blob: 9c8eb9b6db6aace3ef93edc5c8369ba48ecfb4f4 [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>
Christoph Hellwigae259a92016-06-21 09:23:11 +100024#include <linux/iomap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <linux/mm.h>
26#include <linux/percpu.h>
27#include <linux/slab.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080028#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/blkdev.h>
30#include <linux/file.h>
31#include <linux/quotaops.h>
32#include <linux/highmem.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050033#include <linux/export.h>
Tejun Heobafc0db2015-06-02 08:37:23 -060034#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070035#include <linux/writeback.h>
36#include <linux/hash.h>
37#include <linux/suspend.h>
38#include <linux/buffer_head.h>
Andrew Morton55e829a2006-12-10 02:19:27 -080039#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070040#include <linux/bio.h>
41#include <linux/notifier.h>
42#include <linux/cpu.h>
43#include <linux/bitops.h>
44#include <linux/mpage.h>
Ingo Molnarfb1c8f92005-09-10 00:25:56 -070045#include <linux/bit_spinlock.h>
Tejun Heo5305cb82013-01-11 13:06:36 -080046#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070047
48static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
Mike Christie2a222ca2016-06-05 14:31:43 -050049static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Tejun Heob16b1de2015-06-02 08:39:48 -060050 unsigned long bio_flags,
51 struct writeback_control *wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
53#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
54
Yan Honga3f3c292012-12-12 13:52:15 -080055void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -070056{
57 bh->b_end_io = handler;
58 bh->b_private = private;
59}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070060EXPORT_SYMBOL(init_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070061
Tejun Heof0059af2013-01-11 13:06:35 -080062inline void touch_buffer(struct buffer_head *bh)
63{
Tejun Heo5305cb82013-01-11 13:06:36 -080064 trace_block_touch_buffer(bh);
Tejun Heof0059af2013-01-11 13:06:35 -080065 mark_page_accessed(bh->b_page);
66}
67EXPORT_SYMBOL(touch_buffer);
68
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080069void __lock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070070{
NeilBrown74316202014-07-07 15:16:04 +100071 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070072}
73EXPORT_SYMBOL(__lock_buffer);
74
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080075void unlock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070076{
Nick Piggin51b07fc2008-10-18 20:27:00 -070077 clear_bit_unlock(BH_Lock, &bh->b_state);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010078 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -070079 wake_up_bit(&bh->b_state, BH_Lock);
80}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070081EXPORT_SYMBOL(unlock_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070082
83/*
Mel Gormanb4597222013-07-03 15:02:05 -070084 * Returns if the page has dirty or writeback buffers. If all the buffers
85 * are unlocked and clean then the PageDirty information is stale. If
86 * any of the pages are locked, it is assumed they are locked for IO.
87 */
88void buffer_check_dirty_writeback(struct page *page,
89 bool *dirty, bool *writeback)
90{
91 struct buffer_head *head, *bh;
92 *dirty = false;
93 *writeback = false;
94
95 BUG_ON(!PageLocked(page));
96
97 if (!page_has_buffers(page))
98 return;
99
100 if (PageWriteback(page))
101 *writeback = true;
102
103 head = page_buffers(page);
104 bh = head;
105 do {
106 if (buffer_locked(bh))
107 *writeback = true;
108
109 if (buffer_dirty(bh))
110 *dirty = true;
111
112 bh = bh->b_this_page;
113 } while (bh != head);
114}
115EXPORT_SYMBOL(buffer_check_dirty_writeback);
116
117/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700118 * Block until a buffer comes unlocked. This doesn't stop it
119 * from becoming locked again - you have to lock it yourself
120 * if you want to preserve its state.
121 */
122void __wait_on_buffer(struct buffer_head * bh)
123{
NeilBrown74316202014-07-07 15:16:04 +1000124 wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700126EXPORT_SYMBOL(__wait_on_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127
128static void
129__clear_page_buffers(struct page *page)
130{
131 ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700132 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300133 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700134}
135
Robert Elliottb744c2a2014-10-21 13:55:09 -0600136static void buffer_io_error(struct buffer_head *bh, char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137{
Robert Elliott432f16e2014-10-21 13:55:11 -0600138 if (!test_bit(BH_Quiet, &bh->b_state))
139 printk_ratelimited(KERN_ERR
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400140 "Buffer I/O error on dev %pg, logical block %llu%s\n",
141 bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142}
143
144/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700145 * End-of-IO handler helper function which does not touch the bh after
146 * unlocking it.
147 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
148 * a race there is benign: unlock_buffer() only use the bh's address for
149 * hashing after unlocking the buffer, so it doesn't actually touch the bh
150 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700151 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700152static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700153{
154 if (uptodate) {
155 set_buffer_uptodate(bh);
156 } else {
Christoph Hellwig70246282016-07-19 11:28:41 +0200157 /* This happens, due to failed read-ahead attempts. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158 clear_buffer_uptodate(bh);
159 }
160 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700161}
162
163/*
164 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
165 * unlock the buffer. This is what ll_rw_block uses too.
166 */
167void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
168{
169 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170 put_bh(bh);
171}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700172EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700173
174void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
175{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700176 if (uptodate) {
177 set_buffer_uptodate(bh);
178 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600179 buffer_io_error(bh, ", lost sync page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 set_buffer_write_io_error(bh);
181 clear_buffer_uptodate(bh);
182 }
183 unlock_buffer(bh);
184 put_bh(bh);
185}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700186EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187
188/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700189 * Various filesystems appear to want __find_get_block to be non-blocking.
190 * But it's the page lock which protects the buffers. To get around this,
191 * we get exclusion from try_to_free_buffers with the blockdev mapping's
192 * private_lock.
193 *
194 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
195 * may be quite high. This code could TryLock the page, and if that
196 * succeeds, there is no need to take private_lock. (But if
197 * private_lock is contended then so is mapping->tree_lock).
198 */
199static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800200__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201{
202 struct inode *bd_inode = bdev->bd_inode;
203 struct address_space *bd_mapping = bd_inode->i_mapping;
204 struct buffer_head *ret = NULL;
205 pgoff_t index;
206 struct buffer_head *bh;
207 struct buffer_head *head;
208 struct page *page;
209 int all_mapped = 1;
210
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300211 index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman2457aec2014-06-04 16:10:31 -0700212 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700213 if (!page)
214 goto out;
215
216 spin_lock(&bd_mapping->private_lock);
217 if (!page_has_buffers(page))
218 goto out_unlock;
219 head = page_buffers(page);
220 bh = head;
221 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700222 if (!buffer_mapped(bh))
223 all_mapped = 0;
224 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225 ret = bh;
226 get_bh(bh);
227 goto out_unlock;
228 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229 bh = bh->b_this_page;
230 } while (bh != head);
231
232 /* we might be here because some of the buffers on this page are
233 * not mapped. This is due to various races between
234 * file io on the block device and getblk. It gets dealt with
235 * elsewhere, don't buffer_error if we had some unmapped buffers
236 */
237 if (all_mapped) {
238 printk("__find_get_block_slow() failed. "
239 "block=%llu, b_blocknr=%llu\n",
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800240 (unsigned long long)block,
241 (unsigned long long)bh->b_blocknr);
242 printk("b_state=0x%08lx, b_size=%zu\n",
243 bh->b_state, bh->b_size);
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400244 printk("device %pg blocksize: %d\n", bdev,
Tao Ma72a2ebd2011-10-31 17:09:00 -0700245 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 }
247out_unlock:
248 spin_unlock(&bd_mapping->private_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300249 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250out:
251 return ret;
252}
253
Linus Torvalds1da177e2005-04-16 15:20:36 -0700254/*
Jens Axboe5b0830c2009-09-23 19:37:09 +0200255 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700256 */
257static void free_more_memory(void)
258{
Mel Gormanc33d6c02016-05-19 17:14:10 -0700259 struct zoneref *z;
Mel Gorman0e884602008-04-28 02:12:14 -0700260 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600262 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700263 yield();
264
Mel Gorman0e884602008-04-28 02:12:14 -0700265 for_each_online_node(nid) {
Mel Gormanc33d6c02016-05-19 17:14:10 -0700266
267 z = first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
268 gfp_zone(GFP_NOFS), NULL);
269 if (z->zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700270 try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -0700271 GFP_NOFS, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700272 }
273}
274
275/*
276 * I/O completion handler for block_read_full_page() - pages
277 * which come unlocked at the end of I/O.
278 */
279static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
280{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700282 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283 struct buffer_head *tmp;
284 struct page *page;
285 int page_uptodate = 1;
286
287 BUG_ON(!buffer_async_read(bh));
288
289 page = bh->b_page;
290 if (uptodate) {
291 set_buffer_uptodate(bh);
292 } else {
293 clear_buffer_uptodate(bh);
Robert Elliott432f16e2014-10-21 13:55:11 -0600294 buffer_io_error(bh, ", async page read");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 SetPageError(page);
296 }
297
298 /*
299 * Be _very_ careful from here on. Bad things can happen if
300 * two buffer heads end IO at almost the same time and both
301 * decide that the page is now completely done.
302 */
Nick Piggina3972202005-07-07 17:56:56 -0700303 first = page_buffers(page);
304 local_irq_save(flags);
305 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700306 clear_buffer_async_read(bh);
307 unlock_buffer(bh);
308 tmp = bh;
309 do {
310 if (!buffer_uptodate(tmp))
311 page_uptodate = 0;
312 if (buffer_async_read(tmp)) {
313 BUG_ON(!buffer_locked(tmp));
314 goto still_busy;
315 }
316 tmp = tmp->b_this_page;
317 } while (tmp != bh);
Nick Piggina3972202005-07-07 17:56:56 -0700318 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
319 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700320
321 /*
322 * If none of the buffers had errors and they are all
323 * uptodate then we can set the page uptodate.
324 */
325 if (page_uptodate && !PageError(page))
326 SetPageUptodate(page);
327 unlock_page(page);
328 return;
329
330still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700331 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
332 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700333 return;
334}
335
336/*
337 * Completion handler for block_write_full_page() - pages which are unlocked
338 * during I/O, and which have PageWriteback cleared upon I/O completion.
339 */
Chris Mason35c80d52009-04-15 13:22:38 -0400340void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700341{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700343 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 struct buffer_head *tmp;
345 struct page *page;
346
347 BUG_ON(!buffer_async_write(bh));
348
349 page = bh->b_page;
350 if (uptodate) {
351 set_buffer_uptodate(bh);
352 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600353 buffer_io_error(bh, ", lost async page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354 set_bit(AS_EIO, &page->mapping->flags);
Jan Kara58ff4072006-10-17 00:10:19 -0700355 set_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700356 clear_buffer_uptodate(bh);
357 SetPageError(page);
358 }
359
Nick Piggina3972202005-07-07 17:56:56 -0700360 first = page_buffers(page);
361 local_irq_save(flags);
362 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
363
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 clear_buffer_async_write(bh);
365 unlock_buffer(bh);
366 tmp = bh->b_this_page;
367 while (tmp != bh) {
368 if (buffer_async_write(tmp)) {
369 BUG_ON(!buffer_locked(tmp));
370 goto still_busy;
371 }
372 tmp = tmp->b_this_page;
373 }
Nick Piggina3972202005-07-07 17:56:56 -0700374 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
375 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700376 end_page_writeback(page);
377 return;
378
379still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700380 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
381 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700382 return;
383}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700384EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385
386/*
387 * If a page's buffers are under async readin (end_buffer_async_read
388 * completion) then there is a possibility that another thread of
389 * control could lock one of the buffers after it has completed
390 * but while some of the other buffers have not completed. This
391 * locked buffer would confuse end_buffer_async_read() into not unlocking
392 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
393 * that this buffer is not under async I/O.
394 *
395 * The page comes unlocked when it has no locked buffer_async buffers
396 * left.
397 *
398 * PageLocked prevents anyone starting new async I/O reads any of
399 * the buffers.
400 *
401 * PageWriteback is used to prevent simultaneous writeout of the same
402 * page.
403 *
404 * PageLocked prevents anyone from starting writeback of a page which is
405 * under read I/O (PageWriteback is only ever set against a locked page).
406 */
407static void mark_buffer_async_read(struct buffer_head *bh)
408{
409 bh->b_end_io = end_buffer_async_read;
410 set_buffer_async_read(bh);
411}
412
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700413static void mark_buffer_async_write_endio(struct buffer_head *bh,
414 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400415{
416 bh->b_end_io = handler;
417 set_buffer_async_write(bh);
418}
419
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420void mark_buffer_async_write(struct buffer_head *bh)
421{
Chris Mason35c80d52009-04-15 13:22:38 -0400422 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700423}
424EXPORT_SYMBOL(mark_buffer_async_write);
425
426
427/*
428 * fs/buffer.c contains helper functions for buffer-backed address space's
429 * fsync functions. A common requirement for buffer-based filesystems is
430 * that certain data from the backing blockdev needs to be written out for
431 * a successful fsync(). For example, ext2 indirect blocks need to be
432 * written back and waited upon before fsync() returns.
433 *
434 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
435 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
436 * management of a list of dependent buffers at ->i_mapping->private_list.
437 *
438 * Locking is a little subtle: try_to_free_buffers() will remove buffers
439 * from their controlling inode's queue when they are being freed. But
440 * try_to_free_buffers() will be operating against the *blockdev* mapping
441 * at the time, not against the S_ISREG file which depends on those buffers.
442 * So the locking for private_list is via the private_lock in the address_space
443 * which backs the buffers. Which is different from the address_space
444 * against which the buffers are listed. So for a particular address_space,
445 * mapping->private_lock does *not* protect mapping->private_list! In fact,
446 * mapping->private_list will always be protected by the backing blockdev's
447 * ->private_lock.
448 *
449 * Which introduces a requirement: all buffers on an address_space's
450 * ->private_list must be from the same address_space: the blockdev's.
451 *
452 * address_spaces which do not place buffers at ->private_list via these
453 * utility functions are free to use private_lock and private_list for
454 * whatever they want. The only requirement is that list_empty(private_list)
455 * be true at clear_inode() time.
