blob: c24201e78492f1650bff12eeb340c5d4522e55bf [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);
Mike Christie2a222ca2016-06-05 14:31:43 -050048static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Tejun Heob16b1de2015-06-02 08:39:48 -060049 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);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300132 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
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{
Robert Elliott432f16e2014-10-21 13:55:11 -0600137 if (!test_bit(BH_Quiet, &bh->b_state))
138 printk_ratelimited(KERN_ERR
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400139 "Buffer I/O error on dev %pg, logical block %llu%s\n",
140 bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141}
142
143/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700144 * End-of-IO handler helper function which does not touch the bh after
145 * unlocking it.
146 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
147 * a race there is benign: unlock_buffer() only use the bh's address for
148 * hashing after unlocking the buffer, so it doesn't actually touch the bh
149 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700151static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152{
153 if (uptodate) {
154 set_buffer_uptodate(bh);
155 } else {
Christoph Hellwig70246282016-07-19 11:28:41 +0200156 /* This happens, due to failed read-ahead attempts. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157 clear_buffer_uptodate(bh);
158 }
159 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700160}
161
162/*
163 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
164 * unlock the buffer. This is what ll_rw_block uses too.
165 */
166void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
167{
168 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700169 put_bh(bh);
170}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700171EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172
173void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
174{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700175 if (uptodate) {
176 set_buffer_uptodate(bh);
177 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600178 buffer_io_error(bh, ", lost sync page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 set_buffer_write_io_error(bh);
180 clear_buffer_uptodate(bh);
181 }
182 unlock_buffer(bh);
183 put_bh(bh);
184}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700185EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186
187/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 * Various filesystems appear to want __find_get_block to be non-blocking.
189 * But it's the page lock which protects the buffers. To get around this,
190 * we get exclusion from try_to_free_buffers with the blockdev mapping's
191 * private_lock.
192 *
193 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
194 * may be quite high. This code could TryLock the page, and if that
195 * succeeds, there is no need to take private_lock. (But if
196 * private_lock is contended then so is mapping->tree_lock).
197 */
198static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800199__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200{
201 struct inode *bd_inode = bdev->bd_inode;
202 struct address_space *bd_mapping = bd_inode->i_mapping;
203 struct buffer_head *ret = NULL;
204 pgoff_t index;
205 struct buffer_head *bh;
206 struct buffer_head *head;
207 struct page *page;
208 int all_mapped = 1;
209
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300210 index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman2457aec2014-06-04 16:10:31 -0700211 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700212 if (!page)
213 goto out;
214
215 spin_lock(&bd_mapping->private_lock);
216 if (!page_has_buffers(page))
217 goto out_unlock;
218 head = page_buffers(page);
219 bh = head;
220 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700221 if (!buffer_mapped(bh))
222 all_mapped = 0;
223 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700224 ret = bh;
225 get_bh(bh);
226 goto out_unlock;
227 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700228 bh = bh->b_this_page;
229 } while (bh != head);
230
231 /* we might be here because some of the buffers on this page are
232 * not mapped. This is due to various races between
233 * file io on the block device and getblk. It gets dealt with
234 * elsewhere, don't buffer_error if we had some unmapped buffers
235 */
236 if (all_mapped) {
237 printk("__find_get_block_slow() failed. "
238 "block=%llu, b_blocknr=%llu\n",
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800239 (unsigned long long)block,
240 (unsigned long long)bh->b_blocknr);
241 printk("b_state=0x%08lx, b_size=%zu\n",
242 bh->b_state, bh->b_size);
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400243 printk("device %pg blocksize: %d\n", bdev,
Tao Ma72a2ebd2011-10-31 17:09:00 -0700244 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245 }
246out_unlock:
247 spin_unlock(&bd_mapping->private_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300248 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249out:
250 return ret;
251}
252
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253/*
Jens Axboe5b0830c2009-09-23 19:37:09 +0200254 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255 */
256static void free_more_memory(void)
257{
Mel Gormanc33d6c02016-05-19 17:14:10 -0700258 struct zoneref *z;
Mel Gorman0e884602008-04-28 02:12:14 -0700259 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600261 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262 yield();
263
Mel Gorman0e884602008-04-28 02:12:14 -0700264 for_each_online_node(nid) {
Mel Gormanc33d6c02016-05-19 17:14:10 -0700265
266 z = first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
267 gfp_zone(GFP_NOFS), NULL);
268 if (z->zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700269 try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -0700270 GFP_NOFS, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 }
272}
273
274/*
275 * I/O completion handler for block_read_full_page() - pages
276 * which come unlocked at the end of I/O.
277 */
278static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
279{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700280 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700281 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 struct buffer_head *tmp;
283 struct page *page;
284 int page_uptodate = 1;
285
286 BUG_ON(!buffer_async_read(bh));
287
288 page = bh->b_page;
289 if (uptodate) {
290 set_buffer_uptodate(bh);
291 } else {
292 clear_buffer_uptodate(bh);
Robert Elliott432f16e2014-10-21 13:55:11 -0600293 buffer_io_error(bh, ", async page read");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294 SetPageError(page);
295 }
296
297 /*
298 * Be _very_ careful from here on. Bad things can happen if
299 * two buffer heads end IO at almost the same time and both
300 * decide that the page is now completely done.
301 */
Nick Piggina3972202005-07-07 17:56:56 -0700302 first = page_buffers(page);
303 local_irq_save(flags);
304 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700305 clear_buffer_async_read(bh);
306 unlock_buffer(bh);
307 tmp = bh;
308 do {
309 if (!buffer_uptodate(tmp))
310 page_uptodate = 0;
311 if (buffer_async_read(tmp)) {
312 BUG_ON(!buffer_locked(tmp));
313 goto still_busy;
314 }
315 tmp = tmp->b_this_page;
316 } while (tmp != bh);
Nick Piggina3972202005-07-07 17:56:56 -0700317 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
318 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700319
320 /*
321 * If none of the buffers had errors and they are all
322 * uptodate then we can set the page uptodate.
323 */
324 if (page_uptodate && !PageError(page))
325 SetPageUptodate(page);
326 unlock_page(page);
327 return;
328
329still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700330 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
331 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 return;
333}
334
335/*
336 * Completion handler for block_write_full_page() - pages which are unlocked
337 * during I/O, and which have PageWriteback cleared upon I/O completion.
338 */
Chris Mason35c80d52009-04-15 13:22:38 -0400339void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700340{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700342 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 struct buffer_head *tmp;
344 struct page *page;
345
346 BUG_ON(!buffer_async_write(bh));
347
348 page = bh->b_page;
349 if (uptodate) {
350 set_buffer_uptodate(bh);
351 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600352 buffer_io_error(bh, ", lost async page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700353 set_bit(AS_EIO, &page->mapping->flags);
Jan Kara58ff4072006-10-17 00:10:19 -0700354 set_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700355 clear_buffer_uptodate(bh);
356 SetPageError(page);
357 }
358
Nick Piggina3972202005-07-07 17:56:56 -0700359 first = page_buffers(page);
360 local_irq_save(flags);
361 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
362
Linus Torvalds1da177e2005-04-16 15:20:36 -0700363 clear_buffer_async_write(bh);
364 unlock_buffer(bh);
365 tmp = bh->b_this_page;
366 while (tmp != bh) {
367 if (buffer_async_write(tmp)) {
368 BUG_ON(!buffer_locked(tmp));
369 goto still_busy;
370 }
371 tmp = tmp->b_this_page;
372 }
Nick Piggina3972202005-07-07 17:56:56 -0700373 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
374 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 end_page_writeback(page);
376 return;
377
378still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700379 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
380 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700381 return;
382}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700383EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700384
385/*
386 * If a page's buffers are under async readin (end_buffer_async_read
387 * completion) then there is a possibility that another thread of
388 * control could lock one of the buffers after it has completed
389 * but while some of the other buffers have not completed. This
390 * locked buffer would confuse end_buffer_async_read() into not unlocking
391 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
392 * that this buffer is not under async I/O.
393 *
394 * The page comes unlocked when it has no locked buffer_async buffers
395 * left.
396 *
397 * PageLocked prevents anyone starting new async I/O reads any of
398 * the buffers.
399 *
400 * PageWriteback is used to prevent simultaneous writeout of the same
401 * page.
402 *
403 * PageLocked prevents anyone from starting writeback of a page which is
404 * under read I/O (PageWriteback is only ever set against a locked page).
405 */
406static void mark_buffer_async_read(struct buffer_head *bh)
407{
408 bh->b_end_io = end_buffer_async_read;
409 set_buffer_async_read(bh);
410}
411
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700412static void mark_buffer_async_write_endio(struct buffer_head *bh,
413 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400414{
415 bh->b_end_io = handler;
416 set_buffer_async_write(bh);
417}
418
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419void mark_buffer_async_write(struct buffer_head *bh)
420{
Chris Mason35c80d52009-04-15 13:22:38 -0400421 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422}
423EXPORT_SYMBOL(mark_buffer_async_write);
424
425
426/*
427 * fs/buffer.c contains helper functions for buffer-backed address space's
428 * fsync functions. A common requirement for buffer-based filesystems is
429 * that certain data from the backing blockdev needs to be written out for
430 * a successful fsync(). For example, ext2 indirect blocks need to be
431 * written back and waited upon before fsync() returns.
432 *
433 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
434 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
435 * management of a list of dependent buffers at ->i_mapping->private_list.
436 *
437 * Locking is a little subtle: try_to_free_buffers() will remove buffers
438 * from their controlling inode's queue when they are being freed. But
439 * try_to_free_buffers() will be operating against the *blockdev* mapping
440 * at the time, not against the S_ISREG file which depends on those buffers.
441 * So the locking for private_list is via the private_lock in the address_space
442 * which backs the buffers. Which is different from the address_space
443 * against which the buffers are listed. So for a particular address_space,
444 * mapping->private_lock does *not* protect mapping->private_list! In fact,
445 * mapping->private_list will always be protected by the backing blockdev's
446 * ->private_lock.
447 *
448 * Which introduces a requirement: all buffers on an address_space's
449 * ->private_list must be from the same address_space: the blockdev's.
450 *
451 * address_spaces which do not place buffers at ->private_list via these
452 * utility functions are free to use private_lock and private_list for
453 * whatever they want. The only requirement is that list_empty(private_list)
454 * be true at clear_inode() time.
455 *
456 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
457 * filesystems should do that. invalidate_inode_buffers() should just go
458 * BUG_ON(!list_empty).
459 *
460 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
461 * take an address_space, not an inode. And it should be called
462 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
463 * queued up.
464 *
465 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
466 * list if it is already on a list. Because if the buffer is on a list,
467 * it *must* already be on the right one. If not, the filesystem is being
468 * silly. This will save a ton of locking. But first we have to ensure
469 * that buffers are taken *off* the old inode's list when they are freed
470 * (presumably in truncate). That requires careful auditing of all
471 * filesystems (do it inside bforget()). It could also be done by bringing
472 * b_inode back.
473 */
474
475/*
476 * The buffer's backing address_space's private_lock must be held
477 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700478static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700479{
480 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -0700481 WARN_ON(!bh->b_assoc_map);
482 if (buffer_write_io_error(bh))
483 set_bit(AS_EIO, &bh->b_assoc_map->flags);
484 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700485}
486
487int inode_has_buffers(struct inode *inode)
488{
489 return !list_empty(&inode->i_data.private_list);
490}
491
492/*
493 * osync is designed to support O_SYNC io. It waits synchronously for
494 * all already-submitted IO to complete, but does not queue any new
495 * writes to the disk.
496 *
497 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
498 * you dirty the buffers, and then use osync_inode_buffers to wait for
499 * completion. Any other dirty buffers which are not yet queued for
500 * write will not be flushed to disk by the osync.
501 */
502static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
503{
504 struct buffer_head *bh;
505 struct list_head *p;
506 int err = 0;
507
508 spin_lock(lock);
509repeat:
510 list_for_each_prev(p, list) {
511 bh = BH_ENTRY(p);
512 if (buffer_locked(bh)) {
513 get_bh(bh);
514 spin_unlock(lock);
515 wait_on_buffer(bh);
516 if (!buffer_uptodate(bh))
517 err = -EIO;
518 brelse(bh);
519 spin_lock(lock);
520 goto repeat;
521 }
522 }
523 spin_unlock(lock);
524 return err;
525}
526
Al Viro01a05b32010-03-23 06:06:58 -0400527static void do_thaw_one(struct super_block *sb, void *unused)
528{
Al Viro01a05b32010-03-23 06:06:58 -0400529 while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400530 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
Al Viro01a05b32010-03-23 06:06:58 -0400531}
532
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700533static void do_thaw_all(struct work_struct *work)
Eric Sandeenc2d75432009-03-31 15:23:46 -0700534{
Al Viro01a05b32010-03-23 06:06:58 -0400535 iterate_supers(do_thaw_one, NULL);
Jens Axboe053c5252009-04-08 13:44:08 +0200536 kfree(work);
Eric Sandeenc2d75432009-03-31 15:23:46 -0700537 printk(KERN_WARNING "Emergency Thaw complete\n");
538}
539
540/**
541 * emergency_thaw_all -- forcibly thaw every frozen filesystem
542 *
543 * Used for emergency unfreeze of all filesystems via SysRq
544 */
545void emergency_thaw_all(void)
546{
Jens Axboe053c5252009-04-08 13:44:08 +0200547 struct work_struct *work;
548
549 work = kmalloc(sizeof(*work), GFP_ATOMIC);
550 if (work) {
551 INIT_WORK(work, do_thaw_all);
552 schedule_work(work);
553 }
Eric Sandeenc2d75432009-03-31 15:23:46 -0700554}
555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800557 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700558 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559 *
560 * Starts I/O against the buffers at mapping->private_list, and waits upon
561 * that I/O.
