blob: d805dddedf5140544c6ef093ebcb5cc1902a4d93 [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>
Eric Biggers0957cd62022-12-23 12:36:37 -080051#include <linux/fsverity.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070052
Ben Dooks2b211dc2019-11-30 17:49:18 -080053#include "internal.h"
54
Linus Torvalds1da177e2005-04-16 15:20:36 -070055static int fsync_buffers_list(spinlock_t *lock, struct list_head *list);
Mike Christie2a222ca2016-06-05 14:31:43 -050056static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Jens Axboe8e8f9292017-06-27 09:30:05 -060057 enum rw_hint hint, struct writeback_control *wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -070058
59#define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers)
60
Tejun Heof0059af2013-01-11 13:06:35 -080061inline void touch_buffer(struct buffer_head *bh)
62{
Tejun Heo5305cb82013-01-11 13:06:36 -080063 trace_block_touch_buffer(bh);
Tejun Heof0059af2013-01-11 13:06:35 -080064 mark_page_accessed(bh->b_page);
65}
66EXPORT_SYMBOL(touch_buffer);
67
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080068void __lock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070069{
NeilBrown74316202014-07-07 15:16:04 +100070 wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -070071}
72EXPORT_SYMBOL(__lock_buffer);
73
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -080074void unlock_buffer(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -070075{
Nick Piggin51b07fc2008-10-18 20:27:00 -070076 clear_bit_unlock(BH_Lock, &bh->b_state);
Peter Zijlstra4e857c52014-03-17 18:06:10 +010077 smp_mb__after_atomic();
Linus Torvalds1da177e2005-04-16 15:20:36 -070078 wake_up_bit(&bh->b_state, BH_Lock);
79}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -070080EXPORT_SYMBOL(unlock_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -070081
82/*
Mel Gormanb4597222013-07-03 15:02:05 -070083 * Returns if the page has dirty or writeback buffers. If all the buffers
84 * are unlocked and clean then the PageDirty information is stale. If
85 * any of the pages are locked, it is assumed they are locked for IO.
86 */
87void buffer_check_dirty_writeback(struct page *page,
88 bool *dirty, bool *writeback)
89{
90 struct buffer_head *head, *bh;
91 *dirty = false;
92 *writeback = false;
93
94 BUG_ON(!PageLocked(page));
95
96 if (!page_has_buffers(page))
97 return;
98
99 if (PageWriteback(page))
100 *writeback = true;
101
102 head = page_buffers(page);
103 bh = head;
104 do {
105 if (buffer_locked(bh))
106 *writeback = true;
107
108 if (buffer_dirty(bh))
109 *dirty = true;
110
111 bh = bh->b_this_page;
112 } while (bh != head);
113}
114EXPORT_SYMBOL(buffer_check_dirty_writeback);
115
116/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700117 * Block until a buffer comes unlocked. This doesn't stop it
118 * from becoming locked again - you have to lock it yourself
119 * if you want to preserve its state.
120 */
121void __wait_on_buffer(struct buffer_head * bh)
122{
NeilBrown74316202014-07-07 15:16:04 +1000123 wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700124}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700125EXPORT_SYMBOL(__wait_on_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700126
Robert Elliottb744c2a2014-10-21 13:55:09 -0600127static void buffer_io_error(struct buffer_head *bh, char *msg)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128{
Robert Elliott432f16e2014-10-21 13:55:11 -0600129 if (!test_bit(BH_Quiet, &bh->b_state))
130 printk_ratelimited(KERN_ERR
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400131 "Buffer I/O error on dev %pg, logical block %llu%s\n",
132 bh->b_bdev, (unsigned long long)bh->b_blocknr, msg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133}
134
135/*
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700136 * End-of-IO handler helper function which does not touch the bh after
137 * unlocking it.
138 * Note: unlock_buffer() sort-of does touch the bh after unlocking it, but
139 * a race there is benign: unlock_buffer() only use the bh's address for
140 * hashing after unlocking the buffer, so it doesn't actually touch the bh
141 * itself.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142 */
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700143static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700144{
145 if (uptodate) {
146 set_buffer_uptodate(bh);
147 } else {
Christoph Hellwig70246282016-07-19 11:28:41 +0200148 /* This happens, due to failed read-ahead attempts. */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149 clear_buffer_uptodate(bh);
150 }
151 unlock_buffer(bh);
Dmitry Monakhov68671f32007-10-16 01:24:47 -0700152}
153
154/*
155 * Default synchronous end-of-IO handler.. Just mark it up-to-date and
156 * unlock the buffer. This is what ll_rw_block uses too.
157 */
158void end_buffer_read_sync(struct buffer_head *bh, int uptodate)
159{
160 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700161 put_bh(bh);
162}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700163EXPORT_SYMBOL(end_buffer_read_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164
165void end_buffer_write_sync(struct buffer_head *bh, int uptodate)
166{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700167 if (uptodate) {
168 set_buffer_uptodate(bh);
169 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600170 buffer_io_error(bh, ", lost sync page write");
Jeff Layton87354e52017-07-06 07:02:21 -0400171 mark_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700172 clear_buffer_uptodate(bh);
173 }
174 unlock_buffer(bh);
175 put_bh(bh);
176}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700177EXPORT_SYMBOL(end_buffer_write_sync);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700178
179/*
Linus Torvalds1da177e2005-04-16 15:20:36 -0700180 * Various filesystems appear to want __find_get_block to be non-blocking.
181 * But it's the page lock which protects the buffers. To get around this,
182 * we get exclusion from try_to_free_buffers with the blockdev mapping's
183 * private_lock.
184 *
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700185 * Hack idea: for the blockdev mapping, private_lock contention
Linus Torvalds1da177e2005-04-16 15:20:36 -0700186 * may be quite high. This code could TryLock the page, and if that
Matthew Wilcoxb93b0162018-04-10 16:36:56 -0700187 * succeeds, there is no need to take private_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700188 */
189static struct buffer_head *
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -0800190__find_get_block_slow(struct block_device *bdev, sector_t block)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700191{
192 struct inode *bd_inode = bdev->bd_inode;
193 struct address_space *bd_mapping = bd_inode->i_mapping;
194 struct buffer_head *ret = NULL;
195 pgoff_t index;
196 struct buffer_head *bh;
197 struct buffer_head *head;
198 struct page *page;
199 int all_mapped = 1;
Tetsuo Handa43636c82019-01-21 22:49:37 +0900200 static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700201
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300202 index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman2457aec2014-06-04 16:10:31 -0700203 page = find_get_page_flags(bd_mapping, index, FGP_ACCESSED);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700204 if (!page)
205 goto out;
206
207 spin_lock(&bd_mapping->private_lock);
208 if (!page_has_buffers(page))
209 goto out_unlock;
210 head = page_buffers(page);
211 bh = head;
212 do {
Nikanth Karthikesan97f76d32009-04-02 16:56:46 -0700213 if (!buffer_mapped(bh))
214 all_mapped = 0;
215 else if (bh->b_blocknr == block) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700216 ret = bh;
217 get_bh(bh);
218 goto out_unlock;
219 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220 bh = bh->b_this_page;
221 } while (bh != head);
222
223 /* we might be here because some of the buffers on this page are
224 * not mapped. This is due to various races between
225 * file io on the block device and getblk. It gets dealt with
226 * elsewhere, don't buffer_error if we had some unmapped buffers
227 */
Tetsuo Handa43636c82019-01-21 22:49:37 +0900228 ratelimit_set_flags(&last_warned, RATELIMIT_MSG_ON_RELEASE);
229 if (all_mapped && __ratelimit(&last_warned)) {
230 printk("__find_get_block_slow() failed. block=%llu, "
231 "b_blocknr=%llu, b_state=0x%08lx, b_size=%zu, "
232 "device %pg blocksize: %d\n",
233 (unsigned long long)block,
234 (unsigned long long)bh->b_blocknr,
235 bh->b_state, bh->b_size, bdev,
236 1 << bd_inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237 }
238out_unlock:
239 spin_unlock(&bd_mapping->private_lock);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +0300240 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700241out:
242 return ret;
243}
244
Linus Torvalds1da177e2005-04-16 15:20:36 -0700245static void end_buffer_async_read(struct buffer_head *bh, int uptodate)
246{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700248 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249 struct buffer_head *tmp;
250 struct page *page;
251 int page_uptodate = 1;
252
253 BUG_ON(!buffer_async_read(bh));
254
255 page = bh->b_page;
256 if (uptodate) {
257 set_buffer_uptodate(bh);
258 } else {
259 clear_buffer_uptodate(bh);
Robert Elliott432f16e2014-10-21 13:55:11 -0600260 buffer_io_error(bh, ", async page read");
Linus Torvalds1da177e2005-04-16 15:20:36 -0700261 SetPageError(page);
262 }
263
264 /*
265 * Be _very_ careful from here on. Bad things can happen if
266 * two buffer heads end IO at almost the same time and both
267 * decide that the page is now completely done.
268 */
Nick Piggina3972202005-07-07 17:56:56 -0700269 first = page_buffers(page);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100270 spin_lock_irqsave(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700271 clear_buffer_async_read(bh);
272 unlock_buffer(bh);
273 tmp = bh;
274 do {
275 if (!buffer_uptodate(tmp))
276 page_uptodate = 0;
277 if (buffer_async_read(tmp)) {
278 BUG_ON(!buffer_locked(tmp));
279 goto still_busy;
280 }
281 tmp = tmp->b_this_page;
282 } while (tmp != bh);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100283 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284
285 /*
286 * If none of the buffers had errors and they are all
287 * uptodate then we can set the page uptodate.
288 */
289 if (page_uptodate && !PageError(page))
290 SetPageUptodate(page);
291 unlock_page(page);
292 return;
293
294still_busy:
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100295 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700296 return;
297}
298
Eric Biggers0957cd62022-12-23 12:36:37 -0800299struct postprocess_bh_ctx {
Eric Biggers31fb9922019-10-22 20:33:11 -0700300 struct work_struct work;
301 struct buffer_head *bh;
302};
303
Eric Biggers0957cd62022-12-23 12:36:37 -0800304static void verify_bh(struct work_struct *work)
305{
306 struct postprocess_bh_ctx *ctx =
307 container_of(work, struct postprocess_bh_ctx, work);
308 struct buffer_head *bh = ctx->bh;
309 bool valid;
310
311 valid = fsverity_verify_blocks(bh->b_page, bh->b_size, bh_offset(bh));
312 end_buffer_async_read(bh, valid);
313 kfree(ctx);
314}
315
316static bool need_fsverity(struct buffer_head *bh)
317{
318 struct page *page = bh->b_page;
319 struct inode *inode = page->mapping->host;
320
321 return fsverity_active(inode) &&
322 /* needed by ext4 */
323 page->index < DIV_ROUND_UP(inode->i_size, PAGE_SIZE);
324}
325
Eric Biggers31fb9922019-10-22 20:33:11 -0700326static void decrypt_bh(struct work_struct *work)
327{
Eric Biggers0957cd62022-12-23 12:36:37 -0800328 struct postprocess_bh_ctx *ctx =
329 container_of(work, struct postprocess_bh_ctx, work);
Eric Biggers31fb9922019-10-22 20:33:11 -0700330 struct buffer_head *bh = ctx->bh;
331 int err;
332
333 err = fscrypt_decrypt_pagecache_blocks(bh->b_page, bh->b_size,
334 bh_offset(bh));
Eric Biggers0957cd62022-12-23 12:36:37 -0800335 if (err == 0 && need_fsverity(bh)) {
336 /*
337 * We use different work queues for decryption and for verity
338 * because verity may require reading metadata pages that need
339 * decryption, and we shouldn't recurse to the same workqueue.
340 */
341 INIT_WORK(&ctx->work, verify_bh);
342 fsverity_enqueue_verify_work(&ctx->work);
343 return;
344 }
Eric Biggers31fb9922019-10-22 20:33:11 -0700345 end_buffer_async_read(bh, err == 0);
346 kfree(ctx);
347}
348
349/*
350 * I/O completion handler for block_read_full_page() - pages
351 * which come unlocked at the end of I/O.
352 */
353static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate)
354{
Eric Biggers0957cd62022-12-23 12:36:37 -0800355 struct inode *inode = bh->b_page->mapping->host;
356 bool decrypt = fscrypt_inode_uses_fs_layer_crypto(inode);
357 bool verify = need_fsverity(bh);
358
359 /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */
360 if (uptodate && (decrypt || verify)) {
361 struct postprocess_bh_ctx *ctx =
362 kmalloc(sizeof(*ctx), GFP_ATOMIC);
Eric Biggers31fb9922019-10-22 20:33:11 -0700363
364 if (ctx) {
Eric Biggers31fb9922019-10-22 20:33:11 -0700365 ctx->bh = bh;
Eric Biggers0957cd62022-12-23 12:36:37 -0800366 if (decrypt) {
367 INIT_WORK(&ctx->work, decrypt_bh);
368 fscrypt_enqueue_decrypt_work(&ctx->work);
369 } else {
370 INIT_WORK(&ctx->work, verify_bh);
371 fsverity_enqueue_verify_work(&ctx->work);
372 }
Eric Biggers31fb9922019-10-22 20:33:11 -0700373 return;
374 }
375 uptodate = 0;
376 }
377 end_buffer_async_read(bh, uptodate);
378}
379
Linus Torvalds1da177e2005-04-16 15:20:36 -0700380/*
381 * Completion handler for block_write_full_page() - pages which are unlocked
382 * during I/O, and which have PageWriteback cleared upon I/O completion.
