blob: d21771fcf7d345ab4299cb7fa25881ffcc61ef52 [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>
Jan Kara29f3ad72016-11-04 18:08:11 +010046#include <linux/pagevec.h>
Tejun Heo5305cb82013-01-11 13:06:36 -080047#include <trace/events/block.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070048
49static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
Mike Christie2a222ca2016-06-05 14:31:43 -050050static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Tejun Heob16b1de2015-06-02 08:39:48 -060051 unsigned long bio_flags,
52 struct writeback_control *wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070053
54#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
55
Yan Honga3f3c292012-12-12 13:52:15 -080056void init_buffer(struct buffer_head *bh, bh_end_io_t *handler, void *private)
Linus Torvalds1da177e2005-04-16 15:20:36 -070057{
58 bh->b_end_io = handler;
59 bh->b_private = private;
60}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070061EXPORT_SYMBOL(init_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070062
Tejun Heof0059af2013-01-11 13:06:35 -080063inline void touch_buffer(struct buffer_head *bh)
64{
Tejun Heo5305cb82013-01-11 13:06:36 -080065 trace_block_touch_buffer(bh);
Tejun Heof0059af2013-01-11 13:06:35 -080066 mark_page_accessed(bh->b_page);
67}
68EXPORT_SYMBOL(touch_buffer);
69
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080070void __lock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070071{
NeilBrown74316202014-07-07 15:16:04 +100072 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070073}
74EXPORT_SYMBOL(__lock_buffer);
75
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080076void unlock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070077{
Nick Piggin51b07fc2008-10-18 20:27:00 -070078 clear_bit_unlock(BH_Lock, &bh->b_state);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010079 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -070080 wake_up_bit(&bh->b_state, BH_Lock);
81}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070082EXPORT_SYMBOL(unlock_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070083
84/*
Mel Gormanb4597222013-07-03 15:02:05 -070085 * Returns if the page has dirty or writeback buffers. If all the buffers
86 * are unlocked and clean then the PageDirty information is stale. If
87 * any of the pages are locked, it is assumed they are locked for IO.
88 */
89void buffer_check_dirty_writeback(struct page *page,
90 bool *dirty, bool *writeback)
91{
92 struct buffer_head *head, *bh;
93 *dirty = false;
94 *writeback = false;
95
96 BUG_ON(!PageLocked(page));
97
98 if (!page_has_buffers(page))
99 return;
100
101 if (PageWriteback(page))
102 *writeback = true;
103
104 head = page_buffers(page);
105 bh = head;
106 do {
107 if (buffer_locked(bh))
108 *writeback = true;
109
110 if (buffer_dirty(bh))
111 *dirty = true;
112
113 bh = bh->b_this_page;
114 } while (bh != head);
115}
116EXPORT_SYMBOL(buffer_check_dirty_writeback);
117
118/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700119 * Block until a buffer comes unlocked. This doesn't stop it
120 * from becoming locked again - you have to lock it yourself
121 * if you want to preserve its state.
122 */
123void __wait_on_buffer(struct buffer_head * bh)
124{
NeilBrown74316202014-07-07 15:16:04 +1000125 wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700127EXPORT_SYMBOL(__wait_on_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128
129static void
130__clear_page_buffers(struct page *page)
131{
132 ClearPagePrivate(page);
Hugh Dickins4c21e2f2005-10-29 18:16:40 -0700133 set_page_private(page, 0);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300134 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135}
136
Robert Elliottb744c2a2014-10-21 13:55:09 -0600137static void buffer_io_error(struct buffer_head *bh, char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700138{
Robert Elliott432f16e2014-10-21 13:55:11 -0600139 if (!test_bit(BH_Quiet, &bh->b_state))
140 printk_ratelimited(KERN_ERR
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400141 "Buffer I/O error on dev %pg, logical block %llu%s\n",
142 bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143}
144
145/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700146 * End-of-IO handler helper function which does not touch the bh after
147 * unlocking it.
148 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
149 * a race there is benign: unlock_buffer() only use the bh's address for
150 * hashing after unlocking the buffer, so it doesn't actually touch the bh
151 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700152 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700153static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154{
155 if (uptodate) {
156 set_buffer_uptodate(bh);
157 } else {
Christoph Hellwig70246282016-07-19 11:28:41 +0200158 /* This happens, due to failed read-ahead attempts. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700159 clear_buffer_uptodate(bh);
160 }
161 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700162}
163
164/*
165 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
166 * unlock the buffer. This is what ll_rw_block uses too.
167 */
168void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
169{
170 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 put_bh(bh);
172}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700173EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700174
175void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
176{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177 if (uptodate) {
178 set_buffer_uptodate(bh);
179 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600180 buffer_io_error(bh, ", lost sync page write");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700181 set_buffer_write_io_error(bh);
182 clear_buffer_uptodate(bh);
183 }
184 unlock_buffer(bh);
185 put_bh(bh);
186}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700187EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188
189/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190 * Various filesystems appear to want __find_get_block to be non-blocking.
191 * But it's the page lock which protects the buffers. To get around this,
192 * we get exclusion from try_to_free_buffers with the blockdev mapping's
193 * private_lock.
194 *
195 * Hack idea: for the blockdev mapping, i_bufferlist_lock contention
196 * may be quite high. This code could TryLock the page, and if that
197 * succeeds, there is no need to take private_lock. (But if
198 * private_lock is contended then so is mapping->tree_lock).
199 */
200static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800201__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700202{
203 struct inode *bd_inode = bdev->bd_inode;
204 struct address_space *bd_mapping = bd_inode->i_mapping;
205 struct buffer_head *ret = NULL;
206 pgoff_t index;
207 struct buffer_head *bh;
208 struct buffer_head *head;
209 struct page *page;
210 int all_mapped = 1;
211
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300212 index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman2457aec2014-06-04 16:10:31 -0700213 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700214 if (!page)
215 goto out;
216
217 spin_lock(&bd_mapping->private_lock);
218 if (!page_has_buffers(page))
219 goto out_unlock;
220 head = page_buffers(page);
221 bh = head;
222 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700223 if (!buffer_mapped(bh))
224 all_mapped = 0;
225 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700226 ret = bh;
227 get_bh(bh);
228 goto out_unlock;
229 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700230 bh = bh->b_this_page;
231 } while (bh != head);
232
233 /* we might be here because some of the buffers on this page are
234 * not mapped. This is due to various races between
235 * file io on the block device and getblk. It gets dealt with
236 * elsewhere, don't buffer_error if we had some unmapped buffers
237 */
238 if (all_mapped) {
239 printk("__find_get_block_slow() failed. "
240 "block=%llu, b_blocknr=%llu\n",
Badari Pulavarty205f87f2006-03-26 01:38:00 -0800241 (unsigned long long)block,
242 (unsigned long long)bh->b_blocknr);
243 printk("b_state=0x%08lx, b_size=%zu\n",
244 bh->b_state, bh->b_size);
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400245 printk("device %pg blocksize: %d\n", bdev,
Tao Ma72a2ebd2011-10-31 17:09:00 -0700246 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 }
248out_unlock:
249 spin_unlock(&bd_mapping->private_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300250 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251out:
252 return ret;
253}
254
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255/*
Jens Axboe5b0830c2009-09-23 19:37:09 +0200256 * Kick the writeback threads then try to free up some ZONE_NORMAL memory.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700257 */
258static void free_more_memory(void)
259{
Mel Gormanc33d6c02016-05-19 17:14:10 -0700260 struct zoneref *z;
Mel Gorman0e884602008-04-28 02:12:14 -0700261 int nid;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700262
Curt Wohlgemuth0e175a12011-10-07 21:54:10 -0600263 wakeup_flusher_threads(1024, WB_REASON_FREE_MORE_MEM);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700264 yield();
265
Mel Gorman0e884602008-04-28 02:12:14 -0700266 for_each_online_node(nid) {
Mel Gormanc33d6c02016-05-19 17:14:10 -0700267
268 z = first_zones_zonelist(node_zonelist(nid, GFP_NOFS),
269 gfp_zone(GFP_NOFS), NULL);
270 if (z->zone)
Mel Gorman54a6eb52008-04-28 02:12:16 -0700271 try_to_free_pages(node_zonelist(nid, GFP_NOFS), 0,
KAMEZAWA Hiroyuki327c0e92009-03-31 15:23:31 -0700272 GFP_NOFS, NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700273 }
274}
275
276/*
277 * I/O completion handler for block_read_full_page() - pages
278 * which come unlocked at the end of I/O.
279 */
280static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
281{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700282 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700283 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284 struct buffer_head *tmp;
285 struct page *page;
286 int page_uptodate = 1;
287
288 BUG_ON(!buffer_async_read(bh));
289
290 page = bh->b_page;
291 if (uptodate) {
292 set_buffer_uptodate(bh);
293 } else {
294 clear_buffer_uptodate(bh);
Robert Elliott432f16e2014-10-21 13:55:11 -0600295 buffer_io_error(bh, ", async page read");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 SetPageError(page);
297 }
298
299 /*
300 * Be _very_ careful from here on. Bad things can happen if
301 * two buffer heads end IO at almost the same time and both
302 * decide that the page is now completely done.
303 */
Nick Piggina3972202005-07-07 17:56:56 -0700304 first = page_buffers(page);
305 local_irq_save(flags);
306 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700307 clear_buffer_async_read(bh);
308 unlock_buffer(bh);
309 tmp = bh;
310 do {
311 if (!buffer_uptodate(tmp))
312 page_uptodate = 0;
313 if (buffer_async_read(tmp)) {
314 BUG_ON(!buffer_locked(tmp));
315 goto still_busy;
316 }
317 tmp = tmp->b_this_page;
318 } while (tmp != bh);
Nick Piggina3972202005-07-07 17:56:56 -0700319 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
320 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700321
322 /*
323 * If none of the buffers had errors and they are all
324 * uptodate then we can set the page uptodate.
325 */
326 if (page_uptodate && !PageError(page))
327 SetPageUptodate(page);
328 unlock_page(page);
329 return;
330
331still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700332 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
333 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700334 return;
335}
336
337/*
338 * Completion handler for block_write_full_page() - pages which are unlocked
339 * during I/O, and which have PageWriteback cleared upon I/O completion.
340 */
Chris Mason35c80d52009-04-15 13:22:38 -0400341void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700342{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700344 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700345 struct buffer_head *tmp;
346 struct page *page;
347
348 BUG_ON(!buffer_async_write(bh));
349
350 page = bh->b_page;
351 if (uptodate) {
352 set_buffer_uptodate(bh);
353 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600354 buffer_io_error(bh, ", lost async page write");
Michal Hocko5114a972016-10-11 13:56:01 -0700355 mapping_set_error(page->mapping, -EIO);
Jan Kara58ff4072006-10-17 00:10:19 -0700356 set_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700357 clear_buffer_uptodate(bh);
358 SetPageError(page);
359 }
360
Nick Piggina3972202005-07-07 17:56:56 -0700361 first = page_buffers(page);
362 local_irq_save(flags);
363 bit_spin_lock(BH_Uptodate_Lock, &first->b_state);
364
Linus Torvalds1da177e2005-04-16 15:20:36 -0700365 clear_buffer_async_write(bh);
366 unlock_buffer(bh);
367 tmp = bh->b_this_page;
368 while (tmp != bh) {
369 if (buffer_async_write(tmp)) {
370 BUG_ON(!buffer_locked(tmp));
371 goto still_busy;
372 }
373 tmp = tmp->b_this_page;
374 }
Nick Piggina3972202005-07-07 17:56:56 -0700375 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
376 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700377 end_page_writeback(page);
378 return;
379
380still_busy:
Nick Piggina3972202005-07-07 17:56:56 -0700381 bit_spin_unlock(BH_Uptodate_Lock, &first->b_state);
382 local_irq_restore(flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383 return;
384}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700385EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386
387/*
388 * If a page's buffers are under async readin (end_buffer_async_read
389 * completion) then there is a possibility that another thread of
390 * control could lock one of the buffers after it has completed
391 * but while some of the other buffers have not completed. This
392 * locked buffer would confuse end_buffer_async_read() into not unlocking
393 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
394 * that this buffer is not under async I/O.
395 *
396 * The page comes unlocked when it has no locked buffer_async buffers
397 * left.
398 *
399 * PageLocked prevents anyone starting new async I/O reads any of
400 * the buffers.