562 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700563 * Basically, this is a convenience function for fsync().
564 * @mapping is a file or directory which needs those buffers to be written for
565 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 */
567int sync_mapping_buffers(struct address_space *mapping)
568{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800569 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570
571 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
572 return 0;
573
574 return fsync_buffers_list(&buffer_mapping->private_lock,
575 &mapping->private_list);
576}
577EXPORT_SYMBOL(sync_mapping_buffers);
578
579/*
580 * Called when we've recently written block `bblock', and it is known that
581 * `bblock' was for a buffer_boundary() buffer. This means that the block at
582 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
583 * dirty, schedule it for IO. So that indirects merge nicely with their data.
584 */
585void write_boundary_block(struct block_device *bdev,
586 sector_t bblock, unsigned blocksize)
587{
588 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
589 if (bh) {
590 if (buffer_dirty(bh))
Mike Christiedfec8a12016-06-05 14:31:44 -0500591 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 put_bh(bh);
593 }
594}
595
596void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
597{
598 struct address_space *mapping = inode->i_mapping;
599 struct address_space *buffer_mapping = bh->b_page->mapping;
600
601 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800602 if (!mapping->private_data) {
603 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800605 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800607 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 spin_lock(&buffer_mapping->private_lock);
609 list_move_tail(&bh->b_assoc_buffers,
610 &mapping->private_list);
Jan Kara58ff4072006-10-17 00:10:19 -0700611 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612 spin_unlock(&buffer_mapping->private_lock);
613 }
614}
615EXPORT_SYMBOL(mark_buffer_dirty_inode);
616
617/*
Nick Piggin787d2212007-07-17 04:03:34 -0700618 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
619 * dirty.
620 *
621 * If warn is true, then emit a warning if the page is not uptodate and has
622 * not been truncated.
Greg Thelenc4843a72015-05-22 17:13:16 -0400623 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700624 * The caller must hold lock_page_memcg().
Nick Piggin787d2212007-07-17 04:03:34 -0700625 */
Greg Thelenc4843a72015-05-22 17:13:16 -0400626static void __set_page_dirty(struct page *page, struct address_space *mapping,
Johannes Weiner62cccb82016-03-15 14:57:22 -0700627 int warn)
Nick Piggin787d2212007-07-17 04:03:34 -0700628{
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800629 unsigned long flags;
630
631 spin_lock_irqsave(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700632 if (page->mapping) { /* Race with truncate? */
633 WARN_ON_ONCE(warn && !PageUptodate(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700634 account_page_dirtied(page, mapping);
Nick Piggin787d2212007-07-17 04:03:34 -0700635 radix_tree_tag_set(&mapping->page_tree,
636 page_index(page), PAGECACHE_TAG_DIRTY);
637 }
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800638 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700639}
640
641/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 * Add a page to the dirty page list.
643 *
644 * It is a sad fact of life that this function is called from several places
645 * deeply under spinlocking. It may not sleep.
646 *
647 * If the page has buffers, the uptodate buffers are set dirty, to preserve
648 * dirty-state coherency between the page and the buffers. It the page does
649 * not have buffers then when they are later attached they will all be set
650 * dirty.
651 *
652 * The buffers are dirtied before the page is dirtied. There's a small race
653 * window in which a writepage caller may see the page cleanness but not the
654 * buffer dirtiness. That's fine. If this code were to set the page dirty
655 * before the buffers, a concurrent writepage caller could clear the page dirty
656 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
657 * page on the dirty page list.
658 *
659 * We use private_lock to lock against try_to_free_buffers while using the
660 * page's buffer list. Also use this to protect against clean buffers being
661 * added to the page after it was set dirty.
662 *
663 * FIXME: may need to call ->reservepage here as well. That's rather up to the
664 * address_space though.
665 */
666int __set_page_dirty_buffers(struct page *page)
667{
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700668 int newly_dirty;
Nick Piggin787d2212007-07-17 04:03:34 -0700669 struct address_space *mapping = page_mapping(page);
Nick Pigginebf7a222006-10-10 04:36:54 +0200670
671 if (unlikely(!mapping))
672 return !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700673
674 spin_lock(&mapping->private_lock);
675 if (page_has_buffers(page)) {
676 struct buffer_head *head = page_buffers(page);
677 struct buffer_head *bh = head;
678
679 do {
680 set_buffer_dirty(bh);
681 bh = bh->b_this_page;
682 } while (bh != head);
683 }
Greg Thelenc4843a72015-05-22 17:13:16 -0400684 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700685 * Lock out page->mem_cgroup migration to keep PageDirty
686 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -0400687 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700688 lock_page_memcg(page);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700689 newly_dirty = !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700690 spin_unlock(&mapping->private_lock);
691
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700692 if (newly_dirty)
Johannes Weiner62cccb82016-03-15 14:57:22 -0700693 __set_page_dirty(page, mapping, 1);
Greg Thelenc4843a72015-05-22 17:13:16 -0400694
Johannes Weiner62cccb82016-03-15 14:57:22 -0700695 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400696
697 if (newly_dirty)
698 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
699
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700700 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701}
702EXPORT_SYMBOL(__set_page_dirty_buffers);
703
704/*
705 * Write out and wait upon a list of buffers.
706 *
707 * We have conflicting pressures: we want to make sure that all
708 * initially dirty buffers get waited on, but that any subsequently
709 * dirtied buffers don't. After all, we don't want fsync to last
710 * forever if somebody is actively writing to the file.
711 *
712 * Do this in two main stages: first we copy dirty buffers to a
713 * temporary inode list, queueing the writes as we go. Then we clean
714 * up, waiting for those writes to complete.
715 *
716 * During this second stage, any subsequent updates to the file may end
717 * up refiling the buffer on the original inode's dirty list again, so
718 * there is a chance we will end up with a buffer queued for write but
719 * not yet completed on that list. So, as a final cleanup we go through
720 * the osync code to catch these locked, dirty buffers without requeuing
721 * any newly dirty buffers for write.
722 */
723static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
724{
725 struct buffer_head *bh;
726 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100727 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100729 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730
731 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100732 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733
734 spin_lock(lock);
735 while (!list_empty(list)) {
736 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800737 mapping = bh->b_assoc_map;
Jan Kara58ff4072006-10-17 00:10:19 -0700738 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800739 /* Avoid race with mark_buffer_dirty_inode() which does
740 * a lockless check and we rely on seeing the dirty bit */
741 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700742 if (buffer_dirty(bh) || buffer_locked(bh)) {
743 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800744 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700745 if (buffer_dirty(bh)) {
746 get_bh(bh);
747 spin_unlock(lock);
748 /*
749 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200750 * write_dirty_buffer() actually writes the
751 * current contents - it is a noop if I/O is
752 * still in flight on potentially older
753 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700754 */
Jens Axboe721a9602011-03-09 11:56:30 +0100755 write_dirty_buffer(bh, WRITE_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200756
757 /*
758 * Kick off IO for the previous mapping. Note
759 * that we will not run the very last mapping,
760 * wait_on_buffer() will do that for us
761 * through sync_buffer().
762 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700763 brelse(bh);
764 spin_lock(lock);
765 }
766 }
767 }
768
Jens Axboe4ee24912011-03-17 10:51:40 +0100769 spin_unlock(lock);
770 blk_finish_plug(&plug);
771 spin_lock(lock);
772
Linus Torvalds1da177e2005-04-16 15:20:36 -0700773 while (!list_empty(&tmp)) {
774 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800776 mapping = bh->b_assoc_map;
777 __remove_assoc_queue(bh);
778 /* Avoid race with mark_buffer_dirty_inode() which does
779 * a lockless check and we rely on seeing the dirty bit */
780 smp_mb();
781 if (buffer_dirty(bh)) {
782 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800783 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800784 bh->b_assoc_map = mapping;
785 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 spin_unlock(lock);
787 wait_on_buffer(bh);
788 if (!buffer_uptodate(bh))
789 err = -EIO;
790 brelse(bh);
791 spin_lock(lock);
792 }
793
794 spin_unlock(lock);
795 err2 = osync_buffers_list(lock, list);
796 if (err)
797 return err;
798 else
799 return err2;
800}
801
802/*
803 * Invalidate any and all dirty buffers on a given inode. We are
804 * probably unmounting the fs, but that doesn't mean we have already
805 * done a sync(). Just drop the buffers from the inode list.
806 *
807 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
808 * assumes that all the buffers are against the blockdev. Not true
809 * for reiserfs.
810 */
811void invalidate_inode_buffers(struct inode *inode)
812{
813 if (inode_has_buffers(inode)) {
814 struct address_space *mapping = &inode->i_data;
815 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800816 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818 spin_lock(&buffer_mapping->private_lock);
819 while (!list_empty(list))
820 __remove_assoc_queue(BH_ENTRY(list->next));
821 spin_unlock(&buffer_mapping->private_lock);
822 }
823}
Jan Kara52b19ac2008-09-23 18:24:08 +0200824EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700825
826/*
827 * Remove any clean buffers from the inode's buffer list. This is called
828 * when we're trying to free the inode itself. Those buffers can pin it.
829 *
830 * Returns true if all buffers were removed.
831 */
832int remove_inode_buffers(struct inode *inode)
833{
834 int ret = 1;
835
836 if (inode_has_buffers(inode)) {
837 struct address_space *mapping = &inode->i_data;
838 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800839 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840
841 spin_lock(&buffer_mapping->private_lock);
842 while (!list_empty(list)) {
843 struct buffer_head *bh = BH_ENTRY(list->next);
844 if (buffer_dirty(bh)) {
845 ret = 0;
846 break;
847 }
848 __remove_assoc_queue(bh);
849 }
850 spin_unlock(&buffer_mapping->private_lock);
851 }
852 return ret;
853}
854
855/*
856 * Create the appropriate buffers when given a page for data area and
857 * the size of each buffer.. Use the bh->b_this_page linked list to
858 * follow the buffers created. Return NULL if unable to create more
859 * buffers.
860 *
861 * The retry flag is used to differentiate async IO (paging, swapping)
862 * which may not fail from ordinary buffer allocations.
863 */
864struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
865 int retry)
866{
867 struct buffer_head *bh, *head;
868 long offset;
869
870try_again:
871 head = NULL;
872 offset = PAGE_SIZE;
873 while ((offset -= size) >= 0) {
874 bh = alloc_buffer_head(GFP_NOFS);
875 if (!bh)
876 goto no_grow;
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 bh->b_this_page = head;
879 bh->b_blocknr = -1;
880 head = bh;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 bh->b_size = size;
883
884 /* Link the buffer to its page */
885 set_bh_page(bh, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
887 return head;
888/*
889 * In case anything failed, we just free everything we got.
890 */
891no_grow:
892 if (head) {
893 do {
894 bh = head;
895 head = head->b_this_page;
896 free_buffer_head(bh);
897 } while (head);
898 }
899
900 /*
901 * Return failure for non-async IO requests. Async IO requests
902 * are not allowed to fail, so we have to wait until buffer heads
903 * become available. But we don't want tasks sleeping with
904 * partially complete buffers, so all were released above.
905 */
906 if (!retry)
907 return NULL;
908
909 /* We're _really_ low on memory. Now we just
910 * wait for old buffer heads to become free due to
911 * finishing IO. Since this is an async request and
912 * the reserve list is empty, we're sure there are
913 * async buffer heads in use.
914 */
915 free_more_memory();
916 goto try_again;
917}
918EXPORT_SYMBOL_GPL(alloc_page_buffers);
919
920static inline void
921link_dev_buffers(struct page *page, struct buffer_head *head)
922{
923 struct buffer_head *bh, *tail;
924
925 bh = head;
926 do {
927 tail = bh;
928 bh = bh->b_this_page;
929 } while (bh);
930 tail->b_this_page = head;
931 attach_page_buffers(page, head);
932}
933
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800934static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
935{
936 sector_t retval = ~((sector_t)0);
937 loff_t sz = i_size_read(bdev->bd_inode);
938
939 if (sz) {
940 unsigned int sizebits = blksize_bits(size);
941 retval = (sz >> sizebits);
942 }
943 return retval;
944}
945
Linus Torvalds1da177e2005-04-16 15:20:36 -0700946/*
947 * Initialise the state of a blockdev page's buffers.