456 *
457 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
458 * filesystems should do that. invalidate_inode_buffers() should just go
459 * BUG_ON(!list_empty).
460 *
461 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
462 * take an address_space, not an inode. And it should be called
463 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
464 * queued up.
465 *
466 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
467 * list if it is already on a list. Because if the buffer is on a list,
468 * it *must* already be on the right one. If not, the filesystem is being
469 * silly. This will save a ton of locking. But first we have to ensure
470 * that buffers are taken *off* the old inode's list when they are freed
471 * (presumably in truncate). That requires careful auditing of all
472 * filesystems (do it inside bforget()). It could also be done by bringing
473 * b_inode back.
474 */
475
476/*
477 * The buffer's backing address_space's private_lock must be held
478 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700479static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700480{
481 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -0700482 WARN_ON(!bh->b_assoc_map);
483 if (buffer_write_io_error(bh))
484 set_bit(AS_EIO, &bh->b_assoc_map->flags);
485 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700486}
487
488int inode_has_buffers(struct inode *inode)
489{
490 return !list_empty(&inode->i_data.private_list);
491}
492
493/*
494 * osync is designed to support O_SYNC io. It waits synchronously for
495 * all already-submitted IO to complete, but does not queue any new
496 * writes to the disk.
497 *
498 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
499 * you dirty the buffers, and then use osync_inode_buffers to wait for
500 * completion. Any other dirty buffers which are not yet queued for
501 * write will not be flushed to disk by the osync.
502 */
503static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
504{
505 struct buffer_head *bh;
506 struct list_head *p;
507 int err = 0;
508
509 spin_lock(lock);
510repeat:
511 list_for_each_prev(p, list) {
512 bh = BH_ENTRY(p);
513 if (buffer_locked(bh)) {
514 get_bh(bh);
515 spin_unlock(lock);
516 wait_on_buffer(bh);
517 if (!buffer_uptodate(bh))
518 err = -EIO;
519 brelse(bh);
520 spin_lock(lock);
521 goto repeat;
522 }
523 }
524 spin_unlock(lock);
525 return err;
526}
527
Al Viro01a05b32010-03-23 06:06:58 -0400528static void do_thaw_one(struct super_block *sb, void *unused)
529{
Al Viro01a05b32010-03-23 06:06:58 -0400530 while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400531 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
Al Viro01a05b32010-03-23 06:06:58 -0400532}
533
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700534static void do_thaw_all(struct work_struct *work)
Eric Sandeenc2d75432009-03-31 15:23:46 -0700535{
Al Viro01a05b32010-03-23 06:06:58 -0400536 iterate_supers(do_thaw_one, NULL);
Jens Axboe053c5252009-04-08 13:44:08 +0200537 kfree(work);
Eric Sandeenc2d75432009-03-31 15:23:46 -0700538 printk(KERN_WARNING "Emergency Thaw complete\n");
539}
540
541/**
542 * emergency_thaw_all -- forcibly thaw every frozen filesystem
543 *
544 * Used for emergency unfreeze of all filesystems via SysRq
545 */
546void emergency_thaw_all(void)
547{
Jens Axboe053c5252009-04-08 13:44:08 +0200548 struct work_struct *work;
549
550 work = kmalloc(sizeof(*work), GFP_ATOMIC);
551 if (work) {
552 INIT_WORK(work, do_thaw_all);
553 schedule_work(work);
554 }
Eric Sandeenc2d75432009-03-31 15:23:46 -0700555}
556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800558 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700559 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 *
561 * Starts I/O against the buffers at mapping->private_list, and waits upon
562 * that I/O.
563 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700564 * Basically, this is a convenience function for fsync().
565 * @mapping is a file or directory which needs those buffers to be written for
566 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 */
568int sync_mapping_buffers(struct address_space *mapping)
569{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800570 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700571
572 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
573 return 0;
574
575 return fsync_buffers_list(&buffer_mapping->private_lock,
576 &mapping->private_list);
577}
578EXPORT_SYMBOL(sync_mapping_buffers);
579
580/*
581 * Called when we've recently written block `bblock', and it is known that
582 * `bblock' was for a buffer_boundary() buffer. This means that the block at
583 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
584 * dirty, schedule it for IO. So that indirects merge nicely with their data.
585 */
586void write_boundary_block(struct block_device *bdev,
587 sector_t bblock, unsigned blocksize)
588{
589 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
590 if (bh) {
591 if (buffer_dirty(bh))
Mike Christiedfec8a12016-06-05 14:31:44 -0500592 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700593 put_bh(bh);
594 }
595}
596
597void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
598{
599 struct address_space *mapping = inode->i_mapping;
600 struct address_space *buffer_mapping = bh->b_page->mapping;
601
602 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800603 if (!mapping->private_data) {
604 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800606 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800608 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700609 spin_lock(&buffer_mapping->private_lock);
610 list_move_tail(&bh->b_assoc_buffers,
611 &mapping->private_list);
Jan Kara58ff4072006-10-17 00:10:19 -0700612 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700613 spin_unlock(&buffer_mapping->private_lock);
614 }
615}
616EXPORT_SYMBOL(mark_buffer_dirty_inode);
617
618/*
Nick Piggin787d2212007-07-17 04:03:34 -0700619 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
620 * dirty.
621 *
622 * If warn is true, then emit a warning if the page is not uptodate and has
623 * not been truncated.
Greg Thelenc4843a72015-05-22 17:13:16 -0400624 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700625 * The caller must hold lock_page_memcg().
Nick Piggin787d2212007-07-17 04:03:34 -0700626 */
Greg Thelenc4843a72015-05-22 17:13:16 -0400627static void __set_page_dirty(struct page *page, struct address_space *mapping,
Johannes Weiner62cccb82016-03-15 14:57:22 -0700628 int warn)
Nick Piggin787d2212007-07-17 04:03:34 -0700629{
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800630 unsigned long flags;
631
632 spin_lock_irqsave(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700633 if (page->mapping) { /* Race with truncate? */
634 WARN_ON_ONCE(warn && !PageUptodate(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700635 account_page_dirtied(page, mapping);
Nick Piggin787d2212007-07-17 04:03:34 -0700636 radix_tree_tag_set(&mapping->page_tree,
637 page_index(page), PAGECACHE_TAG_DIRTY);
638 }
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800639 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700640}
641
642/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 * Add a page to the dirty page list.
644 *
645 * It is a sad fact of life that this function is called from several places
646 * deeply under spinlocking. It may not sleep.
647 *
648 * If the page has buffers, the uptodate buffers are set dirty, to preserve
649 * dirty-state coherency between the page and the buffers. It the page does
650 * not have buffers then when they are later attached they will all be set
651 * dirty.
652 *
653 * The buffers are dirtied before the page is dirtied. There's a small race
654 * window in which a writepage caller may see the page cleanness but not the
655 * buffer dirtiness. That's fine. If this code were to set the page dirty
656 * before the buffers, a concurrent writepage caller could clear the page dirty
657 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
658 * page on the dirty page list.
659 *
660 * We use private_lock to lock against try_to_free_buffers while using the
661 * page's buffer list. Also use this to protect against clean buffers being
662 * added to the page after it was set dirty.
663 *
664 * FIXME: may need to call ->reservepage here as well. That's rather up to the
665 * address_space though.
666 */
667int __set_page_dirty_buffers(struct page *page)
668{
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700669 int newly_dirty;
Nick Piggin787d2212007-07-17 04:03:34 -0700670 struct address_space *mapping = page_mapping(page);
Nick Pigginebf7a222006-10-10 04:36:54 +0200671
672 if (unlikely(!mapping))
673 return !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
675 spin_lock(&mapping->private_lock);
676 if (page_has_buffers(page)) {
677 struct buffer_head *head = page_buffers(page);
678 struct buffer_head *bh = head;
679
680 do {
681 set_buffer_dirty(bh);
682 bh = bh->b_this_page;
683 } while (bh != head);
684 }
Greg Thelenc4843a72015-05-22 17:13:16 -0400685 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700686 * Lock out page->mem_cgroup migration to keep PageDirty
687 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -0400688 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700689 lock_page_memcg(page);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700690 newly_dirty = !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700691 spin_unlock(&mapping->private_lock);
692
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700693 if (newly_dirty)
Johannes Weiner62cccb82016-03-15 14:57:22 -0700694 __set_page_dirty(page, mapping, 1);
Greg Thelenc4843a72015-05-22 17:13:16 -0400695
Johannes Weiner62cccb82016-03-15 14:57:22 -0700696 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400697
698 if (newly_dirty)
699 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
700
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700701 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700702}
703EXPORT_SYMBOL(__set_page_dirty_buffers);
704
705/*
706 * Write out and wait upon a list of buffers.
707 *
708 * We have conflicting pressures: we want to make sure that all
709 * initially dirty buffers get waited on, but that any subsequently
710 * dirtied buffers don't. After all, we don't want fsync to last
711 * forever if somebody is actively writing to the file.
712 *
713 * Do this in two main stages: first we copy dirty buffers to a
714 * temporary inode list, queueing the writes as we go. Then we clean
715 * up, waiting for those writes to complete.
716 *
717 * During this second stage, any subsequent updates to the file may end
718 * up refiling the buffer on the original inode's dirty list again, so
719 * there is a chance we will end up with a buffer queued for write but
720 * not yet completed on that list. So, as a final cleanup we go through
721 * the osync code to catch these locked, dirty buffers without requeuing
722 * any newly dirty buffers for write.
723 */
724static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
725{
726 struct buffer_head *bh;
727 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100728 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100730 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731
732 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100733 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734
735 spin_lock(lock);
736 while (!list_empty(list)) {
737 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800738 mapping = bh->b_assoc_map;
Jan Kara58ff4072006-10-17 00:10:19 -0700739 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800740 /* Avoid race with mark_buffer_dirty_inode() which does
741 * a lockless check and we rely on seeing the dirty bit */
742 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700743 if (buffer_dirty(bh) || buffer_locked(bh)) {
744 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800745 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 if (buffer_dirty(bh)) {
747 get_bh(bh);
748 spin_unlock(lock);
749 /*
750 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200751 * write_dirty_buffer() actually writes the
752 * current contents - it is a noop if I/O is
753 * still in flight on potentially older
754 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 */
Jens Axboe721a9602011-03-09 11:56:30 +0100756 write_dirty_buffer(bh, WRITE_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200757
758 /*
759 * Kick off IO for the previous mapping. Note
760 * that we will not run the very last mapping,
761 * wait_on_buffer() will do that for us
762 * through sync_buffer().
763 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764 brelse(bh);
765 spin_lock(lock);
766 }
767 }
768 }
769
Jens Axboe4ee24912011-03-17 10:51:40 +0100770 spin_unlock(lock);
771 blk_finish_plug(&plug);
772 spin_lock(lock);
773
Linus Torvalds1da177e2005-04-16 15:20:36 -0700774 while (!list_empty(&tmp)) {
775 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800777 mapping = bh->b_assoc_map;
778 __remove_assoc_queue(bh);
779 /* Avoid race with mark_buffer_dirty_inode() which does
780 * a lockless check and we rely on seeing the dirty bit */
781 smp_mb();
782 if (buffer_dirty(bh)) {
783 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800784 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800785 bh->b_assoc_map = mapping;
786 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 spin_unlock(lock);
788 wait_on_buffer(bh);
789 if (!buffer_uptodate(bh))
790 err = -EIO;
791 brelse(bh);
792 spin_lock(lock);
793 }
794
795 spin_unlock(lock);
796 err2 = osync_buffers_list(lock, list);
797 if (err)
798 return err;
799 else
800 return err2;
801}
802
803/*
804 * Invalidate any and all dirty buffers on a given inode. We are
805 * probably unmounting the fs, but that doesn't mean we have already
806 * done a sync(). Just drop the buffers from the inode list.
807 *
808 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
809 * assumes that all the buffers are against the blockdev. Not true
810 * for reiserfs.
811 */
812void invalidate_inode_buffers(struct inode *inode)
813{
814 if (inode_has_buffers(inode)) {
815 struct address_space *mapping = &inode->i_data;
816 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800817 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
819 spin_lock(&buffer_mapping->private_lock);
820 while (!list_empty(list))
821 __remove_assoc_queue(BH_ENTRY(list->next));
822 spin_unlock(&buffer_mapping->private_lock);
823 }
824}
Jan Kara52b19ac2008-09-23 18:24:08 +0200825EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826
827/*
828 * Remove any clean buffers from the inode's buffer list. This is called
829 * when we're trying to free the inode itself. Those buffers can pin it.
830 *
831 * Returns true if all buffers were removed.
832 */
833int remove_inode_buffers(struct inode *inode)
834{
835 int ret = 1;
836
837 if (inode_has_buffers(inode)) {
838 struct address_space *mapping = &inode->i_data;
839 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800840 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841
842 spin_lock(&buffer_mapping->private_lock);
843 while (!list_empty(list)) {
844 struct buffer_head *bh = BH_ENTRY(list->next);
845 if (buffer_dirty(bh)) {
846 ret = 0;
847 break;
848 }
849 __remove_assoc_queue(bh);
850 }
851 spin_unlock(&buffer_mapping->private_lock);
852 }
853 return ret;
854}
855
856/*
857 * Create the appropriate buffers when given a page for data area and
858 * the size of each buffer.. Use the bh->b_this_page linked list to
859 * follow the buffers created. Return NULL if unable to create more
860 * buffers.
861 *
862 * The retry flag is used to differentiate async IO (paging, swapping)
863 * which may not fail from ordinary buffer allocations.
864 */
865struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
866 int retry)
867{
868 struct buffer_head *bh, *head;
869 long offset;
870
871try_again:
872 head = NULL;
873 offset = PAGE_SIZE;
874 while ((offset -= size) >= 0) {
875 bh = alloc_buffer_head(GFP_NOFS);
876 if (!bh)
877 goto no_grow;
878
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879 bh->b_this_page = head;
880 bh->b_blocknr = -1;
881 head = bh;
882
Linus Torvalds1da177e2005-04-16 15:20:36 -0700883 bh->b_size = size;
884
885 /* Link the buffer to its page */
886 set_bh_page(bh, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887 }
888 return head;
889/*
890 * In case anything failed, we just free everything we got.