401 *
402 * PageWriteback is used to prevent simultaneous writeout of the same
403 * page.
404 *
405 * PageLocked prevents anyone from starting writeback of a page which is
406 * under read I/O (PageWriteback is only ever set against a locked page).
407 */
408static void mark_buffer_async_read(struct buffer_head *bh)
409{
410 bh->b_end_io = end_buffer_async_read;
411 set_buffer_async_read(bh);
412}
413
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700414static void mark_buffer_async_write_endio(struct buffer_head *bh,
415 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400416{
417 bh->b_end_io = handler;
418 set_buffer_async_write(bh);
419}
420
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421void mark_buffer_async_write(struct buffer_head *bh)
422{
Chris Mason35c80d52009-04-15 13:22:38 -0400423 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700424}
425EXPORT_SYMBOL(mark_buffer_async_write);
426
427
428/*
429 * fs/buffer.c contains helper functions for buffer-backed address space's
430 * fsync functions. A common requirement for buffer-based filesystems is
431 * that certain data from the backing blockdev needs to be written out for
432 * a successful fsync(). For example, ext2 indirect blocks need to be
433 * written back and waited upon before fsync() returns.
434 *
435 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
436 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
437 * management of a list of dependent buffers at ->i_mapping->private_list.
438 *
439 * Locking is a little subtle: try_to_free_buffers() will remove buffers
440 * from their controlling inode's queue when they are being freed. But
441 * try_to_free_buffers() will be operating against the *blockdev* mapping
442 * at the time, not against the S_ISREG file which depends on those buffers.
443 * So the locking for private_list is via the private_lock in the address_space
444 * which backs the buffers. Which is different from the address_space
445 * against which the buffers are listed. So for a particular address_space,
446 * mapping->private_lock does *not* protect mapping->private_list! In fact,
447 * mapping->private_list will always be protected by the backing blockdev's
448 * ->private_lock.
449 *
450 * Which introduces a requirement: all buffers on an address_space's
451 * ->private_list must be from the same address_space: the blockdev's.
452 *
453 * address_spaces which do not place buffers at ->private_list via these
454 * utility functions are free to use private_lock and private_list for
455 * whatever they want. The only requirement is that list_empty(private_list)
456 * be true at clear_inode() time.
457 *
458 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
459 * filesystems should do that. invalidate_inode_buffers() should just go
460 * BUG_ON(!list_empty).
461 *
462 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
463 * take an address_space, not an inode. And it should be called
464 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
465 * queued up.
466 *
467 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
468 * list if it is already on a list. Because if the buffer is on a list,
469 * it *must* already be on the right one. If not, the filesystem is being
470 * silly. This will save a ton of locking. But first we have to ensure
471 * that buffers are taken *off* the old inode's list when they are freed
472 * (presumably in truncate). That requires careful auditing of all
473 * filesystems (do it inside bforget()). It could also be done by bringing
474 * b_inode back.
475 */
476
477/*
478 * The buffer's backing address_space's private_lock must be held
479 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700480static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700481{
482 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -0700483 WARN_ON(!bh->b_assoc_map);
484 if (buffer_write_io_error(bh))
485 set_bit(AS_EIO, &bh->b_assoc_map->flags);
486 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487}
488
489int inode_has_buffers(struct inode *inode)
490{
491 return !list_empty(&inode->i_data.private_list);
492}
493
494/*
495 * osync is designed to support O_SYNC io. It waits synchronously for
496 * all already-submitted IO to complete, but does not queue any new
497 * writes to the disk.
498 *
499 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
500 * you dirty the buffers, and then use osync_inode_buffers to wait for
501 * completion. Any other dirty buffers which are not yet queued for
502 * write will not be flushed to disk by the osync.
503 */
504static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
505{
506 struct buffer_head *bh;
507 struct list_head *p;
508 int err = 0;
509
510 spin_lock(lock);
511repeat:
512 list_for_each_prev(p, list) {
513 bh = BH_ENTRY(p);
514 if (buffer_locked(bh)) {
515 get_bh(bh);
516 spin_unlock(lock);
517 wait_on_buffer(bh);
518 if (!buffer_uptodate(bh))
519 err = -EIO;
520 brelse(bh);
521 spin_lock(lock);
522 goto repeat;
523 }
524 }
525 spin_unlock(lock);
526 return err;
527}
528
Al Viro01a05b32010-03-23 06:06:58 -0400529static void do_thaw_one(struct super_block *sb, void *unused)
530{
Al Viro01a05b32010-03-23 06:06:58 -0400531 while (sb->s_bdev && !thaw_bdev(sb->s_bdev, sb))
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400532 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
Al Viro01a05b32010-03-23 06:06:58 -0400533}
534
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700535static void do_thaw_all(struct work_struct *work)
Eric Sandeenc2d75432009-03-31 15:23:46 -0700536{
Al Viro01a05b32010-03-23 06:06:58 -0400537 iterate_supers(do_thaw_one, NULL);
Jens Axboe053c5252009-04-08 13:44:08 +0200538 kfree(work);
Eric Sandeenc2d75432009-03-31 15:23:46 -0700539 printk(KERN_WARNING "Emergency Thaw complete\n");
540}
541
542/**
543 * emergency_thaw_all -- forcibly thaw every frozen filesystem
544 *
545 * Used for emergency unfreeze of all filesystems via SysRq
546 */
547void emergency_thaw_all(void)
548{
Jens Axboe053c5252009-04-08 13:44:08 +0200549 struct work_struct *work;
550
551 work = kmalloc(sizeof(*work), GFP_ATOMIC);
552 if (work) {
553 INIT_WORK(work, do_thaw_all);
554 schedule_work(work);
555 }
Eric Sandeenc2d75432009-03-31 15:23:46 -0700556}
557
Linus Torvalds1da177e2005-04-16 15:20:36 -0700558/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800559 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700560 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561 *
562 * Starts I/O against the buffers at mapping->private_list, and waits upon
563 * that I/O.
564 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700565 * Basically, this is a convenience function for fsync().
566 * @mapping is a file or directory which needs those buffers to be written for
567 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 */
569int sync_mapping_buffers(struct address_space *mapping)
570{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800571 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572
573 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
574 return 0;
575
576 return fsync_buffers_list(&buffer_mapping->private_lock,
577 &mapping->private_list);
578}
579EXPORT_SYMBOL(sync_mapping_buffers);
580
581/*
582 * Called when we've recently written block `bblock', and it is known that
583 * `bblock' was for a buffer_boundary() buffer. This means that the block at
584 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
585 * dirty, schedule it for IO. So that indirects merge nicely with their data.
586 */
587void write_boundary_block(struct block_device *bdev,
588 sector_t bblock, unsigned blocksize)
589{
590 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
591 if (bh) {
592 if (buffer_dirty(bh))
Mike Christiedfec8a12016-06-05 14:31:44 -0500593 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 put_bh(bh);
595 }
596}
597
598void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
599{
600 struct address_space *mapping = inode->i_mapping;
601 struct address_space *buffer_mapping = bh->b_page->mapping;
602
603 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800604 if (!mapping->private_data) {
605 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700606 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800607 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800609 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610 spin_lock(&buffer_mapping->private_lock);
611 list_move_tail(&bh->b_assoc_buffers,
612 &mapping->private_list);
Jan Kara58ff4072006-10-17 00:10:19 -0700613 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614 spin_unlock(&buffer_mapping->private_lock);
615 }
616}
617EXPORT_SYMBOL(mark_buffer_dirty_inode);
618
619/*
Nick Piggin787d2212007-07-17 04:03:34 -0700620 * Mark the page dirty, and set it dirty in the radix tree, and mark the inode
621 * dirty.
622 *
623 * If warn is true, then emit a warning if the page is not uptodate and has
624 * not been truncated.
Greg Thelenc4843a72015-05-22 17:13:16 -0400625 *
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700626 * The caller must hold lock_page_memcg().
Nick Piggin787d2212007-07-17 04:03:34 -0700627 */
Greg Thelenc4843a72015-05-22 17:13:16 -0400628static void __set_page_dirty(struct page *page, struct address_space *mapping,
Johannes Weiner62cccb82016-03-15 14:57:22 -0700629 int warn)
Nick Piggin787d2212007-07-17 04:03:34 -0700630{
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800631 unsigned long flags;
632
633 spin_lock_irqsave(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700634 if (page->mapping) { /* Race with truncate? */
635 WARN_ON_ONCE(warn && !PageUptodate(page));
Johannes Weiner62cccb82016-03-15 14:57:22 -0700636 account_page_dirtied(page, mapping);
Nick Piggin787d2212007-07-17 04:03:34 -0700637 radix_tree_tag_set(&mapping->page_tree,
638 page_index(page), PAGECACHE_TAG_DIRTY);
639 }
KOSAKI Motohiro227d53b32014-02-06 12:04:28 -0800640 spin_unlock_irqrestore(&mapping->tree_lock, flags);
Nick Piggin787d2212007-07-17 04:03:34 -0700641}
642
643/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644 * Add a page to the dirty page list.
645 *
646 * It is a sad fact of life that this function is called from several places
647 * deeply under spinlocking. It may not sleep.
648 *
649 * If the page has buffers, the uptodate buffers are set dirty, to preserve
650 * dirty-state coherency between the page and the buffers. It the page does
651 * not have buffers then when they are later attached they will all be set
652 * dirty.
653 *
654 * The buffers are dirtied before the page is dirtied. There's a small race
655 * window in which a writepage caller may see the page cleanness but not the
656 * buffer dirtiness. That's fine. If this code were to set the page dirty
657 * before the buffers, a concurrent writepage caller could clear the page dirty
658 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
659 * page on the dirty page list.
660 *
661 * We use private_lock to lock against try_to_free_buffers while using the
662 * page's buffer list. Also use this to protect against clean buffers being
663 * added to the page after it was set dirty.
664 *
665 * FIXME: may need to call ->reservepage here as well. That's rather up to the
666 * address_space though.
667 */
668int __set_page_dirty_buffers(struct page *page)
669{
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700670 int newly_dirty;
Nick Piggin787d2212007-07-17 04:03:34 -0700671 struct address_space *mapping = page_mapping(page);
Nick Pigginebf7a222006-10-10 04:36:54 +0200672
673 if (unlikely(!mapping))
674 return !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
676 spin_lock(&mapping->private_lock);
677 if (page_has_buffers(page)) {
678 struct buffer_head *head = page_buffers(page);
679 struct buffer_head *bh = head;
680
681 do {
682 set_buffer_dirty(bh);
683 bh = bh->b_this_page;
684 } while (bh != head);
685 }
Greg Thelenc4843a72015-05-22 17:13:16 -0400686 /*
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700687 * Lock out page->mem_cgroup migration to keep PageDirty
688 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -0400689 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700690 lock_page_memcg(page);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700691 newly_dirty = !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 spin_unlock(&mapping->private_lock);
693
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700694 if (newly_dirty)
Johannes Weiner62cccb82016-03-15 14:57:22 -0700695 __set_page_dirty(page, mapping, 1);
Greg Thelenc4843a72015-05-22 17:13:16 -0400696
Johannes Weiner62cccb82016-03-15 14:57:22 -0700697 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400698
699 if (newly_dirty)
700 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
701
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700702 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703}
704EXPORT_SYMBOL(__set_page_dirty_buffers);
705
706/*
707 * Write out and wait upon a list of buffers.
708 *
709 * We have conflicting pressures: we want to make sure that all
710 * initially dirty buffers get waited on, but that any subsequently
711 * dirtied buffers don't. After all, we don't want fsync to last
712 * forever if somebody is actively writing to the file.
713 *
714 * Do this in two main stages: first we copy dirty buffers to a
715 * temporary inode list, queueing the writes as we go. Then we clean
716 * up, waiting for those writes to complete.
717 *
718 * During this second stage, any subsequent updates to the file may end
719 * up refiling the buffer on the original inode's dirty list again, so
720 * there is a chance we will end up with a buffer queued for write but
721 * not yet completed on that list. So, as a final cleanup we go through
722 * the osync code to catch these locked, dirty buffers without requeuing
723 * any newly dirty buffers for write.
