blob: 82283abb2795d95bd31d595809e7881ca43942a3 [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>
Tejun Heobafc0db2015-06-02 08:37:23 -060033#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070034#include <linux/writeback.h>
35#include <linux/hash.h>
36#include <linux/suspend.h>
37#include <linux/buffer_head.h>
Andrew Morton55e829a2006-12-10 02:19:27 -080038#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070039#include <linux/bio.h>
40#include <linux/notifier.h>
41#include <linux/cpu.h>
42#include <linux/bitops.h>
43#include <linux/mpage.h>
Ingo Molnarfb1c8f92005-09-10 00:25:56 -070044#include <linux/bit_spinlock.h>
Tejun Heo5305cb82013-01-11 13:06:36 -080045#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070046
47static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
Tejun Heob16b1de2015-06-02 08:39:48 -060048static int submit_bh_wbc(int rw, struct buffer_head *bh,
49 unsigned long bio_flags,
50 struct writeback_control *wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
52#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
53
Yan Honga3f3c292012-12-12 13:52:15 -080054void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -070055{
56 bh->b_end_io = handler;
57 bh->b_private = private;
58}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070059EXPORT_SYMBOL(init_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070060
Tejun Heof0059af2013-01-11 13:06:35 -080061inline void touch_buffer(struct buffer_head *bh)
62{
Tejun Heo5305cb82013-01-11 13:06:36 -080063 trace_block_touch_buffer(bh);
Tejun Heof0059af2013-01-11 13:06:35 -080064 mark_page_accessed(bh->b_page);
65}
66EXPORT_SYMBOL(touch_buffer);
67
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080068void __lock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069{
NeilBrown74316202014-07-07 15:16:04 +100070 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
72EXPORT_SYMBOL(__lock_buffer);
73
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080074void unlock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Nick Piggin51b07fc2008-10-18 20:27:00 -070076 clear_bit_unlock(BH_Lock, &bh->b_state);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010077 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 wake_up_bit(&bh->b_state, BH_Lock);
79}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070080EXPORT_SYMBOL(unlock_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82/*
Mel Gormanb4597222013-07-03 15:02:05 -070083 * Returns if the page has dirty or writeback buffers. If all the buffers
84 * are unlocked and clean then the PageDirty information is stale. If
85 * any of the pages are locked, it is assumed they are locked for IO.
86 */
87void buffer_check_dirty_writeback(struct page *page,
88 bool *dirty, bool *writeback)
89{
90 struct buffer_head *head, *bh;
91 *dirty = false;
92 *writeback = false;
93
94 BUG_ON(!PageLocked(page));
95
96 if (!page_has_buffers(page))
97 return;
98
99 if (PageWriteback(page))
100 *writeback = true;
101
102 head = page_buffers(page);
103 bh = head;
104 do {
105 if (buffer_locked(bh))
106 *writeback = true;
107
108 if (buffer_dirty(bh))
109 *dirty = true;
110
111 bh = bh->b_this_page;
112 } while (bh != head);
113}
114EXPORT_SYMBOL(buffer_check_dirty_writeback);
115
116/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 * Block until a buffer comes unlocked. This doesn't stop it
118 * from becoming locked again - you have to lock it yourself
119 * if you want to preserve its state.
120 */
121void __wait_on_buffer(struct buffer_head * bh)
122{
NeilBrown74316202014-07-07 15:16:04 +1000123 wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700125EXPORT_SYMBOL(__wait_on_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
127static void
128__clear_page_buffers(struct page *page)
129{
130 ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700131 set_page_private(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132 page_cache_release(page);
133}
134
Robert Elliottb744c2a2014-10-21 13:55:09 -0600135static void buffer_io_error(struct buffer_head *bh, char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700136{
137 char b[BDEVNAME_SIZE];
Robert Elliott432f16e2014-10-21 13:55:11 -0600138
139 if (!test_bit(BH_Quiet, &bh->b_state))
140 printk_ratelimited(KERN_ERR
141 "Buffer I/O error on dev %s, logical block %llu%s\n",
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 bdevname(bh->b_bdev, b),
Robert Elliottb744c2a2014-10-21 13:55:09 -0600143 (unsigned long long)bh->b_blocknr, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144}
145
146/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700147 * End-of-IO handler helper function which does not touch the bh after
148 * unlocking it.
149 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
150 * a race there is benign: unlock_buffer() only use the bh's address for
151 * hashing after unlocking the buffer, so it doesn't actually touch the bh
152 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700154static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700155{
156 if (uptodate) {
157 set_buffer_uptodate(bh);
158 } else {
159 /* This happens, due to failed READA attempts. */
160 clear_buffer_uptodate(bh);
161 }
162 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700163}
164
165/*
166 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
167 * unlock the buffer. This is what ll_rw_block uses too.
168 */
169void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
170{
171 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 put_bh(bh);
173}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700174EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175
176void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
177{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178 if (uptodate) {
179 set_buffer_uptodate(bh);
180 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600181 buffer_io_error(bh, ", lost sync page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700182 set_buffer_write_io_error(bh);
183 clear_buffer_uptodate(bh);
184 }
185 unlock_buffer(bh);
186 put_bh(bh);
187}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700188EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189
190/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191 * Various filesystems appear to want __find_get_block to be non-blocking.
192 * But it's the page lock which protects the buffers. To get around this,
193 * we get exclusion from try_to_free_buffers with the blockdev mapping's
194 * private_lock.
195 *
196 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
197 * may be quite high. This code could TryLock the page, and if that
198 * succeeds, there is no need to take private_lock. (But if
199 * private_lock is contended then so is mapping->tree_lock).
200 */
201static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800202__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203{
204 struct inode *bd_inode = bdev->bd_inode;
205 struct address_space *bd_mapping = bd_inode->i_mapping;
206 struct buffer_head *ret = NULL;
207 pgoff_t index;
208 struct buffer_head *bh;
209 struct buffer_head *head;
210 struct page *page;
211 int all_mapped = 1;
212
213 index = block >> (PAGE_CACHE_SHIFT - bd_inode->i_blkbits);
Mel Gorman2457aec2014-06-04 16:10:31 -0700214 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 if (!page)
216 goto out;
217
218 spin_lock(&bd_mapping->private_lock);
219 if (!page_has_buffers(page))
220 goto out_unlock;
221 head = page_buffers(page);
222 bh = head;
223 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700224 if (!buffer_mapped(bh))
225 all_mapped = 0;
226 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700227 ret = bh;
228 get_bh(bh);
229 goto out_unlock;
230 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700231 bh = bh->b_this_page;
232 } while (bh != head);
233
234 /* we might be here because some of the buffers on this page are
235 * not mapped. This is due to various races between
236 * file io on the block device and getblk. It gets dealt with
237 * elsewhere, don't buffer_error if we had some unmapped buffers
238 */
239 if (all_mapped) {
Tao Ma72a2ebd2011-10-31 17:09:00 -0700240 char b[BDEVNAME_SIZE];
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242 printk("__find_get_block_slow() failed. "
243 "block=%llu, b_blocknr=%llu\n",
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800244 (unsigned long long)block,
245 (unsigned long long)bh->b_blocknr);
246 printk("b_state=0x%08lx, b_size=%zu\n",
247 bh->b_state, bh->b_size);
Tao Ma72a2ebd2011-10-31 17:09:00 -0700248 printk("device %s blocksize: %d\n", bdevname(bdev, b),
249 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250 }
251out_unlock:
252 spin_unlock(&bd_mapping->private_lock);
253 page_cache_release(page);
254out:
255 return ret;
256}
257
Linus Torvalds1da177e2005-04-16 15:20:36 -0700258/*
Jens Axboe5b0830c2009-09-23 19:37:09 +0200259 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 */
261static void free_more_memory(void)
262{
Mel Gorman19770b32008-04-28 02:12:18 -0700263 struct zone *zone;
Mel Gorman0e884602008-04-28 02:12:14 -0700264 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700265
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600266 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700267 yield();
268
Mel Gorman0e884602008-04-28 02:12:14 -0700269 for_each_online_node(nid) {
Mel Gorman19770b32008-04-28 02:12:18 -0700270 (void)first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
271 gfp_zone(GFP_NOFS), NULL,
272 &zone);
273 if (zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700274 try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -0700275 GFP_NOFS, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276 }
277}
278
279/*
280 * I/O completion handler for block_read_full_page() - pages
281 * which come unlocked at the end of I/O.
282 */
283static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
284{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700285 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700286 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700287 struct buffer_head *tmp;
288 struct page *page;
289 int page_uptodate = 1;
290
291 BUG_ON(!buffer_async_read(bh));
292
293 page = bh->b_page;
294 if (uptodate) {
295 set_buffer_uptodate(bh);
296 } else {
297 clear_buffer_uptodate(bh);
Robert Elliott432f16e2014-10-21 13:55:11 -0600298 buffer_io_error(bh, ", async page read");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700299 SetPageError(page);
300 }
301
302 /*
303 * Be _very_ careful from here on. Bad things can happen if
304 * two buffer heads end IO at almost the same time and both
305 * decide that the page is now completely done.
306 */
Nick Piggina3972202005-07-07 17:56:56 -0700307 first = page_buffers(page);
308 local_irq_save(flags);
309 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 clear_buffer_async_read(bh);
311 unlock_buffer(bh);
312 tmp = bh;
313 do {
314 if (!buffer_uptodate(tmp))
315 page_uptodate = 0;
316 if (buffer_async_read(tmp)) {
317 BUG_ON(!buffer_locked(tmp));
318 goto still_busy;
319 }
320 tmp = tmp->b_this_page;
321 } while (tmp != bh);
Nick Piggina3972202005-07-07 17:56:56 -0700322 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
323 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700324
325 /*
326 * If none of the buffers had errors and they are all
327 * uptodate then we can set the page uptodate.
328 */
329 if (page_uptodate && !PageError(page))
330 SetPageUptodate(page);
331 unlock_page(page);
332 return;
333
334still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700335 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
336 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700337 return;
338}
339
340/*
341 * Completion handler for block_write_full_page() - pages which are unlocked
342 * during I/O, and which have PageWriteback cleared upon I/O completion.
343 */
Chris Mason35c80d52009-04-15 13:22:38 -0400344void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700347 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700348 struct buffer_head *tmp;
349 struct page *page;
350
351 BUG_ON(!buffer_async_write(bh));
352
353 page = bh->b_page;
354 if (uptodate) {
355 set_buffer_uptodate(bh);
356 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600357 buffer_io_error(bh, ", lost async page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700358 set_bit(AS_EIO, &page->mapping->flags);
Jan Kara58ff4072006-10-17 00:10:19 -0700359 set_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 clear_buffer_uptodate(bh);
361 SetPageError(page);
362 }
363
Nick Piggina3972202005-07-07 17:56:56 -0700364 first = page_buffers(page);
365 local_irq_save(flags);
366 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
367
Linus Torvalds1da177e2005-04-16 15:20:36 -0700368 clear_buffer_async_write(bh);
369 unlock_buffer(bh);
370 tmp = bh->b_this_page;
371 while (tmp != bh) {
372 if (buffer_async_write(tmp)) {
373 BUG_ON(!buffer_locked(tmp));
374 goto still_busy;
375 }
376 tmp = tmp->b_this_page;
377 }
Nick Piggina3972202005-07-07 17:56:56 -0700378 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
379 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 end_page_writeback(page);
381 return;
382
383still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700384 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
385 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 return;
387}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700388EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700389
390/*
391 * If a page's buffers are under async readin (end_buffer_async_read
392 * completion) then there is a possibility that another thread of
393 * control could lock one of the buffers after it has completed
394 * but while some of the other buffers have not completed. This
395 * locked buffer would confuse end_buffer_async_read() into not unlocking
396 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
397 * that this buffer is not under async I/O.
398 *
399 * The page comes unlocked when it has no locked buffer_async buffers
400 * left.
401 *
402 * PageLocked prevents anyone starting new async I/O reads any of
403 * the buffers.
404 *
405 * PageWriteback is used to prevent simultaneous writeout of the same
406 * page.
407 *
408 * PageLocked prevents anyone from starting writeback of a page which is
409 * under read I/O (PageWriteback is only ever set against a locked page).
410 */
411static void mark_buffer_async_read(struct buffer_head *bh)
412{
413 bh->b_end_io = end_buffer_async_read;
414 set_buffer_async_read(bh);
415}
416
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700417static void mark_buffer_async_write_endio(struct buffer_head *bh,
418 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400419{
420 bh->b_end_io = handler;
421 set_buffer_async_write(bh);
422}
423
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424void mark_buffer_async_write(struct buffer_head *bh)
425{
Chris Mason35c80d52009-04-15 13:22:38 -0400426 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427}
428EXPORT_SYMBOL(mark_buffer_async_write);
429
430
431/*
432 * fs/buffer.c contains helper functions for buffer-backed address space's
433 * fsync functions. A common requirement for buffer-based filesystems is
434 * that certain data from the backing blockdev needs to be written out for
435 * a successful fsync(). For example, ext2 indirect blocks need to be
436 * written back and waited upon before fsync() returns.
437 *
438 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
439 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
440 * management of a list of dependent buffers at ->i_mapping->private_list.
441 *
442 * Locking is a little subtle: try_to_free_buffers() will remove buffers
443 * from their controlling inode's queue when they are being freed. But
444 * try_to_free_buffers() will be operating against the *blockdev* mapping
445 * at the time, not against the S_ISREG file which depends on those buffers.
446 * So the locking for private_list is via the private_lock in the address_space
447 * which backs the buffers. Which is different from the address_space
448 * against which the buffers are listed. So for a particular address_space,
449 * mapping->private_lock does *not* protect mapping->private_list! In fact,
450 * mapping->private_list will always be protected by the backing blockdev's
451 * ->private_lock.
452 *
453 * Which introduces a requirement: all buffers on an address_space's
454 * ->private_list must be from the same address_space: the blockdev's.
