 if (buffer_async_write(tmp)) {
354 BUG_ON(!buffer_locked(tmp));
355 goto still_busy;
356 }
357 tmp = tmp->b_this_page;
358 }
Nick Piggina3972202005-07-07 17:56:56 -0700359 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
360 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700361 end_page_writeback(page);
362 return;
363
364still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700365 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
366 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700367 return;
368}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700369EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700370
371/*
372 * If a page's buffers are under async readin (end_buffer_async_read
373 * completion) then there is a possibility that another thread of
374 * control could lock one of the buffers after it has completed
375 * but while some of the other buffers have not completed. This
376 * locked buffer would confuse end_buffer_async_read() into not unlocking
377 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
378 * that this buffer is not under async I/O.
379 *
380 * The page comes unlocked when it has no locked buffer_async buffers
381 * left.
382 *
383 * PageLocked prevents anyone starting new async I/O reads any of
384 * the buffers.
385 *
386 * PageWriteback is used to prevent simultaneous writeout of the same
387 * page.
388 *
389 * PageLocked prevents anyone from starting writeback of a page which is
390 * under read I/O (PageWriteback is only ever set against a locked page).
391 */
392static void mark_buffer_async_read(struct buffer_head *bh)
393{
394 bh->b_end_io = end_buffer_async_read;
395 set_buffer_async_read(bh);
396}
397
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700398static void mark_buffer_async_write_endio(struct buffer_head *bh,
399 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400400{
401 bh->b_end_io = handler;
402 set_buffer_async_write(bh);
403}
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405void mark_buffer_async_write(struct buffer_head *bh)
406{
Chris Mason35c80d52009-04-15 13:22:38 -0400407 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408}
409EXPORT_SYMBOL(mark_buffer_async_write);
410
411
412/*
413 * fs/buffer.c contains helper functions for buffer-backed address space's
414 * fsync functions. A common requirement for buffer-based filesystems is
415 * that certain data from the backing blockdev needs to be written out for
416 * a successful fsync(). For example, ext2 indirect blocks need to be
417 * written back and waited upon before fsync() returns.
418 *
419 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
420 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
421 * management of a list of dependent buffers at ->i_mapping->private_list.
422 *
423 * Locking is a little subtle: try_to_free_buffers() will remove buffers
424 * from their controlling inode's queue when they are being freed. But
425 * try_to_free_buffers() will be operating against the *blockdev* mapping
426 * at the time, not against the S_ISREG file which depends on those buffers.
427 * So the locking for private_list is via the private_lock in the address_space
428 * which backs the buffers. Which is different from the address_space
429 * against which the buffers are listed. So for a particular address_space,
430 * mapping->private_lock does *not* protect mapping->private_list! In fact,
431 * mapping->private_list will always be protected by the backing blockdev's
432 * ->private_lock.
433 *
434 * Which introduces a requirement: all buffers on an address_space's
435 * ->private_list must be from the same address_space: the blockdev's.
436 *
437 * address_spaces which do not place buffers at ->private_list via these
438 * utility functions are free to use private_lock and private_list for
439 * whatever they want. The only requirement is that list_empty(private_list)
440 * be true at clear_inode() time.
441 *
442 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
443 * filesystems should do that. invalidate_inode_buffers() should just go
444 * BUG_ON(!list_empty).
445 *
446 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
447 * take an address_space, not an inode. And it should be called
448 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
449 * queued up.
450 *
451 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
452 * list if it is already on a list. Because if the buffer is on a list,
453 * it *must* already be on the right one. If not, the filesystem is being
454 * silly. This will save a ton of locking. But first we have to ensure
455 * that buffers are taken *off* the old inode's list when they are freed
456 * (presumably in truncate). That requires careful auditing of all
457 * filesystems (do it inside bforget()). It could also be done by bringing
458 * b_inode back.
459 */
460
461/*
462 * The buffer's backing address_space's private_lock must be held
463 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700464static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700465{
466 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -0700467 WARN_ON(!bh->b_assoc_map);
468 if (buffer_write_io_error(bh))
469 set_bit(AS_EIO, &bh->b_assoc_map->flags);
470 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700471}
472
473int inode_has_buffers(struct inode *inode)
474{
475 return !list_empty(&inode->i_data.private_list);
476}
477
478/*
479 * osync is designed to support O_SYNC io. It waits synchronously for
480 * all already-submitted IO to complete, but does not queue any new
481 * writes to the disk.
482 *
483 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
484 * you dirty the buffers, and then use osync_inode_buffers to wait for
485 * completion. Any other dirty buffers which are not yet queued for
486 * write will not be flushed to disk by the osync.
487 */
488static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
489{
490 struct buffer_head *bh;
491 struct list_head *p;
492 int err = 0;
493
494 spin_lock(lock);
495repeat:
496 list_for_each_prev(p, list) {
497 bh = BH_ENTRY(p);
498 if (buffer_locked(bh)) {
499 get_bh(bh);
500 spin_unlock(lock);
501 wait_on_buffer(bh);
502 if (!buffer_uptodate(bh))
503 err = -EIO;
504 brelse(bh);
505 spin_lock(lock);
506 goto repeat;
507 }
508 }
509 spin_unlock(lock);
510 return err;
511}
512
Al Viro01a05b32010-03-23 06:06:58 -0400513static void do_thaw_one(struct super_block *sb, void *unused)
514{
515 char b[BDEVNAME_SIZE];
516 while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
517 printk(KERN_WARNING "Emergency Thaw on %s\n",
518 bdevname(sb->s_bdev, b));
519}
520
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700521static void do_thaw_all(struct work_struct *work)
Eric Sandeenc2d75432009-03-31 15:23:46 -0700522{
Al Viro01a05b32010-03-23 06:06:58 -0400523 iterate_supers(do_thaw_one, NULL);
Jens Axboe053c5252009-04-08 13:44:08 +0200524 kfree(work);
Eric Sandeenc2d75432009-03-31 15:23:46 -0700525 printk(KERN_WARNING "Emergency Thaw complete\n");
526}
527
528/**
529 * emergency_thaw_all -- forcibly thaw every frozen filesystem
530 *
531 * Used for emergency unfreeze of all filesystems via SysRq
532 */
533void emergency_thaw_all(void)
534{
Jens Axboe053c5252009-04-08 13:44:08 +0200535 struct work_struct *work;
536
537 work = kmalloc(sizeof(*work), GFP_ATOMIC);
538 if (work) {
539 INIT_WORK(work, do_thaw_all);
540 schedule_work(work);
541 }
Eric Sandeenc2d75432009-03-31 15:23:46 -0700542}
543
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800545 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700546 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547 *
548 * Starts I/O against the buffers at mapping->private_list, and waits upon
549 * that I/O.
550 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700551 * Basically, this is a convenience function for fsync().
552 * @mapping is a file or directory which needs those buffers to be written for
553 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700554 */
555int sync_mapping_buffers(struct address_space *mapping)
556{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800557 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558
559 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
560 return 0;
561
562 return fsync_buffers_list(&buffer_mapping->private_lock,
563 &mapping->private_list);
564}
565EXPORT_SYMBOL(sync_mapping_buffers);
566
567/*
568 * Called when we've recently written block `bblock', and it is known that
569 * `bblock' was for a buffer_boundary() buffer. This means that the block at
570 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
571 * dirty, schedule it for IO. So that indirects merge nicely with their data.
572 */
573void write_boundary_block(struct block_device *bdev,
574 sector_t bblock, unsigned blocksize)
575{
576 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
577 if (bh) {
578 if (buffer_dirty(bh))
579 ll_rw_block(WRITE, 1, &bh);
580 put_bh(bh);
581 }
582}
583
584void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
585{
586 struct address_space *mapping = inode->i_mapping;
587 struct address_space *buffer_mapping = bh->b_page->mapping;
588
589 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800590 if (!mapping->private_data) {
591 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800593 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800595 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 spin_lock(&buffer_mapping->private_lock);
597 list_move_tail(&bh->b_assoc_buffers,
598 &mapping->private_list);
Jan Kara58ff4072006-10-17 00:10:19 -0700599 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 spin_unlock(&buffer_mapping->private_lock);
601 }
602}
603EXPORT_SYMBOL(mark_buffer_dirty_inode);
604
605/*
Nick Piggin787d2212007-07-17 04:03:34 -0700606 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
607 * dirty.
608 *
609 * If warn is true, then emit a warning if the page is not uptodate and has
610 * not been truncated.
611 */
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700612static void __set_page_dirty(struct page *page,
Nick Piggin787d2212007-07-17 04:03:34 -0700613 struct address_space *mapping, int warn)
614{
Nick Piggin19fd6232008-07-25 19:45:32 -0700615 spin_lock_irq(&mapping->tree_lock);
Nick Piggin787d2212007-07-17 04:03:34 -0700616 if (page->mapping) { /* Race with truncate? */
617 WARN_ON_ONCE(warn && !PageUptodate(page));
Edward Shishkine3a7cca2009-03-31 15:19:39 -0700618 account_page_dirtied(page, mapping);
Nick Piggin787d2212007-07-17 04:03:34 -0700619 radix_tree_tag_set(&mapping->page_tree,
620 page_index(page), PAGECACHE_TAG_DIRTY);
621 }
Nick Piggin19fd6232008-07-25 19:45:32 -0700622 spin_unlock_irq(&mapping->tree_lock);
Nick Piggin787d2212007-07-17 04:03:34 -0700623 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Nick Piggin787d2212007-07-17 04:03:34 -0700624}
625
626/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700627 * Add a page to the dirty page list.
628 *
629 * It is a sad fact of life that this function is called from several places
630 * deeply under spinlocking. It may not sleep.
631 *
632 * If the page has buffers, the uptodate buffers are set dirty, to preserve
633 * dirty-state coherency between the page and the buffers. It the page does
634 * not have buffers then when they are later attached they will all be set
635 * dirty.
636 *
637 * The buffers are dirtied before the page is dirtied. There's a small race
638 * window in which a writepage caller may see the page cleanness but not the
639 * buffer dirtiness. That's fine. If this code were to set the page dirty
640 * before the buffers, a concurrent writepage caller could clear the page dirty
641 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
642 * page on the dirty page list.
643 *
644 * We use private_lock to lock against try_to_free_buffers while using the
645 * page's buffer list. Also use this to protect against clean buffers being
646 * added to the page after it was set dirty.
647 *
648 * FIXME: may need to call ->reservepage here as well. That's rather up to the
649 * address_space though.
650 */
651int __set_page_dirty_buffers(struct page *page)
652{
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700653 int newly_dirty;
Nick Piggin787d2212007-07-17 04:03:34 -0700654 struct address_space *mapping = page_mapping(page);
Nick Pigginebf7a222006-10-10 04:36:54 +0200655
656 if (unlikely(!mapping))
657 return !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700658
659 spin_lock(&mapping->private_lock);
660 if (page_has_buffers(page)) {
661 struct buffer_head *head = page_buffers(page);
662 struct buffer_head *bh = head;
663
664 do {
665 set_buffer_dirty(bh);
666 bh = bh->b_this_page;
667 } while (bh != head);
668 }
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700669 newly_dirty = !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700670 spin_unlock(&mapping->private_lock);
671
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700672 if (newly_dirty)
673 __set_page_dirty(page, mapping, 1);
674 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675}
676EXPORT_SYMBOL(__set_page_dirty_buffers);
677
678/*
679 * Write out and wait upon a list of buffers.
680 *
681 * We have conflicting pressures: we want to make sure that all
682 * initially dirty buffers get waited on, but that any subsequently
683 * dirtied buffers don't. After all, we don't want fsync to last
684 * forever if somebody is actively writing to the file.
685 *
686 * Do this in two main stages: first we copy dirty buffers to a
687 * temporary inode list, queueing the writes as we go. Then we clean
688 * up, waiting for those writes to complete.