724 */
725static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
726{
727 struct buffer_head *bh;
728 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100729 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100731 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732
733 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100734 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
736 spin_lock(lock);
737 while (!list_empty(list)) {
738 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800739 mapping = bh->b_assoc_map;
Jan Kara58ff4072006-10-17 00:10:19 -0700740 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800741 /* Avoid race with mark_buffer_dirty_inode() which does
742 * a lockless check and we rely on seeing the dirty bit */
743 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700744 if (buffer_dirty(bh) || buffer_locked(bh)) {
745 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800746 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700747 if (buffer_dirty(bh)) {
748 get_bh(bh);
749 spin_unlock(lock);
750 /*
751 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200752 * write_dirty_buffer() actually writes the
753 * current contents - it is a noop if I/O is
754 * still in flight on potentially older
755 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700756 */
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600757 write_dirty_buffer(bh, REQ_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200758
759 /*
760 * Kick off IO for the previous mapping. Note
761 * that we will not run the very last mapping,
762 * wait_on_buffer() will do that for us
763 * through sync_buffer().
764 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 brelse(bh);
766 spin_lock(lock);
767 }
768 }
769 }
770
Jens Axboe4ee24912011-03-17 10:51:40 +0100771 spin_unlock(lock);
772 blk_finish_plug(&plug);
773 spin_lock(lock);
774
Linus Torvalds1da177e2005-04-16 15:20:36 -0700775 while (!list_empty(&tmp)) {
776 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700777 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800778 mapping = bh->b_assoc_map;
779 __remove_assoc_queue(bh);
780 /* Avoid race with mark_buffer_dirty_inode() which does
781 * a lockless check and we rely on seeing the dirty bit */
782 smp_mb();
783 if (buffer_dirty(bh)) {
784 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800785 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800786 bh->b_assoc_map = mapping;
787 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 spin_unlock(lock);
789 wait_on_buffer(bh);
790 if (!buffer_uptodate(bh))
791 err = -EIO;
792 brelse(bh);
793 spin_lock(lock);
794 }
795
796 spin_unlock(lock);
797 err2 = osync_buffers_list(lock, list);
798 if (err)
799 return err;
800 else
801 return err2;
802}
803
804/*
805 * Invalidate any and all dirty buffers on a given inode. We are
806 * probably unmounting the fs, but that doesn't mean we have already
807 * done a sync(). Just drop the buffers from the inode list.
808 *
809 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
810 * assumes that all the buffers are against the blockdev. Not true
811 * for reiserfs.
812 */
813void invalidate_inode_buffers(struct inode *inode)
814{
815 if (inode_has_buffers(inode)) {
816 struct address_space *mapping = &inode->i_data;
817 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800818 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700819
820 spin_lock(&buffer_mapping->private_lock);
821 while (!list_empty(list))
822 __remove_assoc_queue(BH_ENTRY(list->next));
823 spin_unlock(&buffer_mapping->private_lock);
824 }
825}
Jan Kara52b19ac2008-09-23 18:24:08 +0200826EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700827
828/*
829 * Remove any clean buffers from the inode's buffer list. This is called
830 * when we're trying to free the inode itself. Those buffers can pin it.
831 *
832 * Returns true if all buffers were removed.
833 */
834int remove_inode_buffers(struct inode *inode)
835{
836 int ret = 1;
837
838 if (inode_has_buffers(inode)) {
839 struct address_space *mapping = &inode->i_data;
840 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800841 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700842
843 spin_lock(&buffer_mapping->private_lock);
844 while (!list_empty(list)) {
845 struct buffer_head *bh = BH_ENTRY(list->next);
846 if (buffer_dirty(bh)) {
847 ret = 0;
848 break;
849 }
850 __remove_assoc_queue(bh);
851 }
852 spin_unlock(&buffer_mapping->private_lock);
853 }
854 return ret;
855}
856
857/*
858 * Create the appropriate buffers when given a page for data area and
859 * the size of each buffer.. Use the bh->b_this_page linked list to
860 * follow the buffers created. Return NULL if unable to create more
861 * buffers.
862 *
863 * The retry flag is used to differentiate async IO (paging, swapping)
864 * which may not fail from ordinary buffer allocations.
865 */
866struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
867 int retry)
868{
869 struct buffer_head *bh, *head;
870 long offset;
871
872try_again:
873 head = NULL;
874 offset = PAGE_SIZE;
875 while ((offset -= size) >= 0) {
876 bh = alloc_buffer_head(GFP_NOFS);
877 if (!bh)
878 goto no_grow;
879
Linus Torvalds1da177e2005-04-16 15:20:36 -0700880 bh->b_this_page = head;
881 bh->b_blocknr = -1;
882 head = bh;
883
Linus Torvalds1da177e2005-04-16 15:20:36 -0700884 bh->b_size = size;
885
886 /* Link the buffer to its page */
887 set_bh_page(bh, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888 }
889 return head;
890/*
891 * In case anything failed, we just free everything we got.
892 */
893no_grow:
894 if (head) {
895 do {
896 bh = head;
897 head = head->b_this_page;
898 free_buffer_head(bh);
899 } while (head);
900 }
901
902 /*
903 * Return failure for non-async IO requests. Async IO requests
904 * are not allowed to fail, so we have to wait until buffer heads
905 * become available. But we don't want tasks sleeping with
906 * partially complete buffers, so all were released above.
907 */
908 if (!retry)
909 return NULL;
910
911 /* We're _really_ low on memory. Now we just
912 * wait for old buffer heads to become free due to
913 * finishing IO. Since this is an async request and
914 * the reserve list is empty, we're sure there are
915 * async buffer heads in use.
916 */
917 free_more_memory();
918 goto try_again;
919}
920EXPORT_SYMBOL_GPL(alloc_page_buffers);
921
922static inline void
923link_dev_buffers(struct page *page, struct buffer_head *head)
924{
925 struct buffer_head *bh, *tail;
926
927 bh = head;
928 do {
929 tail = bh;
930 bh = bh->b_this_page;
931 } while (bh);
932 tail->b_this_page = head;
933 attach_page_buffers(page, head);
934}
935
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800936static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
937{
938 sector_t retval = ~((sector_t)0);
939 loff_t sz = i_size_read(bdev->bd_inode);
940
941 if (sz) {
942 unsigned int sizebits = blksize_bits(size);
943 retval = (sz >> sizebits);
944 }
945 return retval;
946}
947
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948/*
949 * Initialise the state of a blockdev page's buffers.
950 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200951static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700952init_page_buffers(struct page *page, struct block_device *bdev,
953 sector_t block, int size)
954{
955 struct buffer_head *head = page_buffers(page);
956 struct buffer_head *bh = head;
957 int uptodate = PageUptodate(page);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800958 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700959
960 do {
961 if (!buffer_mapped(bh)) {
962 init_buffer(bh, NULL, NULL);
963 bh->b_bdev = bdev;
964 bh->b_blocknr = block;
965 if (uptodate)
966 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200967 if (block < end_block)
968 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 }
970 block++;
971 bh = bh->b_this_page;
972 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200973
974 /*
975 * Caller needs to validate requested block against end of device.
976 */
977 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700978}
979
980/*
981 * Create the page-cache page that contains the requested block.
982 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200983 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200985static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986grow_dev_page(struct block_device *bdev, sector_t block,
Gioh Kim3b5e6452014-09-04 22:04:42 -0400987 pgoff_t index, int size, int sizebits, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988{
989 struct inode *inode = bdev->bd_inode;
990 struct page *page;
991 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200992 sector_t end_block;
993 int ret = 0; /* Will call free_more_memory() */
Johannes Weiner84235de2013-10-16 13:47:00 -0700994 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700995
Michal Hockoc62d2552015-11-06 16:28:49 -0800996 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
Gioh Kim3b5e6452014-09-04 22:04:42 -0400997
Johannes Weiner84235de2013-10-16 13:47:00 -0700998 /*
999 * XXX: __getblk_slow() can not really deal with failure and
1000 * will endlessly loop on improvised global reclaim. Prefer
1001 * looping in the allocator rather than here, at least that
1002 * code knows what it's doing.
1003 */
1004 gfp_mask |= __GFP_NOFAIL;
1005
1006 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001007 if (!page)
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001008 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001009
Eric Sesterhenne827f922006-03-26 18:24:46 +02001010 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
1012 if (page_has_buffers(page)) {
1013 bh = page_buffers(page);
1014 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001015 end_block = init_page_buffers(page, bdev,
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001016 (sector_t)index << sizebits,
1017 size);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001018 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001019 }
1020 if (!try_to_free_buffers(page))
1021 goto failed;
1022 }
1023
1024 /*
1025 * Allocate some buffers for this page
1026 */
1027 bh = alloc_page_buffers(page, size, 0);
1028 if (!bh)
1029 goto failed;
1030
1031 /*
1032 * Link the page to the buffers and initialise them. Take the
1033 * lock to be atomic wrt __find_get_block(), which does not
1034 * run under the page lock.
1035 */
1036 spin_lock(&inode->i_mapping->private_lock);
1037 link_dev_buffers(page, bh);
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001038 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1039 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001040 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001041done:
1042 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001043failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001044 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001045 put_page(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001046 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001047}
1048
1049/*
1050 * Create buffers for the specified block device block's page. If
1051 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001052 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001053static int
Gioh Kim3b5e6452014-09-04 22:04:42 -04001054grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001056 pgoff_t index;
1057 int sizebits;
1058
1059 sizebits = -1;
1060 do {
1061 sizebits++;
1062 } while ((size << sizebits) < PAGE_SIZE);
1063
1064 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065
Andrew Mortone5657932006-10-11 01:21:46 -07001066 /*
1067 * Check for a block which wants to lie outside our maximum possible
1068 * pagecache index. (this comparison is done using sector_t types).
1069 */
1070 if (unlikely(index != block >> sizebits)) {
Andrew Mortone5657932006-10-11 01:21:46 -07001071 printk(KERN_ERR "%s: requested out-of-range block %llu for "
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001072 "device %pg\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001073 __func__, (unsigned long long)block,
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001074 bdev);
Andrew Mortone5657932006-10-11 01:21:46 -07001075 return -EIO;
1076 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001077
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 /* Create a page with the proper size buffers.. */
Gioh Kim3b5e6452014-09-04 22:04:42 -04001079 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080}
1081
Eric Biggers0026ba42016-09-12 13:30:41 -07001082static struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001083__getblk_slow(struct block_device *bdev, sector_t block,
1084 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001085{
1086 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001087 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001088 (size < 512 || size > PAGE_SIZE))) {
1089 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1090 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001091 printk(KERN_ERR "logical block size: %d\n",
1092 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001093
1094 dump_stack();
1095 return NULL;
1096 }
1097
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001098 for (;;) {
1099 struct buffer_head *bh;
1100 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001101
1102 bh = __find_get_block(bdev, block, size);
1103 if (bh)
1104 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001105
Gioh Kim3b5e6452014-09-04 22:04:42 -04001106 ret = grow_buffers(bdev, block, size, gfp);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001107 if (ret < 0)
1108 return NULL;
1109 if (ret == 0)
1110 free_more_memory();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111 }
1112}
1113
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
Jan Kara29f3ad72016-11-04 18:08:11 +01001608/**
1609 * clean_bdev_aliases: clean a range of buffers in block device
1610 * @bdev: Block device to clean buffers in
1611 * @block: Start of a range of blocks to clean
1612 * @len: Number of blocks to clean
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613 *
Jan Kara29f3ad72016-11-04 18:08:11 +01001614 * We are taking a range of blocks for data and we don't want writeback of any
1615 * buffer-cache aliases starting from return from this function and until the
1616 * moment when something will explicitly mark the buffer dirty (hopefully that
1617 * will not happen until we will free that block ;-) We don't even need to mark
1618 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1619 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1620 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1621 * would confuse anyone who might pick it with bread() afterwards...