948 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200949static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700950init_page_buffers(struct page *page, struct block_device *bdev,
951 sector_t block, int size)
952{
953 struct buffer_head *head = page_buffers(page);
954 struct buffer_head *bh = head;
955 int uptodate = PageUptodate(page);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800956 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957
958 do {
959 if (!buffer_mapped(bh)) {
960 init_buffer(bh, NULL, NULL);
961 bh->b_bdev = bdev;
962 bh->b_blocknr = block;
963 if (uptodate)
964 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200965 if (block < end_block)
966 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700967 }
968 block++;
969 bh = bh->b_this_page;
970 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200971
972 /*
973 * Caller needs to validate requested block against end of device.
974 */
975 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976}
977
978/*
979 * Create the page-cache page that contains the requested block.
980 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200981 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700982 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200983static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984grow_dev_page(struct block_device *bdev, sector_t block,
Gioh Kim3b5e6452014-09-04 22:04:42 -0400985 pgoff_t index, int size, int sizebits, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986{
987 struct inode *inode = bdev->bd_inode;
988 struct page *page;
989 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200990 sector_t end_block;
991 int ret = 0; /* Will call free_more_memory() */
Johannes Weiner84235de2013-10-16 13:47:00 -0700992 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700993
Michal Hockoc62d2552015-11-06 16:28:49 -0800994 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
Gioh Kim3b5e6452014-09-04 22:04:42 -0400995
Johannes Weiner84235de2013-10-16 13:47:00 -0700996 /*
997 * XXX: __getblk_slow() can not really deal with failure and
998 * will endlessly loop on improvised global reclaim. Prefer
999 * looping in the allocator rather than here, at least that
1000 * code knows what it's doing.
1001 */
1002 gfp_mask |= __GFP_NOFAIL;
1003
1004 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001005 if (!page)
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001006 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007
Eric Sesterhenne827f922006-03-26 18:24:46 +02001008 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
1010 if (page_has_buffers(page)) {
1011 bh = page_buffers(page);
1012 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001013 end_block = init_page_buffers(page, bdev,
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001014 (sector_t)index << sizebits,
1015 size);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001016 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017 }
1018 if (!try_to_free_buffers(page))
1019 goto failed;
1020 }
1021
1022 /*
1023 * Allocate some buffers for this page
1024 */
1025 bh = alloc_page_buffers(page, size, 0);
1026 if (!bh)
1027 goto failed;
1028
1029 /*
1030 * Link the page to the buffers and initialise them. Take the
1031 * lock to be atomic wrt __find_get_block(), which does not
1032 * run under the page lock.
1033 */
1034 spin_lock(&inode->i_mapping->private_lock);
1035 link_dev_buffers(page, bh);
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001036 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1037 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001038 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001039done:
1040 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001043 put_page(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001044 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001045}
1046
1047/*
1048 * Create buffers for the specified block device block's page. If
1049 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001050 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001051static int
Gioh Kim3b5e6452014-09-04 22:04:42 -04001052grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001053{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054 pgoff_t index;
1055 int sizebits;
1056
1057 sizebits = -1;
1058 do {
1059 sizebits++;
1060 } while ((size << sizebits) < PAGE_SIZE);
1061
1062 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Andrew Mortone5657932006-10-11 01:21:46 -07001064 /*
1065 * Check for a block which wants to lie outside our maximum possible
1066 * pagecache index. (this comparison is done using sector_t types).
1067 */
1068 if (unlikely(index != block >> sizebits)) {
Andrew Mortone5657932006-10-11 01:21:46 -07001069 printk(KERN_ERR "%s: requested out-of-range block %llu for "
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001070 "device %pg\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001071 __func__, (unsigned long long)block,
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001072 bdev);
Andrew Mortone5657932006-10-11 01:21:46 -07001073 return -EIO;
1074 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001075
Linus Torvalds1da177e2005-04-16 15:20:36 -07001076 /* Create a page with the proper size buffers.. */
Gioh Kim3b5e6452014-09-04 22:04:42 -04001077 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078}
1079
Gioh Kim3b5e6452014-09-04 22:04:42 -04001080struct buffer_head *
1081__getblk_slow(struct block_device *bdev, sector_t block,
1082 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001083{
1084 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001085 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 (size < 512 || size > PAGE_SIZE))) {
1087 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1088 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001089 printk(KERN_ERR "logical block size: %d\n",
1090 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001091
1092 dump_stack();
1093 return NULL;
1094 }
1095
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001096 for (;;) {
1097 struct buffer_head *bh;
1098 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001099
1100 bh = __find_get_block(bdev, block, size);
1101 if (bh)
1102 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001103
Gioh Kim3b5e6452014-09-04 22:04:42 -04001104 ret = grow_buffers(bdev, block, size, gfp);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001105 if (ret < 0)
1106 return NULL;
1107 if (ret == 0)
1108 free_more_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109 }
1110}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001111EXPORT_SYMBOL(__getblk_slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001112
1113/*
1114 * The relationship between dirty buffers and dirty pages:
1115 *
1116 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1117 * the page is tagged dirty in its radix tree.
1118 *
1119 * At all times, the dirtiness of the buffers represents the dirtiness of
1120 * subsections of the page. If the page has buffers, the page dirty bit is
1121 * merely a hint about the true dirty state.
1122 *
1123 * When a page is set dirty in its entirety, all its buffers are marked dirty
1124 * (if the page has buffers).
1125 *
1126 * When a buffer is marked dirty, its page is dirtied, but the page's other
1127 * buffers are not.
1128 *
1129 * Also. When blockdev buffers are explicitly read with bread(), they
1130 * individually become uptodate. But their backing page remains not
1131 * uptodate - even if all of its buffers are uptodate. A subsequent
1132 * block_read_full_page() against that page will discover all the uptodate
1133 * buffers, will set the page uptodate and will perform no I/O.
1134 */
1135
1136/**
1137 * mark_buffer_dirty - mark a buffer_head as needing writeout
Martin Waitz67be2dd2005-05-01 08:59:26 -07001138 * @bh: the buffer_head to mark dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001139 *
1140 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1141 * backing page dirty, then tag the page as dirty in its address_space's radix
1142 * tree and then attach the address_space's inode to its superblock's dirty
1143 * inode list.
1144 *
1145 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
Dave Chinner250df6e2011-03-22 22:23:36 +11001146 * mapping->tree_lock and mapping->host->i_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001147 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001148void mark_buffer_dirty(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001149{
Nick Piggin787d2212007-07-17 04:03:34 -07001150 WARN_ON_ONCE(!buffer_uptodate(bh));
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001151
Tejun Heo5305cb82013-01-11 13:06:36 -08001152 trace_block_dirty_buffer(bh);
1153
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001154 /*
1155 * Very *carefully* optimize the it-is-already-dirty case.
1156 *
1157 * Don't let the final "is it dirty" escape to before we
1158 * perhaps modified the buffer.
1159 */
1160 if (buffer_dirty(bh)) {
1161 smp_mb();
1162 if (buffer_dirty(bh))
1163 return;
1164 }
1165
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001166 if (!test_set_buffer_dirty(bh)) {
1167 struct page *page = bh->b_page;
Greg Thelenc4843a72015-05-22 17:13:16 -04001168 struct address_space *mapping = NULL;
Greg Thelenc4843a72015-05-22 17:13:16 -04001169
Johannes Weiner62cccb82016-03-15 14:57:22 -07001170 lock_page_memcg(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001171 if (!TestSetPageDirty(page)) {
Greg Thelenc4843a72015-05-22 17:13:16 -04001172 mapping = page_mapping(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001173 if (mapping)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001174 __set_page_dirty(page, mapping, 0);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001175 }
Johannes Weiner62cccb82016-03-15 14:57:22 -07001176 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04001177 if (mapping)
1178 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001179 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001180}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001181EXPORT_SYMBOL(mark_buffer_dirty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182
1183/*
1184 * Decrement a buffer_head's reference count. If all buffers against a page
1185 * have zero reference count, are clean and unlocked, and if the page is clean
1186 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1187 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1188 * a page but it ends up not being freed, and buffers may later be reattached).
1189 */
1190void __brelse(struct buffer_head * buf)
1191{
1192 if (atomic_read(&buf->b_count)) {
1193 put_bh(buf);
1194 return;
1195 }
Arjan van de Ven5c752ad2008-07-25 19:45:40 -07001196 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001198EXPORT_SYMBOL(__brelse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199
1200/*
1201 * bforget() is like brelse(), except it discards any
1202 * potentially dirty data.
1203 */
1204void __bforget(struct buffer_head *bh)
1205{
1206 clear_buffer_dirty(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -08001207 if (bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208 struct address_space *buffer_mapping = bh->b_page->mapping;
1209
1210 spin_lock(&buffer_mapping->private_lock);
1211 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -07001212 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001213 spin_unlock(&buffer_mapping->private_lock);
1214 }
1215 __brelse(bh);
1216}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001217EXPORT_SYMBOL(__bforget);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218
1219static struct buffer_head *__bread_slow(struct buffer_head *bh)
1220{
1221 lock_buffer(bh);
1222 if (buffer_uptodate(bh)) {
1223 unlock_buffer(bh);
1224 return bh;
1225 } else {
1226 get_bh(bh);
1227 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05001228 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 wait_on_buffer(bh);
1230 if (buffer_uptodate(bh))
1231 return bh;
1232 }
1233 brelse(bh);
1234 return NULL;
1235}
1236
1237/*
1238 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1239 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1240 * refcount elevated by one when they're in an LRU. A buffer can only appear
1241 * once in a particular CPU's LRU. A single buffer can be present in multiple
1242 * CPU's LRUs at the same time.
1243 *
1244 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1245 * sb_find_get_block().
1246 *
1247 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1248 * a local interrupt disable for that.
1249 */
1250
Sebastien Buisson86cf78d2014-10-09 15:29:38 -07001251#define BH_LRU_SIZE 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252
1253struct bh_lru {
1254 struct buffer_head *bhs[BH_LRU_SIZE];
1255};
1256
1257static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1258
1259#ifdef CONFIG_SMP
1260#define bh_lru_lock() local_irq_disable()
1261#define bh_lru_unlock() local_irq_enable()
1262#else
1263#define bh_lru_lock() preempt_disable()
1264#define bh_lru_unlock() preempt_enable()
1265#endif
1266
1267static inline void check_irqs_on(void)
1268{
1269#ifdef irqs_disabled
1270 BUG_ON(irqs_disabled());
1271#endif
1272}
1273
1274/*
1275 * The LRU management algorithm is dopey-but-simple. Sorry.
1276 */
1277static void bh_lru_install(struct buffer_head *bh)
1278{
1279 struct buffer_head *evictee = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001280
1281 check_irqs_on();
1282 bh_lru_lock();
Christoph Lameterc7b92512010-12-06 11:16:28 -06001283 if (__this_cpu_read(bh_lrus.bhs[0]) != bh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001284 struct buffer_head *bhs[BH_LRU_SIZE];
1285 int in;
1286 int out = 0;
1287
1288 get_bh(bh);
1289 bhs[out++] = bh;
1290 for (in = 0; in < BH_LRU_SIZE; in++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001291 struct buffer_head *bh2 =
1292 __this_cpu_read(bh_lrus.bhs[in]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001293
1294 if (bh2 == bh) {
1295 __brelse(bh2);
1296 } else {
1297 if (out >= BH_LRU_SIZE) {
1298 BUG_ON(evictee != NULL);
1299 evictee = bh2;
1300 } else {
1301 bhs[out++] = bh2;
1302 }
1303 }
1304 }
1305 while (out < BH_LRU_SIZE)
1306 bhs[out++] = NULL;
Christoph Lameterca6673b02013-12-03 17:32:53 -06001307 memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 }
1309 bh_lru_unlock();
1310
1311 if (evictee)
1312 __brelse(evictee);
1313}
1314
1315/*
1316 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1317 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001318static struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001319lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001320{
1321 struct buffer_head *ret = NULL;
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001322 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323
1324 check_irqs_on();
1325 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326 for (i = 0; i < BH_LRU_SIZE; i++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001327 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328
Zach Brown9470dd52014-10-13 15:55:05 -07001329 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1330 bh->b_size == size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001331 if (i) {
1332 while (i) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001333 __this_cpu_write(bh_lrus.bhs[i],
1334 __this_cpu_read(bh_lrus.bhs[i - 1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001335 i--;
1336 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06001337 __this_cpu_write(bh_lrus.bhs[0], bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 }
1339 get_bh(bh);
1340 ret = bh;
1341 break;
1342 }
1343 }
1344 bh_lru_unlock();
1345 return ret;
1346}
1347
1348/*
1349 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1350 * it in the LRU and mark it as accessed. If it is not present then return
1351 * NULL
1352 */
1353struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001354__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001355{
1356 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1357
1358 if (bh == NULL) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001359 /* __find_get_block_slow will mark the page accessed */
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001360 bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 if (bh)
1362 bh_lru_install(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001363 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001364 touch_buffer(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001365
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366 return bh;
1367}
1368EXPORT_SYMBOL(__find_get_block);
1369
1370/*
Gioh Kim3b5e6452014-09-04 22:04:42 -04001371 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 * which corresponds to the passed block_device, block and size. The
1373 * returned buffer has its reference count incremented.