383 */
Chris Mason35c80d52009-04-15 13:22:38 -0400384void end_buffer_async_write(struct buffer_head *bh, int uptodate)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700385{
Linus Torvalds1da177e2005-04-16 15:20:36 -0700386 unsigned long flags;
Nick Piggina3972202005-07-07 17:56:56 -0700387 struct buffer_head *first;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700388 struct buffer_head *tmp;
389 struct page *page;
390
391 BUG_ON(!buffer_async_write(bh));
392
393 page = bh->b_page;
394 if (uptodate) {
395 set_buffer_uptodate(bh);
396 } else {
Robert Elliott432f16e2014-10-21 13:55:11 -0600397 buffer_io_error(bh, ", lost async page write");
Jeff Layton87354e52017-07-06 07:02:21 -0400398 mark_buffer_write_io_error(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700399 clear_buffer_uptodate(bh);
400 SetPageError(page);
401 }
402
Nick Piggina3972202005-07-07 17:56:56 -0700403 first = page_buffers(page);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100404 spin_lock_irqsave(&first->b_uptodate_lock, flags);
Nick Piggina3972202005-07-07 17:56:56 -0700405
Linus Torvalds1da177e2005-04-16 15:20:36 -0700406 clear_buffer_async_write(bh);
407 unlock_buffer(bh);
408 tmp = bh->b_this_page;
409 while (tmp != bh) {
410 if (buffer_async_write(tmp)) {
411 BUG_ON(!buffer_locked(tmp));
412 goto still_busy;
413 }
414 tmp = tmp->b_this_page;
415 }
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100416 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700417 end_page_writeback(page);
418 return;
419
420still_busy:
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +0100421 spin_unlock_irqrestore(&first->b_uptodate_lock, flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700422 return;
423}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700424EXPORT_SYMBOL(end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700425
426/*
427 * If a page's buffers are under async readin (end_buffer_async_read
428 * completion) then there is a possibility that another thread of
429 * control could lock one of the buffers after it has completed
430 * but while some of the other buffers have not completed. This
431 * locked buffer would confuse end_buffer_async_read() into not unlocking
432 * the page. So the absence of BH_Async_Read tells end_buffer_async_read()
433 * that this buffer is not under async I/O.
434 *
435 * The page comes unlocked when it has no locked buffer_async buffers
436 * left.
437 *
438 * PageLocked prevents anyone starting new async I/O reads any of
439 * the buffers.
440 *
441 * PageWriteback is used to prevent simultaneous writeout of the same
442 * page.
443 *
444 * PageLocked prevents anyone from starting writeback of a page which is
445 * under read I/O (PageWriteback is only ever set against a locked page).
446 */
447static void mark_buffer_async_read(struct buffer_head *bh)
448{
Eric Biggers31fb9922019-10-22 20:33:11 -0700449 bh->b_end_io = end_buffer_async_read_io;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700450 set_buffer_async_read(bh);
451}
452
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -0700453static void mark_buffer_async_write_endio(struct buffer_head *bh,
454 bh_end_io_t *handler)
Chris Mason35c80d52009-04-15 13:22:38 -0400455{
456 bh->b_end_io = handler;
457 set_buffer_async_write(bh);
458}
459
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460void mark_buffer_async_write(struct buffer_head *bh)
461{
Chris Mason35c80d52009-04-15 13:22:38 -0400462 mark_buffer_async_write_endio(bh, end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700463}
464EXPORT_SYMBOL(mark_buffer_async_write);
465
466
467/*
468 * fs/buffer.c contains helper functions for buffer-backed address space's
469 * fsync functions. A common requirement for buffer-based filesystems is
470 * that certain data from the backing blockdev needs to be written out for
471 * a successful fsync(). For example, ext2 indirect blocks need to be
472 * written back and waited upon before fsync() returns.
473 *
474 * The functions mark_buffer_inode_dirty(), fsync_inode_buffers(),
475 * inode_has_buffers() and invalidate_inode_buffers() are provided for the
476 * management of a list of dependent buffers at ->i_mapping->private_list.
477 *
478 * Locking is a little subtle: try_to_free_buffers() will remove buffers
479 * from their controlling inode's queue when they are being freed. But
480 * try_to_free_buffers() will be operating against the *blockdev* mapping
481 * at the time, not against the S_ISREG file which depends on those buffers.
482 * So the locking for private_list is via the private_lock in the address_space
483 * which backs the buffers. Which is different from the address_space
484 * against which the buffers are listed. So for a particular address_space,
485 * mapping->private_lock does *not* protect mapping->private_list! In fact,
486 * mapping->private_list will always be protected by the backing blockdev's
487 * ->private_lock.
488 *
489 * Which introduces a requirement: all buffers on an address_space's
490 * ->private_list must be from the same address_space: the blockdev's.
491 *
492 * address_spaces which do not place buffers at ->private_list via these
493 * utility functions are free to use private_lock and private_list for
494 * whatever they want. The only requirement is that list_empty(private_list)
495 * be true at clear_inode() time.
496 *
497 * FIXME: clear_inode should not call invalidate_inode_buffers(). The
498 * filesystems should do that. invalidate_inode_buffers() should just go
499 * BUG_ON(!list_empty).
500 *
501 * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should
502 * take an address_space, not an inode. And it should be called
503 * mark_buffer_dirty_fsync() to clearly define why those buffers are being
504 * queued up.
505 *
506 * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the
507 * list if it is already on a list. Because if the buffer is on a list,
508 * it *must* already be on the right one. If not, the filesystem is being
509 * silly. This will save a ton of locking. But first we have to ensure
510 * that buffers are taken *off* the old inode's list when they are freed
511 * (presumably in truncate). That requires careful auditing of all
512 * filesystems (do it inside bforget()). It could also be done by bringing
513 * b_inode back.
514 */
515
516/*
517 * The buffer's backing address_space's private_lock must be held
518 */
Thomas Petazzonidbacefc2008-07-29 22:33:47 -0700519static void __remove_assoc_queue(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700520{
521 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -0700522 WARN_ON(!bh->b_assoc_map);
Jan Kara58ff4072006-10-17 00:10:19 -0700523 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524}
525
526int inode_has_buffers(struct inode *inode)
527{
528 return !list_empty(&inode->i_data.private_list);
529}
530
531/*
532 * osync is designed to support O_SYNC io. It waits synchronously for
533 * all already-submitted IO to complete, but does not queue any new
534 * writes to the disk.
535 *
536 * To do O_SYNC writes, just queue the buffer writes with ll_rw_block as
537 * you dirty the buffers, and then use osync_inode_buffers to wait for
538 * completion. Any other dirty buffers which are not yet queued for
539 * write will not be flushed to disk by the osync.
540 */
541static int osync_buffers_list(spinlock_t *lock, struct list_head *list)
542{
543 struct buffer_head *bh;
544 struct list_head *p;
545 int err = 0;
546
547 spin_lock(lock);
548repeat:
549 list_for_each_prev(p, list) {
550 bh = BH_ENTRY(p);
551 if (buffer_locked(bh)) {
552 get_bh(bh);
553 spin_unlock(lock);
554 wait_on_buffer(bh);
555 if (!buffer_uptodate(bh))
556 err = -EIO;
557 brelse(bh);
558 spin_lock(lock);
559 goto repeat;
560 }
561 }
562 spin_unlock(lock);
563 return err;
564}
565
Mateusz Guzik08fdc8a2017-10-03 18:17:41 +0200566void emergency_thaw_bdev(struct super_block *sb)
Al Viro01a05b32010-03-23 06:06:58 -0400567{
Christoph Hellwig040f04b2020-11-24 11:54:06 +0100568 while (sb->s_bdev && !thaw_bdev(sb->s_bdev))
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +0400569 printk(KERN_WARNING "Emergency Thaw on %pg\n", sb->s_bdev);
Al Viro01a05b32010-03-23 06:06:58 -0400570}
571
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572/**
Randy Dunlap78a4a502008-02-29 22:02:31 -0800573 * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers
Martin Waitz67be2dd2005-05-01 08:59:26 -0700574 * @mapping: the mapping which wants those buffers written
Linus Torvalds1da177e2005-04-16 15:20:36 -0700575 *
576 * Starts I/O against the buffers at mapping->private_list, and waits upon
577 * that I/O.
578 *
Martin Waitz67be2dd2005-05-01 08:59:26 -0700579 * Basically, this is a convenience function for fsync().
580 * @mapping is a file or directory which needs those buffers to be written for
581 * a successful fsync().
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 */
583int sync_mapping_buffers(struct address_space *mapping)
584{
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800585 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586
587 if (buffer_mapping == NULL || list_empty(&mapping->private_list))
588 return 0;
589
590 return fsync_buffers_list(&buffer_mapping->private_lock,
591 &mapping->private_list);
592}
593EXPORT_SYMBOL(sync_mapping_buffers);
594
595/*
596 * Called when we've recently written block `bblock', and it is known that
597 * `bblock' was for a buffer_boundary() buffer. This means that the block at
598 * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's
599 * dirty, schedule it for IO. So that indirects merge nicely with their data.
600 */
601void write_boundary_block(struct block_device *bdev,
602 sector_t bblock, unsigned blocksize)
603{
604 struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize);
605 if (bh) {
606 if (buffer_dirty(bh))
Mike Christiedfec8a12016-06-05 14:31:44 -0500607 ll_rw_block(REQ_OP_WRITE, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700608 put_bh(bh);
609 }
610}
611
612void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode)
613{
614 struct address_space *mapping = inode->i_mapping;
615 struct address_space *buffer_mapping = bh->b_page->mapping;
616
617 mark_buffer_dirty(bh);
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800618 if (!mapping->private_data) {
619 mapping->private_data = buffer_mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620 } else {
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800621 BUG_ON(mapping->private_data != buffer_mapping);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 }
Jan Kara535ee2f2008-02-08 04:21:59 -0800623 if (!bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 spin_lock(&buffer_mapping->private_lock);
625 list_move_tail(&bh->b_assoc_buffers,
626 &mapping->private_list);
Jan Kara58ff4072006-10-17 00:10:19 -0700627 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 spin_unlock(&buffer_mapping->private_lock);
629 }
630}
631EXPORT_SYMBOL(mark_buffer_dirty_inode);
632
633/*
634 * Add a page to the dirty page list.
635 *
636 * It is a sad fact of life that this function is called from several places
637 * deeply under spinlocking. It may not sleep.
638 *
639 * If the page has buffers, the uptodate buffers are set dirty, to preserve
640 * dirty-state coherency between the page and the buffers. It the page does
641 * not have buffers then when they are later attached they will all be set
642 * dirty.
643 *
644 * The buffers are dirtied before the page is dirtied. There's a small race
645 * window in which a writepage caller may see the page cleanness but not the
646 * buffer dirtiness. That's fine. If this code were to set the page dirty
647 * before the buffers, a concurrent writepage caller could clear the page dirty
648 * bit, see a bunch of clean buffers and we'd end up with dirty buffers/clean
649 * page on the dirty page list.
650 *
651 * We use private_lock to lock against try_to_free_buffers while using the
652 * page's buffer list. Also use this to protect against clean buffers being
653 * added to the page after it was set dirty.
654 *
655 * FIXME: may need to call ->reservepage here as well. That's rather up to the
656 * address_space though.
657 */
658int __set_page_dirty_buffers(struct page *page)
659{
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700660 int newly_dirty;
Nick Piggin787d2212007-07-17 04:03:34 -0700661 struct address_space *mapping = page_mapping(page);
Nick Pigginebf7a222006-10-10 04:36:54 +0200662
663 if (unlikely(!mapping))
664 return !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700665
666 spin_lock(&mapping->private_lock);
667 if (page_has_buffers(page)) {
668 struct buffer_head *head = page_buffers(page);
669 struct buffer_head *bh = head;
670
671 do {
672 set_buffer_dirty(bh);
673 bh = bh->b_this_page;
674 } while (bh != head);
675 }
Greg Thelenc4843a72015-05-22 17:13:16 -0400676 /*
Roman Gushchinbcfe06b2020-12-01 13:58:27 -0800677 * Lock out page's memcg migration to keep PageDirty
Johannes Weiner81f8c3a2016-03-15 14:57:04 -0700678 * synchronized with per-memcg dirty page counters.
Greg Thelenc4843a72015-05-22 17:13:16 -0400679 */
Johannes Weiner62cccb82016-03-15 14:57:22 -0700680 lock_page_memcg(page);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700681 newly_dirty = !TestSetPageDirty(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 spin_unlock(&mapping->private_lock);
683
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700684 if (newly_dirty)
Johannes Weiner62cccb82016-03-15 14:57:22 -0700685 __set_page_dirty(page, mapping, 1);
Greg Thelenc4843a72015-05-22 17:13:16 -0400686
Johannes Weiner62cccb82016-03-15 14:57:22 -0700687 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -0400688
689 if (newly_dirty)
690 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
691
Linus Torvaldsa8e7d492009-03-19 11:32:05 -0700692 return newly_dirty;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700693}
694EXPORT_SYMBOL(__set_page_dirty_buffers);
695
696/*
697 * Write out and wait upon a list of buffers.
698 *
699 * We have conflicting pressures: we want to make sure that all
700 * initially dirty buffers get waited on, but that any subsequently
701 * dirtied buffers don't. After all, we don't want fsync to last
702 * forever if somebody is actively writing to the file.
703 *
704 * Do this in two main stages: first we copy dirty buffers to a
705 * temporary inode list, queueing the writes as we go. Then we clean
706 * up, waiting for those writes to complete.
707 *
708 * During this second stage, any subsequent updates to the file may end
709 * up refiling the buffer on the original inode's dirty list again, so
710 * there is a chance we will end up with a buffer queued for write but
711 * not yet completed on that list. So, as a final cleanup we go through
712 * the osync code to catch these locked, dirty buffers without requeuing
713 * any newly dirty buffers for write.