1622 *
1623 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1624 * writeout I/O going on against recently-freed buffers. We don't wait on that
1625 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1626 * need to. That happens here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 */
Jan Kara29f3ad72016-11-04 18:08:11 +01001628void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001629{
Jan Kara29f3ad72016-11-04 18:08:11 +01001630 struct inode *bd_inode = bdev->bd_inode;
1631 struct address_space *bd_mapping = bd_inode->i_mapping;
1632 struct pagevec pvec;
1633 pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1634 pgoff_t end;
1635 int i;
1636 struct buffer_head *bh;
1637 struct buffer_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638
Jan Kara29f3ad72016-11-04 18:08:11 +01001639 end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
1640 pagevec_init(&pvec, 0);
1641 while (index <= end && pagevec_lookup(&pvec, bd_mapping, index,
1642 min(end - index, (pgoff_t)PAGEVEC_SIZE - 1) + 1)) {
1643 for (i = 0; i < pagevec_count(&pvec); i++) {
1644 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645
Jan Kara29f3ad72016-11-04 18:08:11 +01001646 index = page->index;
1647 if (index > end)
1648 break;
1649 if (!page_has_buffers(page))
1650 continue;
1651 /*
1652 * We use page lock instead of bd_mapping->private_lock
1653 * to pin buffers here since we can afford to sleep and
1654 * it scales better than a global spinlock lock.
1655 */
1656 lock_page(page);
1657 /* Recheck when the page is locked which pins bhs */
1658 if (!page_has_buffers(page))
1659 goto unlock_page;
1660 head = page_buffers(page);
1661 bh = head;
1662 do {
1663 if (!buffer_mapped(bh))
1664 goto next;
1665 if (bh->b_blocknr >= block + len)
1666 break;
1667 clear_buffer_dirty(bh);
1668 wait_on_buffer(bh);
1669 clear_buffer_req(bh);
1670next:
1671 bh = bh->b_this_page;
1672 } while (bh != head);
1673unlock_page:
1674 unlock_page(page);
1675 }
1676 pagevec_release(&pvec);
1677 cond_resched();
1678 index++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001679 }
1680}
Jan Kara29f3ad72016-11-04 18:08:11 +01001681EXPORT_SYMBOL(clean_bdev_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001682
1683/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001684 * Size is a power-of-two in the range 512..PAGE_SIZE,
1685 * and the case we care about most is PAGE_SIZE.
1686 *
1687 * So this *could* possibly be written with those
1688 * constraints in mind (relevant mostly if some
1689 * architecture has a slow bit-scan instruction)
1690 */
1691static inline int block_size_bits(unsigned int blocksize)
1692{
1693 return ilog2(blocksize);
1694}
1695
1696static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1697{
1698 BUG_ON(!PageLocked(page));
1699
1700 if (!page_has_buffers(page))
1701 create_empty_buffers(page, 1 << ACCESS_ONCE(inode->i_blkbits), b_state);
1702 return page_buffers(page);
1703}
1704
1705/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706 * NOTE! All mapped/uptodate combinations are valid:
1707 *
1708 * Mapped Uptodate Meaning
1709 *
1710 * No No "unknown" - must do get_block()
1711 * No Yes "hole" - zero-filled
1712 * Yes No "allocated" - allocated on disk, not read in
1713 * Yes Yes "valid" - allocated and up-to-date in memory.
1714 *
1715 * "Dirty" is valid only with the last case (mapped+uptodate).
1716 */
1717
1718/*
1719 * While block_write_full_page is writing back the dirty buffers under
1720 * the page lock, whoever dirtied the buffers may decide to clean them
1721 * again at any time. We handle that by only looking at the buffer
1722 * state inside lock_buffer().
1723 *
1724 * If block_write_full_page() is called for regular writeback
1725 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1726 * locked buffer. This only can happen if someone has written the buffer
1727 * directly, with submit_bh(). At the address_space level PageWriteback
1728 * prevents this contention from occurring.
Theodore Ts'o6e34eed2009-04-07 18:12:43 -04001729 *
1730 * If block_write_full_page() is called with wbc->sync_mode ==
Christoph Hellwig70fd7612016-11-01 07:40:10 -06001731 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
Jens Axboe721a9602011-03-09 11:56:30 +01001732 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733 */
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001734int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001735 get_block_t *get_block, struct writeback_control *wbc,
1736 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001737{
1738 int err;
1739 sector_t block;
1740 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001741 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001742 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001743 int nr_underway = 0;
Jens Axboe76372412016-11-01 10:00:38 -06001744 int write_flags = wbc_to_write_flags(wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001745
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001746 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748
1749 /*
1750 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1751 * here, and the (potentially unmapped) buffers may become dirty at
1752 * any time. If a buffer becomes dirty here after we've inspected it
1753 * then we just miss that fact, and the page stays dirty.
1754 *
1755 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1756 * handle that here by just cleaning them.
1757 */
1758
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001760 blocksize = bh->b_size;
1761 bbits = block_size_bits(blocksize);
1762
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001763 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001764 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001765
1766 /*
1767 * Get all the dirty buffers mapped to disk addresses and
1768 * handle any aliases from the underlying blockdev's mapping.
1769 */
1770 do {
1771 if (block > last_block) {
1772 /*
1773 * mapped buffers outside i_size will occur, because
1774 * this page can be outside i_size when there is a
1775 * truncate in progress.
1776 */
1777 /*
1778 * The buffer was zeroed by block_write_full_page()
1779 */
1780 clear_buffer_dirty(bh);
1781 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001782 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1783 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001784 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 err = get_block(inode, block, bh, 1);
1786 if (err)
1787 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001788 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789 if (buffer_new(bh)) {
1790 /* blockdev mappings never come here */
1791 clear_buffer_new(bh);
Jan Karae64855c2016-11-04 18:08:15 +01001792 clean_bdev_bh_alias(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793 }
1794 }
1795 bh = bh->b_this_page;
1796 block++;
1797 } while (bh != head);
1798
1799 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800 if (!buffer_mapped(bh))
1801 continue;
1802 /*
1803 * If it's a fully non-blocking write attempt and we cannot
1804 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001805 * potentially cause a busy-wait loop from writeback threads
1806 * and kswapd activity, but those code paths have their own
1807 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001809 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001811 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 redirty_page_for_writepage(wbc, page);
1813 continue;
1814 }
1815 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001816 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001817 } else {
1818 unlock_buffer(bh);
1819 }
1820 } while ((bh = bh->b_this_page) != head);
1821
1822 /*
1823 * The page and its buffers are protected by PageWriteback(), so we can
1824 * drop the bh refcounts early.
1825 */
1826 BUG_ON(PageWriteback(page));
1827 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828
1829 do {
1830 struct buffer_head *next = bh->b_this_page;
1831 if (buffer_async_write(bh)) {
Mike Christie2a222ca2016-06-05 14:31:43 -05001832 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833 nr_underway++;
1834 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 bh = next;
1836 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001837 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838
1839 err = 0;
1840done:
1841 if (nr_underway == 0) {
1842 /*
1843 * The page was marked dirty, but the buffers were
1844 * clean. Someone wrote them back by hand with
1845 * ll_rw_block/submit_bh. A rare case.
1846 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849 /*
1850 * The page and buffer_heads can be released at any time from
1851 * here on.
1852 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 }
1854 return err;
1855
1856recover:
1857 /*
1858 * ENOSPC, or some other error. We may already have added some
1859 * blocks to the file, so we need to write these out to avoid
1860 * exposing stale data.
1861 * The page is currently locked and not marked for writeback
1862 */
1863 bh = head;
1864 /* Recovery: lock and submit the mapped buffers */
1865 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001866 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1867 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001869 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870 } else {
1871 /*
1872 * The buffer may have been set dirty during
1873 * attachment to a dirty page.
1874 */
1875 clear_buffer_dirty(bh);
1876 }
1877 } while ((bh = bh->b_this_page) != head);
1878 SetPageError(page);
1879 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001880 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882 do {
1883 struct buffer_head *next = bh->b_this_page;
1884 if (buffer_async_write(bh)) {
1885 clear_buffer_dirty(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05001886 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, 0, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001887 nr_underway++;
1888 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001889 bh = next;
1890 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001891 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892 goto done;
1893}
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001894EXPORT_SYMBOL(__block_write_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001895
Nick Pigginafddba42007-10-16 01:25:01 -07001896/*
1897 * If a page has any new buffers, zero them out here, and mark them uptodate
1898 * and dirty so they'll be written out (in order to prevent uninitialised
1899 * block data from leaking). And clear the new bit.
1900 */
1901void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1902{
1903 unsigned int block_start, block_end;
1904 struct buffer_head *head, *bh;
1905
1906 BUG_ON(!PageLocked(page));
1907 if (!page_has_buffers(page))
1908 return;
1909
1910 bh = head = page_buffers(page);
1911 block_start = 0;
1912 do {
1913 block_end = block_start + bh->b_size;
1914
1915 if (buffer_new(bh)) {
1916 if (block_end > from && block_start < to) {
1917 if (!PageUptodate(page)) {
1918 unsigned start, size;
1919
1920 start = max(from, block_start);
1921 size = min(to, block_end) - start;
1922
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001923 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001924 set_buffer_uptodate(bh);
1925 }
1926
1927 clear_buffer_new(bh);
1928 mark_buffer_dirty(bh);
1929 }
1930 }
1931
1932 block_start = block_end;
1933 bh = bh->b_this_page;
1934 } while (bh != head);
1935}
1936EXPORT_SYMBOL(page_zero_new_buffers);
1937
Christoph Hellwigae259a92016-06-21 09:23:11 +10001938static void
1939iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
1940 struct iomap *iomap)
1941{
1942 loff_t offset = block << inode->i_blkbits;
1943
1944 bh->b_bdev = iomap->bdev;
1945
1946 /*
1947 * Block points to offset in file we need to map, iomap contains
1948 * the offset at which the map starts. If the map ends before the
1949 * current block, then do not map the buffer and let the caller
1950 * handle it.
1951 */
1952 BUG_ON(offset >= iomap->offset + iomap->length);
1953
1954 switch (iomap->type) {
1955 case IOMAP_HOLE:
1956 /*
1957 * If the buffer is not up to date or beyond the current EOF,
1958 * we need to mark it as new to ensure sub-block zeroing is
1959 * executed if necessary.
1960 */
1961 if (!buffer_uptodate(bh) ||
1962 (offset >= i_size_read(inode)))
1963 set_buffer_new(bh);
1964 break;
1965 case IOMAP_DELALLOC:
1966 if (!buffer_uptodate(bh) ||
1967 (offset >= i_size_read(inode)))
1968 set_buffer_new(bh);
1969 set_buffer_uptodate(bh);
1970 set_buffer_mapped(bh);
1971 set_buffer_delay(bh);
1972 break;
1973 case IOMAP_UNWRITTEN:
1974 /*
1975 * For unwritten regions, we always need to ensure that
1976 * sub-block writes cause the regions in the block we are not
1977 * writing to are zeroed. Set the buffer as new to ensure this.
1978 */
1979 set_buffer_new(bh);
1980 set_buffer_unwritten(bh);
1981 /* FALLTHRU */
1982 case IOMAP_MAPPED:
1983 if (offset >= i_size_read(inode))
1984 set_buffer_new(bh);
1985 bh->b_blocknr = (iomap->blkno >> (inode->i_blkbits - 9)) +
1986 ((offset - iomap->offset) >> inode->i_blkbits);
1987 set_buffer_mapped(bh);
1988 break;
1989 }
1990}
1991
1992int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
1993 get_block_t *get_block, struct iomap *iomap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001995 unsigned from = pos & (PAGE_SIZE - 1);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001996 unsigned to = from + len;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02001997 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998 unsigned block_start, block_end;
1999 sector_t block;
2000 int err = 0;
2001 unsigned blocksize, bbits;
2002 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
2003
2004 BUG_ON(!PageLocked(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002005 BUG_ON(from > PAGE_SIZE);
2006 BUG_ON(to > PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 BUG_ON(from > to);
2008
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002009 head = create_page_buffers(page, inode, 0);
2010 blocksize = head->b_size;
2011 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002013 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002014
2015 for(bh = head, block_start = 0; bh != head || !block_start;
2016 block++, block_start=block_end, bh = bh->b_this_page) {
2017 block_end = block_start + blocksize;
2018 if (block_end <= from || block_start >= to) {
2019 if (PageUptodate(page)) {
2020 if (!buffer_uptodate(bh))
2021 set_buffer_uptodate(bh);
2022 }
2023 continue;
2024 }
2025 if (buffer_new(bh))
2026 clear_buffer_new(bh);
2027 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002028 WARN_ON(bh->b_size != blocksize);
Christoph Hellwigae259a92016-06-21 09:23:11 +10002029 if (get_block) {
2030 err = get_block(inode, block, bh, 1);
2031 if (err)
2032 break;
2033 } else {
2034 iomap_to_bh(inode, block, bh, iomap);
2035 }
2036
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 if (buffer_new(bh)) {
Jan Karae64855c2016-11-04 18:08:15 +01002038 clean_bdev_bh_alias(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002039 if (PageUptodate(page)) {
Nick Piggin637aff42007-10-16 01:25:00 -07002040 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002041 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07002042 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002043 continue;
2044 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002045 if (block_end > to || block_start < from)
2046 zero_user_segments(page,
2047 to, block_end,
2048 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049 continue;
2050 }
2051 }
2052 if (PageUptodate(page)) {
2053 if (!buffer_uptodate(bh))
2054 set_buffer_uptodate(bh);
2055 continue;
2056 }
2057 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08002058 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 (block_start < from || block_end > to)) {
Mike Christiedfec8a12016-06-05 14:31:44 -05002060 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 *wait_bh++=bh;
2062 }
2063 }
2064 /*
2065 * If we issued read requests - let them complete.