455 *
456 * address_spaces which do not place buffers at ->private_list via these
457 * utility functions are free to use private_lock and private_list for
458 * whatever they want. The only requirement is that list_empty(private_list)
459 * be true at clear_inode() time.
460 *
461 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
462 * filesystems should do that. invalidate_inode_buffers() should just go
463 * BUG_ON(!list_empty).
464 *
465 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
466 * take an address_space, not an inode. And it should be called
467 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
468 * queued up.
469 *
470 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
471 * list if it is already on a list. Because if the buffer is on a list,
472 * it *must* already be on the right one. If not, the filesystem is being
473 * silly. This will save a ton of locking. But first we have to ensure
474 * that buffers are taken *off* the old inode's list when they are freed
475 * (presumably in truncate). That requires careful auditing of all
476 * filesystems (do it inside bforget()). It could also be done by bringing
477 * b_inode back.
478 */
479
480/*
481 * The buffer's backing address_space's private_lock must be held
482 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700483static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700484{
485 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -0700486 WARN_ON(!bh->b_assoc_map);
487 if (buffer_write_io_error(bh))
488 set_bit(AS_EIO, &bh->b_assoc_map->flags);
489 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490}
491
492int inode_has_buffers(struct inode *inode)
493{
494 return !list_empty(&inode->i_data.private_list);
495}
496
497/*
498 * osync is designed to support O_SYNC io. It waits synchronously for
499 * all already-submitted IO to complete, but does not queue any new
500 * writes to the disk.
501 *
502 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
503 * you dirty the buffers, and then use osync_inode_buffers to wait for
504 * completion. Any other dirty buffers which are not yet queued for
505 * write will not be flushed to disk by the osync.
506 */
507static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
508{
509 struct buffer_head *bh;
510 struct list_head *p;
511 int err = 0;
512
513 spin_lock(lock);
514repeat:
515 list_for_each_prev(p, list) {
516 bh = BH_ENTRY(p);
517 if (buffer_locked(bh)) {
518 get_bh(bh);
519 spin_unlock(lock);
520 wait_on_buffer(bh);
521 if (!buffer_uptodate(bh))
522 err = -EIO;
523 brelse(bh);
524 spin_lock(lock);
525 goto repeat;
526 }
527 }
528 spin_unlock(lock);
529 return err;
530}
531
Al Viro01a05b32010-03-23 06:06:58 -0400532static void do_thaw_one(struct super_block *sb, void *unused)
533{
534 char b[BDEVNAME_SIZE];
535 while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
536 printk(KERN_WARNING "Emergency Thaw on %s\n",
537 bdevname(sb->s_bdev, b));
538}
539
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700540static void do_thaw_all(struct work_struct *work)
Eric Sandeenc2d75432009-03-31 15:23:46 -0700541{
Al Viro01a05b32010-03-23 06:06:58 -0400542 iterate_supers(do_thaw_one, NULL);
Jens Axboe053c5252009-04-08 13:44:08 +0200543 kfree(work);
Eric Sandeenc2d75432009-03-31 15:23:46 -0700544 printk(KERN_WARNING "Emergency Thaw complete\n");
545}
546
547/**
548 * emergency_thaw_all -- forcibly thaw every frozen filesystem
549 *
550 * Used for emergency unfreeze of all filesystems via SysRq
551 */
552void emergency_thaw_all(void)
553{
Jens Axboe053c5252009-04-08 13:44:08 +0200554 struct work_struct *work;
555
556 work = kmalloc(sizeof(*work), GFP_ATOMIC);
557 if (work) {
558 INIT_WORK(work, do_thaw_all);
559 schedule_work(work);
560 }
Eric Sandeenc2d75432009-03-31 15:23:46 -0700561}
562
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800564 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700565 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 *
567 * Starts I/O against the buffers at mapping->private_list, and waits upon
568 * that I/O.
569 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700570 * Basically, this is a convenience function for fsync().
571 * @mapping is a file or directory which needs those buffers to be written for
572 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 */
574int sync_mapping_buffers(struct address_space *mapping)
575{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800576 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577
578 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
579 return 0;
580
581 return fsync_buffers_list(&buffer_mapping->private_lock,
582 &mapping->private_list);
583}
584EXPORT_SYMBOL(sync_mapping_buffers);
585
586/*
587 * Called when we've recently written block `bblock', and it is known that
588 * `bblock' was for a buffer_boundary() buffer. This means that the block at
589 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
590 * dirty, schedule it for IO. So that indirects merge nicely with their data.
591 */
592void write_boundary_block(struct block_device *bdev,
593 sector_t bblock, unsigned blocksize)
594{
595 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
596 if (bh) {
597 if (buffer_dirty(bh))
598 ll_rw_block(WRITE, 1, &bh);
599 put_bh(bh);
600 }
601}
602
603void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
604{
605 struct address_space *mapping = inode->i_mapping;
606 struct address_space *buffer_mapping = bh->b_page->mapping;
607
608 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800609 if (!mapping->private_data) {
610 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800612 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800614 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615 spin_lock(&buffer_mapping->private_lock);
616 list_move_tail(&bh->b_assoc_buffers,
617 &mapping->private_list);
Jan Kara58ff4072006-10-17 00:10:19 -0700618 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700619 spin_unlock(&buffer_mapping->private_lock);
620 }
621}
622EXPORT_SYMBOL(mark_buffer_dirty_inode);
623
624/*
Nick Piggin787d2212007-07-17 04:03:34 -0700625 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
626 * dirty.
627 *
628 * If warn is true, then emit a warning if the page is not uptodate and has
629 * not been truncated.
Greg Thelenc4843a72015-05-22 17:13:16 -0400630 *
631 * The caller must hold mem_cgroup_begin_page_stat() lock.
Nick Piggin787d2212007-07-17 04:03:34 -0700632 */
Greg Thelenc4843a72015-05-22 17:13:16 -0400633static void __set_page_dirty(struct page *page, struct address_space *mapping,
634 struct mem_cgroup *memcg, int warn)
Nick Piggin787d2212007-07-17 04:03:34 -0700635{
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800636 unsigned long flags;
637
638 spin_lock_irqsave(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700639 if (page->mapping) { /* Race with truncate? */
640 WARN_ON_ONCE(warn && !PageUptodate(page));
Greg Thelenc4843a72015-05-22 17:13:16 -0400641 account_page_dirtied(page, mapping, memcg);
Nick Piggin787d2212007-07-17 04:03:34 -0700642 radix_tree_tag_set(&mapping->page_tree,
643 page_index(page), PAGECACHE_TAG_DIRTY);
644 }
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800645 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700646}
647
648/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649 * Add a page to the dirty page list.
650 *
651 * It is a sad fact of life that this function is called from several places
652 * deeply under spinlocking. It may not sleep.
653 *
654 * If the page has buffers, the uptodate buffers are set dirty, to preserve
655 * dirty-state coherency between the page and the buffers. It the page does
656 * not have buffers then when they are later attached they will all be set
657 * dirty.
658 *
659 * The buffers are dirtied before the page is dirtied. There's a small race
660 * window in which a writepage caller may see the page cleanness but not the
661 * buffer dirtiness. That's fine. If this code were to set the page dirty
662 * before the buffers, a concurrent writepage caller could clear the page dirty
663 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
664 * page on the dirty page list.
665 *
666 * We use private_lock to lock against try_to_free_buffers while using the
667 * page's buffer list. Also use this to protect against clean buffers being
668 * added to the page after it was set dirty.
669 *
670 * FIXME: may need to call ->reservepage here as well. That's rather up to the
671 * address_space though.
672 */
673int __set_page_dirty_buffers(struct page *page)
674{
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700675 int newly_dirty;
Greg Thelenc4843a72015-05-22 17:13:16 -0400676 struct mem_cgroup *memcg;
Nick Piggin787d2212007-07-17 04:03:34 -0700677 struct address_space *mapping = page_mapping(page);
Nick Pigginebf7a222006-10-10 04:36:54 +0200678
679 if (unlikely(!mapping))
680 return !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681
682 spin_lock(&mapping->private_lock);
683 if (page_has_buffers(page)) {
684 struct buffer_head *head = page_buffers(page);
685 struct buffer_head *bh = head;
686
687 do {
688 set_buffer_dirty(bh);
689 bh = bh->b_this_page;
690 } while (bh != head);
691 }
Greg Thelenc4843a72015-05-22 17:13:16 -0400692 /*
693 * Use mem_group_begin_page_stat() to keep PageDirty synchronized with
694 * per-memcg dirty page counters.
695 */
696 memcg = mem_cgroup_begin_page_stat(page);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700697 newly_dirty = !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700698 spin_unlock(&mapping->private_lock);
699
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700700 if (newly_dirty)
Greg Thelenc4843a72015-05-22 17:13:16 -0400701 __set_page_dirty(page, mapping, memcg, 1);
702
703 mem_cgroup_end_page_stat(memcg);
704
705 if (newly_dirty)
706 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
707
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700708 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709}
710EXPORT_SYMBOL(__set_page_dirty_buffers);
711
712/*
713 * Write out and wait upon a list of buffers.
714 *
715 * We have conflicting pressures: we want to make sure that all
716 * initially dirty buffers get waited on, but that any subsequently
717 * dirtied buffers don't. After all, we don't want fsync to last
718 * forever if somebody is actively writing to the file.
719 *
720 * Do this in two main stages: first we copy dirty buffers to a
721 * temporary inode list, queueing the writes as we go. Then we clean
722 * up, waiting for those writes to complete.
723 *
724 * During this second stage, any subsequent updates to the file may end
725 * up refiling the buffer on the original inode's dirty list again, so
726 * there is a chance we will end up with a buffer queued for write but
727 * not yet completed on that list. So, as a final cleanup we go through
728 * the osync code to catch these locked, dirty buffers without requeuing
729 * any newly dirty buffers for write.
730 */
731static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
732{
733 struct buffer_head *bh;
734 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100735 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100737 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700738
739 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100740 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700741
742 spin_lock(lock);
743 while (!list_empty(list)) {
744 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800745 mapping = bh->b_assoc_map;
Jan Kara58ff4072006-10-17 00:10:19 -0700746 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800747 /* Avoid race with mark_buffer_dirty_inode() which does
748 * a lockless check and we rely on seeing the dirty bit */
749 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700750 if (buffer_dirty(bh) || buffer_locked(bh)) {
751 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800752 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700753 if (buffer_dirty(bh)) {
754 get_bh(bh);
755 spin_unlock(lock);
756 /*
757 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200758 * write_dirty_buffer() actually writes the
759 * current contents - it is a noop if I/O is
760 * still in flight on potentially older
761 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700762 */
Jens Axboe721a9602011-03-09 11:56:30 +0100763 write_dirty_buffer(bh, WRITE_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200764
765 /*
766 * Kick off IO for the previous mapping. Note
767 * that we will not run the very last mapping,
768 * wait_on_buffer() will do that for us
769 * through sync_buffer().
770 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700771 brelse(bh);
772 spin_lock(lock);
773 }
774 }
775 }
776
Jens Axboe4ee24912011-03-17 10:51:40 +0100777 spin_unlock(lock);
778 blk_finish_plug(&plug);
779 spin_lock(lock);
780
Linus Torvalds1da177e2005-04-16 15:20:36 -0700781 while (!list_empty(&tmp)) {
782 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800784 mapping = bh->b_assoc_map;
785 __remove_assoc_queue(bh);
786 /* Avoid race with mark_buffer_dirty_inode() which does
787 * a lockless check and we rely on seeing the dirty bit */
788 smp_mb();
789 if (buffer_dirty(bh)) {
790 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800791 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800792 bh->b_assoc_map = mapping;
793 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700794 spin_unlock(lock);
795 wait_on_buffer(bh);
796 if (!buffer_uptodate(bh))
797 err = -EIO;
798 brelse(bh);
799 spin_lock(lock);
800 }
801
802 spin_unlock(lock);
803 err2 = osync_buffers_list(lock, list);
804 if (err)
805 return err;
806 else
807 return err2;
808}
809
810/*
811 * Invalidate any and all dirty buffers on a given inode. We are
812 * probably unmounting the fs, but that doesn't mean we have already
813 * done a sync(). Just drop the buffers from the inode list.
814 *
815 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
816 * assumes that all the buffers are against the blockdev. Not true
817 * for reiserfs.
818 */
819void invalidate_inode_buffers(struct inode *inode)
820{
821 if (inode_has_buffers(inode)) {
822 struct address_space *mapping = &inode->i_data;
823 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800824 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826 spin_lock(&buffer_mapping->private_lock);
827 while (!list_empty(list))
828 __remove_assoc_queue(BH_ENTRY(list->next));
829 spin_unlock(&buffer_mapping->private_lock);
830 }
831}
Jan Kara52b19ac2008-09-23 18:24:08 +0200832EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833
834/*
835 * Remove any clean buffers from the inode's buffer list. This is called
836 * when we're trying to free the inode itself. Those buffers can pin it.
837 *
838 * Returns true if all buffers were removed.
839 */
840int remove_inode_buffers(struct inode *inode)
841{
842 int ret = 1;
843
844 if (inode_has_buffers(inode)) {
845 struct address_space *mapping = &inode->i_data;
846 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800847 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
849 spin_lock(&buffer_mapping->private_lock);
850 while (!list_empty(list)) {
851 struct buffer_head *bh = BH_ENTRY(list->next);
852 if (buffer_dirty(bh)) {
853 ret = 0;
854 break;
855 }
856 __remove_assoc_queue(bh);
857 }
858 spin_unlock(&buffer_mapping->private_lock);
859 }
860 return ret;
861}
862
863/*
864 * Create the appropriate buffers when given a page for data area and
865 * the size of each buffer.. Use the bh->b_this_page linked list to
866 * follow the buffers created. Return NULL if unable to create more
867 * buffers.
868 *
869 * The retry flag is used to differentiate async IO (paging, swapping)
870 * which may not fail from ordinary buffer allocations.