689 *
690 * During this second stage, any subsequent updates to the file may end
691 * up refiling the buffer on the original inode's dirty list again, so
692 * there is a chance we will end up with a buffer queued for write but
693 * not yet completed on that list. So, as a final cleanup we go through
694 * the osync code to catch these locked, dirty buffers without requeuing
695 * any newly dirty buffers for write.
696 */
697static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
698{
699 struct buffer_head *bh;
700 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100701 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100703 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700704
705 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100706 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707
708 spin_lock(lock);
709 while (!list_empty(list)) {
710 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800711 mapping = bh->b_assoc_map;
Jan Kara58ff4072006-10-17 00:10:19 -0700712 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800713 /* Avoid race with mark_buffer_dirty_inode() which does
714 * a lockless check and we rely on seeing the dirty bit */
715 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700716 if (buffer_dirty(bh) || buffer_locked(bh)) {
717 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800718 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700719 if (buffer_dirty(bh)) {
720 get_bh(bh);
721 spin_unlock(lock);
722 /*
723 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200724 * write_dirty_buffer() actually writes the
725 * current contents - it is a noop if I/O is
726 * still in flight on potentially older
727 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 */
Jens Axboe721a9602011-03-09 11:56:30 +0100729 write_dirty_buffer(bh, WRITE_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200730
731 /*
732 * Kick off IO for the previous mapping. Note
733 * that we will not run the very last mapping,
734 * wait_on_buffer() will do that for us
735 * through sync_buffer().
736 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 brelse(bh);
738 spin_lock(lock);
739 }
740 }
741 }
742
Jens Axboe4ee24912011-03-17 10:51:40 +0100743 spin_unlock(lock);
744 blk_finish_plug(&plug);
745 spin_lock(lock);
746
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 while (!list_empty(&tmp)) {
748 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700749 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800750 mapping = bh->b_assoc_map;
751 __remove_assoc_queue(bh);
752 /* Avoid race with mark_buffer_dirty_inode() which does
753 * a lockless check and we rely on seeing the dirty bit */
754 smp_mb();
755 if (buffer_dirty(bh)) {
756 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800757 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800758 bh->b_assoc_map = mapping;
759 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700760 spin_unlock(lock);
761 wait_on_buffer(bh);
762 if (!buffer_uptodate(bh))
763 err = -EIO;
764 brelse(bh);
765 spin_lock(lock);
766 }
767
768 spin_unlock(lock);
769 err2 = osync_buffers_list(lock, list);
770 if (err)
771 return err;
772 else
773 return err2;
774}
775
776/*
777 * Invalidate any and all dirty buffers on a given inode. We are
778 * probably unmounting the fs, but that doesn't mean we have already
779 * done a sync(). Just drop the buffers from the inode list.
780 *
781 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
782 * assumes that all the buffers are against the blockdev. Not true
783 * for reiserfs.
784 */
785void invalidate_inode_buffers(struct inode *inode)
786{
787 if (inode_has_buffers(inode)) {
788 struct address_space *mapping = &inode->i_data;
789 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800790 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791
792 spin_lock(&buffer_mapping->private_lock);
793 while (!list_empty(list))
794 __remove_assoc_queue(BH_ENTRY(list->next));
795 spin_unlock(&buffer_mapping->private_lock);
796 }
797}
Jan Kara52b19ac2008-09-23 18:24:08 +0200798EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
800/*
801 * Remove any clean buffers from the inode's buffer list. This is called
802 * when we're trying to free the inode itself. Those buffers can pin it.
803 *
804 * Returns true if all buffers were removed.
805 */
806int remove_inode_buffers(struct inode *inode)
807{
808 int ret = 1;
809
810 if (inode_has_buffers(inode)) {
811 struct address_space *mapping = &inode->i_data;
812 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800813 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
815 spin_lock(&buffer_mapping->private_lock);
816 while (!list_empty(list)) {
817 struct buffer_head *bh = BH_ENTRY(list->next);
818 if (buffer_dirty(bh)) {
819 ret = 0;
820 break;
821 }
822 __remove_assoc_queue(bh);
823 }
824 spin_unlock(&buffer_mapping->private_lock);
825 }
826 return ret;
827}
828
829/*
830 * Create the appropriate buffers when given a page for data area and
831 * the size of each buffer.. Use the bh->b_this_page linked list to
832 * follow the buffers created. Return NULL if unable to create more
833 * buffers.
834 *
835 * The retry flag is used to differentiate async IO (paging, swapping)
836 * which may not fail from ordinary buffer allocations.
837 */
838struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
839 int retry)
840{
841 struct buffer_head *bh, *head;
842 long offset;
843
844try_again:
845 head = NULL;
846 offset = PAGE_SIZE;
847 while ((offset -= size) >= 0) {
848 bh = alloc_buffer_head(GFP_NOFS);
849 if (!bh)
850 goto no_grow;
851
Linus Torvalds1da177e2005-04-16 15:20:36 -0700852 bh->b_this_page = head;
853 bh->b_blocknr = -1;
854 head = bh;
855
Linus Torvalds1da177e2005-04-16 15:20:36 -0700856 bh->b_size = size;
857
858 /* Link the buffer to its page */
859 set_bh_page(bh, page, offset);
860
Nathan Scott01ffe332006-01-17 09:02:07 +1100861 init_buffer(bh, NULL, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700862 }
863 return head;
864/*
865 * In case anything failed, we just free everything we got.
866 */
867no_grow:
868 if (head) {
869 do {
870 bh = head;
871 head = head->b_this_page;
872 free_buffer_head(bh);
873 } while (head);
874 }
875
876 /*
877 * Return failure for non-async IO requests. Async IO requests
878 * are not allowed to fail, so we have to wait until buffer heads
879 * become available. But we don't want tasks sleeping with
880 * partially complete buffers, so all were released above.
881 */
882 if (!retry)
883 return NULL;
884
885 /* We're _really_ low on memory. Now we just
886 * wait for old buffer heads to become free due to
887 * finishing IO. Since this is an async request and
888 * the reserve list is empty, we're sure there are
889 * async buffer heads in use.
890 */
891 free_more_memory();
892 goto try_again;
893}
894EXPORT_SYMBOL_GPL(alloc_page_buffers);
895
896static inline void
897link_dev_buffers(struct page *page, struct buffer_head *head)
898{
899 struct buffer_head *bh, *tail;
900
901 bh = head;
902 do {
903 tail = bh;
904 bh = bh->b_this_page;
905 } while (bh);
906 tail->b_this_page = head;
907 attach_page_buffers(page, head);
908}
909
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800910static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
911{
912 sector_t retval = ~((sector_t)0);
913 loff_t sz = i_size_read(bdev->bd_inode);
914
915 if (sz) {
916 unsigned int sizebits = blksize_bits(size);
917 retval = (sz >> sizebits);
918 }
919 return retval;
920}
921
Linus Torvalds1da177e2005-04-16 15:20:36 -0700922/*
923 * Initialise the state of a blockdev page's buffers.
924 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200925static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926init_page_buffers(struct page *page, struct block_device *bdev,
927 sector_t block, int size)
928{
929 struct buffer_head *head = page_buffers(page);
930 struct buffer_head *bh = head;
931 int uptodate = PageUptodate(page);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800932 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700933
934 do {
935 if (!buffer_mapped(bh)) {
936 init_buffer(bh, NULL, NULL);
937 bh->b_bdev = bdev;
938 bh->b_blocknr = block;
939 if (uptodate)
940 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200941 if (block < end_block)
942 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943 }
944 block++;
945 bh = bh->b_this_page;
946 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200947
948 /*
949 * Caller needs to validate requested block against end of device.
950 */
951 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952}
953
954/*
955 * Create the page-cache page that contains the requested block.
956 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200957 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200959static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700960grow_dev_page(struct block_device *bdev, sector_t block,
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200961 pgoff_t index, int size, int sizebits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700962{
963 struct inode *inode = bdev->bd_inode;
964 struct page *page;
965 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200966 sector_t end_block;
967 int ret = 0; /* Will call free_more_memory() */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Christoph Lameterea125892007-05-16 22:11:21 -0700969 page = find_or_create_page(inode->i_mapping, index,
Mel Gorman769848c2007-07-17 04:03:05 -0700970 (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS)|__GFP_MOVABLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971 if (!page)
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200972 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Eric Sesterhenne827f922006-03-26 18:24:46 +0200974 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975
976 if (page_has_buffers(page)) {
977 bh = page_buffers(page);
978 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200979 end_block = init_page_buffers(page, bdev,
980 index << sizebits, size);
981 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 }
983 if (!try_to_free_buffers(page))
984 goto failed;
985 }
986
987 /*
988 * Allocate some buffers for this page
989 */
990 bh = alloc_page_buffers(page, size, 0);
991 if (!bh)
992 goto failed;
993
994 /*
995 * Link the page to the buffers and initialise them. Take the
996 * lock to be atomic wrt __find_get_block(), which does not
997 * run under the page lock.
998 */
999 spin_lock(&inode->i_mapping->private_lock);
1000 link_dev_buffers(page, bh);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001001 end_block = init_page_buffers(page, bdev, index << sizebits, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001003done:
1004 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 unlock_page(page);
1007 page_cache_release(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001008 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009}
1010
1011/*
1012 * Create buffers for the specified block device block's page. If
1013 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001014 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001015static int
Linus Torvalds1da177e2005-04-16 15:20:36 -07001016grow_buffers(struct block_device *bdev, sector_t block, int size)
1017{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 pgoff_t index;
1019 int sizebits;
1020
1021 sizebits = -1;
1022 do {
1023 sizebits++;
1024 } while ((size << sizebits) < PAGE_SIZE);
1025
1026 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001027
Andrew Mortone5657932006-10-11 01:21:46 -07001028 /*
1029 * Check for a block which wants to lie outside our maximum possible
1030 * pagecache index. (this comparison is done using sector_t types).
1031 */
1032 if (unlikely(index != block >> sizebits)) {
1033 char b[BDEVNAME_SIZE];
1034
1035 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1036 "device %s\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001037 __func__, (unsigned long long)block,
Andrew Mortone5657932006-10-11 01:21:46 -07001038 bdevname(bdev, b));
1039 return -EIO;
1040 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001041
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 /* Create a page with the proper size buffers.. */
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001043 return grow_dev_page(bdev, block, index, size, sizebits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044}
1045
Adrian Bunk75c96f82005-05-05 16:16:09 -07001046static struct buffer_head *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047__getblk_slow(struct block_device *bdev, sector_t block, int size)
1048{
1049 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001050 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 (size < 512 || size > PAGE_SIZE))) {
1052 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1053 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001054 printk(KERN_ERR "logical block size: %d\n",
1055 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056
1057 dump_stack();
1058 return NULL;
1059 }
1060
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001061 for (;;) {
1062 struct buffer_head *bh;
1063 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
1065 bh = __find_get_block(bdev, block, size);
1066 if (bh)
1067 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001068
1069 ret = grow_buffers(bdev, block, size);
1070 if (ret < 0)
1071 return NULL;
1072 if (ret == 0)
1073 free_more_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 }
1075}
1076
1077/*
1078 * The relationship between dirty buffers and dirty pages:
1079 *
1080 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1081 * the page is tagged dirty in its radix tree.
1082 *
1083 * At all times, the dirtiness of the buffers represents the dirtiness of
1084 * subsections of the page. If the page has buffers, the page dirty bit is
1085 * merely a hint about the true dirty state.
1086 *
1087 * When a page is set dirty in its entirety, all its buffers are marked dirty
1088 * (if the page has buffers).
1089 *
1090 * When a buffer is marked dirty, its page is dirtied, but the page's other
1091 * buffers are not.
1092 *
1093 * Also. When blockdev buffers are explicitly read with bread(), they
1094 * individually become uptodate. But their backing page remains not
1095 * uptodate - even if all of its buffers are uptodate. A subsequent
1096 * block_read_full_page() against that page will discover all the uptodate
1097 * buffers, will set the page uptodate and will perform no I/O.