2066 */
2067 while(wait_bh > wait) {
2068 wait_on_buffer(*--wait_bh);
2069 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07002070 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 }
Jan Karaf9f07b62011-06-14 00:58:27 +02002072 if (unlikely(err))
Nick Pigginafddba42007-10-16 01:25:01 -07002073 page_zero_new_buffers(page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 return err;
2075}
Christoph Hellwigae259a92016-06-21 09:23:11 +10002076
2077int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2078 get_block_t *get_block)
2079{
2080 return __block_write_begin_int(page, pos, len, get_block, NULL);
2081}
Christoph Hellwigebdec242010-10-06 10:47:23 +02002082EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083
2084static int __block_commit_write(struct inode *inode, struct page *page,
2085 unsigned from, unsigned to)
2086{
2087 unsigned block_start, block_end;
2088 int partial = 0;
2089 unsigned blocksize;
2090 struct buffer_head *bh, *head;
2091
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002092 bh = head = page_buffers(page);
2093 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002094
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002095 block_start = 0;
2096 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002097 block_end = block_start + blocksize;
2098 if (block_end <= from || block_start >= to) {
2099 if (!buffer_uptodate(bh))
2100 partial = 1;
2101 } else {
2102 set_buffer_uptodate(bh);
2103 mark_buffer_dirty(bh);
2104 }
Nick Pigginafddba42007-10-16 01:25:01 -07002105 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002106
2107 block_start = block_end;
2108 bh = bh->b_this_page;
2109 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002110
2111 /*
2112 * If this is a partial write which happened to make all buffers
2113 * uptodate then we can optimize away a bogus readpage() for
2114 * the next read(). Here we 'discover' whether the page went
2115 * uptodate as a result of this (potentially partial) write.
2116 */
2117 if (!partial)
2118 SetPageUptodate(page);
2119 return 0;
2120}
2121
2122/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02002123 * block_write_begin takes care of the basic task of block allocation and
2124 * bringing partial write blocks uptodate first.
2125 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002126 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07002127 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02002128int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2129 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07002130{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002131 pgoff_t index = pos >> PAGE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07002132 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002133 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07002134
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002135 page = grab_cache_page_write_begin(mapping, index, flags);
2136 if (!page)
2137 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07002138
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002139 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07002140 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002141 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002142 put_page(page);
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002143 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07002144 }
2145
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002146 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07002147 return status;
2148}
2149EXPORT_SYMBOL(block_write_begin);
2150
2151int block_write_end(struct file *file, struct address_space *mapping,
2152 loff_t pos, unsigned len, unsigned copied,
2153 struct page *page, void *fsdata)
2154{
2155 struct inode *inode = mapping->host;
2156 unsigned start;
2157
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002158 start = pos & (PAGE_SIZE - 1);
Nick Pigginafddba42007-10-16 01:25:01 -07002159
2160 if (unlikely(copied < len)) {
2161 /*
2162 * The buffers that were written will now be uptodate, so we
2163 * don't have to worry about a readpage reading them and
2164 * overwriting a partial write. However if we have encountered
2165 * a short write and only partially written into a buffer, it
2166 * will not be marked uptodate, so a readpage might come in and
2167 * destroy our partial write.
2168 *
2169 * Do the simplest thing, and just treat any short write to a
2170 * non uptodate page as a zero-length write, and force the
2171 * caller to redo the whole thing.
2172 */
2173 if (!PageUptodate(page))
2174 copied = 0;
2175
2176 page_zero_new_buffers(page, start+copied, start+len);
2177 }
2178 flush_dcache_page(page);
2179
2180 /* This could be a short (even 0-length) commit */
2181 __block_commit_write(inode, page, start, start+copied);
2182
2183 return copied;
2184}
2185EXPORT_SYMBOL(block_write_end);
2186
2187int generic_write_end(struct file *file, struct address_space *mapping,
2188 loff_t pos, unsigned len, unsigned copied,
2189 struct page *page, void *fsdata)
2190{
2191 struct inode *inode = mapping->host;
Jan Kara90a80202014-10-01 21:49:18 -04002192 loff_t old_size = inode->i_size;
Jan Karac7d206b2008-07-11 19:27:31 -04002193 int i_size_changed = 0;
Nick Pigginafddba42007-10-16 01:25:01 -07002194
2195 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
2196
2197 /*
2198 * No need to use i_size_read() here, the i_size
2199 * cannot change under us because we hold i_mutex.
2200 *
2201 * But it's important to update i_size while still holding page lock:
2202 * page writeout could otherwise come in and zero beyond i_size.
2203 */
2204 if (pos+copied > inode->i_size) {
2205 i_size_write(inode, pos+copied);
Jan Karac7d206b2008-07-11 19:27:31 -04002206 i_size_changed = 1;
Nick Pigginafddba42007-10-16 01:25:01 -07002207 }
2208
2209 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002210 put_page(page);
Nick Pigginafddba42007-10-16 01:25:01 -07002211
Jan Kara90a80202014-10-01 21:49:18 -04002212 if (old_size < pos)
2213 pagecache_isize_extended(inode, old_size, pos);
Jan Karac7d206b2008-07-11 19:27:31 -04002214 /*
2215 * Don't mark the inode dirty under page lock. First, it unnecessarily
2216 * makes the holding time of page lock longer. Second, it forces lock
2217 * ordering of page lock and transaction start for journaling
2218 * filesystems.
2219 */
2220 if (i_size_changed)
2221 mark_inode_dirty(inode);
2222
Nick Pigginafddba42007-10-16 01:25:01 -07002223 return copied;
2224}
2225EXPORT_SYMBOL(generic_write_end);
2226
2227/*
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002228 * block_is_partially_uptodate checks whether buffers within a page are
2229 * uptodate or not.
2230 *
2231 * Returns true if all buffers which correspond to a file portion
2232 * we want to read are uptodate.
2233 */
Al Viroc186afb42014-02-02 21:16:54 -05002234int block_is_partially_uptodate(struct page *page, unsigned long from,
2235 unsigned long count)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002236{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002237 unsigned block_start, block_end, blocksize;
2238 unsigned to;
2239 struct buffer_head *bh, *head;
2240 int ret = 1;
2241
2242 if (!page_has_buffers(page))
2243 return 0;
2244
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002245 head = page_buffers(page);
2246 blocksize = head->b_size;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002247 to = min_t(unsigned, PAGE_SIZE - from, count);
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002248 to = from + to;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002249 if (from < blocksize && to > PAGE_SIZE - blocksize)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002250 return 0;
2251
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002252 bh = head;
2253 block_start = 0;
2254 do {
2255 block_end = block_start + blocksize;
2256 if (block_end > from && block_start < to) {
2257 if (!buffer_uptodate(bh)) {
2258 ret = 0;
2259 break;
2260 }
2261 if (block_end >= to)
2262 break;
2263 }
2264 block_start = block_end;
2265 bh = bh->b_this_page;
2266 } while (bh != head);
2267
2268 return ret;
2269}
2270EXPORT_SYMBOL(block_is_partially_uptodate);
2271
2272/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273 * Generic "read page" function for block devices that have the normal
2274 * get_block functionality. This is most of the block device filesystems.
2275 * Reads the page asynchronously --- the unlock_buffer() and
2276 * set/clear_buffer_uptodate() functions propagate buffer state into the
2277 * page struct once IO has completed.
2278 */
2279int block_read_full_page(struct page *page, get_block_t *get_block)
2280{
2281 struct inode *inode = page->mapping->host;
2282 sector_t iblock, lblock;
2283 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002284 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 int nr, i;
2286 int fully_mapped = 1;
2287
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002288 head = create_page_buffers(page, inode, 0);
2289 blocksize = head->b_size;
2290 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002292 iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002293 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002294 bh = head;
2295 nr = 0;
2296 i = 0;
2297
2298 do {
2299 if (buffer_uptodate(bh))
2300 continue;
2301
2302 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002303 int err = 0;
2304
Linus Torvalds1da177e2005-04-16 15:20:36 -07002305 fully_mapped = 0;
2306 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002307 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002308 err = get_block(inode, iblock, bh, 0);
2309 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002310 SetPageError(page);
2311 }
2312 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002313 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002314 if (!err)
2315 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002316 continue;
2317 }
2318 /*
2319 * get_block() might have updated the buffer
2320 * synchronously
2321 */
2322 if (buffer_uptodate(bh))
2323 continue;
2324 }
2325 arr[nr++] = bh;
2326 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2327
2328 if (fully_mapped)
2329 SetPageMappedToDisk(page);
2330
2331 if (!nr) {
2332 /*
2333 * All buffers are uptodate - we can set the page uptodate
2334 * as well. But not if get_block() returned an error.
2335 */
2336 if (!PageError(page))
2337 SetPageUptodate(page);
2338 unlock_page(page);
2339 return 0;
2340 }
2341
2342 /* Stage two: lock the buffers */
2343 for (i = 0; i < nr; i++) {
2344 bh = arr[i];
2345 lock_buffer(bh);
2346 mark_buffer_async_read(bh);
2347 }
2348
2349 /*
2350 * Stage 3: start the IO. Check for uptodateness
2351 * inside the buffer lock in case another process reading
2352 * the underlying blockdev brought it uptodate (the sct fix).
2353 */
2354 for (i = 0; i < nr; i++) {
2355 bh = arr[i];
2356 if (buffer_uptodate(bh))
2357 end_buffer_async_read(bh, 1);
2358 else
Mike Christie2a222ca2016-06-05 14:31:43 -05002359 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002360 }
2361 return 0;
2362}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002363EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364
2365/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002366 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002367 * deal with the hole.
2368 */
Nick Piggin89e10782007-10-16 01:25:07 -07002369int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002370{
2371 struct address_space *mapping = inode->i_mapping;
2372 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002373 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 int err;
2375
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002376 err = inode_newsize_ok(inode, size);
2377 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002378 goto out;
2379
Nick Piggin89e10782007-10-16 01:25:07 -07002380 err = pagecache_write_begin(NULL, mapping, size, 0,
2381 AOP_FLAG_UNINTERRUPTIBLE|AOP_FLAG_CONT_EXPAND,
2382 &page, &fsdata);
2383 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002384 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002385
Nick Piggin89e10782007-10-16 01:25:07 -07002386 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2387 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002388
Linus Torvalds1da177e2005-04-16 15:20:36 -07002389out:
2390 return err;
2391}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002392EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002393
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002394static int cont_expand_zero(struct file *file, struct address_space *mapping,
2395 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002396{
Nick Piggin89e10782007-10-16 01:25:07 -07002397 struct inode *inode = mapping->host;
2398 unsigned blocksize = 1 << inode->i_blkbits;
2399 struct page *page;
2400 void *fsdata;
2401 pgoff_t index, curidx;
2402 loff_t curpos;
2403 unsigned zerofrom, offset, len;
2404 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002405
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002406 index = pos >> PAGE_SHIFT;
2407 offset = pos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002408
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002409 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2410 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002411 if (zerofrom & (blocksize-1)) {
2412 *bytes |= (blocksize-1);
2413 (*bytes)++;
2414 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002415 len = PAGE_SIZE - zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002416
2417 err = pagecache_write_begin(file, mapping, curpos, len,
2418 AOP_FLAG_UNINTERRUPTIBLE,
2419 &page, &fsdata);
2420 if (err)
2421 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002422 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002423 err = pagecache_write_end(file, mapping, curpos, len, len,
2424 page, fsdata);
2425 if (err < 0)
2426 goto out;
2427 BUG_ON(err != len);
2428 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002429
2430 balance_dirty_pages_ratelimited(mapping);
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002431
2432 if (unlikely(fatal_signal_pending(current))) {
2433 err = -EINTR;
2434 goto out;
2435 }
Nick Piggin89e10782007-10-16 01:25:07 -07002436 }
2437
2438 /* page covers the boundary, find the boundary offset */
2439 if (index == curidx) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002440 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002441 /* if we will expand the thing last block will be filled */
2442 if (offset <= zerofrom) {
2443 goto out;
2444 }
2445 if (zerofrom & (blocksize-1)) {
2446 *bytes |= (blocksize-1);
2447 (*bytes)++;
2448 }
2449 len = offset - zerofrom;
2450
2451 err = pagecache_write_begin(file, mapping, curpos, len,
2452 AOP_FLAG_UNINTERRUPTIBLE,
2453 &page, &fsdata);
2454 if (err)
2455 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002456 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002457 err = pagecache_write_end(file, mapping, curpos, len, len,
2458 page, fsdata);
2459 if (err < 0)
2460 goto out;
2461 BUG_ON(err != len);
2462 err = 0;
2463 }
2464out:
2465 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002466}
2467
Linus Torvalds1da177e2005-04-16 15:20:36 -07002468/*
2469 * For moronic filesystems that do not allow holes in file.