891 */
892no_grow:
893 if (head) {
894 do {
895 bh = head;
896 head = head->b_this_page;
897 free_buffer_head(bh);
898 } while (head);
899 }
900
901 /*
902 * Return failure for non-async IO requests. Async IO requests
903 * are not allowed to fail, so we have to wait until buffer heads
904 * become available. But we don't want tasks sleeping with
905 * partially complete buffers, so all were released above.
906 */
907 if (!retry)
908 return NULL;
909
910 /* We're _really_ low on memory. Now we just
911 * wait for old buffer heads to become free due to
912 * finishing IO. Since this is an async request and
913 * the reserve list is empty, we're sure there are
914 * async buffer heads in use.
915 */
916 free_more_memory();
917 goto try_again;
918}
919EXPORT_SYMBOL_GPL(alloc_page_buffers);
920
921static inline void
922link_dev_buffers(struct page *page, struct buffer_head *head)
923{
924 struct buffer_head *bh, *tail;
925
926 bh = head;
927 do {
928 tail = bh;
929 bh = bh->b_this_page;
930 } while (bh);
931 tail->b_this_page = head;
932 attach_page_buffers(page, head);
933}
934
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800935static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
936{
937 sector_t retval = ~((sector_t)0);
938 loff_t sz = i_size_read(bdev->bd_inode);
939
940 if (sz) {
941 unsigned int sizebits = blksize_bits(size);
942 retval = (sz >> sizebits);
943 }
944 return retval;
945}
946
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947/*
948 * Initialise the state of a blockdev page's buffers.
949 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200950static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700951init_page_buffers(struct page *page, struct block_device *bdev,
952 sector_t block, int size)
953{
954 struct buffer_head *head = page_buffers(page);
955 struct buffer_head *bh = head;
956 int uptodate = PageUptodate(page);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800957 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700958
959 do {
960 if (!buffer_mapped(bh)) {
961 init_buffer(bh, NULL, NULL);
962 bh->b_bdev = bdev;
963 bh->b_blocknr = block;
964 if (uptodate)
965 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200966 if (block < end_block)
967 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968 }
969 block++;
970 bh = bh->b_this_page;
971 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200972
973 /*
974 * Caller needs to validate requested block against end of device.
975 */
976 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700977}
978
979/*
980 * Create the page-cache page that contains the requested block.
981 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200982 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200984static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700985grow_dev_page(struct block_device *bdev, sector_t block,
Gioh Kim3b5e6452014-09-04 22:04:42 -0400986 pgoff_t index, int size, int sizebits, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987{
988 struct inode *inode = bdev->bd_inode;
989 struct page *page;
990 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200991 sector_t end_block;
992 int ret = 0; /* Will call free_more_memory() */
Johannes Weiner84235de2013-10-16 13:47:00 -0700993 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
Michal Hockoc62d2552015-11-06 16:28:49 -0800995 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
Gioh Kim3b5e6452014-09-04 22:04:42 -0400996
Johannes Weiner84235de2013-10-16 13:47:00 -0700997 /*
998 * XXX: __getblk_slow() can not really deal with failure and
999 * will endlessly loop on improvised global reclaim. Prefer
1000 * looping in the allocator rather than here, at least that
1001 * code knows what it's doing.
1002 */
1003 gfp_mask |= __GFP_NOFAIL;
1004
1005 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001006 if (!page)
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001007 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001008
Eric Sesterhenne827f922006-03-26 18:24:46 +02001009 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010
1011 if (page_has_buffers(page)) {
1012 bh = page_buffers(page);
1013 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001014 end_block = init_page_buffers(page, bdev,
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001015 (sector_t)index << sizebits,
1016 size);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001017 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001018 }
1019 if (!try_to_free_buffers(page))
1020 goto failed;
1021 }
1022
1023 /*
1024 * Allocate some buffers for this page
1025 */
1026 bh = alloc_page_buffers(page, size, 0);
1027 if (!bh)
1028 goto failed;
1029
1030 /*
1031 * Link the page to the buffers and initialise them. Take the
1032 * lock to be atomic wrt __find_get_block(), which does not
1033 * run under the page lock.
1034 */
1035 spin_lock(&inode->i_mapping->private_lock);
1036 link_dev_buffers(page, bh);
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001037 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1038 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001039 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001040done:
1041 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001044 put_page(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001045 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001046}
1047
1048/*
1049 * Create buffers for the specified block device block's page. If
1050 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001051 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001052static int
Gioh Kim3b5e6452014-09-04 22:04:42 -04001053grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001054{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 pgoff_t index;
1056 int sizebits;
1057
1058 sizebits = -1;
1059 do {
1060 sizebits++;
1061 } while ((size << sizebits) < PAGE_SIZE);
1062
1063 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001064
Andrew Mortone5657932006-10-11 01:21:46 -07001065 /*
1066 * Check for a block which wants to lie outside our maximum possible
1067 * pagecache index. (this comparison is done using sector_t types).
1068 */
1069 if (unlikely(index != block >> sizebits)) {
Andrew Mortone5657932006-10-11 01:21:46 -07001070 printk(KERN_ERR "%s: requested out-of-range block %llu for "
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001071 "device %pg\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001072 __func__, (unsigned long long)block,
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001073 bdev);
Andrew Mortone5657932006-10-11 01:21:46 -07001074 return -EIO;
1075 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001076
Linus Torvalds1da177e2005-04-16 15:20:36 -07001077 /* Create a page with the proper size buffers.. */
Gioh Kim3b5e6452014-09-04 22:04:42 -04001078 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079}
1080
Gioh Kim3b5e6452014-09-04 22:04:42 -04001081struct buffer_head *
1082__getblk_slow(struct block_device *bdev, sector_t block,
1083 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084{
1085 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001086 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087 (size < 512 || size > PAGE_SIZE))) {
1088 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1089 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001090 printk(KERN_ERR "logical block size: %d\n",
1091 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001092
1093 dump_stack();
1094 return NULL;
1095 }
1096
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001097 for (;;) {
1098 struct buffer_head *bh;
1099 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100
1101 bh = __find_get_block(bdev, block, size);
1102 if (bh)
1103 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001104
Gioh Kim3b5e6452014-09-04 22:04:42 -04001105 ret = grow_buffers(bdev, block, size, gfp);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001106 if (ret < 0)
1107 return NULL;
1108 if (ret == 0)
1109 free_more_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001110 }
1111}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001112EXPORT_SYMBOL(__getblk_slow);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
1114/*
1115 * The relationship between dirty buffers and dirty pages:
1116 *
1117 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
1118 * the page is tagged dirty in its radix tree.
1119 *
1120 * At all times, the dirtiness of the buffers represents the dirtiness of
1121 * subsections of the page. If the page has buffers, the page dirty bit is
1122 * merely a hint about the true dirty state.
1123 *
1124 * When a page is set dirty in its entirety, all its buffers are marked dirty
1125 * (if the page has buffers).
1126 *
1127 * When a buffer is marked dirty, its page is dirtied, but the page's other
1128 * buffers are not.
1129 *
1130 * Also. When blockdev buffers are explicitly read with bread(), they
1131 * individually become uptodate. But their backing page remains not
1132 * uptodate - even if all of its buffers are uptodate. A subsequent
1133 * block_read_full_page() against that page will discover all the uptodate
1134 * buffers, will set the page uptodate and will perform no I/O.
1135 */
1136
1137/**
1138 * mark_buffer_dirty - mark a buffer_head as needing writeout
Martin Waitz67be2dd2005-05-01 08:59:26 -07001139 * @bh: the buffer_head to mark dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001140 *
1141 * mark_buffer_dirty() will set the dirty bit against the buffer, then set its
1142 * backing page dirty, then tag the page as dirty in its address_space's radix
1143 * tree and then attach the address_space's inode to its superblock's dirty
1144 * inode list.
1145 *
1146 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
Dave Chinner250df6e2011-03-22 22:23:36 +11001147 * mapping->tree_lock and mapping->host->i_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001149void mark_buffer_dirty(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001150{
Nick Piggin787d2212007-07-17 04:03:34 -07001151 WARN_ON_ONCE(!buffer_uptodate(bh));
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001152
Tejun Heo5305cb82013-01-11 13:06:36 -08001153 trace_block_dirty_buffer(bh);
1154
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001155 /*
1156 * Very *carefully* optimize the it-is-already-dirty case.
1157 *
1158 * Don't let the final "is it dirty" escape to before we
1159 * perhaps modified the buffer.
1160 */
1161 if (buffer_dirty(bh)) {
1162 smp_mb();
1163 if (buffer_dirty(bh))
1164 return;
1165 }
1166
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001167 if (!test_set_buffer_dirty(bh)) {
1168 struct page *page = bh->b_page;
Greg Thelenc4843a72015-05-22 17:13:16 -04001169 struct address_space *mapping = NULL;
Greg Thelenc4843a72015-05-22 17:13:16 -04001170
Johannes Weiner62cccb82016-03-15 14:57:22 -07001171 lock_page_memcg(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001172 if (!TestSetPageDirty(page)) {
Greg Thelenc4843a72015-05-22 17:13:16 -04001173 mapping = page_mapping(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001174 if (mapping)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001175 __set_page_dirty(page, mapping, 0);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001176 }
Johannes Weiner62cccb82016-03-15 14:57:22 -07001177 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04001178 if (mapping)
1179 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001180 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001181}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001182EXPORT_SYMBOL(mark_buffer_dirty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001183
1184/*
1185 * Decrement a buffer_head's reference count. If all buffers against a page
1186 * have zero reference count, are clean and unlocked, and if the page is clean
1187 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1188 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1189 * a page but it ends up not being freed, and buffers may later be reattached).
1190 */
1191void __brelse(struct buffer_head * buf)
1192{
1193 if (atomic_read(&buf->b_count)) {
1194 put_bh(buf);
1195 return;
1196 }
Arjan van de Ven5c752ad2008-07-25 19:45:40 -07001197 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001199EXPORT_SYMBOL(__brelse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200
1201/*
1202 * bforget() is like brelse(), except it discards any
1203 * potentially dirty data.
1204 */
1205void __bforget(struct buffer_head *bh)
1206{
1207 clear_buffer_dirty(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -08001208 if (bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 struct address_space *buffer_mapping = bh->b_page->mapping;
1210
1211 spin_lock(&buffer_mapping->private_lock);
1212 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -07001213 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001214 spin_unlock(&buffer_mapping->private_lock);
1215 }
1216 __brelse(bh);
1217}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001218EXPORT_SYMBOL(__bforget);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219
1220static struct buffer_head *__bread_slow(struct buffer_head *bh)
1221{
1222 lock_buffer(bh);
1223 if (buffer_uptodate(bh)) {
1224 unlock_buffer(bh);
1225 return bh;
1226 } else {
1227 get_bh(bh);
1228 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05001229 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 wait_on_buffer(bh);
1231 if (buffer_uptodate(bh))
1232 return bh;
1233 }
1234 brelse(bh);
1235 return NULL;
1236}
1237
1238/*
1239 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1240 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1241 * refcount elevated by one when they're in an LRU. A buffer can only appear
1242 * once in a particular CPU's LRU. A single buffer can be present in multiple
1243 * CPU's LRUs at the same time.
1244 *
1245 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1246 * sb_find_get_block().
1247 *
1248 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1249 * a local interrupt disable for that.
1250 */
1251
Sebastien Buisson86cf78d2014-10-09 15:29:38 -07001252#define BH_LRU_SIZE 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253
1254struct bh_lru {
1255 struct buffer_head *bhs[BH_LRU_SIZE];
1256};
1257
1258static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1259
1260#ifdef CONFIG_SMP
1261#define bh_lru_lock() local_irq_disable()
1262#define bh_lru_unlock() local_irq_enable()
1263#else
1264#define bh_lru_lock() preempt_disable()
1265#define bh_lru_unlock() preempt_enable()
1266#endif
1267
1268static inline void check_irqs_on(void)
1269{
1270#ifdef irqs_disabled
1271 BUG_ON(irqs_disabled());
1272#endif
1273}
1274
1275/*
1276 * The LRU management algorithm is dopey-but-simple. Sorry.
1277 */
1278static void bh_lru_install(struct buffer_head *bh)
1279{
1280 struct buffer_head *evictee = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001281
1282 check_irqs_on();
1283 bh_lru_lock();
Christoph Lameterc7b92512010-12-06 11:16:28 -06001284 if (__this_cpu_read(bh_lrus.bhs[0]) != bh) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 struct buffer_head *bhs[BH_LRU_SIZE];
1286 int in;
1287 int out = 0;
1288
1289 get_bh(bh);
1290 bhs[out++] = bh;
1291 for (in = 0; in < BH_LRU_SIZE; in++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001292 struct buffer_head *bh2 =
1293 __this_cpu_read(bh_lrus.bhs[in]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001294
1295 if (bh2 == bh) {
1296 __brelse(bh2);
1297 } else {
1298 if (out >= BH_LRU_SIZE) {
1299 BUG_ON(evictee != NULL);
1300 evictee = bh2;
1301 } else {
1302 bhs[out++] = bh2;
1303 }
1304 }
1305 }
1306 while (out < BH_LRU_SIZE)
1307 bhs[out++] = NULL;
Christoph Lameterca6673b02013-12-03 17:32:53 -06001308 memcpy(this_cpu_ptr(&bh_lrus.bhs), bhs, sizeof(bhs));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 }
1310 bh_lru_unlock();
1311
1312 if (evictee)
1313 __brelse(evictee);
1314}
1315
1316/*
1317 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1318 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001319static struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001320lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321{
1322 struct buffer_head *ret = NULL;
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001323 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324
1325 check_irqs_on();
1326 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001327 for (i = 0; i < BH_LRU_SIZE; i++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001328 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001329
Zach Brown9470dd52014-10-13 15:55:05 -07001330 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1331 bh->b_size == size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332 if (i) {
1333 while (i) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001334 __this_cpu_write(bh_lrus.bhs[i],
1335 __this_cpu_read(bh_lrus.bhs[i - 1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336 i--;
1337 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06001338 __this_cpu_write(bh_lrus.bhs[0], bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 }
1340 get_bh(bh);
1341 ret = bh;
1342 break;
1343 }
1344 }
1345 bh_lru_unlock();
1346 return ret;
1347}
1348
1349/*
1350 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1351 * it in the LRU and mark it as accessed. If it is not present then return
1352 * NULL
1353 */
1354struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001355__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356{
1357 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1358
1359 if (bh == NULL) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001360 /* __find_get_block_slow will mark the page accessed */
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001361 bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 if (bh)
1363 bh_lru_install(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001364 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 touch_buffer(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001366
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 return bh;
1368}
1369EXPORT_SYMBOL(__find_get_block);
1370
1371/*
Gioh Kim3b5e6452014-09-04 22:04:42 -04001372 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001373 * which corresponds to the passed block_device, block and size. The
1374 * returned buffer has its reference count incremented.