714 */
715static int fsync_buffers_list(spinlock_t *lock, struct list_head *list)
716{
717 struct buffer_head *bh;
718 struct list_head tmp;
Jens Axboe7eaceac2011-03-10 08:52:07 +0100719 struct address_space *mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720 int err = 0, err2;
Jens Axboe4ee24912011-03-17 10:51:40 +0100721 struct blk_plug plug;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700722
723 INIT_LIST_HEAD(&tmp);
Jens Axboe4ee24912011-03-17 10:51:40 +0100724 blk_start_plug(&plug);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725
726 spin_lock(lock);
727 while (!list_empty(list)) {
728 bh = BH_ENTRY(list->next);
Jan Kara535ee2f2008-02-08 04:21:59 -0800729 mapping = bh->b_assoc_map;
Jan Kara58ff4072006-10-17 00:10:19 -0700730 __remove_assoc_queue(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800731 /* Avoid race with mark_buffer_dirty_inode() which does
732 * a lockless check and we rely on seeing the dirty bit */
733 smp_mb();
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734 if (buffer_dirty(bh) || buffer_locked(bh)) {
735 list_add(&bh->b_assoc_buffers, &tmp);
Jan Kara535ee2f2008-02-08 04:21:59 -0800736 bh->b_assoc_map = mapping;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737 if (buffer_dirty(bh)) {
738 get_bh(bh);
739 spin_unlock(lock);
740 /*
741 * Ensure any pending I/O completes so that
Christoph Hellwig9cb569d2010-08-11 17:06:24 +0200742 * write_dirty_buffer() actually writes the
743 * current contents - it is a noop if I/O is
744 * still in flight on potentially older
745 * contents.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700746 */
Christoph Hellwig70fd7612016-11-01 07:40:10 -0600747 write_dirty_buffer(bh, REQ_SYNC);
Jens Axboe9cf6b722009-04-06 14:48:03 +0200748
749 /*
750 * Kick off IO for the previous mapping. Note
751 * that we will not run the very last mapping,
752 * wait_on_buffer() will do that for us
753 * through sync_buffer().
754 */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755 brelse(bh);
756 spin_lock(lock);
757 }
758 }
759 }
760
Jens Axboe4ee24912011-03-17 10:51:40 +0100761 spin_unlock(lock);
762 blk_finish_plug(&plug);
763 spin_lock(lock);
764
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765 while (!list_empty(&tmp)) {
766 bh = BH_ENTRY(tmp.prev);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767 get_bh(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -0800768 mapping = bh->b_assoc_map;
769 __remove_assoc_queue(bh);
770 /* Avoid race with mark_buffer_dirty_inode() which does
771 * a lockless check and we rely on seeing the dirty bit */
772 smp_mb();
773 if (buffer_dirty(bh)) {
774 list_add(&bh->b_assoc_buffers,
Jan Karae3892292008-03-04 14:28:33 -0800775 &mapping->private_list);
Jan Kara535ee2f2008-02-08 04:21:59 -0800776 bh->b_assoc_map = mapping;
777 }
Linus Torvalds1da177e2005-04-16 15:20:36 -0700778 spin_unlock(lock);
779 wait_on_buffer(bh);
780 if (!buffer_uptodate(bh))
781 err = -EIO;
782 brelse(bh);
783 spin_lock(lock);
784 }
785
786 spin_unlock(lock);
787 err2 = osync_buffers_list(lock, list);
788 if (err)
789 return err;
790 else
791 return err2;
792}
793
794/*
795 * Invalidate any and all dirty buffers on a given inode. We are
796 * probably unmounting the fs, but that doesn't mean we have already
797 * done a sync(). Just drop the buffers from the inode list.
798 *
799 * NOTE: we take the inode's blockdev's mapping's private_lock. Which
800 * assumes that all the buffers are against the blockdev. Not true
801 * for reiserfs.
802 */
803void invalidate_inode_buffers(struct inode *inode)
804{
805 if (inode_has_buffers(inode)) {
806 struct address_space *mapping = &inode->i_data;
807 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800808 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700809
810 spin_lock(&buffer_mapping->private_lock);
811 while (!list_empty(list))
812 __remove_assoc_queue(BH_ENTRY(list->next));
813 spin_unlock(&buffer_mapping->private_lock);
814 }
815}
Jan Kara52b19ac2008-09-23 18:24:08 +0200816EXPORT_SYMBOL(invalidate_inode_buffers);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700817
818/*
819 * Remove any clean buffers from the inode's buffer list. This is called
820 * when we're trying to free the inode itself. Those buffers can pin it.
821 *
822 * Returns true if all buffers were removed.
823 */
824int remove_inode_buffers(struct inode *inode)
825{
826 int ret = 1;
827
828 if (inode_has_buffers(inode)) {
829 struct address_space *mapping = &inode->i_data;
830 struct list_head *list = &mapping->private_list;
Rafael Aquini252aa6f2012-12-11 16:02:35 -0800831 struct address_space *buffer_mapping = mapping->private_data;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700832
833 spin_lock(&buffer_mapping->private_lock);
834 while (!list_empty(list)) {
835 struct buffer_head *bh = BH_ENTRY(list->next);
836 if (buffer_dirty(bh)) {
837 ret = 0;
838 break;
839 }
840 __remove_assoc_queue(bh);
841 }
842 spin_unlock(&buffer_mapping->private_lock);
843 }
844 return ret;
845}
846
847/*
848 * Create the appropriate buffers when given a page for data area and
849 * the size of each buffer.. Use the bh->b_this_page linked list to
850 * follow the buffers created. Return NULL if unable to create more
851 * buffers.
852 *
853 * The retry flag is used to differentiate async IO (paging, swapping)
854 * which may not fail from ordinary buffer allocations.
855 */
856struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size,
Jens Axboe640ab982017-09-27 05:40:16 -0600857 bool retry)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700858{
859 struct buffer_head *bh, *head;
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700860 gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861 long offset;
Roman Gushchinb87d8ce2020-10-17 16:13:40 -0700862 struct mem_cgroup *memcg, *old_memcg;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700863
Jens Axboe640ab982017-09-27 05:40:16 -0600864 if (retry)
865 gfp |= __GFP_NOFAIL;
866
Johannes Weiner6eeb1042021-02-24 12:04:15 -0800867 /* The page lock pins the memcg */
868 memcg = page_memcg(page);
Roman Gushchinb87d8ce2020-10-17 16:13:40 -0700869 old_memcg = set_active_memcg(memcg);
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700870
Linus Torvalds1da177e2005-04-16 15:20:36 -0700871 head = NULL;
872 offset = PAGE_SIZE;
873 while ((offset -= size) >= 0) {
Jens Axboe640ab982017-09-27 05:40:16 -0600874 bh = alloc_buffer_head(gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700875 if (!bh)
876 goto no_grow;
877
Linus Torvalds1da177e2005-04-16 15:20:36 -0700878 bh->b_this_page = head;
879 bh->b_blocknr = -1;
880 head = bh;
881
Linus Torvalds1da177e2005-04-16 15:20:36 -0700882 bh->b_size = size;
883
884 /* Link the buffer to its page */
885 set_bh_page(bh, page, offset);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700886 }
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700887out:
Roman Gushchinb87d8ce2020-10-17 16:13:40 -0700888 set_active_memcg(old_memcg);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889 return head;
890/*
891 * In case anything failed, we just free everything we got.
892 */
893no_grow:
894 if (head) {
895 do {
896 bh = head;
897 head = head->b_this_page;
898 free_buffer_head(bh);
899 } while (head);
900 }
901
Shakeel Buttf745c6f2018-08-17 15:46:44 -0700902 goto out;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700903}
904EXPORT_SYMBOL_GPL(alloc_page_buffers);
905
906static inline void
907link_dev_buffers(struct page *page, struct buffer_head *head)
908{
909 struct buffer_head *bh, *tail;
910
911 bh = head;
912 do {
913 tail = bh;
914 bh = bh->b_this_page;
915 } while (bh);
916 tail->b_this_page = head;
Guoqing Jiang45dcfc22020-06-01 21:47:48 -0700917 attach_page_private(page, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700918}
919
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800920static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size)
921{
922 sector_t retval = ~((sector_t)0);
923 loff_t sz = i_size_read(bdev->bd_inode);
924
925 if (sz) {
926 unsigned int sizebits = blksize_bits(size);
927 retval = (sz >> sizebits);
928 }
929 return retval;
930}
931
Linus Torvalds1da177e2005-04-16 15:20:36 -0700932/*
933 * Initialise the state of a blockdev page's buffers.
934 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200935static sector_t
Linus Torvalds1da177e2005-04-16 15:20:36 -0700936init_page_buffers(struct page *page, struct block_device *bdev,
937 sector_t block, int size)
938{
939 struct buffer_head *head = page_buffers(page);
940 struct buffer_head *bh = head;
941 int uptodate = PageUptodate(page);
Linus Torvaldsbbec02702012-11-29 12:31:52 -0800942 sector_t end_block = blkdev_max_block(I_BDEV(bdev->bd_inode), size);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700943
944 do {
945 if (!buffer_mapped(bh)) {
Eric Biggers01950a32018-01-16 22:25:12 -0800946 bh->b_end_io = NULL;
947 bh->b_private = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700948 bh->b_bdev = bdev;
949 bh->b_blocknr = block;
950 if (uptodate)
951 set_buffer_uptodate(bh);
Jeff Moyer080399a2012-05-11 16:34:10 +0200952 if (block < end_block)
953 set_buffer_mapped(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700954 }
955 block++;
956 bh = bh->b_this_page;
957 } while (bh != head);
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200958
959 /*
960 * Caller needs to validate requested block against end of device.
961 */
962 return end_block;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700963}
964
965/*
966 * Create the page-cache page that contains the requested block.
967 *
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200968 * This is used purely for blockdev mappings.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 */
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200970static int
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971grow_dev_page(struct block_device *bdev, sector_t block,
Gioh Kim3b5e6452014-09-04 22:04:42 -0400972 pgoff_t index, int size, int sizebits, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973{
974 struct inode *inode = bdev->bd_inode;
975 struct page *page;
976 struct buffer_head *bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200977 sector_t end_block;
Zhiqiang Liuc4b4c2a2020-04-13 13:12:10 +0800978 int ret = 0;
Johannes Weiner84235de2013-10-16 13:47:00 -0700979 gfp_t gfp_mask;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700980
Michal Hockoc62d2552015-11-06 16:28:49 -0800981 gfp_mask = mapping_gfp_constraint(inode->i_mapping, ~__GFP_FS) | gfp;
Gioh Kim3b5e6452014-09-04 22:04:42 -0400982
Johannes Weiner84235de2013-10-16 13:47:00 -0700983 /*
984 * XXX: __getblk_slow() can not really deal with failure and
985 * will endlessly loop on improvised global reclaim. Prefer
986 * looping in the allocator rather than here, at least that
987 * code knows what it's doing.
988 */
989 gfp_mask |= __GFP_NOFAIL;
990
991 page = find_or_create_page(inode->i_mapping, index, gfp_mask);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992
Eric Sesterhenne827f922006-03-26 18:24:46 +0200993 BUG_ON(!PageLocked(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -0700994
995 if (page_has_buffers(page)) {
996 bh = page_buffers(page);
997 if (bh->b_size == size) {
Hugh Dickins676ce6d2012-08-23 12:17:36 +0200998 end_block = init_page_buffers(page, bdev,
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +0100999 (sector_t)index << sizebits,
1000 size);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001001 goto done;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001002 }
1003 if (!try_to_free_buffers(page))
1004 goto failed;
1005 }
1006
1007 /*
1008 * Allocate some buffers for this page
1009 */
Jens Axboe94dc24c2017-09-27 05:45:36 -06001010 bh = alloc_page_buffers(page, size, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001011
1012 /*
1013 * Link the page to the buffers and initialise them. Take the
1014 * lock to be atomic wrt __find_get_block(), which does not
1015 * run under the page lock.
1016 */
1017 spin_lock(&inode->i_mapping->private_lock);
1018 link_dev_buffers(page, bh);
Anton Altaparmakovf2d5a942014-09-22 01:53:03 +01001019 end_block = init_page_buffers(page, bdev, (sector_t)index << sizebits,
1020 size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001021 spin_unlock(&inode->i_mapping->private_lock);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001022done:
1023 ret = (block < end_block) ? 1 : -ENXIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001024failed:
Linus Torvalds1da177e2005-04-16 15:20:36 -07001025 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001026 put_page(page);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001027 return ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001028}
1029
1030/*
1031 * Create buffers for the specified block device block's page. If
1032 * that page was dirty, the buffers are set dirty also.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001033 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001034static int
Gioh Kim3b5e6452014-09-04 22:04:42 -04001035grow_buffers(struct block_device *bdev, sector_t block, int size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001036{
Linus Torvalds1da177e2005-04-16 15:20:36 -07001037 pgoff_t index;
1038 int sizebits;
1039
Mikulas Patocka90432e62021-03-22 10:05:05 -04001040 sizebits = PAGE_SHIFT - __ffs(size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001041 index = block >> sizebits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001042
Andrew Mortone5657932006-10-11 01:21:46 -07001043 /*
1044 * Check for a block which wants to lie outside our maximum possible
1045 * pagecache index. (this comparison is done using sector_t types).
