blob: 2b5561ae5d0b32b038131a77db6073a0eccddb7f [file] [log] [blame]
Thomas Gleixner457c8992019-05-19 13:08:55 +01001// SPDX-License-Identifier: GPL-2.0-only
Linus Torvalds1da177e2005-04-16 15:20:36 -07002/*
3 * linux/fs/buffer.c
4 *
5 * Copyright (C) 1991, 1992, 2002 Linus Torvalds
6 */
7
8/*
9 * Start bdflush() with kernel_thread not syscall - Paul Gortmaker, 12/95
10 *
11 * Removed a lot of unnecessary code and simplified things now that
12 * the buffer cache isn't our primary cache - Andrew Tridgell 12/96
13 *
14 * Speed up hash, lru, and free list operations. Use gfp() for allocating
15 * hash table, use SLAB cache for buffer heads. SMP threading. -DaveM
16 *
17 * Added 32k buffer block sizes - these are required older ARM systems. - RMK
18 *
19 * async buffer flushing, 1999 Andrea Arcangeli <andrea@suse.de>
20 */
21
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/kernel.h>
Ingo Molnarf361bf42017-02-03 23:47:37 +010023#include <linux/sched/signal.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070024#include <linux/syscalls.h>
25#include <linux/fs.h>
Christoph Hellwigae259a92016-06-21 09:23:11 +100026#include <linux/iomap.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/mm.h>
28#include <linux/percpu.h>
29#include <linux/slab.h>
Randy Dunlap16f7e0f2006-01-11 12:17:46 -080030#include <linux/capability.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <linux/blkdev.h>
32#include <linux/file.h>
33#include <linux/quotaops.h>
34#include <linux/highmem.h>
Paul Gortmaker630d9c42011-11-16 23:57:37 -050035#include <linux/export.h>
Tejun Heobafc0db2015-06-02 08:37:23 -060036#include <linux/backing-dev.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037#include <linux/writeback.h>
38#include <linux/hash.h>
39#include <linux/suspend.h>
40#include <linux/buffer_head.h>
Andrew Morton55e829a2006-12-10 02:19:27 -080041#include <linux/task_io_accounting_ops.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070042#include <linux/bio.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070043#include <linux/cpu.h>
44#include <linux/bitops.h>
45#include <linux/mpage.h>
Ingo Molnarfb1c8f92005-09-10 00:25:56 -070046#include <linux/bit_spinlock.h>
Jan Kara29f3ad72016-11-04 18:08:11 +010047#include <linux/pagevec.h>
Shakeel Buttf745c6f2018-08-17 15:46:44 -070048#include <linux/sched/mm.h>
Tejun Heo5305cb82013-01-11 13:06:36 -080049#include <trace/events/block.h>
Eric Biggers31fb9922019-10-22 20:33:11 -070050#include <linux/fscrypt.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070051
Ben Dooks2b211dc2019-11-30 17:49:18 -080052#include "internal.h"
53
Linus Torvalds1da177e2005-04-16 15:20:36 -070054static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
Mike Christie2a222ca2016-06-05 14:31:43 -050055static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Christoph Hellwigc75e7072022-03-04 18:55:56 +010056 struct writeback_control *wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070057
58#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
59
Tejun Heof0059af2013-01-11 13:06:35 -080060inline void touch_buffer(struct buffer_head *bh)
61{
Tejun Heo5305cb82013-01-11 13:06:36 -080062 trace_block_touch_buffer(bh);
Tejun Heof0059af2013-01-11 13:06:35 -080063 mark_page_accessed(bh->b_page);
64}
65EXPORT_SYMBOL(touch_buffer);
66
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080067void __lock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070068{
NeilBrown74316202014-07-07 15:16:04 +100069 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070070}
71EXPORT_SYMBOL(__lock_buffer);
72
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080073void unlock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070074{
Nick Piggin51b07fc2008-10-18 20:27:00 -070075 clear_bit_unlock(BH_Lock, &bh->b_state);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010076 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -070077 wake_up_bit(&bh->b_state, BH_Lock);
78}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070079EXPORT_SYMBOL(unlock_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070080
81/*
Mel Gormanb4597222013-07-03 15:02:05 -070082 * Returns if the page has dirty or writeback buffers. If all the buffers
83 * are unlocked and clean then the PageDirty information is stale. If
84 * any of the pages are locked, it is assumed they are locked for IO.
85 */
86void buffer_check_dirty_writeback(struct page *page,
87 bool *dirty, bool *writeback)
88{
89 struct buffer_head *head, *bh;
90 *dirty = false;
91 *writeback = false;
92
93 BUG_ON(!PageLocked(page));
94
95 if (!page_has_buffers(page))
96 return;
97
98 if (PageWriteback(page))
99 *writeback = true;
100
101 head = page_buffers(page);
102 bh = head;
103 do {
104 if (buffer_locked(bh))
105 *writeback = true;
106
107 if (buffer_dirty(bh))
108 *dirty = true;
109
110 bh = bh->b_this_page;
111 } while (bh != head);
112}
113EXPORT_SYMBOL(buffer_check_dirty_writeback);
114
115/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700116 * Block until a buffer comes unlocked. This doesn't stop it
117 * from becoming locked again - you have to lock it yourself
118 * if you want to preserve its state.
119 */
120void __wait_on_buffer(struct buffer_head * bh)
121{
NeilBrown74316202014-07-07 15:16:04 +1000122 wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700123}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700124EXPORT_SYMBOL(__wait_on_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700125
Robert Elliottb744c2a2014-10-21 13:55:09 -0600126static void buffer_io_error(struct buffer_head *bh, char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700127{
Robert Elliott432f16e2014-10-21 13:55:11 -0600128 if (!test_bit(BH_Quiet, &bh->b_state))
129 printk_ratelimited(KERN_ERR
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400130 "Buffer I/O error on dev %pg, logical block %llu%s\n",
131 bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700132}
133
134/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700135 * End-of-IO handler helper function which does not touch the bh after
136 * unlocking it.
137 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
138 * a race there is benign: unlock_buffer() only use the bh's address for
139 * hashing after unlocking the buffer, so it doesn't actually touch the bh
140 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700141 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700142static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700143{
144 if (uptodate) {
145 set_buffer_uptodate(bh);
146 } else {
Christoph Hellwig70246282016-07-19 11:28:41 +0200147 /* This happens, due to failed read-ahead attempts. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148 clear_buffer_uptodate(bh);
149 }
150 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700151}
152
153/*
154 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
155 * unlock the buffer. This is what ll_rw_block uses too.
156 */
157void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
158{
159 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160 put_bh(bh);
161}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700162EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700163
164void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
165{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166 if (uptodate) {
167 set_buffer_uptodate(bh);
168 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600169 buffer_io_error(bh, ", lost sync page write");
Jeff Layton87354e52017-07-06 07:02:21 -0400170 mark_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171 clear_buffer_uptodate(bh);
172 }
173 unlock_buffer(bh);
174 put_bh(bh);
175}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700176EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700177
178/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700179 * Various filesystems appear to want __find_get_block to be non-blocking.
180 * But it's the page lock which protects the buffers. To get around this,
181 * we get exclusion from try_to_free_buffers with the blockdev mapping's
182 * private_lock.
183 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700184 * Hack idea: for the blockdev mapping, private_lock contention
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185 * may be quite high. This code could TryLock the page, and if that
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700186 * succeeds, there is no need to take private_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700187 */
188static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800189__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700190{
191 struct inode *bd_inode = bdev->bd_inode;
192 struct address_space *bd_mapping = bd_inode->i_mapping;
193 struct buffer_head *ret = NULL;
194 pgoff_t index;
195 struct buffer_head *bh;
196 struct buffer_head *head;
197 struct page *page;
198 int all_mapped = 1;
Tetsuo Handa43636c82019-01-21 22:49:37 +0900199 static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700200
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300201 index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman2457aec2014-06-04 16:10:31 -0700202 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700203 if (!page)
204 goto out;
205
206 spin_lock(&bd_mapping->private_lock);
207 if (!page_has_buffers(page))
208 goto out_unlock;
209 head = page_buffers(page);
210 bh = head;
211 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700212 if (!buffer_mapped(bh))
213 all_mapped = 0;
214 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700215 ret = bh;
216 get_bh(bh);
217 goto out_unlock;
218 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700219 bh = bh->b_this_page;
220 } while (bh != head);
221
222 /* we might be here because some of the buffers on this page are
223 * not mapped. This is due to various races between
224 * file io on the block device and getblk. It gets dealt with
225 * elsewhere, don't buffer_error if we had some unmapped buffers
226 */
Tetsuo Handa43636c82019-01-21 22:49:37 +0900227 ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE);
228 if (all_mapped && __ratelimit(&last_warned)) {
229 printk("__find_get_block_slow() failed. block=%llu, "
230 "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
231 "device %pg blocksize: %d\n",
232 (unsigned long long)block,
233 (unsigned long long)bh->b_blocknr,
234 bh->b_state, bh->b_size, bdev,
235 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236 }
237out_unlock:
238 spin_unlock(&bd_mapping->private_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300239 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700240out:
241 return ret;
242}
243
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
245{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700247 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248 struct buffer_head *tmp;
249 struct page *page;
250 int page_uptodate = 1;
251
252 BUG_ON(!buffer_async_read(bh));
253
254 page = bh->b_page;
255 if (uptodate) {
256 set_buffer_uptodate(bh);
257 } else {
258 clear_buffer_uptodate(bh);
Robert Elliott432f16e2014-10-21 13:55:11 -0600259 buffer_io_error(bh, ", async page read");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700260 SetPageError(page);
261 }
262
263 /*
264 * Be _very_ careful from here on. Bad things can happen if
265 * two buffer heads end IO at almost the same time and both
266 * decide that the page is now completely done.
267 */
Nick Piggina3972202005-07-07 17:56:56 -0700268 first = page_buffers(page);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100269 spin_lock_irqsave(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700270 clear_buffer_async_read(bh);
271 unlock_buffer(bh);
272 tmp = bh;
273 do {
274 if (!buffer_uptodate(tmp))
275 page_uptodate = 0;
276 if (buffer_async_read(tmp)) {
277 BUG_ON(!buffer_locked(tmp));
278 goto still_busy;
279 }
280 tmp = tmp->b_this_page;
281 } while (tmp != bh);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100282 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700283
284 /*
285 * If none of the buffers had errors and they are all
286 * uptodate then we can set the page uptodate.
287 */
288 if (page_uptodate && !PageError(page))
289 SetPageUptodate(page);
290 unlock_page(page);
291 return;
292
293still_busy:
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100294 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700295 return;
296}
297
Eric Biggers31fb9922019-10-22 20:33:11 -0700298struct decrypt_bh_ctx {
299 struct work_struct work;
300 struct buffer_head *bh;
301};
302
303static void decrypt_bh(struct work_struct *work)
304{
305 struct decrypt_bh_ctx *ctx =
306 container_of(work, struct decrypt_bh_ctx, work);
307 struct buffer_head *bh = ctx->bh;
308 int err;
309
310 err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size,
311 bh_offset(bh));
312 end_buffer_async_read(bh, err == 0);
313 kfree(ctx);
314}
315
316/*
317 * I/O completion handler for block_read_full_page() - pages
318 * which come unlocked at the end of I/O.
319 */
320static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate)
321{
322 /* Decrypt if needed */
Eric Biggers4f74d152020-07-02 01:56:07 +0000323 if (uptodate &&
324 fscrypt_inode_uses_fs_layer_crypto(bh->b_page->mapping->host)) {
Eric Biggers31fb9922019-10-22 20:33:11 -0700325 struct decrypt_bh_ctx *ctx = kmalloc(sizeof(*ctx), GFP_ATOMIC);
326
327 if (ctx) {
328 INIT_WORK(&ctx->work, decrypt_bh);
329 ctx->bh = bh;
330 fscrypt_enqueue_decrypt_work(&ctx->work);
331 return;
332 }
333 uptodate = 0;
334 }
335 end_buffer_async_read(bh, uptodate);
336}
337
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338/*
339 * Completion handler for block_write_full_page() - pages which are unlocked
340 * during I/O, and which have PageWriteback cleared upon I/O completion.
341 */
Chris Mason35c80d52009-04-15 13:22:38 -0400342void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700343{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700344 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700345 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700346 struct buffer_head *tmp;
347 struct page *page;
348
349 BUG_ON(!buffer_async_write(bh));
350
351 page = bh->b_page;
352 if (uptodate) {
353 set_buffer_uptodate(bh);
354 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600355 buffer_io_error(bh, ", lost async page write");
Jeff Layton87354e52017-07-06 07:02:21 -0400356 mark_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);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100362 spin_lock_irqsave(&first->b_uptodate_lock, flags);
Nick Piggina3972202005-07-07 17:56:56 -0700363
Linus Torvalds1da177e2005-04-16 15:20:36 -0700364 clear_buffer_async_write(bh);
365 unlock_buffer(bh);
366 tmp = bh->b_this_page;
367 while (tmp != bh) {
368 if (buffer_async_write(tmp)) {
369 BUG_ON(!buffer_locked(tmp));
370 goto still_busy;
371 }
372 tmp = tmp->b_this_page;
373 }
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100374 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700375 end_page_writeback(page);
376 return;
377
378still_busy:
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100379 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380 return;
381}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700382EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700383
384/*
385 * If a page's buffers are under async readin (end_buffer_async_read
386 * completion) then there is a possibility that another thread of
387 * control could lock one of the buffers after it has completed
388 * but while some of the other buffers have not completed. This
389 * locked buffer would confuse end_buffer_async_read() into not unlocking
390 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
391 * that this buffer is not under async I/O.
392 *
393 * The page comes unlocked when it has no locked buffer_async buffers
394 * left.
395 *
396 * PageLocked prevents anyone starting new async I/O reads any of
397 * the buffers.
398 *
399 * PageWriteback is used to prevent simultaneous writeout of the same
400 * page.
401 *
402 * PageLocked prevents anyone from starting writeback of a page which is
403 * under read I/O (PageWriteback is only ever set against a locked page).