1098 */
1099
1100/**
1101 * mark_buffer_dirty - mark a buffer_head as needing writeout
Martin Waitz67be2dd2005-05-01 08:59:26 -07001102 * @bh: the buffer_head to mark dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001103 *
1104 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1105 * backing page dirty, then tag the page as dirty in its address_space's radix
1106 * tree and then attach the address_space's inode to its superblock's dirty
1107 * inode list.
1108 *
1109 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
Dave Chinner250df6e2011-03-22 22:23:36 +11001110 * mapping->tree_lock and mapping->host->i_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001112void mark_buffer_dirty(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113{
Nick Piggin787d2212007-07-17 04:03:34 -07001114 WARN_ON_ONCE(!buffer_uptodate(bh));
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001115
1116 /*
1117 * Very *carefully* optimize the it-is-already-dirty case.
1118 *
1119 * Don't let the final "is it dirty" escape to before we
1120 * perhaps modified the buffer.
1121 */
1122 if (buffer_dirty(bh)) {
1123 smp_mb();
1124 if (buffer_dirty(bh))
1125 return;
1126 }
1127
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001128 if (!test_set_buffer_dirty(bh)) {
1129 struct page *page = bh->b_page;
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001130 if (!TestSetPageDirty(page)) {
1131 struct address_space *mapping = page_mapping(page);
1132 if (mapping)
1133 __set_page_dirty(page, mapping, 0);
1134 }
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001135 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001137EXPORT_SYMBOL(mark_buffer_dirty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001138
1139/*
1140 * Decrement a buffer_head's reference count. If all buffers against a page
1141 * have zero reference count, are clean and unlocked, and if the page is clean
1142 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1143 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1144 * a page but it ends up not being freed, and buffers may later be reattached).
1145 */
1146void __brelse(struct buffer_head * buf)
1147{
1148 if (atomic_read(&buf->b_count)) {
1149 put_bh(buf);
1150 return;
1151 }
Arjan van de Ven5c752ad2008-07-25 19:45:40 -07001152 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001153}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001154EXPORT_SYMBOL(__brelse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001155
1156/*
1157 * bforget() is like brelse(), except it discards any
1158 * potentially dirty data.
1159 */
1160void __bforget(struct buffer_head *bh)
1161{
1162 clear_buffer_dirty(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -08001163 if (bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001164 struct address_space *buffer_mapping = bh->b_page->mapping;
1165
1166 spin_lock(&buffer_mapping->private_lock);
1167 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -07001168 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001169 spin_unlock(&buffer_mapping->private_lock);
1170 }
1171 __brelse(bh);
1172}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001173EXPORT_SYMBOL(__bforget);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001174
1175static struct buffer_head *__bread_slow(struct buffer_head *bh)
1176{
1177 lock_buffer(bh);
1178 if (buffer_uptodate(bh)) {
1179 unlock_buffer(bh);
1180 return bh;
1181 } else {
1182 get_bh(bh);
1183 bh->b_end_io = end_buffer_read_sync;
1184 submit_bh(READ, bh);
1185 wait_on_buffer(bh);
1186 if (buffer_uptodate(bh))
1187 return bh;
1188 }
1189 brelse(bh);
1190 return NULL;
1191}
1192
1193/*
1194 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1195 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1196 * refcount elevated by one when they're in an LRU. A buffer can only appear
1197 * once in a particular CPU's LRU. A single buffer can be present in multiple
1198 * CPU's LRUs at the same time.
1199 *
1200 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1201 * sb_find_get_block().
1202 *
1203 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1204 * a local interrupt disable for that.
1205 */
1206
1207#define BH_LRU_SIZE 8
1208
1209struct bh_lru {
1210 struct buffer_head *bhs[BH_LRU_SIZE];
1211};
1212
1213static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1214
1215#ifdef CONFIG_SMP
1216#define bh_lru_lock() local_irq_disable()
1217#define bh_lru_unlock() local_irq_enable()
1218#else
1219#define bh_lru_lock() preempt_disable()
1220#define bh_lru_unlock() preempt_enable()
1221#endif
1222
1223static inline void check_irqs_on(void)
1224{
1225#ifdef irqs_disabled
1226 BUG_ON(irqs_disabled());
1227#endif
1228}
1229
1230/*
1231 * The LRU management algorithm is dopey-but-simple. Sorry.
1232 */
1233static void bh_lru_install(struct buffer_head *bh)
1234{
1235 struct buffer_head *evictee = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001236
1237 check_irqs_on();
1238 bh_lru_lock();
Christoph Lameterc7b92512010-12-06 11:16:28 -06001239 if (__this_cpu_read(bh_lrus.bhs[0]) != bh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001240 struct buffer_head *bhs[BH_LRU_SIZE];
1241 int in;
1242 int out = 0;
1243
1244 get_bh(bh);
1245 bhs[out++] = bh;
1246 for (in = 0; in < BH_LRU_SIZE; in++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001247 struct buffer_head *bh2 =
1248 __this_cpu_read(bh_lrus.bhs[in]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001249
1250 if (bh2 == bh) {
1251 __brelse(bh2);
1252 } else {
1253 if (out >= BH_LRU_SIZE) {
1254 BUG_ON(evictee != NULL);
1255 evictee = bh2;
1256 } else {
1257 bhs[out++] = bh2;
1258 }
1259 }
1260 }
1261 while (out < BH_LRU_SIZE)
1262 bhs[out++] = NULL;
Christoph Lameterc7b92512010-12-06 11:16:28 -06001263 memcpy(__this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001264 }
1265 bh_lru_unlock();
1266
1267 if (evictee)
1268 __brelse(evictee);
1269}
1270
1271/*
1272 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1273 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001274static struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001275lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276{
1277 struct buffer_head *ret = NULL;
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001278 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001279
1280 check_irqs_on();
1281 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 for (i = 0; i < BH_LRU_SIZE; i++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001283 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284
1285 if (bh && bh->b_bdev == bdev &&
1286 bh->b_blocknr == block && bh->b_size == size) {
1287 if (i) {
1288 while (i) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001289 __this_cpu_write(bh_lrus.bhs[i],
1290 __this_cpu_read(bh_lrus.bhs[i - 1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 i--;
1292 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06001293 __this_cpu_write(bh_lrus.bhs[0], bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294 }
1295 get_bh(bh);
1296 ret = bh;
1297 break;
1298 }
1299 }
1300 bh_lru_unlock();
1301 return ret;
1302}
1303
1304/*
1305 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1306 * it in the LRU and mark it as accessed. If it is not present then return
1307 * NULL
1308 */
1309struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001310__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311{
1312 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1313
1314 if (bh == NULL) {
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001315 bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 if (bh)
1317 bh_lru_install(bh);
1318 }
1319 if (bh)
1320 touch_buffer(bh);
1321 return bh;
1322}
1323EXPORT_SYMBOL(__find_get_block);
1324
1325/*
1326 * __getblk will locate (and, if necessary, create) the buffer_head
1327 * which corresponds to the passed block_device, block and size. The
1328 * returned buffer has its reference count incremented.
1329 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 * __getblk() will lock up the machine if grow_dev_page's try_to_free_buffers()
1331 * attempt is failing. FIXME, perhaps?
1332 */
1333struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001334__getblk(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335{
1336 struct buffer_head *bh = __find_get_block(bdev, block, size);
1337
1338 might_sleep();
1339 if (bh == NULL)
1340 bh = __getblk_slow(bdev, block, size);
1341 return bh;
1342}
1343EXPORT_SYMBOL(__getblk);
1344
1345/*
1346 * Do async read-ahead on a buffer..
1347 */
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001348void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349{
1350 struct buffer_head *bh = __getblk(bdev, block, size);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001351 if (likely(bh)) {
1352 ll_rw_block(READA, 1, &bh);
1353 brelse(bh);
1354 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355}
1356EXPORT_SYMBOL(__breadahead);
1357
1358/**
1359 * __bread() - reads a specified block and returns the bh
Martin Waitz67be2dd2005-05-01 08:59:26 -07001360 * @bdev: the block_device to read from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * @block: number of block
1362 * @size: size (in bytes) to read
1363 *
1364 * Reads a specified block, and returns buffer head that contains it.
1365 * It returns NULL if the block was unreadable.
1366 */
1367struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001368__bread(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369{
1370 struct buffer_head *bh = __getblk(bdev, block, size);
1371
Andrew Mortona3e713b2005-10-30 15:03:15 -08001372 if (likely(bh) && !buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 bh = __bread_slow(bh);
1374 return bh;
1375}
1376EXPORT_SYMBOL(__bread);
1377
1378/*
1379 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1380 * This doesn't race because it runs in each cpu either in irq
1381 * or with preempt disabled.
1382 */
1383static void invalidate_bh_lru(void *arg)
1384{
1385 struct bh_lru *b = &get_cpu_var(bh_lrus);
1386 int i;
1387
1388 for (i = 0; i < BH_LRU_SIZE; i++) {
1389 brelse(b->bhs[i]);
1390 b->bhs[i] = NULL;
1391 }
1392 put_cpu_var(bh_lrus);
1393}
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001394
1395static bool has_bh_in_lru(int cpu, void *dummy)
1396{
1397 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1398 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001400 for (i = 0; i < BH_LRU_SIZE; i++) {
1401 if (b->bhs[i])
1402 return 1;
1403 }
1404
1405 return 0;
1406}
1407
Peter Zijlstraf9a14392007-05-06 14:49:55 -07001408void invalidate_bh_lrus(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409{
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001410 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411}
Nick Piggin9db55792008-02-08 04:19:49 -08001412EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413
1414void set_bh_page(struct buffer_head *bh,
1415 struct page *page, unsigned long offset)
1416{
1417 bh->b_page = page;
Eric Sesterhenne827f922006-03-26 18:24:46 +02001418 BUG_ON(offset >= PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419 if (PageHighMem(page))
1420 /*
1421 * This catches illegal uses and preserves the offset:
1422 */
1423 bh->b_data = (char *)(0 + offset);
1424 else
1425 bh->b_data = page_address(page) + offset;
1426}
1427EXPORT_SYMBOL(set_bh_page);
1428
1429/*
1430 * Called when truncating a buffer on a page completely.
1431 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001432static void discard_buffer(struct buffer_head * bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433{
1434 lock_buffer(bh);
1435 clear_buffer_dirty(bh);
1436 bh->b_bdev = NULL;
1437 clear_buffer_mapped(bh);
1438 clear_buffer_req(bh);
1439 clear_buffer_new(bh);
1440 clear_buffer_delay(bh);
David Chinner33a266d2007-02-12 00:51:41 -08001441 clear_buffer_unwritten(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 unlock_buffer(bh);
1443}
1444
1445/**
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001446 * block_invalidatepage - invalidate part or all of a buffer-backed page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 *
1448 * @page: the page which is affected
1449 * @offset: the index of the truncation point
1450 *
1451 * block_invalidatepage() is called when all or part of the page has become
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001452 * invalidated by a truncate operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 *
1454 * block_invalidatepage() does not have to release all buffers, but it must
1455 * ensure that no dirty buffer is left outside @offset and that no I/O
1456 * is underway against any of the blocks which are outside the truncation
1457 * point. Because the caller is about to free (and possibly reuse) those
1458 * blocks on-disk.
1459 */
NeilBrown2ff28e22006-03-26 01:37:18 -08001460void block_invalidatepage(struct page *page, unsigned long offset)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461{
1462 struct buffer_head *head, *bh, *next;
1463 unsigned int curr_off = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465 BUG_ON(!PageLocked(page));
1466 if (!page_has_buffers(page))
1467 goto out;
1468
1469 head = page_buffers(page);
1470 bh = head;
1471 do {
1472 unsigned int next_off = curr_off + bh->b_size;
1473 next = bh->b_this_page;
1474
1475 /*
1476 * is this block fully invalidated?