871 */
872struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
873 int retry)
874{
875 struct buffer_head *bh, *head;
876 long offset;
877
878try_again:
879 head = NULL;
880 offset = PAGE_SIZE;
881 while ((offset -= size) >= 0) {
882 bh = alloc_buffer_head(GFP_NOFS);
883 if (!bh)
884 goto no_grow;
885
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 bh->b_this_page = head;
887 bh->b_blocknr = -1;
888 head = bh;
889
Linus Torvalds1da177e2005-04-16 15:20:36 -0700890 bh->b_size = size;
891
892 /* Link the buffer to its page */
893 set_bh_page(bh, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700894 }
895 return head;
896/*
897 * In case anything failed, we just free everything we got.
898 */
899no_grow:
900 if (head) {
901 do {
902 bh = head;
903 head = head->b_this_page;
904 free_buffer_head(bh);
905 } while (head);
906 }
907
908 /*
909 * Return failure for non-async IO requests. Async IO requests
910 * are not allowed to fail, so we have to wait until buffer heads
911 * become available. But we don't want tasks sleeping with
912 * partially complete buffers, so all were released above.
913 */
914 if (!retry)
915 return NULL;
916
917 /* We're _really_ low on memory. Now we just
918 * wait for old buffer heads to become free due to
919 * finishing IO. Since this is an async request and
920 * the reserve list is empty, we're sure there are
921 * async buffer heads in use.
922 */
923 free_more_memory();
924 goto try_again;
925}
926EXPORT_SYMBOL_GPL(alloc_page_buffers);
927
928static inline void
929link_dev_buffers(struct page *page, struct buffer_head *head)
930{
931 struct buffer_head *bh, *tail;
932
933 bh = head;
934 do {
935 tail = bh;
936 bh = bh->b_this_page;
937 } while (bh);
938 tail->b_this_page = head;
939 attach_page_buffers(page, head);
940}
941
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800942static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
943{
944 sector_t retval = ~((sector_t)0);
945 loff_t sz = i_size_read(bdev->bd_inode);
946
947 if (sz) {
948 unsigned int sizebits = blksize_bits(size);
949 retval = (sz >> sizebits);
950 }
951 return retval;
952}
953
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954/*
955 * Initialise the state of a blockdev page's buffers.
956 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200957static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958init_page_buffers(struct page *page, struct block_device *bdev,
959 sector_t block, int size)
960{
961 struct buffer_head *head = page_buffers(page);
962 struct buffer_head *bh = head;
963 int uptodate = PageUptodate(page);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800964 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700965
966 do {
967 if (!buffer_mapped(bh)) {
968 init_buffer(bh, NULL, NULL);
969 bh->b_bdev = bdev;
970 bh->b_blocknr = block;
971 if (uptodate)
972 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200973 if (block < end_block)
974 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700975 }
976 block++;
977 bh = bh->b_this_page;
978 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200979
980 /*
981 * Caller needs to validate requested block against end of device.
982 */
983 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984}
985
986/*
987 * Create the page-cache page that contains the requested block.
988 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200989 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700990 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200991static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992grow_dev_page(struct block_device *bdev, sector_t block,
Gioh Kim3b5e6452014-09-04 22:04:42 -0400993 pgoff_t index, int size, int sizebits, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994{
995 struct inode *inode = bdev->bd_inode;
996 struct page *page;
997 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200998 sector_t end_block;
999 int ret = 0; /* Will call free_more_memory() */
Johannes Weiner84235de2013-10-16 13:47:00 -07001000 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001
Gioh Kim3b5e6452014-09-04 22:04:42 -04001002 gfp_mask = (mapping_gfp_mask(inode->i_mapping) & ~__GFP_FS) | gfp;
1003
Johannes Weiner84235de2013-10-16 13:47:00 -07001004 /*
1005 * XXX: __getblk_slow() can not really deal with failure and
1006 * will endlessly loop on improvised global reclaim. Prefer
1007 * looping in the allocator rather than here, at least that
1008 * code knows what it's doing.
1009 */
1010 gfp_mask |= __GFP_NOFAIL;
1011
1012 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001013 if (!page)
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001014 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001015
Eric Sesterhenne827f922006-03-26 18:24:46 +02001016 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017
1018 if (page_has_buffers(page)) {
1019 bh = page_buffers(page);
1020 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001021 end_block = init_page_buffers(page, bdev,
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001022 (sector_t)index << sizebits,
1023 size);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001024 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 }
1026 if (!try_to_free_buffers(page))
1027 goto failed;
1028 }
1029
1030 /*
1031 * Allocate some buffers for this page
1032 */
1033 bh = alloc_page_buffers(page, size, 0);
1034 if (!bh)
1035 goto failed;
1036
1037 /*
1038 * Link the page to the buffers and initialise them. Take the
1039 * lock to be atomic wrt __find_get_block(), which does not
1040 * run under the page lock.
1041 */
1042 spin_lock(&inode->i_mapping->private_lock);
1043 link_dev_buffers(page, bh);
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001044 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1045 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001047done:
1048 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 unlock_page(page);
1051 page_cache_release(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001052 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053}
1054
1055/*
1056 * Create buffers for the specified block device block's page. If
1057 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001058 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001059static int
Gioh Kim3b5e6452014-09-04 22:04:42 -04001060grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001061{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062 pgoff_t index;
1063 int sizebits;
1064
1065 sizebits = -1;
1066 do {
1067 sizebits++;
1068 } while ((size << sizebits) < PAGE_SIZE);
1069
1070 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001071
Andrew Mortone5657932006-10-11 01:21:46 -07001072 /*
1073 * Check for a block which wants to lie outside our maximum possible
1074 * pagecache index. (this comparison is done using sector_t types).
1075 */
1076 if (unlikely(index != block >> sizebits)) {
1077 char b[BDEVNAME_SIZE];
1078
1079 printk(KERN_ERR "%s: requested out-of-range block %llu for "
1080 "device %s\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001081 __func__, (unsigned long long)block,
Andrew Mortone5657932006-10-11 01:21:46 -07001082 bdevname(bdev, b));
1083 return -EIO;
1084 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001085
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 /* Create a page with the proper size buffers.. */
Gioh Kim3b5e6452014-09-04 22:04:42 -04001087 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088}
1089
Gioh Kim3b5e6452014-09-04 22:04:42 -04001090struct buffer_head *
1091__getblk_slow(struct block_device *bdev, sector_t block,
1092 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093{
1094 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001095 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096 (size < 512 || size > PAGE_SIZE))) {
1097 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1098 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001099 printk(KERN_ERR "logical block size: %d\n",
1100 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 dump_stack();
1103 return NULL;
1104 }
1105
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001106 for (;;) {
1107 struct buffer_head *bh;
1108 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109
1110 bh = __find_get_block(bdev, block, size);
1111 if (bh)
1112 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001113
Gioh Kim3b5e6452014-09-04 22:04:42 -04001114 ret = grow_buffers(bdev, block, size, gfp);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001115 if (ret < 0)
1116 return NULL;
1117 if (ret == 0)
1118 free_more_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 }
1120}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001121EXPORT_SYMBOL(__getblk_slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001122
1123/*
1124 * The relationship between dirty buffers and dirty pages:
1125 *
1126 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1127 * the page is tagged dirty in its radix tree.
1128 *
1129 * At all times, the dirtiness of the buffers represents the dirtiness of
1130 * subsections of the page. If the page has buffers, the page dirty bit is
1131 * merely a hint about the true dirty state.
1132 *
1133 * When a page is set dirty in its entirety, all its buffers are marked dirty
1134 * (if the page has buffers).
1135 *
1136 * When a buffer is marked dirty, its page is dirtied, but the page's other
1137 * buffers are not.
1138 *
1139 * Also. When blockdev buffers are explicitly read with bread(), they
1140 * individually become uptodate. But their backing page remains not
1141 * uptodate - even if all of its buffers are uptodate. A subsequent
1142 * block_read_full_page() against that page will discover all the uptodate
1143 * buffers, will set the page uptodate and will perform no I/O.
1144 */
1145
1146/**
1147 * mark_buffer_dirty - mark a buffer_head as needing writeout
Martin Waitz67be2dd2005-05-01 08:59:26 -07001148 * @bh: the buffer_head to mark dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149 *
1150 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1151 * backing page dirty, then tag the page as dirty in its address_space's radix
1152 * tree and then attach the address_space's inode to its superblock's dirty
1153 * inode list.
1154 *
1155 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
Dave Chinner250df6e2011-03-22 22:23:36 +11001156 * mapping->tree_lock and mapping->host->i_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001158void mark_buffer_dirty(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159{
Nick Piggin787d2212007-07-17 04:03:34 -07001160 WARN_ON_ONCE(!buffer_uptodate(bh));
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001161
Tejun Heo5305cb82013-01-11 13:06:36 -08001162 trace_block_dirty_buffer(bh);
1163
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001164 /*
1165 * Very *carefully* optimize the it-is-already-dirty case.
1166 *
1167 * Don't let the final "is it dirty" escape to before we
1168 * perhaps modified the buffer.
1169 */
1170 if (buffer_dirty(bh)) {
1171 smp_mb();
1172 if (buffer_dirty(bh))
1173 return;
1174 }
1175
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001176 if (!test_set_buffer_dirty(bh)) {
1177 struct page *page = bh->b_page;
Greg Thelenc4843a72015-05-22 17:13:16 -04001178 struct address_space *mapping = NULL;
1179 struct mem_cgroup *memcg;
1180
1181 memcg = mem_cgroup_begin_page_stat(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001182 if (!TestSetPageDirty(page)) {
Greg Thelenc4843a72015-05-22 17:13:16 -04001183 mapping = page_mapping(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001184 if (mapping)
Greg Thelenc4843a72015-05-22 17:13:16 -04001185 __set_page_dirty(page, mapping, memcg, 0);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001186 }
Greg Thelenc4843a72015-05-22 17:13:16 -04001187 mem_cgroup_end_page_stat(memcg);
1188 if (mapping)
1189 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001190 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001192EXPORT_SYMBOL(mark_buffer_dirty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194/*
1195 * Decrement a buffer_head's reference count. If all buffers against a page
1196 * have zero reference count, are clean and unlocked, and if the page is clean
1197 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1198 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1199 * a page but it ends up not being freed, and buffers may later be reattached).
1200 */
1201void __brelse(struct buffer_head * buf)
1202{
1203 if (atomic_read(&buf->b_count)) {
1204 put_bh(buf);
1205 return;
1206 }
Arjan van de Ven5c752ad2008-07-25 19:45:40 -07001207 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001209EXPORT_SYMBOL(__brelse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001210
1211/*
1212 * bforget() is like brelse(), except it discards any
1213 * potentially dirty data.
1214 */
1215void __bforget(struct buffer_head *bh)
1216{
1217 clear_buffer_dirty(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -08001218 if (bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219 struct address_space *buffer_mapping = bh->b_page->mapping;
1220
1221 spin_lock(&buffer_mapping->private_lock);
1222 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -07001223 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001224 spin_unlock(&buffer_mapping->private_lock);
1225 }
1226 __brelse(bh);
1227}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001228EXPORT_SYMBOL(__bforget);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229
1230static struct buffer_head *__bread_slow(struct buffer_head *bh)
1231{
1232 lock_buffer(bh);
1233 if (buffer_uptodate(bh)) {
1234 unlock_buffer(bh);
1235 return bh;
1236 } else {
1237 get_bh(bh);
1238 bh->b_end_io = end_buffer_read_sync;
1239 submit_bh(READ, bh);
1240 wait_on_buffer(bh);
1241 if (buffer_uptodate(bh))
1242 return bh;
1243 }
1244 brelse(bh);
1245 return NULL;
1246}
1247
1248/*
1249 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1250 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1251 * refcount elevated by one when they're in an LRU. A buffer can only appear
1252 * once in a particular CPU's LRU. A single buffer can be present in multiple
1253 * CPU's LRUs at the same time.
1254 *
1255 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1256 * sb_find_get_block().
1257 *
1258 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1259 * a local interrupt disable for that.
1260 */
1261
Sebastien Buisson86cf78d2014-10-09 15:29:38 -07001262#define BH_LRU_SIZE 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07001263
1264struct bh_lru {
1265 struct buffer_head *bhs[BH_LRU_SIZE];
1266};
1267
1268static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1269
1270#ifdef CONFIG_SMP
1271#define bh_lru_lock() local_irq_disable()
1272#define bh_lru_unlock() local_irq_enable()
1273#else
1274#define bh_lru_lock() preempt_disable()
1275#define bh_lru_unlock() preempt_enable()
1276#endif
1277
1278static inline void check_irqs_on(void)
1279{
1280#ifdef irqs_disabled
1281 BUG_ON(irqs_disabled());
1282#endif
1283}
1284
1285/*
1286 * The LRU management algorithm is dopey-but-simple. Sorry.