1046 */
1047 if (unlikely(index != block >> sizebits)) {
Andrew Mortone5657932006-10-11 01:21:46 -07001048 printk(KERN_ERR "%s: requested out-of-range block %llu for "
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001049 "device %pg\n",
Harvey Harrison8e24eea2008-04-30 00:55:09 -07001050 __func__, (unsigned long long)block,
Dmitry Monakhova1c6f0572015-04-13 16:31:37 +04001051 bdev);
Andrew Mortone5657932006-10-11 01:21:46 -07001052 return -EIO;
1053 }
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001054
Linus Torvalds1da177e2005-04-16 15:20:36 -07001055 /* Create a page with the proper size buffers.. */
Gioh Kim3b5e6452014-09-04 22:04:42 -04001056 return grow_dev_page(bdev, block, index, size, sizebits, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001057}
1058
Eric Biggers0026ba42016-09-12 13:30:41 -07001059static struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001060__getblk_slow(struct block_device *bdev, sector_t block,
1061 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001062{
1063 /* Size must be multiple of hard sectorsize */
Martin K. Petersene1defc42009-05-22 17:17:49 -04001064 if (unlikely(size & (bdev_logical_block_size(bdev)-1) ||
Linus Torvalds1da177e2005-04-16 15:20:36 -07001065 (size < 512 || size > PAGE_SIZE))) {
1066 printk(KERN_ERR "getblk(): invalid block size %d requested\n",
1067 size);
Martin K. Petersene1defc42009-05-22 17:17:49 -04001068 printk(KERN_ERR "logical block size: %d\n",
1069 bdev_logical_block_size(bdev));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
1071 dump_stack();
1072 return NULL;
1073 }
1074
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001075 for (;;) {
1076 struct buffer_head *bh;
1077 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001078
1079 bh = __find_get_block(bdev, block, size);
1080 if (bh)
1081 return bh;
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001082
Gioh Kim3b5e6452014-09-04 22:04:42 -04001083 ret = grow_buffers(bdev, block, size, gfp);
Hugh Dickins676ce6d2012-08-23 12:17:36 +02001084 if (ret < 0)
1085 return NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001086 }
1087}
1088
1089/*
1090 * The relationship between dirty buffers and dirty pages:
1091 *
1092 * Whenever a page has any dirty buffers, the page's dirty bit is set, and
Matthew Wilcoxec82e1c2017-12-04 10:40:41 -05001093 * the page is tagged dirty in the page cache.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001094 *
1095 * At all times, the dirtiness of the buffers represents the dirtiness of
1096 * subsections of the page. If the page has buffers, the page dirty bit is
1097 * merely a hint about the true dirty state.
1098 *
1099 * When a page is set dirty in its entirety, all its buffers are marked dirty
1100 * (if the page has buffers).
1101 *
1102 * When a buffer is marked dirty, its page is dirtied, but the page's other
1103 * buffers are not.
1104 *
1105 * Also. When blockdev buffers are explicitly read with bread(), they
1106 * individually become uptodate. But their backing page remains not
1107 * uptodate - even if all of its buffers are uptodate. A subsequent
1108 * block_read_full_page() against that page will discover all the uptodate
1109 * buffers, will set the page uptodate and will perform no I/O.
1110 */
1111
1112/**
1113 * mark_buffer_dirty - mark a buffer_head as needing writeout
Martin Waitz67be2dd2005-05-01 08:59:26 -07001114 * @bh: the buffer_head to mark dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001115 *
Matthew Wilcoxec82e1c2017-12-04 10:40:41 -05001116 * mark_buffer_dirty() will set the dirty bit against the buffer, then set
1117 * its backing page dirty, then tag the page as dirty in the page cache
1118 * and then attach the address_space's inode to its superblock's dirty
Linus Torvalds1da177e2005-04-16 15:20:36 -07001119 * inode list.
1120 *
1121 * mark_buffer_dirty() is atomic. It takes bh->b_page->mapping->private_lock,
Matthew Wilcoxb93b0162018-04-10 16:36:56 -07001122 * i_pages lock and mapping->host->i_lock.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001123 */
Harvey Harrisonfc9b52c2008-02-08 04:19:52 -08001124void mark_buffer_dirty(struct buffer_head *bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001125{
Nick Piggin787d2212007-07-17 04:03:34 -07001126 WARN_ON_ONCE(!buffer_uptodate(bh));
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001127
Tejun Heo5305cb82013-01-11 13:06:36 -08001128 trace_block_dirty_buffer(bh);
1129
Linus Torvalds1be62dc2008-04-04 14:38:17 -07001130 /*
1131 * Very *carefully* optimize the it-is-already-dirty case.
1132 *
1133 * Don't let the final "is it dirty" escape to before we
1134 * perhaps modified the buffer.
1135 */
1136 if (buffer_dirty(bh)) {
1137 smp_mb();
1138 if (buffer_dirty(bh))
1139 return;
1140 }
1141
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001142 if (!test_set_buffer_dirty(bh)) {
1143 struct page *page = bh->b_page;
Greg Thelenc4843a72015-05-22 17:13:16 -04001144 struct address_space *mapping = NULL;
Greg Thelenc4843a72015-05-22 17:13:16 -04001145
Johannes Weiner62cccb82016-03-15 14:57:22 -07001146 lock_page_memcg(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001147 if (!TestSetPageDirty(page)) {
Greg Thelenc4843a72015-05-22 17:13:16 -04001148 mapping = page_mapping(page);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001149 if (mapping)
Johannes Weiner62cccb82016-03-15 14:57:22 -07001150 __set_page_dirty(page, mapping, 0);
Linus Torvalds8e9d78e2009-08-21 17:40:08 -07001151 }
Johannes Weiner62cccb82016-03-15 14:57:22 -07001152 unlock_page_memcg(page);
Greg Thelenc4843a72015-05-22 17:13:16 -04001153 if (mapping)
1154 __mark_inode_dirty(mapping->host, I_DIRTY_PAGES);
Linus Torvaldsa8e7d492009-03-19 11:32:05 -07001155 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001156}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001157EXPORT_SYMBOL(mark_buffer_dirty);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001158
Jeff Layton87354e52017-07-06 07:02:21 -04001159void mark_buffer_write_io_error(struct buffer_head *bh)
1160{
Jeff Layton485e9602020-06-01 21:45:40 -07001161 struct super_block *sb;
1162
Jeff Layton87354e52017-07-06 07:02:21 -04001163 set_buffer_write_io_error(bh);
1164 /* FIXME: do we need to set this in both places? */
1165 if (bh->b_page && bh->b_page->mapping)
1166 mapping_set_error(bh->b_page->mapping, -EIO);
1167 if (bh->b_assoc_map)
1168 mapping_set_error(bh->b_assoc_map, -EIO);
Jeff Layton485e9602020-06-01 21:45:40 -07001169 rcu_read_lock();
1170 sb = READ_ONCE(bh->b_bdev->bd_super);
1171 if (sb)
1172 errseq_set(&sb->s_wb_err, -EIO);
1173 rcu_read_unlock();
Jeff Layton87354e52017-07-06 07:02:21 -04001174}
1175EXPORT_SYMBOL(mark_buffer_write_io_error);
1176
Linus Torvalds1da177e2005-04-16 15:20:36 -07001177/*
1178 * Decrement a buffer_head's reference count. If all buffers against a page
1179 * have zero reference count, are clean and unlocked, and if the page is clean
1180 * and unlocked then try_to_free_buffers() may strip the buffers from the page
1181 * in preparation for freeing it (sometimes, rarely, buffers are removed from
1182 * a page but it ends up not being freed, and buffers may later be reattached).
1183 */
1184void __brelse(struct buffer_head * buf)
1185{
1186 if (atomic_read(&buf->b_count)) {
1187 put_bh(buf);
1188 return;
1189 }
Arjan van de Ven5c752ad2008-07-25 19:45:40 -07001190 WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n");
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001192EXPORT_SYMBOL(__brelse);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
1194/*
1195 * bforget() is like brelse(), except it discards any
1196 * potentially dirty data.
1197 */
1198void __bforget(struct buffer_head *bh)
1199{
1200 clear_buffer_dirty(bh);
Jan Kara535ee2f2008-02-08 04:21:59 -08001201 if (bh->b_assoc_map) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202 struct address_space *buffer_mapping = bh->b_page->mapping;
1203
1204 spin_lock(&buffer_mapping->private_lock);
1205 list_del_init(&bh->b_assoc_buffers);
Jan Kara58ff4072006-10-17 00:10:19 -07001206 bh->b_assoc_map = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207 spin_unlock(&buffer_mapping->private_lock);
1208 }
1209 __brelse(bh);
1210}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07001211EXPORT_SYMBOL(__bforget);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001212
1213static struct buffer_head *__bread_slow(struct buffer_head *bh)
1214{
1215 lock_buffer(bh);
1216 if (buffer_uptodate(bh)) {
1217 unlock_buffer(bh);
1218 return bh;
1219 } else {
1220 get_bh(bh);
1221 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05001222 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223 wait_on_buffer(bh);
1224 if (buffer_uptodate(bh))
1225 return bh;
1226 }
1227 brelse(bh);
1228 return NULL;
1229}
1230
1231/*
1232 * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block().
1233 * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their
1234 * refcount elevated by one when they're in an LRU. A buffer can only appear
1235 * once in a particular CPU's LRU. A single buffer can be present in multiple
1236 * CPU's LRUs at the same time.
1237 *
1238 * This is a transparent caching front-end to sb_bread(), sb_getblk() and
1239 * sb_find_get_block().
1240 *
1241 * The LRUs themselves only need locking against invalidate_bh_lrus. We use
1242 * a local interrupt disable for that.
1243 */
1244
Sebastien Buisson86cf78d2014-10-09 15:29:38 -07001245#define BH_LRU_SIZE 16
Linus Torvalds1da177e2005-04-16 15:20:36 -07001246
1247struct bh_lru {
1248 struct buffer_head *bhs[BH_LRU_SIZE];
1249};
1250
1251static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }};
1252
1253#ifdef CONFIG_SMP
1254#define bh_lru_lock() local_irq_disable()
1255#define bh_lru_unlock() local_irq_enable()
1256#else
1257#define bh_lru_lock() preempt_disable()
1258#define bh_lru_unlock() preempt_enable()
1259#endif
1260
1261static inline void check_irqs_on(void)
1262{
1263#ifdef irqs_disabled
1264 BUG_ON(irqs_disabled());
1265#endif
1266}
1267
1268/*
Eric Biggers241f01f2017-07-10 15:47:29 -07001269 * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is
1270 * inserted at the front, and the buffer_head at the back if any is evicted.
1271 * Or, if already in the LRU it is moved to the front.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 */
1273static void bh_lru_install(struct buffer_head *bh)
1274{
Eric Biggers241f01f2017-07-10 15:47:29 -07001275 struct buffer_head *evictee = bh;
1276 struct bh_lru *b;
1277 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001278
1279 check_irqs_on();
Minchan Kim8cc621d2021-05-04 18:37:00 -07001280 /*
1281 * the refcount of buffer_head in bh_lru prevents dropping the
1282 * attached page(i.e., try_to_free_buffers) so it could cause
1283 * failing page migration.
1284 * Skip putting upcoming bh into bh_lru until migration is done.
1285 */
1286 if (lru_cache_disabled())
1287 return;
1288
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290
Eric Biggers241f01f2017-07-10 15:47:29 -07001291 b = this_cpu_ptr(&bh_lrus);
1292 for (i = 0; i < BH_LRU_SIZE; i++) {
1293 swap(evictee, b->bhs[i]);
1294 if (evictee == bh) {
1295 bh_lru_unlock();
1296 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001297 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001299
Eric Biggers241f01f2017-07-10 15:47:29 -07001300 get_bh(bh);
1301 bh_lru_unlock();
1302 brelse(evictee);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001303}
1304
1305/*
1306 * Look up the bh in this cpu's LRU. If it's there, move it to the head.
1307 */
Arjan van de Ven858119e2006-01-14 13:20:43 -08001308static struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001309lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001310{
1311 struct buffer_head *ret = NULL;
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001312 unsigned int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313
1314 check_irqs_on();
1315 bh_lru_lock();
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 for (i = 0; i < BH_LRU_SIZE; i++) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001317 struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001318
Zach Brown9470dd52014-10-13 15:55:05 -07001319 if (bh && bh->b_blocknr == block && bh->b_bdev == bdev &&
1320 bh->b_size == size) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001321 if (i) {
1322 while (i) {
Christoph Lameterc7b92512010-12-06 11:16:28 -06001323 __this_cpu_write(bh_lrus.bhs[i],
1324 __this_cpu_read(bh_lrus.bhs[i - 1]));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001325 i--;
1326 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06001327 __this_cpu_write(bh_lrus.bhs[0], bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001328 }
1329 get_bh(bh);
1330 ret = bh;
1331 break;
1332 }
1333 }
1334 bh_lru_unlock();
1335 return ret;
1336}
1337
1338/*
1339 * Perform a pagecache lookup for the matching buffer. If it's there, refresh
1340 * it in the LRU and mark it as accessed. If it is not present then return
1341 * NULL
1342 */
1343struct buffer_head *
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001344__find_get_block(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345{
1346 struct buffer_head *bh = lookup_bh_lru(bdev, block, size);
1347
1348 if (bh == NULL) {
Mel Gorman2457aec2014-06-04 16:10:31 -07001349 /* __find_get_block_slow will mark the page accessed */
Coywolf Qi Hunt385fd4c2005-11-07 00:59:39 -08001350 bh = __find_get_block_slow(bdev, block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 if (bh)
1352 bh_lru_install(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001353 } else
Linus Torvalds1da177e2005-04-16 15:20:36 -07001354 touch_buffer(bh);
Mel Gorman2457aec2014-06-04 16:10:31 -07001355
Linus Torvalds1da177e2005-04-16 15:20:36 -07001356 return bh;
1357}
1358EXPORT_SYMBOL(__find_get_block);
1359
1360/*
Gioh Kim3b5e6452014-09-04 22:04:42 -04001361 * __getblk_gfp() will locate (and, if necessary, create) the buffer_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 * which corresponds to the passed block_device, block and size. The
1363 * returned buffer has its reference count incremented.