1375 *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001376 * __getblk_gfp() will lock up the machine if grow_dev_page's
1377 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378 */
1379struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001380__getblk_gfp(struct block_device *bdev, sector_t block,
1381 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001382{
1383 struct buffer_head *bh = __find_get_block(bdev, block, size);
1384
1385 might_sleep();
1386 if (bh == NULL)
Gioh Kim3b5e6452014-09-04 22:04:42 -04001387 bh = __getblk_slow(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 return bh;
1389}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001390EXPORT_SYMBOL(__getblk_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391
1392/*
1393 * Do async read-ahead on a buffer..
1394 */
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001395void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396{
1397 struct buffer_head *bh = __getblk(bdev, block, size);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001398 if (likely(bh)) {
Christoph Hellwig70246282016-07-19 11:28:41 +02001399 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001400 brelse(bh);
1401 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001402}
1403EXPORT_SYMBOL(__breadahead);
1404
1405/**
Gioh Kim3b5e6452014-09-04 22:04:42 -04001406 * __bread_gfp() - reads a specified block and returns the bh
Martin Waitz67be2dd2005-05-01 08:59:26 -07001407 * @bdev: the block_device to read from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 * @block: number of block
1409 * @size: size (in bytes) to read
Gioh Kim3b5e6452014-09-04 22:04:42 -04001410 * @gfp: page allocation flag
1411 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 * Reads a specified block, and returns buffer head that contains it.
Gioh Kim3b5e6452014-09-04 22:04:42 -04001413 * The page cache can be allocated from non-movable area
1414 * not to prevent page migration if you set gfp to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 * It returns NULL if the block was unreadable.
1416 */
1417struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001418__bread_gfp(struct block_device *bdev, sector_t block,
1419 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420{
Gioh Kim3b5e6452014-09-04 22:04:42 -04001421 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Andrew Mortona3e713b2005-10-30 15:03:15 -08001423 if (likely(bh) && !buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 bh = __bread_slow(bh);
1425 return bh;
1426}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001427EXPORT_SYMBOL(__bread_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
1429/*
1430 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1431 * This doesn't race because it runs in each cpu either in irq
1432 * or with preempt disabled.
1433 */
1434static void invalidate_bh_lru(void *arg)
1435{
1436 struct bh_lru *b = &get_cpu_var(bh_lrus);
1437 int i;
1438
1439 for (i = 0; i < BH_LRU_SIZE; i++) {
1440 brelse(b->bhs[i]);
1441 b->bhs[i] = NULL;
1442 }
1443 put_cpu_var(bh_lrus);
1444}
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001445
1446static bool has_bh_in_lru(int cpu, void *dummy)
1447{
1448 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1449 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001450
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001451 for (i = 0; i < BH_LRU_SIZE; i++) {
1452 if (b->bhs[i])
1453 return 1;
1454 }
1455
1456 return 0;
1457}
1458
Peter Zijlstraf9a14392007-05-06 14:49:55 -07001459void invalidate_bh_lrus(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001460{
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001461 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1, GFP_KERNEL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001462}
Nick Piggin9db55792008-02-08 04:19:49 -08001463EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464
1465void set_bh_page(struct buffer_head *bh,
1466 struct page *page, unsigned long offset)
1467{
1468 bh->b_page = page;
Eric Sesterhenne827f922006-03-26 18:24:46 +02001469 BUG_ON(offset >= PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470 if (PageHighMem(page))
1471 /*
1472 * This catches illegal uses and preserves the offset:
1473 */
1474 bh->b_data = (char *)(0 + offset);
1475 else
1476 bh->b_data = page_address(page) + offset;
1477}
1478EXPORT_SYMBOL(set_bh_page);
1479
1480/*
1481 * Called when truncating a buffer on a page completely.
1482 */
Mel Gormane7470ee2014-06-04 16:10:29 -07001483
1484/* Bits that are cleared during an invalidate */
1485#define BUFFER_FLAGS_DISCARD \
1486 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1487 1 << BH_Delay | 1 << BH_Unwritten)
1488
Arjan van de Ven858119e2006-01-14 13:20:43 -08001489static void discard_buffer(struct buffer_head * bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001490{
Mel Gormane7470ee2014-06-04 16:10:29 -07001491 unsigned long b_state, b_state_old;
1492
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 lock_buffer(bh);
1494 clear_buffer_dirty(bh);
1495 bh->b_bdev = NULL;
Mel Gormane7470ee2014-06-04 16:10:29 -07001496 b_state = bh->b_state;
1497 for (;;) {
1498 b_state_old = cmpxchg(&bh->b_state, b_state,
1499 (b_state & ~BUFFER_FLAGS_DISCARD));
1500 if (b_state_old == b_state)
1501 break;
1502 b_state = b_state_old;
1503 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001504 unlock_buffer(bh);
1505}
1506
1507/**
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001508 * block_invalidatepage - invalidate part or all of a buffer-backed page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509 *
1510 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -04001511 * @offset: start of the range to invalidate
1512 * @length: length of the range to invalidate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001513 *
1514 * block_invalidatepage() is called when all or part of the page has become
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001515 * invalidated by a truncate operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001516 *
1517 * block_invalidatepage() does not have to release all buffers, but it must
1518 * ensure that no dirty buffer is left outside @offset and that no I/O
1519 * is underway against any of the blocks which are outside the truncation
1520 * point. Because the caller is about to free (and possibly reuse) those
1521 * blocks on-disk.
1522 */
Lukas Czernerd47992f2013-05-21 23:17:23 -04001523void block_invalidatepage(struct page *page, unsigned int offset,
1524 unsigned int length)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001525{
1526 struct buffer_head *head, *bh, *next;
1527 unsigned int curr_off = 0;
Lukas Czernerd47992f2013-05-21 23:17:23 -04001528 unsigned int stop = length + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001529
1530 BUG_ON(!PageLocked(page));
1531 if (!page_has_buffers(page))
1532 goto out;
1533
Lukas Czernerd47992f2013-05-21 23:17:23 -04001534 /*
1535 * Check for overflow
1536 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001537 BUG_ON(stop > PAGE_SIZE || stop < length);
Lukas Czernerd47992f2013-05-21 23:17:23 -04001538
Linus Torvalds1da177e2005-04-16 15:20:36 -07001539 head = page_buffers(page);
1540 bh = head;
1541 do {
1542 unsigned int next_off = curr_off + bh->b_size;
1543 next = bh->b_this_page;
1544
1545 /*
Lukas Czernerd47992f2013-05-21 23:17:23 -04001546 * Are we still fully in range ?
1547 */
1548 if (next_off > stop)
1549 goto out;
1550
1551 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 * is this block fully invalidated?
1553 */
1554 if (offset <= curr_off)
1555 discard_buffer(bh);
1556 curr_off = next_off;
1557 bh = next;
1558 } while (bh != head);
1559
1560 /*
1561 * We release buffers only if the entire page is being invalidated.
1562 * The get_block cached value has been unconditionally invalidated,
1563 * so real IO is not possible anymore.
1564 */
1565 if (offset == 0)
NeilBrown2ff28e22006-03-26 01:37:18 -08001566 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567out:
NeilBrown2ff28e22006-03-26 01:37:18 -08001568 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570EXPORT_SYMBOL(block_invalidatepage);
1571
Lukas Czernerd47992f2013-05-21 23:17:23 -04001572
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573/*
1574 * We attach and possibly dirty the buffers atomically wrt
1575 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1576 * is already excluded via the page lock.
1577 */
1578void create_empty_buffers(struct page *page,
1579 unsigned long blocksize, unsigned long b_state)
1580{
1581 struct buffer_head *bh, *head, *tail;
1582
1583 head = alloc_page_buffers(page, blocksize, 1);
1584 bh = head;
1585 do {
1586 bh->b_state |= b_state;
1587 tail = bh;
1588 bh = bh->b_this_page;
1589 } while (bh);
1590 tail->b_this_page = head;
1591
1592 spin_lock(&page->mapping->private_lock);
1593 if (PageUptodate(page) || PageDirty(page)) {
1594 bh = head;
1595 do {
1596 if (PageDirty(page))
1597 set_buffer_dirty(bh);
1598 if (PageUptodate(page))
1599 set_buffer_uptodate(bh);
1600 bh = bh->b_this_page;
1601 } while (bh != head);
1602 }
1603 attach_page_buffers(page, head);
1604 spin_unlock(&page->mapping->private_lock);
1605}
1606EXPORT_SYMBOL(create_empty_buffers);
1607
1608/*
1609 * We are taking a block for data and we don't want any output from any
1610 * buffer-cache aliases starting from return from that function and
1611 * until the moment when something will explicitly mark the buffer
1612 * dirty (hopefully that will not happen until we will free that block ;-)
1613 * We don't even need to mark it not-uptodate - nobody can expect
1614 * anything from a newly allocated buffer anyway. We used to used
1615 * unmap_buffer() for such invalidation, but that was wrong. We definitely
1616 * don't want to mark the alias unmapped, for example - it would confuse
1617 * anyone who might pick it with bread() afterwards...
1618 *
1619 * Also.. Note that bforget() doesn't lock the buffer. So there can
1620 * be writeout I/O going on against recently-freed buffers. We don't
1621 * wait on that I/O in bforget() - it's more efficient to wait on the I/O
1622 * only if we really need to. That happens here.
1623 */
1624void unmap_underlying_metadata(struct block_device *bdev, sector_t block)
1625{
1626 struct buffer_head *old_bh;
1627
1628 might_sleep();
1629
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001630 old_bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 if (old_bh) {
1632 clear_buffer_dirty(old_bh);
1633 wait_on_buffer(old_bh);
1634 clear_buffer_req(old_bh);
1635 __brelse(old_bh);
1636 }
1637}
1638EXPORT_SYMBOL(unmap_underlying_metadata);
1639
1640/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001641 * Size is a power-of-two in the range 512..PAGE_SIZE,
1642 * and the case we care about most is PAGE_SIZE.
1643 *
1644 * So this *could* possibly be written with those
1645 * constraints in mind (relevant mostly if some
1646 * architecture has a slow bit-scan instruction)
1647 */
1648static inline int block_size_bits(unsigned int blocksize)
1649{
1650 return ilog2(blocksize);
1651}
1652
1653static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1654{
1655 BUG_ON(!PageLocked(page));
1656
1657 if (!page_has_buffers(page))
1658 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1659 return page_buffers(page);
1660}
1661
1662/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 * NOTE! All mapped/uptodate combinations are valid:
1664 *
1665 * Mapped Uptodate Meaning
1666 *
1667 * No No "unknown" - must do get_block()
1668 * No Yes "hole" - zero-filled
1669 * Yes No "allocated" - allocated on disk, not read in
1670 * Yes Yes "valid" - allocated and up-to-date in memory.
1671 *
1672 * "Dirty" is valid only with the last case (mapped+uptodate).
1673 */
1674
1675/*
1676 * While block_write_full_page is writing back the dirty buffers under
1677 * the page lock, whoever dirtied the buffers may decide to clean them
1678 * again at any time. We handle that by only looking at the buffer
1679 * state inside lock_buffer().
1680 *
1681 * If block_write_full_page() is called for regular writeback
1682 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1683 * locked buffer. This only can happen if someone has written the buffer
1684 * directly, with submit_bh(). At the address_space level PageWriteback
1685 * prevents this contention from occurring.
Theodore Ts'o6e34eed2009-04-07 18:12:43 -04001686 *
1687 * If block_write_full_page() is called with wbc->sync_mode ==
Jens Axboe721a9602011-03-09 11:56:30 +01001688 * WB_SYNC_ALL, the writes are posted using WRITE_SYNC; this
1689 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690 */
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001691int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001692 get_block_t *get_block, struct writeback_control *wbc,
1693 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001694{
1695 int err;
1696 sector_t block;
1697 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001698 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001699 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700 int nr_underway = 0;
Mike Christie2a222ca2016-06-05 14:31:43 -05001701 int write_flags = (wbc->sync_mode == WB_SYNC_ALL ? WRITE_SYNC : 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001703 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705
1706 /*
1707 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1708 * here, and the (potentially unmapped) buffers may become dirty at
1709 * any time. If a buffer becomes dirty here after we've inspected it
1710 * then we just miss that fact, and the page stays dirty.
1711 *
1712 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1713 * handle that here by just cleaning them.
1714 */
1715
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001717 blocksize = bh->b_size;
1718 bbits = block_size_bits(blocksize);
1719
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001720 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001721 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722
1723 /*
1724 * Get all the dirty buffers mapped to disk addresses and
1725 * handle any aliases from the underlying blockdev's mapping.
1726 */
1727 do {
1728 if (block > last_block) {
1729 /*
1730 * mapped buffers outside i_size will occur, because
1731 * this page can be outside i_size when there is a
1732 * truncate in progress.