2470 * We may have to extend the file.
2471 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002472int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002473 loff_t pos, unsigned len, unsigned flags,
2474 struct page **pagep, void **fsdata,
2475 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002476{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002478 unsigned blocksize = 1 << inode->i_blkbits;
Nick Piggin89e10782007-10-16 01:25:07 -07002479 unsigned zerofrom;
2480 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481
Nick Piggin89e10782007-10-16 01:25:07 -07002482 err = cont_expand_zero(file, mapping, pos, bytes);
2483 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002484 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002485
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002486 zerofrom = *bytes & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002487 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2488 *bytes |= (blocksize-1);
2489 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002490 }
2491
Christoph Hellwig155130a2010-06-04 11:29:58 +02002492 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002493}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002494EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495
Linus Torvalds1da177e2005-04-16 15:20:36 -07002496int block_commit_write(struct page *page, unsigned from, unsigned to)
2497{
2498 struct inode *inode = page->mapping->host;
2499 __block_commit_write(inode,page,from,to);
2500 return 0;
2501}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002502EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503
David Chinner54171692007-07-19 17:39:55 +10002504/*
2505 * block_page_mkwrite() is not allowed to change the file size as it gets
2506 * called from a page fault handler when a page is first dirtied. Hence we must
2507 * be careful to check for EOF conditions here. We set the page up correctly
2508 * for a written page which means we get ENOSPC checking when writing into
2509 * holes and correct delalloc and unwritten extent mapping on filesystems that
2510 * support these features.
2511 *
2512 * We are not allowed to take the i_mutex here so we have to play games to
2513 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002514 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002515 * page lock we can determine safely if the page is beyond EOF. If it is not
2516 * beyond EOF, then the page is guaranteed safe against truncation until we
2517 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002518 *
Jan Kara14da9202012-06-12 16:20:37 +02002519 * Direct callers of this function should protect against filesystem freezing
Ross Zwisler5c500022015-10-13 16:51:02 -06002520 * using sb_start_pagefault() - sb_end_pagefault() functions.
David Chinner54171692007-07-19 17:39:55 +10002521 */
Ross Zwisler5c500022015-10-13 16:51:02 -06002522int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
Jan Kara24da4fa2011-05-24 00:23:34 +02002523 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002524{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002525 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002526 struct inode *inode = file_inode(vma->vm_file);
David Chinner54171692007-07-19 17:39:55 +10002527 unsigned long end;
2528 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002529 int ret;
David Chinner54171692007-07-19 17:39:55 +10002530
2531 lock_page(page);
2532 size = i_size_read(inode);
2533 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002534 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002535 /* We overload EFAULT to mean page got truncated */
2536 ret = -EFAULT;
2537 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002538 }
2539
2540 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002541 if (((page->index + 1) << PAGE_SHIFT) > size)
2542 end = size & ~PAGE_MASK;
David Chinner54171692007-07-19 17:39:55 +10002543 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002544 end = PAGE_SIZE;
David Chinner54171692007-07-19 17:39:55 +10002545
Christoph Hellwigebdec242010-10-06 10:47:23 +02002546 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002547 if (!ret)
2548 ret = block_commit_write(page, 0, end);
2549
Jan Kara24da4fa2011-05-24 00:23:34 +02002550 if (unlikely(ret < 0))
2551 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002552 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002553 wait_for_stable_page(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002554 return 0;
2555out_unlock:
2556 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002557 return ret;
2558}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002559EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
2561/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002562 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 * immediately, while under the page lock. So it needs a special end_io
2564 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002565 */
2566static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2567{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002568 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569}
2570
2571/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002572 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2573 * the page (converting it to circular linked list and taking care of page
2574 * dirty races).
2575 */
2576static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2577{
2578 struct buffer_head *bh;
2579
2580 BUG_ON(!PageLocked(page));
2581
2582 spin_lock(&page->mapping->private_lock);
2583 bh = head;
2584 do {
2585 if (PageDirty(page))
2586 set_buffer_dirty(bh);
2587 if (!bh->b_this_page)
2588 bh->b_this_page = head;
2589 bh = bh->b_this_page;
2590 } while (bh != head);
2591 attach_page_buffers(page, head);
2592 spin_unlock(&page->mapping->private_lock);
2593}
2594
2595/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002596 * On entry, the page is fully not uptodate.
2597 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002598 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002599 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002600int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002601 loff_t pos, unsigned len, unsigned flags,
2602 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002603 get_block_t *get_block)
2604{
Nick Piggin03158cd2007-10-16 01:25:25 -07002605 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002606 const unsigned blkbits = inode->i_blkbits;
2607 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002608 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002609 struct page *page;
2610 pgoff_t index;
2611 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002612 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002613 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002614 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002615 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002616 int ret = 0;
2617 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002618
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002619 index = pos >> PAGE_SHIFT;
2620 from = pos & (PAGE_SIZE - 1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002621 to = from + len;
2622
Nick Piggin54566b22009-01-04 12:00:53 -08002623 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002624 if (!page)
2625 return -ENOMEM;
2626 *pagep = page;
2627 *fsdata = NULL;
2628
2629 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002630 ret = __block_write_begin(page, pos, len, get_block);
2631 if (unlikely(ret))
2632 goto out_release;
2633 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002634 }
Nick Piggina4b06722007-10-16 01:24:48 -07002635
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 if (PageMappedToDisk(page))
2637 return 0;
2638
Nick Piggina4b06722007-10-16 01:24:48 -07002639 /*
2640 * Allocate buffers so that we can keep track of state, and potentially
2641 * attach them to the page if an error occurs. In the common case of
2642 * no error, they will just be freed again without ever being attached
2643 * to the page (which is all OK, because we're under the page lock).
2644 *
2645 * Be careful: the buffer linked list is a NULL terminated one, rather
2646 * than the circular one we're used to.
2647 */
2648 head = alloc_page_buffers(page, blocksize, 0);
Nick Piggin03158cd2007-10-16 01:25:25 -07002649 if (!head) {
2650 ret = -ENOMEM;
2651 goto out_release;
2652 }
Nick Piggina4b06722007-10-16 01:24:48 -07002653
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002654 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655
2656 /*
2657 * We loop across all blocks in the page, whether or not they are
2658 * part of the affected region. This is so we can discover if the
2659 * page is fully mapped-to-disk.
2660 */
Nick Piggina4b06722007-10-16 01:24:48 -07002661 for (block_start = 0, block_in_page = 0, bh = head;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002662 block_start < PAGE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002663 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002664 int create;
2665
Nick Piggina4b06722007-10-16 01:24:48 -07002666 block_end = block_start + blocksize;
2667 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668 create = 1;
2669 if (block_start >= to)
2670 create = 0;
2671 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002672 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673 if (ret)
2674 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002675 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002677 if (buffer_new(bh))
Jan Karae64855c2016-11-04 18:08:15 +01002678 clean_bdev_bh_alias(bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002679 if (PageUptodate(page)) {
2680 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002681 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002682 }
2683 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002684 zero_user_segments(page, block_start, from,
2685 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 continue;
2687 }
Nick Piggina4b06722007-10-16 01:24:48 -07002688 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 continue; /* reiserfs does this */
2690 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002691 lock_buffer(bh);
2692 bh->b_end_io = end_buffer_read_nobh;
Mike Christie2a222ca2016-06-05 14:31:43 -05002693 submit_bh(REQ_OP_READ, 0, bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002694 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695 }
2696 }
2697
2698 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 /*
2700 * The page is locked, so these buffers are protected from
2701 * any VM or truncate activity. Hence we don't need to care
2702 * for the buffer_head refcounts.
2703 */
Nick Piggina4b06722007-10-16 01:24:48 -07002704 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 wait_on_buffer(bh);
2706 if (!buffer_uptodate(bh))
2707 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708 }
2709 if (ret)
2710 goto failed;
2711 }
2712
2713 if (is_mapped_to_disk)
2714 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715
Nick Piggin03158cd2007-10-16 01:25:25 -07002716 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002717
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 return 0;
2719
2720failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002721 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002723 * Error recovery is a bit difficult. We need to zero out blocks that
2724 * were newly allocated, and dirty them to ensure they get written out.
2725 * Buffers need to be attached to the page at this point, otherwise
2726 * the handling of potential IO errors during writeout would be hard
2727 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002729 attach_nobh_buffers(page, head);
2730 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002731
Nick Piggin03158cd2007-10-16 01:25:25 -07002732out_release:
2733 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002734 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002735 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002736
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002737 return ret;
2738}
Nick Piggin03158cd2007-10-16 01:25:25 -07002739EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740
Nick Piggin03158cd2007-10-16 01:25:25 -07002741int nobh_write_end(struct file *file, struct address_space *mapping,
2742 loff_t pos, unsigned len, unsigned copied,
2743 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744{
2745 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002746 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002747 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002748 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002749
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002750 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002751 attach_nobh_buffers(page, head);
2752 if (page_has_buffers(page))
2753 return generic_write_end(file, mapping, pos, len,
2754 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002755
Nick Piggin22c8ca72007-02-20 13:58:09 -08002756 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002757 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002758 if (pos+copied > inode->i_size) {
2759 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760 mark_inode_dirty(inode);
2761 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002762
2763 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002764 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002765
Nick Piggin03158cd2007-10-16 01:25:25 -07002766 while (head) {
2767 bh = head;
2768 head = head->b_this_page;
2769 free_buffer_head(bh);
2770 }
2771
2772 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773}
Nick Piggin03158cd2007-10-16 01:25:25 -07002774EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002775
2776/*
2777 * nobh_writepage() - based on block_full_write_page() except
2778 * that it tries to operate without attaching bufferheads to
2779 * the page.
2780 */
2781int nobh_writepage(struct page *page, get_block_t *get_block,
2782 struct writeback_control *wbc)
2783{
2784 struct inode * const inode = page->mapping->host;
2785 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002786 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788 int ret;
2789
2790 /* Is the page fully inside i_size? */
2791 if (page->index < end_index)
2792 goto out;
2793
2794 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002795 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796 if (page->index >= end_index+1 || !offset) {
2797 /*
2798 * The page may have dirty, unmapped buffers. For example,
2799 * they may have been added in ext3_writepage(). Make them
2800 * freeable here, so the page does not leak.
2801 */
2802#if 0
2803 /* Not really sure about this - do we need this ? */
2804 if (page->mapping->a_ops->invalidatepage)
2805 page->mapping->a_ops->invalidatepage(page, offset);
2806#endif
2807 unlock_page(page);
2808 return 0; /* don't care */
2809 }
2810
2811 /*
2812 * The page straddles i_size. It must be zeroed out on each and every
2813 * writepage invocation because it may be mmapped. "A file is mapped
2814 * in multiples of the page size. For a file that is not a multiple of
2815 * the page size, the remaining memory is zeroed when mapped, and
2816 * writes to that region are not written out to the file."