1287 */
1288static void bh_lru_install(struct buffer_head *bh)
1289{
1290 struct buffer_head *evictee = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291
1292 check_irqs_on();
1293 bh_lru_lock();
Christoph Lameterc7b92512010-12-06 11:16:28 -06001294 if (__this_cpu_read(bh_lrus.bhs[0]) != bh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295 struct buffer_head *bhs[BH_LRU_SIZE];
1296 int in;
1297 int out = 0;
1298
1299 get_bh(bh);
1300 bhs[out++] = bh;
1301 for (in = 0; in < BH_LRU_SIZE; in++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001302 struct buffer_head *bh2 =
1303 __this_cpu_read(bh_lrus.bhs[in]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001304
1305 if (bh2 == bh) {
1306 __brelse(bh2);
1307 } else {
1308 if (out >= BH_LRU_SIZE) {
1309 BUG_ON(evictee != NULL);
1310 evictee = bh2;
1311 } else {
1312 bhs[out++] = bh2;
1313 }
1314 }
1315 }
1316 while (out < BH_LRU_SIZE)
1317 bhs[out++] = NULL;
Christoph Lameterca6673b02013-12-03 17:32:53 -06001318 memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 }
1320 bh_lru_unlock();
1321
1322 if (evictee)
1323 __brelse(evictee);
1324}
1325
1326/*
1327 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1328 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001329static struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001330lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331{
1332 struct buffer_head *ret = NULL;
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001333 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334
1335 check_irqs_on();
1336 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337 for (i = 0; i < BH_LRU_SIZE; i++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001338 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339
Zach Brown9470dd52014-10-13 15:55:05 -07001340 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1341 bh->b_size == size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342 if (i) {
1343 while (i) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001344 __this_cpu_write(bh_lrus.bhs[i],
1345 __this_cpu_read(bh_lrus.bhs[i - 1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 i--;
1347 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06001348 __this_cpu_write(bh_lrus.bhs[0], bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001349 }
1350 get_bh(bh);
1351 ret = bh;
1352 break;
1353 }
1354 }
1355 bh_lru_unlock();
1356 return ret;
1357}
1358
1359/*
1360 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1361 * it in the LRU and mark it as accessed. If it is not present then return
1362 * NULL
1363 */
1364struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001365__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366{
1367 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1368
1369 if (bh == NULL) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001370 /* __find_get_block_slow will mark the page accessed */
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001371 bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 if (bh)
1373 bh_lru_install(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001374 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001375 touch_buffer(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001376
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return bh;
1378}
1379EXPORT_SYMBOL(__find_get_block);
1380
1381/*
Gioh Kim3b5e6452014-09-04 22:04:42 -04001382 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383 * which corresponds to the passed block_device, block and size. The
1384 * returned buffer has its reference count incremented.
1385 *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001386 * __getblk_gfp() will lock up the machine if grow_dev_page's
1387 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 */
1389struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001390__getblk_gfp(struct block_device *bdev, sector_t block,
1391 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392{
1393 struct buffer_head *bh = __find_get_block(bdev, block, size);
1394
1395 might_sleep();
1396 if (bh == NULL)
Gioh Kim3b5e6452014-09-04 22:04:42 -04001397 bh = __getblk_slow(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 return bh;
1399}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001400EXPORT_SYMBOL(__getblk_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401
1402/*
1403 * Do async read-ahead on a buffer..
1404 */
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001405void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001406{
1407 struct buffer_head *bh = __getblk(bdev, block, size);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001408 if (likely(bh)) {
1409 ll_rw_block(READA, 1, &bh);
1410 brelse(bh);
1411 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412}
1413EXPORT_SYMBOL(__breadahead);
1414
1415/**
Gioh Kim3b5e6452014-09-04 22:04:42 -04001416 * __bread_gfp() - reads a specified block and returns the bh
Martin Waitz67be2dd2005-05-01 08:59:26 -07001417 * @bdev: the block_device to read from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418 * @block: number of block
1419 * @size: size (in bytes) to read
Gioh Kim3b5e6452014-09-04 22:04:42 -04001420 * @gfp: page allocation flag
1421 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422 * Reads a specified block, and returns buffer head that contains it.
Gioh Kim3b5e6452014-09-04 22:04:42 -04001423 * The page cache can be allocated from non-movable area
1424 * not to prevent page migration if you set gfp to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 * It returns NULL if the block was unreadable.
1426 */
1427struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001428__bread_gfp(struct block_device *bdev, sector_t block,
1429 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
Gioh Kim3b5e6452014-09-04 22:04:42 -04001431 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432
Andrew Mortona3e713b2005-10-30 15:03:15 -08001433 if (likely(bh) && !buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 bh = __bread_slow(bh);
1435 return bh;
1436}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001437EXPORT_SYMBOL(__bread_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438
1439/*
1440 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1441 * This doesn't race because it runs in each cpu either in irq
1442 * or with preempt disabled.
1443 */
1444static void invalidate_bh_lru(void *arg)
1445{
1446 struct bh_lru *b = &get_cpu_var(bh_lrus);
1447 int i;
1448
1449 for (i = 0; i < BH_LRU_SIZE; i++) {
1450 brelse(b->bhs[i]);
1451 b->bhs[i] = NULL;
1452 }
1453 put_cpu_var(bh_lrus);
1454}
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001455
1456static bool has_bh_in_lru(int cpu, void *dummy)
1457{
1458 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1459 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001461 for (i = 0; i < BH_LRU_SIZE; i++) {
1462 if (b->bhs[i])
1463 return 1;
1464 }
1465
1466 return 0;
1467}
1468
Peter Zijlstraf9a14392007-05-06 14:49:55 -07001469void invalidate_bh_lrus(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470{
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001471 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472}
Nick Piggin9db55792008-02-08 04:19:49 -08001473EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001474
1475void set_bh_page(struct buffer_head *bh,
1476 struct page *page, unsigned long offset)
1477{
1478 bh->b_page = page;
Eric Sesterhenne827f922006-03-26 18:24:46 +02001479 BUG_ON(offset >= PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 if (PageHighMem(page))
1481 /*
1482 * This catches illegal uses and preserves the offset:
1483 */
1484 bh->b_data = (char *)(0 + offset);
1485 else
1486 bh->b_data = page_address(page) + offset;
1487}
1488EXPORT_SYMBOL(set_bh_page);
1489
1490/*
1491 * Called when truncating a buffer on a page completely.
1492 */
Mel Gormane7470ee2014-06-04 16:10:29 -07001493
1494/* Bits that are cleared during an invalidate */
1495#define BUFFER_FLAGS_DISCARD \
1496 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1497 1 << BH_Delay | 1 << BH_Unwritten)
1498
Arjan van de Ven858119e2006-01-14 13:20:43 -08001499static void discard_buffer(struct buffer_head * bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500{
Mel Gormane7470ee2014-06-04 16:10:29 -07001501 unsigned long b_state, b_state_old;
1502
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 lock_buffer(bh);
1504 clear_buffer_dirty(bh);
1505 bh->b_bdev = NULL;
Mel Gormane7470ee2014-06-04 16:10:29 -07001506 b_state = bh->b_state;
1507 for (;;) {
1508 b_state_old = cmpxchg(&bh->b_state, b_state,
1509 (b_state & ~BUFFER_FLAGS_DISCARD));
1510 if (b_state_old == b_state)
1511 break;
1512 b_state = b_state_old;
1513 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 unlock_buffer(bh);
1515}
1516
1517/**
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001518 * block_invalidatepage - invalidate part or all of a buffer-backed page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519 *
1520 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -04001521 * @offset: start of the range to invalidate
1522 * @length: length of the range to invalidate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 *
1524 * block_invalidatepage() is called when all or part of the page has become
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001525 * invalidated by a truncate operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001526 *
1527 * block_invalidatepage() does not have to release all buffers, but it must
1528 * ensure that no dirty buffer is left outside @offset and that no I/O
1529 * is underway against any of the blocks which are outside the truncation
1530 * point. Because the caller is about to free (and possibly reuse) those
1531 * blocks on-disk.
1532 */
Lukas Czernerd47992f2013-05-21 23:17:23 -04001533void block_invalidatepage(struct page *page, unsigned int offset,
1534 unsigned int length)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535{
1536 struct buffer_head *head, *bh, *next;
1537 unsigned int curr_off = 0;
Lukas Czernerd47992f2013-05-21 23:17:23 -04001538 unsigned int stop = length + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539
1540 BUG_ON(!PageLocked(page));
1541 if (!page_has_buffers(page))
1542 goto out;
1543
Lukas Czernerd47992f2013-05-21 23:17:23 -04001544 /*
1545 * Check for overflow
1546 */
1547 BUG_ON(stop > PAGE_CACHE_SIZE || stop < length);
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549 head = page_buffers(page);
1550 bh = head;
1551 do {
1552 unsigned int next_off = curr_off + bh->b_size;
1553 next = bh->b_this_page;
1554
1555 /*
Lukas Czernerd47992f2013-05-21 23:17:23 -04001556 * Are we still fully in range ?
1557 */
1558 if (next_off > stop)
1559 goto out;
1560
1561 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562 * is this block fully invalidated?
1563 */
1564 if (offset <= curr_off)
1565 discard_buffer(bh);
1566 curr_off = next_off;
1567 bh = next;
1568 } while (bh != head);
1569
1570 /*
1571 * We release buffers only if the entire page is being invalidated.
1572 * The get_block cached value has been unconditionally invalidated,
1573 * so real IO is not possible anymore.
1574 */
1575 if (offset == 0)
NeilBrown2ff28e22006-03-26 01:37:18 -08001576 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001577out:
NeilBrown2ff28e22006-03-26 01:37:18 -08001578 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579}
1580EXPORT_SYMBOL(block_invalidatepage);
1581
Lukas Czernerd47992f2013-05-21 23:17:23 -04001582
Linus Torvalds1da177e2005-04-16 15:20:36 -07001583/*
1584 * We attach and possibly dirty the buffers atomically wrt
1585 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1586 * is already excluded via the page lock.
1587 */
1588void create_empty_buffers(struct page *page,
1589 unsigned long blocksize, unsigned long b_state)
1590{
1591 struct buffer_head *bh, *head, *tail;
1592
1593 head = alloc_page_buffers(page, blocksize, 1);
1594 bh = head;
1595 do {
1596 bh->b_state |= b_state;
1597 tail = bh;
1598 bh = bh->b_this_page;
1599 } while (bh);
1600 tail->b_this_page = head;
1601
1602 spin_lock(&page->mapping->private_lock);
1603 if (PageUptodate(page) || PageDirty(page)) {
1604 bh = head;
1605 do {
1606 if (PageDirty(page))
1607 set_buffer_dirty(bh);
1608 if (PageUptodate(page))
1609 set_buffer_uptodate(bh);
1610 bh = bh->b_this_page;
1611 } while (bh != head);
1612 }
1613 attach_page_buffers(page, head);
1614 spin_unlock(&page->mapping->private_lock);
1615}
1616EXPORT_SYMBOL(create_empty_buffers);
1617
1618/*
1619 * We are taking a block for data and we don't want any output from any
1620 * buffer-cache aliases starting from return from that function and
1621 * until the moment when something will explicitly mark the buffer
1622 * dirty (hopefully that will not happen until we will free that block ;-)
1623 * We don't even need to mark it not-uptodate - nobody can expect
1624 * anything from a newly allocated buffer anyway. We used to used
1625 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1626 * don't want to mark the alias unmapped, for example - it would confuse
1627 * anyone who might pick it with bread() afterwards...
1628 *
1629 * Also.. Note that bforget() doesn't lock the buffer. So there can
1630 * be writeout I/O going on against recently-freed buffers. We don't
1631 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1632 * only if we really need to. That happens here.
1633 */
1634void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
1635{
1636 struct buffer_head *old_bh;
1637
1638 might_sleep();
1639
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001640 old_bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001641 if (old_bh) {
1642 clear_buffer_dirty(old_bh);
1643 wait_on_buffer(old_bh);
1644 clear_buffer_req(old_bh);
1645 __brelse(old_bh);
1646 }
1647}
1648EXPORT_SYMBOL(unmap_underlying_metadata);
1649
1650/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001651 * Size is a power-of-two in the range 512..PAGE_SIZE,
1652 * and the case we care about most is PAGE_SIZE.
1653 *
1654 * So this *could* possibly be written with those
1655 * constraints in mind (relevant mostly if some
1656 * architecture has a slow bit-scan instruction)
1657 */
1658static inline int block_size_bits(unsigned int blocksize)
1659{
1660 return ilog2(blocksize);
1661}
1662
1663static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1664{
1665 BUG_ON(!PageLocked(page));
1666
1667 if (!page_has_buffers(page))
1668 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1669 return page_buffers(page);
1670}
1671
1672/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 * NOTE! All mapped/uptodate combinations are valid:
1674 *
1675 * Mapped Uptodate Meaning
1676 *
1677 * No No "unknown" - must do get_block()
1678 * No Yes "hole" - zero-filled
1679 * Yes No "allocated" - allocated on disk, not read in
1680 * Yes Yes "valid" - allocated and up-to-date in memory.
1681 *
1682 * "Dirty" is valid only with the last case (mapped+uptodate).
1683 */
1684
1685/*
1686 * While block_write_full_page is writing back the dirty buffers under
1687 * the page lock, whoever dirtied the buffers may decide to clean them
1688 * again at any time. We handle that by only looking at the buffer
1689 * state inside lock_buffer().
1690 *
1691 * If block_write_full_page() is called for regular writeback
1692 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1693 * locked buffer. This only can happen if someone has written the buffer
1694 * directly, with submit_bh(). At the address_space level PageWriteback
1695 * prevents this contention from occurring.
Theodore Ts'o6e34eed2009-04-07 18:12:43 -04001696 *
1697 * If block_write_full_page() is called with wbc->sync_mode ==
Jens Axboe721a9602011-03-09 11:56:30 +01001698 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1699 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 */
1701static int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001702 get_block_t *get_block, struct writeback_control *wbc,
1703 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
1705 int err;
1706 sector_t block;
1707 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001708 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001709 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001710 int nr_underway = 0;
Tejun Heobafc0db2015-06-02 08:37:23 -06001711 int write_op = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : WRITE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001713 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715
1716 /*
1717 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1718 * here, and the (potentially unmapped) buffers may become dirty at
1719 * any time. If a buffer becomes dirty here after we've inspected it
1720 * then we just miss that fact, and the page stays dirty.
1721 *
1722 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1723 * handle that here by just cleaning them.
1724 */
1725
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001727 blocksize = bh->b_size;
1728 bbits = block_size_bits(blocksize);
1729
1730 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1731 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732
1733 /*
1734 * Get all the dirty buffers mapped to disk addresses and
1735 * handle any aliases from the underlying blockdev's mapping.
1736 */
1737 do {
1738 if (block > last_block) {
1739 /*
1740 * mapped buffers outside i_size will occur, because
1741 * this page can be outside i_size when there is a
1742 * truncate in progress.