404 */
405static void mark_buffer_async_read(struct buffer_head *bh)
406{
Eric Biggers31fb9922019-10-22 20:33:11 -0700407 bh->b_end_io = end_buffer_async_read_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408 set_buffer_async_read(bh);
409}
410
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700411static void mark_buffer_async_write_endio(struct buffer_head *bh,
412 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400413{
414 bh->b_end_io = handler;
415 set_buffer_async_write(bh);
416}
417
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418void mark_buffer_async_write(struct buffer_head *bh)
419{
Chris Mason35c80d52009-04-15 13:22:38 -0400420 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700421}
422EXPORT_SYMBOL(mark_buffer_async_write);
423
424
425/*
426 * fs/buffer.c contains helper functions for buffer-backed address space's
427 * fsync functions. A common requirement for buffer-based filesystems is
428 * that certain data from the backing blockdev needs to be written out for
429 * a successful fsync(). For example, ext2 indirect blocks need to be
430 * written back and waited upon before fsync() returns.
431 *
432 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
433 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
434 * management of a list of dependent buffers at ->i_mapping->private_list.
435 *
436 * Locking is a little subtle: try_to_free_buffers() will remove buffers
437 * from their controlling inode's queue when they are being freed. But
438 * try_to_free_buffers() will be operating against the *blockdev* mapping
439 * at the time, not against the S_ISREG file which depends on those buffers.
440 * So the locking for private_list is via the private_lock in the address_space
441 * which backs the buffers. Which is different from the address_space
442 * against which the buffers are listed. So for a particular address_space,
443 * mapping->private_lock does *not* protect mapping->private_list! In fact,
444 * mapping->private_list will always be protected by the backing blockdev's
445 * ->private_lock.
446 *
447 * Which introduces a requirement: all buffers on an address_space's
448 * ->private_list must be from the same address_space: the blockdev's.
449 *
450 * address_spaces which do not place buffers at ->private_list via these
451 * utility functions are free to use private_lock and private_list for
452 * whatever they want. The only requirement is that list_empty(private_list)
453 * be true at clear_inode() time.
454 *
455 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
456 * filesystems should do that. invalidate_inode_buffers() should just go
457 * BUG_ON(!list_empty).
458 *
459 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
460 * take an address_space, not an inode. And it should be called
461 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
462 * queued up.
463 *
464 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
465 * list if it is already on a list. Because if the buffer is on a list,
466 * it *must* already be on the right one. If not, the filesystem is being
467 * silly. This will save a ton of locking. But first we have to ensure
468 * that buffers are taken *off* the old inode's list when they are freed
469 * (presumably in truncate). That requires careful auditing of all
470 * filesystems (do it inside bforget()). It could also be done by bringing
471 * b_inode back.
472 */
473
474/*
475 * The buffer's backing address_space's private_lock must be held
476 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700477static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478{
479 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff407b2006-10-17 00:10:19 -0700480 WARN_ON(!bh->b_assoc_map);
Jan Kara58ff407b2006-10-17 00:10:19 -0700481 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700482}
483
484int inode_has_buffers(struct inode *inode)
485{
486 return !list_empty(&inode->i_data.private_list);
487}
488
489/*
490 * osync is designed to support O_SYNC io. It waits synchronously for
491 * all already-submitted IO to complete, but does not queue any new
492 * writes to the disk.
493 *
494 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
495 * you dirty the buffers, and then use osync_inode_buffers to wait for
496 * completion. Any other dirty buffers which are not yet queued for
497 * write will not be flushed to disk by the osync.
498 */
499static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
500{
501 struct buffer_head *bh;
502 struct list_head *p;
503 int err = 0;
504
505 spin_lock(lock);
506repeat:
507 list_for_each_prev(p, list) {
508 bh = BH_ENTRY(p);
509 if (buffer_locked(bh)) {
510 get_bh(bh);
511 spin_unlock(lock);
512 wait_on_buffer(bh);
513 if (!buffer_uptodate(bh))
514 err = -EIO;
515 brelse(bh);
516 spin_lock(lock);
517 goto repeat;
518 }
519 }
520 spin_unlock(lock);
521 return err;
522}
523
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +0200524void emergency_thaw_bdev(struct super_block *sb)
Al Viro01a05b32010-03-23 06:06:58 -0400525{
Christoph Hellwig040f04b2020-11-24 11:54:06 +0100526 while (sb->s_bdev && !thaw_bdev(sb->s_bdev))
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400527 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
Al Viro01a05b32010-03-23 06:06:58 -0400528}
529
Linus Torvalds1da177e2005-04-16 15:20:36 -0700530/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800531 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700532 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533 *
534 * Starts I/O against the buffers at mapping->private_list, and waits upon
535 * that I/O.
536 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700537 * Basically, this is a convenience function for fsync().
538 * @mapping is a file or directory which needs those buffers to be written for
539 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700540 */
541int sync_mapping_buffers(struct address_space *mapping)
542{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800543 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700544
545 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
546 return 0;
547
548 return fsync_buffers_list(&buffer_mapping->private_lock,
549 &mapping->private_list);
550}
551EXPORT_SYMBOL(sync_mapping_buffers);
552
553/*
554 * Called when we've recently written block `bblock', and it is known that
555 * `bblock' was for a buffer_boundary() buffer. This means that the block at
556 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
557 * dirty, schedule it for IO. So that indirects merge nicely with their data.
558 */
559void write_boundary_block(struct block_device *bdev,
560 sector_t bblock, unsigned blocksize)
561{
562 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
563 if (bh) {
564 if (buffer_dirty(bh))
Mike Christiedfec8a12016-06-05 14:31:44 -0500565 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 put_bh(bh);
567 }
568}
569
570void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
571{
572 struct address_space *mapping = inode->i_mapping;
573 struct address_space *buffer_mapping = bh->b_page->mapping;
574
575 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800576 if (!mapping->private_data) {
577 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800579 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800581 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 spin_lock(&buffer_mapping->private_lock);
583 list_move_tail(&bh->b_assoc_buffers,
584 &mapping->private_list);
Jan Kara58ff407b2006-10-17 00:10:19 -0700585 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 spin_unlock(&buffer_mapping->private_lock);
587 }
588}
589EXPORT_SYMBOL(mark_buffer_dirty_inode);
590
591/*
592 * Add a page to the dirty page list.
593 *
594 * It is a sad fact of life that this function is called from several places
595 * deeply under spinlocking. It may not sleep.
596 *
597 * If the page has buffers, the uptodate buffers are set dirty, to preserve
598 * dirty-state coherency between the page and the buffers. It the page does
599 * not have buffers then when they are later attached they will all be set
600 * dirty.
601 *
602 * The buffers are dirtied before the page is dirtied. There's a small race
603 * window in which a writepage caller may see the page cleanness but not the
604 * buffer dirtiness. That's fine. If this code were to set the page dirty
605 * before the buffers, a concurrent writepage caller could clear the page dirty
606 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
607 * page on the dirty page list.
608 *
609 * We use private_lock to lock against try_to_free_buffers while using the
610 * page's buffer list. Also use this to protect against clean buffers being
611 * added to the page after it was set dirty.
612 *
613 * FIXME: may need to call ->reservepage here as well. That's rather up to the
614 * address_space though.
615 */
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000616bool block_dirty_folio(struct address_space *mapping, struct folio *folio)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700617{
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000618 struct buffer_head *head;
619 bool newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
621 spin_lock(&mapping->private_lock);
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000622 head = folio_buffers(folio);
623 if (head) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 struct buffer_head *bh = head;
625
626 do {
627 set_buffer_dirty(bh);
628 bh = bh->b_this_page;
629 } while (bh != head);
630 }
Greg Thelenc4843a72015-05-22 17:13:16 -0400631 /*
Roman Gushchinbcfe06b2020-12-01 13:58:27 -0800632 * Lock out page's memcg migration to keep PageDirty
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700633 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -0400634 */
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000635 folio_memcg_lock(folio);
636 newly_dirty = !folio_test_set_dirty(folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637 spin_unlock(&mapping->private_lock);
638
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700639 if (newly_dirty)
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000640 __folio_mark_dirty(folio, mapping, 1);
Greg Thelenc4843a72015-05-22 17:13:16 -0400641
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000642 folio_memcg_unlock(folio);
Greg Thelenc4843a72015-05-22 17:13:16 -0400643
644 if (newly_dirty)
645 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
646
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700647 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648}
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +0000649EXPORT_SYMBOL(block_dirty_folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
651/*
652 * Write out and wait upon a list of buffers.
653 *
654 * We have conflicting pressures: we want to make sure that all
655 * initially dirty buffers get waited on, but that any subsequently
656 * dirtied buffers don't. After all, we don't want fsync to last
657 * forever if somebody is actively writing to the file.
658 *
659 * Do this in two main stages: first we copy dirty buffers to a
660 * temporary inode list, queueing the writes as we go. Then we clean
661 * up, waiting for those writes to complete.
662 *
663 * During this second stage, any subsequent updates to the file may end
664 * up refiling the buffer on the original inode's dirty list again, so
665 * there is a chance we will end up with a buffer queued for write but
666 * not yet completed on that list. So, as a final cleanup we go through
667 * the osync code to catch these locked, dirty buffers without requeuing
668 * any newly dirty buffers for write.
669 */
670static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
671{
672 struct buffer_head *bh;
673 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100674 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100676 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
678 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100679 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
681 spin_lock(lock);
682 while (!list_empty(list)) {
683 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800684 mapping = bh->b_assoc_map;
Jan Kara58ff407b2006-10-17 00:10:19 -0700685 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800686 /* Avoid race with mark_buffer_dirty_inode() which does
687 * a lockless check and we rely on seeing the dirty bit */
688 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689 if (buffer_dirty(bh) || buffer_locked(bh)) {
690 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800691 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700692 if (buffer_dirty(bh)) {
693 get_bh(bh);
694 spin_unlock(lock);
695 /*
696 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200697 * write_dirty_buffer() actually writes the
698 * current contents - it is a noop if I/O is
699 * still in flight on potentially older
700 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701 */
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600702 write_dirty_buffer(bh, REQ_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200703
704 /*
705 * Kick off IO for the previous mapping. Note
706 * that we will not run the very last mapping,
707 * wait_on_buffer() will do that for us
708 * through sync_buffer().
709 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710 brelse(bh);
711 spin_lock(lock);
712 }
713 }
714 }
715
Jens Axboe4ee24912011-03-17 10:51:40 +0100716 spin_unlock(lock);
717 blk_finish_plug(&plug);
718 spin_lock(lock);
719
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 while (!list_empty(&tmp)) {
721 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800723 mapping = bh->b_assoc_map;
724 __remove_assoc_queue(bh);
725 /* Avoid race with mark_buffer_dirty_inode() which does
726 * a lockless check and we rely on seeing the dirty bit */
727 smp_mb();
728 if (buffer_dirty(bh)) {
729 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800730 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800731 bh->b_assoc_map = mapping;
732 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733 spin_unlock(lock);
734 wait_on_buffer(bh);
735 if (!buffer_uptodate(bh))
736 err = -EIO;
737 brelse(bh);
738 spin_lock(lock);
739 }
740
741 spin_unlock(lock);
742 err2 = osync_buffers_list(lock, list);
743 if (err)
744 return err;
745 else
746 return err2;
747}
748
749/*
750 * Invalidate any and all dirty buffers on a given inode. We are
751 * probably unmounting the fs, but that doesn't mean we have already
752 * done a sync(). Just drop the buffers from the inode list.
753 *
754 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
755 * assumes that all the buffers are against the blockdev. Not true
756 * for reiserfs.
757 */
758void invalidate_inode_buffers(struct inode *inode)
759{
760 if (inode_has_buffers(inode)) {
761 struct address_space *mapping = &inode->i_data;
762 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800763 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
765 spin_lock(&buffer_mapping->private_lock);
766 while (!list_empty(list))
767 __remove_assoc_queue(BH_ENTRY(list->next));
768 spin_unlock(&buffer_mapping->private_lock);
769 }
770}
Jan Kara52b19ac2008-09-23 18:24:08 +0200771EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700772
773/*
774 * Remove any clean buffers from the inode's buffer list. This is called
775 * when we're trying to free the inode itself. Those buffers can pin it.
776 *
777 * Returns true if all buffers were removed.
778 */
779int remove_inode_buffers(struct inode *inode)
780{
781 int ret = 1;
782
783 if (inode_has_buffers(inode)) {
784 struct address_space *mapping = &inode->i_data;
785 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800786 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787
788 spin_lock(&buffer_mapping->private_lock);
789 while (!list_empty(list)) {
790 struct buffer_head *bh = BH_ENTRY(list->next);
791 if (buffer_dirty(bh)) {
792 ret = 0;
793 break;
794 }
795 __remove_assoc_queue(bh);
796 }
797 spin_unlock(&buffer_mapping->private_lock);
798 }
799 return ret;
800}
801
802/*
803 * Create the appropriate buffers when given a page for data area and
804 * the size of each buffer.. Use the bh->b_this_page linked list to
805 * follow the buffers created. Return NULL if unable to create more
806 * buffers.
807 *
808 * The retry flag is used to differentiate async IO (paging, swapping)
809 * which may not fail from ordinary buffer allocations.
810 */
811struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
Jens Axboe640ab982017-09-27 05:40:16 -0600812 bool retry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700813{
814 struct buffer_head *bh, *head;
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700815 gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700816 long offset;
Roman Gushchinb87d8ce2020-10-17 16:13:40 -0700817 struct mem_cgroup *memcg, *old_memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700818
Jens Axboe640ab982017-09-27 05:40:16 -0600819 if (retry)
820 gfp |= __GFP_NOFAIL;
821
Johannes Weiner6eeb1042021-02-24 12:04:15 -0800822 /* The page lock pins the memcg */
823 memcg = page_memcg(page);
Roman Gushchinb87d8ce2020-10-17 16:13:40 -0700824 old_memcg = set_active_memcg(memcg);
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700825
Linus Torvalds1da177e2005-04-16 15:20:36 -0700826 head = NULL;
827 offset = PAGE_SIZE;
828 while ((offset -= size) >= 0) {
Jens Axboe640ab982017-09-27 05:40:16 -0600829 bh = alloc_buffer_head(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700830 if (!bh)
831 goto no_grow;
832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 bh->b_this_page = head;
834 bh->b_blocknr = -1;
835 head = bh;
836
Linus Torvalds1da177e2005-04-16 15:20:36 -0700837 bh->b_size = size;
838
839 /* Link the buffer to its page */
840 set_bh_page(bh, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700841 }
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700842out:
Roman Gushchinb87d8ce2020-10-17 16:13:40 -0700843 set_active_memcg(old_memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700844 return head;
845/*
846 * In case anything failed, we just free everything we got.