1364 *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001365 * __getblk_gfp() will lock up the machine if grow_dev_page's
1366 * try_to_free_buffers() attempt is failing. FIXME, perhaps?
Linus Torvalds1da177e2005-04-16 15:20:36 -07001367 */
1368struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001369__getblk_gfp(struct block_device *bdev, sector_t block,
1370 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371{
1372 struct buffer_head *bh = __find_get_block(bdev, block, size);
1373
1374 might_sleep();
1375 if (bh == NULL)
Gioh Kim3b5e6452014-09-04 22:04:42 -04001376 bh = __getblk_slow(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001377 return bh;
1378}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001379EXPORT_SYMBOL(__getblk_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001380
1381/*
1382 * Do async read-ahead on a buffer..
1383 */
Tomasz Kvarsin3991d3b2007-02-12 00:52:14 -08001384void __breadahead(struct block_device *bdev, sector_t block, unsigned size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385{
1386 struct buffer_head *bh = __getblk(bdev, block, size);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001387 if (likely(bh)) {
Christoph Hellwig70246282016-07-19 11:28:41 +02001388 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
Andrew Mortona3e713b2005-10-30 15:03:15 -08001389 brelse(bh);
1390 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391}
1392EXPORT_SYMBOL(__breadahead);
1393
Roman Gushchind87f6392020-02-28 16:14:11 -08001394void __breadahead_gfp(struct block_device *bdev, sector_t block, unsigned size,
1395 gfp_t gfp)
1396{
1397 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
1398 if (likely(bh)) {
1399 ll_rw_block(REQ_OP_READ, REQ_RAHEAD, 1, &bh);
1400 brelse(bh);
1401 }
1402}
1403EXPORT_SYMBOL(__breadahead_gfp);
1404
Linus Torvalds1da177e2005-04-16 15:20:36 -07001405/**
Gioh Kim3b5e6452014-09-04 22:04:42 -04001406 * __bread_gfp() - reads a specified block and returns the bh
Martin Waitz67be2dd2005-05-01 08:59:26 -07001407 * @bdev: the block_device to read from
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 * @block: number of block
1409 * @size: size (in bytes) to read
Gioh Kim3b5e6452014-09-04 22:04:42 -04001410 * @gfp: page allocation flag
1411 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001412 * Reads a specified block, and returns buffer head that contains it.
Gioh Kim3b5e6452014-09-04 22:04:42 -04001413 * The page cache can be allocated from non-movable area
1414 * not to prevent page migration if you set gfp to zero.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001415 * It returns NULL if the block was unreadable.
1416 */
1417struct buffer_head *
Gioh Kim3b5e6452014-09-04 22:04:42 -04001418__bread_gfp(struct block_device *bdev, sector_t block,
1419 unsigned size, gfp_t gfp)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420{
Gioh Kim3b5e6452014-09-04 22:04:42 -04001421 struct buffer_head *bh = __getblk_gfp(bdev, block, size, gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422
Andrew Mortona3e713b2005-10-30 15:03:15 -08001423 if (likely(bh) && !buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 bh = __bread_slow(bh);
1425 return bh;
1426}
Gioh Kim3b5e6452014-09-04 22:04:42 -04001427EXPORT_SYMBOL(__bread_gfp);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428
Minchan Kim8cc621d2021-05-04 18:37:00 -07001429static void __invalidate_bh_lrus(struct bh_lru *b)
1430{
1431 int i;
1432
1433 for (i = 0; i < BH_LRU_SIZE; i++) {
1434 brelse(b->bhs[i]);
1435 b->bhs[i] = NULL;
1436 }
1437}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438/*
1439 * invalidate_bh_lrus() is called rarely - but not only at unmount.
1440 * This doesn't race because it runs in each cpu either in irq
1441 * or with preempt disabled.
1442 */
1443static void invalidate_bh_lru(void *arg)
1444{
1445 struct bh_lru *b = &get_cpu_var(bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001446
Minchan Kim8cc621d2021-05-04 18:37:00 -07001447 __invalidate_bh_lrus(b);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448 put_cpu_var(bh_lrus);
1449}
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001450
Minchan Kim8cc621d2021-05-04 18:37:00 -07001451bool has_bh_in_lru(int cpu, void *dummy)
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001452{
1453 struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu);
1454 int i;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001455
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001456 for (i = 0; i < BH_LRU_SIZE; i++) {
1457 if (b->bhs[i])
Saurav Girepunje1d706672019-11-30 17:49:15 -08001458 return true;
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001459 }
1460
Saurav Girepunje1d706672019-11-30 17:49:15 -08001461 return false;
Gilad Ben-Yossef42be35d2012-03-28 14:42:45 -07001462}
1463
Peter Zijlstraf9a14392007-05-06 14:49:55 -07001464void invalidate_bh_lrus(void)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465{
Sebastian Andrzej Siewiorcb923152020-01-17 10:01:37 +01001466 on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001467}
Nick Piggin9db55792008-02-08 04:19:49 -08001468EXPORT_SYMBOL_GPL(invalidate_bh_lrus);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469
Minchan Kim243418e2021-09-24 15:43:47 -07001470/*
1471 * It's called from workqueue context so we need a bh_lru_lock to close
1472 * the race with preemption/irq.
1473 */
1474void invalidate_bh_lrus_cpu(void)
Minchan Kim8cc621d2021-05-04 18:37:00 -07001475{
1476 struct bh_lru *b;
1477
1478 bh_lru_lock();
Minchan Kim243418e2021-09-24 15:43:47 -07001479 b = this_cpu_ptr(&bh_lrus);
Minchan Kim8cc621d2021-05-04 18:37:00 -07001480 __invalidate_bh_lrus(b);
1481 bh_lru_unlock();
1482}
1483
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484void set_bh_page(struct buffer_head *bh,
1485 struct page *page, unsigned long offset)
1486{
1487 bh->b_page = page;
Eric Sesterhenne827f922006-03-26 18:24:46 +02001488 BUG_ON(offset >= PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489 if (PageHighMem(page))
1490 /*
1491 * This catches illegal uses and preserves the offset:
1492 */
1493 bh->b_data = (char *)(0 + offset);
1494 else
1495 bh->b_data = page_address(page) + offset;
1496}
1497EXPORT_SYMBOL(set_bh_page);
1498
1499/*
1500 * Called when truncating a buffer on a page completely.
1501 */
Mel Gormane7470ee2014-06-04 16:10:29 -07001502
1503/* Bits that are cleared during an invalidate */
1504#define BUFFER_FLAGS_DISCARD \
1505 (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \
1506 1 << BH_Delay | 1 << BH_Unwritten)
1507
Arjan van de Ven858119e2006-01-14 13:20:43 -08001508static void discard_buffer(struct buffer_head * bh)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001509{
Mel Gormane7470ee2014-06-04 16:10:29 -07001510 unsigned long b_state, b_state_old;
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512 lock_buffer(bh);
1513 clear_buffer_dirty(bh);
1514 bh->b_bdev = NULL;
Mel Gormane7470ee2014-06-04 16:10:29 -07001515 b_state = bh->b_state;
1516 for (;;) {
1517 b_state_old = cmpxchg(&bh->b_state, b_state,
1518 (b_state & ~BUFFER_FLAGS_DISCARD));
1519 if (b_state_old == b_state)
1520 break;
1521 b_state = b_state_old;
1522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001523 unlock_buffer(bh);
1524}
1525
1526/**
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001527 * block_invalidatepage - invalidate part or all of a buffer-backed page
Linus Torvalds1da177e2005-04-16 15:20:36 -07001528 *
1529 * @page: the page which is affected
Lukas Czernerd47992f2013-05-21 23:17:23 -04001530 * @offset: start of the range to invalidate
1531 * @length: length of the range to invalidate
Linus Torvalds1da177e2005-04-16 15:20:36 -07001532 *
1533 * block_invalidatepage() is called when all or part of the page has become
Wang Sheng-Hui814e1d22011-09-01 08:22:57 +08001534 * invalidated by a truncate operation.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535 *
1536 * block_invalidatepage() does not have to release all buffers, but it must
1537 * ensure that no dirty buffer is left outside @offset and that no I/O
1538 * is underway against any of the blocks which are outside the truncation
1539 * point. Because the caller is about to free (and possibly reuse) those
1540 * blocks on-disk.
1541 */
Lukas Czernerd47992f2013-05-21 23:17:23 -04001542void block_invalidatepage(struct page *page, unsigned int offset,
1543 unsigned int length)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544{
1545 struct buffer_head *head, *bh, *next;
1546 unsigned int curr_off = 0;
Lukas Czernerd47992f2013-05-21 23:17:23 -04001547 unsigned int stop = length + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548
1549 BUG_ON(!PageLocked(page));
1550 if (!page_has_buffers(page))
1551 goto out;
1552
Lukas Czernerd47992f2013-05-21 23:17:23 -04001553 /*
1554 * Check for overflow
1555 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001556 BUG_ON(stop > PAGE_SIZE || stop < length);
Lukas Czernerd47992f2013-05-21 23:17:23 -04001557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 head = page_buffers(page);
1559 bh = head;
1560 do {
1561 unsigned int next_off = curr_off + bh->b_size;
1562 next = bh->b_this_page;
1563
1564 /*
Lukas Czernerd47992f2013-05-21 23:17:23 -04001565 * Are we still fully in range ?
1566 */
1567 if (next_off > stop)
1568 goto out;
1569
1570 /*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001571 * is this block fully invalidated?
1572 */
1573 if (offset <= curr_off)
1574 discard_buffer(bh);
1575 curr_off = next_off;
1576 bh = next;
1577 } while (bh != head);
1578
1579 /*
1580 * We release buffers only if the entire page is being invalidated.
1581 * The get_block cached value has been unconditionally invalidated,
1582 * so real IO is not possible anymore.
1583 */
Jeff Moyer31724852018-04-05 16:25:01 -07001584 if (length == PAGE_SIZE)
NeilBrown2ff28e22006-03-26 01:37:18 -08001585 try_to_release_page(page, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001586out:
NeilBrown2ff28e22006-03-26 01:37:18 -08001587 return;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588}
1589EXPORT_SYMBOL(block_invalidatepage);
1590
Lukas Czernerd47992f2013-05-21 23:17:23 -04001591
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592/*
1593 * We attach and possibly dirty the buffers atomically wrt
1594 * __set_page_dirty_buffers() via private_lock. try_to_free_buffers
1595 * is already excluded via the page lock.
1596 */
1597void create_empty_buffers(struct page *page,
1598 unsigned long blocksize, unsigned long b_state)
1599{
1600 struct buffer_head *bh, *head, *tail;
1601
Jens Axboe640ab982017-09-27 05:40:16 -06001602 head = alloc_page_buffers(page, blocksize, true);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001603 bh = head;
1604 do {
1605 bh->b_state |= b_state;
1606 tail = bh;
1607 bh = bh->b_this_page;
1608 } while (bh);
1609 tail->b_this_page = head;
1610
1611 spin_lock(&page->mapping->private_lock);
1612 if (PageUptodate(page) || PageDirty(page)) {
1613 bh = head;
1614 do {
1615 if (PageDirty(page))
1616 set_buffer_dirty(bh);
1617 if (PageUptodate(page))
1618 set_buffer_uptodate(bh);
1619 bh = bh->b_this_page;
1620 } while (bh != head);
1621 }
Guoqing Jiang45dcfc22020-06-01 21:47:48 -07001622 attach_page_private(page, head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623 spin_unlock(&page->mapping->private_lock);
1624}
1625EXPORT_SYMBOL(create_empty_buffers);
1626
Jan Kara29f3ad72016-11-04 18:08:11 +01001627/**
1628 * clean_bdev_aliases: clean a range of buffers in block device
1629 * @bdev: Block device to clean buffers in
1630 * @block: Start of a range of blocks to clean
1631 * @len: Number of blocks to clean
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632 *
Jan Kara29f3ad72016-11-04 18:08:11 +01001633 * We are taking a range of blocks for data and we don't want writeback of any
1634 * buffer-cache aliases starting from return from this function and until the
1635 * moment when something will explicitly mark the buffer dirty (hopefully that
1636 * will not happen until we will free that block ;-) We don't even need to mark
1637 * it not-uptodate - nobody can expect anything from a newly allocated buffer
1638 * anyway. We used to use unmap_buffer() for such invalidation, but that was
1639 * wrong. We definitely don't want to mark the alias unmapped, for example - it
1640 * would confuse anyone who might pick it with bread() afterwards...
1641 *
1642 * Also.. Note that bforget() doesn't lock the buffer. So there can be
1643 * writeout I/O going on against recently-freed buffers. We don't wait on that
1644 * I/O in bforget() - it's more efficient to wait on the I/O only if we really
1645 * need to. That happens here.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646 */
Jan Kara29f3ad72016-11-04 18:08:11 +01001647void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648{
Jan Kara29f3ad72016-11-04 18:08:11 +01001649 struct inode *bd_inode = bdev->bd_inode;
1650 struct address_space *bd_mapping = bd_inode->i_mapping;
1651 struct pagevec pvec;
1652 pgoff_t index = block >> (PAGE_SHIFT - bd_inode->i_blkbits);
1653 pgoff_t end;
Jan Karac10f7782017-09-06 16:21:24 -07001654 int i, count;
Jan Kara29f3ad72016-11-04 18:08:11 +01001655 struct buffer_head *bh;
1656 struct buffer_head *head;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657
Jan Kara29f3ad72016-11-04 18:08:11 +01001658 end = (block + len - 1) >> (PAGE_SHIFT - bd_inode->i_blkbits);
Mel Gorman86679822017-11-15 17:37:52 -08001659 pagevec_init(&pvec);
Jan Kara397162f2017-09-06 16:21:43 -07001660 while (pagevec_lookup_range(&pvec, bd_mapping, &index, end)) {
Jan Karac10f7782017-09-06 16:21:24 -07001661 count = pagevec_count(&pvec);
1662 for (i = 0; i < count; i++) {
Jan Kara29f3ad72016-11-04 18:08:11 +01001663 struct page *page = pvec.pages[i];
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664
Jan Kara29f3ad72016-11-04 18:08:11 +01001665 if (!page_has_buffers(page))
1666 continue;
1667 /*
1668 * We use page lock instead of bd_mapping->private_lock
1669 * to pin buffers here since we can afford to sleep and
1670 * it scales better than a global spinlock lock.