1743 */
1744 /*
1745 * The buffer was zeroed by block_write_full_page()
1746 */
1747 clear_buffer_dirty(bh);
1748 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001749 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1750 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001751 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 err = get_block(inode, block, bh, 1);
1753 if (err)
1754 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001755 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756 if (buffer_new(bh)) {
1757 /* blockdev mappings never come here */
1758 clear_buffer_new(bh);
1759 unmap_underlying_metadata(bh->b_bdev,
1760 bh->b_blocknr);
1761 }
1762 }
1763 bh = bh->b_this_page;
1764 block++;
1765 } while (bh != head);
1766
1767 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 if (!buffer_mapped(bh))
1769 continue;
1770 /*
1771 * If it's a fully non-blocking write attempt and we cannot
1772 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001773 * potentially cause a busy-wait loop from writeback threads
1774 * and kswapd activity, but those code paths have their own
1775 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001777 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001779 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001780 redirty_page_for_writepage(wbc, page);
1781 continue;
1782 }
1783 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001784 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 } else {
1786 unlock_buffer(bh);
1787 }
1788 } while ((bh = bh->b_this_page) != head);
1789
1790 /*
1791 * The page and its buffers are protected by PageWriteback(), so we can
1792 * drop the bh refcounts early.
1793 */
1794 BUG_ON(PageWriteback(page));
1795 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
1797 do {
1798 struct buffer_head *next = bh->b_this_page;
1799 if (buffer_async_write(bh)) {
Tejun Heob16b1de2015-06-02 08:39:48 -06001800 submit_bh_wbc(write_op, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 nr_underway++;
1802 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 bh = next;
1804 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001805 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806
1807 err = 0;
1808done:
1809 if (nr_underway == 0) {
1810 /*
1811 * The page was marked dirty, but the buffers were
1812 * clean. Someone wrote them back by hand with
1813 * ll_rw_block/submit_bh. A rare case.
1814 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001815 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001816
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 /*
1818 * The page and buffer_heads can be released at any time from
1819 * here on.
1820 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001821 }
1822 return err;
1823
1824recover:
1825 /*
1826 * ENOSPC, or some other error. We may already have added some
1827 * blocks to the file, so we need to write these out to avoid
1828 * exposing stale data.
1829 * The page is currently locked and not marked for writeback
1830 */
1831 bh = head;
1832 /* Recovery: lock and submit the mapped buffers */
1833 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001834 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1835 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001837 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 } else {
1839 /*
1840 * The buffer may have been set dirty during
1841 * attachment to a dirty page.
1842 */
1843 clear_buffer_dirty(bh);
1844 }
1845 } while ((bh = bh->b_this_page) != head);
1846 SetPageError(page);
1847 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001848 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 do {
1851 struct buffer_head *next = bh->b_this_page;
1852 if (buffer_async_write(bh)) {
1853 clear_buffer_dirty(bh);
Tejun Heob16b1de2015-06-02 08:39:48 -06001854 submit_bh_wbc(write_op, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 nr_underway++;
1856 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 bh = next;
1858 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001859 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001860 goto done;
1861}
1862
Nick Pigginafddba42007-10-16 01:25:01 -07001863/*
1864 * If a page has any new buffers, zero them out here, and mark them uptodate
1865 * and dirty so they'll be written out (in order to prevent uninitialised
1866 * block data from leaking). And clear the new bit.
1867 */
1868void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1869{
1870 unsigned int block_start, block_end;
1871 struct buffer_head *head, *bh;
1872
1873 BUG_ON(!PageLocked(page));
1874 if (!page_has_buffers(page))
1875 return;
1876
1877 bh = head = page_buffers(page);
1878 block_start = 0;
1879 do {
1880 block_end = block_start + bh->b_size;
1881
1882 if (buffer_new(bh)) {
1883 if (block_end > from && block_start < to) {
1884 if (!PageUptodate(page)) {
1885 unsigned start, size;
1886
1887 start = max(from, block_start);
1888 size = min(to, block_end) - start;
1889
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001890 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001891 set_buffer_uptodate(bh);
1892 }
1893
1894 clear_buffer_new(bh);
1895 mark_buffer_dirty(bh);
1896 }
1897 }
1898
1899 block_start = block_end;
1900 bh = bh->b_this_page;
1901 } while (bh != head);
1902}
1903EXPORT_SYMBOL(page_zero_new_buffers);
1904
Christoph Hellwigebdec242010-10-06 10:47:23 +02001905int __block_write_begin(struct page *page, loff_t pos, unsigned len,
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001906 get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001907{
Christoph Hellwigebdec242010-10-06 10:47:23 +02001908 unsigned from = pos & (PAGE_CACHE_SIZE - 1);
1909 unsigned to = from + len;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001910 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911 unsigned block_start, block_end;
1912 sector_t block;
1913 int err = 0;
1914 unsigned blocksize, bbits;
1915 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1916
1917 BUG_ON(!PageLocked(page));
1918 BUG_ON(from > PAGE_CACHE_SIZE);
1919 BUG_ON(to > PAGE_CACHE_SIZE);
1920 BUG_ON(from > to);
1921
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001922 head = create_page_buffers(page, inode, 0);
1923 blocksize = head->b_size;
1924 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925
Linus Torvalds1da177e2005-04-16 15:20:36 -07001926 block = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
1927
1928 for(bh = head, block_start = 0; bh != head || !block_start;
1929 block++, block_start=block_end, bh = bh->b_this_page) {
1930 block_end = block_start + blocksize;
1931 if (block_end <= from || block_start >= to) {
1932 if (PageUptodate(page)) {
1933 if (!buffer_uptodate(bh))
1934 set_buffer_uptodate(bh);
1935 }
1936 continue;
1937 }
1938 if (buffer_new(bh))
1939 clear_buffer_new(bh);
1940 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001941 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942 err = get_block(inode, block, bh, 1);
1943 if (err)
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07001944 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001945 if (buffer_new(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946 unmap_underlying_metadata(bh->b_bdev,
1947 bh->b_blocknr);
1948 if (PageUptodate(page)) {
Nick Piggin637aff42007-10-16 01:25:00 -07001949 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07001951 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952 continue;
1953 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001954 if (block_end > to || block_start < from)
1955 zero_user_segments(page,
1956 to, block_end,
1957 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001958 continue;
1959 }
1960 }
1961 if (PageUptodate(page)) {
1962 if (!buffer_uptodate(bh))
1963 set_buffer_uptodate(bh);
1964 continue;
1965 }
1966 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08001967 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968 (block_start < from || block_end > to)) {
1969 ll_rw_block(READ, 1, &bh);
1970 *wait_bh++=bh;
1971 }
1972 }
1973 /*
1974 * If we issued read requests - let them complete.
1975 */
1976 while(wait_bh > wait) {
1977 wait_on_buffer(*--wait_bh);
1978 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07001979 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001980 }
Jan Karaf9f07b62011-06-14 00:58:27 +02001981 if (unlikely(err))
Nick Pigginafddba42007-10-16 01:25:01 -07001982 page_zero_new_buffers(page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 return err;
1984}
Christoph Hellwigebdec242010-10-06 10:47:23 +02001985EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
1987static int __block_commit_write(struct inode *inode, struct page *page,
1988 unsigned from, unsigned to)
1989{
1990 unsigned block_start, block_end;
1991 int partial = 0;
1992 unsigned blocksize;
1993 struct buffer_head *bh, *head;
1994
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001995 bh = head = page_buffers(page);
1996 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001997
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001998 block_start = 0;
1999 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 block_end = block_start + blocksize;
2001 if (block_end <= from || block_start >= to) {
2002 if (!buffer_uptodate(bh))
2003 partial = 1;
2004 } else {
2005 set_buffer_uptodate(bh);
2006 mark_buffer_dirty(bh);
2007 }
Nick Pigginafddba42007-10-16 01:25:01 -07002008 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002009
2010 block_start = block_end;
2011 bh = bh->b_this_page;
2012 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013
2014 /*
2015 * If this is a partial write which happened to make all buffers
2016 * uptodate then we can optimize away a bogus readpage() for
2017 * the next read(). Here we 'discover' whether the page went
2018 * uptodate as a result of this (potentially partial) write.
2019 */
2020 if (!partial)
2021 SetPageUptodate(page);
2022 return 0;
2023}
2024
2025/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02002026 * block_write_begin takes care of the basic task of block allocation and
2027 * bringing partial write blocks uptodate first.
2028 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002029 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07002030 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02002031int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2032 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07002033{
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002034 pgoff_t index = pos >> PAGE_CACHE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07002035 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002036 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07002037
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002038 page = grab_cache_page_write_begin(mapping, index, flags);
2039 if (!page)
2040 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07002041
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002042 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07002043 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002044 unlock_page(page);
2045 page_cache_release(page);
2046 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07002047 }
2048
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002049 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07002050 return status;
2051}
2052EXPORT_SYMBOL(block_write_begin);
2053
2054int block_write_end(struct file *file, struct address_space *mapping,
2055 loff_t pos, unsigned len, unsigned copied,
2056 struct page *page, void *fsdata)
2057{
2058 struct inode *inode = mapping->host;
2059 unsigned start;
2060
2061 start = pos & (PAGE_CACHE_SIZE - 1);
2062
2063 if (unlikely(copied < len)) {
2064 /*
2065 * The buffers that were written will now be uptodate, so we
2066 * don't have to worry about a readpage reading them and
2067 * overwriting a partial write. However if we have encountered
2068 * a short write and only partially written into a buffer, it
2069 * will not be marked uptodate, so a readpage might come in and
2070 * destroy our partial write.
2071 *
2072 * Do the simplest thing, and just treat any short write to a
2073 * non uptodate page as a zero-length write, and force the
2074 * caller to redo the whole thing.
2075 */
2076 if (!PageUptodate(page))
2077 copied = 0;
2078
2079 page_zero_new_buffers(page, start+copied, start+len);
2080 }
2081 flush_dcache_page(page);
2082
2083 /* This could be a short (even 0-length) commit */
2084 __block_commit_write(inode, page, start, start+copied);
2085
2086 return copied;
2087}
2088EXPORT_SYMBOL(block_write_end);
2089
2090int generic_write_end(struct file *file, struct address_space *mapping,
2091 loff_t pos, unsigned len, unsigned copied,
2092 struct page *page, void *fsdata)
2093{
2094 struct inode *inode = mapping->host;
Jan Kara90a80202014-10-01 21:49:18 -04002095 loff_t old_size = inode->i_size;
Jan Karac7d206b2008-07-11 19:27:31 -04002096 int i_size_changed = 0;
Nick Pigginafddba42007-10-16 01:25:01 -07002097
2098 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2099
2100 /*
2101 * No need to use i_size_read() here, the i_size
2102 * cannot change under us because we hold i_mutex.
2103 *
2104 * But it's important to update i_size while still holding page lock:
2105 * page writeout could otherwise come in and zero beyond i_size.
2106 */
2107 if (pos+copied > inode->i_size) {
2108 i_size_write(inode, pos+copied);
Jan Karac7d206b2008-07-11 19:27:31 -04002109 i_size_changed = 1;
Nick Pigginafddba42007-10-16 01:25:01 -07002110 }
2111
2112 unlock_page(page);
2113 page_cache_release(page);
2114
Jan Kara90a80202014-10-01 21:49:18 -04002115 if (old_size < pos)
2116 pagecache_isize_extended(inode, old_size, pos);
Jan Karac7d206b2008-07-11 19:27:31 -04002117 /*
2118 * Don't mark the inode dirty under page lock. First, it unnecessarily
2119 * makes the holding time of page lock longer. Second, it forces lock
2120 * ordering of page lock and transaction start for journaling
2121 * filesystems.
2122 */
2123 if (i_size_changed)
2124 mark_inode_dirty(inode);
2125
Nick Pigginafddba42007-10-16 01:25:01 -07002126 return copied;
2127}
2128EXPORT_SYMBOL(generic_write_end);
2129
2130/*
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002131 * block_is_partially_uptodate checks whether buffers within a page are
2132 * uptodate or not.
2133 *
2134 * Returns true if all buffers which correspond to a file portion
2135 * we want to read are uptodate.
2136 */
Al Viroc186afb42014-02-02 21:16:54 -05002137int block_is_partially_uptodate(struct page *page, unsigned long from,
2138 unsigned long count)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002139{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002140 unsigned block_start, block_end, blocksize;
2141 unsigned to;
2142 struct buffer_head *bh, *head;
2143 int ret = 1;
2144
2145 if (!page_has_buffers(page))
2146 return 0;
2147
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002148 head = page_buffers(page);
2149 blocksize = head->b_size;
Al Viroc186afb42014-02-02 21:16:54 -05002150 to = min_t(unsigned, PAGE_CACHE_SIZE - from, count);
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002151 to = from + to;
2152 if (from < blocksize && to > PAGE_CACHE_SIZE - blocksize)
2153 return 0;
2154
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002155 bh = head;
2156 block_start = 0;
2157 do {
2158 block_end = block_start + blocksize;
2159 if (block_end > from && block_start < to) {
2160 if (!buffer_uptodate(bh)) {
2161 ret = 0;
2162 break;
2163 }
2164 if (block_end >= to)
2165 break;
2166 }
2167 block_start = block_end;
2168 bh = bh->b_this_page;
2169 } while (bh != head);
2170
2171 return ret;
2172}
2173EXPORT_SYMBOL(block_is_partially_uptodate);
2174
2175/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002176 * Generic "read page" function for block devices that have the normal
2177 * get_block functionality. This is most of the block device filesystems.
2178 * Reads the page asynchronously --- the unlock_buffer() and
2179 * set/clear_buffer_uptodate() functions propagate buffer state into the
2180 * page struct once IO has completed.