1477 */
1478 if (offset <= curr_off)
1479 discard_buffer(bh);
1480 curr_off = next_off;
1481 bh = next;
1482 } while (bh != head);
1483
1484 /*
1485 * We release buffers only if the entire page is being invalidated.
1486 * The get_block cached value has been unconditionally invalidated,
1487 * so real IO is not possible anymore.
1488 */
1489 if (offset == 0)
NeilBrown2ff28e22006-03-26 01:37:18 -08001490 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491out:
NeilBrown2ff28e22006-03-26 01:37:18 -08001492 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493}
1494EXPORT_SYMBOL(block_invalidatepage);
1495
1496/*
1497 * We attach and possibly dirty the buffers atomically wrt
1498 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1499 * is already excluded via the page lock.
1500 */
1501void create_empty_buffers(struct page *page,
1502 unsigned long blocksize, unsigned long b_state)
1503{
1504 struct buffer_head *bh, *head, *tail;
1505
1506 head = alloc_page_buffers(page, blocksize, 1);
1507 bh = head;
1508 do {
1509 bh->b_state |= b_state;
1510 tail = bh;
1511 bh = bh->b_this_page;
1512 } while (bh);
1513 tail->b_this_page = head;
1514
1515 spin_lock(&page->mapping->private_lock);
1516 if (PageUptodate(page) || PageDirty(page)) {
1517 bh = head;
1518 do {
1519 if (PageDirty(page))
1520 set_buffer_dirty(bh);
1521 if (PageUptodate(page))
1522 set_buffer_uptodate(bh);
1523 bh = bh->b_this_page;
1524 } while (bh != head);
1525 }
1526 attach_page_buffers(page, head);
1527 spin_unlock(&page->mapping->private_lock);
1528}
1529EXPORT_SYMBOL(create_empty_buffers);
1530
1531/*
1532 * We are taking a block for data and we don't want any output from any
1533 * buffer-cache aliases starting from return from that function and
1534 * until the moment when something will explicitly mark the buffer
1535 * dirty (hopefully that will not happen until we will free that block ;-)
1536 * We don't even need to mark it not-uptodate - nobody can expect
1537 * anything from a newly allocated buffer anyway. We used to used
1538 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1539 * don't want to mark the alias unmapped, for example - it would confuse
1540 * anyone who might pick it with bread() afterwards...
1541 *
1542 * Also.. Note that bforget() doesn't lock the buffer. So there can
1543 * be writeout I/O going on against recently-freed buffers. We don't
1544 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1545 * only if we really need to. That happens here.
1546 */
1547void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
1548{
1549 struct buffer_head *old_bh;
1550
1551 might_sleep();
1552
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001553 old_bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001554 if (old_bh) {
1555 clear_buffer_dirty(old_bh);
1556 wait_on_buffer(old_bh);
1557 clear_buffer_req(old_bh);
1558 __brelse(old_bh);
1559 }
1560}
1561EXPORT_SYMBOL(unmap_underlying_metadata);
1562
1563/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001564 * Size is a power-of-two in the range 512..PAGE_SIZE,
1565 * and the case we care about most is PAGE_SIZE.
1566 *
1567 * So this *could* possibly be written with those
1568 * constraints in mind (relevant mostly if some
1569 * architecture has a slow bit-scan instruction)
1570 */
1571static inline int block_size_bits(unsigned int blocksize)
1572{
1573 return ilog2(blocksize);
1574}
1575
1576static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1577{
1578 BUG_ON(!PageLocked(page));
1579
1580 if (!page_has_buffers(page))
1581 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1582 return page_buffers(page);
1583}
1584
1585/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586 * NOTE! All mapped/uptodate combinations are valid:
1587 *
1588 * Mapped Uptodate Meaning
1589 *
1590 * No No "unknown" - must do get_block()
1591 * No Yes "hole" - zero-filled
1592 * Yes No "allocated" - allocated on disk, not read in
1593 * Yes Yes "valid" - allocated and up-to-date in memory.
1594 *
1595 * "Dirty" is valid only with the last case (mapped+uptodate).
1596 */
1597
1598/*
1599 * While block_write_full_page is writing back the dirty buffers under
1600 * the page lock, whoever dirtied the buffers may decide to clean them
1601 * again at any time. We handle that by only looking at the buffer
1602 * state inside lock_buffer().
1603 *
1604 * If block_write_full_page() is called for regular writeback
1605 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1606 * locked buffer. This only can happen if someone has written the buffer
1607 * directly, with submit_bh(). At the address_space level PageWriteback
1608 * prevents this contention from occurring.
Theodore Ts'o6e34eed2009-04-07 18:12:43 -04001609 *
1610 * If block_write_full_page() is called with wbc->sync_mode ==
Jens Axboe721a9602011-03-09 11:56:30 +01001611 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1612 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 */
1614static int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001615 get_block_t *get_block, struct writeback_control *wbc,
1616 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001617{
1618 int err;
1619 sector_t block;
1620 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001621 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001622 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 int nr_underway = 0;
Theodore Ts'o6e34eed2009-04-07 18:12:43 -04001624 int write_op = (wbc->sync_mode == WB_SYNC_ALL ?
Jens Axboe721a9602011-03-09 11:56:30 +01001625 WRITE_SYNC : WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001626
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001627 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629
1630 /*
1631 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1632 * here, and the (potentially unmapped) buffers may become dirty at
1633 * any time. If a buffer becomes dirty here after we've inspected it
1634 * then we just miss that fact, and the page stays dirty.
1635 *
1636 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1637 * handle that here by just cleaning them.
1638 */
1639
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001641 blocksize = bh->b_size;
1642 bbits = block_size_bits(blocksize);
1643
1644 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1645 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646
1647 /*
1648 * Get all the dirty buffers mapped to disk addresses and
1649 * handle any aliases from the underlying blockdev's mapping.
1650 */
1651 do {
1652 if (block > last_block) {
1653 /*
1654 * mapped buffers outside i_size will occur, because
1655 * this page can be outside i_size when there is a
1656 * truncate in progress.
1657 */
1658 /*
1659 * The buffer was zeroed by block_write_full_page()
1660 */
1661 clear_buffer_dirty(bh);
1662 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001663 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1664 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001665 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001666 err = get_block(inode, block, bh, 1);
1667 if (err)
1668 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001669 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 if (buffer_new(bh)) {
1671 /* blockdev mappings never come here */
1672 clear_buffer_new(bh);
1673 unmap_underlying_metadata(bh->b_bdev,
1674 bh->b_blocknr);
1675 }
1676 }
1677 bh = bh->b_this_page;
1678 block++;
1679 } while (bh != head);
1680
1681 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682 if (!buffer_mapped(bh))
1683 continue;
1684 /*
1685 * If it's a fully non-blocking write attempt and we cannot
1686 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001687 * potentially cause a busy-wait loop from writeback threads
1688 * and kswapd activity, but those code paths have their own
1689 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001691 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001693 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694 redirty_page_for_writepage(wbc, page);
1695 continue;
1696 }
1697 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001698 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 } else {
1700 unlock_buffer(bh);
1701 }
1702 } while ((bh = bh->b_this_page) != head);
1703
1704 /*
1705 * The page and its buffers are protected by PageWriteback(), so we can
1706 * drop the bh refcounts early.
1707 */
1708 BUG_ON(PageWriteback(page));
1709 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710
1711 do {
1712 struct buffer_head *next = bh->b_this_page;
1713 if (buffer_async_write(bh)) {
Theodore Ts'oa64c8612009-03-27 22:14:10 -04001714 submit_bh(write_op, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 nr_underway++;
1716 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717 bh = next;
1718 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001719 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
1721 err = 0;
1722done:
1723 if (nr_underway == 0) {
1724 /*
1725 * The page was marked dirty, but the buffers were
1726 * clean. Someone wrote them back by hand with
1727 * ll_rw_block/submit_bh. A rare case.
1728 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001730
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 /*
1732 * The page and buffer_heads can be released at any time from
1733 * here on.
1734 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001735 }
1736 return err;
1737
1738recover:
1739 /*
1740 * ENOSPC, or some other error. We may already have added some
1741 * blocks to the file, so we need to write these out to avoid
1742 * exposing stale data.
1743 * The page is currently locked and not marked for writeback
1744 */
1745 bh = head;
1746 /* Recovery: lock and submit the mapped buffers */
1747 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001748 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1749 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001751 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 } else {
1753 /*
1754 * The buffer may have been set dirty during
1755 * attachment to a dirty page.
1756 */
1757 clear_buffer_dirty(bh);
1758 }
1759 } while ((bh = bh->b_this_page) != head);
1760 SetPageError(page);
1761 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001762 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 do {
1765 struct buffer_head *next = bh->b_this_page;
1766 if (buffer_async_write(bh)) {
1767 clear_buffer_dirty(bh);
Theodore Ts'oa64c8612009-03-27 22:14:10 -04001768 submit_bh(write_op, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 nr_underway++;
1770 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771 bh = next;
1772 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001773 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 goto done;
1775}
1776
Nick Pigginafddba42007-10-16 01:25:01 -07001777/*
1778 * If a page has any new buffers, zero them out here, and mark them uptodate
1779 * and dirty so they'll be written out (in order to prevent uninitialised
1780 * block data from leaking). And clear the new bit.
1781 */
1782void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1783{
1784 unsigned int block_start, block_end;
1785 struct buffer_head *head, *bh;
1786
1787 BUG_ON(!PageLocked(page));
1788 if (!page_has_buffers(page))
1789 return;
1790
1791 bh = head = page_buffers(page);
1792 block_start = 0;
1793 do {
1794 block_end = block_start + bh->b_size;
1795
1796 if (buffer_new(bh)) {
1797 if (block_end > from && block_start < to) {
1798 if (!PageUptodate(page)) {
1799 unsigned start, size;
1800
1801 start = max(from, block_start);
1802 size = min(to, block_end) - start;
1803
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001804 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001805 set_buffer_uptodate(bh);
1806 }
1807
1808 clear_buffer_new(bh);
1809 mark_buffer_dirty(bh);
1810 }
1811 }
1812
1813 block_start = block_end;
1814 bh = bh->b_this_page;
1815 } while (bh != head);
1816}
1817EXPORT_SYMBOL(page_zero_new_buffers);
1818
Christoph Hellwigebdec242010-10-06 10:47:23 +02001819int __block_write_begin(struct page *page, loff_t pos, unsigned len,
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001820 get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821{
Christoph Hellwigebdec242010-10-06 10:47:23 +02001822 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1823 unsigned to = from + len;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001824 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 unsigned block_start, block_end;
1826 sector_t block;
1827 int err = 0;
1828 unsigned blocksize, bbits;
1829 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1830
1831 BUG_ON(!PageLocked(page));
1832 BUG_ON(from > PAGE_CACHE_SIZE);
1833 BUG_ON(to > PAGE_CACHE_SIZE);
1834 BUG_ON(from > to);
1835
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001836 head = create_page_buffers(page, inode, 0);
1837 blocksize = head->b_size;
1838 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1841
1842 for(bh = head, block_start = 0; bh != head || !block_start;
1843 block++, block_start=block_end, bh = bh->b_this_page) {
1844 block_end = block_start + blocksize;
1845 if (block_end <= from || block_start >= to) {
1846 if (PageUptodate(page)) {
1847 if (!buffer_uptodate(bh))
1848 set_buffer_uptodate(bh);
1849 }
1850 continue;
1851 }
1852 if (buffer_new(bh))
1853 clear_buffer_new(bh);
1854 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001855 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 err = get_block(inode, block, bh, 1);
1857 if (err)
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07001858 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001859 if (buffer_new(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 unmap_underlying_metadata(bh->b_bdev,
1861 bh->b_blocknr);
1862 if (PageUptodate(page)) {
Nick Piggin637aff42007-10-16 01:25:00 -07001863 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07001865 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866 continue;
1867 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001868 if (block_end > to || block_start < from)
1869 zero_user_segments(page,
1870 to, block_end,
1871 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872 continue;
1873 }
1874 }
1875 if (PageUptodate(page)) {
1876 if (!buffer_uptodate(bh))
1877 set_buffer_uptodate(bh);
1878 continue;
1879 }
1880 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08001881 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 (block_start < from || block_end > to)) {
1883 ll_rw_block(READ, 1, &bh);
1884 *wait_bh++=bh;
1885 }
1886 }
1887 /*
1888 * If we issued read requests - let them complete.