1374 *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001375 * __getblk_gfp() will lock up the machine if grow_dev_page's
1376 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 */
1378struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001379__getblk_gfp(struct block_device *bdev, sector_t block,
1380 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381{
1382 struct buffer_head *bh = __find_get_block(bdev, block, size);
1383
1384 might_sleep();
1385 if (bh == NULL)
Gioh Kim3b5e6452014-09-04 22:04:42 -04001386 bh = __getblk_slow(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387 return bh;
1388}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001389EXPORT_SYMBOL(__getblk_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390
1391/*
1392 * Do async read-ahead on a buffer..
1393 */
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001394void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395{
1396 struct buffer_head *bh = __getblk(bdev, block, size);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001397 if (likely(bh)) {
Christoph Hellwig70246282016-07-19 11:28:41 +02001398 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001399 brelse(bh);
1400 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401}
1402EXPORT_SYMBOL(__breadahead);
1403
1404/**
Gioh Kim3b5e6452014-09-04 22:04:42 -04001405 * __bread_gfp() - reads a specified block and returns the bh
Martin Waitz67be2dd2005-05-01 08:59:26 -07001406 * @bdev: the block_device to read from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 * @block: number of block
1408 * @size: size (in bytes) to read
Gioh Kim3b5e6452014-09-04 22:04:42 -04001409 * @gfp: page allocation flag
1410 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411 * Reads a specified block, and returns buffer head that contains it.
Gioh Kim3b5e6452014-09-04 22:04:42 -04001412 * The page cache can be allocated from non-movable area
1413 * not to prevent page migration if you set gfp to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001414 * It returns NULL if the block was unreadable.
1415 */
1416struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001417__bread_gfp(struct block_device *bdev, sector_t block,
1418 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001419{
Gioh Kim3b5e6452014-09-04 22:04:42 -04001420 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001421
Andrew Mortona3e713b2005-10-30 15:03:15 -08001422 if (likely(bh) && !buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423 bh = __bread_slow(bh);
1424 return bh;
1425}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001426EXPORT_SYMBOL(__bread_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427
1428/*
1429 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1430 * This doesn't race because it runs in each cpu either in irq
1431 * or with preempt disabled.
1432 */
1433static void invalidate_bh_lru(void *arg)
1434{
1435 struct bh_lru *b = &get_cpu_var(bh_lrus);
1436 int i;
1437
1438 for (i = 0; i < BH_LRU_SIZE; i++) {
1439 brelse(b->bhs[i]);
1440 b->bhs[i] = NULL;
1441 }
1442 put_cpu_var(bh_lrus);
1443}
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001444
1445static bool has_bh_in_lru(int cpu, void *dummy)
1446{
1447 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1448 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001449
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001450 for (i = 0; i < BH_LRU_SIZE; i++) {
1451 if (b->bhs[i])
1452 return 1;
1453 }
1454
1455 return 0;
1456}
1457
Peter Zijlstraf9a14392007-05-06 14:49:55 -07001458void invalidate_bh_lrus(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459{
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001460 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
Nick Piggin9db55792008-02-08 04:19:49 -08001462EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463
1464void set_bh_page(struct buffer_head *bh,
1465 struct page *page, unsigned long offset)
1466{
1467 bh->b_page = page;
Eric Sesterhenne827f922006-03-26 18:24:46 +02001468 BUG_ON(offset >= PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 if (PageHighMem(page))
1470 /*
1471 * This catches illegal uses and preserves the offset:
1472 */
1473 bh->b_data = (char *)(0 + offset);
1474 else
1475 bh->b_data = page_address(page) + offset;
1476}
1477EXPORT_SYMBOL(set_bh_page);
1478
1479/*
1480 * Called when truncating a buffer on a page completely.
1481 */
Mel Gormane7470ee2014-06-04 16:10:29 -07001482
1483/* Bits that are cleared during an invalidate */
1484#define BUFFER_FLAGS_DISCARD \
1485 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1486 1 << BH_Delay | 1 << BH_Unwritten)
1487
Arjan van de Ven858119e2006-01-14 13:20:43 -08001488static void discard_buffer(struct buffer_head * bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489{
Mel Gormane7470ee2014-06-04 16:10:29 -07001490 unsigned long b_state, b_state_old;
1491
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492 lock_buffer(bh);
1493 clear_buffer_dirty(bh);
1494 bh->b_bdev = NULL;
Mel Gormane7470ee2014-06-04 16:10:29 -07001495 b_state = bh->b_state;
1496 for (;;) {
1497 b_state_old = cmpxchg(&bh->b_state, b_state,
1498 (b_state & ~BUFFER_FLAGS_DISCARD));
1499 if (b_state_old == b_state)
1500 break;
1501 b_state = b_state_old;
1502 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 unlock_buffer(bh);
1504}
1505
1506/**
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001507 * block_invalidatepage - invalidate part or all of a buffer-backed page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001508 *
1509 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -04001510 * @offset: start of the range to invalidate
1511 * @length: length of the range to invalidate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 *
1513 * block_invalidatepage() is called when all or part of the page has become
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001514 * invalidated by a truncate operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 *
1516 * block_invalidatepage() does not have to release all buffers, but it must
1517 * ensure that no dirty buffer is left outside @offset and that no I/O
1518 * is underway against any of the blocks which are outside the truncation
1519 * point. Because the caller is about to free (and possibly reuse) those
1520 * blocks on-disk.
1521 */
Lukas Czernerd47992f2013-05-21 23:17:23 -04001522void block_invalidatepage(struct page *page, unsigned int offset,
1523 unsigned int length)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001524{
1525 struct buffer_head *head, *bh, *next;
1526 unsigned int curr_off = 0;
Lukas Czernerd47992f2013-05-21 23:17:23 -04001527 unsigned int stop = length + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528
1529 BUG_ON(!PageLocked(page));
1530 if (!page_has_buffers(page))
1531 goto out;
1532
Lukas Czernerd47992f2013-05-21 23:17:23 -04001533 /*
1534 * Check for overflow
1535 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001536 BUG_ON(stop > PAGE_SIZE || stop < length);
Lukas Czernerd47992f2013-05-21 23:17:23 -04001537
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538 head = page_buffers(page);
1539 bh = head;
1540 do {
1541 unsigned int next_off = curr_off + bh->b_size;
1542 next = bh->b_this_page;
1543
1544 /*
Lukas Czernerd47992f2013-05-21 23:17:23 -04001545 * Are we still fully in range ?
1546 */
1547 if (next_off > stop)
1548 goto out;
1549
1550 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 * is this block fully invalidated?
1552 */
1553 if (offset <= curr_off)
1554 discard_buffer(bh);
1555 curr_off = next_off;
1556 bh = next;
1557 } while (bh != head);
1558
1559 /*
1560 * We release buffers only if the entire page is being invalidated.
1561 * The get_block cached value has been unconditionally invalidated,
1562 * so real IO is not possible anymore.
1563 */
1564 if (offset == 0)
NeilBrown2ff28e22006-03-26 01:37:18 -08001565 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566out:
NeilBrown2ff28e22006-03-26 01:37:18 -08001567 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568}
1569EXPORT_SYMBOL(block_invalidatepage);
1570
Lukas Czernerd47992f2013-05-21 23:17:23 -04001571
Linus Torvalds1da177e2005-04-16 15:20:36 -07001572/*
1573 * We attach and possibly dirty the buffers atomically wrt
1574 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1575 * is already excluded via the page lock.
1576 */
1577void create_empty_buffers(struct page *page,
1578 unsigned long blocksize, unsigned long b_state)
1579{
1580 struct buffer_head *bh, *head, *tail;
1581
1582 head = alloc_page_buffers(page, blocksize, 1);
1583 bh = head;
1584 do {
1585 bh->b_state |= b_state;
1586 tail = bh;
1587 bh = bh->b_this_page;
1588 } while (bh);
1589 tail->b_this_page = head;
1590
1591 spin_lock(&page->mapping->private_lock);
1592 if (PageUptodate(page) || PageDirty(page)) {
1593 bh = head;
1594 do {
1595 if (PageDirty(page))
1596 set_buffer_dirty(bh);
1597 if (PageUptodate(page))
1598 set_buffer_uptodate(bh);
1599 bh = bh->b_this_page;
1600 } while (bh != head);
1601 }
1602 attach_page_buffers(page, head);
1603 spin_unlock(&page->mapping->private_lock);
1604}
1605EXPORT_SYMBOL(create_empty_buffers);
1606
1607/*
1608 * We are taking a block for data and we don't want any output from any
1609 * buffer-cache aliases starting from return from that function and
1610 * until the moment when something will explicitly mark the buffer
1611 * dirty (hopefully that will not happen until we will free that block ;-)
1612 * We don't even need to mark it not-uptodate - nobody can expect
1613 * anything from a newly allocated buffer anyway. We used to used
1614 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1615 * don't want to mark the alias unmapped, for example - it would confuse
1616 * anyone who might pick it with bread() afterwards...
1617 *
1618 * Also.. Note that bforget() doesn't lock the buffer. So there can
1619 * be writeout I/O going on against recently-freed buffers. We don't
1620 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1621 * only if we really need to. That happens here.
1622 */
1623void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
1624{
1625 struct buffer_head *old_bh;
1626
1627 might_sleep();
1628
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001629 old_bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630 if (old_bh) {
1631 clear_buffer_dirty(old_bh);
1632 wait_on_buffer(old_bh);
1633 clear_buffer_req(old_bh);
1634 __brelse(old_bh);
1635 }
1636}
1637EXPORT_SYMBOL(unmap_underlying_metadata);
1638
1639/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001640 * Size is a power-of-two in the range 512..PAGE_SIZE,
1641 * and the case we care about most is PAGE_SIZE.
1642 *
1643 * So this *could* possibly be written with those
1644 * constraints in mind (relevant mostly if some
1645 * architecture has a slow bit-scan instruction)
1646 */
1647static inline int block_size_bits(unsigned int blocksize)
1648{
1649 return ilog2(blocksize);
1650}
1651
1652static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1653{
1654 BUG_ON(!PageLocked(page));
1655
1656 if (!page_has_buffers(page))
1657 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1658 return page_buffers(page);
1659}
1660
1661/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001662 * NOTE! All mapped/uptodate combinations are valid:
1663 *
1664 * Mapped Uptodate Meaning
1665 *
1666 * No No "unknown" - must do get_block()
1667 * No Yes "hole" - zero-filled
1668 * Yes No "allocated" - allocated on disk, not read in
1669 * Yes Yes "valid" - allocated and up-to-date in memory.
1670 *
1671 * "Dirty" is valid only with the last case (mapped+uptodate).
1672 */
1673
1674/*
1675 * While block_write_full_page is writing back the dirty buffers under
1676 * the page lock, whoever dirtied the buffers may decide to clean them
1677 * again at any time. We handle that by only looking at the buffer
1678 * state inside lock_buffer().
1679 *
1680 * If block_write_full_page() is called for regular writeback
1681 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1682 * locked buffer. This only can happen if someone has written the buffer
1683 * directly, with submit_bh(). At the address_space level PageWriteback
1684 * prevents this contention from occurring.
Theodore Ts'o6e34eed2009-04-07 18:12:43 -04001685 *
1686 * If block_write_full_page() is called with wbc->sync_mode ==
Jens Axboe721a9602011-03-09 11:56:30 +01001687 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1688 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689 */
1690static int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001691 get_block_t *get_block, struct writeback_control *wbc,
1692 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001693{
1694 int err;
1695 sector_t block;
1696 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001697 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001698 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001699 int nr_underway = 0;
Mike Christie2a222ca2016-06-05 14:31:43 -05001700 int write_flags = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001702 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001703 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704
1705 /*
1706 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1707 * here, and the (potentially unmapped) buffers may become dirty at
1708 * any time. If a buffer becomes dirty here after we've inspected it
1709 * then we just miss that fact, and the page stays dirty.
1710 *
1711 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1712 * handle that here by just cleaning them.
1713 */
1714
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001716 blocksize = bh->b_size;
1717 bbits = block_size_bits(blocksize);
1718
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001719 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001720 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721
1722 /*
1723 * Get all the dirty buffers mapped to disk addresses and
1724 * handle any aliases from the underlying blockdev's mapping.
1725 */
1726 do {
1727 if (block > last_block) {
1728 /*
1729 * mapped buffers outside i_size will occur, because
1730 * this page can be outside i_size when there is a
1731 * truncate in progress.
1732 */
1733 /*
1734 * The buffer was zeroed by block_write_full_page()
1735 */
1736 clear_buffer_dirty(bh);
1737 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001738 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1739 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001740 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741 err = get_block(inode, block, bh, 1);
1742 if (err)
1743 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001744 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745 if (buffer_new(bh)) {
1746 /* blockdev mappings never come here */
1747 clear_buffer_new(bh);
1748 unmap_underlying_metadata(bh->b_bdev,
1749 bh->b_blocknr);
1750 }
1751 }
1752 bh = bh->b_this_page;
1753 block++;
1754 } while (bh != head);
1755
1756 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757 if (!buffer_mapped(bh))
1758 continue;
1759 /*
1760 * If it's a fully non-blocking write attempt and we cannot
1761 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001762 * potentially cause a busy-wait loop from writeback threads
1763 * and kswapd activity, but those code paths have their own
1764 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001766 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001768 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001769 redirty_page_for_writepage(wbc, page);
1770 continue;
1771 }
1772 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001773 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 } else {
1775 unlock_buffer(bh);
1776 }
1777 } while ((bh = bh->b_this_page) != head);
1778
1779 /*
1780 * The page and its buffers are protected by PageWriteback(), so we can
1781 * drop the bh refcounts early.