1671 */
1672 lock_page(page);
1673 /* Recheck when the page is locked which pins bhs */
1674 if (!page_has_buffers(page))
1675 goto unlock_page;
1676 head = page_buffers(page);
1677 bh = head;
1678 do {
Chandan Rajendra6c006a92016-12-25 19:01:03 +05301679 if (!buffer_mapped(bh) || (bh->b_blocknr < block))
Jan Kara29f3ad72016-11-04 18:08:11 +01001680 goto next;
1681 if (bh->b_blocknr >= block + len)
1682 break;
1683 clear_buffer_dirty(bh);
1684 wait_on_buffer(bh);
1685 clear_buffer_req(bh);
1686next:
1687 bh = bh->b_this_page;
1688 } while (bh != head);
1689unlock_page:
1690 unlock_page(page);
1691 }
1692 pagevec_release(&pvec);
1693 cond_resched();
Jan Karac10f7782017-09-06 16:21:24 -07001694 /* End of range already reached? */
1695 if (index > end || !index)
1696 break;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001697 }
1698}
Jan Kara29f3ad72016-11-04 18:08:11 +01001699EXPORT_SYMBOL(clean_bdev_aliases);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700
1701/*
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001702 * Size is a power-of-two in the range 512..PAGE_SIZE,
1703 * and the case we care about most is PAGE_SIZE.
1704 *
1705 * So this *could* possibly be written with those
1706 * constraints in mind (relevant mostly if some
1707 * architecture has a slow bit-scan instruction)
1708 */
1709static inline int block_size_bits(unsigned int blocksize)
1710{
1711 return ilog2(blocksize);
1712}
1713
1714static struct buffer_head *create_page_buffers(struct page *page, struct inode *inode, unsigned int b_state)
1715{
1716 BUG_ON(!PageLocked(page));
1717
1718 if (!page_has_buffers(page))
Mark Rutland6aa7de02017-10-23 14:07:29 -07001719 create_empty_buffers(page, 1 << READ_ONCE(inode->i_blkbits),
1720 b_state);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001721 return page_buffers(page);
1722}
1723
1724/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001725 * NOTE! All mapped/uptodate combinations are valid:
1726 *
1727 * Mapped Uptodate Meaning
1728 *
1729 * No No "unknown" - must do get_block()
1730 * No Yes "hole" - zero-filled
1731 * Yes No "allocated" - allocated on disk, not read in
1732 * Yes Yes "valid" - allocated and up-to-date in memory.
1733 *
1734 * "Dirty" is valid only with the last case (mapped+uptodate).
1735 */
1736
1737/*
1738 * While block_write_full_page is writing back the dirty buffers under
1739 * the page lock, whoever dirtied the buffers may decide to clean them
1740 * again at any time. We handle that by only looking at the buffer
1741 * state inside lock_buffer().
1742 *
1743 * If block_write_full_page() is called for regular writeback
1744 * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a
1745 * locked buffer. This only can happen if someone has written the buffer
1746 * directly, with submit_bh(). At the address_space level PageWriteback
1747 * prevents this contention from occurring.
Theodore Ts'o6e34eedd2009-04-07 18:12:43 -04001748 *
1749 * If block_write_full_page() is called with wbc->sync_mode ==
Christoph Hellwig70fd7612016-11-01 07:40:10 -06001750 * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this
Jens Axboe721a9602011-03-09 11:56:30 +01001751 * causes the writes to be flagged as synchronous writes.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752 */
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001753int __block_write_full_page(struct inode *inode, struct page *page,
Chris Mason35c80d52009-04-15 13:22:38 -04001754 get_block_t *get_block, struct writeback_control *wbc,
1755 bh_end_io_t *handler)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756{
1757 int err;
1758 sector_t block;
1759 sector_t last_block;
Andrew Mortonf0fbd5f2005-05-05 16:15:48 -07001760 struct buffer_head *bh, *head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001761 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762 int nr_underway = 0;
Jens Axboe76372412016-11-01 10:00:38 -06001763 int write_flags = wbc_to_write_flags(wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001764
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001765 head = create_page_buffers(page, inode,
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766 (1 << BH_Dirty)|(1 << BH_Uptodate));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767
1768 /*
1769 * Be very careful. We have no exclusion from __set_page_dirty_buffers
1770 * here, and the (potentially unmapped) buffers may become dirty at
1771 * any time. If a buffer becomes dirty here after we've inspected it
1772 * then we just miss that fact, and the page stays dirty.
1773 *
1774 * Buffers outside i_size may be dirtied by __set_page_dirty_buffers;
1775 * handle that here by just cleaning them.
1776 */
1777
Linus Torvalds1da177e2005-04-16 15:20:36 -07001778 bh = head;
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001779 blocksize = bh->b_size;
1780 bbits = block_size_bits(blocksize);
1781
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03001782 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08001783 last_block = (i_size_read(inode) - 1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001784
1785 /*
1786 * Get all the dirty buffers mapped to disk addresses and
1787 * handle any aliases from the underlying blockdev's mapping.
1788 */
1789 do {
1790 if (block > last_block) {
1791 /*
1792 * mapped buffers outside i_size will occur, because
1793 * this page can be outside i_size when there is a
1794 * truncate in progress.
1795 */
1796 /*
1797 * The buffer was zeroed by block_write_full_page()
1798 */
1799 clear_buffer_dirty(bh);
1800 set_buffer_uptodate(bh);
Alex Tomas29a814d2008-07-11 19:27:31 -04001801 } else if ((!buffer_mapped(bh) || buffer_delay(bh)) &&
1802 buffer_dirty(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08001803 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 err = get_block(inode, block, bh, 1);
1805 if (err)
1806 goto recover;
Alex Tomas29a814d2008-07-11 19:27:31 -04001807 clear_buffer_delay(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001808 if (buffer_new(bh)) {
1809 /* blockdev mappings never come here */
1810 clear_buffer_new(bh);
Jan Karae64855c2016-11-04 18:08:15 +01001811 clean_bdev_bh_alias(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812 }
1813 }
1814 bh = bh->b_this_page;
1815 block++;
1816 } while (bh != head);
1817
1818 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 if (!buffer_mapped(bh))
1820 continue;
1821 /*
1822 * If it's a fully non-blocking write attempt and we cannot
1823 * lock the buffer then redirty the page. Note that this can
Jens Axboe5b0830c2009-09-23 19:37:09 +02001824 * potentially cause a busy-wait loop from writeback threads
1825 * and kswapd activity, but those code paths have their own
1826 * higher-level throttling.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 */
Wu Fengguang1b430be2010-10-26 14:21:26 -07001828 if (wbc->sync_mode != WB_SYNC_NONE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 lock_buffer(bh);
Nick Pigginca5de402008-08-02 12:02:13 +02001830 } else if (!trylock_buffer(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 redirty_page_for_writepage(wbc, page);
1832 continue;
1833 }
1834 if (test_clear_buffer_dirty(bh)) {
Chris Mason35c80d52009-04-15 13:22:38 -04001835 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836 } else {
1837 unlock_buffer(bh);
1838 }
1839 } while ((bh = bh->b_this_page) != head);
1840
1841 /*
1842 * The page and its buffers are protected by PageWriteback(), so we can
1843 * drop the bh refcounts early.
1844 */
1845 BUG_ON(PageWriteback(page));
1846 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001847
1848 do {
1849 struct buffer_head *next = bh->b_this_page;
1850 if (buffer_async_write(bh)) {
Jens Axboe8e8f9292017-06-27 09:30:05 -06001851 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1852 inode->i_write_hint, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001853 nr_underway++;
1854 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001855 bh = next;
1856 } while (bh != head);
Andrew Morton05937ba2005-05-05 16:15:47 -07001857 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859 err = 0;
1860done:
1861 if (nr_underway == 0) {
1862 /*
1863 * The page was marked dirty, but the buffers were
1864 * clean. Someone wrote them back by hand with
1865 * ll_rw_block/submit_bh. A rare case.
1866 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867 end_page_writeback(page);
Nick Piggin3d67f2d2007-05-06 14:49:05 -07001868
Linus Torvalds1da177e2005-04-16 15:20:36 -07001869 /*
1870 * The page and buffer_heads can be released at any time from
1871 * here on.
1872 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001873 }
1874 return err;
1875
1876recover:
1877 /*
1878 * ENOSPC, or some other error. We may already have added some
1879 * blocks to the file, so we need to write these out to avoid
1880 * exposing stale data.
1881 * The page is currently locked and not marked for writeback
1882 */
1883 bh = head;
1884 /* Recovery: lock and submit the mapped buffers */
1885 do {
Alex Tomas29a814d2008-07-11 19:27:31 -04001886 if (buffer_mapped(bh) && buffer_dirty(bh) &&
1887 !buffer_delay(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888 lock_buffer(bh);
Chris Mason35c80d52009-04-15 13:22:38 -04001889 mark_buffer_async_write_endio(bh, handler);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890 } else {
1891 /*
1892 * The buffer may have been set dirty during
1893 * attachment to a dirty page.
1894 */
1895 clear_buffer_dirty(bh);
1896 }
1897 } while ((bh = bh->b_this_page) != head);
1898 SetPageError(page);
1899 BUG_ON(PageWriteback(page));
Andrew Morton7e4c3692007-05-08 00:23:27 -07001900 mapping_set_error(page->mapping, err);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001901 set_page_writeback(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902 do {
1903 struct buffer_head *next = bh->b_this_page;
1904 if (buffer_async_write(bh)) {
1905 clear_buffer_dirty(bh);
Jens Axboe8e8f9292017-06-27 09:30:05 -06001906 submit_bh_wbc(REQ_OP_WRITE, write_flags, bh,
1907 inode->i_write_hint, wbc);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001908 nr_underway++;
1909 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001910 bh = next;
1911 } while (bh != head);
Nick Pigginffda9d32007-02-20 13:57:54 -08001912 unlock_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913 goto done;
1914}
Benjamin Marzinskib4bba382016-06-27 09:58:40 -05001915EXPORT_SYMBOL(__block_write_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916
Nick Pigginafddba42007-10-16 01:25:01 -07001917/*
1918 * If a page has any new buffers, zero them out here, and mark them uptodate
1919 * and dirty so they'll be written out (in order to prevent uninitialised
1920 * block data from leaking). And clear the new bit.
1921 */
1922void page_zero_new_buffers(struct page *page, unsigned from, unsigned to)
1923{
1924 unsigned int block_start, block_end;
1925 struct buffer_head *head, *bh;
1926
1927 BUG_ON(!PageLocked(page));
1928 if (!page_has_buffers(page))
1929 return;
1930
1931 bh = head = page_buffers(page);
1932 block_start = 0;
1933 do {
1934 block_end = block_start + bh->b_size;
1935
1936 if (buffer_new(bh)) {
1937 if (block_end > from && block_start < to) {
1938 if (!PageUptodate(page)) {
1939 unsigned start, size;
1940
1941 start = max(from, block_start);
1942 size = min(to, block_end) - start;
1943
Christoph Lametereebd2aa2008-02-04 22:28:29 -08001944 zero_user(page, start, size);
Nick Pigginafddba42007-10-16 01:25:01 -07001945 set_buffer_uptodate(bh);
1946 }
1947
1948 clear_buffer_new(bh);
1949 mark_buffer_dirty(bh);
1950 }
1951 }
1952
1953 block_start = block_end;
1954 bh = bh->b_this_page;
1955 } while (bh != head);
1956}
1957EXPORT_SYMBOL(page_zero_new_buffers);
1958
Christoph Hellwigae259a92016-06-21 09:23:11 +10001959static void
1960iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh,
Christoph Hellwig6d49cc82021-08-10 18:33:05 -07001961 const struct iomap *iomap)
Christoph Hellwigae259a92016-06-21 09:23:11 +10001962{
1963 loff_t offset = block << inode->i_blkbits;
1964
1965 bh->b_bdev = iomap->bdev;
1966
1967 /*
1968 * Block points to offset in file we need to map, iomap contains
1969 * the offset at which the map starts. If the map ends before the
1970 * current block, then do not map the buffer and let the caller
1971 * handle it.
1972 */
1973 BUG_ON(offset >= iomap->offset + iomap->length);
1974
1975 switch (iomap->type) {
1976 case IOMAP_HOLE:
1977 /*
1978 * If the buffer is not up to date or beyond the current EOF,
1979 * we need to mark it as new to ensure sub-block zeroing is
1980 * executed if necessary.