2817 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002818 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002819out:
2820 ret = mpage_writepage(page, get_block, wbc);
2821 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002822 ret = __block_write_full_page(inode, page, get_block, wbc,
2823 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824 return ret;
2825}
2826EXPORT_SYMBOL(nobh_writepage);
2827
Nick Piggin03158cd2007-10-16 01:25:25 -07002828int nobh_truncate_page(struct address_space *mapping,
2829 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002831 pgoff_t index = from >> PAGE_SHIFT;
2832 unsigned offset = from & (PAGE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002833 unsigned blocksize;
2834 sector_t iblock;
2835 unsigned length, pos;
2836 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002838 struct buffer_head map_bh;
2839 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002840
Nick Piggin03158cd2007-10-16 01:25:25 -07002841 blocksize = 1 << inode->i_blkbits;
2842 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843
Nick Piggin03158cd2007-10-16 01:25:25 -07002844 /* Block boundary? Nothing to do */
2845 if (!length)
2846 return 0;
2847
2848 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002849 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Nick Piggin03158cd2007-10-16 01:25:25 -07002850
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002852 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 if (!page)
2854 goto out;
2855
Nick Piggin03158cd2007-10-16 01:25:25 -07002856 if (page_has_buffers(page)) {
2857has_buffers:
2858 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002859 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002860 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002861 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002862
2863 /* Find the buffer that contains "offset" */
2864 pos = blocksize;
2865 while (offset >= pos) {
2866 iblock++;
2867 pos += blocksize;
2868 }
2869
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002870 map_bh.b_size = blocksize;
2871 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002872 err = get_block(inode, iblock, &map_bh, 0);
2873 if (err)
2874 goto unlock;
2875 /* unmapped? It's a hole - nothing to do */
2876 if (!buffer_mapped(&map_bh))
2877 goto unlock;
2878
2879 /* Ok, it's mapped. Make sure it's up-to-date */
2880 if (!PageUptodate(page)) {
2881 err = mapping->a_ops->readpage(NULL, page);
2882 if (err) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002883 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002884 goto out;
2885 }
2886 lock_page(page);
2887 if (!PageUptodate(page)) {
2888 err = -EIO;
2889 goto unlock;
2890 }
2891 if (page_has_buffers(page))
2892 goto has_buffers;
2893 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002894 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002895 set_page_dirty(page);
2896 err = 0;
2897
2898unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002900 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002901out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002902 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002903}
2904EXPORT_SYMBOL(nobh_truncate_page);
2905
2906int block_truncate_page(struct address_space *mapping,
2907 loff_t from, get_block_t *get_block)
2908{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002909 pgoff_t index = from >> PAGE_SHIFT;
2910 unsigned offset = from & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002911 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002912 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 unsigned length, pos;
2914 struct inode *inode = mapping->host;
2915 struct page *page;
2916 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917 int err;
2918
2919 blocksize = 1 << inode->i_blkbits;
2920 length = offset & (blocksize - 1);
2921
2922 /* Block boundary? Nothing to do */
2923 if (!length)
2924 return 0;
2925
2926 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002927 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
2929 page = grab_cache_page(mapping, index);
2930 err = -ENOMEM;
2931 if (!page)
2932 goto out;
2933
2934 if (!page_has_buffers(page))
2935 create_empty_buffers(page, blocksize, 0);
2936
2937 /* Find the buffer that contains "offset" */
2938 bh = page_buffers(page);
2939 pos = blocksize;
2940 while (offset >= pos) {
2941 bh = bh->b_this_page;
2942 iblock++;
2943 pos += blocksize;
2944 }
2945
2946 err = 0;
2947 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002948 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002949 err = get_block(inode, iblock, bh, 0);
2950 if (err)
2951 goto unlock;
2952 /* unmapped? It's a hole - nothing to do */
2953 if (!buffer_mapped(bh))
2954 goto unlock;
2955 }
2956
2957 /* Ok, it's mapped. Make sure it's up-to-date */
2958 if (PageUptodate(page))
2959 set_buffer_uptodate(bh);
2960
David Chinner33a266d2007-02-12 00:51:41 -08002961 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 err = -EIO;
Mike Christiedfec8a12016-06-05 14:31:44 -05002963 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002964 wait_on_buffer(bh);
2965 /* Uhhuh. Read error. Complain and punt. */
2966 if (!buffer_uptodate(bh))
2967 goto unlock;
2968 }
2969
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002970 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002971 mark_buffer_dirty(bh);
2972 err = 0;
2973
2974unlock:
2975 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002976 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002977out:
2978 return err;
2979}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002980EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002981
2982/*
2983 * The generic ->writepage function for buffer-backed address_spaces
2984 */
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002985int block_write_full_page(struct page *page, get_block_t *get_block,
2986 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987{
2988 struct inode * const inode = page->mapping->host;
2989 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002990 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002992
2993 /* Is the page fully inside i_size? */
2994 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04002995 return __block_write_full_page(inode, page, get_block, wbc,
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002996 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997
2998 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002999 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 if (page->index >= end_index+1 || !offset) {
3001 /*
3002 * The page may have dirty, unmapped buffers. For example,
3003 * they may have been added in ext3_writepage(). Make them
3004 * freeable here, so the page does not leak.
3005 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003006 do_invalidatepage(page, 0, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007 unlock_page(page);
3008 return 0; /* don't care */
3009 }
3010
3011 /*
3012 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05003013 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 * in multiples of the page size. For a file that is not a multiple of
3015 * the page size, the remaining memory is zeroed when mapped, and
3016 * writes to that region are not written out to the file."
3017 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003018 zero_user_segment(page, offset, PAGE_SIZE);
Matthew Wilcox1b938c02014-06-04 16:07:43 -07003019 return __block_write_full_page(inode, page, get_block, wbc,
3020 end_buffer_async_write);
Chris Mason35c80d52009-04-15 13:22:38 -04003021}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003022EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04003023
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
3025 get_block_t *get_block)
3026{
3027 struct buffer_head tmp;
3028 struct inode *inode = mapping->host;
3029 tmp.b_state = 0;
3030 tmp.b_blocknr = 0;
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08003031 tmp.b_size = 1 << inode->i_blkbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032 get_block(inode, block, &tmp, 0);
3033 return tmp.b_blocknr;
3034}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003035EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003036
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003037static void end_bio_bh_io_sync(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038{
3039 struct buffer_head *bh = bio->bi_private;
3040
Jens Axboeb7c44ed2015-07-24 12:37:59 -06003041 if (unlikely(bio_flagged(bio, BIO_QUIET)))
Keith Mannthey08bafc02008-11-25 10:24:35 +01003042 set_bit(BH_Quiet, &bh->b_state);
3043
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003044 bh->b_end_io(bh, !bio->bi_error);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003045 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046}
3047
Linus Torvalds57302e02012-12-04 08:25:11 -08003048/*
3049 * This allows us to do IO even on the odd last sectors
Akinobu Mita59d43912014-10-09 15:26:53 -07003050 * of a device, even if the block size is some multiple
Linus Torvalds57302e02012-12-04 08:25:11 -08003051 * of the physical sector size.
3052 *
3053 * We'll just truncate the bio to the size of the device,
3054 * and clear the end of the buffer head manually.
3055 *
3056 * Truly out-of-range accesses will turn into actual IO
3057 * errors, this only handles the "we need to be able to
3058 * do IO at the final sector" case.
3059 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003060void guard_bio_eod(int op, struct bio *bio)
Linus Torvalds57302e02012-12-04 08:25:11 -08003061{
3062 sector_t maxsector;
Akinobu Mita59d43912014-10-09 15:26:53 -07003063 struct bio_vec *bvec = &bio->bi_io_vec[bio->bi_vcnt - 1];
3064 unsigned truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08003065
3066 maxsector = i_size_read(bio->bi_bdev->bd_inode) >> 9;
3067 if (!maxsector)
3068 return;
3069
3070 /*
3071 * If the *whole* IO is past the end of the device,
3072 * let it through, and the IO layer will turn it into
3073 * an EIO.
3074 */
Kent Overstreet4f024f32013-10-11 15:44:27 -07003075 if (unlikely(bio->bi_iter.bi_sector >= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08003076 return;
3077
Kent Overstreet4f024f32013-10-11 15:44:27 -07003078 maxsector -= bio->bi_iter.bi_sector;
Akinobu Mita59d43912014-10-09 15:26:53 -07003079 if (likely((bio->bi_iter.bi_size >> 9) <= maxsector))
Linus Torvalds57302e02012-12-04 08:25:11 -08003080 return;
3081
Akinobu Mita59d43912014-10-09 15:26:53 -07003082 /* Uhhuh. We've got a bio that straddles the device size! */
3083 truncated_bytes = bio->bi_iter.bi_size - (maxsector << 9);
Linus Torvalds57302e02012-12-04 08:25:11 -08003084
3085 /* Truncate the bio.. */
Akinobu Mita59d43912014-10-09 15:26:53 -07003086 bio->bi_iter.bi_size -= truncated_bytes;
3087 bvec->bv_len -= truncated_bytes;
Linus Torvalds57302e02012-12-04 08:25:11 -08003088
3089 /* ..and clear the end of the buffer for reads */
Mike Christie2a222ca2016-06-05 14:31:43 -05003090 if (op == REQ_OP_READ) {
Akinobu Mita59d43912014-10-09 15:26:53 -07003091 zero_user(bvec->bv_page, bvec->bv_offset + bvec->bv_len,
3092 truncated_bytes);
Linus Torvalds57302e02012-12-04 08:25:11 -08003093 }
3094}
3095
Mike Christie2a222ca2016-06-05 14:31:43 -05003096static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Tejun Heob16b1de2015-06-02 08:39:48 -06003097 unsigned long bio_flags, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098{
3099 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003100
3101 BUG_ON(!buffer_locked(bh));
3102 BUG_ON(!buffer_mapped(bh));
3103 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04003104 BUG_ON(buffer_delay(bh));
3105 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106
Jens Axboe48fd4f92008-08-22 10:00:36 +02003107 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02003108 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003110 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003111 clear_buffer_write_io_error(bh);
3112
3113 /*
3114 * from here on down, it's all bio -- do the initial mapping,
3115 * submit_bio -> generic_make_request may further map this bio around
3116 */
3117 bio = bio_alloc(GFP_NOIO, 1);
3118
Tejun Heo2a814902015-05-28 14:50:51 -04003119 if (wbc) {
Tejun Heob16b1de2015-06-02 08:39:48 -06003120 wbc_init_bio(wbc, bio);
Tejun Heo2a814902015-05-28 14:50:51 -04003121 wbc_account_io(wbc, bh->b_page, bh->b_size);
3122 }
Tejun Heobafc0db2015-06-02 08:37:23 -06003123
Kent Overstreet4f024f32013-10-11 15:44:27 -07003124 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003125 bio->bi_bdev = bh->b_bdev;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003126
Kent Overstreet6cf66b42014-12-22 12:48:42 +01003127 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3128 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129
3130 bio->bi_end_io = end_bio_bh_io_sync;
3131 bio->bi_private = bh;
Darrick J. Wong7136851112013-04-29 15:07:25 -07003132 bio->bi_flags |= bio_flags;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003133
Linus Torvalds57302e02012-12-04 08:25:11 -08003134 /* Take care of bh's that straddle the end of the device */
Mike Christie2a222ca2016-06-05 14:31:43 -05003135 guard_bio_eod(op, bio);
Linus Torvalds57302e02012-12-04 08:25:11 -08003136
Theodore Ts'o877f9622013-04-20 19:58:37 -04003137 if (buffer_meta(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003138 op_flags |= REQ_META;
Theodore Ts'o877f9622013-04-20 19:58:37 -04003139 if (buffer_prio(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003140 op_flags |= REQ_PRIO;
3141 bio_set_op_attrs(bio, op, op_flags);
Theodore Ts'o877f9622013-04-20 19:58:37 -04003142
Mike Christie4e49ea42016-06-05 14:31:41 -05003143 submit_bio(bio);
Julia Lawallf6454b02015-05-26 21:59:53 +02003144 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145}
Tejun Heobafc0db2015-06-02 08:37:23 -06003146
Mike Christie2a222ca2016-06-05 14:31:43 -05003147int _submit_bh(int op, int op_flags, struct buffer_head *bh,
3148 unsigned long bio_flags)
Tejun Heobafc0db2015-06-02 08:37:23 -06003149{
Mike Christie2a222ca2016-06-05 14:31:43 -05003150 return submit_bh_wbc(op, op_flags, bh, bio_flags, NULL);
Tejun Heobafc0db2015-06-02 08:37:23 -06003151}
Darrick J. Wong7136851112013-04-29 15:07:25 -07003152EXPORT_SYMBOL_GPL(_submit_bh);
3153
Mike Christie2a222ca2016-06-05 14:31:43 -05003154int submit_bh(int op, int op_flags, struct buffer_head *bh)
Darrick J. Wong7136851112013-04-29 15:07:25 -07003155{
Mike Christie2a222ca2016-06-05 14:31:43 -05003156 return submit_bh_wbc(op, op_flags, bh, 0, NULL);
Darrick J. Wong7136851112013-04-29 15:07:25 -07003157}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003158EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003159
3160/**
3161 * ll_rw_block: low-level access to block devices (DEPRECATED)
Mike Christiedfec8a12016-06-05 14:31:44 -05003162 * @op: whether to %READ or %WRITE
Christoph Hellwigef295ec2016-10-28 08:48:16 -06003163 * @op_flags: req_flag_bits
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 * @nr: number of &struct buffer_heads in the array
3165 * @bhs: array of pointers to &struct buffer_head
3166 *
Jan Karaa7662232005-09-06 15:19:10 -07003167 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
Christoph Hellwig70246282016-07-19 11:28:41 +02003168 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3169 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3170 * %REQ_RAHEAD.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003171 *
3172 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003173 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3174 * request, and any buffer that appears to be up-to-date when doing read
3175 * request. Further it marks as clean buffers that are processed for
3176 * writing (the buffer cache won't assume that they are actually clean
3177 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003178 *
3179 * ll_rw_block sets b_end_io to simple completion handler that marks
Masanari Iidae2278672014-02-18 22:54:36 +09003180 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
Linus Torvalds1da177e2005-04-16 15:20:36 -07003181 * any waiters.