1782 */
1783 BUG_ON(PageWriteback(page));
1784 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785
1786 do {
1787 struct buffer_head *next = bh->b_this_page;
1788 if (buffer_async_write(bh)) {
Mike Christie2a222ca2016-06-05 14:31:43 -05001789 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001790 nr_underway++;
1791 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 bh = next;
1793 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001794 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795
1796 err = 0;
1797done:
1798 if (nr_underway == 0) {
1799 /*
1800 * The page was marked dirty, but the buffers were
1801 * clean. Someone wrote them back by hand with
1802 * ll_rw_block/submit_bh. A rare case.
1803 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001805
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806 /*
1807 * The page and buffer_heads can be released at any time from
1808 * here on.
1809 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 }
1811 return err;
1812
1813recover:
1814 /*
1815 * ENOSPC, or some other error. We may already have added some
1816 * blocks to the file, so we need to write these out to avoid
1817 * exposing stale data.
1818 * The page is currently locked and not marked for writeback
1819 */
1820 bh = head;
1821 /* Recovery: lock and submit the mapped buffers */
1822 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001823 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1824 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001826 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 } else {
1828 /*
1829 * The buffer may have been set dirty during
1830 * attachment to a dirty page.
1831 */
1832 clear_buffer_dirty(bh);
1833 }
1834 } while ((bh = bh->b_this_page) != head);
1835 SetPageError(page);
1836 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001837 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 do {
1840 struct buffer_head *next = bh->b_this_page;
1841 if (buffer_async_write(bh)) {
1842 clear_buffer_dirty(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05001843 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844 nr_underway++;
1845 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846 bh = next;
1847 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001848 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 goto done;
1850}
1851
Nick Pigginafddba42007-10-16 01:25:01 -07001852/*
1853 * If a page has any new buffers, zero them out here, and mark them uptodate
1854 * and dirty so they'll be written out (in order to prevent uninitialised
1855 * block data from leaking). And clear the new bit.
1856 */
1857void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1858{
1859 unsigned int block_start, block_end;
1860 struct buffer_head *head, *bh;
1861
1862 BUG_ON(!PageLocked(page));
1863 if (!page_has_buffers(page))
1864 return;
1865
1866 bh = head = page_buffers(page);
1867 block_start = 0;
1868 do {
1869 block_end = block_start + bh->b_size;
1870
1871 if (buffer_new(bh)) {
1872 if (block_end > from && block_start < to) {
1873 if (!PageUptodate(page)) {
1874 unsigned start, size;
1875
1876 start = max(from, block_start);
1877 size = min(to, block_end) - start;
1878
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001879 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001880 set_buffer_uptodate(bh);
1881 }
1882
1883 clear_buffer_new(bh);
1884 mark_buffer_dirty(bh);
1885 }
1886 }
1887
1888 block_start = block_end;
1889 bh = bh->b_this_page;
1890 } while (bh != head);
1891}
1892EXPORT_SYMBOL(page_zero_new_buffers);
1893
Christoph Hellwigebdec242010-10-06 10:47:23 +02001894int __block_write_begin(struct page *page, loff_t pos, unsigned len,
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001895 get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001897 unsigned from = pos & (PAGE_SIZE - 1);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001898 unsigned to = from + len;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001899 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900 unsigned block_start, block_end;
1901 sector_t block;
1902 int err = 0;
1903 unsigned blocksize, bbits;
1904 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1905
1906 BUG_ON(!PageLocked(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001907 BUG_ON(from > PAGE_SIZE);
1908 BUG_ON(to > PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909 BUG_ON(from > to);
1910
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001911 head = create_page_buffers(page, inode, 0);
1912 blocksize = head->b_size;
1913 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001914
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001915 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
1917 for(bh = head, block_start = 0; bh != head || !block_start;
1918 block++, block_start=block_end, bh = bh->b_this_page) {
1919 block_end = block_start + blocksize;
1920 if (block_end <= from || block_start >= to) {
1921 if (PageUptodate(page)) {
1922 if (!buffer_uptodate(bh))
1923 set_buffer_uptodate(bh);
1924 }
1925 continue;
1926 }
1927 if (buffer_new(bh))
1928 clear_buffer_new(bh);
1929 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001930 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001931 err = get_block(inode, block, bh, 1);
1932 if (err)
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07001933 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934 if (buffer_new(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935 unmap_underlying_metadata(bh->b_bdev,
1936 bh->b_blocknr);
1937 if (PageUptodate(page)) {
Nick Piggin637aff42007-10-16 01:25:00 -07001938 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001939 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07001940 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941 continue;
1942 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001943 if (block_end > to || block_start < from)
1944 zero_user_segments(page,
1945 to, block_end,
1946 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001947 continue;
1948 }
1949 }
1950 if (PageUptodate(page)) {
1951 if (!buffer_uptodate(bh))
1952 set_buffer_uptodate(bh);
1953 continue;
1954 }
1955 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08001956 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 (block_start < from || block_end > to)) {
Mike Christiedfec8a12016-06-05 14:31:44 -05001958 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001959 *wait_bh++=bh;
1960 }
1961 }
1962 /*
1963 * If we issued read requests - let them complete.
1964 */
1965 while(wait_bh > wait) {
1966 wait_on_buffer(*--wait_bh);
1967 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07001968 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001969 }
Jan Karaf9f07b62011-06-14 00:58:27 +02001970 if (unlikely(err))
Nick Pigginafddba42007-10-16 01:25:01 -07001971 page_zero_new_buffers(page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972 return err;
1973}
Christoph Hellwigebdec242010-10-06 10:47:23 +02001974EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975
1976static int __block_commit_write(struct inode *inode, struct page *page,
1977 unsigned from, unsigned to)
1978{
1979 unsigned block_start, block_end;
1980 int partial = 0;
1981 unsigned blocksize;
1982 struct buffer_head *bh, *head;
1983
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001984 bh = head = page_buffers(page);
1985 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001987 block_start = 0;
1988 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989 block_end = block_start + blocksize;
1990 if (block_end <= from || block_start >= to) {
1991 if (!buffer_uptodate(bh))
1992 partial = 1;
1993 } else {
1994 set_buffer_uptodate(bh);
1995 mark_buffer_dirty(bh);
1996 }
Nick Pigginafddba42007-10-16 01:25:01 -07001997 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001998
1999 block_start = block_end;
2000 bh = bh->b_this_page;
2001 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002
2003 /*
2004 * If this is a partial write which happened to make all buffers
2005 * uptodate then we can optimize away a bogus readpage() for
2006 * the next read(). Here we 'discover' whether the page went
2007 * uptodate as a result of this (potentially partial) write.
2008 */
2009 if (!partial)
2010 SetPageUptodate(page);
2011 return 0;
2012}
2013
2014/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02002015 * block_write_begin takes care of the basic task of block allocation and
2016 * bringing partial write blocks uptodate first.
2017 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002018 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07002019 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02002020int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2021 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07002022{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002023 pgoff_t index = pos >> PAGE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07002024 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002025 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07002026
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002027 page = grab_cache_page_write_begin(mapping, index, flags);
2028 if (!page)
2029 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07002030
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002031 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07002032 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002033 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002034 put_page(page);
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002035 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07002036 }
2037
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002038 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07002039 return status;
2040}
2041EXPORT_SYMBOL(block_write_begin);
2042
2043int block_write_end(struct file *file, struct address_space *mapping,
2044 loff_t pos, unsigned len, unsigned copied,
2045 struct page *page, void *fsdata)
2046{
2047 struct inode *inode = mapping->host;
2048 unsigned start;
2049
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002050 start = pos & (PAGE_SIZE - 1);
Nick Pigginafddba42007-10-16 01:25:01 -07002051
2052 if (unlikely(copied < len)) {
2053 /*
2054 * The buffers that were written will now be uptodate, so we
2055 * don't have to worry about a readpage reading them and
2056 * overwriting a partial write. However if we have encountered
2057 * a short write and only partially written into a buffer, it
2058 * will not be marked uptodate, so a readpage might come in and
2059 * destroy our partial write.
2060 *
2061 * Do the simplest thing, and just treat any short write to a
2062 * non uptodate page as a zero-length write, and force the
2063 * caller to redo the whole thing.
2064 */
2065 if (!PageUptodate(page))
2066 copied = 0;
2067
2068 page_zero_new_buffers(page, start+copied, start+len);
2069 }
2070 flush_dcache_page(page);
2071
2072 /* This could be a short (even 0-length) commit */
2073 __block_commit_write(inode, page, start, start+copied);
2074
2075 return copied;
2076}
2077EXPORT_SYMBOL(block_write_end);
2078
2079int generic_write_end(struct file *file, struct address_space *mapping,
2080 loff_t pos, unsigned len, unsigned copied,
2081 struct page *page, void *fsdata)
2082{
2083 struct inode *inode = mapping->host;
Jan Kara90a80202014-10-01 21:49:18 -04002084 loff_t old_size = inode->i_size;
Jan Karac7d206b2008-07-11 19:27:31 -04002085 int i_size_changed = 0;
Nick Pigginafddba42007-10-16 01:25:01 -07002086
2087 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2088
2089 /*
2090 * No need to use i_size_read() here, the i_size
2091 * cannot change under us because we hold i_mutex.
2092 *
2093 * But it's important to update i_size while still holding page lock:
2094 * page writeout could otherwise come in and zero beyond i_size.
2095 */
2096 if (pos+copied > inode->i_size) {
2097 i_size_write(inode, pos+copied);
Jan Karac7d206b2008-07-11 19:27:31 -04002098 i_size_changed = 1;
Nick Pigginafddba42007-10-16 01:25:01 -07002099 }
2100
2101 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002102 put_page(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002103
Jan Kara90a80202014-10-01 21:49:18 -04002104 if (old_size < pos)
2105 pagecache_isize_extended(inode, old_size, pos);
Jan Karac7d206b2008-07-11 19:27:31 -04002106 /*
2107 * Don't mark the inode dirty under page lock. First, it unnecessarily
2108 * makes the holding time of page lock longer. Second, it forces lock
2109 * ordering of page lock and transaction start for journaling
2110 * filesystems.
2111 */
2112 if (i_size_changed)
2113 mark_inode_dirty(inode);
2114
Nick Pigginafddba42007-10-16 01:25:01 -07002115 return copied;
2116}
2117EXPORT_SYMBOL(generic_write_end);
2118
2119/*
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002120 * block_is_partially_uptodate checks whether buffers within a page are
2121 * uptodate or not.
2122 *
2123 * Returns true if all buffers which correspond to a file portion
2124 * we want to read are uptodate.
2125 */
Al Viroc186afb42014-02-02 21:16:54 -05002126int block_is_partially_uptodate(struct page *page, unsigned long from,
2127 unsigned long count)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002128{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002129 unsigned block_start, block_end, blocksize;
2130 unsigned to;
2131 struct buffer_head *bh, *head;
2132 int ret = 1;
2133
2134 if (!page_has_buffers(page))
2135 return 0;
2136
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002137 head = page_buffers(page);
2138 blocksize = head->b_size;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002139 to = min_t(unsigned, PAGE_SIZE - from, count);
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002140 to = from + to;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002141 if (from < blocksize && to > PAGE_SIZE - blocksize)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002142 return 0;
2143
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002144 bh = head;
2145 block_start = 0;
2146 do {
2147 block_end = block_start + blocksize;
2148 if (block_end > from && block_start < to) {
2149 if (!buffer_uptodate(bh)) {
2150 ret = 0;
2151 break;
2152 }
2153 if (block_end >= to)
2154 break;
2155 }
2156 block_start = block_end;
2157 bh = bh->b_this_page;
2158 } while (bh != head);
2159
2160 return ret;
2161}
2162EXPORT_SYMBOL(block_is_partially_uptodate);
2163
2164/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002165 * Generic "read page" function for block devices that have the normal
2166 * get_block functionality. This is most of the block device filesystems.
2167 * Reads the page asynchronously --- the unlock_buffer() and
2168 * set/clear_buffer_uptodate() functions propagate buffer state into the
2169 * page struct once IO has completed.
2170 */
2171int block_read_full_page(struct page *page, get_block_t *get_block)
2172{
2173 struct inode *inode = page->mapping->host;
2174 sector_t iblock, lblock;
2175 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002176 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177 int nr, i;
2178 int fully_mapped = 1;
2179
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002180 head = create_page_buffers(page, inode, 0);
2181 blocksize = head->b_size;
2182 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002183
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002184 iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002185 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186 bh = head;
2187 nr = 0;
2188 i = 0;
2189
2190 do {
2191 if (buffer_uptodate(bh))
2192 continue;
2193
2194 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002195 int err = 0;
2196
Linus Torvalds1da177e2005-04-16 15:20:36 -07002197 fully_mapped = 0;
2198 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002199 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002200 err = get_block(inode, iblock, bh, 0);
2201 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002202 SetPageError(page);
2203 }
2204 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002205 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002206 if (!err)
2207 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002208 continue;
2209 }
2210 /*
2211 * get_block() might have updated the buffer
2212 * synchronously
2213 */
2214 if (buffer_uptodate(bh))
2215 continue;
2216 }
2217 arr[nr++] = bh;
2218 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2219
2220 if (fully_mapped)
2221 SetPageMappedToDisk(page);
2222
2223 if (!nr) {
2224 /*
2225 * All buffers are uptodate - we can set the page uptodate
2226 * as well. But not if get_block() returned an error.