2181 */
2182int block_read_full_page(struct page *page, get_block_t *get_block)
2183{
2184 struct inode *inode = page->mapping->host;
2185 sector_t iblock, lblock;
2186 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002187 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002188 int nr, i;
2189 int fully_mapped = 1;
2190
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002191 head = create_page_buffers(page, inode, 0);
2192 blocksize = head->b_size;
2193 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002194
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002195 iblock = (sector_t)page->index << (PAGE_CACHE_SHIFT - bbits);
2196 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 bh = head;
2198 nr = 0;
2199 i = 0;
2200
2201 do {
2202 if (buffer_uptodate(bh))
2203 continue;
2204
2205 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002206 int err = 0;
2207
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 fully_mapped = 0;
2209 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002210 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002211 err = get_block(inode, iblock, bh, 0);
2212 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213 SetPageError(page);
2214 }
2215 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002216 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002217 if (!err)
2218 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002219 continue;
2220 }
2221 /*
2222 * get_block() might have updated the buffer
2223 * synchronously
2224 */
2225 if (buffer_uptodate(bh))
2226 continue;
2227 }
2228 arr[nr++] = bh;
2229 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2230
2231 if (fully_mapped)
2232 SetPageMappedToDisk(page);
2233
2234 if (!nr) {
2235 /*
2236 * All buffers are uptodate - we can set the page uptodate
2237 * as well. But not if get_block() returned an error.
2238 */
2239 if (!PageError(page))
2240 SetPageUptodate(page);
2241 unlock_page(page);
2242 return 0;
2243 }
2244
2245 /* Stage two: lock the buffers */
2246 for (i = 0; i < nr; i++) {
2247 bh = arr[i];
2248 lock_buffer(bh);
2249 mark_buffer_async_read(bh);
2250 }
2251
2252 /*
2253 * Stage 3: start the IO. Check for uptodateness
2254 * inside the buffer lock in case another process reading
2255 * the underlying blockdev brought it uptodate (the sct fix).
2256 */
2257 for (i = 0; i < nr; i++) {
2258 bh = arr[i];
2259 if (buffer_uptodate(bh))
2260 end_buffer_async_read(bh, 1);
2261 else
2262 submit_bh(READ, bh);
2263 }
2264 return 0;
2265}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002266EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267
2268/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002269 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 * deal with the hole.
2271 */
Nick Piggin89e10782007-10-16 01:25:07 -07002272int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273{
2274 struct address_space *mapping = inode->i_mapping;
2275 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002276 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002277 int err;
2278
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002279 err = inode_newsize_ok(inode, size);
2280 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281 goto out;
2282
Nick Piggin89e10782007-10-16 01:25:07 -07002283 err = pagecache_write_begin(NULL, mapping, size, 0,
2284 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2285 &page, &fsdata);
2286 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002287 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002288
Nick Piggin89e10782007-10-16 01:25:07 -07002289 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2290 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002291
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292out:
2293 return err;
2294}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002295EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002297static int cont_expand_zero(struct file *file, struct address_space *mapping,
2298 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002299{
Nick Piggin89e10782007-10-16 01:25:07 -07002300 struct inode *inode = mapping->host;
2301 unsigned blocksize = 1 << inode->i_blkbits;
2302 struct page *page;
2303 void *fsdata;
2304 pgoff_t index, curidx;
2305 loff_t curpos;
2306 unsigned zerofrom, offset, len;
2307 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002308
Nick Piggin89e10782007-10-16 01:25:07 -07002309 index = pos >> PAGE_CACHE_SHIFT;
2310 offset = pos & ~PAGE_CACHE_MASK;
2311
2312 while (index > (curidx = (curpos = *bytes)>>PAGE_CACHE_SHIFT)) {
2313 zerofrom = curpos & ~PAGE_CACHE_MASK;
2314 if (zerofrom & (blocksize-1)) {
2315 *bytes |= (blocksize-1);
2316 (*bytes)++;
2317 }
2318 len = PAGE_CACHE_SIZE - zerofrom;
2319
2320 err = pagecache_write_begin(file, mapping, curpos, len,
2321 AOP_FLAG_UNINTERRUPTIBLE,
2322 &page, &fsdata);
2323 if (err)
2324 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002325 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002326 err = pagecache_write_end(file, mapping, curpos, len, len,
2327 page, fsdata);
2328 if (err < 0)
2329 goto out;
2330 BUG_ON(err != len);
2331 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002332
2333 balance_dirty_pages_ratelimited(mapping);
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002334
2335 if (unlikely(fatal_signal_pending(current))) {
2336 err = -EINTR;
2337 goto out;
2338 }
Nick Piggin89e10782007-10-16 01:25:07 -07002339 }
2340
2341 /* page covers the boundary, find the boundary offset */
2342 if (index == curidx) {
2343 zerofrom = curpos & ~PAGE_CACHE_MASK;
2344 /* if we will expand the thing last block will be filled */
2345 if (offset <= zerofrom) {
2346 goto out;
2347 }
2348 if (zerofrom & (blocksize-1)) {
2349 *bytes |= (blocksize-1);
2350 (*bytes)++;
2351 }
2352 len = offset - zerofrom;
2353
2354 err = pagecache_write_begin(file, mapping, curpos, len,
2355 AOP_FLAG_UNINTERRUPTIBLE,
2356 &page, &fsdata);
2357 if (err)
2358 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002359 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002360 err = pagecache_write_end(file, mapping, curpos, len, len,
2361 page, fsdata);
2362 if (err < 0)
2363 goto out;
2364 BUG_ON(err != len);
2365 err = 0;
2366 }
2367out:
2368 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002369}
2370
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371/*
2372 * For moronic filesystems that do not allow holes in file.
2373 * We may have to extend the file.
2374 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002375int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002376 loff_t pos, unsigned len, unsigned flags,
2377 struct page **pagep, void **fsdata,
2378 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002379{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002380 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002381 unsigned blocksize = 1 << inode->i_blkbits;
Nick Piggin89e10782007-10-16 01:25:07 -07002382 unsigned zerofrom;
2383 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384
Nick Piggin89e10782007-10-16 01:25:07 -07002385 err = cont_expand_zero(file, mapping, pos, bytes);
2386 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002387 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002388
2389 zerofrom = *bytes & ~PAGE_CACHE_MASK;
2390 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2391 *bytes |= (blocksize-1);
2392 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393 }
2394
Christoph Hellwig155130a2010-06-04 11:29:58 +02002395 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002396}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002397EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002398
Linus Torvalds1da177e2005-04-16 15:20:36 -07002399int block_commit_write(struct page *page, unsigned from, unsigned to)
2400{
2401 struct inode *inode = page->mapping->host;
2402 __block_commit_write(inode,page,from,to);
2403 return 0;
2404}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002405EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002406
David Chinner54171692007-07-19 17:39:55 +10002407/*
2408 * block_page_mkwrite() is not allowed to change the file size as it gets
2409 * called from a page fault handler when a page is first dirtied. Hence we must
2410 * be careful to check for EOF conditions here. We set the page up correctly
2411 * for a written page which means we get ENOSPC checking when writing into
2412 * holes and correct delalloc and unwritten extent mapping on filesystems that
2413 * support these features.
2414 *
2415 * We are not allowed to take the i_mutex here so we have to play games to
2416 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002417 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002418 * page lock we can determine safely if the page is beyond EOF. If it is not
2419 * beyond EOF, then the page is guaranteed safe against truncation until we
2420 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002421 *
Jan Kara14da9202012-06-12 16:20:37 +02002422 * Direct callers of this function should protect against filesystem freezing
2423 * using sb_start_write() - sb_end_write() functions.
David Chinner54171692007-07-19 17:39:55 +10002424 */
Jan Kara24da4fa2011-05-24 00:23:34 +02002425int __block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2426 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002427{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002428 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002429 struct inode *inode = file_inode(vma->vm_file);
David Chinner54171692007-07-19 17:39:55 +10002430 unsigned long end;
2431 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002432 int ret;
David Chinner54171692007-07-19 17:39:55 +10002433
2434 lock_page(page);
2435 size = i_size_read(inode);
2436 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002437 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002438 /* We overload EFAULT to mean page got truncated */
2439 ret = -EFAULT;
2440 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002441 }
2442
2443 /* page is wholly or partially inside EOF */
2444 if (((page->index + 1) << PAGE_CACHE_SHIFT) > size)
2445 end = size & ~PAGE_CACHE_MASK;
2446 else
2447 end = PAGE_CACHE_SIZE;
2448
Christoph Hellwigebdec242010-10-06 10:47:23 +02002449 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002450 if (!ret)
2451 ret = block_commit_write(page, 0, end);
2452
Jan Kara24da4fa2011-05-24 00:23:34 +02002453 if (unlikely(ret < 0))
2454 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002455 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002456 wait_for_stable_page(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002457 return 0;
2458out_unlock:
2459 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002460 return ret;
2461}
Jan Kara24da4fa2011-05-24 00:23:34 +02002462EXPORT_SYMBOL(__block_page_mkwrite);
2463
2464int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
2465 get_block_t get_block)
2466{
Jan Karaea13a862011-05-24 00:23:35 +02002467 int ret;
Al Viro496ad9a2013-01-23 17:07:38 -05002468 struct super_block *sb = file_inode(vma->vm_file)->i_sb;
Jan Kara24da4fa2011-05-24 00:23:34 +02002469
Jan Kara14da9202012-06-12 16:20:37 +02002470 sb_start_pagefault(sb);
Theodore Ts'o041bbb6d2012-09-30 23:04:56 -04002471
2472 /*
2473 * Update file times before taking page lock. We may end up failing the
2474 * fault so this update may be superfluous but who really cares...
2475 */
2476 file_update_time(vma->vm_file);
2477
Jan Karaea13a862011-05-24 00:23:35 +02002478 ret = __block_page_mkwrite(vma, vmf, get_block);
Jan Kara14da9202012-06-12 16:20:37 +02002479 sb_end_pagefault(sb);
Jan Kara24da4fa2011-05-24 00:23:34 +02002480 return block_page_mkwrite_return(ret);
2481}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002482EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002483
2484/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002485 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002486 * immediately, while under the page lock. So it needs a special end_io
2487 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002488 */
2489static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2490{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002491 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492}
2493
2494/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002495 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2496 * the page (converting it to circular linked list and taking care of page
2497 * dirty races).
2498 */
2499static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2500{
2501 struct buffer_head *bh;
2502
2503 BUG_ON(!PageLocked(page));
2504
2505 spin_lock(&page->mapping->private_lock);
2506 bh = head;
2507 do {
2508 if (PageDirty(page))
2509 set_buffer_dirty(bh);
2510 if (!bh->b_this_page)
2511 bh->b_this_page = head;
2512 bh = bh->b_this_page;
2513 } while (bh != head);
2514 attach_page_buffers(page, head);
2515 spin_unlock(&page->mapping->private_lock);
2516}
2517
2518/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002519 * On entry, the page is fully not uptodate.
2520 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002521 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002523int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002524 loff_t pos, unsigned len, unsigned flags,
2525 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002526 get_block_t *get_block)
2527{
Nick Piggin03158cd2007-10-16 01:25:25 -07002528 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002529 const unsigned blkbits = inode->i_blkbits;
2530 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002531 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002532 struct page *page;
2533 pgoff_t index;
2534 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002535 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002536 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002537 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002538 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539 int ret = 0;
2540 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002541
Nick Piggin03158cd2007-10-16 01:25:25 -07002542 index = pos >> PAGE_CACHE_SHIFT;
2543 from = pos & (PAGE_CACHE_SIZE - 1);
2544 to = from + len;
2545
Nick Piggin54566b22009-01-04 12:00:53 -08002546 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002547 if (!page)
2548 return -ENOMEM;
2549 *pagep = page;
2550 *fsdata = NULL;
2551
2552 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002553 ret = __block_write_begin(page, pos, len, get_block);
2554 if (unlikely(ret))
2555 goto out_release;
2556 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002557 }
Nick Piggina4b06722007-10-16 01:24:48 -07002558
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 if (PageMappedToDisk(page))
2560 return 0;
2561
Nick Piggina4b06722007-10-16 01:24:48 -07002562 /*
2563 * Allocate buffers so that we can keep track of state, and potentially
2564 * attach them to the page if an error occurs. In the common case of
2565 * no error, they will just be freed again without ever being attached
2566 * to the page (which is all OK, because we're under the page lock).
2567 *
2568 * Be careful: the buffer linked list is a NULL terminated one, rather
2569 * than the circular one we're used to.
2570 */
2571 head = alloc_page_buffers(page, blocksize, 0);
Nick Piggin03158cd2007-10-16 01:25:25 -07002572 if (!head) {
2573 ret = -ENOMEM;
2574 goto out_release;
2575 }
Nick Piggina4b06722007-10-16 01:24:48 -07002576
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 block_in_file = (sector_t)page->index << (PAGE_CACHE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578
2579 /*
2580 * We loop across all blocks in the page, whether or not they are
2581 * part of the affected region. This is so we can discover if the
2582 * page is fully mapped-to-disk.
2583 */
Nick Piggina4b06722007-10-16 01:24:48 -07002584 for (block_start = 0, block_in_page = 0, bh = head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 block_start < PAGE_CACHE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002586 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 int create;
2588
Nick Piggina4b06722007-10-16 01:24:48 -07002589 block_end = block_start + blocksize;
2590 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 create = 1;
2592 if (block_start >= to)
2593 create = 0;
2594 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002595 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596 if (ret)
2597 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002598 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002600 if (buffer_new(bh))
2601 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
2602 if (PageUptodate(page)) {
2603 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002604 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002605 }
2606 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002607 zero_user_segments(page, block_start, from,
2608 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 continue;
2610 }
Nick Piggina4b06722007-10-16 01:24:48 -07002611 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 continue; /* reiserfs does this */
2613 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002614 lock_buffer(bh);
2615 bh->b_end_io = end_buffer_read_nobh;
2616 submit_bh(READ, bh);
2617 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618 }
2619 }
2620
2621 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622 /*
2623 * The page is locked, so these buffers are protected from
2624 * any VM or truncate activity. Hence we don't need to care
2625 * for the buffer_head refcounts.