1733 */
1734 /*
1735 * The buffer was zeroed by block_write_full_page()
1736 */
1737 clear_buffer_dirty(bh);
1738 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001739 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1740 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001741 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742 err = get_block(inode, block, bh, 1);
1743 if (err)
1744 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001745 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001746 if (buffer_new(bh)) {
1747 /* blockdev mappings never come here */
1748 clear_buffer_new(bh);
1749 unmap_underlying_metadata(bh->b_bdev,
1750 bh->b_blocknr);
1751 }
1752 }
1753 bh = bh->b_this_page;
1754 block++;
1755 } while (bh != head);
1756
1757 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001758 if (!buffer_mapped(bh))
1759 continue;
1760 /*
1761 * If it's a fully non-blocking write attempt and we cannot
1762 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001763 * potentially cause a busy-wait loop from writeback threads
1764 * and kswapd activity, but those code paths have their own
1765 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001767 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001769 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001770 redirty_page_for_writepage(wbc, page);
1771 continue;
1772 }
1773 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001774 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 } else {
1776 unlock_buffer(bh);
1777 }
1778 } while ((bh = bh->b_this_page) != head);
1779
1780 /*
1781 * The page and its buffers are protected by PageWriteback(), so we can
1782 * drop the bh refcounts early.
1783 */
1784 BUG_ON(PageWriteback(page));
1785 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001786
1787 do {
1788 struct buffer_head *next = bh->b_this_page;
1789 if (buffer_async_write(bh)) {
Mike Christie2a222ca2016-06-05 14:31:43 -05001790 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001791 nr_underway++;
1792 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 bh = next;
1794 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001795 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796
1797 err = 0;
1798done:
1799 if (nr_underway == 0) {
1800 /*
1801 * The page was marked dirty, but the buffers were
1802 * clean. Someone wrote them back by hand with
1803 * ll_rw_block/submit_bh. A rare case.
1804 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001806
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 /*
1808 * The page and buffer_heads can be released at any time from
1809 * here on.
1810 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811 }
1812 return err;
1813
1814recover:
1815 /*
1816 * ENOSPC, or some other error. We may already have added some
1817 * blocks to the file, so we need to write these out to avoid
1818 * exposing stale data.
1819 * The page is currently locked and not marked for writeback
1820 */
1821 bh = head;
1822 /* Recovery: lock and submit the mapped buffers */
1823 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001824 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1825 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001827 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 } else {
1829 /*
1830 * The buffer may have been set dirty during
1831 * attachment to a dirty page.
1832 */
1833 clear_buffer_dirty(bh);
1834 }
1835 } while ((bh = bh->b_this_page) != head);
1836 SetPageError(page);
1837 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001838 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 do {
1841 struct buffer_head *next = bh->b_this_page;
1842 if (buffer_async_write(bh)) {
1843 clear_buffer_dirty(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05001844 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 nr_underway++;
1846 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 bh = next;
1848 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001849 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001850 goto done;
1851}
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001852EXPORT_SYMBOL(__block_write_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853
Nick Pigginafddba42007-10-16 01:25:01 -07001854/*
1855 * If a page has any new buffers, zero them out here, and mark them uptodate
1856 * and dirty so they'll be written out (in order to prevent uninitialised
1857 * block data from leaking). And clear the new bit.
1858 */
1859void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1860{
1861 unsigned int block_start, block_end;
1862 struct buffer_head *head, *bh;
1863
1864 BUG_ON(!PageLocked(page));
1865 if (!page_has_buffers(page))
1866 return;
1867
1868 bh = head = page_buffers(page);
1869 block_start = 0;
1870 do {
1871 block_end = block_start + bh->b_size;
1872
1873 if (buffer_new(bh)) {
1874 if (block_end > from && block_start < to) {
1875 if (!PageUptodate(page)) {
1876 unsigned start, size;
1877
1878 start = max(from, block_start);
1879 size = min(to, block_end) - start;
1880
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001881 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001882 set_buffer_uptodate(bh);
1883 }
1884
1885 clear_buffer_new(bh);
1886 mark_buffer_dirty(bh);
1887 }
1888 }
1889
1890 block_start = block_end;
1891 bh = bh->b_this_page;
1892 } while (bh != head);
1893}
1894EXPORT_SYMBOL(page_zero_new_buffers);
1895
Christoph Hellwigae259a92016-06-21 09:23:11 +10001896static void
1897iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
1898 struct iomap *iomap)
1899{
1900 loff_t offset = block << inode->i_blkbits;
1901
1902 bh->b_bdev = iomap->bdev;
1903
1904 /*
1905 * Block points to offset in file we need to map, iomap contains
1906 * the offset at which the map starts. If the map ends before the
1907 * current block, then do not map the buffer and let the caller
1908 * handle it.
1909 */
1910 BUG_ON(offset >= iomap->offset + iomap->length);
1911
1912 switch (iomap->type) {
1913 case IOMAP_HOLE:
1914 /*
1915 * If the buffer is not up to date or beyond the current EOF,
1916 * we need to mark it as new to ensure sub-block zeroing is
1917 * executed if necessary.
1918 */
1919 if (!buffer_uptodate(bh) ||
1920 (offset >= i_size_read(inode)))
1921 set_buffer_new(bh);
1922 break;
1923 case IOMAP_DELALLOC:
1924 if (!buffer_uptodate(bh) ||
1925 (offset >= i_size_read(inode)))
1926 set_buffer_new(bh);
1927 set_buffer_uptodate(bh);
1928 set_buffer_mapped(bh);
1929 set_buffer_delay(bh);
1930 break;
1931 case IOMAP_UNWRITTEN:
1932 /*
1933 * For unwritten regions, we always need to ensure that
1934 * sub-block writes cause the regions in the block we are not
1935 * writing to are zeroed. Set the buffer as new to ensure this.
1936 */
1937 set_buffer_new(bh);
1938 set_buffer_unwritten(bh);
1939 /* FALLTHRU */
1940 case IOMAP_MAPPED:
1941 if (offset >= i_size_read(inode))
1942 set_buffer_new(bh);
1943 bh->b_blocknr = (iomap->blkno >> (inode->i_blkbits - 9)) +
1944 ((offset - iomap->offset) >> inode->i_blkbits);
1945 set_buffer_mapped(bh);
1946 break;
1947 }
1948}
1949
1950int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1951 get_block_t *get_block, struct iomap *iomap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001952{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001953 unsigned from = pos & (PAGE_SIZE - 1);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001954 unsigned to = from + len;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001955 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956 unsigned block_start, block_end;
1957 sector_t block;
1958 int err = 0;
1959 unsigned blocksize, bbits;
1960 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1961
1962 BUG_ON(!PageLocked(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001963 BUG_ON(from > PAGE_SIZE);
1964 BUG_ON(to > PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001965 BUG_ON(from > to);
1966
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001967 head = create_page_buffers(page, inode, 0);
1968 blocksize = head->b_size;
1969 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001971 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001972
1973 for(bh = head, block_start = 0; bh != head || !block_start;
1974 block++, block_start=block_end, bh = bh->b_this_page) {
1975 block_end = block_start + blocksize;
1976 if (block_end <= from || block_start >= to) {
1977 if (PageUptodate(page)) {
1978 if (!buffer_uptodate(bh))
1979 set_buffer_uptodate(bh);
1980 }
1981 continue;
1982 }
1983 if (buffer_new(bh))
1984 clear_buffer_new(bh);
1985 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001986 WARN_ON(bh->b_size != blocksize);
Christoph Hellwigae259a92016-06-21 09:23:11 +10001987 if (get_block) {
1988 err = get_block(inode, block, bh, 1);
1989 if (err)
1990 break;
1991 } else {
1992 iomap_to_bh(inode, block, bh, iomap);
1993 }
1994
Linus Torvalds1da177e2005-04-16 15:20:36 -07001995 if (buffer_new(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 unmap_underlying_metadata(bh->b_bdev,
1997 bh->b_blocknr);
1998 if (PageUptodate(page)) {
Nick Piggin637aff42007-10-16 01:25:00 -07001999 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07002001 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002002 continue;
2003 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002004 if (block_end > to || block_start < from)
2005 zero_user_segments(page,
2006 to, block_end,
2007 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002008 continue;
2009 }
2010 }
2011 if (PageUptodate(page)) {
2012 if (!buffer_uptodate(bh))
2013 set_buffer_uptodate(bh);
2014 continue;
2015 }
2016 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08002017 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002018 (block_start < from || block_end > to)) {
Mike Christiedfec8a12016-06-05 14:31:44 -05002019 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 *wait_bh++=bh;
2021 }
2022 }
2023 /*
2024 * If we issued read requests - let them complete.
2025 */
2026 while(wait_bh > wait) {
2027 wait_on_buffer(*--wait_bh);
2028 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07002029 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002030 }
Jan Karaf9f07b62011-06-14 00:58:27 +02002031 if (unlikely(err))
Nick Pigginafddba42007-10-16 01:25:01 -07002032 page_zero_new_buffers(page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002033 return err;
2034}
Christoph Hellwigae259a92016-06-21 09:23:11 +10002035
2036int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2037 get_block_t *get_block)
2038{
2039 return __block_write_begin_int(page, pos, len, get_block, NULL);
2040}
Christoph Hellwigebdec242010-10-06 10:47:23 +02002041EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002042
2043static int __block_commit_write(struct inode *inode, struct page *page,
2044 unsigned from, unsigned to)
2045{
2046 unsigned block_start, block_end;
2047 int partial = 0;
2048 unsigned blocksize;
2049 struct buffer_head *bh, *head;
2050
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002051 bh = head = page_buffers(page);
2052 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002053
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002054 block_start = 0;
2055 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056 block_end = block_start + blocksize;
2057 if (block_end <= from || block_start >= to) {
2058 if (!buffer_uptodate(bh))
2059 partial = 1;
2060 } else {
2061 set_buffer_uptodate(bh);
2062 mark_buffer_dirty(bh);
2063 }
Nick Pigginafddba42007-10-16 01:25:01 -07002064 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002065
2066 block_start = block_end;
2067 bh = bh->b_this_page;
2068 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069
2070 /*
2071 * If this is a partial write which happened to make all buffers
2072 * uptodate then we can optimize away a bogus readpage() for
2073 * the next read(). Here we 'discover' whether the page went
2074 * uptodate as a result of this (potentially partial) write.
2075 */
2076 if (!partial)
2077 SetPageUptodate(page);
2078 return 0;
2079}
2080
2081/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02002082 * block_write_begin takes care of the basic task of block allocation and
2083 * bringing partial write blocks uptodate first.
2084 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002085 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07002086 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02002087int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2088 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07002089{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002090 pgoff_t index = pos >> PAGE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07002091 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002092 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07002093
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002094 page = grab_cache_page_write_begin(mapping, index, flags);
2095 if (!page)
2096 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07002097
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002098 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07002099 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002100 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002101 put_page(page);
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002102 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07002103 }
2104
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002105 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07002106 return status;
2107}
2108EXPORT_SYMBOL(block_write_begin);
2109
2110int block_write_end(struct file *file, struct address_space *mapping,
2111 loff_t pos, unsigned len, unsigned copied,
2112 struct page *page, void *fsdata)
2113{
2114 struct inode *inode = mapping->host;
2115 unsigned start;
2116
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002117 start = pos & (PAGE_SIZE - 1);
Nick Pigginafddba42007-10-16 01:25:01 -07002118
2119 if (unlikely(copied < len)) {
2120 /*
2121 * The buffers that were written will now be uptodate, so we
2122 * don't have to worry about a readpage reading them and
2123 * overwriting a partial write. However if we have encountered
2124 * a short write and only partially written into a buffer, it
2125 * will not be marked uptodate, so a readpage might come in and
2126 * destroy our partial write.
2127 *
2128 * Do the simplest thing, and just treat any short write to a
2129 * non uptodate page as a zero-length write, and force the
2130 * caller to redo the whole thing.
2131 */
2132 if (!PageUptodate(page))
2133 copied = 0;
2134
2135 page_zero_new_buffers(page, start+copied, start+len);
2136 }
2137 flush_dcache_page(page);
2138
2139 /* This could be a short (even 0-length) commit */
2140 __block_commit_write(inode, page, start, start+copied);
2141
2142 return copied;
2143}
2144EXPORT_SYMBOL(block_write_end);
2145
2146int generic_write_end(struct file *file, struct address_space *mapping,
2147 loff_t pos, unsigned len, unsigned copied,
2148 struct page *page, void *fsdata)
2149{
2150 struct inode *inode = mapping->host;
Jan Kara90a80202014-10-01 21:49:18 -04002151 loff_t old_size = inode->i_size;
Jan Karac7d206b2008-07-11 19:27:31 -04002152 int i_size_changed = 0;
Nick Pigginafddba42007-10-16 01:25:01 -07002153
2154 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2155
2156 /*
2157 * No need to use i_size_read() here, the i_size
2158 * cannot change under us because we hold i_mutex.
2159 *
2160 * But it's important to update i_size while still holding page lock:
2161 * page writeout could otherwise come in and zero beyond i_size.
2162 */
2163 if (pos+copied > inode->i_size) {
2164 i_size_write(inode, pos+copied);
Jan Karac7d206b2008-07-11 19:27:31 -04002165 i_size_changed = 1;
Nick Pigginafddba42007-10-16 01:25:01 -07002166 }
2167
2168 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002169 put_page(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002170
Jan Kara90a80202014-10-01 21:49:18 -04002171 if (old_size < pos)
2172 pagecache_isize_extended(inode, old_size, pos);
Jan Karac7d206b2008-07-11 19:27:31 -04002173 /*
2174 * Don't mark the inode dirty under page lock. First, it unnecessarily
2175 * makes the holding time of page lock longer. Second, it forces lock
2176 * ordering of page lock and transaction start for journaling
2177 * filesystems.
2178 */
2179 if (i_size_changed)
2180 mark_inode_dirty(inode);
2181
Nick Pigginafddba42007-10-16 01:25:01 -07002182 return copied;
2183}
2184EXPORT_SYMBOL(generic_write_end);
2185
2186/*
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002187 * block_is_partially_uptodate checks whether buffers within a page are
2188 * uptodate or not.