847 */
848no_grow:
849 if (head) {
850 do {
851 bh = head;
852 head = head->b_this_page;
853 free_buffer_head(bh);
854 } while (head);
855 }
856
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700857 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858}
859EXPORT_SYMBOL_GPL(alloc_page_buffers);
860
861static inline void
862link_dev_buffers(struct page *page, struct buffer_head *head)
863{
864 struct buffer_head *bh, *tail;
865
866 bh = head;
867 do {
868 tail = bh;
869 bh = bh->b_this_page;
870 } while (bh);
871 tail->b_this_page = head;
Guoqing Jiang45dcfc22020-06-01 21:47:48 -0700872 attach_page_private(page, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700873}
874
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800875static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
876{
877 sector_t retval = ~((sector_t)0);
Christoph Hellwigb86058f2021-10-18 12:11:09 +0200878 loff_t sz = bdev_nr_bytes(bdev);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800879
880 if (sz) {
881 unsigned int sizebits = blksize_bits(size);
882 retval = (sz >> sizebits);
883 }
884 return retval;
885}
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887/*
888 * Initialise the state of a blockdev page's buffers.
889 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200890static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700891init_page_buffers(struct page *page, struct block_device *bdev,
892 sector_t block, int size)
893{
894 struct buffer_head *head = page_buffers(page);
895 struct buffer_head *bh = head;
896 int uptodate = PageUptodate(page);
Christoph Hellwigbcd1d062021-10-18 12:11:10 +0200897 sector_t end_block = blkdev_max_block(bdev, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700898
899 do {
900 if (!buffer_mapped(bh)) {
Eric Biggers01950a32018-01-16 22:25:12 -0800901 bh->b_end_io = NULL;
902 bh->b_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903 bh->b_bdev = bdev;
904 bh->b_blocknr = block;
905 if (uptodate)
906 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200907 if (block < end_block)
908 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700909 }
910 block++;
911 bh = bh->b_this_page;
912 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200913
914 /*
915 * Caller needs to validate requested block against end of device.
916 */
917 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918}
919
920/*
921 * Create the page-cache page that contains the requested block.
922 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200923 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700924 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200925static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700926grow_dev_page(struct block_device *bdev, sector_t block,
Gioh Kim3b5e6452014-09-04 22:04:42 -0400927 pgoff_t index, int size, int sizebits, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700928{
929 struct inode *inode = bdev->bd_inode;
930 struct page *page;
931 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200932 sector_t end_block;
Zhiqiang Liuc4b4c2a2020-04-13 13:12:10 +0800933 int ret = 0;
Johannes Weiner84235de2013-10-16 13:47:00 -0700934 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700935
Michal Hockoc62d2552015-11-06 16:28:49 -0800936 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
Gioh Kim3b5e6452014-09-04 22:04:42 -0400937
Johannes Weiner84235de2013-10-16 13:47:00 -0700938 /*
939 * XXX: __getblk_slow() can not really deal with failure and
940 * will endlessly loop on improvised global reclaim. Prefer
941 * looping in the allocator rather than here, at least that
942 * code knows what it's doing.
943 */
944 gfp_mask |= __GFP_NOFAIL;
945
946 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700947
Eric Sesterhenne827f922006-03-26 18:24:46 +0200948 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700949
950 if (page_has_buffers(page)) {
951 bh = page_buffers(page);
952 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200953 end_block = init_page_buffers(page, bdev,
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +0100954 (sector_t)index << sizebits,
955 size);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200956 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700957 }
958 if (!try_to_free_buffers(page))
959 goto failed;
960 }
961
962 /*
963 * Allocate some buffers for this page
964 */
Jens Axboe94dc24c2017-09-27 05:45:36 -0600965 bh = alloc_page_buffers(page, size, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700966
967 /*
968 * Link the page to the buffers and initialise them. Take the
969 * lock to be atomic wrt __find_get_block(), which does not
970 * run under the page lock.
971 */
972 spin_lock(&inode->i_mapping->private_lock);
973 link_dev_buffers(page, bh);
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +0100974 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
975 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700976 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200977done:
978 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700979failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300981 put_page(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200982 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700983}
984
985/*
986 * Create buffers for the specified block device block's page. If
987 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700988 */
Arjan van de Ven858119e2006-01-14 13:20:43 -0800989static int
Gioh Kim3b5e6452014-09-04 22:04:42 -0400990grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700991{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992 pgoff_t index;
993 int sizebits;
994
Mikulas Patocka90432e62021-03-22 10:05:05 -0400995 sizebits = PAGE_SHIFT - __ffs(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700996 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700997
Andrew Mortone5657932006-10-11 01:21:46 -0700998 /*
999 * Check for a block which wants to lie outside our maximum possible
1000 * pagecache index. (this comparison is done using sector_t types).
1001 */
1002 if (unlikely(index != block >> sizebits)) {
Andrew Mortone5657932006-10-11 01:21:46 -07001003 printk(KERN_ERR "%s: requested out-of-range block %llu for "
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001004 "device %pg\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001005 __func__, (unsigned long long)block,
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001006 bdev);
Andrew Mortone5657932006-10-11 01:21:46 -07001007 return -EIO;
1008 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010 /* Create a page with the proper size buffers.. */
Gioh Kim3b5e6452014-09-04 22:04:42 -04001011 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001012}
1013
Eric Biggers0026ba42016-09-12 13:30:41 -07001014static struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001015__getblk_slow(struct block_device *bdev, sector_t block,
1016 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001017{
1018 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001019 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001020 (size < 512 || size > PAGE_SIZE))) {
1021 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1022 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001023 printk(KERN_ERR "logical block size: %d\n",
1024 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025
1026 dump_stack();
1027 return NULL;
1028 }
1029
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001030 for (;;) {
1031 struct buffer_head *bh;
1032 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033
1034 bh = __find_get_block(bdev, block, size);
1035 if (bh)
1036 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001037
Gioh Kim3b5e6452014-09-04 22:04:42 -04001038 ret = grow_buffers(bdev, block, size, gfp);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001039 if (ret < 0)
1040 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 }
1042}
1043
1044/*
1045 * The relationship between dirty buffers and dirty pages:
1046 *
1047 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
Matthew Wilcoxec82e1c2017-12-04 10:40:41 -05001048 * the page is tagged dirty in the page cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001049 *
1050 * At all times, the dirtiness of the buffers represents the dirtiness of
1051 * subsections of the page. If the page has buffers, the page dirty bit is
1052 * merely a hint about the true dirty state.
1053 *
1054 * When a page is set dirty in its entirety, all its buffers are marked dirty
1055 * (if the page has buffers).
1056 *
1057 * When a buffer is marked dirty, its page is dirtied, but the page's other
1058 * buffers are not.
1059 *
1060 * Also. When blockdev buffers are explicitly read with bread(), they
1061 * individually become uptodate. But their backing page remains not
1062 * uptodate - even if all of its buffers are uptodate. A subsequent
1063 * block_read_full_page() against that page will discover all the uptodate
1064 * buffers, will set the page uptodate and will perform no I/O.
1065 */
1066
1067/**
1068 * mark_buffer_dirty - mark a buffer_head as needing writeout
Martin Waitz67be2dd2005-05-01 08:59:26 -07001069 * @bh: the buffer_head to mark dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070 *
Matthew Wilcoxec82e1c2017-12-04 10:40:41 -05001071 * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1072 * its backing page dirty, then tag the page as dirty in the page cache
1073 * and then attach the address_space's inode to its superblock's dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001074 * inode list.
1075 *
1076 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001077 * i_pages lock and mapping->host->i_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001079void mark_buffer_dirty(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001080{
Nick Piggin787d22142007-07-17 04:03:34 -07001081 WARN_ON_ONCE(!buffer_uptodate(bh));
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001082
Tejun Heo5305cb82013-01-11 13:06:36 -08001083 trace_block_dirty_buffer(bh);
1084
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001085 /*
1086 * Very *carefully* optimize the it-is-already-dirty case.
1087 *
1088 * Don't let the final "is it dirty" escape to before we
1089 * perhaps modified the buffer.
1090 */
1091 if (buffer_dirty(bh)) {
1092 smp_mb();
1093 if (buffer_dirty(bh))
1094 return;
1095 }
1096
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001097 if (!test_set_buffer_dirty(bh)) {
1098 struct page *page = bh->b_page;
Greg Thelenc4843a72015-05-22 17:13:16 -04001099 struct address_space *mapping = NULL;
Greg Thelenc4843a72015-05-22 17:13:16 -04001100
Johannes Weiner62cccb82016-03-15 14:57:22 -07001101 lock_page_memcg(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001102 if (!TestSetPageDirty(page)) {
Greg Thelenc4843a72015-05-22 17:13:16 -04001103 mapping = page_mapping(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001104 if (mapping)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001105 __set_page_dirty(page, mapping, 0);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001106 }
Johannes Weiner62cccb82016-03-15 14:57:22 -07001107 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04001108 if (mapping)
1109 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001110 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001112EXPORT_SYMBOL(mark_buffer_dirty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001113
Jeff Layton87354e52017-07-06 07:02:21 -04001114void mark_buffer_write_io_error(struct buffer_head *bh)
1115{
Jeff Layton485e9602020-06-01 21:45:40 -07001116 struct super_block *sb;
1117
Jeff Layton87354e52017-07-06 07:02:21 -04001118 set_buffer_write_io_error(bh);
1119 /* FIXME: do we need to set this in both places? */
1120 if (bh->b_page && bh->b_page->mapping)
1121 mapping_set_error(bh->b_page->mapping, -EIO);
1122 if (bh->b_assoc_map)
1123 mapping_set_error(bh->b_assoc_map, -EIO);
Jeff Layton485e9602020-06-01 21:45:40 -07001124 rcu_read_lock();
1125 sb = READ_ONCE(bh->b_bdev->bd_super);
1126 if (sb)
1127 errseq_set(&sb->s_wb_err, -EIO);
1128 rcu_read_unlock();
Jeff Layton87354e52017-07-06 07:02:21 -04001129}
1130EXPORT_SYMBOL(mark_buffer_write_io_error);
1131
Linus Torvalds1da177e2005-04-16 15:20:36 -07001132/*
1133 * Decrement a buffer_head's reference count. If all buffers against a page
1134 * have zero reference count, are clean and unlocked, and if the page is clean
1135 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1136 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1137 * a page but it ends up not being freed, and buffers may later be reattached).
1138 */
1139void __brelse(struct buffer_head * buf)
1140{
1141 if (atomic_read(&buf->b_count)) {
1142 put_bh(buf);
1143 return;
1144 }
Arjan van de Ven5c752ad2008-07-25 19:45:40 -07001145 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001146}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001147EXPORT_SYMBOL(__brelse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001148
1149/*
1150 * bforget() is like brelse(), except it discards any
1151 * potentially dirty data.
1152 */
1153void __bforget(struct buffer_head *bh)
1154{
1155 clear_buffer_dirty(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -08001156 if (bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 struct address_space *buffer_mapping = bh->b_page->mapping;
1158
1159 spin_lock(&buffer_mapping->private_lock);
1160 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff407b2006-10-17 00:10:19 -07001161 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001162 spin_unlock(&buffer_mapping->private_lock);
1163 }
1164 __brelse(bh);
1165}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001166EXPORT_SYMBOL(__bforget);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001167
1168static struct buffer_head *__bread_slow(struct buffer_head *bh)
1169{
1170 lock_buffer(bh);
1171 if (buffer_uptodate(bh)) {
1172 unlock_buffer(bh);
1173 return bh;
1174 } else {
1175 get_bh(bh);
1176 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05001177 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178 wait_on_buffer(bh);
1179 if (buffer_uptodate(bh))
1180 return bh;
1181 }
1182 brelse(bh);
1183 return NULL;
1184}
1185
1186/*
1187 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1188 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1189 * refcount elevated by one when they're in an LRU. A buffer can only appear
1190 * once in a particular CPU's LRU. A single buffer can be present in multiple
1191 * CPU's LRUs at the same time.
1192 *
1193 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1194 * sb_find_get_block().
1195 *
1196 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1197 * a local interrupt disable for that.
1198 */
1199
Sebastien Buisson86cf78d2014-10-09 15:29:38 -07001200#define BH_LRU_SIZE 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07001201
1202struct bh_lru {
1203 struct buffer_head *bhs[BH_LRU_SIZE];
1204};
1205
1206static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1207
1208#ifdef CONFIG_SMP
1209#define bh_lru_lock() local_irq_disable()
1210#define bh_lru_unlock() local_irq_enable()
1211#else
1212#define bh_lru_lock() preempt_disable()
1213#define bh_lru_unlock() preempt_enable()
1214#endif
1215
1216static inline void check_irqs_on(void)
1217{
1218#ifdef irqs_disabled
1219 BUG_ON(irqs_disabled());
1220#endif
1221}
1222
1223/*
Eric Biggers241f01f2017-07-10 15:47:29 -07001224 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1225 * inserted at the front, and the buffer_head at the back if any is evicted.
1226 * Or, if already in the LRU it is moved to the front.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001227 */
1228static void bh_lru_install(struct buffer_head *bh)
1229{
Eric Biggers241f01f2017-07-10 15:47:29 -07001230 struct buffer_head *evictee = bh;
1231 struct bh_lru *b;
1232 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001233
1234 check_irqs_on();
Minchan Kimc0226eb2022-03-22 14:39:34 -07001235 bh_lru_lock();
1236
Minchan Kim8cc621d2021-05-04 18:37:00 -07001237 /*
1238 * the refcount of buffer_head in bh_lru prevents dropping the
1239 * attached page(i.e., try_to_free_buffers) so it could cause
1240 * failing page migration.
1241 * Skip putting upcoming bh into bh_lru until migration is done.