1981 */
1982 if (!buffer_uptodate(bh) ||
1983 (offset >= i_size_read(inode)))
1984 set_buffer_new(bh);
1985 break;
1986 case IOMAP_DELALLOC:
1987 if (!buffer_uptodate(bh) ||
1988 (offset >= i_size_read(inode)))
1989 set_buffer_new(bh);
1990 set_buffer_uptodate(bh);
1991 set_buffer_mapped(bh);
1992 set_buffer_delay(bh);
1993 break;
1994 case IOMAP_UNWRITTEN:
1995 /*
Andreas Gruenbacher3d7b6b212018-06-19 15:10:55 -07001996 * For unwritten regions, we always need to ensure that regions
1997 * in the block we are not writing to are zeroed. Mark the
1998 * buffer as new to ensure this.
Christoph Hellwigae259a92016-06-21 09:23:11 +10001999 */
2000 set_buffer_new(bh);
2001 set_buffer_unwritten(bh);
Gustavo A. R. Silvadf561f662020-08-23 17:36:59 -05002002 fallthrough;
Christoph Hellwigae259a92016-06-21 09:23:11 +10002003 case IOMAP_MAPPED:
Andreas Gruenbacher3d7b6b212018-06-19 15:10:55 -07002004 if ((iomap->flags & IOMAP_F_NEW) ||
2005 offset >= i_size_read(inode))
Christoph Hellwigae259a92016-06-21 09:23:11 +10002006 set_buffer_new(bh);
Andreas Gruenbacher19fe5f62017-10-01 17:55:54 -04002007 bh->b_blocknr = (iomap->addr + offset - iomap->offset) >>
2008 inode->i_blkbits;
Christoph Hellwigae259a92016-06-21 09:23:11 +10002009 set_buffer_mapped(bh);
2010 break;
2011 }
2012}
2013
2014int __block_write_begin_int(struct page *page, loff_t pos, unsigned len,
Christoph Hellwig6d49cc82021-08-10 18:33:05 -07002015 get_block_t *get_block, const struct iomap *iomap)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002016{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002017 unsigned from = pos & (PAGE_SIZE - 1);
Christoph Hellwigebdec242010-10-06 10:47:23 +02002018 unsigned to = from + len;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002019 struct inode *inode = page->mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002020 unsigned block_start, block_end;
2021 sector_t block;
2022 int err = 0;
2023 unsigned blocksize, bbits;
2024 struct buffer_head *bh, *head, *wait[2], **wait_bh=wait;
2025
2026 BUG_ON(!PageLocked(page));
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002027 BUG_ON(from > PAGE_SIZE);
2028 BUG_ON(to > PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002029 BUG_ON(from > to);
2030
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002031 head = create_page_buffers(page, inode, 0);
2032 blocksize = head->b_size;
2033 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002034
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002035 block = (sector_t)page->index << (PAGE_SHIFT - bbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002036
2037 for(bh = head, block_start = 0; bh != head || !block_start;
2038 block++, block_start=block_end, bh = bh->b_this_page) {
2039 block_end = block_start + blocksize;
2040 if (block_end <= from || block_start >= to) {
2041 if (PageUptodate(page)) {
2042 if (!buffer_uptodate(bh))
2043 set_buffer_uptodate(bh);
2044 }
2045 continue;
2046 }
2047 if (buffer_new(bh))
2048 clear_buffer_new(bh);
2049 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002050 WARN_ON(bh->b_size != blocksize);
Christoph Hellwigae259a92016-06-21 09:23:11 +10002051 if (get_block) {
2052 err = get_block(inode, block, bh, 1);
2053 if (err)
2054 break;
2055 } else {
2056 iomap_to_bh(inode, block, bh, iomap);
2057 }
2058
Linus Torvalds1da177e2005-04-16 15:20:36 -07002059 if (buffer_new(bh)) {
Jan Karae64855c2016-11-04 18:08:15 +01002060 clean_bdev_bh_alias(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061 if (PageUptodate(page)) {
Nick Piggin637aff42007-10-16 01:25:00 -07002062 clear_buffer_new(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002063 set_buffer_uptodate(bh);
Nick Piggin637aff42007-10-16 01:25:00 -07002064 mark_buffer_dirty(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002065 continue;
2066 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002067 if (block_end > to || block_start < from)
2068 zero_user_segments(page,
2069 to, block_end,
2070 block_start, from);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002071 continue;
2072 }
2073 }
2074 if (PageUptodate(page)) {
2075 if (!buffer_uptodate(bh))
2076 set_buffer_uptodate(bh);
2077 continue;
2078 }
2079 if (!buffer_uptodate(bh) && !buffer_delay(bh) &&
David Chinner33a266d2007-02-12 00:51:41 -08002080 !buffer_unwritten(bh) &&
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081 (block_start < from || block_end > to)) {
Mike Christiedfec8a12016-06-05 14:31:44 -05002082 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083 *wait_bh++=bh;
2084 }
2085 }
2086 /*
2087 * If we issued read requests - let them complete.
2088 */
2089 while(wait_bh > wait) {
2090 wait_on_buffer(*--wait_bh);
2091 if (!buffer_uptodate(*wait_bh))
Nick Pigginf3ddbdc2005-05-05 16:15:45 -07002092 err = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002093 }
Jan Karaf9f07b62011-06-14 00:58:27 +02002094 if (unlikely(err))
Nick Pigginafddba42007-10-16 01:25:01 -07002095 page_zero_new_buffers(page, from, to);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 return err;
2097}
Christoph Hellwigae259a92016-06-21 09:23:11 +10002098
2099int __block_write_begin(struct page *page, loff_t pos, unsigned len,
2100 get_block_t *get_block)
2101{
2102 return __block_write_begin_int(page, pos, len, get_block, NULL);
2103}
Christoph Hellwigebdec242010-10-06 10:47:23 +02002104EXPORT_SYMBOL(__block_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002105
2106static int __block_commit_write(struct inode *inode, struct page *page,
2107 unsigned from, unsigned to)
2108{
2109 unsigned block_start, block_end;
2110 int partial = 0;
2111 unsigned blocksize;
2112 struct buffer_head *bh, *head;
2113
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002114 bh = head = page_buffers(page);
2115 blocksize = bh->b_size;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002116
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002117 block_start = 0;
2118 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002119 block_end = block_start + blocksize;
2120 if (block_end <= from || block_start >= to) {
2121 if (!buffer_uptodate(bh))
2122 partial = 1;
2123 } else {
2124 set_buffer_uptodate(bh);
2125 mark_buffer_dirty(bh);
2126 }
Yang Guo4ebd3ae2021-02-24 12:02:48 -08002127 if (buffer_new(bh))
2128 clear_buffer_new(bh);
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002129
2130 block_start = block_end;
2131 bh = bh->b_this_page;
2132 } while (bh != head);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002133
2134 /*
2135 * If this is a partial write which happened to make all buffers
2136 * uptodate then we can optimize away a bogus readpage() for
2137 * the next read(). Here we 'discover' whether the page went
2138 * uptodate as a result of this (potentially partial) write.
2139 */
2140 if (!partial)
2141 SetPageUptodate(page);
2142 return 0;
2143}
2144
2145/*
Christoph Hellwig155130a2010-06-04 11:29:58 +02002146 * block_write_begin takes care of the basic task of block allocation and
2147 * bringing partial write blocks uptodate first.
2148 *
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002149 * The filesystem needs to handle block truncation upon failure.
Nick Pigginafddba42007-10-16 01:25:01 -07002150 */
Christoph Hellwig155130a2010-06-04 11:29:58 +02002151int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len,
2152 unsigned flags, struct page **pagep, get_block_t *get_block)
Nick Pigginafddba42007-10-16 01:25:01 -07002153{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002154 pgoff_t index = pos >> PAGE_SHIFT;
Nick Pigginafddba42007-10-16 01:25:01 -07002155 struct page *page;
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002156 int status;
Nick Pigginafddba42007-10-16 01:25:01 -07002157
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002158 page = grab_cache_page_write_begin(mapping, index, flags);
2159 if (!page)
2160 return -ENOMEM;
Nick Pigginafddba42007-10-16 01:25:01 -07002161
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002162 status = __block_write_begin(page, pos, len, get_block);
Nick Pigginafddba42007-10-16 01:25:01 -07002163 if (unlikely(status)) {
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002164 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002165 put_page(page);
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002166 page = NULL;
Nick Pigginafddba42007-10-16 01:25:01 -07002167 }
2168
Christoph Hellwig6e1db882010-06-04 11:29:57 +02002169 *pagep = page;
Nick Pigginafddba42007-10-16 01:25:01 -07002170 return status;
2171}
2172EXPORT_SYMBOL(block_write_begin);
2173
2174int block_write_end(struct file *file, struct address_space *mapping,
2175 loff_t pos, unsigned len, unsigned copied,
2176 struct page *page, void *fsdata)
2177{
2178 struct inode *inode = mapping->host;
2179 unsigned start;
2180
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002181 start = pos & (PAGE_SIZE - 1);
Nick Pigginafddba42007-10-16 01:25:01 -07002182
2183 if (unlikely(copied < len)) {
2184 /*
2185 * The buffers that were written will now be uptodate, so we
2186 * don't have to worry about a readpage reading them and
2187 * overwriting a partial write. However if we have encountered
2188 * a short write and only partially written into a buffer, it
2189 * will not be marked uptodate, so a readpage might come in and
2190 * destroy our partial write.
2191 *
2192 * Do the simplest thing, and just treat any short write to a
2193 * non uptodate page as a zero-length write, and force the
2194 * caller to redo the whole thing.
2195 */
2196 if (!PageUptodate(page))
2197 copied = 0;
2198
2199 page_zero_new_buffers(page, start+copied, start+len);
2200 }
2201 flush_dcache_page(page);
2202
2203 /* This could be a short (even 0-length) commit */
2204 __block_commit_write(inode, page, start, start+copied);
2205
2206 return copied;
2207}
2208EXPORT_SYMBOL(block_write_end);
2209
2210int generic_write_end(struct file *file, struct address_space *mapping,
2211 loff_t pos, unsigned len, unsigned copied,
2212 struct page *page, void *fsdata)
2213{
Christoph Hellwig8af54f22019-06-27 17:28:40 -07002214 struct inode *inode = mapping->host;
2215 loff_t old_size = inode->i_size;
2216 bool i_size_changed = false;
2217
Nick Pigginafddba42007-10-16 01:25:01 -07002218 copied = block_write_end(file, mapping, pos, len, copied, page, fsdata);
Christoph Hellwig8af54f22019-06-27 17:28:40 -07002219
2220 /*
2221 * No need to use i_size_read() here, the i_size cannot change under us
2222 * because we hold i_rwsem.
2223 *
2224 * But it's important to update i_size while still holding page lock:
2225 * page writeout could otherwise come in and zero beyond i_size.
2226 */
2227 if (pos + copied > inode->i_size) {
2228 i_size_write(inode, pos + copied);
2229 i_size_changed = true;
2230 }
2231
2232 unlock_page(page);
Andreas Gruenbacher7a77dad2019-04-30 08:45:34 -07002233 put_page(page);
Christoph Hellwig8af54f22019-06-27 17:28:40 -07002234
2235 if (old_size < pos)
2236 pagecache_isize_extended(inode, old_size, pos);
2237 /*
2238 * Don't mark the inode dirty under page lock. First, it unnecessarily
2239 * makes the holding time of page lock longer. Second, it forces lock
2240 * ordering of page lock and transaction start for journaling
2241 * filesystems.
2242 */
2243 if (i_size_changed)
2244 mark_inode_dirty(inode);
Andreas Gruenbacher26ddb1f2019-04-30 08:45:33 -07002245 return copied;
Nick Pigginafddba42007-10-16 01:25:01 -07002246}
2247EXPORT_SYMBOL(generic_write_end);
2248
2249/*
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002250 * block_is_partially_uptodate checks whether buffers within a page are
2251 * uptodate or not.
2252 *
2253 * Returns true if all buffers which correspond to a file portion
2254 * we want to read are uptodate.
2255 */
Al Viroc186afb42014-02-02 21:16:54 -05002256int block_is_partially_uptodate(struct page *page, unsigned long from,
2257 unsigned long count)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002258{
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002259 unsigned block_start, block_end, blocksize;
2260 unsigned to;
2261 struct buffer_head *bh, *head;
2262 int ret = 1;
2263
2264 if (!page_has_buffers(page))
2265 return 0;
2266
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002267 head = page_buffers(page);
2268 blocksize = head->b_size;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002269 to = min_t(unsigned, PAGE_SIZE - from, count);
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002270 to = from + to;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002271 if (from < blocksize && to > PAGE_SIZE - blocksize)
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002272 return 0;
2273
Hisashi Hifumi8ab22b92008-07-28 15:46:36 -07002274 bh = head;
2275 block_start = 0;
2276 do {
2277 block_end = block_start + blocksize;
2278 if (block_end > from && block_start < to) {
2279 if (!buffer_uptodate(bh)) {
2280 ret = 0;
2281 break;
2282 }
2283 if (block_end >= to)
2284 break;
2285 }
2286 block_start = block_end;
2287 bh = bh->b_this_page;
2288 } while (bh != head);
2289
2290 return ret;
2291}
2292EXPORT_SYMBOL(block_is_partially_uptodate);
2293
2294/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07002295 * Generic "read page" function for block devices that have the normal
2296 * get_block functionality. This is most of the block device filesystems.
2297 * Reads the page asynchronously --- the unlock_buffer() and
2298 * set/clear_buffer_uptodate() functions propagate buffer state into the
2299 * page struct once IO has completed.