1889 */
1890 while(wait_bh > wait) {
1891 wait_on_buffer(*--wait_bh);
1892 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07001893 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 }
Jan Karaf9f07b62011-06-14 00:58:27 +02001895 if (unlikely(err))
Nick Pigginafddba42007-10-16 01:25:01 -07001896 page_zero_new_buffers(page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001897 return err;
1898}
Christoph Hellwigebdec242010-10-06 10:47:23 +02001899EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900
1901static int __block_commit_write(struct inode *inode, struct page *page,
1902 unsigned from, unsigned to)
1903{
1904 unsigned block_start, block_end;
1905 int partial = 0;
1906 unsigned blocksize;
1907 struct buffer_head *bh, *head;
1908
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001909 bh = head = page_buffers(page);
1910 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001912 block_start = 0;
1913 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914 block_end = block_start + blocksize;
1915 if (block_end <= from || block_start >= to) {
1916 if (!buffer_uptodate(bh))
1917 partial = 1;
1918 } else {
1919 set_buffer_uptodate(bh);
1920 mark_buffer_dirty(bh);
1921 }
Nick Pigginafddba42007-10-16 01:25:01 -07001922 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001923
1924 block_start = block_end;
1925 bh = bh->b_this_page;
1926 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
1928 /*
1929 * If this is a partial write which happened to make all buffers
1930 * uptodate then we can optimize away a bogus readpage() for
1931 * the next read(). Here we 'discover' whether the page went
1932 * uptodate as a result of this (potentially partial) write.
1933 */
1934 if (!partial)
1935 SetPageUptodate(page);
1936 return 0;
1937}
1938
1939/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02001940 * block_write_begin takes care of the basic task of block allocation and
1941 * bringing partial write blocks uptodate first.
1942 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10001943 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07001944 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02001945int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
1946 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07001947{
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001948 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07001949 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001950 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07001951
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001952 page = grab_cache_page_write_begin(mapping, index, flags);
1953 if (!page)
1954 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07001955
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001956 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07001957 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001958 unlock_page(page);
1959 page_cache_release(page);
1960 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07001961 }
1962
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001963 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07001964 return status;
1965}
1966EXPORT_SYMBOL(block_write_begin);
1967
1968int block_write_end(struct file *file, struct address_space *mapping,
1969 loff_t pos, unsigned len, unsigned copied,
1970 struct page *page, void *fsdata)
1971{
1972 struct inode *inode = mapping->host;
1973 unsigned start;
1974
1975 start = pos & (PAGE_CACHE_SIZE - 1);
1976
1977 if (unlikely(copied < len)) {
1978 /*
1979 * The buffers that were written will now be uptodate, so we
1980 * don't have to worry about a readpage reading them and
1981 * overwriting a partial write. However if we have encountered
1982 * a short write and only partially written into a buffer, it
1983 * will not be marked uptodate, so a readpage might come in and
1984 * destroy our partial write.
1985 *
1986 * Do the simplest thing, and just treat any short write to a
1987 * non uptodate page as a zero-length write, and force the
1988 * caller to redo the whole thing.
1989 */
1990 if (!PageUptodate(page))
1991 copied = 0;
1992
1993 page_zero_new_buffers(page, start+copied, start+len);
1994 }
1995 flush_dcache_page(page);
1996
1997 /* This could be a short (even 0-length) commit */
1998 __block_commit_write(inode, page, start, start+copied);
1999
2000 return copied;
2001}
2002EXPORT_SYMBOL(block_write_end);
2003
2004int generic_write_end(struct file *file, struct address_space *mapping,
2005 loff_t pos, unsigned len, unsigned copied,
2006 struct page *page, void *fsdata)
2007{
2008 struct inode *inode = mapping->host;
Jan Karac7d206b2008-07-11 19:27:31 -04002009 int i_size_changed = 0;
Nick Pigginafddba42007-10-16 01:25:01 -07002010
2011 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2012
2013 /*
2014 * No need to use i_size_read() here, the i_size
2015 * cannot change under us because we hold i_mutex.
2016 *
2017 * But it's important to update i_size while still holding page lock:
2018 * page writeout could otherwise come in and zero beyond i_size.
2019 */
2020 if (pos+copied > inode->i_size) {
2021 i_size_write(inode, pos+copied);
Jan Karac7d206b2008-07-11 19:27:31 -04002022 i_size_changed = 1;
Nick Pigginafddba42007-10-16 01:25:01 -07002023 }
2024
2025 unlock_page(page);
2026 page_cache_release(page);
2027
Jan Karac7d206b2008-07-11 19:27:31 -04002028 /*
2029 * Don't mark the inode dirty under page lock. First, it unnecessarily
2030 * makes the holding time of page lock longer. Second, it forces lock
2031 * ordering of page lock and transaction start for journaling
2032 * filesystems.
2033 */
2034 if (i_size_changed)
2035 mark_inode_dirty(inode);
2036
Nick Pigginafddba42007-10-16 01:25:01 -07002037 return copied;
2038}
2039EXPORT_SYMBOL(generic_write_end);
2040
2041/*
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002042 * block_is_partially_uptodate checks whether buffers within a page are
2043 * uptodate or not.
2044 *
2045 * Returns true if all buffers which correspond to a file portion
2046 * we want to read are uptodate.
2047 */
2048int block_is_partially_uptodate(struct page *page, read_descriptor_t *desc,
2049 unsigned long from)
2050{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002051 unsigned block_start, block_end, blocksize;
2052 unsigned to;
2053 struct buffer_head *bh, *head;
2054 int ret = 1;
2055
2056 if (!page_has_buffers(page))
2057 return 0;
2058
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002059 head = page_buffers(page);
2060 blocksize = head->b_size;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002061 to = min_t(unsigned, PAGE_CACHE_SIZE - from, desc->count);
2062 to = from + to;
2063 if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize)
2064 return 0;
2065
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002066 bh = head;
2067 block_start = 0;
2068 do {
2069 block_end = block_start + blocksize;
2070 if (block_end > from && block_start < to) {
2071 if (!buffer_uptodate(bh)) {
2072 ret = 0;
2073 break;
2074 }
2075 if (block_end >= to)
2076 break;
2077 }
2078 block_start = block_end;
2079 bh = bh->b_this_page;
2080 } while (bh != head);
2081
2082 return ret;
2083}
2084EXPORT_SYMBOL(block_is_partially_uptodate);
2085
2086/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002087 * Generic "read page" function for block devices that have the normal
2088 * get_block functionality. This is most of the block device filesystems.
2089 * Reads the page asynchronously --- the unlock_buffer() and
2090 * set/clear_buffer_uptodate() functions propagate buffer state into the
2091 * page struct once IO has completed.
2092 */
2093int block_read_full_page(struct page *page, get_block_t *get_block)
2094{
2095 struct inode *inode = page->mapping->host;
2096 sector_t iblock, lblock;
2097 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002098 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002099 int nr, i;
2100 int fully_mapped = 1;
2101
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002102 head = create_page_buffers(page, inode, 0);
2103 blocksize = head->b_size;
2104 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002106 iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
2107 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002108 bh = head;
2109 nr = 0;
2110 i = 0;
2111
2112 do {
2113 if (buffer_uptodate(bh))
2114 continue;
2115
2116 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002117 int err = 0;
2118
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 fully_mapped = 0;
2120 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002121 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002122 err = get_block(inode, iblock, bh, 0);
2123 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002124 SetPageError(page);
2125 }
2126 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002127 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002128 if (!err)
2129 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002130 continue;
2131 }
2132 /*
2133 * get_block() might have updated the buffer
2134 * synchronously
2135 */
2136 if (buffer_uptodate(bh))
2137 continue;
2138 }
2139 arr[nr++] = bh;
2140 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2141
2142 if (fully_mapped)
2143 SetPageMappedToDisk(page);
2144
2145 if (!nr) {
2146 /*
2147 * All buffers are uptodate - we can set the page uptodate
2148 * as well. But not if get_block() returned an error.
2149 */
2150 if (!PageError(page))
2151 SetPageUptodate(page);
2152 unlock_page(page);
2153 return 0;
2154 }
2155
2156 /* Stage two: lock the buffers */
2157 for (i = 0; i < nr; i++) {
2158 bh = arr[i];
2159 lock_buffer(bh);
2160 mark_buffer_async_read(bh);
2161 }
2162
2163 /*
2164 * Stage 3: start the IO. Check for uptodateness
2165 * inside the buffer lock in case another process reading
2166 * the underlying blockdev brought it uptodate (the sct fix).
2167 */
2168 for (i = 0; i < nr; i++) {
2169 bh = arr[i];
2170 if (buffer_uptodate(bh))
2171 end_buffer_async_read(bh, 1);
2172 else
2173 submit_bh(READ, bh);
2174 }
2175 return 0;
2176}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002177EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002178
2179/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002180 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002181 * deal with the hole.
2182 */
Nick Piggin89e10782007-10-16 01:25:07 -07002183int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002184{
2185 struct address_space *mapping = inode->i_mapping;
2186 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002187 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 int err;
2189
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002190 err = inode_newsize_ok(inode, size);
2191 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002192 goto out;
2193
Nick Piggin89e10782007-10-16 01:25:07 -07002194 err = pagecache_write_begin(NULL, mapping, size, 0,
2195 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2196 &page, &fsdata);
2197 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002198 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002199
Nick Piggin89e10782007-10-16 01:25:07 -07002200 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2201 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002202
Linus Torvalds1da177e2005-04-16 15:20:36 -07002203out:
2204 return err;
2205}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002206EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002208static int cont_expand_zero(struct file *file, struct address_space *mapping,
2209 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002210{
Nick Piggin89e10782007-10-16 01:25:07 -07002211 struct inode *inode = mapping->host;
2212 unsigned blocksize = 1 << inode->i_blkbits;
2213 struct page *page;
2214 void *fsdata;
2215 pgoff_t index, curidx;
2216 loff_t curpos;
2217 unsigned zerofrom, offset, len;
2218 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002219
Nick Piggin89e10782007-10-16 01:25:07 -07002220 index = pos >> PAGE_CACHE_SHIFT;
2221 offset = pos & ~PAGE_CACHE_MASK;
2222
2223 while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) {
2224 zerofrom = curpos & ~PAGE_CACHE_MASK;
2225 if (zerofrom & (blocksize-1)) {
2226 *bytes |= (blocksize-1);
2227 (*bytes)++;
2228 }
2229 len = PAGE_CACHE_SIZE - zerofrom;
2230
2231 err = pagecache_write_begin(file, mapping, curpos, len,
2232 AOP_FLAG_UNINTERRUPTIBLE,
2233 &page, &fsdata);
2234 if (err)
2235 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002236 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002237 err = pagecache_write_end(file, mapping, curpos, len, len,
2238 page, fsdata);
2239 if (err < 0)
2240 goto out;
2241 BUG_ON(err != len);
2242 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002243
2244 balance_dirty_pages_ratelimited(mapping);
Nick Piggin89e10782007-10-16 01:25:07 -07002245 }
2246
2247 /* page covers the boundary, find the boundary offset */
2248 if (index == curidx) {
2249 zerofrom = curpos & ~PAGE_CACHE_MASK;
2250 /* if we will expand the thing last block will be filled */
2251 if (offset <= zerofrom) {
2252 goto out;
2253 }
2254 if (zerofrom & (blocksize-1)) {
2255 *bytes |= (blocksize-1);
2256 (*bytes)++;
2257 }
2258 len = offset - zerofrom;
2259
2260 err = pagecache_write_begin(file, mapping, curpos, len,
2261 AOP_FLAG_UNINTERRUPTIBLE,
2262 &page, &fsdata);
2263 if (err)
2264 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002265 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002266 err = pagecache_write_end(file, mapping, curpos, len, len,
2267 page, fsdata);
2268 if (err < 0)
2269 goto out;
2270 BUG_ON(err != len);
2271 err = 0;
2272 }
2273out:
2274 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002275}
2276
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277/*
2278 * For moronic filesystems that do not allow holes in file.