1242 */
Minchan Kimc0226eb2022-03-22 14:39:34 -07001243 if (lru_cache_disabled()) {
1244 bh_lru_unlock();
Minchan Kim8cc621d2021-05-04 18:37:00 -07001245 return;
Minchan Kimc0226eb2022-03-22 14:39:34 -07001246 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247
Eric Biggers241f01f2017-07-10 15:47:29 -07001248 b = this_cpu_ptr(&bh_lrus);
1249 for (i = 0; i < BH_LRU_SIZE; i++) {
1250 swap(evictee, b->bhs[i]);
1251 if (evictee == bh) {
1252 bh_lru_unlock();
1253 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001254 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001255 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001256
Eric Biggers241f01f2017-07-10 15:47:29 -07001257 get_bh(bh);
1258 bh_lru_unlock();
1259 brelse(evictee);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001260}
1261
1262/*
1263 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1264 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001265static struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001266lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267{
1268 struct buffer_head *ret = NULL;
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001269 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270
1271 check_irqs_on();
1272 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 for (i = 0; i < BH_LRU_SIZE; i++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001274 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001275
Zach Brown9470dd52014-10-13 15:55:05 -07001276 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1277 bh->b_size == size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278 if (i) {
1279 while (i) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001280 __this_cpu_write(bh_lrus.bhs[i],
1281 __this_cpu_read(bh_lrus.bhs[i - 1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 i--;
1283 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06001284 __this_cpu_write(bh_lrus.bhs[0], bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001285 }
1286 get_bh(bh);
1287 ret = bh;
1288 break;
1289 }
1290 }
1291 bh_lru_unlock();
1292 return ret;
1293}
1294
1295/*
1296 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1297 * it in the LRU and mark it as accessed. If it is not present then return
1298 * NULL
1299 */
1300struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001301__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001302{
1303 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1304
1305 if (bh == NULL) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001306 /* __find_get_block_slow will mark the page accessed */
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001307 bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001308 if (bh)
1309 bh_lru_install(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001310 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311 touch_buffer(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313 return bh;
1314}
1315EXPORT_SYMBOL(__find_get_block);
1316
1317/*
Gioh Kim3b5e6452014-09-04 22:04:42 -04001318 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001319 * which corresponds to the passed block_device, block and size. The
1320 * returned buffer has its reference count incremented.
1321 *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001322 * __getblk_gfp() will lock up the machine if grow_dev_page's
1323 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324 */
1325struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001326__getblk_gfp(struct block_device *bdev, sector_t block,
1327 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328{
1329 struct buffer_head *bh = __find_get_block(bdev, block, size);
1330
1331 might_sleep();
1332 if (bh == NULL)
Gioh Kim3b5e6452014-09-04 22:04:42 -04001333 bh = __getblk_slow(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334 return bh;
1335}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001336EXPORT_SYMBOL(__getblk_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001337
1338/*
1339 * Do async read-ahead on a buffer..
1340 */
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001341void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
1343 struct buffer_head *bh = __getblk(bdev, block, size);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001344 if (likely(bh)) {
Christoph Hellwig70246282016-07-19 11:28:41 +02001345 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001346 brelse(bh);
1347 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348}
1349EXPORT_SYMBOL(__breadahead);
1350
Roman Gushchind87f6392020-02-28 16:14:11 -08001351void __breadahead_gfp(struct block_device *bdev, sector_t block, unsigned size,
1352 gfp_t gfp)
1353{
1354 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1355 if (likely(bh)) {
1356 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1357 brelse(bh);
1358 }
1359}
1360EXPORT_SYMBOL(__breadahead_gfp);
1361
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362/**
Gioh Kim3b5e6452014-09-04 22:04:42 -04001363 * __bread_gfp() - reads a specified block and returns the bh
Martin Waitz67be2dd2005-05-01 08:59:26 -07001364 * @bdev: the block_device to read from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001365 * @block: number of block
1366 * @size: size (in bytes) to read
Gioh Kim3b5e6452014-09-04 22:04:42 -04001367 * @gfp: page allocation flag
1368 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * Reads a specified block, and returns buffer head that contains it.
Gioh Kim3b5e6452014-09-04 22:04:42 -04001370 * The page cache can be allocated from non-movable area
1371 * not to prevent page migration if you set gfp to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372 * It returns NULL if the block was unreadable.
1373 */
1374struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001375__bread_gfp(struct block_device *bdev, sector_t block,
1376 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377{
Gioh Kim3b5e6452014-09-04 22:04:42 -04001378 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379
Andrew Mortona3e713b2005-10-30 15:03:15 -08001380 if (likely(bh) && !buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001381 bh = __bread_slow(bh);
1382 return bh;
1383}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001384EXPORT_SYMBOL(__bread_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385
Minchan Kim8cc621d2021-05-04 18:37:00 -07001386static void __invalidate_bh_lrus(struct bh_lru *b)
1387{
1388 int i;
1389
1390 for (i = 0; i < BH_LRU_SIZE; i++) {
1391 brelse(b->bhs[i]);
1392 b->bhs[i] = NULL;
1393 }
1394}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395/*
1396 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1397 * This doesn't race because it runs in each cpu either in irq
1398 * or with preempt disabled.
1399 */
1400static void invalidate_bh_lru(void *arg)
1401{
1402 struct bh_lru *b = &get_cpu_var(bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403
Minchan Kim8cc621d2021-05-04 18:37:00 -07001404 __invalidate_bh_lrus(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405 put_cpu_var(bh_lrus);
1406}
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001407
Minchan Kim8cc621d2021-05-04 18:37:00 -07001408bool has_bh_in_lru(int cpu, void *dummy)
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001409{
1410 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1411 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001413 for (i = 0; i < BH_LRU_SIZE; i++) {
1414 if (b->bhs[i])
Saurav Girepunje1d706672019-11-30 17:49:15 -08001415 return true;
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001416 }
1417
Saurav Girepunje1d706672019-11-30 17:49:15 -08001418 return false;
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001419}
1420
Peter Zijlstraf9a14392007-05-06 14:49:55 -07001421void invalidate_bh_lrus(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422{
Sebastian Andrzej Siewiorcb923152020-01-17 10:01:37 +01001423 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424}
Nick Piggin9db55792008-02-08 04:19:49 -08001425EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426
Minchan Kim243418e2021-09-24 15:43:47 -07001427/*
1428 * It's called from workqueue context so we need a bh_lru_lock to close
1429 * the race with preemption/irq.
1430 */
1431void invalidate_bh_lrus_cpu(void)
Minchan Kim8cc621d2021-05-04 18:37:00 -07001432{
1433 struct bh_lru *b;
1434
1435 bh_lru_lock();
Minchan Kim243418e2021-09-24 15:43:47 -07001436 b = this_cpu_ptr(&bh_lrus);
Minchan Kim8cc621d2021-05-04 18:37:00 -07001437 __invalidate_bh_lrus(b);
1438 bh_lru_unlock();
1439}
1440
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441void set_bh_page(struct buffer_head *bh,
1442 struct page *page, unsigned long offset)
1443{
1444 bh->b_page = page;
Eric Sesterhenne827f922006-03-26 18:24:46 +02001445 BUG_ON(offset >= PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446 if (PageHighMem(page))
1447 /*
1448 * This catches illegal uses and preserves the offset:
1449 */
1450 bh->b_data = (char *)(0 + offset);
1451 else
1452 bh->b_data = page_address(page) + offset;
1453}
1454EXPORT_SYMBOL(set_bh_page);
1455
1456/*
1457 * Called when truncating a buffer on a page completely.
1458 */
Mel Gormane7470ee2014-06-04 16:10:29 -07001459
1460/* Bits that are cleared during an invalidate */
1461#define BUFFER_FLAGS_DISCARD \
1462 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1463 1 << BH_Delay | 1 << BH_Unwritten)
1464
Arjan van de Ven858119e2006-01-14 13:20:43 -08001465static void discard_buffer(struct buffer_head * bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001466{
Mel Gormane7470ee2014-06-04 16:10:29 -07001467 unsigned long b_state, b_state_old;
1468
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 lock_buffer(bh);
1470 clear_buffer_dirty(bh);
1471 bh->b_bdev = NULL;
Mel Gormane7470ee2014-06-04 16:10:29 -07001472 b_state = bh->b_state;
1473 for (;;) {
1474 b_state_old = cmpxchg(&bh->b_state, b_state,
1475 (b_state & ~BUFFER_FLAGS_DISCARD));
1476 if (b_state_old == b_state)
1477 break;
1478 b_state = b_state_old;
1479 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 unlock_buffer(bh);
1481}
1482
1483/**
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001484 * block_invalidate_folio - Invalidate part or all of a buffer-backed folio.
1485 * @folio: The folio which is affected.
Lukas Czernerd47992f2013-05-21 23:17:23 -04001486 * @offset: start of the range to invalidate
1487 * @length: length of the range to invalidate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 *
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001489 * block_invalidate_folio() is called when all or part of the folio has been
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001490 * invalidated by a truncate operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001491 *
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001492 * block_invalidate_folio() does not have to release all buffers, but it must
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 * ensure that no dirty buffer is left outside @offset and that no I/O
1494 * is underway against any of the blocks which are outside the truncation
1495 * point. Because the caller is about to free (and possibly reuse) those
1496 * blocks on-disk.
1497 */
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001498void block_invalidate_folio(struct folio *folio, size_t offset, size_t length)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499{
1500 struct buffer_head *head, *bh, *next;
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001501 size_t curr_off = 0;
1502 size_t stop = length + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001504 BUG_ON(!folio_test_locked(folio));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505
Lukas Czernerd47992f2013-05-21 23:17:23 -04001506 /*
1507 * Check for overflow
1508 */
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001509 BUG_ON(stop > folio_size(folio) || stop < length);
Lukas Czernerd47992f2013-05-21 23:17:23 -04001510
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001511 head = folio_buffers(folio);
1512 if (!head)
1513 return;
1514
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 bh = head;
1516 do {
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001517 size_t next_off = curr_off + bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001518 next = bh->b_this_page;
1519
1520 /*
Lukas Czernerd47992f2013-05-21 23:17:23 -04001521 * Are we still fully in range ?
1522 */
1523 if (next_off > stop)
1524 goto out;
1525
1526 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001527 * is this block fully invalidated?
1528 */
1529 if (offset <= curr_off)
1530 discard_buffer(bh);
1531 curr_off = next_off;
1532 bh = next;
1533 } while (bh != head);
1534
1535 /*
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001536 * We release buffers only if the entire folio is being invalidated.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537 * The get_block cached value has been unconditionally invalidated,
1538 * so real IO is not possible anymore.
1539 */
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001540 if (length == folio_size(folio))
1541 filemap_release_folio(folio, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542out:
NeilBrown2ff28e22006-03-26 01:37:18 -08001543 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544}
Matthew Wilcox (Oracle)7ba13ab2022-02-09 20:21:34 +00001545EXPORT_SYMBOL(block_invalidate_folio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001546
Lukas Czernerd47992f2013-05-21 23:17:23 -04001547
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548/*
1549 * We attach and possibly dirty the buffers atomically wrt
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +00001550 * block_dirty_folio() via private_lock. try_to_free_buffers
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 * is already excluded via the page lock.
1552 */
1553void create_empty_buffers(struct page *page,
1554 unsigned long blocksize, unsigned long b_state)
1555{
1556 struct buffer_head *bh, *head, *tail;
1557
Jens Axboe640ab982017-09-27 05:40:16 -06001558 head = alloc_page_buffers(page, blocksize, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 bh = head;
1560 do {
1561 bh->b_state |= b_state;
1562 tail = bh;
1563 bh = bh->b_this_page;
1564 } while (bh);
1565 tail->b_this_page = head;
1566
1567 spin_lock(&page->mapping->private_lock);
1568 if (PageUptodate(page) || PageDirty(page)) {
1569 bh = head;
1570 do {
1571 if (PageDirty(page))
1572 set_buffer_dirty(bh);
1573 if (PageUptodate(page))
1574 set_buffer_uptodate(bh);
1575 bh = bh->b_this_page;
1576 } while (bh != head);
1577 }
Guoqing Jiang45dcfc22020-06-01 21:47:48 -07001578 attach_page_private(page, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579 spin_unlock(&page->mapping->private_lock);
1580}
1581EXPORT_SYMBOL(create_empty_buffers);
1582
Jan Kara29f3ad72016-11-04 18:08:11 +01001583/**
1584 * clean_bdev_aliases: clean a range of buffers in block device
1585 * @bdev: Block device to clean buffers in
1586 * @block: Start of a range of blocks to clean
1587 * @len: Number of blocks to clean
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588 *
Jan Kara29f3ad72016-11-04 18:08:11 +01001589 * We are taking a range of blocks for data and we don't want writeback of any
1590 * buffer-cache aliases starting from return from this function and until the
1591 * moment when something will explicitly mark the buffer dirty (hopefully that
1592 * will not happen until we will free that block ;-) We don't even need to mark
1593 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1594 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1595 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1596 * would confuse anyone who might pick it with bread() afterwards...
1597 *
1598 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1599 * writeout I/O going on against recently-freed buffers. We don't wait on that
1600 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1601 * need to. That happens here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602 */
Jan Kara29f3ad72016-11-04 18:08:11 +01001603void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604{
Jan Kara29f3ad72016-11-04 18:08:11 +01001605 struct inode *bd_inode = bdev->bd_inode;
1606 struct address_space *bd_mapping = bd_inode->i_mapping;
1607 struct pagevec pvec;
1608 pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1609 pgoff_t end;
Jan Karac10f7782017-09-06 16:21:24 -07001610 int i, count;
Jan Kara29f3ad72016-11-04 18:08:11 +01001611 struct buffer_head *bh;
1612 struct buffer_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001613
Jan Kara29f3ad72016-11-04 18:08:11 +01001614 end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman86679822017-11-15 17:37:52 -08001615 pagevec_init(&pvec);
Jan Kara397162f2017-09-06 16:21:43 -07001616 while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
Jan Karac10f7782017-09-06 16:21:24 -07001617 count = pagevec_count(&pvec);
1618 for (i = 0; i < count; i++) {
Jan Kara29f3ad72016-11-04 18:08:11 +01001619 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001620
Jan Kara29f3ad72016-11-04 18:08:11 +01001621 if (!page_has_buffers(page))
1622 continue;
1623 /*
1624 * We use page lock instead of bd_mapping->private_lock
1625 * to pin buffers here since we can afford to sleep and
1626 * it scales better than a global spinlock lock.