2189 *
2190 * Returns true if all buffers which correspond to a file portion
2191 * we want to read are uptodate.
2192 */
Al Viroc186afb42014-02-02 21:16:54 -05002193int block_is_partially_uptodate(struct page *page, unsigned long from,
2194 unsigned long count)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002195{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002196 unsigned block_start, block_end, blocksize;
2197 unsigned to;
2198 struct buffer_head *bh, *head;
2199 int ret = 1;
2200
2201 if (!page_has_buffers(page))
2202 return 0;
2203
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002204 head = page_buffers(page);
2205 blocksize = head->b_size;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002206 to = min_t(unsigned, PAGE_SIZE - from, count);
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002207 to = from + to;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002208 if (from < blocksize && to > PAGE_SIZE - blocksize)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002209 return 0;
2210
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002211 bh = head;
2212 block_start = 0;
2213 do {
2214 block_end = block_start + blocksize;
2215 if (block_end > from && block_start < to) {
2216 if (!buffer_uptodate(bh)) {
2217 ret = 0;
2218 break;
2219 }
2220 if (block_end >= to)
2221 break;
2222 }
2223 block_start = block_end;
2224 bh = bh->b_this_page;
2225 } while (bh != head);
2226
2227 return ret;
2228}
2229EXPORT_SYMBOL(block_is_partially_uptodate);
2230
2231/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002232 * Generic "read page" function for block devices that have the normal
2233 * get_block functionality. This is most of the block device filesystems.
2234 * Reads the page asynchronously --- the unlock_buffer() and
2235 * set/clear_buffer_uptodate() functions propagate buffer state into the
2236 * page struct once IO has completed.
2237 */
2238int block_read_full_page(struct page *page, get_block_t *get_block)
2239{
2240 struct inode *inode = page->mapping->host;
2241 sector_t iblock, lblock;
2242 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002243 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002244 int nr, i;
2245 int fully_mapped = 1;
2246
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002247 head = create_page_buffers(page, inode, 0);
2248 blocksize = head->b_size;
2249 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002250
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002251 iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002252 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002253 bh = head;
2254 nr = 0;
2255 i = 0;
2256
2257 do {
2258 if (buffer_uptodate(bh))
2259 continue;
2260
2261 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002262 int err = 0;
2263
Linus Torvalds1da177e2005-04-16 15:20:36 -07002264 fully_mapped = 0;
2265 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002266 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002267 err = get_block(inode, iblock, bh, 0);
2268 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 SetPageError(page);
2270 }
2271 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002272 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002273 if (!err)
2274 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002275 continue;
2276 }
2277 /*
2278 * get_block() might have updated the buffer
2279 * synchronously
2280 */
2281 if (buffer_uptodate(bh))
2282 continue;
2283 }
2284 arr[nr++] = bh;
2285 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2286
2287 if (fully_mapped)
2288 SetPageMappedToDisk(page);
2289
2290 if (!nr) {
2291 /*
2292 * All buffers are uptodate - we can set the page uptodate
2293 * as well. But not if get_block() returned an error.
2294 */
2295 if (!PageError(page))
2296 SetPageUptodate(page);
2297 unlock_page(page);
2298 return 0;
2299 }
2300
2301 /* Stage two: lock the buffers */
2302 for (i = 0; i < nr; i++) {
2303 bh = arr[i];
2304 lock_buffer(bh);
2305 mark_buffer_async_read(bh);
2306 }
2307
2308 /*
2309 * Stage 3: start the IO. Check for uptodateness
2310 * inside the buffer lock in case another process reading
2311 * the underlying blockdev brought it uptodate (the sct fix).
2312 */
2313 for (i = 0; i < nr; i++) {
2314 bh = arr[i];
2315 if (buffer_uptodate(bh))
2316 end_buffer_async_read(bh, 1);
2317 else
Mike Christie2a222ca2016-06-05 14:31:43 -05002318 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002319 }
2320 return 0;
2321}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002322EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323
2324/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002325 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 * deal with the hole.
2327 */
Nick Piggin89e10782007-10-16 01:25:07 -07002328int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329{
2330 struct address_space *mapping = inode->i_mapping;
2331 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002332 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333 int err;
2334
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002335 err = inode_newsize_ok(inode, size);
2336 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 goto out;
2338
Nick Piggin89e10782007-10-16 01:25:07 -07002339 err = pagecache_write_begin(NULL, mapping, size, 0,
2340 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2341 &page, &fsdata);
2342 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002344
Nick Piggin89e10782007-10-16 01:25:07 -07002345 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2346 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002347
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348out:
2349 return err;
2350}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002351EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002352
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002353static int cont_expand_zero(struct file *file, struct address_space *mapping,
2354 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002355{
Nick Piggin89e10782007-10-16 01:25:07 -07002356 struct inode *inode = mapping->host;
2357 unsigned blocksize = 1 << inode->i_blkbits;
2358 struct page *page;
2359 void *fsdata;
2360 pgoff_t index, curidx;
2361 loff_t curpos;
2362 unsigned zerofrom, offset, len;
2363 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002364
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002365 index = pos >> PAGE_SHIFT;
2366 offset = pos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002367
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002368 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2369 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002370 if (zerofrom & (blocksize-1)) {
2371 *bytes |= (blocksize-1);
2372 (*bytes)++;
2373 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002374 len = PAGE_SIZE - zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002375
2376 err = pagecache_write_begin(file, mapping, curpos, len,
2377 AOP_FLAG_UNINTERRUPTIBLE,
2378 &page, &fsdata);
2379 if (err)
2380 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002381 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002382 err = pagecache_write_end(file, mapping, curpos, len, len,
2383 page, fsdata);
2384 if (err < 0)
2385 goto out;
2386 BUG_ON(err != len);
2387 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002388
2389 balance_dirty_pages_ratelimited(mapping);
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002390
2391 if (unlikely(fatal_signal_pending(current))) {
2392 err = -EINTR;
2393 goto out;
2394 }
Nick Piggin89e10782007-10-16 01:25:07 -07002395 }
2396
2397 /* page covers the boundary, find the boundary offset */
2398 if (index == curidx) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002399 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002400 /* if we will expand the thing last block will be filled */
2401 if (offset <= zerofrom) {
2402 goto out;
2403 }
2404 if (zerofrom & (blocksize-1)) {
2405 *bytes |= (blocksize-1);
2406 (*bytes)++;
2407 }
2408 len = offset - zerofrom;
2409
2410 err = pagecache_write_begin(file, mapping, curpos, len,
2411 AOP_FLAG_UNINTERRUPTIBLE,
2412 &page, &fsdata);
2413 if (err)
2414 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002415 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002416 err = pagecache_write_end(file, mapping, curpos, len, len,
2417 page, fsdata);
2418 if (err < 0)
2419 goto out;
2420 BUG_ON(err != len);
2421 err = 0;
2422 }
2423out:
2424 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002425}
2426
Linus Torvalds1da177e2005-04-16 15:20:36 -07002427/*
2428 * For moronic filesystems that do not allow holes in file.
2429 * We may have to extend the file.
2430 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002431int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002432 loff_t pos, unsigned len, unsigned flags,
2433 struct page **pagep, void **fsdata,
2434 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002435{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002437 unsigned blocksize = 1 << inode->i_blkbits;
Nick Piggin89e10782007-10-16 01:25:07 -07002438 unsigned zerofrom;
2439 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002440
Nick Piggin89e10782007-10-16 01:25:07 -07002441 err = cont_expand_zero(file, mapping, pos, bytes);
2442 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002443 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002444
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002445 zerofrom = *bytes & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002446 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2447 *bytes |= (blocksize-1);
2448 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002449 }
2450
Christoph Hellwig155130a2010-06-04 11:29:58 +02002451 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002453EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002454
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455int block_commit_write(struct page *page, unsigned from, unsigned to)
2456{
2457 struct inode *inode = page->mapping->host;
2458 __block_commit_write(inode,page,from,to);
2459 return 0;
2460}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002461EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462
David Chinner54171692007-07-19 17:39:55 +10002463/*
2464 * block_page_mkwrite() is not allowed to change the file size as it gets
2465 * called from a page fault handler when a page is first dirtied. Hence we must
2466 * be careful to check for EOF conditions here. We set the page up correctly
2467 * for a written page which means we get ENOSPC checking when writing into
2468 * holes and correct delalloc and unwritten extent mapping on filesystems that
2469 * support these features.
2470 *
2471 * We are not allowed to take the i_mutex here so we have to play games to
2472 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002473 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002474 * page lock we can determine safely if the page is beyond EOF. If it is not
2475 * beyond EOF, then the page is guaranteed safe against truncation until we
2476 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002477 *
Jan Kara14da9202012-06-12 16:20:37 +02002478 * Direct callers of this function should protect against filesystem freezing
Ross Zwisler5c500022015-10-13 16:51:02 -06002479 * using sb_start_pagefault() - sb_end_pagefault() functions.
David Chinner54171692007-07-19 17:39:55 +10002480 */
Ross Zwisler5c500022015-10-13 16:51:02 -06002481int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
Jan Kara24da4fa2011-05-24 00:23:34 +02002482 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002483{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002484 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002485 struct inode *inode = file_inode(vma->vm_file);
David Chinner54171692007-07-19 17:39:55 +10002486 unsigned long end;
2487 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002488 int ret;
David Chinner54171692007-07-19 17:39:55 +10002489
2490 lock_page(page);
2491 size = i_size_read(inode);
2492 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002493 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002494 /* We overload EFAULT to mean page got truncated */
2495 ret = -EFAULT;
2496 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002497 }
2498
2499 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002500 if (((page->index + 1) << PAGE_SHIFT) > size)
2501 end = size & ~PAGE_MASK;
David Chinner54171692007-07-19 17:39:55 +10002502 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002503 end = PAGE_SIZE;
David Chinner54171692007-07-19 17:39:55 +10002504
Christoph Hellwigebdec242010-10-06 10:47:23 +02002505 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002506 if (!ret)
2507 ret = block_commit_write(page, 0, end);
2508
Jan Kara24da4fa2011-05-24 00:23:34 +02002509 if (unlikely(ret < 0))
2510 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002511 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002512 wait_for_stable_page(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002513 return 0;
2514out_unlock:
2515 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002516 return ret;
2517}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002518EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
2520/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002521 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002522 * immediately, while under the page lock. So it needs a special end_io
2523 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 */
2525static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2526{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002527 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002528}
2529
2530/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002531 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2532 * the page (converting it to circular linked list and taking care of page
2533 * dirty races).
2534 */
2535static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2536{
2537 struct buffer_head *bh;
2538
2539 BUG_ON(!PageLocked(page));
2540
2541 spin_lock(&page->mapping->private_lock);
2542 bh = head;
2543 do {
2544 if (PageDirty(page))
2545 set_buffer_dirty(bh);
2546 if (!bh->b_this_page)
2547 bh->b_this_page = head;
2548 bh = bh->b_this_page;
2549 } while (bh != head);
2550 attach_page_buffers(page, head);
2551 spin_unlock(&page->mapping->private_lock);
2552}
2553
2554/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002555 * On entry, the page is fully not uptodate.
2556 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002557 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002558 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002559int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002560 loff_t pos, unsigned len, unsigned flags,
2561 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002562 get_block_t *get_block)
2563{
Nick Piggin03158cd2007-10-16 01:25:25 -07002564 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 const unsigned blkbits = inode->i_blkbits;
2566 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002567 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002568 struct page *page;
2569 pgoff_t index;
2570 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002572 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002573 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002575 int ret = 0;
2576 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002578 index = pos >> PAGE_SHIFT;
2579 from = pos & (PAGE_SIZE - 1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002580 to = from + len;
2581
Nick Piggin54566b22009-01-04 12:00:53 -08002582 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002583 if (!page)
2584 return -ENOMEM;
2585 *pagep = page;
2586 *fsdata = NULL;
2587
2588 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002589 ret = __block_write_begin(page, pos, len, get_block);
2590 if (unlikely(ret))
2591 goto out_release;
2592 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002593 }
Nick Piggina4b06722007-10-16 01:24:48 -07002594
Linus Torvalds1da177e2005-04-16 15:20:36 -07002595 if (PageMappedToDisk(page))
2596 return 0;
2597
Nick Piggina4b06722007-10-16 01:24:48 -07002598 /*
2599 * Allocate buffers so that we can keep track of state, and potentially
2600 * attach them to the page if an error occurs. In the common case of
2601 * no error, they will just be freed again without ever being attached
2602 * to the page (which is all OK, because we're under the page lock).
2603 *
2604 * Be careful: the buffer linked list is a NULL terminated one, rather
2605 * than the circular one we're used to.
2606 */
2607 head = alloc_page_buffers(page, blocksize, 0);
Nick Piggin03158cd2007-10-16 01:25:25 -07002608 if (!head) {
2609 ret = -ENOMEM;
2610 goto out_release;
2611 }
Nick Piggina4b06722007-10-16 01:24:48 -07002612
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002613 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614
2615 /*
2616 * We loop across all blocks in the page, whether or not they are
2617 * part of the affected region. This is so we can discover if the
2618 * page is fully mapped-to-disk.
2619 */
Nick Piggina4b06722007-10-16 01:24:48 -07002620 for (block_start = 0, block_in_page = 0, bh = head;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002621 block_start < PAGE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002622 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 int create;
2624
Nick Piggina4b06722007-10-16 01:24:48 -07002625 block_end = block_start + blocksize;
2626 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 create = 1;
2628 if (block_start >= to)
2629 create = 0;
2630 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002631 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632 if (ret)
2633 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002634 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002636 if (buffer_new(bh))
2637 unmap_underlying_metadata(bh->b_bdev, bh->b_blocknr);
2638 if (PageUptodate(page)) {
2639 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002641 }
2642 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002643 zero_user_segments(page, block_start, from,
2644 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002645 continue;
2646 }
Nick Piggina4b06722007-10-16 01:24:48 -07002647 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002648 continue; /* reiserfs does this */
2649 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002650 lock_buffer(bh);
2651 bh->b_end_io = end_buffer_read_nobh;
Mike Christie2a222ca2016-06-05 14:31:43 -05002652 submit_bh(REQ_OP_READ, 0, bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002653 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002654 }
2655 }
2656
2657 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002658 /*
2659 * The page is locked, so these buffers are protected from
2660 * any VM or truncate activity. Hence we don't need to care
2661 * for the buffer_head refcounts.