2279 * We may have to extend the file.
2280 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002281int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002282 loff_t pos, unsigned len, unsigned flags,
2283 struct page **pagep, void **fsdata,
2284 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002286 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 unsigned blocksize = 1 << inode->i_blkbits;
Nick Piggin89e10782007-10-16 01:25:07 -07002288 unsigned zerofrom;
2289 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290
Nick Piggin89e10782007-10-16 01:25:07 -07002291 err = cont_expand_zero(file, mapping, pos, bytes);
2292 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002293 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002294
2295 zerofrom = *bytes & ~PAGE_CACHE_MASK;
2296 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2297 *bytes |= (blocksize-1);
2298 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002299 }
2300
Christoph Hellwig155130a2010-06-04 11:29:58 +02002301 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002302}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002303EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305int block_commit_write(struct page *page, unsigned from, unsigned to)
2306{
2307 struct inode *inode = page->mapping->host;
2308 __block_commit_write(inode,page,from,to);
2309 return 0;
2310}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002311EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002312
David Chinner54171692007-07-19 17:39:55 +10002313/*
2314 * block_page_mkwrite() is not allowed to change the file size as it gets
2315 * called from a page fault handler when a page is first dirtied. Hence we must
2316 * be careful to check for EOF conditions here. We set the page up correctly
2317 * for a written page which means we get ENOSPC checking when writing into
2318 * holes and correct delalloc and unwritten extent mapping on filesystems that
2319 * support these features.
2320 *
2321 * We are not allowed to take the i_mutex here so we have to play games to
2322 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002323 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002324 * page lock we can determine safely if the page is beyond EOF. If it is not
2325 * beyond EOF, then the page is guaranteed safe against truncation until we
2326 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002327 *
Jan Kara14da9202012-06-12 16:20:37 +02002328 * Direct callers of this function should protect against filesystem freezing
2329 * using sb_start_write() - sb_end_write() functions.
David Chinner54171692007-07-19 17:39:55 +10002330 */
Jan Kara24da4fa2011-05-24 00:23:34 +02002331int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2332 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002333{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002334 struct page *page = vmf->page;
David Chinner54171692007-07-19 17:39:55 +10002335 struct inode *inode = vma->vm_file->f_path.dentry->d_inode;
2336 unsigned long end;
2337 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002338 int ret;
David Chinner54171692007-07-19 17:39:55 +10002339
2340 lock_page(page);
2341 size = i_size_read(inode);
2342 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002343 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002344 /* We overload EFAULT to mean page got truncated */
2345 ret = -EFAULT;
2346 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002347 }
2348
2349 /* page is wholly or partially inside EOF */
2350 if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
2351 end = size & ~PAGE_CACHE_MASK;
2352 else
2353 end = PAGE_CACHE_SIZE;
2354
Christoph Hellwigebdec242010-10-06 10:47:23 +02002355 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002356 if (!ret)
2357 ret = block_commit_write(page, 0, end);
2358
Jan Kara24da4fa2011-05-24 00:23:34 +02002359 if (unlikely(ret < 0))
2360 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002361 set_page_dirty(page);
Darrick J. Wongd76ee182011-05-27 12:23:41 -07002362 wait_on_page_writeback(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002363 return 0;
2364out_unlock:
2365 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002366 return ret;
2367}
Jan Kara24da4fa2011-05-24 00:23:34 +02002368EXPORT_SYMBOL(__block_page_mkwrite);
2369
2370int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2371 get_block_t get_block)
2372{
Jan Karaea13a862011-05-24 00:23:35 +02002373 int ret;
2374 struct super_block *sb = vma->vm_file->f_path.dentry->d_inode->i_sb;
Jan Kara24da4fa2011-05-24 00:23:34 +02002375
Jan Kara14da9202012-06-12 16:20:37 +02002376 sb_start_pagefault(sb);
Theodore Ts'o041bbb6d2012-09-30 23:04:56 -04002377
2378 /*
2379 * Update file times before taking page lock. We may end up failing the
2380 * fault so this update may be superfluous but who really cares...
2381 */
2382 file_update_time(vma->vm_file);
2383
Jan Karaea13a862011-05-24 00:23:35 +02002384 ret = __block_page_mkwrite(vma, vmf, get_block);
Jan Kara14da9202012-06-12 16:20:37 +02002385 sb_end_pagefault(sb);
Jan Kara24da4fa2011-05-24 00:23:34 +02002386 return block_page_mkwrite_return(ret);
2387}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002388EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389
2390/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002391 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002392 * immediately, while under the page lock. So it needs a special end_io
2393 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 */
2395static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2396{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002397 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398}
2399
2400/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002401 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2402 * the page (converting it to circular linked list and taking care of page
2403 * dirty races).
2404 */
2405static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2406{
2407 struct buffer_head *bh;
2408
2409 BUG_ON(!PageLocked(page));
2410
2411 spin_lock(&page->mapping->private_lock);
2412 bh = head;
2413 do {
2414 if (PageDirty(page))
2415 set_buffer_dirty(bh);
2416 if (!bh->b_this_page)
2417 bh->b_this_page = head;
2418 bh = bh->b_this_page;
2419 } while (bh != head);
2420 attach_page_buffers(page, head);
2421 spin_unlock(&page->mapping->private_lock);
2422}
2423
2424/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002425 * On entry, the page is fully not uptodate.
2426 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002427 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002428 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002429int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002430 loff_t pos, unsigned len, unsigned flags,
2431 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002432 get_block_t *get_block)
2433{
Nick Piggin03158cd2007-10-16 01:25:25 -07002434 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435 const unsigned blkbits = inode->i_blkbits;
2436 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002437 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002438 struct page *page;
2439 pgoff_t index;
2440 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002441 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002442 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002443 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002444 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002445 int ret = 0;
2446 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447
Nick Piggin03158cd2007-10-16 01:25:25 -07002448 index = pos >> PAGE_CACHE_SHIFT;
2449 from = pos & (PAGE_CACHE_SIZE - 1);
2450 to = from + len;
2451
Nick Piggin54566b22009-01-04 12:00:53 -08002452 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002453 if (!page)
2454 return -ENOMEM;
2455 *pagep = page;
2456 *fsdata = NULL;
2457
2458 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002459 ret = __block_write_begin(page, pos, len, get_block);
2460 if (unlikely(ret))
2461 goto out_release;
2462 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002463 }
Nick Piggina4b06722007-10-16 01:24:48 -07002464
Linus Torvalds1da177e2005-04-16 15:20:36 -07002465 if (PageMappedToDisk(page))
2466 return 0;
2467
Nick Piggina4b06722007-10-16 01:24:48 -07002468 /*
2469 * Allocate buffers so that we can keep track of state, and potentially
2470 * attach them to the page if an error occurs. In the common case of
2471 * no error, they will just be freed again without ever being attached
2472 * to the page (which is all OK, because we're under the page lock).
2473 *
2474 * Be careful: the buffer linked list is a NULL terminated one, rather
2475 * than the circular one we're used to.
2476 */
2477 head = alloc_page_buffers(page, blocksize, 0);
Nick Piggin03158cd2007-10-16 01:25:25 -07002478 if (!head) {
2479 ret = -ENOMEM;
2480 goto out_release;
2481 }
Nick Piggina4b06722007-10-16 01:24:48 -07002482
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002484
2485 /*
2486 * We loop across all blocks in the page, whether or not they are
2487 * part of the affected region. This is so we can discover if the
2488 * page is fully mapped-to-disk.
2489 */
Nick Piggina4b06722007-10-16 01:24:48 -07002490 for (block_start = 0, block_in_page = 0, bh = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 block_start < PAGE_CACHE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002492 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493 int create;
2494
Nick Piggina4b06722007-10-16 01:24:48 -07002495 block_end = block_start + blocksize;
2496 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002497 create = 1;
2498 if (block_start >= to)
2499 create = 0;
2500 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002501 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002502 if (ret)
2503 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002504 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002506 if (buffer_new(bh))
2507 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
2508 if (PageUptodate(page)) {
2509 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002511 }
2512 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002513 zero_user_segments(page, block_start, from,
2514 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002515 continue;
2516 }
Nick Piggina4b06722007-10-16 01:24:48 -07002517 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518 continue; /* reiserfs does this */
2519 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002520 lock_buffer(bh);
2521 bh->b_end_io = end_buffer_read_nobh;
2522 submit_bh(READ, bh);
2523 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 }
2525 }
2526
2527 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 /*
2529 * The page is locked, so these buffers are protected from
2530 * any VM or truncate activity. Hence we don't need to care
2531 * for the buffer_head refcounts.
2532 */
Nick Piggina4b06722007-10-16 01:24:48 -07002533 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 wait_on_buffer(bh);
2535 if (!buffer_uptodate(bh))
2536 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 }
2538 if (ret)
2539 goto failed;
2540 }
2541
2542 if (is_mapped_to_disk)
2543 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002544
Nick Piggin03158cd2007-10-16 01:25:25 -07002545 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002546
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547 return 0;
2548
2549failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002550 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002551 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002552 * Error recovery is a bit difficult. We need to zero out blocks that
2553 * were newly allocated, and dirty them to ensure they get written out.
2554 * Buffers need to be attached to the page at this point, otherwise
2555 * the handling of potential IO errors during writeout would be hard
2556 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002558 attach_nobh_buffers(page, head);
2559 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002560
Nick Piggin03158cd2007-10-16 01:25:25 -07002561out_release:
2562 unlock_page(page);
2563 page_cache_release(page);
2564 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002565
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002566 return ret;
2567}
Nick Piggin03158cd2007-10-16 01:25:25 -07002568EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Nick Piggin03158cd2007-10-16 01:25:25 -07002570int nobh_write_end(struct file *file, struct address_space *mapping,
2571 loff_t pos, unsigned len, unsigned copied,
2572 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573{
2574 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002575 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002576 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002577 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002579 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002580 attach_nobh_buffers(page, head);
2581 if (page_has_buffers(page))
2582 return generic_write_end(file, mapping, pos, len,
2583 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002584
Nick Piggin22c8ca72007-02-20 13:58:09 -08002585 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002587 if (pos+copied > inode->i_size) {
2588 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 mark_inode_dirty(inode);
2590 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002591
2592 unlock_page(page);
2593 page_cache_release(page);
2594
Nick Piggin03158cd2007-10-16 01:25:25 -07002595 while (head) {
2596 bh = head;
2597 head = head->b_this_page;
2598 free_buffer_head(bh);
2599 }
2600
2601 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002602}
Nick Piggin03158cd2007-10-16 01:25:25 -07002603EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604
2605/*
2606 * nobh_writepage() - based on block_full_write_page() except
2607 * that it tries to operate without attaching bufferheads to
2608 * the page.
2609 */
2610int nobh_writepage(struct page *page, get_block_t *get_block,
2611 struct writeback_control *wbc)
2612{
2613 struct inode * const inode = page->mapping->host;
2614 loff_t i_size = i_size_read(inode);
2615 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2616 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617 int ret;
2618
2619 /* Is the page fully inside i_size? */
2620 if (page->index < end_index)
2621 goto out;
2622
2623 /* Is the page fully outside i_size? (truncate in progress) */
2624 offset = i_size & (PAGE_CACHE_SIZE-1);
2625 if (page->index >= end_index+1 || !offset) {
2626 /*
2627 * The page may have dirty, unmapped buffers. For example,
2628 * they may have been added in ext3_writepage(). Make them
2629 * freeable here, so the page does not leak.