2227 */
2228 if (!PageError(page))
2229 SetPageUptodate(page);
2230 unlock_page(page);
2231 return 0;
2232 }
2233
2234 /* Stage two: lock the buffers */
2235 for (i = 0; i < nr; i++) {
2236 bh = arr[i];
2237 lock_buffer(bh);
2238 mark_buffer_async_read(bh);
2239 }
2240
2241 /*
2242 * Stage 3: start the IO. Check for uptodateness
2243 * inside the buffer lock in case another process reading
2244 * the underlying blockdev brought it uptodate (the sct fix).
2245 */
2246 for (i = 0; i < nr; i++) {
2247 bh = arr[i];
2248 if (buffer_uptodate(bh))
2249 end_buffer_async_read(bh, 1);
2250 else
Mike Christie2a222ca2016-06-05 14:31:43 -05002251 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 }
2253 return 0;
2254}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002255EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002256
2257/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002258 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002259 * deal with the hole.
2260 */
Nick Piggin89e10782007-10-16 01:25:07 -07002261int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002262{
2263 struct address_space *mapping = inode->i_mapping;
2264 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002265 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266 int err;
2267
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002268 err = inode_newsize_ok(inode, size);
2269 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002270 goto out;
2271
Nick Piggin89e10782007-10-16 01:25:07 -07002272 err = pagecache_write_begin(NULL, mapping, size, 0,
2273 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2274 &page, &fsdata);
2275 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002276 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002277
Nick Piggin89e10782007-10-16 01:25:07 -07002278 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2279 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002280
Linus Torvalds1da177e2005-04-16 15:20:36 -07002281out:
2282 return err;
2283}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002284EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002286static int cont_expand_zero(struct file *file, struct address_space *mapping,
2287 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002288{
Nick Piggin89e10782007-10-16 01:25:07 -07002289 struct inode *inode = mapping->host;
2290 unsigned blocksize = 1 << inode->i_blkbits;
2291 struct page *page;
2292 void *fsdata;
2293 pgoff_t index, curidx;
2294 loff_t curpos;
2295 unsigned zerofrom, offset, len;
2296 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002297
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002298 index = pos >> PAGE_SHIFT;
2299 offset = pos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002300
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002301 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2302 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002303 if (zerofrom & (blocksize-1)) {
2304 *bytes |= (blocksize-1);
2305 (*bytes)++;
2306 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002307 len = PAGE_SIZE - zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002308
2309 err = pagecache_write_begin(file, mapping, curpos, len,
2310 AOP_FLAG_UNINTERRUPTIBLE,
2311 &page, &fsdata);
2312 if (err)
2313 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002314 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002315 err = pagecache_write_end(file, mapping, curpos, len, len,
2316 page, fsdata);
2317 if (err < 0)
2318 goto out;
2319 BUG_ON(err != len);
2320 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002321
2322 balance_dirty_pages_ratelimited(mapping);
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002323
2324 if (unlikely(fatal_signal_pending(current))) {
2325 err = -EINTR;
2326 goto out;
2327 }
Nick Piggin89e10782007-10-16 01:25:07 -07002328 }
2329
2330 /* page covers the boundary, find the boundary offset */
2331 if (index == curidx) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002332 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002333 /* if we will expand the thing last block will be filled */
2334 if (offset <= zerofrom) {
2335 goto out;
2336 }
2337 if (zerofrom & (blocksize-1)) {
2338 *bytes |= (blocksize-1);
2339 (*bytes)++;
2340 }
2341 len = offset - zerofrom;
2342
2343 err = pagecache_write_begin(file, mapping, curpos, len,
2344 AOP_FLAG_UNINTERRUPTIBLE,
2345 &page, &fsdata);
2346 if (err)
2347 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002348 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002349 err = pagecache_write_end(file, mapping, curpos, len, len,
2350 page, fsdata);
2351 if (err < 0)
2352 goto out;
2353 BUG_ON(err != len);
2354 err = 0;
2355 }
2356out:
2357 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002358}
2359
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360/*
2361 * For moronic filesystems that do not allow holes in file.
2362 * We may have to extend the file.
2363 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002364int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002365 loff_t pos, unsigned len, unsigned flags,
2366 struct page **pagep, void **fsdata,
2367 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370 unsigned blocksize = 1 << inode->i_blkbits;
Nick Piggin89e10782007-10-16 01:25:07 -07002371 unsigned zerofrom;
2372 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002373
Nick Piggin89e10782007-10-16 01:25:07 -07002374 err = cont_expand_zero(file, mapping, pos, bytes);
2375 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002376 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002377
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002378 zerofrom = *bytes & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002379 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2380 *bytes |= (blocksize-1);
2381 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002382 }
2383
Christoph Hellwig155130a2010-06-04 11:29:58 +02002384 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002385}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002386EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387
Linus Torvalds1da177e2005-04-16 15:20:36 -07002388int block_commit_write(struct page *page, unsigned from, unsigned to)
2389{
2390 struct inode *inode = page->mapping->host;
2391 __block_commit_write(inode,page,from,to);
2392 return 0;
2393}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002394EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002395
David Chinner54171692007-07-19 17:39:55 +10002396/*
2397 * block_page_mkwrite() is not allowed to change the file size as it gets
2398 * called from a page fault handler when a page is first dirtied. Hence we must
2399 * be careful to check for EOF conditions here. We set the page up correctly
2400 * for a written page which means we get ENOSPC checking when writing into
2401 * holes and correct delalloc and unwritten extent mapping on filesystems that
2402 * support these features.
2403 *
2404 * We are not allowed to take the i_mutex here so we have to play games to
2405 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002406 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002407 * page lock we can determine safely if the page is beyond EOF. If it is not
2408 * beyond EOF, then the page is guaranteed safe against truncation until we
2409 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002410 *
Jan Kara14da9202012-06-12 16:20:37 +02002411 * Direct callers of this function should protect against filesystem freezing
Ross Zwisler5c500022015-10-13 16:51:02 -06002412 * using sb_start_pagefault() - sb_end_pagefault() functions.
David Chinner54171692007-07-19 17:39:55 +10002413 */
Ross Zwisler5c500022015-10-13 16:51:02 -06002414int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
Jan Kara24da4fa2011-05-24 00:23:34 +02002415 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002416{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002417 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002418 struct inode *inode = file_inode(vma->vm_file);
David Chinner54171692007-07-19 17:39:55 +10002419 unsigned long end;
2420 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002421 int ret;
David Chinner54171692007-07-19 17:39:55 +10002422
2423 lock_page(page);
2424 size = i_size_read(inode);
2425 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002426 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002427 /* We overload EFAULT to mean page got truncated */
2428 ret = -EFAULT;
2429 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002430 }
2431
2432 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002433 if (((page->index + 1) << PAGE_SHIFT) > size)
2434 end = size & ~PAGE_MASK;
David Chinner54171692007-07-19 17:39:55 +10002435 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002436 end = PAGE_SIZE;
David Chinner54171692007-07-19 17:39:55 +10002437
Christoph Hellwigebdec242010-10-06 10:47:23 +02002438 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002439 if (!ret)
2440 ret = block_commit_write(page, 0, end);
2441
Jan Kara24da4fa2011-05-24 00:23:34 +02002442 if (unlikely(ret < 0))
2443 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002444 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002445 wait_for_stable_page(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002446 return 0;
2447out_unlock:
2448 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002449 return ret;
2450}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002451EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
2453/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002454 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455 * immediately, while under the page lock. So it needs a special end_io
2456 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002457 */
2458static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2459{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002460 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461}
2462
2463/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002464 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2465 * the page (converting it to circular linked list and taking care of page
2466 * dirty races).
2467 */
2468static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2469{
2470 struct buffer_head *bh;
2471
2472 BUG_ON(!PageLocked(page));
2473
2474 spin_lock(&page->mapping->private_lock);
2475 bh = head;
2476 do {
2477 if (PageDirty(page))
2478 set_buffer_dirty(bh);
2479 if (!bh->b_this_page)
2480 bh->b_this_page = head;
2481 bh = bh->b_this_page;
2482 } while (bh != head);
2483 attach_page_buffers(page, head);
2484 spin_unlock(&page->mapping->private_lock);
2485}
2486
2487/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002488 * On entry, the page is fully not uptodate.
2489 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002490 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002491 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002492int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002493 loff_t pos, unsigned len, unsigned flags,
2494 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495 get_block_t *get_block)
2496{
Nick Piggin03158cd2007-10-16 01:25:25 -07002497 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498 const unsigned blkbits = inode->i_blkbits;
2499 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002500 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002501 struct page *page;
2502 pgoff_t index;
2503 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002504 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002505 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002506 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002508 int ret = 0;
2509 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002511 index = pos >> PAGE_SHIFT;
2512 from = pos & (PAGE_SIZE - 1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002513 to = from + len;
2514
Nick Piggin54566b22009-01-04 12:00:53 -08002515 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002516 if (!page)
2517 return -ENOMEM;
2518 *pagep = page;
2519 *fsdata = NULL;
2520
2521 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002522 ret = __block_write_begin(page, pos, len, get_block);
2523 if (unlikely(ret))
2524 goto out_release;
2525 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002526 }
Nick Piggina4b06722007-10-16 01:24:48 -07002527
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528 if (PageMappedToDisk(page))
2529 return 0;
2530
Nick Piggina4b06722007-10-16 01:24:48 -07002531 /*
2532 * Allocate buffers so that we can keep track of state, and potentially
2533 * attach them to the page if an error occurs. In the common case of
2534 * no error, they will just be freed again without ever being attached
2535 * to the page (which is all OK, because we're under the page lock).
2536 *
2537 * Be careful: the buffer linked list is a NULL terminated one, rather
2538 * than the circular one we're used to.
2539 */
2540 head = alloc_page_buffers(page, blocksize, 0);
Nick Piggin03158cd2007-10-16 01:25:25 -07002541 if (!head) {
2542 ret = -ENOMEM;
2543 goto out_release;
2544 }
Nick Piggina4b06722007-10-16 01:24:48 -07002545
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002546 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002547
2548 /*
2549 * We loop across all blocks in the page, whether or not they are
2550 * part of the affected region. This is so we can discover if the
2551 * page is fully mapped-to-disk.
2552 */
Nick Piggina4b06722007-10-16 01:24:48 -07002553 for (block_start = 0, block_in_page = 0, bh = head;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002554 block_start < PAGE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002555 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002556 int create;
2557
Nick Piggina4b06722007-10-16 01:24:48 -07002558 block_end = block_start + blocksize;
2559 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560 create = 1;
2561 if (block_start >= to)
2562 create = 0;
2563 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002564 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 if (ret)
2566 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002567 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002568 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002569 if (buffer_new(bh))
2570 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
2571 if (PageUptodate(page)) {
2572 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002574 }
2575 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002576 zero_user_segments(page, block_start, from,
2577 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002578 continue;
2579 }
Nick Piggina4b06722007-10-16 01:24:48 -07002580 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002581 continue; /* reiserfs does this */
2582 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002583 lock_buffer(bh);
2584 bh->b_end_io = end_buffer_read_nobh;
Mike Christie2a222ca2016-06-05 14:31:43 -05002585 submit_bh(REQ_OP_READ, 0, bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002586 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 }
2588 }
2589
2590 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002591 /*
2592 * The page is locked, so these buffers are protected from
2593 * any VM or truncate activity. Hence we don't need to care
2594 * for the buffer_head refcounts.
2595 */
Nick Piggina4b06722007-10-16 01:24:48 -07002596 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002597 wait_on_buffer(bh);
2598 if (!buffer_uptodate(bh))
2599 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002600 }
2601 if (ret)
2602 goto failed;
2603 }
2604
2605 if (is_mapped_to_disk)
2606 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607
Nick Piggin03158cd2007-10-16 01:25:25 -07002608 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002609
Linus Torvalds1da177e2005-04-16 15:20:36 -07002610 return 0;
2611
2612failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002613 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002615 * Error recovery is a bit difficult. We need to zero out blocks that
2616 * were newly allocated, and dirty them to ensure they get written out.