3182 *
3183 * All of the buffers must be for the same device, and must also be a
3184 * multiple of the current approved size for the device.
3185 */
Mike Christiedfec8a12016-06-05 14:31:44 -05003186void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187{
3188 int i;
3189
3190 for (i = 0; i < nr; i++) {
3191 struct buffer_head *bh = bhs[i];
3192
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003193 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194 continue;
Mike Christiedfec8a12016-06-05 14:31:44 -05003195 if (op == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003196 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003197 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003198 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003199 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 continue;
3201 }
3202 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003204 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003205 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003206 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003207 continue;
3208 }
3209 }
3210 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 }
3212}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003213EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003214
Mike Christie2a222ca2016-06-05 14:31:43 -05003215void write_dirty_buffer(struct buffer_head *bh, int op_flags)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003216{
3217 lock_buffer(bh);
3218 if (!test_clear_buffer_dirty(bh)) {
3219 unlock_buffer(bh);
3220 return;
3221 }
3222 bh->b_end_io = end_buffer_write_sync;
3223 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05003224 submit_bh(REQ_OP_WRITE, op_flags, bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003225}
3226EXPORT_SYMBOL(write_dirty_buffer);
3227
Linus Torvalds1da177e2005-04-16 15:20:36 -07003228/*
3229 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3230 * and then start new I/O and then wait upon it. The caller must have a ref on
3231 * the buffer_head.
3232 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003233int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003234{
3235 int ret = 0;
3236
3237 WARN_ON(atomic_read(&bh->b_count) < 1);
3238 lock_buffer(bh);
3239 if (test_clear_buffer_dirty(bh)) {
3240 get_bh(bh);
3241 bh->b_end_io = end_buffer_write_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003242 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003243 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003244 if (!ret && !buffer_uptodate(bh))
3245 ret = -EIO;
3246 } else {
3247 unlock_buffer(bh);
3248 }
3249 return ret;
3250}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003251EXPORT_SYMBOL(__sync_dirty_buffer);
3252
3253int sync_dirty_buffer(struct buffer_head *bh)
3254{
Christoph Hellwig70fd7612016-11-01 07:40:10 -06003255 return __sync_dirty_buffer(bh, REQ_SYNC);
Christoph Hellwig87e99512010-08-11 17:05:45 +02003256}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003257EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003258
3259/*
3260 * try_to_free_buffers() checks if all the buffers on this particular page
3261 * are unused, and releases them if so.
3262 *
3263 * Exclusion against try_to_free_buffers may be obtained by either
3264 * locking the page or by holding its mapping's private_lock.
3265 *
3266 * If the page is dirty but all the buffers are clean then we need to
3267 * be sure to mark the page clean as well. This is because the page
3268 * may be against a block device, and a later reattachment of buffers
3269 * to a dirty page will set *all* buffers dirty. Which would corrupt
3270 * filesystem data on the same device.
3271 *
3272 * The same applies to regular filesystem pages: if all the buffers are
3273 * clean then we set the page clean and proceed. To do that, we require
3274 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3275 * private_lock.
3276 *
3277 * try_to_free_buffers() is non-blocking.
3278 */
3279static inline int buffer_busy(struct buffer_head *bh)
3280{
3281 return atomic_read(&bh->b_count) |
3282 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3283}
3284
3285static int
3286drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3287{
3288 struct buffer_head *head = page_buffers(page);
3289 struct buffer_head *bh;
3290
3291 bh = head;
3292 do {
akpm@osdl.orgde7d5a32005-05-01 08:58:39 -07003293 if (buffer_write_io_error(bh) && page->mapping)
Michal Hocko5114a972016-10-11 13:56:01 -07003294 mapping_set_error(page->mapping, -EIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003295 if (buffer_busy(bh))
3296 goto failed;
3297 bh = bh->b_this_page;
3298 } while (bh != head);
3299
3300 do {
3301 struct buffer_head *next = bh->b_this_page;
3302
Jan Kara535ee2f2008-02-08 04:21:59 -08003303 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304 __remove_assoc_queue(bh);
3305 bh = next;
3306 } while (bh != head);
3307 *buffers_to_free = head;
3308 __clear_page_buffers(page);
3309 return 1;
3310failed:
3311 return 0;
3312}
3313
3314int try_to_free_buffers(struct page *page)
3315{
3316 struct address_space * const mapping = page->mapping;
3317 struct buffer_head *buffers_to_free = NULL;
3318 int ret = 0;
3319
3320 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003321 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 return 0;
3323
3324 if (mapping == NULL) { /* can this still happen? */
3325 ret = drop_buffers(page, &buffers_to_free);
3326 goto out;
3327 }
3328
3329 spin_lock(&mapping->private_lock);
3330 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003331
3332 /*
3333 * If the filesystem writes its buffers by hand (eg ext3)
3334 * then we can have clean buffers against a dirty page. We
3335 * clean the page here; otherwise the VM will never notice
3336 * that the filesystem did any IO at all.
3337 *
3338 * Also, during truncate, discard_buffer will have marked all
3339 * the page's buffers clean. We discover that here and clean
3340 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003341 *
3342 * private_lock must be held over this entire operation in order
3343 * to synchronise against __set_page_dirty_buffers and prevent the
3344 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003345 */
Tejun Heo11f81be2015-05-22 17:13:15 -04003346 if (ret)
3347 cancel_dirty_page(page);
Nick Piggin87df7242007-01-30 14:36:27 +11003348 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003349out:
3350 if (buffers_to_free) {
3351 struct buffer_head *bh = buffers_to_free;
3352
3353 do {
3354 struct buffer_head *next = bh->b_this_page;
3355 free_buffer_head(bh);
3356 bh = next;
3357 } while (bh != buffers_to_free);
3358 }
3359 return ret;
3360}
3361EXPORT_SYMBOL(try_to_free_buffers);
3362
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363/*
3364 * There are no bdflush tunables left. But distributions are
3365 * still running obsolete flush daemons, so we terminate them here.
3366 *
3367 * Use of bdflush() is deprecated and will be removed in a future kernel.
Jens Axboe5b0830c2009-09-23 19:37:09 +02003368 * The `flush-X' kernel threads fully replace bdflush daemons and this call.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003369 */
Heiko Carstensbdc480e2009-01-14 14:14:12 +01003370SYSCALL_DEFINE2(bdflush, int, func, long, data)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003371{
3372 static int msg_count;
3373
3374 if (!capable(CAP_SYS_ADMIN))
3375 return -EPERM;
3376
3377 if (msg_count < 5) {
3378 msg_count++;
3379 printk(KERN_INFO
3380 "warning: process `%s' used the obsolete bdflush"
3381 " system call\n", current->comm);
3382 printk(KERN_INFO "Fix your initscripts?\n");
3383 }
3384
3385 if (func == 1)
3386 do_exit(0);
3387 return 0;
3388}
3389
3390/*
3391 * Buffer-head allocation
3392 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003393static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003394
3395/*
3396 * Once the number of bh's in the machine exceeds this level, we start
3397 * stripping them in writeback.
3398 */
Zhang Yanfei43be5942013-02-22 16:35:46 -08003399static unsigned long max_buffer_heads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003400
3401int buffer_heads_over_limit;
3402
3403struct bh_accounting {
3404 int nr; /* Number of live bh's */
3405 int ratelimit; /* Limit cacheline bouncing */
3406};
3407
3408static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3409
3410static void recalc_bh_state(void)
3411{
3412 int i;
3413 int tot = 0;
3414
Christoph Lameteree1be862010-12-06 11:40:05 -06003415 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003416 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003417 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003418 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003419 tot += per_cpu(bh_accounting, i).nr;
3420 buffer_heads_over_limit = (tot > max_buffer_heads);
3421}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003422
Al Virodd0fc662005-10-07 07:46:04 +01003423struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003424{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003425 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003426 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003427 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003428 preempt_disable();
3429 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003430 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003431 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003432 }
3433 return ret;
3434}
3435EXPORT_SYMBOL(alloc_buffer_head);
3436
3437void free_buffer_head(struct buffer_head *bh)
3438{
3439 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3440 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003441 preempt_disable();
3442 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003443 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003444 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445}
3446EXPORT_SYMBOL(free_buffer_head);
3447
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003448static int buffer_exit_cpu_dead(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003449{
3450 int i;
3451 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3452
3453 for (i = 0; i < BH_LRU_SIZE; i++) {
3454 brelse(b->bhs[i]);
3455 b->bhs[i] = NULL;
3456 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003457 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003458 per_cpu(bh_accounting, cpu).nr = 0;
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003459 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003461
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003462/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003463 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003464 * @bh: struct buffer_head
3465 *
3466 * Return true if the buffer is up-to-date and false,
3467 * with the buffer locked, if not.
3468 */
3469int bh_uptodate_or_lock(struct buffer_head *bh)
3470{
3471 if (!buffer_uptodate(bh)) {
3472 lock_buffer(bh);
3473 if (!buffer_uptodate(bh))
3474 return 0;
3475 unlock_buffer(bh);
3476 }
3477 return 1;
3478}
3479EXPORT_SYMBOL(bh_uptodate_or_lock);
3480
3481/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003482 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003483 * @bh: struct buffer_head
3484 *
3485 * Returns zero on success and -EIO on error.
3486 */
3487int bh_submit_read(struct buffer_head *bh)
3488{
3489 BUG_ON(!buffer_locked(bh));
3490
3491 if (buffer_uptodate(bh)) {
3492 unlock_buffer(bh);
3493 return 0;
3494 }
3495
3496 get_bh(bh);
3497 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003498 submit_bh(REQ_OP_READ, 0, bh);
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003499 wait_on_buffer(bh);
3500 if (buffer_uptodate(bh))
3501 return 0;
3502 return -EIO;
3503}
3504EXPORT_SYMBOL(bh_submit_read);
3505
Linus Torvalds1da177e2005-04-16 15:20:36 -07003506void __init buffer_init(void)
3507{
Zhang Yanfei43be5942013-02-22 16:35:46 -08003508 unsigned long nrpages;
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003509 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003510
Christoph Lameterb98938c2008-02-04 22:28:36 -08003511 bh_cachep = kmem_cache_create("buffer_head",
3512 sizeof(struct buffer_head), 0,
3513 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3514 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003515 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003516
3517 /*
3518 * Limit the bh occupancy to 10% of ZONE_NORMAL
3519 */
3520 nrpages = (nr_free_buffer_pages() * 10) / 100;
3521 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003522 ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3523 NULL, buffer_exit_cpu_dead);
3524 WARN_ON(ret < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003525}