1627 */
1628 lock_page(page);
1629 /* Recheck when the page is locked which pins bhs */
1630 if (!page_has_buffers(page))
1631 goto unlock_page;
1632 head = page_buffers(page);
1633 bh = head;
1634 do {
Chandan Rajendra6c006a92016-12-25 19:01:03 +05301635 if (!buffer_mapped(bh) || (bh->b_blocknr < block))
Jan Kara29f3ad72016-11-04 18:08:11 +01001636 goto next;
1637 if (bh->b_blocknr >= block + len)
1638 break;
1639 clear_buffer_dirty(bh);
1640 wait_on_buffer(bh);
1641 clear_buffer_req(bh);
1642next:
1643 bh = bh->b_this_page;
1644 } while (bh != head);
1645unlock_page:
1646 unlock_page(page);
1647 }
1648 pagevec_release(&pvec);
1649 cond_resched();
Jan Karac10f7782017-09-06 16:21:24 -07001650 /* End of range already reached? */
1651 if (index > end || !index)
1652 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001653 }
1654}
Jan Kara29f3ad72016-11-04 18:08:11 +01001655EXPORT_SYMBOL(clean_bdev_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656
1657/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001658 * Size is a power-of-two in the range 512..PAGE_SIZE,
1659 * and the case we care about most is PAGE_SIZE.
1660 *
1661 * So this *could* possibly be written with those
1662 * constraints in mind (relevant mostly if some
1663 * architecture has a slow bit-scan instruction)
1664 */
1665static inline int block_size_bits(unsigned int blocksize)
1666{
1667 return ilog2(blocksize);
1668}
1669
1670static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1671{
1672 BUG_ON(!PageLocked(page));
1673
1674 if (!page_has_buffers(page))
Mark Rutland6aa7de02017-10-23 14:07:29 -07001675 create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits),
1676 b_state);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001677 return page_buffers(page);
1678}
1679
1680/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 * NOTE! All mapped/uptodate combinations are valid:
1682 *
1683 * Mapped Uptodate Meaning
1684 *
1685 * No No "unknown" - must do get_block()
1686 * No Yes "hole" - zero-filled
1687 * Yes No "allocated" - allocated on disk, not read in
1688 * Yes Yes "valid" - allocated and up-to-date in memory.
1689 *
1690 * "Dirty" is valid only with the last case (mapped+uptodate).
1691 */
1692
1693/*
1694 * While block_write_full_page is writing back the dirty buffers under
1695 * the page lock, whoever dirtied the buffers may decide to clean them
1696 * again at any time. We handle that by only looking at the buffer
1697 * state inside lock_buffer().
1698 *
1699 * If block_write_full_page() is called for regular writeback
1700 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1701 * locked buffer. This only can happen if someone has written the buffer
1702 * directly, with submit_bh(). At the address_space level PageWriteback
1703 * prevents this contention from occurring.
Theodore Ts'o6e34eedd2009-04-07 18:12:43 -04001704 *
1705 * If block_write_full_page() is called with wbc->sync_mode ==
Christoph Hellwig70fd7612016-11-01 07:40:10 -06001706 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
Jens Axboe721a9602011-03-09 11:56:30 +01001707 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 */
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001709int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001710 get_block_t *get_block, struct writeback_control *wbc,
1711 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712{
1713 int err;
1714 sector_t block;
1715 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001716 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001717 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 int nr_underway = 0;
Jens Axboe76372412016-11-01 10:00:38 -06001719 int write_flags = wbc_to_write_flags(wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001721 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001723
1724 /*
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +00001725 * Be very careful. We have no exclusion from block_dirty_folio
Linus Torvalds1da177e2005-04-16 15:20:36 -07001726 * here, and the (potentially unmapped) buffers may become dirty at
1727 * any time. If a buffer becomes dirty here after we've inspected it
1728 * then we just miss that fact, and the page stays dirty.
1729 *
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +00001730 * Buffers outside i_size may be dirtied by block_dirty_folio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001731 * handle that here by just cleaning them.
1732 */
1733
Linus Torvalds1da177e2005-04-16 15:20:36 -07001734 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001735 blocksize = bh->b_size;
1736 bbits = block_size_bits(blocksize);
1737
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001738 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001739 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
1741 /*
1742 * Get all the dirty buffers mapped to disk addresses and
1743 * handle any aliases from the underlying blockdev's mapping.
1744 */
1745 do {
1746 if (block > last_block) {
1747 /*
1748 * mapped buffers outside i_size will occur, because
1749 * this page can be outside i_size when there is a
1750 * truncate in progress.
1751 */
1752 /*
1753 * The buffer was zeroed by block_write_full_page()
1754 */
1755 clear_buffer_dirty(bh);
1756 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001757 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1758 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001759 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760 err = get_block(inode, block, bh, 1);
1761 if (err)
1762 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001763 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764 if (buffer_new(bh)) {
1765 /* blockdev mappings never come here */
1766 clear_buffer_new(bh);
Jan Karae64855c2016-11-04 18:08:15 +01001767 clean_bdev_bh_alias(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001768 }
1769 }
1770 bh = bh->b_this_page;
1771 block++;
1772 } while (bh != head);
1773
1774 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 if (!buffer_mapped(bh))
1776 continue;
1777 /*
1778 * If it's a fully non-blocking write attempt and we cannot
1779 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001780 * potentially cause a busy-wait loop from writeback threads
1781 * and kswapd activity, but those code paths have their own
1782 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001783 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001784 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001786 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 redirty_page_for_writepage(wbc, page);
1788 continue;
1789 }
1790 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001791 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 } else {
1793 unlock_buffer(bh);
1794 }
1795 } while ((bh = bh->b_this_page) != head);
1796
1797 /*
1798 * The page and its buffers are protected by PageWriteback(), so we can
1799 * drop the bh refcounts early.
1800 */
1801 BUG_ON(PageWriteback(page));
1802 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803
1804 do {
1805 struct buffer_head *next = bh->b_this_page;
1806 if (buffer_async_write(bh)) {
Christoph Hellwigc75e7072022-03-04 18:55:56 +01001807 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 nr_underway++;
1809 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810 bh = next;
1811 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001812 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001813
1814 err = 0;
1815done:
1816 if (nr_underway == 0) {
1817 /*
1818 * The page was marked dirty, but the buffers were
1819 * clean. Someone wrote them back by hand with
1820 * ll_rw_block/submit_bh. A rare case.
1821 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001822 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001823
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824 /*
1825 * The page and buffer_heads can be released at any time from
1826 * here on.
1827 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 }
1829 return err;
1830
1831recover:
1832 /*
1833 * ENOSPC, or some other error. We may already have added some
1834 * blocks to the file, so we need to write these out to avoid
1835 * exposing stale data.
1836 * The page is currently locked and not marked for writeback
1837 */
1838 bh = head;
1839 /* Recovery: lock and submit the mapped buffers */
1840 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001841 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1842 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001843 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001844 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845 } else {
1846 /*
1847 * The buffer may have been set dirty during
1848 * attachment to a dirty page.
1849 */
1850 clear_buffer_dirty(bh);
1851 }
1852 } while ((bh = bh->b_this_page) != head);
1853 SetPageError(page);
1854 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001855 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001856 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001857 do {
1858 struct buffer_head *next = bh->b_this_page;
1859 if (buffer_async_write(bh)) {
1860 clear_buffer_dirty(bh);
Christoph Hellwigc75e7072022-03-04 18:55:56 +01001861 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001862 nr_underway++;
1863 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 bh = next;
1865 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001866 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 goto done;
1868}
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001869EXPORT_SYMBOL(__block_write_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001870
Nick Pigginafddba42007-10-16 01:25:01 -07001871/*
1872 * If a page has any new buffers, zero them out here, and mark them uptodate
1873 * and dirty so they'll be written out (in order to prevent uninitialised
1874 * block data from leaking). And clear the new bit.
1875 */
1876void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1877{
1878 unsigned int block_start, block_end;
1879 struct buffer_head *head, *bh;
1880
1881 BUG_ON(!PageLocked(page));
1882 if (!page_has_buffers(page))
1883 return;
1884
1885 bh = head = page_buffers(page);
1886 block_start = 0;
1887 do {
1888 block_end = block_start + bh->b_size;
1889
1890 if (buffer_new(bh)) {
1891 if (block_end > from && block_start < to) {
1892 if (!PageUptodate(page)) {
1893 unsigned start, size;
1894
1895 start = max(from, block_start);
1896 size = min(to, block_end) - start;
1897
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001898 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001899 set_buffer_uptodate(bh);
1900 }
1901
1902 clear_buffer_new(bh);
1903 mark_buffer_dirty(bh);
1904 }
1905 }
1906
1907 block_start = block_end;
1908 bh = bh->b_this_page;
1909 } while (bh != head);
1910}
1911EXPORT_SYMBOL(page_zero_new_buffers);
1912
Christoph Hellwigae259a92016-06-21 09:23:11 +10001913static void
1914iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
Christoph Hellwig6d49cc82021-08-10 18:33:05 -07001915 const struct iomap *iomap)
Christoph Hellwigae259a92016-06-21 09:23:11 +10001916{
1917 loff_t offset = block << inode->i_blkbits;
1918
1919 bh->b_bdev = iomap->bdev;
1920
1921 /*
1922 * Block points to offset in file we need to map, iomap contains
1923 * the offset at which the map starts. If the map ends before the
1924 * current block, then do not map the buffer and let the caller
1925 * handle it.
1926 */
1927 BUG_ON(offset >= iomap->offset + iomap->length);
1928
1929 switch (iomap->type) {
1930 case IOMAP_HOLE:
1931 /*
1932 * If the buffer is not up to date or beyond the current EOF,
1933 * we need to mark it as new to ensure sub-block zeroing is
1934 * executed if necessary.
1935 */
1936 if (!buffer_uptodate(bh) ||
1937 (offset >= i_size_read(inode)))
1938 set_buffer_new(bh);
1939 break;
1940 case IOMAP_DELALLOC:
1941 if (!buffer_uptodate(bh) ||
1942 (offset >= i_size_read(inode)))
1943 set_buffer_new(bh);
1944 set_buffer_uptodate(bh);
1945 set_buffer_mapped(bh);
1946 set_buffer_delay(bh);
1947 break;
1948 case IOMAP_UNWRITTEN:
1949 /*
Andreas Gruenbacher3d7b6b212018-06-19 15:10:55 -07001950 * For unwritten regions, we always need to ensure that regions
1951 * in the block we are not writing to are zeroed. Mark the
1952 * buffer as new to ensure this.
Christoph Hellwigae259a92016-06-21 09:23:11 +10001953 */
1954 set_buffer_new(bh);
1955 set_buffer_unwritten(bh);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05001956 fallthrough;
Christoph Hellwigae259a92016-06-21 09:23:11 +10001957 case IOMAP_MAPPED:
Andreas Gruenbacher3d7b6b212018-06-19 15:10:55 -07001958 if ((iomap->flags & IOMAP_F_NEW) ||
1959 offset >= i_size_read(inode))
Christoph Hellwigae259a92016-06-21 09:23:11 +10001960 set_buffer_new(bh);
Andreas Gruenbacher19fe5f62017-10-01 17:55:54 -04001961 bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
1962 inode->i_blkbits;
Christoph Hellwigae259a92016-06-21 09:23:11 +10001963 set_buffer_mapped(bh);
1964 break;
1965 }
1966}
1967
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04001968int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len,
Christoph Hellwig6d49cc82021-08-10 18:33:05 -07001969 get_block_t *get_block, const struct iomap *iomap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001970{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001971 unsigned from = pos & (PAGE_SIZE - 1);
Christoph Hellwigebdec242010-10-06 10:47:23 +02001972 unsigned to = from + len;
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04001973 struct inode *inode = folio->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 unsigned block_start, block_end;
1975 sector_t block;
1976 int err = 0;
1977 unsigned blocksize, bbits;
1978 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
1979
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04001980 BUG_ON(!folio_test_locked(folio));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001981 BUG_ON(from > PAGE_SIZE);
1982 BUG_ON(to > PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983 BUG_ON(from > to);
1984
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04001985 head = create_page_buffers(&folio->page, inode, 0);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001986 blocksize = head->b_size;
1987 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04001989 block = (sector_t)folio->index << (PAGE_SHIFT - bbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001990
1991 for(bh = head, block_start = 0; bh != head || !block_start;
1992 block++, block_start=block_end, bh = bh->b_this_page) {
1993 block_end = block_start + blocksize;
1994 if (block_end <= from || block_start >= to) {
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04001995 if (folio_test_uptodate(folio)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001996 if (!buffer_uptodate(bh))
1997 set_buffer_uptodate(bh);
1998 }
1999 continue;
2000 }
2001 if (buffer_new(bh))
2002 clear_buffer_new(bh);
2003 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002004 WARN_ON(bh->b_size != blocksize);
Christoph Hellwigae259a92016-06-21 09:23:11 +10002005 if (get_block) {
2006 err = get_block(inode, block, bh, 1);
2007 if (err)
2008 break;
2009 } else {
2010 iomap_to_bh(inode, block, bh, iomap);
2011 }
2012
Linus Torvalds1da177e2005-04-16 15:20:36 -07002013 if (buffer_new(bh)) {
Jan Karae64855c2016-11-04 18:08:15 +01002014 clean_bdev_bh_alias(bh);
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04002015 if (folio_test_uptodate(folio)) {
Nick Piggin637aff42007-10-16 01:25:00 -07002016 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002017 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07002018 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002019 continue;
2020 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002021 if (block_end > to || block_start < from)
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04002022 folio_zero_segments(folio,
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002023 to, block_end,
2024 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002025 continue;
2026 }
2027 }
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04002028 if (folio_test_uptodate(folio)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 if (!buffer_uptodate(bh))
2030 set_buffer_uptodate(bh);
2031 continue;
2032 }
2033 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08002034 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002035 (block_start < from || block_end > to)) {
Mike Christiedfec8a12016-06-05 14:31:44 -05002036 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037 *wait_bh++=bh;
2038 }
2039 }
2040 /*
2041 * If we issued read requests - let them complete.