2617 * Buffers need to be attached to the page at this point, otherwise
2618 * the handling of potential IO errors during writeout would be hard
2619 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002621 attach_nobh_buffers(page, head);
2622 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002623
Nick Piggin03158cd2007-10-16 01:25:25 -07002624out_release:
2625 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002626 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002627 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002628
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002629 return ret;
2630}
Nick Piggin03158cd2007-10-16 01:25:25 -07002631EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632
Nick Piggin03158cd2007-10-16 01:25:25 -07002633int nobh_write_end(struct file *file, struct address_space *mapping,
2634 loff_t pos, unsigned len, unsigned copied,
2635 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636{
2637 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002638 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002639 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002640 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002642 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002643 attach_nobh_buffers(page, head);
2644 if (page_has_buffers(page))
2645 return generic_write_end(file, mapping, pos, len,
2646 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002647
Nick Piggin22c8ca72007-02-20 13:58:09 -08002648 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002649 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002650 if (pos+copied > inode->i_size) {
2651 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 mark_inode_dirty(inode);
2653 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002654
2655 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002656 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002657
Nick Piggin03158cd2007-10-16 01:25:25 -07002658 while (head) {
2659 bh = head;
2660 head = head->b_this_page;
2661 free_buffer_head(bh);
2662 }
2663
2664 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665}
Nick Piggin03158cd2007-10-16 01:25:25 -07002666EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667
2668/*
2669 * nobh_writepage() - based on block_full_write_page() except
2670 * that it tries to operate without attaching bufferheads to
2671 * the page.
2672 */
2673int nobh_writepage(struct page *page, get_block_t *get_block,
2674 struct writeback_control *wbc)
2675{
2676 struct inode * const inode = page->mapping->host;
2677 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002678 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 int ret;
2681
2682 /* Is the page fully inside i_size? */
2683 if (page->index < end_index)
2684 goto out;
2685
2686 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002687 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 if (page->index >= end_index+1 || !offset) {
2689 /*
2690 * The page may have dirty, unmapped buffers. For example,
2691 * they may have been added in ext3_writepage(). Make them
2692 * freeable here, so the page does not leak.
2693 */
2694#if 0
2695 /* Not really sure about this - do we need this ? */
2696 if (page->mapping->a_ops->invalidatepage)
2697 page->mapping->a_ops->invalidatepage(page, offset);
2698#endif
2699 unlock_page(page);
2700 return 0; /* don't care */
2701 }
2702
2703 /*
2704 * The page straddles i_size. It must be zeroed out on each and every
2705 * writepage invocation because it may be mmapped. "A file is mapped
2706 * in multiples of the page size. For a file that is not a multiple of
2707 * the page size, the remaining memory is zeroed when mapped, and
2708 * writes to that region are not written out to the file."
2709 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002710 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711out:
2712 ret = mpage_writepage(page, get_block, wbc);
2713 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002714 ret = __block_write_full_page(inode, page, get_block, wbc,
2715 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 return ret;
2717}
2718EXPORT_SYMBOL(nobh_writepage);
2719
Nick Piggin03158cd2007-10-16 01:25:25 -07002720int nobh_truncate_page(struct address_space *mapping,
2721 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002723 pgoff_t index = from >> PAGE_SHIFT;
2724 unsigned offset = from & (PAGE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002725 unsigned blocksize;
2726 sector_t iblock;
2727 unsigned length, pos;
2728 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002730 struct buffer_head map_bh;
2731 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732
Nick Piggin03158cd2007-10-16 01:25:25 -07002733 blocksize = 1 << inode->i_blkbits;
2734 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002735
Nick Piggin03158cd2007-10-16 01:25:25 -07002736 /* Block boundary? Nothing to do */
2737 if (!length)
2738 return 0;
2739
2740 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002741 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Nick Piggin03158cd2007-10-16 01:25:25 -07002742
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002744 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745 if (!page)
2746 goto out;
2747
Nick Piggin03158cd2007-10-16 01:25:25 -07002748 if (page_has_buffers(page)) {
2749has_buffers:
2750 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002751 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002752 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002753 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002754
2755 /* Find the buffer that contains "offset" */
2756 pos = blocksize;
2757 while (offset >= pos) {
2758 iblock++;
2759 pos += blocksize;
2760 }
2761
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002762 map_bh.b_size = blocksize;
2763 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002764 err = get_block(inode, iblock, &map_bh, 0);
2765 if (err)
2766 goto unlock;
2767 /* unmapped? It's a hole - nothing to do */
2768 if (!buffer_mapped(&map_bh))
2769 goto unlock;
2770
2771 /* Ok, it's mapped. Make sure it's up-to-date */
2772 if (!PageUptodate(page)) {
2773 err = mapping->a_ops->readpage(NULL, page);
2774 if (err) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002775 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002776 goto out;
2777 }
2778 lock_page(page);
2779 if (!PageUptodate(page)) {
2780 err = -EIO;
2781 goto unlock;
2782 }
2783 if (page_has_buffers(page))
2784 goto has_buffers;
2785 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002786 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002787 set_page_dirty(page);
2788 err = 0;
2789
2790unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002792 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002793out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002794 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002795}
2796EXPORT_SYMBOL(nobh_truncate_page);
2797
2798int block_truncate_page(struct address_space *mapping,
2799 loff_t from, get_block_t *get_block)
2800{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002801 pgoff_t index = from >> PAGE_SHIFT;
2802 unsigned offset = from & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002804 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 unsigned length, pos;
2806 struct inode *inode = mapping->host;
2807 struct page *page;
2808 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809 int err;
2810
2811 blocksize = 1 << inode->i_blkbits;
2812 length = offset & (blocksize - 1);
2813
2814 /* Block boundary? Nothing to do */
2815 if (!length)
2816 return 0;
2817
2818 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002819 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820
2821 page = grab_cache_page(mapping, index);
2822 err = -ENOMEM;
2823 if (!page)
2824 goto out;
2825
2826 if (!page_has_buffers(page))
2827 create_empty_buffers(page, blocksize, 0);
2828
2829 /* Find the buffer that contains "offset" */
2830 bh = page_buffers(page);
2831 pos = blocksize;
2832 while (offset >= pos) {
2833 bh = bh->b_this_page;
2834 iblock++;
2835 pos += blocksize;
2836 }
2837
2838 err = 0;
2839 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002840 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 err = get_block(inode, iblock, bh, 0);
2842 if (err)
2843 goto unlock;
2844 /* unmapped? It's a hole - nothing to do */
2845 if (!buffer_mapped(bh))
2846 goto unlock;
2847 }
2848
2849 /* Ok, it's mapped. Make sure it's up-to-date */
2850 if (PageUptodate(page))
2851 set_buffer_uptodate(bh);
2852
David Chinner33a266d2007-02-12 00:51:41 -08002853 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854 err = -EIO;
Mike Christiedfec8a12016-06-05 14:31:44 -05002855 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856 wait_on_buffer(bh);
2857 /* Uhhuh. Read error. Complain and punt. */
2858 if (!buffer_uptodate(bh))
2859 goto unlock;
2860 }
2861
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002862 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002863 mark_buffer_dirty(bh);
2864 err = 0;
2865
2866unlock:
2867 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002868 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002869out:
2870 return err;
2871}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002872EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002873
2874/*
2875 * The generic ->writepage function for buffer-backed address_spaces
2876 */
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002877int block_write_full_page(struct page *page, get_block_t *get_block,
2878 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002879{
2880 struct inode * const inode = page->mapping->host;
2881 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002882 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002883 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884
2885 /* Is the page fully inside i_size? */
2886 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04002887 return __block_write_full_page(inode, page, get_block, wbc,
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002888 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
2890 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002891 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892 if (page->index >= end_index+1 || !offset) {
2893 /*
2894 * The page may have dirty, unmapped buffers. For example,
2895 * they may have been added in ext3_writepage(). Make them
2896 * freeable here, so the page does not leak.
2897 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002898 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 unlock_page(page);
2900 return 0; /* don't care */
2901 }
2902
2903 /*
2904 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05002905 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906 * in multiples of the page size. For a file that is not a multiple of
2907 * the page size, the remaining memory is zeroed when mapped, and
2908 * writes to that region are not written out to the file."
2909 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002910 zero_user_segment(page, offset, PAGE_SIZE);
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002911 return __block_write_full_page(inode, page, get_block, wbc,
2912 end_buffer_async_write);
Chris Mason35c80d52009-04-15 13:22:38 -04002913}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002914EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04002915
Linus Torvalds1da177e2005-04-16 15:20:36 -07002916sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2917 get_block_t *get_block)
2918{
2919 struct buffer_head tmp;
2920 struct inode *inode = mapping->host;
2921 tmp.b_state = 0;
2922 tmp.b_blocknr = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002923 tmp.b_size = 1 << inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 get_block(inode, block, &tmp, 0);
2925 return tmp.b_blocknr;
2926}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002927EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002929static void end_bio_bh_io_sync(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930{
2931 struct buffer_head *bh = bio->bi_private;
2932
Jens Axboeb7c44ed2015-07-24 12:37:59 -06002933 if (unlikely(bio_flagged(bio, BIO_QUIET)))
Keith Mannthey08bafc02008-11-25 10:24:35 +01002934 set_bit(BH_Quiet, &bh->b_state);
2935
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002936 bh->b_end_io(bh, !bio->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002937 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938}
2939
Linus Torvalds57302e02012-12-04 08:25:11 -08002940/*
2941 * This allows us to do IO even on the odd last sectors
Akinobu Mita59d43912014-10-09 15:26:53 -07002942 * of a device, even if the block size is some multiple
Linus Torvalds57302e02012-12-04 08:25:11 -08002943 * of the physical sector size.
2944 *
2945 * We'll just truncate the bio to the size of the device,
2946 * and clear the end of the buffer head manually.
2947 *
2948 * Truly out-of-range accesses will turn into actual IO
2949 * errors, this only handles the "we need to be able to
2950 * do IO at the final sector" case.
2951 */
Mike Christie2a222ca2016-06-05 14:31:43 -05002952void guard_bio_eod(int op, struct bio *bio)
Linus Torvalds57302e02012-12-04 08:25:11 -08002953{
2954 sector_t maxsector;
Akinobu Mita59d43912014-10-09 15:26:53 -07002955 struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
2956 unsigned truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08002957
2958 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
2959 if (!maxsector)
2960 return;
2961
2962 /*
2963 * If the *whole* IO is past the end of the device,
2964 * let it through, and the IO layer will turn it into
2965 * an EIO.
2966 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07002967 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08002968 return;
2969
Kent Overstreet4f024f32013-10-11 15:44:27 -07002970 maxsector -= bio->bi_iter.bi_sector;
Akinobu Mita59d43912014-10-09 15:26:53 -07002971 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08002972 return;
2973
Akinobu Mita59d43912014-10-09 15:26:53 -07002974 /* Uhhuh. We've got a bio that straddles the device size! */
2975 truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
Linus Torvalds57302e02012-12-04 08:25:11 -08002976
2977 /* Truncate the bio.. */
Akinobu Mita59d43912014-10-09 15:26:53 -07002978 bio->bi_iter.bi_size -= truncated_bytes;
2979 bvec->bv_len -= truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08002980
2981 /* ..and clear the end of the buffer for reads */
Mike Christie2a222ca2016-06-05 14:31:43 -05002982 if (op == REQ_OP_READ) {
Akinobu Mita59d43912014-10-09 15:26:53 -07002983 zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len,
2984 truncated_bytes);
Linus Torvalds57302e02012-12-04 08:25:11 -08002985 }
2986}
2987
Mike Christie2a222ca2016-06-05 14:31:43 -05002988static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Tejun Heob16b1de2015-06-02 08:39:48 -06002989 unsigned long bio_flags, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002990{
2991 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
2993 BUG_ON(!buffer_locked(bh));
2994 BUG_ON(!buffer_mapped(bh));
2995 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04002996 BUG_ON(buffer_delay(bh));
2997 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002998
Jens Axboe48fd4f92008-08-22 10:00:36 +02002999 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02003000 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003002 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003003 clear_buffer_write_io_error(bh);
3004
3005 /*
3006 * from here on down, it's all bio -- do the initial mapping,
3007 * submit_bio -> generic_make_request may further map this bio around
3008 */
3009 bio = bio_alloc(GFP_NOIO, 1);
3010
Tejun Heo2a814902015-05-28 14:50:51 -04003011 if (wbc) {
Tejun Heob16b1de2015-06-02 08:39:48 -06003012 wbc_init_bio(wbc, bio);
Tejun Heo2a814902015-05-28 14:50:51 -04003013 wbc_account_io(wbc, bh->b_page, bh->b_size);
3014 }
Tejun Heobafc0db2015-06-02 08:37:23 -06003015
Kent Overstreet4f024f32013-10-11 15:44:27 -07003016 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 bio->bi_bdev = bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018
Kent Overstreet6cf66b42014-12-22 12:48:42 +01003019 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3020 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021
3022 bio->bi_end_io = end_bio_bh_io_sync;
3023 bio->bi_private = bh;
Darrick J. Wong7136851112013-04-29 15:07:25 -07003024 bio->bi_flags |= bio_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003025
Linus Torvalds57302e02012-12-04 08:25:11 -08003026 /* Take care of bh's that straddle the end of the device */
Mike Christie2a222ca2016-06-05 14:31:43 -05003027 guard_bio_eod(op, bio);
Linus Torvalds57302e02012-12-04 08:25:11 -08003028
Theodore Ts'o877f9622013-04-20 19:58:37 -04003029 if (buffer_meta(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003030 op_flags |= REQ_META;
Theodore Ts'o877f9622013-04-20 19:58:37 -04003031 if (buffer_prio(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003032 op_flags |= REQ_PRIO;
3033 bio_set_op_attrs(bio, op, op_flags);
Theodore Ts'o877f9622013-04-20 19:58:37 -04003034
Mike Christie4e49ea42016-06-05 14:31:41 -05003035 submit_bio(bio);
Julia Lawallf6454b02015-05-26 21:59:53 +02003036 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003037}
Tejun Heobafc0db2015-06-02 08:37:23 -06003038
Mike Christie2a222ca2016-06-05 14:31:43 -05003039int _submit_bh(int op, int op_flags, struct buffer_head *bh,
3040 unsigned long bio_flags)
Tejun Heobafc0db2015-06-02 08:37:23 -06003041{
Mike Christie2a222ca2016-06-05 14:31:43 -05003042 return submit_bh_wbc(op, op_flags, bh, bio_flags, NULL);
Tejun Heobafc0db2015-06-02 08:37:23 -06003043}
Darrick J. Wong7136851112013-04-29 15:07:25 -07003044EXPORT_SYMBOL_GPL(_submit_bh);
3045
Mike Christie2a222ca2016-06-05 14:31:43 -05003046int submit_bh(int op, int op_flags, struct buffer_head *bh)
Darrick J. Wong7136851112013-04-29 15:07:25 -07003047{
Mike Christie2a222ca2016-06-05 14:31:43 -05003048 return submit_bh_wbc(op, op_flags, bh, 0, NULL);
Darrick J. Wong7136851112013-04-29 15:07:25 -07003049}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003050EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003051
3052/**
3053 * ll_rw_block: low-level access to block devices (DEPRECATED)
Mike Christiedfec8a12016-06-05 14:31:44 -05003054 * @op: whether to %READ or %WRITE
Christoph Hellwig70246282016-07-19 11:28:41 +02003055 * @op_flags: rq_flag_bits
Linus Torvalds1da177e2005-04-16 15:20:36 -07003056 * @nr: number of &struct buffer_heads in the array
3057 * @bhs: array of pointers to &struct buffer_head
3058 *
Jan Karaa7662232005-09-06 15:19:10 -07003059 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
Christoph Hellwig70246282016-07-19 11:28:41 +02003060 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3061 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3062 * %REQ_RAHEAD.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003063 *
3064 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003065 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3066 * request, and any buffer that appears to be up-to-date when doing read
3067 * request. Further it marks as clean buffers that are processed for
3068 * writing (the buffer cache won't assume that they are actually clean
3069 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 *
3071 * ll_rw_block sets b_end_io to simple completion handler that marks
Masanari Iidae2278672014-02-18 22:54:36 +09003072 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 * any waiters.