2662 */
Nick Piggina4b06722007-10-16 01:24:48 -07002663 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 wait_on_buffer(bh);
2665 if (!buffer_uptodate(bh))
2666 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002667 }
2668 if (ret)
2669 goto failed;
2670 }
2671
2672 if (is_mapped_to_disk)
2673 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002674
Nick Piggin03158cd2007-10-16 01:25:25 -07002675 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002676
Linus Torvalds1da177e2005-04-16 15:20:36 -07002677 return 0;
2678
2679failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002680 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002682 * Error recovery is a bit difficult. We need to zero out blocks that
2683 * were newly allocated, and dirty them to ensure they get written out.
2684 * Buffers need to be attached to the page at this point, otherwise
2685 * the handling of potential IO errors during writeout would be hard
2686 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002687 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002688 attach_nobh_buffers(page, head);
2689 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002690
Nick Piggin03158cd2007-10-16 01:25:25 -07002691out_release:
2692 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002693 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002694 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002695
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002696 return ret;
2697}
Nick Piggin03158cd2007-10-16 01:25:25 -07002698EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699
Nick Piggin03158cd2007-10-16 01:25:25 -07002700int nobh_write_end(struct file *file, struct address_space *mapping,
2701 loff_t pos, unsigned len, unsigned copied,
2702 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002703{
2704 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002705 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002706 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002707 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002709 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002710 attach_nobh_buffers(page, head);
2711 if (page_has_buffers(page))
2712 return generic_write_end(file, mapping, pos, len,
2713 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002714
Nick Piggin22c8ca72007-02-20 13:58:09 -08002715 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002717 if (pos+copied > inode->i_size) {
2718 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 mark_inode_dirty(inode);
2720 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002721
2722 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002723 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002724
Nick Piggin03158cd2007-10-16 01:25:25 -07002725 while (head) {
2726 bh = head;
2727 head = head->b_this_page;
2728 free_buffer_head(bh);
2729 }
2730
2731 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732}
Nick Piggin03158cd2007-10-16 01:25:25 -07002733EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002734
2735/*
2736 * nobh_writepage() - based on block_full_write_page() except
2737 * that it tries to operate without attaching bufferheads to
2738 * the page.
2739 */
2740int nobh_writepage(struct page *page, get_block_t *get_block,
2741 struct writeback_control *wbc)
2742{
2743 struct inode * const inode = page->mapping->host;
2744 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002745 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002747 int ret;
2748
2749 /* Is the page fully inside i_size? */
2750 if (page->index < end_index)
2751 goto out;
2752
2753 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002754 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 if (page->index >= end_index+1 || !offset) {
2756 /*
2757 * The page may have dirty, unmapped buffers. For example,
2758 * they may have been added in ext3_writepage(). Make them
2759 * freeable here, so the page does not leak.
2760 */
2761#if 0
2762 /* Not really sure about this - do we need this ? */
2763 if (page->mapping->a_ops->invalidatepage)
2764 page->mapping->a_ops->invalidatepage(page, offset);
2765#endif
2766 unlock_page(page);
2767 return 0; /* don't care */
2768 }
2769
2770 /*
2771 * The page straddles i_size. It must be zeroed out on each and every
2772 * writepage invocation because it may be mmapped. "A file is mapped
2773 * in multiples of the page size. For a file that is not a multiple of
2774 * the page size, the remaining memory is zeroed when mapped, and
2775 * writes to that region are not written out to the file."
2776 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002777 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002778out:
2779 ret = mpage_writepage(page, get_block, wbc);
2780 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002781 ret = __block_write_full_page(inode, page, get_block, wbc,
2782 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002783 return ret;
2784}
2785EXPORT_SYMBOL(nobh_writepage);
2786
Nick Piggin03158cd2007-10-16 01:25:25 -07002787int nobh_truncate_page(struct address_space *mapping,
2788 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002790 pgoff_t index = from >> PAGE_SHIFT;
2791 unsigned offset = from & (PAGE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002792 unsigned blocksize;
2793 sector_t iblock;
2794 unsigned length, pos;
2795 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002797 struct buffer_head map_bh;
2798 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799
Nick Piggin03158cd2007-10-16 01:25:25 -07002800 blocksize = 1 << inode->i_blkbits;
2801 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802
Nick Piggin03158cd2007-10-16 01:25:25 -07002803 /* Block boundary? Nothing to do */
2804 if (!length)
2805 return 0;
2806
2807 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002808 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Nick Piggin03158cd2007-10-16 01:25:25 -07002809
Linus Torvalds1da177e2005-04-16 15:20:36 -07002810 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002811 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 if (!page)
2813 goto out;
2814
Nick Piggin03158cd2007-10-16 01:25:25 -07002815 if (page_has_buffers(page)) {
2816has_buffers:
2817 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002818 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002819 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002821
2822 /* Find the buffer that contains "offset" */
2823 pos = blocksize;
2824 while (offset >= pos) {
2825 iblock++;
2826 pos += blocksize;
2827 }
2828
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002829 map_bh.b_size = blocksize;
2830 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002831 err = get_block(inode, iblock, &map_bh, 0);
2832 if (err)
2833 goto unlock;
2834 /* unmapped? It's a hole - nothing to do */
2835 if (!buffer_mapped(&map_bh))
2836 goto unlock;
2837
2838 /* Ok, it's mapped. Make sure it's up-to-date */
2839 if (!PageUptodate(page)) {
2840 err = mapping->a_ops->readpage(NULL, page);
2841 if (err) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002842 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002843 goto out;
2844 }
2845 lock_page(page);
2846 if (!PageUptodate(page)) {
2847 err = -EIO;
2848 goto unlock;
2849 }
2850 if (page_has_buffers(page))
2851 goto has_buffers;
2852 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002853 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002854 set_page_dirty(page);
2855 err = 0;
2856
2857unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002858 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002859 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002861 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862}
2863EXPORT_SYMBOL(nobh_truncate_page);
2864
2865int block_truncate_page(struct address_space *mapping,
2866 loff_t from, get_block_t *get_block)
2867{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002868 pgoff_t index = from >> PAGE_SHIFT;
2869 unsigned offset = from & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002870 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002871 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 unsigned length, pos;
2873 struct inode *inode = mapping->host;
2874 struct page *page;
2875 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002876 int err;
2877
2878 blocksize = 1 << inode->i_blkbits;
2879 length = offset & (blocksize - 1);
2880
2881 /* Block boundary? Nothing to do */
2882 if (!length)
2883 return 0;
2884
2885 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002886 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887
2888 page = grab_cache_page(mapping, index);
2889 err = -ENOMEM;
2890 if (!page)
2891 goto out;
2892
2893 if (!page_has_buffers(page))
2894 create_empty_buffers(page, blocksize, 0);
2895
2896 /* Find the buffer that contains "offset" */
2897 bh = page_buffers(page);
2898 pos = blocksize;
2899 while (offset >= pos) {
2900 bh = bh->b_this_page;
2901 iblock++;
2902 pos += blocksize;
2903 }
2904
2905 err = 0;
2906 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002907 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002908 err = get_block(inode, iblock, bh, 0);
2909 if (err)
2910 goto unlock;
2911 /* unmapped? It's a hole - nothing to do */
2912 if (!buffer_mapped(bh))
2913 goto unlock;
2914 }
2915
2916 /* Ok, it's mapped. Make sure it's up-to-date */
2917 if (PageUptodate(page))
2918 set_buffer_uptodate(bh);
2919
David Chinner33a266d2007-02-12 00:51:41 -08002920 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002921 err = -EIO;
Mike Christiedfec8a12016-06-05 14:31:44 -05002922 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 wait_on_buffer(bh);
2924 /* Uhhuh. Read error. Complain and punt. */
2925 if (!buffer_uptodate(bh))
2926 goto unlock;
2927 }
2928
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002929 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002930 mark_buffer_dirty(bh);
2931 err = 0;
2932
2933unlock:
2934 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002935 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002936out:
2937 return err;
2938}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002939EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002940
2941/*
2942 * The generic ->writepage function for buffer-backed address_spaces
2943 */
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002944int block_write_full_page(struct page *page, get_block_t *get_block,
2945 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002946{
2947 struct inode * const inode = page->mapping->host;
2948 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002949 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002950 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002951
2952 /* Is the page fully inside i_size? */
2953 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04002954 return __block_write_full_page(inode, page, get_block, wbc,
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002955 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002956
2957 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002958 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002959 if (page->index >= end_index+1 || !offset) {
2960 /*
2961 * The page may have dirty, unmapped buffers. For example,
2962 * they may have been added in ext3_writepage(). Make them
2963 * freeable here, so the page does not leak.
2964 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002965 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002966 unlock_page(page);
2967 return 0; /* don't care */
2968 }
2969
2970 /*
2971 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05002972 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973 * in multiples of the page size. For a file that is not a multiple of
2974 * the page size, the remaining memory is zeroed when mapped, and
2975 * writes to that region are not written out to the file."
2976 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002977 zero_user_segment(page, offset, PAGE_SIZE);
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002978 return __block_write_full_page(inode, page, get_block, wbc,
2979 end_buffer_async_write);
Chris Mason35c80d52009-04-15 13:22:38 -04002980}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002981EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04002982
Linus Torvalds1da177e2005-04-16 15:20:36 -07002983sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2984 get_block_t *get_block)
2985{
2986 struct buffer_head tmp;
2987 struct inode *inode = mapping->host;
2988 tmp.b_state = 0;
2989 tmp.b_blocknr = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002990 tmp.b_size = 1 << inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 get_block(inode, block, &tmp, 0);
2992 return tmp.b_blocknr;
2993}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002994EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002996static void end_bio_bh_io_sync(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997{
2998 struct buffer_head *bh = bio->bi_private;
2999
Jens Axboeb7c44ed2015-07-24 12:37:59 -06003000 if (unlikely(bio_flagged(bio, BIO_QUIET)))
Keith Mannthey08bafc02008-11-25 10:24:35 +01003001 set_bit(BH_Quiet, &bh->b_state);
3002
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003003 bh->b_end_io(bh, !bio->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005}
3006
Linus Torvalds57302e02012-12-04 08:25:11 -08003007/*
3008 * This allows us to do IO even on the odd last sectors
Akinobu Mita59d43912014-10-09 15:26:53 -07003009 * of a device, even if the block size is some multiple
Linus Torvalds57302e02012-12-04 08:25:11 -08003010 * of the physical sector size.
3011 *
3012 * We'll just truncate the bio to the size of the device,
3013 * and clear the end of the buffer head manually.
3014 *
3015 * Truly out-of-range accesses will turn into actual IO
3016 * errors, this only handles the "we need to be able to
3017 * do IO at the final sector" case.
3018 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003019void guard_bio_eod(int op, struct bio *bio)
Linus Torvalds57302e02012-12-04 08:25:11 -08003020{
3021 sector_t maxsector;
Akinobu Mita59d43912014-10-09 15:26:53 -07003022 struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
3023 unsigned truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08003024
3025 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
3026 if (!maxsector)
3027 return;
3028
3029 /*
3030 * If the *whole* IO is past the end of the device,
3031 * let it through, and the IO layer will turn it into
3032 * an EIO.