2042 */
2043 while(wait_bh > wait) {
2044 wait_on_buffer(*--wait_bh);
2045 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07002046 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002047 }
Jan Karaf9f07b62011-06-14 00:58:27 +02002048 if (unlikely(err))
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04002049 page_zero_new_buffers(&folio->page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002050 return err;
2051}
Christoph Hellwigae259a92016-06-21 09:23:11 +10002052
2053int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2054 get_block_t *get_block)
2055{
Matthew Wilcox (Oracle)d1bd0b42021-11-03 14:05:47 -04002056 return __block_write_begin_int(page_folio(page), pos, len, get_block,
2057 NULL);
Christoph Hellwigae259a92016-06-21 09:23:11 +10002058}
Christoph Hellwigebdec242010-10-06 10:47:23 +02002059EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002060
2061static int __block_commit_write(struct inode *inode, struct page *page,
2062 unsigned from, unsigned to)
2063{
2064 unsigned block_start, block_end;
2065 int partial = 0;
2066 unsigned blocksize;
2067 struct buffer_head *bh, *head;
2068
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002069 bh = head = page_buffers(page);
2070 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002072 block_start = 0;
2073 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074 block_end = block_start + blocksize;
2075 if (block_end <= from || block_start >= to) {
2076 if (!buffer_uptodate(bh))
2077 partial = 1;
2078 } else {
2079 set_buffer_uptodate(bh);
2080 mark_buffer_dirty(bh);
2081 }
Yang Guo4ebd3ae2021-02-24 12:02:48 -08002082 if (buffer_new(bh))
2083 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002084
2085 block_start = block_end;
2086 bh = bh->b_this_page;
2087 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002088
2089 /*
2090 * If this is a partial write which happened to make all buffers
2091 * uptodate then we can optimize away a bogus readpage() for
2092 * the next read(). Here we 'discover' whether the page went
2093 * uptodate as a result of this (potentially partial) write.
2094 */
2095 if (!partial)
2096 SetPageUptodate(page);
2097 return 0;
2098}
2099
2100/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02002101 * block_write_begin takes care of the basic task of block allocation and
2102 * bringing partial write blocks uptodate first.
2103 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002104 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07002105 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02002106int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2107 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07002108{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002109 pgoff_t index = pos >> PAGE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07002110 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002111 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07002112
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002113 page = grab_cache_page_write_begin(mapping, index, flags);
2114 if (!page)
2115 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07002116
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002117 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07002118 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002119 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002120 put_page(page);
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002121 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07002122 }
2123
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002124 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07002125 return status;
2126}
2127EXPORT_SYMBOL(block_write_begin);
2128
2129int block_write_end(struct file *file, struct address_space *mapping,
2130 loff_t pos, unsigned len, unsigned copied,
2131 struct page *page, void *fsdata)
2132{
2133 struct inode *inode = mapping->host;
2134 unsigned start;
2135
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002136 start = pos & (PAGE_SIZE - 1);
Nick Pigginafddba42007-10-16 01:25:01 -07002137
2138 if (unlikely(copied < len)) {
2139 /*
2140 * The buffers that were written will now be uptodate, so we
2141 * don't have to worry about a readpage reading them and
2142 * overwriting a partial write. However if we have encountered
2143 * a short write and only partially written into a buffer, it
2144 * will not be marked uptodate, so a readpage might come in and
2145 * destroy our partial write.
2146 *
2147 * Do the simplest thing, and just treat any short write to a
2148 * non uptodate page as a zero-length write, and force the
2149 * caller to redo the whole thing.
2150 */
2151 if (!PageUptodate(page))
2152 copied = 0;
2153
2154 page_zero_new_buffers(page, start+copied, start+len);
2155 }
2156 flush_dcache_page(page);
2157
2158 /* This could be a short (even 0-length) commit */
2159 __block_commit_write(inode, page, start, start+copied);
2160
2161 return copied;
2162}
2163EXPORT_SYMBOL(block_write_end);
2164
2165int generic_write_end(struct file *file, struct address_space *mapping,
2166 loff_t pos, unsigned len, unsigned copied,
2167 struct page *page, void *fsdata)
2168{
Christoph Hellwig8af54f22019-06-27 17:28:40 -07002169 struct inode *inode = mapping->host;
2170 loff_t old_size = inode->i_size;
2171 bool i_size_changed = false;
2172
Nick Pigginafddba42007-10-16 01:25:01 -07002173 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
Christoph Hellwig8af54f22019-06-27 17:28:40 -07002174
2175 /*
2176 * No need to use i_size_read() here, the i_size cannot change under us
2177 * because we hold i_rwsem.
2178 *
2179 * But it's important to update i_size while still holding page lock:
2180 * page writeout could otherwise come in and zero beyond i_size.
2181 */
2182 if (pos + copied > inode->i_size) {
2183 i_size_write(inode, pos + copied);
2184 i_size_changed = true;
2185 }
2186
2187 unlock_page(page);
Andreas Gruenbacher7a77dad2019-04-30 08:45:34 -07002188 put_page(page);
Christoph Hellwig8af54f22019-06-27 17:28:40 -07002189
2190 if (old_size < pos)
2191 pagecache_isize_extended(inode, old_size, pos);
2192 /*
2193 * Don't mark the inode dirty under page lock. First, it unnecessarily
2194 * makes the holding time of page lock longer. Second, it forces lock
2195 * ordering of page lock and transaction start for journaling
2196 * filesystems.
2197 */
2198 if (i_size_changed)
2199 mark_inode_dirty(inode);
Andreas Gruenbacher26ddb1f2019-04-30 08:45:33 -07002200 return copied;
Nick Pigginafddba42007-10-16 01:25:01 -07002201}
2202EXPORT_SYMBOL(generic_write_end);
2203
2204/*
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002205 * block_is_partially_uptodate checks whether buffers within a folio are
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002206 * uptodate or not.
2207 *
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002208 * Returns true if all buffers which correspond to the specified part
2209 * of the folio are uptodate.
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002210 */
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002211bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002212{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002213 unsigned block_start, block_end, blocksize;
2214 unsigned to;
2215 struct buffer_head *bh, *head;
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002216 bool ret = true;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002217
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002218 head = folio_buffers(folio);
2219 if (!head)
2220 return false;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002221 blocksize = head->b_size;
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002222 to = min_t(unsigned, folio_size(folio) - from, count);
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002223 to = from + to;
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002224 if (from < blocksize && to > folio_size(folio) - blocksize)
2225 return false;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002226
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002227 bh = head;
2228 block_start = 0;
2229 do {
2230 block_end = block_start + blocksize;
2231 if (block_end > from && block_start < to) {
2232 if (!buffer_uptodate(bh)) {
Matthew Wilcox (Oracle)2e7e80f2022-02-09 20:21:27 +00002233 ret = false;
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002234 break;
2235 }
2236 if (block_end >= to)
2237 break;
2238 }
2239 block_start = block_end;
2240 bh = bh->b_this_page;
2241 } while (bh != head);
2242
2243 return ret;
2244}
2245EXPORT_SYMBOL(block_is_partially_uptodate);
2246
2247/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248 * Generic "read page" function for block devices that have the normal
2249 * get_block functionality. This is most of the block device filesystems.
2250 * Reads the page asynchronously --- the unlock_buffer() and
2251 * set/clear_buffer_uptodate() functions propagate buffer state into the
2252 * page struct once IO has completed.
2253 */
2254int block_read_full_page(struct page *page, get_block_t *get_block)
2255{
2256 struct inode *inode = page->mapping->host;
2257 sector_t iblock, lblock;
2258 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002259 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260 int nr, i;
2261 int fully_mapped = 1;
2262
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002263 head = create_page_buffers(page, inode, 0);
2264 blocksize = head->b_size;
2265 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002266
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002267 iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002268 lblock = (i_size_read(inode)+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002269 bh = head;
2270 nr = 0;
2271 i = 0;
2272
2273 do {
2274 if (buffer_uptodate(bh))
2275 continue;
2276
2277 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002278 int err = 0;
2279
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280 fully_mapped = 0;
2281 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002282 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002283 err = get_block(inode, iblock, bh, 0);
2284 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285 SetPageError(page);
2286 }
2287 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002288 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002289 if (!err)
2290 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291 continue;
2292 }
2293 /*
2294 * get_block() might have updated the buffer
2295 * synchronously
2296 */
2297 if (buffer_uptodate(bh))
2298 continue;
2299 }
2300 arr[nr++] = bh;
2301 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2302
2303 if (fully_mapped)
2304 SetPageMappedToDisk(page);
2305
2306 if (!nr) {
2307 /*
2308 * All buffers are uptodate - we can set the page uptodate
2309 * as well. But not if get_block() returned an error.
2310 */
2311 if (!PageError(page))
2312 SetPageUptodate(page);
2313 unlock_page(page);
2314 return 0;
2315 }
2316
2317 /* Stage two: lock the buffers */
2318 for (i = 0; i < nr; i++) {
2319 bh = arr[i];
2320 lock_buffer(bh);
2321 mark_buffer_async_read(bh);
2322 }
2323
2324 /*
2325 * Stage 3: start the IO. Check for uptodateness
2326 * inside the buffer lock in case another process reading
2327 * the underlying blockdev brought it uptodate (the sct fix).
2328 */
2329 for (i = 0; i < nr; i++) {
2330 bh = arr[i];
2331 if (buffer_uptodate(bh))
2332 end_buffer_async_read(bh, 1);
2333 else
Mike Christie2a222ca2016-06-05 14:31:43 -05002334 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002335 }
2336 return 0;
2337}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002338EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339
2340/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002341 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342 * deal with the hole.
2343 */
Nick Piggin89e10782007-10-16 01:25:07 -07002344int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002345{
2346 struct address_space *mapping = inode->i_mapping;
2347 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002348 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002349 int err;
2350
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002351 err = inode_newsize_ok(inode, size);
2352 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002353 goto out;
2354
Matthew Wilcox (Oracle)d7414ba2022-02-20 22:28:03 -05002355 err = pagecache_write_begin(NULL, mapping, size, 0, 0, &page, &fsdata);
Nick Piggin89e10782007-10-16 01:25:07 -07002356 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002357 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002358
Nick Piggin89e10782007-10-16 01:25:07 -07002359 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2360 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002361
Linus Torvalds1da177e2005-04-16 15:20:36 -07002362out:
2363 return err;
2364}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002365EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002367static int cont_expand_zero(struct file *file, struct address_space *mapping,
2368 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002369{
Nick Piggin89e10782007-10-16 01:25:07 -07002370 struct inode *inode = mapping->host;
Fabian Frederick93407472017-02-27 14:28:32 -08002371 unsigned int blocksize = i_blocksize(inode);
Nick Piggin89e10782007-10-16 01:25:07 -07002372 struct page *page;
2373 void *fsdata;
2374 pgoff_t index, curidx;
2375 loff_t curpos;
2376 unsigned zerofrom, offset, len;
2377 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002378
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002379 index = pos >> PAGE_SHIFT;
2380 offset = pos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002381
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002382 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2383 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002384 if (zerofrom & (blocksize-1)) {
2385 *bytes |= (blocksize-1);
2386 (*bytes)++;
2387 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002388 len = PAGE_SIZE - zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002389
Tetsuo Handac718a972017-05-08 15:58:59 -07002390 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2391 &page, &fsdata);
Nick Piggin89e10782007-10-16 01:25:07 -07002392 if (err)
2393 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002394 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002395 err = pagecache_write_end(file, mapping, curpos, len, len,
2396 page, fsdata);
2397 if (err < 0)
2398 goto out;
2399 BUG_ON(err != len);
2400 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002401
2402 balance_dirty_pages_ratelimited(mapping);
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002403
Davidlohr Bueso08d405c2019-01-03 15:28:58 -08002404 if (fatal_signal_pending(current)) {
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002405 err = -EINTR;
2406 goto out;
2407 }
Nick Piggin89e10782007-10-16 01:25:07 -07002408 }
2409
2410 /* page covers the boundary, find the boundary offset */
2411 if (index == curidx) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002412 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002413 /* if we will expand the thing last block will be filled */
2414 if (offset <= zerofrom) {
2415 goto out;
2416 }
2417 if (zerofrom & (blocksize-1)) {
2418 *bytes |= (blocksize-1);
2419 (*bytes)++;
2420 }
2421 len = offset - zerofrom;
2422
Tetsuo Handac718a972017-05-08 15:58:59 -07002423 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2424 &page, &fsdata);
Nick Piggin89e10782007-10-16 01:25:07 -07002425 if (err)
2426 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002427 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002428 err = pagecache_write_end(file, mapping, curpos, len, len,
2429 page, fsdata);
2430 if (err < 0)
2431 goto out;
2432 BUG_ON(err != len);
2433 err = 0;
2434 }
2435out:
2436 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002437}
2438
Linus Torvalds1da177e2005-04-16 15:20:36 -07002439/*
2440 * For moronic filesystems that do not allow holes in file.
2441 * We may have to extend the file.
2442 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002443int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002444 loff_t pos, unsigned len, unsigned flags,
2445 struct page **pagep, void **fsdata,
2446 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002447{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448 struct inode *inode = mapping->host;
Fabian Frederick93407472017-02-27 14:28:32 -08002449 unsigned int blocksize = i_blocksize(inode);
2450 unsigned int zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002451 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452
Nick Piggin89e10782007-10-16 01:25:07 -07002453 err = cont_expand_zero(file, mapping, pos, bytes);
2454 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002455 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002456
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002457 zerofrom = *bytes & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002458 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2459 *bytes |= (blocksize-1);
2460 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002461 }
2462
Christoph Hellwig155130a2010-06-04 11:29:58 +02002463 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002464}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002465EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466
Linus Torvalds1da177e2005-04-16 15:20:36 -07002467int block_commit_write(struct page *page, unsigned from, unsigned to)
2468{
2469 struct inode *inode = page->mapping->host;
2470 __block_commit_write(inode,page,from,to);
2471 return 0;
2472}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002473EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002474
David Chinner54171692007-07-19 17:39:55 +10002475/*
2476 * block_page_mkwrite() is not allowed to change the file size as it gets
2477 * called from a page fault handler when a page is first dirtied. Hence we must
2478 * be careful to check for EOF conditions here. We set the page up correctly
2479 * for a written page which means we get ENOSPC checking when writing into
2480 * holes and correct delalloc and unwritten extent mapping on filesystems that
2481 * support these features.
2482 *
2483 * We are not allowed to take the i_mutex here so we have to play games to
2484 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002485 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002486 * page lock we can determine safely if the page is beyond EOF. If it is not
2487 * beyond EOF, then the page is guaranteed safe against truncation until we
2488 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002489 *
Jan Kara14da9202012-06-12 16:20:37 +02002490 * Direct callers of this function should protect against filesystem freezing
Ross Zwisler5c500022015-10-13 16:51:02 -06002491 * using sb_start_pagefault() - sb_end_pagefault() functions.