3074 *
3075 * All of the buffers must be for the same device, and must also be a
3076 * multiple of the current approved size for the device.
3077 */
Mike Christiedfec8a12016-06-05 14:31:44 -05003078void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
Linus Torvalds1da177e2005-04-16 15:20:36 -07003079{
3080 int i;
3081
3082 for (i = 0; i < nr; i++) {
3083 struct buffer_head *bh = bhs[i];
3084
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003085 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003086 continue;
Mike Christiedfec8a12016-06-05 14:31:44 -05003087 if (op == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003089 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003090 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003091 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092 continue;
3093 }
3094 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003096 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003097 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003098 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003099 continue;
3100 }
3101 }
3102 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103 }
3104}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003105EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106
Mike Christie2a222ca2016-06-05 14:31:43 -05003107void write_dirty_buffer(struct buffer_head *bh, int op_flags)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003108{
3109 lock_buffer(bh);
3110 if (!test_clear_buffer_dirty(bh)) {
3111 unlock_buffer(bh);
3112 return;
3113 }
3114 bh->b_end_io = end_buffer_write_sync;
3115 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05003116 submit_bh(REQ_OP_WRITE, op_flags, bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003117}
3118EXPORT_SYMBOL(write_dirty_buffer);
3119
Linus Torvalds1da177e2005-04-16 15:20:36 -07003120/*
3121 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3122 * and then start new I/O and then wait upon it. The caller must have a ref on
3123 * the buffer_head.
3124 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003125int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126{
3127 int ret = 0;
3128
3129 WARN_ON(atomic_read(&bh->b_count) < 1);
3130 lock_buffer(bh);
3131 if (test_clear_buffer_dirty(bh)) {
3132 get_bh(bh);
3133 bh->b_end_io = end_buffer_write_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003134 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003135 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003136 if (!ret && !buffer_uptodate(bh))
3137 ret = -EIO;
3138 } else {
3139 unlock_buffer(bh);
3140 }
3141 return ret;
3142}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003143EXPORT_SYMBOL(__sync_dirty_buffer);
3144
3145int sync_dirty_buffer(struct buffer_head *bh)
3146{
3147 return __sync_dirty_buffer(bh, WRITE_SYNC);
3148}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003149EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003150
3151/*
3152 * try_to_free_buffers() checks if all the buffers on this particular page
3153 * are unused, and releases them if so.
3154 *
3155 * Exclusion against try_to_free_buffers may be obtained by either
3156 * locking the page or by holding its mapping's private_lock.
3157 *
3158 * If the page is dirty but all the buffers are clean then we need to
3159 * be sure to mark the page clean as well. This is because the page
3160 * may be against a block device, and a later reattachment of buffers
3161 * to a dirty page will set *all* buffers dirty. Which would corrupt
3162 * filesystem data on the same device.
3163 *
3164 * The same applies to regular filesystem pages: if all the buffers are
3165 * clean then we set the page clean and proceed. To do that, we require
3166 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3167 * private_lock.
3168 *
3169 * try_to_free_buffers() is non-blocking.
3170 */
3171static inline int buffer_busy(struct buffer_head *bh)
3172{
3173 return atomic_read(&bh->b_count) |
3174 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3175}
3176
3177static int
3178drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3179{
3180 struct buffer_head *head = page_buffers(page);
3181 struct buffer_head *bh;
3182
3183 bh = head;
3184 do {
akpm@osdl.orgde7d5a32005-05-01 08:58:39 -07003185 if (buffer_write_io_error(bh) && page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003186 set_bit(AS_EIO, &page->mapping->flags);
3187 if (buffer_busy(bh))
3188 goto failed;
3189 bh = bh->b_this_page;
3190 } while (bh != head);
3191
3192 do {
3193 struct buffer_head *next = bh->b_this_page;
3194
Jan Kara535ee2f2008-02-08 04:21:59 -08003195 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 __remove_assoc_queue(bh);
3197 bh = next;
3198 } while (bh != head);
3199 *buffers_to_free = head;
3200 __clear_page_buffers(page);
3201 return 1;
3202failed:
3203 return 0;
3204}
3205
3206int try_to_free_buffers(struct page *page)
3207{
3208 struct address_space * const mapping = page->mapping;
3209 struct buffer_head *buffers_to_free = NULL;
3210 int ret = 0;
3211
3212 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003213 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214 return 0;
3215
3216 if (mapping == NULL) { /* can this still happen? */
3217 ret = drop_buffers(page, &buffers_to_free);
3218 goto out;
3219 }
3220
3221 spin_lock(&mapping->private_lock);
3222 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003223
3224 /*
3225 * If the filesystem writes its buffers by hand (eg ext3)
3226 * then we can have clean buffers against a dirty page. We
3227 * clean the page here; otherwise the VM will never notice
3228 * that the filesystem did any IO at all.
3229 *
3230 * Also, during truncate, discard_buffer will have marked all
3231 * the page's buffers clean. We discover that here and clean
3232 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003233 *
3234 * private_lock must be held over this entire operation in order
3235 * to synchronise against __set_page_dirty_buffers and prevent the
3236 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003237 */
Tejun Heo11f81be2015-05-22 17:13:15 -04003238 if (ret)
3239 cancel_dirty_page(page);
Nick Piggin87df7242007-01-30 14:36:27 +11003240 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003241out:
3242 if (buffers_to_free) {
3243 struct buffer_head *bh = buffers_to_free;
3244
3245 do {
3246 struct buffer_head *next = bh->b_this_page;
3247 free_buffer_head(bh);
3248 bh = next;
3249 } while (bh != buffers_to_free);
3250 }
3251 return ret;
3252}
3253EXPORT_SYMBOL(try_to_free_buffers);
3254
Linus Torvalds1da177e2005-04-16 15:20:36 -07003255/*
3256 * There are no bdflush tunables left. But distributions are
3257 * still running obsolete flush daemons, so we terminate them here.
3258 *
3259 * Use of bdflush() is deprecated and will be removed in a future kernel.
Jens Axboe5b0830c2009-09-23 19:37:09 +02003260 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003261 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01003262SYSCALL_DEFINE2(bdflush, int, func, long, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263{
3264 static int msg_count;
3265
3266 if (!capable(CAP_SYS_ADMIN))
3267 return -EPERM;
3268
3269 if (msg_count < 5) {
3270 msg_count++;
3271 printk(KERN_INFO
3272 "warning: process `%s' used the obsolete bdflush"
3273 " system call\n", current->comm);
3274 printk(KERN_INFO "Fix your initscripts?\n");
3275 }
3276
3277 if (func == 1)
3278 do_exit(0);
3279 return 0;
3280}
3281
3282/*
3283 * Buffer-head allocation
3284 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003285static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003286
3287/*
3288 * Once the number of bh's in the machine exceeds this level, we start
3289 * stripping them in writeback.
3290 */
Zhang Yanfei43be5942013-02-22 16:35:46 -08003291static unsigned long max_buffer_heads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292
3293int buffer_heads_over_limit;
3294
3295struct bh_accounting {
3296 int nr; /* Number of live bh's */
3297 int ratelimit; /* Limit cacheline bouncing */
3298};
3299
3300static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3301
3302static void recalc_bh_state(void)
3303{
3304 int i;
3305 int tot = 0;
3306
Christoph Lameteree1be862010-12-06 11:40:05 -06003307 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003309 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003310 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003311 tot += per_cpu(bh_accounting, i).nr;
3312 buffer_heads_over_limit = (tot > max_buffer_heads);
3313}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003314
Al Virodd0fc662005-10-07 07:46:04 +01003315struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003316{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003317 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003318 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003319 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003320 preempt_disable();
3321 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003323 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003324 }
3325 return ret;
3326}
3327EXPORT_SYMBOL(alloc_buffer_head);
3328
3329void free_buffer_head(struct buffer_head *bh)
3330{
3331 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3332 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003333 preempt_disable();
3334 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003335 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003336 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337}
3338EXPORT_SYMBOL(free_buffer_head);
3339
Linus Torvalds1da177e2005-04-16 15:20:36 -07003340static void buffer_exit_cpu(int cpu)
3341{
3342 int i;
3343 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3344
3345 for (i = 0; i < BH_LRU_SIZE; i++) {
3346 brelse(b->bhs[i]);
3347 b->bhs[i] = NULL;
3348 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003349 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003350 per_cpu(bh_accounting, cpu).nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003351}
3352
3353static int buffer_cpu_notify(struct notifier_block *self,
3354 unsigned long action, void *hcpu)
3355{
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003356 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 buffer_exit_cpu((unsigned long)hcpu);
3358 return NOTIFY_OK;
3359}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003360
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003361/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003362 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003363 * @bh: struct buffer_head
3364 *
3365 * Return true if the buffer is up-to-date and false,
3366 * with the buffer locked, if not.
3367 */
3368int bh_uptodate_or_lock(struct buffer_head *bh)
3369{
3370 if (!buffer_uptodate(bh)) {
3371 lock_buffer(bh);
3372 if (!buffer_uptodate(bh))
3373 return 0;
3374 unlock_buffer(bh);
3375 }
3376 return 1;
3377}
3378EXPORT_SYMBOL(bh_uptodate_or_lock);
3379
3380/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003381 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003382 * @bh: struct buffer_head
3383 *
3384 * Returns zero on success and -EIO on error.
3385 */
3386int bh_submit_read(struct buffer_head *bh)
3387{
3388 BUG_ON(!buffer_locked(bh));
3389
3390 if (buffer_uptodate(bh)) {
3391 unlock_buffer(bh);
3392 return 0;
3393 }
3394
3395 get_bh(bh);
3396 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003397 submit_bh(REQ_OP_READ, 0, bh);
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003398 wait_on_buffer(bh);
3399 if (buffer_uptodate(bh))
3400 return 0;
3401 return -EIO;
3402}
3403EXPORT_SYMBOL(bh_submit_read);
3404
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405void __init buffer_init(void)
3406{
Zhang Yanfei43be5942013-02-22 16:35:46 -08003407 unsigned long nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003408
Christoph Lameterb98938c2008-02-04 22:28:36 -08003409 bh_cachep = kmem_cache_create("buffer_head",
3410 sizeof(struct buffer_head), 0,
3411 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3412 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003413 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003414
3415 /*
3416 * Limit the bh occupancy to 10% of ZONE_NORMAL
3417 */
3418 nrpages = (nr_free_buffer_pages() * 10) / 100;
3419 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3420 hotcpu_notifier(buffer_cpu_notify, 0);
3421}