2630 */
2631#if 0
2632 /* Not really sure about this - do we need this ? */
2633 if (page->mapping->a_ops->invalidatepage)
2634 page->mapping->a_ops->invalidatepage(page, offset);
2635#endif
2636 unlock_page(page);
2637 return 0; /* don't care */
2638 }
2639
2640 /*
2641 * The page straddles i_size. It must be zeroed out on each and every
2642 * writepage invocation because it may be mmapped. "A file is mapped
2643 * in multiples of the page size. For a file that is not a multiple of
2644 * the page size, the remaining memory is zeroed when mapped, and
2645 * writes to that region are not written out to the file."
2646 */
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002647 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648out:
2649 ret = mpage_writepage(page, get_block, wbc);
2650 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002651 ret = __block_write_full_page(inode, page, get_block, wbc,
2652 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 return ret;
2654}
2655EXPORT_SYMBOL(nobh_writepage);
2656
Nick Piggin03158cd2007-10-16 01:25:25 -07002657int nobh_truncate_page(struct address_space *mapping,
2658 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2661 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002662 unsigned blocksize;
2663 sector_t iblock;
2664 unsigned length, pos;
2665 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002667 struct buffer_head map_bh;
2668 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002669
Nick Piggin03158cd2007-10-16 01:25:25 -07002670 blocksize = 1 << inode->i_blkbits;
2671 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Nick Piggin03158cd2007-10-16 01:25:25 -07002673 /* Block boundary? Nothing to do */
2674 if (!length)
2675 return 0;
2676
2677 length = blocksize - length;
2678 iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2679
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002681 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002682 if (!page)
2683 goto out;
2684
Nick Piggin03158cd2007-10-16 01:25:25 -07002685 if (page_has_buffers(page)) {
2686has_buffers:
2687 unlock_page(page);
2688 page_cache_release(page);
2689 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002690 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002691
2692 /* Find the buffer that contains "offset" */
2693 pos = blocksize;
2694 while (offset >= pos) {
2695 iblock++;
2696 pos += blocksize;
2697 }
2698
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002699 map_bh.b_size = blocksize;
2700 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002701 err = get_block(inode, iblock, &map_bh, 0);
2702 if (err)
2703 goto unlock;
2704 /* unmapped? It's a hole - nothing to do */
2705 if (!buffer_mapped(&map_bh))
2706 goto unlock;
2707
2708 /* Ok, it's mapped. Make sure it's up-to-date */
2709 if (!PageUptodate(page)) {
2710 err = mapping->a_ops->readpage(NULL, page);
2711 if (err) {
2712 page_cache_release(page);
2713 goto out;
2714 }
2715 lock_page(page);
2716 if (!PageUptodate(page)) {
2717 err = -EIO;
2718 goto unlock;
2719 }
2720 if (page_has_buffers(page))
2721 goto has_buffers;
2722 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002723 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002724 set_page_dirty(page);
2725 err = 0;
2726
2727unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 unlock_page(page);
2729 page_cache_release(page);
2730out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002731 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732}
2733EXPORT_SYMBOL(nobh_truncate_page);
2734
2735int block_truncate_page(struct address_space *mapping,
2736 loff_t from, get_block_t *get_block)
2737{
2738 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2739 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2740 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002741 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 unsigned length, pos;
2743 struct inode *inode = mapping->host;
2744 struct page *page;
2745 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 int err;
2747
2748 blocksize = 1 << inode->i_blkbits;
2749 length = offset & (blocksize - 1);
2750
2751 /* Block boundary? Nothing to do */
2752 if (!length)
2753 return 0;
2754
2755 length = blocksize - length;
Andrew Morton54b21a72006-01-08 01:03:05 -08002756 iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757
2758 page = grab_cache_page(mapping, index);
2759 err = -ENOMEM;
2760 if (!page)
2761 goto out;
2762
2763 if (!page_has_buffers(page))
2764 create_empty_buffers(page, blocksize, 0);
2765
2766 /* Find the buffer that contains "offset" */
2767 bh = page_buffers(page);
2768 pos = blocksize;
2769 while (offset >= pos) {
2770 bh = bh->b_this_page;
2771 iblock++;
2772 pos += blocksize;
2773 }
2774
2775 err = 0;
2776 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002777 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778 err = get_block(inode, iblock, bh, 0);
2779 if (err)
2780 goto unlock;
2781 /* unmapped? It's a hole - nothing to do */
2782 if (!buffer_mapped(bh))
2783 goto unlock;
2784 }
2785
2786 /* Ok, it's mapped. Make sure it's up-to-date */
2787 if (PageUptodate(page))
2788 set_buffer_uptodate(bh);
2789
David Chinner33a266d2007-02-12 00:51:41 -08002790 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 err = -EIO;
2792 ll_rw_block(READ, 1, &bh);
2793 wait_on_buffer(bh);
2794 /* Uhhuh. Read error. Complain and punt. */
2795 if (!buffer_uptodate(bh))
2796 goto unlock;
2797 }
2798
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002799 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002800 mark_buffer_dirty(bh);
2801 err = 0;
2802
2803unlock:
2804 unlock_page(page);
2805 page_cache_release(page);
2806out:
2807 return err;
2808}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002809EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810
2811/*
2812 * The generic ->writepage function for buffer-backed address_spaces
Chris Mason35c80d52009-04-15 13:22:38 -04002813 * this form passes in the end_io handler used to finish the IO.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 */
Chris Mason35c80d52009-04-15 13:22:38 -04002815int block_write_full_page_endio(struct page *page, get_block_t *get_block,
2816 struct writeback_control *wbc, bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002817{
2818 struct inode * const inode = page->mapping->host;
2819 loff_t i_size = i_size_read(inode);
2820 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2821 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822
2823 /* Is the page fully inside i_size? */
2824 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04002825 return __block_write_full_page(inode, page, get_block, wbc,
2826 handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002827
2828 /* Is the page fully outside i_size? (truncate in progress) */
2829 offset = i_size & (PAGE_CACHE_SIZE-1);
2830 if (page->index >= end_index+1 || !offset) {
2831 /*
2832 * The page may have dirty, unmapped buffers. For example,
2833 * they may have been added in ext3_writepage(). Make them
2834 * freeable here, so the page does not leak.
2835 */
Jan Karaaaa40592005-10-30 15:00:16 -08002836 do_invalidatepage(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 unlock_page(page);
2838 return 0; /* don't care */
2839 }
2840
2841 /*
2842 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05002843 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844 * in multiples of the page size. For a file that is not a multiple of
2845 * the page size, the remaining memory is zeroed when mapped, and
2846 * writes to that region are not written out to the file."
2847 */
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002848 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Chris Mason35c80d52009-04-15 13:22:38 -04002849 return __block_write_full_page(inode, page, get_block, wbc, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002851EXPORT_SYMBOL(block_write_full_page_endio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852
Chris Mason35c80d52009-04-15 13:22:38 -04002853/*
2854 * The generic ->writepage function for buffer-backed address_spaces
2855 */
2856int block_write_full_page(struct page *page, get_block_t *get_block,
2857 struct writeback_control *wbc)
2858{
2859 return block_write_full_page_endio(page, get_block, wbc,
2860 end_buffer_async_write);
2861}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002862EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04002863
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2865 get_block_t *get_block)
2866{
2867 struct buffer_head tmp;
2868 struct inode *inode = mapping->host;
2869 tmp.b_state = 0;
2870 tmp.b_blocknr = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002871 tmp.b_size = 1 << inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 get_block(inode, block, &tmp, 0);
2873 return tmp.b_blocknr;
2874}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002875EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876
NeilBrown6712ecf2007-09-27 12:47:43 +02002877static void end_bio_bh_io_sync(struct bio *bio, int err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878{
2879 struct buffer_head *bh = bio->bi_private;
2880
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 if (err == -EOPNOTSUPP) {
2882 set_bit(BIO_EOPNOTSUPP, &bio->bi_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 }
2884
Keith Mannthey08bafc02008-11-25 10:24:35 +01002885 if (unlikely (test_bit(BIO_QUIET,&bio->bi_flags)))
2886 set_bit(BH_Quiet, &bh->b_state);
2887
Linus Torvalds1da177e2005-04-16 15:20:36 -07002888 bh->b_end_io(bh, test_bit(BIO_UPTODATE, &bio->bi_flags));
2889 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890}
2891
Linus Torvalds57302e02012-12-04 08:25:11 -08002892/*
2893 * This allows us to do IO even on the odd last sectors
2894 * of a device, even if the bh block size is some multiple
2895 * of the physical sector size.
2896 *
2897 * We'll just truncate the bio to the size of the device,
2898 * and clear the end of the buffer head manually.
2899 *
2900 * Truly out-of-range accesses will turn into actual IO
2901 * errors, this only handles the "we need to be able to
2902 * do IO at the final sector" case.
2903 */
2904static void guard_bh_eod(int rw, struct bio *bio, struct buffer_head *bh)
2905{
2906 sector_t maxsector;
2907 unsigned bytes;
2908
2909 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
2910 if (!maxsector)
2911 return;
2912
2913 /*
2914 * If the *whole* IO is past the end of the device,
2915 * let it through, and the IO layer will turn it into
2916 * an EIO.
2917 */
2918 if (unlikely(bio->bi_sector >= maxsector))
2919 return;
2920
2921 maxsector -= bio->bi_sector;
2922 bytes = bio->bi_size;
2923 if (likely((bytes >> 9) <= maxsector))
2924 return;
2925
2926 /* Uhhuh. We've got a bh that straddles the device size! */
2927 bytes = maxsector << 9;
2928
2929 /* Truncate the bio.. */
2930 bio->bi_size = bytes;
2931 bio->bi_io_vec[0].bv_len = bytes;
2932
2933 /* ..and clear the end of the buffer for reads */
Dan Carpenter27d7c2a2012-12-05 20:01:24 +03002934 if ((rw & RW_MASK) == READ) {
Linus Torvalds57302e02012-12-04 08:25:11 -08002935 void *kaddr = kmap_atomic(bh->b_page);
2936 memset(kaddr + bh_offset(bh) + bytes, 0, bh->b_size - bytes);
2937 kunmap_atomic(kaddr);
2938 }
2939}
2940
Linus Torvalds1da177e2005-04-16 15:20:36 -07002941int submit_bh(int rw, struct buffer_head * bh)
2942{
2943 struct bio *bio;
2944 int ret = 0;
2945
2946 BUG_ON(!buffer_locked(bh));
2947 BUG_ON(!buffer_mapped(bh));
2948 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04002949 BUG_ON(buffer_delay(bh));
2950 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951
Jens Axboe48fd4f92008-08-22 10:00:36 +02002952 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02002953 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07002954 */
Jens Axboe48fd4f92008-08-22 10:00:36 +02002955 if (test_set_buffer_req(bh) && (rw & WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956 clear_buffer_write_io_error(bh);
2957
2958 /*
2959 * from here on down, it's all bio -- do the initial mapping,
2960 * submit_bio -> generic_make_request may further map this bio around
2961 */
2962 bio = bio_alloc(GFP_NOIO, 1);
2963
2964 bio->bi_sector = bh->b_blocknr * (bh->b_size >> 9);
2965 bio->bi_bdev = bh->b_bdev;
2966 bio->bi_io_vec[0].bv_page = bh->b_page;
2967 bio->bi_io_vec[0].bv_len = bh->b_size;
2968 bio->bi_io_vec[0].bv_offset = bh_offset(bh);
2969
2970 bio->bi_vcnt = 1;
2971 bio->bi_idx = 0;
2972 bio->bi_size = bh->b_size;
2973
2974 bio->bi_end_io = end_bio_bh_io_sync;
2975 bio->bi_private = bh;
2976
Linus Torvalds57302e02012-12-04 08:25:11 -08002977 /* Take care of bh's that straddle the end of the device */
2978 guard_bh_eod(rw, bio, bh);
2979
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980 bio_get(bio);
2981 submit_bio(rw, bio);
2982
2983 if (bio_flagged(bio, BIO_EOPNOTSUPP))
2984 ret = -EOPNOTSUPP;
2985
2986 bio_put(bio);
2987 return ret;
2988}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002989EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990
2991/**
2992 * ll_rw_block: low-level access to block devices (DEPRECATED)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02002993 * @rw: whether to %READ or %WRITE or maybe %READA (readahead)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994 * @nr: number of &struct buffer_heads in the array
2995 * @bhs: array of pointers to &struct buffer_head
2996 *
Jan Karaa7662232005-09-06 15:19:10 -07002997 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
2998 * requests an I/O operation on them, either a %READ or a %WRITE. The third
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02002999 * %READA option is described in the documentation for generic_make_request()
3000 * which ll_rw_block() calls.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 *
3002 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003003 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3004 * request, and any buffer that appears to be up-to-date when doing read
3005 * request. Further it marks as clean buffers that are processed for
3006 * writing (the buffer cache won't assume that they are actually clean
3007 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003008 *
3009 * ll_rw_block sets b_end_io to simple completion handler that marks
3010 * the buffer up-to-date (if approriate), unlocks the buffer and wakes
3011 * any waiters.