David Chinner54171692007-07-19 17:39:55 +10002492 */
Ross Zwisler5c500022015-10-13 16:51:02 -06002493int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
Jan Kara24da4fa2011-05-24 00:23:34 +02002494 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002495{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002496 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002497 struct inode *inode = file_inode(vma->vm_file);
David Chinner54171692007-07-19 17:39:55 +10002498 unsigned long end;
2499 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002500 int ret;
David Chinner54171692007-07-19 17:39:55 +10002501
2502 lock_page(page);
2503 size = i_size_read(inode);
2504 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002505 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002506 /* We overload EFAULT to mean page got truncated */
2507 ret = -EFAULT;
2508 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002509 }
2510
2511 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002512 if (((page->index + 1) << PAGE_SHIFT) > size)
2513 end = size & ~PAGE_MASK;
David Chinner54171692007-07-19 17:39:55 +10002514 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002515 end = PAGE_SIZE;
David Chinner54171692007-07-19 17:39:55 +10002516
Christoph Hellwigebdec242010-10-06 10:47:23 +02002517 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002518 if (!ret)
2519 ret = block_commit_write(page, 0, end);
2520
Jan Kara24da4fa2011-05-24 00:23:34 +02002521 if (unlikely(ret < 0))
2522 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002523 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002524 wait_for_stable_page(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002525 return 0;
2526out_unlock:
2527 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002528 return ret;
2529}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002530EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002531
2532/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002533 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002534 * immediately, while under the page lock. So it needs a special end_io
2535 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002536 */
2537static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2538{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002539 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002540}
2541
2542/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002543 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2544 * the page (converting it to circular linked list and taking care of page
2545 * dirty races).
2546 */
2547static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2548{
2549 struct buffer_head *bh;
2550
2551 BUG_ON(!PageLocked(page));
2552
2553 spin_lock(&page->mapping->private_lock);
2554 bh = head;
2555 do {
2556 if (PageDirty(page))
2557 set_buffer_dirty(bh);
2558 if (!bh->b_this_page)
2559 bh->b_this_page = head;
2560 bh = bh->b_this_page;
2561 } while (bh != head);
Guoqing Jiang45dcfc22020-06-01 21:47:48 -07002562 attach_page_private(page, head);
Nick Piggin03158cd2007-10-16 01:25:25 -07002563 spin_unlock(&page->mapping->private_lock);
2564}
2565
2566/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002567 * On entry, the page is fully not uptodate.
2568 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002569 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002571int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002572 loff_t pos, unsigned len, unsigned flags,
2573 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574 get_block_t *get_block)
2575{
Nick Piggin03158cd2007-10-16 01:25:25 -07002576 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577 const unsigned blkbits = inode->i_blkbits;
2578 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002579 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002580 struct page *page;
2581 pgoff_t index;
2582 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002584 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 int ret = 0;
2588 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002590 index = pos >> PAGE_SHIFT;
2591 from = pos & (PAGE_SIZE - 1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002592 to = from + len;
2593
Nick Piggin54566b22009-01-04 12:00:53 -08002594 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002595 if (!page)
2596 return -ENOMEM;
2597 *pagep = page;
2598 *fsdata = NULL;
2599
2600 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002601 ret = __block_write_begin(page, pos, len, get_block);
2602 if (unlikely(ret))
2603 goto out_release;
2604 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002605 }
Nick Piggina4b06722007-10-16 01:24:48 -07002606
Linus Torvalds1da177e2005-04-16 15:20:36 -07002607 if (PageMappedToDisk(page))
2608 return 0;
2609
Nick Piggina4b06722007-10-16 01:24:48 -07002610 /*
2611 * Allocate buffers so that we can keep track of state, and potentially
2612 * attach them to the page if an error occurs. In the common case of
2613 * no error, they will just be freed again without ever being attached
2614 * to the page (which is all OK, because we're under the page lock).
2615 *
2616 * Be careful: the buffer linked list is a NULL terminated one, rather
2617 * than the circular one we're used to.
2618 */
Jens Axboe640ab982017-09-27 05:40:16 -06002619 head = alloc_page_buffers(page, blocksize, false);
Nick Piggin03158cd2007-10-16 01:25:25 -07002620 if (!head) {
2621 ret = -ENOMEM;
2622 goto out_release;
2623 }
Nick Piggina4b06722007-10-16 01:24:48 -07002624
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002625 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626
2627 /*
2628 * We loop across all blocks in the page, whether or not they are
2629 * part of the affected region. This is so we can discover if the
2630 * page is fully mapped-to-disk.
2631 */
Nick Piggina4b06722007-10-16 01:24:48 -07002632 for (block_start = 0, block_in_page = 0, bh = head;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002633 block_start < PAGE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002634 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 int create;
2636
Nick Piggina4b06722007-10-16 01:24:48 -07002637 block_end = block_start + blocksize;
2638 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 create = 1;
2640 if (block_start >= to)
2641 create = 0;
2642 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002643 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 if (ret)
2645 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002646 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002648 if (buffer_new(bh))
Jan Karae64855c2016-11-04 18:08:15 +01002649 clean_bdev_bh_alias(bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002650 if (PageUptodate(page)) {
2651 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002652 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002653 }
2654 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002655 zero_user_segments(page, block_start, from,
2656 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002657 continue;
2658 }
Nick Piggina4b06722007-10-16 01:24:48 -07002659 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 continue; /* reiserfs does this */
2661 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002662 lock_buffer(bh);
2663 bh->b_end_io = end_buffer_read_nobh;
Mike Christie2a222ca2016-06-05 14:31:43 -05002664 submit_bh(REQ_OP_READ, 0, bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002665 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002666 }
2667 }
2668
2669 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670 /*
2671 * The page is locked, so these buffers are protected from
2672 * any VM or truncate activity. Hence we don't need to care
2673 * for the buffer_head refcounts.
2674 */
Nick Piggina4b06722007-10-16 01:24:48 -07002675 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002676 wait_on_buffer(bh);
2677 if (!buffer_uptodate(bh))
2678 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 }
2680 if (ret)
2681 goto failed;
2682 }
2683
2684 if (is_mapped_to_disk)
2685 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686
Nick Piggin03158cd2007-10-16 01:25:25 -07002687 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002688
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689 return 0;
2690
2691failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002692 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002694 * Error recovery is a bit difficult. We need to zero out blocks that
2695 * were newly allocated, and dirty them to ensure they get written out.
2696 * Buffers need to be attached to the page at this point, otherwise
2697 * the handling of potential IO errors during writeout would be hard
2698 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002699 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002700 attach_nobh_buffers(page, head);
2701 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002702
Nick Piggin03158cd2007-10-16 01:25:25 -07002703out_release:
2704 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002705 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002706 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002707
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002708 return ret;
2709}
Nick Piggin03158cd2007-10-16 01:25:25 -07002710EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711
Nick Piggin03158cd2007-10-16 01:25:25 -07002712int nobh_write_end(struct file *file, struct address_space *mapping,
2713 loff_t pos, unsigned len, unsigned copied,
2714 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715{
2716 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002717 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002718 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002719 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002721 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002722 attach_nobh_buffers(page, head);
2723 if (page_has_buffers(page))
2724 return generic_write_end(file, mapping, pos, len,
2725 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002726
Nick Piggin22c8ca72007-02-20 13:58:09 -08002727 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002729 if (pos+copied > inode->i_size) {
2730 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 mark_inode_dirty(inode);
2732 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002733
2734 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002735 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002736
Nick Piggin03158cd2007-10-16 01:25:25 -07002737 while (head) {
2738 bh = head;
2739 head = head->b_this_page;
2740 free_buffer_head(bh);
2741 }
2742
2743 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744}
Nick Piggin03158cd2007-10-16 01:25:25 -07002745EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746
2747/*
2748 * nobh_writepage() - based on block_full_write_page() except
2749 * that it tries to operate without attaching bufferheads to
2750 * the page.
2751 */
2752int nobh_writepage(struct page *page, get_block_t *get_block,
2753 struct writeback_control *wbc)
2754{
2755 struct inode * const inode = page->mapping->host;
2756 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002757 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 int ret;
2760
2761 /* Is the page fully inside i_size? */
2762 if (page->index < end_index)
2763 goto out;
2764
2765 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002766 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 if (page->index >= end_index+1 || !offset) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768 unlock_page(page);
2769 return 0; /* don't care */
2770 }
2771
2772 /*
2773 * The page straddles i_size. It must be zeroed out on each and every
2774 * writepage invocation because it may be mmapped. "A file is mapped
2775 * in multiples of the page size. For a file that is not a multiple of
2776 * the page size, the remaining memory is zeroed when mapped, and
2777 * writes to that region are not written out to the file."
2778 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002779 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002780out:
2781 ret = mpage_writepage(page, get_block, wbc);
2782 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002783 ret = __block_write_full_page(inode, page, get_block, wbc,
2784 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 return ret;
2786}
2787EXPORT_SYMBOL(nobh_writepage);
2788
Nick Piggin03158cd2007-10-16 01:25:25 -07002789int nobh_truncate_page(struct address_space *mapping,
2790 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002792 pgoff_t index = from >> PAGE_SHIFT;
2793 unsigned offset = from & (PAGE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002794 unsigned blocksize;
2795 sector_t iblock;
2796 unsigned length, pos;
2797 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002798 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002799 struct buffer_head map_bh;
2800 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002801
Fabian Frederick93407472017-02-27 14:28:32 -08002802 blocksize = i_blocksize(inode);
Nick Piggin03158cd2007-10-16 01:25:25 -07002803 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002804
Nick Piggin03158cd2007-10-16 01:25:25 -07002805 /* Block boundary? Nothing to do */
2806 if (!length)
2807 return 0;
2808
2809 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002810 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Nick Piggin03158cd2007-10-16 01:25:25 -07002811
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002813 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002814 if (!page)
2815 goto out;
2816
Nick Piggin03158cd2007-10-16 01:25:25 -07002817 if (page_has_buffers(page)) {
2818has_buffers:
2819 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002820 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002821 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002823
2824 /* Find the buffer that contains "offset" */
2825 pos = blocksize;
2826 while (offset >= pos) {
2827 iblock++;
2828 pos += blocksize;
2829 }
2830
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002831 map_bh.b_size = blocksize;
2832 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002833 err = get_block(inode, iblock, &map_bh, 0);
2834 if (err)
2835 goto unlock;
2836 /* unmapped? It's a hole - nothing to do */
2837 if (!buffer_mapped(&map_bh))
2838 goto unlock;
2839
2840 /* Ok, it's mapped. Make sure it's up-to-date */
2841 if (!PageUptodate(page)) {
2842 err = mapping->a_ops->readpage(NULL, page);
2843 if (err) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002844 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002845 goto out;
2846 }
2847 lock_page(page);
2848 if (!PageUptodate(page)) {
2849 err = -EIO;
2850 goto unlock;
2851 }
2852 if (page_has_buffers(page))
2853 goto has_buffers;
2854 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002855 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002856 set_page_dirty(page);
2857 err = 0;
2858
2859unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002860 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002861 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002862out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002863 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002864}
2865EXPORT_SYMBOL(nobh_truncate_page);
2866
2867int block_truncate_page(struct address_space *mapping,
2868 loff_t from, get_block_t *get_block)
2869{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002870 pgoff_t index = from >> PAGE_SHIFT;
2871 unsigned offset = from & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002872 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002873 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874 unsigned length, pos;
2875 struct inode *inode = mapping->host;
2876 struct page *page;
2877 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878 int err;
2879
Fabian Frederick93407472017-02-27 14:28:32 -08002880 blocksize = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881 length = offset & (blocksize - 1);
2882
2883 /* Block boundary? Nothing to do */
2884 if (!length)
2885 return 0;
2886
2887 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002888 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002889
2890 page = grab_cache_page(mapping, index);
2891 err = -ENOMEM;
2892 if (!page)
2893 goto out;
2894
2895 if (!page_has_buffers(page))
2896 create_empty_buffers(page, blocksize, 0);
2897
2898 /* Find the buffer that contains "offset" */
2899 bh = page_buffers(page);
2900 pos = blocksize;
2901 while (offset >= pos) {
2902 bh = bh->b_this_page;
2903 iblock++;
2904 pos += blocksize;
2905 }
2906
2907 err = 0;
2908 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002909 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002910 err = get_block(inode, iblock, bh, 0);
2911 if (err)
2912 goto unlock;
2913 /* unmapped? It's a hole - nothing to do */
2914 if (!buffer_mapped(bh))
2915 goto unlock;
2916 }
2917
2918 /* Ok, it's mapped. Make sure it's up-to-date */
2919 if (PageUptodate(page))
2920 set_buffer_uptodate(bh);
2921
David Chinner33a266d2007-02-12 00:51:41 -08002922 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923 err = -EIO;
Mike Christiedfec8a12016-06-05 14:31:44 -05002924 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 wait_on_buffer(bh);
2926 /* Uhhuh. Read error. Complain and punt. */
2927 if (!buffer_uptodate(bh))
2928 goto unlock;
2929 }
2930
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002931 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002932 mark_buffer_dirty(bh);
2933 err = 0;
2934
2935unlock:
2936 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002937 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938out:
2939 return err;
2940}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002941EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
2943/*
2944 * The generic ->writepage function for buffer-backed address_spaces
2945 */
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002946int block_write_full_page(struct page *page, get_block_t *get_block,
2947 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002948{
2949 struct inode * const inode = page->mapping->host;
2950 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002951 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002952 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002953
2954 /* Is the page fully inside i_size? */
2955 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04002956 return __block_write_full_page(inode, page, get_block, wbc,
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002957 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002958
2959 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002960 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002961 if (page->index >= end_index+1 || !offset) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002962 unlock_page(page);
2963 return 0; /* don't care */
2964 }
2965
2966 /*
2967 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05002968 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07002969 * in multiples of the page size. For a file that is not a multiple of
2970 * the page size, the remaining memory is zeroed when mapped, and
2971 * writes to that region are not written out to the file."