3033 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07003034 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08003035 return;
3036
Kent Overstreet4f024f32013-10-11 15:44:27 -07003037 maxsector -= bio->bi_iter.bi_sector;
Akinobu Mita59d43912014-10-09 15:26:53 -07003038 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08003039 return;
3040
Akinobu Mita59d43912014-10-09 15:26:53 -07003041 /* Uhhuh. We've got a bio that straddles the device size! */
3042 truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
Linus Torvalds57302e02012-12-04 08:25:11 -08003043
3044 /* Truncate the bio.. */
Akinobu Mita59d43912014-10-09 15:26:53 -07003045 bio->bi_iter.bi_size -= truncated_bytes;
3046 bvec->bv_len -= truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08003047
3048 /* ..and clear the end of the buffer for reads */
Mike Christie2a222ca2016-06-05 14:31:43 -05003049 if (op == REQ_OP_READ) {
Akinobu Mita59d43912014-10-09 15:26:53 -07003050 zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len,
3051 truncated_bytes);
Linus Torvalds57302e02012-12-04 08:25:11 -08003052 }
3053}
3054
Mike Christie2a222ca2016-06-05 14:31:43 -05003055static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Tejun Heob16b1de2015-06-02 08:39:48 -06003056 unsigned long bio_flags, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003057{
3058 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059
3060 BUG_ON(!buffer_locked(bh));
3061 BUG_ON(!buffer_mapped(bh));
3062 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04003063 BUG_ON(buffer_delay(bh));
3064 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003065
Jens Axboe48fd4f92008-08-22 10:00:36 +02003066 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02003067 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003069 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 clear_buffer_write_io_error(bh);
3071
3072 /*
3073 * from here on down, it's all bio -- do the initial mapping,
3074 * submit_bio -> generic_make_request may further map this bio around
3075 */
3076 bio = bio_alloc(GFP_NOIO, 1);
3077
Tejun Heo2a814902015-05-28 14:50:51 -04003078 if (wbc) {
Tejun Heob16b1de2015-06-02 08:39:48 -06003079 wbc_init_bio(wbc, bio);
Tejun Heo2a814902015-05-28 14:50:51 -04003080 wbc_account_io(wbc, bh->b_page, bh->b_size);
3081 }
Tejun Heobafc0db2015-06-02 08:37:23 -06003082
Kent Overstreet4f024f32013-10-11 15:44:27 -07003083 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003084 bio->bi_bdev = bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003085
Kent Overstreet6cf66b42014-12-22 12:48:42 +01003086 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3087 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003088
3089 bio->bi_end_io = end_bio_bh_io_sync;
3090 bio->bi_private = bh;
Darrick J. Wong7136851112013-04-29 15:07:25 -07003091 bio->bi_flags |= bio_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003092
Linus Torvalds57302e02012-12-04 08:25:11 -08003093 /* Take care of bh's that straddle the end of the device */
Mike Christie2a222ca2016-06-05 14:31:43 -05003094 guard_bio_eod(op, bio);
Linus Torvalds57302e02012-12-04 08:25:11 -08003095
Theodore Ts'o877f9622013-04-20 19:58:37 -04003096 if (buffer_meta(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003097 op_flags |= REQ_META;
Theodore Ts'o877f9622013-04-20 19:58:37 -04003098 if (buffer_prio(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003099 op_flags |= REQ_PRIO;
3100 bio_set_op_attrs(bio, op, op_flags);
Theodore Ts'o877f9622013-04-20 19:58:37 -04003101
Mike Christie4e49ea42016-06-05 14:31:41 -05003102 submit_bio(bio);
Julia Lawallf6454b02015-05-26 21:59:53 +02003103 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003104}
Tejun Heobafc0db2015-06-02 08:37:23 -06003105
Mike Christie2a222ca2016-06-05 14:31:43 -05003106int _submit_bh(int op, int op_flags, struct buffer_head *bh,
3107 unsigned long bio_flags)
Tejun Heobafc0db2015-06-02 08:37:23 -06003108{
Mike Christie2a222ca2016-06-05 14:31:43 -05003109 return submit_bh_wbc(op, op_flags, bh, bio_flags, NULL);
Tejun Heobafc0db2015-06-02 08:37:23 -06003110}
Darrick J. Wong7136851112013-04-29 15:07:25 -07003111EXPORT_SYMBOL_GPL(_submit_bh);
3112
Mike Christie2a222ca2016-06-05 14:31:43 -05003113int submit_bh(int op, int op_flags, struct buffer_head *bh)
Darrick J. Wong7136851112013-04-29 15:07:25 -07003114{
Mike Christie2a222ca2016-06-05 14:31:43 -05003115 return submit_bh_wbc(op, op_flags, bh, 0, NULL);
Darrick J. Wong7136851112013-04-29 15:07:25 -07003116}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003117EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003118
3119/**
3120 * ll_rw_block: low-level access to block devices (DEPRECATED)
Mike Christiedfec8a12016-06-05 14:31:44 -05003121 * @op: whether to %READ or %WRITE
Christoph Hellwig70246282016-07-19 11:28:41 +02003122 * @op_flags: rq_flag_bits
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123 * @nr: number of &struct buffer_heads in the array
3124 * @bhs: array of pointers to &struct buffer_head
3125 *
Jan Karaa7662232005-09-06 15:19:10 -07003126 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
Christoph Hellwig70246282016-07-19 11:28:41 +02003127 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3128 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3129 * %REQ_RAHEAD.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003130 *
3131 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003132 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3133 * request, and any buffer that appears to be up-to-date when doing read
3134 * request. Further it marks as clean buffers that are processed for
3135 * writing (the buffer cache won't assume that they are actually clean
3136 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003137 *
3138 * ll_rw_block sets b_end_io to simple completion handler that marks
Masanari Iidae2278672014-02-18 22:54:36 +09003139 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
Linus Torvalds1da177e2005-04-16 15:20:36 -07003140 * any waiters.
3141 *
3142 * All of the buffers must be for the same device, and must also be a
3143 * multiple of the current approved size for the device.
3144 */
Mike Christiedfec8a12016-06-05 14:31:44 -05003145void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
Linus Torvalds1da177e2005-04-16 15:20:36 -07003146{
3147 int i;
3148
3149 for (i = 0; i < nr; i++) {
3150 struct buffer_head *bh = bhs[i];
3151
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003152 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003153 continue;
Mike Christiedfec8a12016-06-05 14:31:44 -05003154 if (op == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003155 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003156 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003157 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003158 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159 continue;
3160 }
3161 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003163 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003164 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003165 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003166 continue;
3167 }
3168 }
3169 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003170 }
3171}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003172EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003173
Mike Christie2a222ca2016-06-05 14:31:43 -05003174void write_dirty_buffer(struct buffer_head *bh, int op_flags)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003175{
3176 lock_buffer(bh);
3177 if (!test_clear_buffer_dirty(bh)) {
3178 unlock_buffer(bh);
3179 return;
3180 }
3181 bh->b_end_io = end_buffer_write_sync;
3182 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05003183 submit_bh(REQ_OP_WRITE, op_flags, bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003184}
3185EXPORT_SYMBOL(write_dirty_buffer);
3186
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187/*
3188 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3189 * and then start new I/O and then wait upon it. The caller must have a ref on
3190 * the buffer_head.
3191 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003192int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003193{
3194 int ret = 0;
3195
3196 WARN_ON(atomic_read(&bh->b_count) < 1);
3197 lock_buffer(bh);
3198 if (test_clear_buffer_dirty(bh)) {
3199 get_bh(bh);
3200 bh->b_end_io = end_buffer_write_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003201 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003202 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 if (!ret && !buffer_uptodate(bh))
3204 ret = -EIO;
3205 } else {
3206 unlock_buffer(bh);
3207 }
3208 return ret;
3209}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003210EXPORT_SYMBOL(__sync_dirty_buffer);
3211
3212int sync_dirty_buffer(struct buffer_head *bh)
3213{
3214 return __sync_dirty_buffer(bh, WRITE_SYNC);
3215}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003216EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003217
3218/*
3219 * try_to_free_buffers() checks if all the buffers on this particular page
3220 * are unused, and releases them if so.
3221 *
3222 * Exclusion against try_to_free_buffers may be obtained by either
3223 * locking the page or by holding its mapping's private_lock.
3224 *
3225 * If the page is dirty but all the buffers are clean then we need to
3226 * be sure to mark the page clean as well. This is because the page
3227 * may be against a block device, and a later reattachment of buffers
3228 * to a dirty page will set *all* buffers dirty. Which would corrupt
3229 * filesystem data on the same device.
3230 *
3231 * The same applies to regular filesystem pages: if all the buffers are
3232 * clean then we set the page clean and proceed. To do that, we require
3233 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3234 * private_lock.
3235 *
3236 * try_to_free_buffers() is non-blocking.
3237 */
3238static inline int buffer_busy(struct buffer_head *bh)
3239{
3240 return atomic_read(&bh->b_count) |
3241 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3242}
3243
3244static int
3245drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3246{
3247 struct buffer_head *head = page_buffers(page);
3248 struct buffer_head *bh;
3249
3250 bh = head;
3251 do {
akpm@osdl.orgde7d5a32005-05-01 08:58:39 -07003252 if (buffer_write_io_error(bh) && page->mapping)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 set_bit(AS_EIO, &page->mapping->flags);
3254 if (buffer_busy(bh))
3255 goto failed;
3256 bh = bh->b_this_page;
3257 } while (bh != head);
3258
3259 do {
3260 struct buffer_head *next = bh->b_this_page;
3261
Jan Kara535ee2f2008-02-08 04:21:59 -08003262 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263 __remove_assoc_queue(bh);
3264 bh = next;
3265 } while (bh != head);
3266 *buffers_to_free = head;
3267 __clear_page_buffers(page);
3268 return 1;
3269failed:
3270 return 0;
3271}
3272
3273int try_to_free_buffers(struct page *page)
3274{
3275 struct address_space * const mapping = page->mapping;
3276 struct buffer_head *buffers_to_free = NULL;
3277 int ret = 0;
3278
3279 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003280 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003281 return 0;
3282
3283 if (mapping == NULL) { /* can this still happen? */
3284 ret = drop_buffers(page, &buffers_to_free);
3285 goto out;
3286 }
3287
3288 spin_lock(&mapping->private_lock);
3289 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003290
3291 /*
3292 * If the filesystem writes its buffers by hand (eg ext3)
3293 * then we can have clean buffers against a dirty page. We
3294 * clean the page here; otherwise the VM will never notice
3295 * that the filesystem did any IO at all.
3296 *
3297 * Also, during truncate, discard_buffer will have marked all
3298 * the page's buffers clean. We discover that here and clean
3299 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003300 *
3301 * private_lock must be held over this entire operation in order
3302 * to synchronise against __set_page_dirty_buffers and prevent the
3303 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003304 */
Tejun Heo11f81be2015-05-22 17:13:15 -04003305 if (ret)
3306 cancel_dirty_page(page);
Nick Piggin87df7242007-01-30 14:36:27 +11003307 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308out:
3309 if (buffers_to_free) {
3310 struct buffer_head *bh = buffers_to_free;
3311
3312 do {
3313 struct buffer_head *next = bh->b_this_page;
3314 free_buffer_head(bh);
3315 bh = next;
3316 } while (bh != buffers_to_free);
3317 }
3318 return ret;
3319}
3320EXPORT_SYMBOL(try_to_free_buffers);
3321
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322/*
3323 * There are no bdflush tunables left. But distributions are
3324 * still running obsolete flush daemons, so we terminate them here.
3325 *
3326 * Use of bdflush() is deprecated and will be removed in a future kernel.
Jens Axboe5b0830c2009-09-23 19:37:09 +02003327 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003328 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01003329SYSCALL_DEFINE2(bdflush, int, func, long, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003330{
3331 static int msg_count;
3332
3333 if (!capable(CAP_SYS_ADMIN))
3334 return -EPERM;
3335
3336 if (msg_count < 5) {
3337 msg_count++;
3338 printk(KERN_INFO
3339 "warning: process `%s' used the obsolete bdflush"
3340 " system call\n", current->comm);
3341 printk(KERN_INFO "Fix your initscripts?\n");
3342 }
3343
3344 if (func == 1)
3345 do_exit(0);
3346 return 0;
3347}
3348
3349/*
3350 * Buffer-head allocation
3351 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003352static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003353
3354/*
3355 * Once the number of bh's in the machine exceeds this level, we start
3356 * stripping them in writeback.
3357 */
Zhang Yanfei43be5942013-02-22 16:35:46 -08003358static unsigned long max_buffer_heads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003359
3360int buffer_heads_over_limit;
3361
3362struct bh_accounting {
3363 int nr; /* Number of live bh's */
3364 int ratelimit; /* Limit cacheline bouncing */
3365};
3366
3367static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3368
3369static void recalc_bh_state(void)
3370{
3371 int i;
3372 int tot = 0;
3373
Christoph Lameteree1be862010-12-06 11:40:05 -06003374 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003376 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003377 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003378 tot += per_cpu(bh_accounting, i).nr;
3379 buffer_heads_over_limit = (tot > max_buffer_heads);
3380}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003381
Al Virodd0fc662005-10-07 07:46:04 +01003382struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003383{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003384 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003385 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003386 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003387 preempt_disable();
3388 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003389 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003390 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003391 }
3392 return ret;
3393}
3394EXPORT_SYMBOL(alloc_buffer_head);
3395
3396void free_buffer_head(struct buffer_head *bh)
3397{
3398 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3399 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003400 preempt_disable();
3401 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003402 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003403 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003404}
3405EXPORT_SYMBOL(free_buffer_head);
3406
Linus Torvalds1da177e2005-04-16 15:20:36 -07003407static void buffer_exit_cpu(int cpu)
3408{
3409 int i;
3410 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3411
3412 for (i = 0; i < BH_LRU_SIZE; i++) {
3413 brelse(b->bhs[i]);
3414 b->bhs[i] = NULL;
3415 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003416 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003417 per_cpu(bh_accounting, cpu).nr = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003418}
3419
3420static int buffer_cpu_notify(struct notifier_block *self,
3421 unsigned long action, void *hcpu)
3422{
Rafael J. Wysocki8bb78442007-05-09 02:35:10 -07003423 if (action == CPU_DEAD || action == CPU_DEAD_FROZEN)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424 buffer_exit_cpu((unsigned long)hcpu);
3425 return NOTIFY_OK;
3426}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003427
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003428/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003429 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003430 * @bh: struct buffer_head
3431 *
3432 * Return true if the buffer is up-to-date and false,
3433 * with the buffer locked, if not.
3434 */
3435int bh_uptodate_or_lock(struct buffer_head *bh)
3436{
3437 if (!buffer_uptodate(bh)) {
3438 lock_buffer(bh);
3439 if (!buffer_uptodate(bh))
3440 return 0;
3441 unlock_buffer(bh);
3442 }
3443 return 1;
3444}
3445EXPORT_SYMBOL(bh_uptodate_or_lock);
3446
3447/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003448 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003449 * @bh: struct buffer_head
3450 *
3451 * Returns zero on success and -EIO on error.
3452 */
3453int bh_submit_read(struct buffer_head *bh)
3454{
3455 BUG_ON(!buffer_locked(bh));
3456
3457 if (buffer_uptodate(bh)) {
3458 unlock_buffer(bh);
3459 return 0;
3460 }
3461
3462 get_bh(bh);
3463 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003464 submit_bh(REQ_OP_READ, 0, bh);
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003465 wait_on_buffer(bh);
3466 if (buffer_uptodate(bh))
3467 return 0;
3468 return -EIO;
3469}
3470EXPORT_SYMBOL(bh_submit_read);
3471
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472void __init buffer_init(void)
3473{
Zhang Yanfei43be5942013-02-22 16:35:46 -08003474 unsigned long nrpages;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003475
Christoph Lameterb98938c2008-02-04 22:28:36 -08003476 bh_cachep = kmem_cache_create("buffer_head",
3477 sizeof(struct buffer_head), 0,
3478 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3479 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003480 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003481
3482 /*
3483 * Limit the bh occupancy to 10% of ZONE_NORMAL
3484 */
3485 nrpages = (nr_free_buffer_pages() * 10) / 100;
3486 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
3487 hotcpu_notifier(buffer_cpu_notify, 0);
3488}