3012 *
3013 * All of the buffers must be for the same device, and must also be a
3014 * multiple of the current approved size for the device.
3015 */
3016void ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
3017{
3018 int i;
3019
3020 for (i = 0; i < nr; i++) {
3021 struct buffer_head *bh = bhs[i];
3022
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003023 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024 continue;
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003025 if (rw == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003027 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003028 get_bh(bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003029 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030 continue;
3031 }
3032 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003034 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003035 get_bh(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036 submit_bh(rw, bh);
3037 continue;
3038 }
3039 }
3040 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003041 }
3042}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003043EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003045void write_dirty_buffer(struct buffer_head *bh, int rw)
3046{
3047 lock_buffer(bh);
3048 if (!test_clear_buffer_dirty(bh)) {
3049 unlock_buffer(bh);
3050 return;
3051 }
3052 bh->b_end_io = end_buffer_write_sync;
3053 get_bh(bh);
3054 submit_bh(rw, bh);
3055}
3056EXPORT_SYMBOL(write_dirty_buffer);
3057
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058/*
3059 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3060 * and then start new I/O and then wait upon it. The caller must have a ref on
3061 * the buffer_head.
3062 */
Christoph Hellwig87e99512010-08-11 17:05:45 +02003063int __sync_dirty_buffer(struct buffer_head *bh, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003064{
3065 int ret = 0;
3066
3067 WARN_ON(atomic_read(&bh->b_count) < 1);
3068 lock_buffer(bh);
3069 if (test_clear_buffer_dirty(bh)) {
3070 get_bh(bh);
3071 bh->b_end_io = end_buffer_write_sync;
Christoph Hellwig87e99512010-08-11 17:05:45 +02003072 ret = submit_bh(rw, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003074 if (!ret && !buffer_uptodate(bh))
3075 ret = -EIO;
3076 } else {
3077 unlock_buffer(bh);
3078 }
3079 return ret;
3080}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003081EXPORT_SYMBOL(__sync_dirty_buffer);
3082
3083int sync_dirty_buffer(struct buffer_head *bh)
3084{
3085 return __sync_dirty_buffer(bh, WRITE_SYNC);
3086}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003087EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
3089/*
3090 * try_to_free_buffers() checks if all the buffers on this particular page
3091 * are unused, and releases them if so.
3092 *
3093 * Exclusion against try_to_free_buffers may be obtained by either
3094 * locking the page or by holding its mapping's private_lock.
3095 *
3096 * If the page is dirty but all the buffers are clean then we need to
3097 * be sure to mark the page clean as well. This is because the page
3098 * may be against a block device, and a later reattachment of buffers
3099 * to a dirty page will set *all* buffers dirty. Which would corrupt
3100 * filesystem data on the same device.
3101 *
3102 * The same applies to regular filesystem pages: if all the buffers are
3103 * clean then we set the page clean and proceed. To do that, we require
3104 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3105 * private_lock.
3106 *
3107 * try_to_free_buffers() is non-blocking.
3108 */
3109static inline int buffer_busy(struct buffer_head *bh)
3110{
3111 return atomic_read(&bh->b_count) |
3112 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3113}
3114
3115static int
3116drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3117{
3118 struct buffer_head *head = page_buffers(page);
3119 struct buffer_head *bh;
3120
3121 bh = head;
3122 do {
akpm@osdl.orgde7d5a32005-05-01 08:58:39 -07003123 if (buffer_write_io_error(bh) && page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003124 set_bit(AS_EIO, &page->mapping->flags);
3125 if (buffer_busy(bh))
3126 goto failed;
3127 bh = bh->b_this_page;
3128 } while (bh != head);
3129
3130 do {
3131 struct buffer_head *next = bh->b_this_page;
3132
Jan Kara535ee2f2008-02-08 04:21:59 -08003133 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134 __remove_assoc_queue(bh);
3135 bh = next;
3136 } while (bh != head);
3137 *buffers_to_free = head;
3138 __clear_page_buffers(page);
3139 return 1;
3140failed:
3141 return 0;
3142}
3143
3144int try_to_free_buffers(struct page *page)
3145{
3146 struct address_space * const mapping = page->mapping;
3147 struct buffer_head *buffers_to_free = NULL;
3148 int ret = 0;
3149
3150 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003151 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003152 return 0;
3153
3154 if (mapping == NULL) { /* can this still happen? */
3155 ret = drop_buffers(page, &buffers_to_free);
3156 goto out;
3157 }
3158
3159 spin_lock(&mapping->private_lock);
3160 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003161
3162 /*
3163 * If the filesystem writes its buffers by hand (eg ext3)
3164 * then we can have clean buffers against a dirty page. We
3165 * clean the page here; otherwise the VM will never notice
3166 * that the filesystem did any IO at all.
3167 *
3168 * Also, during truncate, discard_buffer will have marked all
3169 * the page's buffers clean. We discover that here and clean
3170 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003171 *
3172 * private_lock must be held over this entire operation in order
3173 * to synchronise against __set_page_dirty_buffers and prevent the
3174 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003175 */
3176 if (ret)
3177 cancel_dirty_page(page, PAGE_CACHE_SIZE);
Nick Piggin87df7242007-01-30 14:36:27 +11003178 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179out:
3180 if (buffers_to_free) {
3181 struct buffer_head *bh = buffers_to_free;
3182
3183 do {
3184 struct buffer_head *next = bh->b_this_page;
3185 free_buffer_head(bh);
3186 bh = next;
3187 } while (bh != buffers_to_free);
3188 }
3189 return ret;
3190}
3191EXPORT_SYMBOL(try_to_free_buffers);
3192
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193/*
3194 * There are no bdflush tunables left. But distributions are
3195 * still running obsolete flush daemons, so we terminate them here.
3196 *
3197 * Use of bdflush() is deprecated and will be removed in a future kernel.
Jens Axboe5b0830c2009-09-23 19:37:09 +02003198 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003199 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01003200SYSCALL_DEFINE2(bdflush, int, func, long, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003201{
3202 static int msg_count;
3203
3204 if (!capable(CAP_SYS_ADMIN))
3205 return -EPERM;
3206
3207 if (msg_count < 5) {
3208 msg_count++;
3209 printk(KERN_INFO
3210 "warning: process `%s' used the obsolete bdflush"
3211 " system call\n", current->comm);
3212 printk(KERN_INFO "Fix your initscripts?\n");
3213 }
3214
3215 if (func == 1)
3216 do_exit(0);
3217 return 0;
3218}
3219
3220/*
3221 * Buffer-head allocation
3222 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003223static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224
3225/*
3226 * Once the number of bh's in the machine exceeds this level, we start
3227 * stripping them in writeback.
3228 */
3229static int max_buffer_heads;
3230
3231int buffer_heads_over_limit;
3232
3233struct bh_accounting {
3234 int nr; /* Number of live bh's */
3235 int ratelimit; /* Limit cacheline bouncing */
3236};
3237
3238static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3239
3240static void recalc_bh_state(void)
3241{
3242 int i;
3243 int tot = 0;
3244
Christoph Lameteree1be862010-12-06 11:40:05 -06003245 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003246 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003247 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003248 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003249 tot += per_cpu(bh_accounting, i).nr;
3250 buffer_heads_over_limit = (tot > max_buffer_heads);
3251}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003252
Al Virodd0fc662005-10-07 07:46:04 +01003253struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003254{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003255 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003256 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003257 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003258 preempt_disable();
3259 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003260 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003261 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 }
3263 return ret;
3264}
3265EXPORT_SYMBOL(alloc_buffer_head);
3266
3267void free_buffer_head(struct buffer_head *bh)
3268{
3269 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3270 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003271 preempt_disable();
3272 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003273 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003274 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275}
3276EXPORT_SYMBOL(free_buffer_head);
3277
Linus Torvalds1da177e2005-04-16 15:20:36 -07003278static void buffer_exit_cpu(int cpu)
3279{
3280 int i;
3281 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3282
3283 for (i = 0; i < BH_LRU_SIZE; i++) {
3284 brelse(b->bhs[i]);
3285 b->bhs[i] = NULL;
3286 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003287 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003288 per_cpu(bh_accounting, cpu).nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289}
3290
3291static int buffer_cpu_notify(struct notifier_block *self,
3292 unsigned long action, void *hcpu)
3293{
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003294 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 buffer_exit_cpu((unsigned long)hcpu);
3296 return NOTIFY_OK;
3297}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003299/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003300 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003301 * @bh: struct buffer_head
3302 *
3303 * Return true if the buffer is up-to-date and false,
3304 * with the buffer locked, if not.
3305 */
3306int bh_uptodate_or_lock(struct buffer_head *bh)
3307{
3308 if (!buffer_uptodate(bh)) {
3309 lock_buffer(bh);
3310 if (!buffer_uptodate(bh))
3311 return 0;
3312 unlock_buffer(bh);
3313 }
3314 return 1;
3315}
3316EXPORT_SYMBOL(bh_uptodate_or_lock);
3317
3318/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003319 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003320 * @bh: struct buffer_head
3321 *
3322 * Returns zero on success and -EIO on error.
3323 */
3324int bh_submit_read(struct buffer_head *bh)
3325{
3326 BUG_ON(!buffer_locked(bh));
3327
3328 if (buffer_uptodate(bh)) {
3329 unlock_buffer(bh);
3330 return 0;
3331 }
3332
3333 get_bh(bh);
3334 bh->b_end_io = end_buffer_read_sync;
3335 submit_bh(READ, bh);
3336 wait_on_buffer(bh);
3337 if (buffer_uptodate(bh))
3338 return 0;
3339 return -EIO;
3340}
3341EXPORT_SYMBOL(bh_submit_read);
3342
Linus Torvalds1da177e2005-04-16 15:20:36 -07003343void __init buffer_init(void)
3344{
3345 int nrpages;
3346
Christoph Lameterb98938c2008-02-04 22:28:36 -08003347 bh_cachep = kmem_cache_create("buffer_head",
3348 sizeof(struct buffer_head), 0,
3349 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3350 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003351 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352
3353 /*
3354 * Limit the bh occupancy to 10% of ZONE_NORMAL
3355 */
3356 nrpages = (nr_free_buffer_pages() * 10) / 100;
3357 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3358 hotcpu_notifier(buffer_cpu_notify, 0);
3359}