2626 */
Nick Piggina4b06722007-10-16 01:24:48 -07002627 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628 wait_on_buffer(bh);
2629 if (!buffer_uptodate(bh))
2630 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002631 }
2632 if (ret)
2633 goto failed;
2634 }
2635
2636 if (is_mapped_to_disk)
2637 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638
Nick Piggin03158cd2007-10-16 01:25:25 -07002639 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002640
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641 return 0;
2642
2643failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002644 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002646 * Error recovery is a bit difficult. We need to zero out blocks that
2647 * were newly allocated, and dirty them to ensure they get written out.
2648 * Buffers need to be attached to the page at this point, otherwise
2649 * the handling of potential IO errors during writeout would be hard
2650 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002651 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002652 attach_nobh_buffers(page, head);
2653 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002654
Nick Piggin03158cd2007-10-16 01:25:25 -07002655out_release:
2656 unlock_page(page);
2657 page_cache_release(page);
2658 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002659
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002660 return ret;
2661}
Nick Piggin03158cd2007-10-16 01:25:25 -07002662EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002663
Nick Piggin03158cd2007-10-16 01:25:25 -07002664int nobh_write_end(struct file *file, struct address_space *mapping,
2665 loff_t pos, unsigned len, unsigned copied,
2666 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667{
2668 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002669 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002670 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002671 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002672
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002673 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002674 attach_nobh_buffers(page, head);
2675 if (page_has_buffers(page))
2676 return generic_write_end(file, mapping, pos, len,
2677 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002678
Nick Piggin22c8ca72007-02-20 13:58:09 -08002679 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002681 if (pos+copied > inode->i_size) {
2682 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683 mark_inode_dirty(inode);
2684 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002685
2686 unlock_page(page);
2687 page_cache_release(page);
2688
Nick Piggin03158cd2007-10-16 01:25:25 -07002689 while (head) {
2690 bh = head;
2691 head = head->b_this_page;
2692 free_buffer_head(bh);
2693 }
2694
2695 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002696}
Nick Piggin03158cd2007-10-16 01:25:25 -07002697EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698
2699/*
2700 * nobh_writepage() - based on block_full_write_page() except
2701 * that it tries to operate without attaching bufferheads to
2702 * the page.
2703 */
2704int nobh_writepage(struct page *page, get_block_t *get_block,
2705 struct writeback_control *wbc)
2706{
2707 struct inode * const inode = page->mapping->host;
2708 loff_t i_size = i_size_read(inode);
2709 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2710 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 int ret;
2712
2713 /* Is the page fully inside i_size? */
2714 if (page->index < end_index)
2715 goto out;
2716
2717 /* Is the page fully outside i_size? (truncate in progress) */
2718 offset = i_size & (PAGE_CACHE_SIZE-1);
2719 if (page->index >= end_index+1 || !offset) {
2720 /*
2721 * The page may have dirty, unmapped buffers. For example,
2722 * they may have been added in ext3_writepage(). Make them
2723 * freeable here, so the page does not leak.
2724 */
2725#if 0
2726 /* Not really sure about this - do we need this ? */
2727 if (page->mapping->a_ops->invalidatepage)
2728 page->mapping->a_ops->invalidatepage(page, offset);
2729#endif
2730 unlock_page(page);
2731 return 0; /* don't care */
2732 }
2733
2734 /*
2735 * The page straddles i_size. It must be zeroed out on each and every
2736 * writepage invocation because it may be mmapped. "A file is mapped
2737 * in multiples of the page size. For a file that is not a multiple of
2738 * the page size, the remaining memory is zeroed when mapped, and
2739 * writes to that region are not written out to the file."
2740 */
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002741 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742out:
2743 ret = mpage_writepage(page, get_block, wbc);
2744 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002745 ret = __block_write_full_page(inode, page, get_block, wbc,
2746 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 return ret;
2748}
2749EXPORT_SYMBOL(nobh_writepage);
2750
Nick Piggin03158cd2007-10-16 01:25:25 -07002751int nobh_truncate_page(struct address_space *mapping,
2752 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2755 unsigned offset = from & (PAGE_CACHE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002756 unsigned blocksize;
2757 sector_t iblock;
2758 unsigned length, pos;
2759 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002761 struct buffer_head map_bh;
2762 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763
Nick Piggin03158cd2007-10-16 01:25:25 -07002764 blocksize = 1 << inode->i_blkbits;
2765 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002766
Nick Piggin03158cd2007-10-16 01:25:25 -07002767 /* Block boundary? Nothing to do */
2768 if (!length)
2769 return 0;
2770
2771 length = blocksize - length;
2772 iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
2773
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002775 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 if (!page)
2777 goto out;
2778
Nick Piggin03158cd2007-10-16 01:25:25 -07002779 if (page_has_buffers(page)) {
2780has_buffers:
2781 unlock_page(page);
2782 page_cache_release(page);
2783 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002785
2786 /* Find the buffer that contains "offset" */
2787 pos = blocksize;
2788 while (offset >= pos) {
2789 iblock++;
2790 pos += blocksize;
2791 }
2792
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002793 map_bh.b_size = blocksize;
2794 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002795 err = get_block(inode, iblock, &map_bh, 0);
2796 if (err)
2797 goto unlock;
2798 /* unmapped? It's a hole - nothing to do */
2799 if (!buffer_mapped(&map_bh))
2800 goto unlock;
2801
2802 /* Ok, it's mapped. Make sure it's up-to-date */
2803 if (!PageUptodate(page)) {
2804 err = mapping->a_ops->readpage(NULL, page);
2805 if (err) {
2806 page_cache_release(page);
2807 goto out;
2808 }
2809 lock_page(page);
2810 if (!PageUptodate(page)) {
2811 err = -EIO;
2812 goto unlock;
2813 }
2814 if (page_has_buffers(page))
2815 goto has_buffers;
2816 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002817 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002818 set_page_dirty(page);
2819 err = 0;
2820
2821unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 unlock_page(page);
2823 page_cache_release(page);
2824out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002825 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826}
2827EXPORT_SYMBOL(nobh_truncate_page);
2828
2829int block_truncate_page(struct address_space *mapping,
2830 loff_t from, get_block_t *get_block)
2831{
2832 pgoff_t index = from >> PAGE_CACHE_SHIFT;
2833 unsigned offset = from & (PAGE_CACHE_SIZE-1);
2834 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002835 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002836 unsigned length, pos;
2837 struct inode *inode = mapping->host;
2838 struct page *page;
2839 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840 int err;
2841
2842 blocksize = 1 << inode->i_blkbits;
2843 length = offset & (blocksize - 1);
2844
2845 /* Block boundary? Nothing to do */
2846 if (!length)
2847 return 0;
2848
2849 length = blocksize - length;
Andrew Morton54b21a72006-01-08 01:03:05 -08002850 iblock = (sector_t)index << (PAGE_CACHE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851
2852 page = grab_cache_page(mapping, index);
2853 err = -ENOMEM;
2854 if (!page)
2855 goto out;
2856
2857 if (!page_has_buffers(page))
2858 create_empty_buffers(page, blocksize, 0);
2859
2860 /* Find the buffer that contains "offset" */
2861 bh = page_buffers(page);
2862 pos = blocksize;
2863 while (offset >= pos) {
2864 bh = bh->b_this_page;
2865 iblock++;
2866 pos += blocksize;
2867 }
2868
2869 err = 0;
2870 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002871 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 err = get_block(inode, iblock, bh, 0);
2873 if (err)
2874 goto unlock;
2875 /* unmapped? It's a hole - nothing to do */
2876 if (!buffer_mapped(bh))
2877 goto unlock;
2878 }
2879
2880 /* Ok, it's mapped. Make sure it's up-to-date */
2881 if (PageUptodate(page))
2882 set_buffer_uptodate(bh);
2883
David Chinner33a266d2007-02-12 00:51:41 -08002884 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 err = -EIO;
2886 ll_rw_block(READ, 1, &bh);
2887 wait_on_buffer(bh);
2888 /* Uhhuh. Read error. Complain and punt. */
2889 if (!buffer_uptodate(bh))
2890 goto unlock;
2891 }
2892
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002893 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002894 mark_buffer_dirty(bh);
2895 err = 0;
2896
2897unlock:
2898 unlock_page(page);
2899 page_cache_release(page);
2900out:
2901 return err;
2902}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002903EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002904
2905/*
2906 * The generic ->writepage function for buffer-backed address_spaces
2907 */
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002908int block_write_full_page(struct page *page, get_block_t *get_block,
2909 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910{
2911 struct inode * const inode = page->mapping->host;
2912 loff_t i_size = i_size_read(inode);
2913 const pgoff_t end_index = i_size >> PAGE_CACHE_SHIFT;
2914 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915
2916 /* Is the page fully inside i_size? */
2917 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04002918 return __block_write_full_page(inode, page, get_block, wbc,
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002919 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920
2921 /* Is the page fully outside i_size? (truncate in progress) */
2922 offset = i_size & (PAGE_CACHE_SIZE-1);
2923 if (page->index >= end_index+1 || !offset) {
2924 /*
2925 * The page may have dirty, unmapped buffers. For example,
2926 * they may have been added in ext3_writepage(). Make them
2927 * freeable here, so the page does not leak.
2928 */
Lukas Czernerd47992f2013-05-21 23:17:23 -04002929 do_invalidatepage(page, 0, PAGE_CACHE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 unlock_page(page);
2931 return 0; /* don't care */
2932 }
2933
2934 /*
2935 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05002936 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 * in multiples of the page size. For a file that is not a multiple of
2938 * the page size, the remaining memory is zeroed when mapped, and
2939 * writes to that region are not written out to the file."
2940 */
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002941 zero_user_segment(page, offset, PAGE_CACHE_SIZE);
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002942 return __block_write_full_page(inode, page, get_block, wbc,
2943 end_buffer_async_write);
Chris Mason35c80d52009-04-15 13:22:38 -04002944}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002945EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04002946
Linus Torvalds1da177e2005-04-16 15:20:36 -07002947sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2948 get_block_t *get_block)
2949{
2950 struct buffer_head tmp;
2951 struct inode *inode = mapping->host;
2952 tmp.b_state = 0;
2953 tmp.b_blocknr = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002954 tmp.b_size = 1 << inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002955 get_block(inode, block, &tmp, 0);
2956 return tmp.b_blocknr;
2957}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002958EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002960static void end_bio_bh_io_sync(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961{
2962 struct buffer_head *bh = bio->bi_private;
2963
Jens Axboeb7c44ed2015-07-24 12:37:59 -06002964 if (unlikely(bio_flagged(bio, BIO_QUIET)))
Keith Mannthey08bafc02008-11-25 10:24:35 +01002965 set_bit(BH_Quiet, &bh->b_state);
2966
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002967 bh->b_end_io(bh, !bio->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002968 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969}
2970
Linus Torvalds57302e02012-12-04 08:25:11 -08002971/*
2972 * This allows us to do IO even on the odd last sectors
Akinobu Mita59d43912014-10-09 15:26:53 -07002973 * of a device, even if the block size is some multiple
Linus Torvalds57302e02012-12-04 08:25:11 -08002974 * of the physical sector size.
2975 *
2976 * We'll just truncate the bio to the size of the device,
2977 * and clear the end of the buffer head manually.
2978 *
2979 * Truly out-of-range accesses will turn into actual IO
2980 * errors, this only handles the "we need to be able to
2981 * do IO at the final sector" case.
2982 */
Akinobu Mita4db96b72014-10-09 15:26:55 -07002983void guard_bio_eod(int rw, struct bio *bio)
Linus Torvalds57302e02012-12-04 08:25:11 -08002984{
2985 sector_t maxsector;
Akinobu Mita59d43912014-10-09 15:26:53 -07002986 struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
2987 unsigned truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08002988
2989 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
2990 if (!maxsector)
2991 return;
2992
2993 /*
2994 * If the *whole* IO is past the end of the device,
2995 * let it through, and the IO layer will turn it into
2996 * an EIO.