2972 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002973 zero_user_segment(page, offset, PAGE_SIZE);
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002974 return __block_write_full_page(inode, page, get_block, wbc,
2975 end_buffer_async_write);
Chris Mason35c80d52009-04-15 13:22:38 -04002976}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002977EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04002978
Linus Torvalds1da177e2005-04-16 15:20:36 -07002979sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
2980 get_block_t *get_block)
2981{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 struct inode *inode = mapping->host;
Alexander Potapenko2a527d62017-07-05 00:56:21 -04002983 struct buffer_head tmp = {
2984 .b_size = i_blocksize(inode),
2985 };
2986
Linus Torvalds1da177e2005-04-16 15:20:36 -07002987 get_block(inode, block, &tmp, 0);
2988 return tmp.b_blocknr;
2989}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002990EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02002992static void end_bio_bh_io_sync(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993{
2994 struct buffer_head *bh = bio->bi_private;
2995
Jens Axboeb7c44ed2015-07-24 12:37:59 -06002996 if (unlikely(bio_flagged(bio, BIO_QUIET)))
Keith Mannthey08bafc02008-11-25 10:24:35 +01002997 set_bit(BH_Quiet, &bh->b_state);
2998
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02002999 bh->b_end_io(bh, !bio->bi_status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003000 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001}
3002
Mike Christie2a222ca2016-06-05 14:31:43 -05003003static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Christoph Hellwigc75e7072022-03-04 18:55:56 +01003004 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005{
3006 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003007
3008 BUG_ON(!buffer_locked(bh));
3009 BUG_ON(!buffer_mapped(bh));
3010 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04003011 BUG_ON(buffer_delay(bh));
3012 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003013
Jens Axboe48fd4f92008-08-22 10:00:36 +02003014 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02003015 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07003016 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003017 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018 clear_buffer_write_io_error(bh);
3019
Christoph Hellwig07888c662022-01-24 10:11:05 +01003020 if (buffer_meta(bh))
3021 op_flags |= REQ_META;
3022 if (buffer_prio(bh))
3023 op_flags |= REQ_PRIO;
3024
3025 bio = bio_alloc(bh->b_bdev, 1, op | op_flags, GFP_NOIO);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026
Eric Biggers4f74d152020-07-02 01:56:07 +00003027 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
3028
Kent Overstreet4f024f32013-10-11 15:44:27 -07003029 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030
Kent Overstreet6cf66b42014-12-22 12:48:42 +01003031 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3032 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033
3034 bio->bi_end_io = end_bio_bh_io_sync;
3035 bio->bi_private = bh;
3036
Ming Lei83c9c542020-01-05 09:41:14 +08003037 /* Take care of bh's that straddle the end of the device */
3038 guard_bio_eod(bio);
3039
Dennis Zhoufd42df32018-12-05 12:10:34 -05003040 if (wbc) {
3041 wbc_init_bio(wbc, bio);
Tejun Heo34e51a52019-06-27 13:39:49 -07003042 wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
Dennis Zhoufd42df32018-12-05 12:10:34 -05003043 }
3044
Mike Christie4e49ea42016-06-05 14:31:41 -05003045 submit_bio(bio);
Julia Lawallf6454b02015-05-26 21:59:53 +02003046 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047}
Tejun Heobafc0db2015-06-02 08:37:23 -06003048
Eric Biggers020c28332017-03-25 21:02:18 -07003049int submit_bh(int op, int op_flags, struct buffer_head *bh)
Tejun Heobafc0db2015-06-02 08:37:23 -06003050{
Christoph Hellwigc75e7072022-03-04 18:55:56 +01003051 return submit_bh_wbc(op, op_flags, bh, NULL);
Darrick J. Wong7136851112013-04-29 15:07:25 -07003052}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003053EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054
3055/**
3056 * ll_rw_block: low-level access to block devices (DEPRECATED)
Mike Christiedfec8a12016-06-05 14:31:44 -05003057 * @op: whether to %READ or %WRITE
Christoph Hellwigef295ec2016-10-28 08:48:16 -06003058 * @op_flags: req_flag_bits
Linus Torvalds1da177e2005-04-16 15:20:36 -07003059 * @nr: number of &struct buffer_heads in the array
3060 * @bhs: array of pointers to &struct buffer_head
3061 *
Jan Karaa7662232005-09-06 15:19:10 -07003062 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
Christoph Hellwig70246282016-07-19 11:28:41 +02003063 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3064 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3065 * %REQ_RAHEAD.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066 *
3067 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003068 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3069 * request, and any buffer that appears to be up-to-date when doing read
3070 * request. Further it marks as clean buffers that are processed for
3071 * writing (the buffer cache won't assume that they are actually clean
3072 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 *
3074 * ll_rw_block sets b_end_io to simple completion handler that marks
Masanari Iidae2278672014-02-18 22:54:36 +09003075 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
Linus Torvalds1da177e2005-04-16 15:20:36 -07003076 * any waiters.
3077 *
3078 * All of the buffers must be for the same device, and must also be a
3079 * multiple of the current approved size for the device.
3080 */
Mike Christiedfec8a12016-06-05 14:31:44 -05003081void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082{
3083 int i;
3084
3085 for (i = 0; i < nr; i++) {
3086 struct buffer_head *bh = bhs[i];
3087
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003088 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003089 continue;
Mike Christiedfec8a12016-06-05 14:31:44 -05003090 if (op == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003092 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003093 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003094 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095 continue;
3096 }
3097 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003098 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003099 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003100 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003101 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003102 continue;
3103 }
3104 }
3105 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003106 }
3107}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003108EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003109
Mike Christie2a222ca2016-06-05 14:31:43 -05003110void write_dirty_buffer(struct buffer_head *bh, int op_flags)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003111{
3112 lock_buffer(bh);
3113 if (!test_clear_buffer_dirty(bh)) {
3114 unlock_buffer(bh);
3115 return;
3116 }
3117 bh->b_end_io = end_buffer_write_sync;
3118 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05003119 submit_bh(REQ_OP_WRITE, op_flags, bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003120}
3121EXPORT_SYMBOL(write_dirty_buffer);
3122
Linus Torvalds1da177e2005-04-16 15:20:36 -07003123/*
3124 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3125 * and then start new I/O and then wait upon it. The caller must have a ref on
3126 * the buffer_head.
3127 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003128int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129{
3130 int ret = 0;
3131
3132 WARN_ON(atomic_read(&bh->b_count) < 1);
3133 lock_buffer(bh);
3134 if (test_clear_buffer_dirty(bh)) {
Xianting Tian377254b2020-07-31 12:10:25 -04003135 /*
3136 * The bh should be mapped, but it might not be if the
3137 * device was hot-removed. Not much we can do but fail the I/O.
3138 */
3139 if (!buffer_mapped(bh)) {
3140 unlock_buffer(bh);
3141 return -EIO;
3142 }
3143
Linus Torvalds1da177e2005-04-16 15:20:36 -07003144 get_bh(bh);
3145 bh->b_end_io = end_buffer_write_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003146 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003148 if (!ret && !buffer_uptodate(bh))
3149 ret = -EIO;
3150 } else {
3151 unlock_buffer(bh);
3152 }
3153 return ret;
3154}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003155EXPORT_SYMBOL(__sync_dirty_buffer);
3156
3157int sync_dirty_buffer(struct buffer_head *bh)
3158{
Christoph Hellwig70fd7612016-11-01 07:40:10 -06003159 return __sync_dirty_buffer(bh, REQ_SYNC);
Christoph Hellwig87e99512010-08-11 17:05:45 +02003160}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003161EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162
3163/*
3164 * try_to_free_buffers() checks if all the buffers on this particular page
3165 * are unused, and releases them if so.
3166 *
3167 * Exclusion against try_to_free_buffers may be obtained by either
3168 * locking the page or by holding its mapping's private_lock.
3169 *
3170 * If the page is dirty but all the buffers are clean then we need to
3171 * be sure to mark the page clean as well. This is because the page
3172 * may be against a block device, and a later reattachment of buffers
3173 * to a dirty page will set *all* buffers dirty. Which would corrupt
3174 * filesystem data on the same device.
3175 *
3176 * The same applies to regular filesystem pages: if all the buffers are
3177 * clean then we set the page clean and proceed. To do that, we require
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +00003178 * total exclusion from block_dirty_folio(). That is obtained with
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179 * private_lock.
3180 *
3181 * try_to_free_buffers() is non-blocking.
3182 */
3183static inline int buffer_busy(struct buffer_head *bh)
3184{
3185 return atomic_read(&bh->b_count) |
3186 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3187}
3188
3189static int
3190drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3191{
3192 struct buffer_head *head = page_buffers(page);
3193 struct buffer_head *bh;
3194
3195 bh = head;
3196 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003197 if (buffer_busy(bh))
3198 goto failed;
3199 bh = bh->b_this_page;
3200 } while (bh != head);
3201
3202 do {
3203 struct buffer_head *next = bh->b_this_page;
3204
Jan Kara535ee2f2008-02-08 04:21:59 -08003205 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003206 __remove_assoc_queue(bh);
3207 bh = next;
3208 } while (bh != head);
3209 *buffers_to_free = head;
Guoqing Jiang45dcfc22020-06-01 21:47:48 -07003210 detach_page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003211 return 1;
3212failed:
3213 return 0;
3214}
3215
3216int try_to_free_buffers(struct page *page)
3217{
3218 struct address_space * const mapping = page->mapping;
3219 struct buffer_head *buffers_to_free = NULL;
3220 int ret = 0;
3221
3222 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003223 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003224 return 0;
3225
3226 if (mapping == NULL) { /* can this still happen? */
3227 ret = drop_buffers(page, &buffers_to_free);
3228 goto out;
3229 }
3230
3231 spin_lock(&mapping->private_lock);
3232 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003233
3234 /*
3235 * If the filesystem writes its buffers by hand (eg ext3)
3236 * then we can have clean buffers against a dirty page. We
3237 * clean the page here; otherwise the VM will never notice
3238 * that the filesystem did any IO at all.
3239 *
3240 * Also, during truncate, discard_buffer will have marked all
3241 * the page's buffers clean. We discover that here and clean
3242 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003243 *
3244 * private_lock must be held over this entire operation in order
Matthew Wilcox (Oracle)e6219002022-02-09 20:22:12 +00003245 * to synchronise against block_dirty_folio and prevent the
Nick Piggin87df7242007-01-30 14:36:27 +11003246 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003247 */
Tejun Heo11f81be2015-05-22 17:13:15 -04003248 if (ret)
3249 cancel_dirty_page(page);
Nick Piggin87df7242007-01-30 14:36:27 +11003250 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003251out:
3252 if (buffers_to_free) {
3253 struct buffer_head *bh = buffers_to_free;
3254
3255 do {
3256 struct buffer_head *next = bh->b_this_page;
3257 free_buffer_head(bh);
3258 bh = next;
3259 } while (bh != buffers_to_free);
3260 }
3261 return ret;
3262}
3263EXPORT_SYMBOL(try_to_free_buffers);
3264
Linus Torvalds1da177e2005-04-16 15:20:36 -07003265/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003266 * Buffer-head allocation
3267 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003268static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003269
3270/*
3271 * Once the number of bh's in the machine exceeds this level, we start
3272 * stripping them in writeback.
3273 */
Zhang Yanfei43be5942013-02-22 16:35:46 -08003274static unsigned long max_buffer_heads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003275
3276int buffer_heads_over_limit;
3277
3278struct bh_accounting {
3279 int nr; /* Number of live bh's */
3280 int ratelimit; /* Limit cacheline bouncing */
3281};
3282
3283static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3284
3285static void recalc_bh_state(void)
3286{
3287 int i;
3288 int tot = 0;
3289
Christoph Lameteree1be862010-12-06 11:40:05 -06003290 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003291 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003292 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003293 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003294 tot += per_cpu(bh_accounting, i).nr;
3295 buffer_heads_over_limit = (tot > max_buffer_heads);
3296}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003297
Al Virodd0fc662005-10-07 07:46:04 +01003298struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003300 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003301 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003302 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +01003303 spin_lock_init(&ret->b_uptodate_lock);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003304 preempt_disable();
3305 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003306 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003307 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003308 }
3309 return ret;
3310}
3311EXPORT_SYMBOL(alloc_buffer_head);
3312
3313void free_buffer_head(struct buffer_head *bh)
3314{
3315 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3316 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003317 preempt_disable();
3318 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003319 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003320 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321}
3322EXPORT_SYMBOL(free_buffer_head);
3323
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003324static int buffer_exit_cpu_dead(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325{
3326 int i;
3327 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3328
3329 for (i = 0; i < BH_LRU_SIZE; i++) {
3330 brelse(b->bhs[i]);
3331 b->bhs[i] = NULL;
3332 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003333 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003334 per_cpu(bh_accounting, cpu).nr = 0;
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003335 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003336}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003337
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003338/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003339 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003340 * @bh: struct buffer_head
3341 *
3342 * Return true if the buffer is up-to-date and false,
3343 * with the buffer locked, if not.
3344 */
3345int bh_uptodate_or_lock(struct buffer_head *bh)
3346{
3347 if (!buffer_uptodate(bh)) {
3348 lock_buffer(bh);
3349 if (!buffer_uptodate(bh))
3350 return 0;
3351 unlock_buffer(bh);
3352 }
3353 return 1;
3354}
3355EXPORT_SYMBOL(bh_uptodate_or_lock);
3356
3357/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003358 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003359 * @bh: struct buffer_head
3360 *
3361 * Returns zero on success and -EIO on error.
3362 */
3363int bh_submit_read(struct buffer_head *bh)
3364{
3365 BUG_ON(!buffer_locked(bh));
3366
3367 if (buffer_uptodate(bh)) {
3368 unlock_buffer(bh);
3369 return 0;
3370 }
3371
3372 get_bh(bh);
3373 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003374 submit_bh(REQ_OP_READ, 0, bh);
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003375 wait_on_buffer(bh);
3376 if (buffer_uptodate(bh))
3377 return 0;
3378 return -EIO;
3379}
3380EXPORT_SYMBOL(bh_submit_read);
3381
Linus Torvalds1da177e2005-04-16 15:20:36 -07003382void __init buffer_init(void)
3383{
Zhang Yanfei43be5942013-02-22 16:35:46 -08003384 unsigned long nrpages;
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003385 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003386
Christoph Lameterb98938c2008-02-04 22:28:36 -08003387 bh_cachep = kmem_cache_create("buffer_head",
3388 sizeof(struct buffer_head), 0,
3389 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3390 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003391 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392
3393 /*
3394 * Limit the bh occupancy to 10% of ZONE_NORMAL
3395 */
3396 nrpages = (nr_free_buffer_pages() * 10) / 100;
3397 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003398 ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3399 NULL, buffer_exit_cpu_dead);
3400 WARN_ON(ret < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003401}