2997 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07002998 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08002999 return;
3000
Kent Overstreet4f024f32013-10-11 15:44:27 -07003001 maxsector -= bio->bi_iter.bi_sector;
Akinobu Mita59d43912014-10-09 15:26:53 -07003002 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08003003 return;
3004
Akinobu Mita59d43912014-10-09 15:26:53 -07003005 /* Uhhuh. We've got a bio that straddles the device size! */
3006 truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
Linus Torvalds57302e02012-12-04 08:25:11 -08003007
3008 /* Truncate the bio.. */
Akinobu Mita59d43912014-10-09 15:26:53 -07003009 bio->bi_iter.bi_size -= truncated_bytes;
3010 bvec->bv_len -= truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08003011
3012 /* ..and clear the end of the buffer for reads */
Dan Carpenter27d7c2a2012-12-05 20:01:24 +03003013 if ((rw & RW_MASK) == READ) {
Akinobu Mita59d43912014-10-09 15:26:53 -07003014 zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len,
3015 truncated_bytes);
Linus Torvalds57302e02012-12-04 08:25:11 -08003016 }
3017}
3018
Tejun Heob16b1de2015-06-02 08:39:48 -06003019static int submit_bh_wbc(int rw, struct buffer_head *bh,
3020 unsigned long bio_flags, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021{
3022 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003023
3024 BUG_ON(!buffer_locked(bh));
3025 BUG_ON(!buffer_mapped(bh));
3026 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04003027 BUG_ON(buffer_delay(bh));
3028 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003029
Jens Axboe48fd4f92008-08-22 10:00:36 +02003030 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02003031 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 */
Jens Axboe48fd4f92008-08-22 10:00:36 +02003033 if (test_set_buffer_req(bh) && (rw & WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003034 clear_buffer_write_io_error(bh);
3035
3036 /*
3037 * from here on down, it's all bio -- do the initial mapping,
3038 * submit_bio -> generic_make_request may further map this bio around
3039 */
3040 bio = bio_alloc(GFP_NOIO, 1);
3041
Tejun Heo2a814902015-05-28 14:50:51 -04003042 if (wbc) {
Tejun Heob16b1de2015-06-02 08:39:48 -06003043 wbc_init_bio(wbc, bio);
Tejun Heo2a814902015-05-28 14:50:51 -04003044 wbc_account_io(wbc, bh->b_page, bh->b_size);
3045 }
Tejun Heobafc0db2015-06-02 08:37:23 -06003046
Kent Overstreet4f024f32013-10-11 15:44:27 -07003047 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003048 bio->bi_bdev = bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003049
Kent Overstreet6cf66b42014-12-22 12:48:42 +01003050 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3051 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003052
3053 bio->bi_end_io = end_bio_bh_io_sync;
3054 bio->bi_private = bh;
Darrick J. Wong7136851112013-04-29 15:07:25 -07003055 bio->bi_flags |= bio_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056
Linus Torvalds57302e02012-12-04 08:25:11 -08003057 /* Take care of bh's that straddle the end of the device */
Akinobu Mita59d43912014-10-09 15:26:53 -07003058 guard_bio_eod(rw, bio);
Linus Torvalds57302e02012-12-04 08:25:11 -08003059
Theodore Ts'o877f9622013-04-20 19:58:37 -04003060 if (buffer_meta(bh))
3061 rw |= REQ_META;
3062 if (buffer_prio(bh))
3063 rw |= REQ_PRIO;
3064
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065 submit_bio(rw, bio);
Julia Lawallf6454b02015-05-26 21:59:53 +02003066 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003067}
Tejun Heobafc0db2015-06-02 08:37:23 -06003068
3069int _submit_bh(int rw, struct buffer_head *bh, unsigned long bio_flags)
3070{
Tejun Heob16b1de2015-06-02 08:39:48 -06003071 return submit_bh_wbc(rw, bh, bio_flags, NULL);
Tejun Heobafc0db2015-06-02 08:37:23 -06003072}
Darrick J. Wong7136851112013-04-29 15:07:25 -07003073EXPORT_SYMBOL_GPL(_submit_bh);
3074
3075int submit_bh(int rw, struct buffer_head *bh)
3076{
Tejun Heob16b1de2015-06-02 08:39:48 -06003077 return submit_bh_wbc(rw, bh, 0, NULL);
Darrick J. Wong7136851112013-04-29 15:07:25 -07003078}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003079EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
3081/**
3082 * ll_rw_block: low-level access to block devices (DEPRECATED)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003083 * @rw: whether to %READ or %WRITE or maybe %READA (readahead)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 * @nr: number of &struct buffer_heads in the array
3085 * @bhs: array of pointers to &struct buffer_head
3086 *
Jan Karaa7662232005-09-06 15:19:10 -07003087 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
3088 * requests an I/O operation on them, either a %READ or a %WRITE. The third
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003089 * %READA option is described in the documentation for generic_make_request()
3090 * which ll_rw_block() calls.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 *
3092 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003093 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3094 * request, and any buffer that appears to be up-to-date when doing read
3095 * request. Further it marks as clean buffers that are processed for
3096 * writing (the buffer cache won't assume that they are actually clean
3097 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 *
3099 * ll_rw_block sets b_end_io to simple completion handler that marks
Masanari Iidae2278672014-02-18 22:54:36 +09003100 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
Linus Torvalds1da177e2005-04-16 15:20:36 -07003101 * any waiters.
3102 *
3103 * All of the buffers must be for the same device, and must also be a
3104 * multiple of the current approved size for the device.
3105 */
3106void ll_rw_block(int rw, int nr, struct buffer_head *bhs[])
3107{
3108 int i;
3109
3110 for (i = 0; i < nr; i++) {
3111 struct buffer_head *bh = bhs[i];
3112
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003113 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003114 continue;
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003115 if (rw == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003116 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003117 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003118 get_bh(bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003119 submit_bh(WRITE, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120 continue;
3121 }
3122 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003124 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003125 get_bh(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126 submit_bh(rw, bh);
3127 continue;
3128 }
3129 }
3130 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131 }
3132}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003133EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003134
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003135void write_dirty_buffer(struct buffer_head *bh, int rw)
3136{
3137 lock_buffer(bh);
3138 if (!test_clear_buffer_dirty(bh)) {
3139 unlock_buffer(bh);
3140 return;
3141 }
3142 bh->b_end_io = end_buffer_write_sync;
3143 get_bh(bh);
3144 submit_bh(rw, bh);
3145}
3146EXPORT_SYMBOL(write_dirty_buffer);
3147
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148/*
3149 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3150 * and then start new I/O and then wait upon it. The caller must have a ref on
3151 * the buffer_head.
3152 */
Christoph Hellwig87e99512010-08-11 17:05:45 +02003153int __sync_dirty_buffer(struct buffer_head *bh, int rw)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154{
3155 int ret = 0;
3156
3157 WARN_ON(atomic_read(&bh->b_count) < 1);
3158 lock_buffer(bh);
3159 if (test_clear_buffer_dirty(bh)) {
3160 get_bh(bh);
3161 bh->b_end_io = end_buffer_write_sync;
Christoph Hellwig87e99512010-08-11 17:05:45 +02003162 ret = submit_bh(rw, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003163 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 if (!ret && !buffer_uptodate(bh))
3165 ret = -EIO;
3166 } else {
3167 unlock_buffer(bh);
3168 }
3169 return ret;
3170}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003171EXPORT_SYMBOL(__sync_dirty_buffer);
3172
3173int sync_dirty_buffer(struct buffer_head *bh)
3174{
3175 return __sync_dirty_buffer(bh, WRITE_SYNC);
3176}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003177EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178
3179/*
3180 * try_to_free_buffers() checks if all the buffers on this particular page
3181 * are unused, and releases them if so.
3182 *
3183 * Exclusion against try_to_free_buffers may be obtained by either
3184 * locking the page or by holding its mapping's private_lock.
3185 *
3186 * If the page is dirty but all the buffers are clean then we need to
3187 * be sure to mark the page clean as well. This is because the page
3188 * may be against a block device, and a later reattachment of buffers
3189 * to a dirty page will set *all* buffers dirty. Which would corrupt
3190 * filesystem data on the same device.
3191 *
3192 * The same applies to regular filesystem pages: if all the buffers are
3193 * clean then we set the page clean and proceed. To do that, we require
3194 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3195 * private_lock.
3196 *
3197 * try_to_free_buffers() is non-blocking.
3198 */
3199static inline int buffer_busy(struct buffer_head *bh)
3200{
3201 return atomic_read(&bh->b_count) |
3202 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3203}
3204
3205static int
3206drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3207{
3208 struct buffer_head *head = page_buffers(page);
3209 struct buffer_head *bh;
3210
3211 bh = head;
3212 do {
akpm@osdl.orgde7d5a32005-05-01 08:58:39 -07003213 if (buffer_write_io_error(bh) && page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 set_bit(AS_EIO, &page->mapping->flags);
3215 if (buffer_busy(bh))
3216 goto failed;
3217 bh = bh->b_this_page;
3218 } while (bh != head);
3219
3220 do {
3221 struct buffer_head *next = bh->b_this_page;
3222
Jan Kara535ee2f2008-02-08 04:21:59 -08003223 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 __remove_assoc_queue(bh);
3225 bh = next;
3226 } while (bh != head);
3227 *buffers_to_free = head;
3228 __clear_page_buffers(page);
3229 return 1;
3230failed:
3231 return 0;
3232}
3233
3234int try_to_free_buffers(struct page *page)
3235{
3236 struct address_space * const mapping = page->mapping;
3237 struct buffer_head *buffers_to_free = NULL;
3238 int ret = 0;
3239
3240 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003241 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003242 return 0;
3243
3244 if (mapping == NULL) { /* can this still happen? */
3245 ret = drop_buffers(page, &buffers_to_free);
3246 goto out;
3247 }
3248
3249 spin_lock(&mapping->private_lock);
3250 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003251
3252 /*
3253 * If the filesystem writes its buffers by hand (eg ext3)
3254 * then we can have clean buffers against a dirty page. We
3255 * clean the page here; otherwise the VM will never notice
3256 * that the filesystem did any IO at all.
3257 *
3258 * Also, during truncate, discard_buffer will have marked all
3259 * the page's buffers clean. We discover that here and clean
3260 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003261 *
3262 * private_lock must be held over this entire operation in order
3263 * to synchronise against __set_page_dirty_buffers and prevent the
3264 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003265 */
Tejun Heo11f81be2015-05-22 17:13:15 -04003266 if (ret)
3267 cancel_dirty_page(page);
Nick Piggin87df7242007-01-30 14:36:27 +11003268 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269out:
3270 if (buffers_to_free) {
3271 struct buffer_head *bh = buffers_to_free;
3272
3273 do {
3274 struct buffer_head *next = bh->b_this_page;
3275 free_buffer_head(bh);
3276 bh = next;
3277 } while (bh != buffers_to_free);
3278 }
3279 return ret;
3280}
3281EXPORT_SYMBOL(try_to_free_buffers);
3282
Linus Torvalds1da177e2005-04-16 15:20:36 -07003283/*
3284 * There are no bdflush tunables left. But distributions are
3285 * still running obsolete flush daemons, so we terminate them here.
3286 *
3287 * Use of bdflush() is deprecated and will be removed in a future kernel.
Jens Axboe5b0830c2009-09-23 19:37:09 +02003288 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003289 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01003290SYSCALL_DEFINE2(bdflush, int, func, long, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291{
3292 static int msg_count;
3293
3294 if (!capable(CAP_SYS_ADMIN))
3295 return -EPERM;
3296
3297 if (msg_count < 5) {
3298 msg_count++;
3299 printk(KERN_INFO
3300 "warning: process `%s' used the obsolete bdflush"
3301 " system call\n", current->comm);
3302 printk(KERN_INFO "Fix your initscripts?\n");
3303 }
3304
3305 if (func == 1)
3306 do_exit(0);
3307 return 0;
3308}
3309
3310/*
3311 * Buffer-head allocation
3312 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003313static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003314
3315/*
3316 * Once the number of bh's in the machine exceeds this level, we start
3317 * stripping them in writeback.
3318 */
Zhang Yanfei43be5942013-02-22 16:35:46 -08003319static unsigned long max_buffer_heads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320
3321int buffer_heads_over_limit;
3322
3323struct bh_accounting {
3324 int nr; /* Number of live bh's */
3325 int ratelimit; /* Limit cacheline bouncing */
3326};
3327
3328static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3329
3330static void recalc_bh_state(void)
3331{
3332 int i;
3333 int tot = 0;
3334
Christoph Lameteree1be862010-12-06 11:40:05 -06003335 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003337 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003338 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003339 tot += per_cpu(bh_accounting, i).nr;
3340 buffer_heads_over_limit = (tot > max_buffer_heads);
3341}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003342
Al Virodd0fc662005-10-07 07:46:04 +01003343struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003344{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003345 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003346 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003347 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003348 preempt_disable();
3349 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003351 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003352 }
3353 return ret;
3354}
3355EXPORT_SYMBOL(alloc_buffer_head);
3356
3357void free_buffer_head(struct buffer_head *bh)
3358{
3359 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3360 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003361 preempt_disable();
3362 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003364 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003365}
3366EXPORT_SYMBOL(free_buffer_head);
3367
Linus Torvalds1da177e2005-04-16 15:20:36 -07003368static void buffer_exit_cpu(int cpu)
3369{
3370 int i;
3371 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3372
3373 for (i = 0; i < BH_LRU_SIZE; i++) {
3374 brelse(b->bhs[i]);
3375 b->bhs[i] = NULL;
3376 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003377 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003378 per_cpu(bh_accounting, cpu).nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003379}
3380
3381static int buffer_cpu_notify(struct notifier_block *self,
3382 unsigned long action, void *hcpu)
3383{
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003384 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 buffer_exit_cpu((unsigned long)hcpu);
3386 return NOTIFY_OK;
3387}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003388
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003389/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003390 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003391 * @bh: struct buffer_head
3392 *
3393 * Return true if the buffer is up-to-date and false,
3394 * with the buffer locked, if not.
3395 */
3396int bh_uptodate_or_lock(struct buffer_head *bh)
3397{
3398 if (!buffer_uptodate(bh)) {
3399 lock_buffer(bh);
3400 if (!buffer_uptodate(bh))
3401 return 0;
3402 unlock_buffer(bh);
3403 }
3404 return 1;
3405}
3406EXPORT_SYMBOL(bh_uptodate_or_lock);
3407
3408/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003409 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003410 * @bh: struct buffer_head
3411 *
3412 * Returns zero on success and -EIO on error.
3413 */
3414int bh_submit_read(struct buffer_head *bh)
3415{
3416 BUG_ON(!buffer_locked(bh));
3417
3418 if (buffer_uptodate(bh)) {
3419 unlock_buffer(bh);
3420 return 0;
3421 }
3422
3423 get_bh(bh);
3424 bh->b_end_io = end_buffer_read_sync;
3425 submit_bh(READ, bh);
3426 wait_on_buffer(bh);
3427 if (buffer_uptodate(bh))
3428 return 0;
3429 return -EIO;
3430}
3431EXPORT_SYMBOL(bh_submit_read);
3432
Linus Torvalds1da177e2005-04-16 15:20:36 -07003433void __init buffer_init(void)
3434{
Zhang Yanfei43be5942013-02-22 16:35:46 -08003435 unsigned long nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003436
Christoph Lameterb98938c2008-02-04 22:28:36 -08003437 bh_cachep = kmem_cache_create("buffer_head",
3438 sizeof(struct buffer_head), 0,
3439 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3440 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003441 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442
3443 /*
3444 * Limit the bh occupancy to 10% of ZONE_NORMAL
3445 */
3446 nrpages = (nr_free_buffer_pages() * 10) / 100;
3447 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3448 hotcpu_notifier(buffer_cpu_notify, 0);
3449}