2300 */
2301int block_read_full_page(struct page *page, get_block_t *get_block)
2302{
2303 struct inode *inode = page->mapping->host;
2304 sector_t iblock, lblock;
2305 struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE];
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002306 unsigned int blocksize, bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002307 int nr, i;
2308 int fully_mapped = 1;
Eric Biggers0957cd62022-12-23 12:36:37 -08002309 loff_t limit = i_size_read(inode);
2310
2311 /* This is needed for ext4. */
2312 if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode))
2313 limit = inode->i_sb->s_maxbytes;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002314
Linus Torvalds45bce8f2012-11-29 10:21:43 -08002315 head = create_page_buffers(page, inode, 0);
2316 blocksize = head->b_size;
2317 bbits = block_size_bits(blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002318
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002319 iblock = (sector_t)page->index << (PAGE_SHIFT - bbits);
Eric Biggers0957cd62022-12-23 12:36:37 -08002320 lblock = (limit+blocksize-1) >> bbits;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 bh = head;
2322 nr = 0;
2323 i = 0;
2324
2325 do {
2326 if (buffer_uptodate(bh))
2327 continue;
2328
2329 if (!buffer_mapped(bh)) {
Andrew Mortonc64610b2005-05-16 21:53:49 -07002330 int err = 0;
2331
Linus Torvalds1da177e2005-04-16 15:20:36 -07002332 fully_mapped = 0;
2333 if (iblock < lblock) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002334 WARN_ON(bh->b_size != blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002335 err = get_block(inode, iblock, bh, 0);
2336 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337 SetPageError(page);
2338 }
2339 if (!buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002340 zero_user(page, i * blocksize, blocksize);
Andrew Mortonc64610b2005-05-16 21:53:49 -07002341 if (!err)
2342 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002343 continue;
2344 }
2345 /*
2346 * get_block() might have updated the buffer
2347 * synchronously
2348 */
2349 if (buffer_uptodate(bh))
2350 continue;
2351 }
2352 arr[nr++] = bh;
2353 } while (i++, iblock++, (bh = bh->b_this_page) != head);
2354
2355 if (fully_mapped)
2356 SetPageMappedToDisk(page);
2357
2358 if (!nr) {
2359 /*
2360 * All buffers are uptodate - we can set the page uptodate
2361 * as well. But not if get_block() returned an error.
2362 */
2363 if (!PageError(page))
2364 SetPageUptodate(page);
2365 unlock_page(page);
2366 return 0;
2367 }
2368
2369 /* Stage two: lock the buffers */
2370 for (i = 0; i < nr; i++) {
2371 bh = arr[i];
2372 lock_buffer(bh);
2373 mark_buffer_async_read(bh);
2374 }
2375
2376 /*
2377 * Stage 3: start the IO. Check for uptodateness
2378 * inside the buffer lock in case another process reading
2379 * the underlying blockdev brought it uptodate (the sct fix).
2380 */
2381 for (i = 0; i < nr; i++) {
2382 bh = arr[i];
2383 if (buffer_uptodate(bh))
2384 end_buffer_async_read(bh, 1);
2385 else
Mike Christie2a222ca2016-06-05 14:31:43 -05002386 submit_bh(REQ_OP_READ, 0, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002387 }
2388 return 0;
2389}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002390EXPORT_SYMBOL(block_read_full_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391
2392/* utility function for filesystems that need to do work on expanding
Nick Piggin89e10782007-10-16 01:25:07 -07002393 * truncates. Uses filesystem pagecache writes to allow the filesystem to
Linus Torvalds1da177e2005-04-16 15:20:36 -07002394 * deal with the hole.
2395 */
Nick Piggin89e10782007-10-16 01:25:07 -07002396int generic_cont_expand_simple(struct inode *inode, loff_t size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397{
2398 struct address_space *mapping = inode->i_mapping;
2399 struct page *page;
Nick Piggin89e10782007-10-16 01:25:07 -07002400 void *fsdata;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002401 int err;
2402
npiggin@suse.dec08d3b02009-08-21 02:35:06 +10002403 err = inode_newsize_ok(inode, size);
2404 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002405 goto out;
2406
Nick Piggin89e10782007-10-16 01:25:07 -07002407 err = pagecache_write_begin(NULL, mapping, size, 0,
Tetsuo Handac718a972017-05-08 15:58:59 -07002408 AOP_FLAG_CONT_EXPAND, &page, &fsdata);
Nick Piggin89e10782007-10-16 01:25:07 -07002409 if (err)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002410 goto out;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002411
Nick Piggin89e10782007-10-16 01:25:07 -07002412 err = pagecache_write_end(NULL, mapping, size, 0, 0, page, fsdata);
2413 BUG_ON(err > 0);
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002414
Linus Torvalds1da177e2005-04-16 15:20:36 -07002415out:
2416 return err;
2417}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002418EXPORT_SYMBOL(generic_cont_expand_simple);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419
Adrian Bunkf1e3af72008-04-29 00:59:01 -07002420static int cont_expand_zero(struct file *file, struct address_space *mapping,
2421 loff_t pos, loff_t *bytes)
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002422{
Nick Piggin89e10782007-10-16 01:25:07 -07002423 struct inode *inode = mapping->host;
Fabian Frederick93407472017-02-27 14:28:32 -08002424 unsigned int blocksize = i_blocksize(inode);
Nick Piggin89e10782007-10-16 01:25:07 -07002425 struct page *page;
2426 void *fsdata;
2427 pgoff_t index, curidx;
2428 loff_t curpos;
2429 unsigned zerofrom, offset, len;
2430 int err = 0;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002431
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002432 index = pos >> PAGE_SHIFT;
2433 offset = pos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002434
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002435 while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) {
2436 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002437 if (zerofrom & (blocksize-1)) {
2438 *bytes |= (blocksize-1);
2439 (*bytes)++;
2440 }
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002441 len = PAGE_SIZE - zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002442
Tetsuo Handac718a972017-05-08 15:58:59 -07002443 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2444 &page, &fsdata);
Nick Piggin89e10782007-10-16 01:25:07 -07002445 if (err)
2446 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002447 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002448 err = pagecache_write_end(file, mapping, curpos, len, len,
2449 page, fsdata);
2450 if (err < 0)
2451 goto out;
2452 BUG_ON(err != len);
2453 err = 0;
OGAWA Hirofumi061e9742008-04-28 02:16:28 -07002454
2455 balance_dirty_pages_ratelimited(mapping);
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002456
Davidlohr Bueso08d405c2019-01-03 15:28:58 -08002457 if (fatal_signal_pending(current)) {
Mikulas Patockac2ca0fc2014-07-27 13:00:41 -04002458 err = -EINTR;
2459 goto out;
2460 }
Nick Piggin89e10782007-10-16 01:25:07 -07002461 }
2462
2463 /* page covers the boundary, find the boundary offset */
2464 if (index == curidx) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002465 zerofrom = curpos & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002466 /* if we will expand the thing last block will be filled */
2467 if (offset <= zerofrom) {
2468 goto out;
2469 }
2470 if (zerofrom & (blocksize-1)) {
2471 *bytes |= (blocksize-1);
2472 (*bytes)++;
2473 }
2474 len = offset - zerofrom;
2475
Tetsuo Handac718a972017-05-08 15:58:59 -07002476 err = pagecache_write_begin(file, mapping, curpos, len, 0,
2477 &page, &fsdata);
Nick Piggin89e10782007-10-16 01:25:07 -07002478 if (err)
2479 goto out;
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002480 zero_user(page, zerofrom, len);
Nick Piggin89e10782007-10-16 01:25:07 -07002481 err = pagecache_write_end(file, mapping, curpos, len, len,
2482 page, fsdata);
2483 if (err < 0)
2484 goto out;
2485 BUG_ON(err != len);
2486 err = 0;
2487 }
2488out:
2489 return err;
OGAWA Hirofumi05eb0b52006-01-08 01:02:13 -08002490}
2491
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492/*
2493 * For moronic filesystems that do not allow holes in file.
2494 * We may have to extend the file.
2495 */
Christoph Hellwig282dc172010-06-04 11:29:55 +02002496int cont_write_begin(struct file *file, struct address_space *mapping,
Nick Piggin89e10782007-10-16 01:25:07 -07002497 loff_t pos, unsigned len, unsigned flags,
2498 struct page **pagep, void **fsdata,
2499 get_block_t *get_block, loff_t *bytes)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002500{
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 struct inode *inode = mapping->host;
Fabian Frederick93407472017-02-27 14:28:32 -08002502 unsigned int blocksize = i_blocksize(inode);
2503 unsigned int zerofrom;
Nick Piggin89e10782007-10-16 01:25:07 -07002504 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002505
Nick Piggin89e10782007-10-16 01:25:07 -07002506 err = cont_expand_zero(file, mapping, pos, bytes);
2507 if (err)
Christoph Hellwig155130a2010-06-04 11:29:58 +02002508 return err;
Nick Piggin89e10782007-10-16 01:25:07 -07002509
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002510 zerofrom = *bytes & ~PAGE_MASK;
Nick Piggin89e10782007-10-16 01:25:07 -07002511 if (pos+len > *bytes && zerofrom & (blocksize-1)) {
2512 *bytes |= (blocksize-1);
2513 (*bytes)++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 }
2515
Christoph Hellwig155130a2010-06-04 11:29:58 +02002516 return block_write_begin(mapping, pos, len, flags, pagep, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002518EXPORT_SYMBOL(cont_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002519
Linus Torvalds1da177e2005-04-16 15:20:36 -07002520int block_commit_write(struct page *page, unsigned from, unsigned to)
2521{
2522 struct inode *inode = page->mapping->host;
2523 __block_commit_write(inode,page,from,to);
2524 return 0;
2525}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002526EXPORT_SYMBOL(block_commit_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002527
David Chinner54171692007-07-19 17:39:55 +10002528/*
2529 * block_page_mkwrite() is not allowed to change the file size as it gets
2530 * called from a page fault handler when a page is first dirtied. Hence we must
2531 * be careful to check for EOF conditions here. We set the page up correctly
2532 * for a written page which means we get ENOSPC checking when writing into
2533 * holes and correct delalloc and unwritten extent mapping on filesystems that
2534 * support these features.
2535 *
2536 * We are not allowed to take the i_mutex here so we have to play games to
2537 * protect against truncate races as the page could now be beyond EOF. Because
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002538 * truncate writes the inode size before removing pages, once we have the
David Chinner54171692007-07-19 17:39:55 +10002539 * page lock we can determine safely if the page is beyond EOF. If it is not
2540 * beyond EOF, then the page is guaranteed safe against truncation until we
2541 * unlock the page.
Jan Karaea13a862011-05-24 00:23:35 +02002542 *
Jan Kara14da9202012-06-12 16:20:37 +02002543 * Direct callers of this function should protect against filesystem freezing
Ross Zwisler5c500022015-10-13 16:51:02 -06002544 * using sb_start_pagefault() - sb_end_pagefault() functions.
David Chinner54171692007-07-19 17:39:55 +10002545 */
Ross Zwisler5c500022015-10-13 16:51:02 -06002546int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf,
Jan Kara24da4fa2011-05-24 00:23:34 +02002547 get_block_t get_block)
David Chinner54171692007-07-19 17:39:55 +10002548{
Nick Pigginc2ec1752009-03-31 15:23:21 -07002549 struct page *page = vmf->page;
Al Viro496ad9a2013-01-23 17:07:38 -05002550 struct inode *inode = file_inode(vma->vm_file);
David Chinner54171692007-07-19 17:39:55 +10002551 unsigned long end;
2552 loff_t size;
Jan Kara24da4fa2011-05-24 00:23:34 +02002553 int ret;
David Chinner54171692007-07-19 17:39:55 +10002554
2555 lock_page(page);
2556 size = i_size_read(inode);
2557 if ((page->mapping != inode->i_mapping) ||
Nick Piggin18336332007-07-20 00:31:45 -07002558 (page_offset(page) > size)) {
Jan Kara24da4fa2011-05-24 00:23:34 +02002559 /* We overload EFAULT to mean page got truncated */
2560 ret = -EFAULT;
2561 goto out_unlock;
David Chinner54171692007-07-19 17:39:55 +10002562 }
2563
2564 /* page is wholly or partially inside EOF */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002565 if (((page->index + 1) << PAGE_SHIFT) > size)
2566 end = size & ~PAGE_MASK;
David Chinner54171692007-07-19 17:39:55 +10002567 else
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002568 end = PAGE_SIZE;
David Chinner54171692007-07-19 17:39:55 +10002569
Christoph Hellwigebdec242010-10-06 10:47:23 +02002570 ret = __block_write_begin(page, 0, end, get_block);
David Chinner54171692007-07-19 17:39:55 +10002571 if (!ret)
2572 ret = block_commit_write(page, 0, end);
2573
Jan Kara24da4fa2011-05-24 00:23:34 +02002574 if (unlikely(ret < 0))
2575 goto out_unlock;
Jan Karaea13a862011-05-24 00:23:35 +02002576 set_page_dirty(page);
Darrick J. Wong1d1d1a72013-02-21 16:42:51 -08002577 wait_for_stable_page(page);
Jan Kara24da4fa2011-05-24 00:23:34 +02002578 return 0;
2579out_unlock:
2580 unlock_page(page);
David Chinner54171692007-07-19 17:39:55 +10002581 return ret;
2582}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002583EXPORT_SYMBOL(block_page_mkwrite);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002584
2585/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002586 * nobh_write_begin()'s prereads are special: the buffer_heads are freed
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587 * immediately, while under the page lock. So it needs a special end_io
2588 * handler which does not touch the bh after unlocking it.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 */
2590static void end_buffer_read_nobh(struct buffer_head *bh, int uptodate)
2591{
Dmitry Monakhov68671f32007-10-16 01:24:47 -07002592 __end_buffer_read_notouch(bh, uptodate);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002593}
2594
2595/*
Nick Piggin03158cd2007-10-16 01:25:25 -07002596 * Attach the singly-linked list of buffers created by nobh_write_begin, to
2597 * the page (converting it to circular linked list and taking care of page
2598 * dirty races).
2599 */
2600static void attach_nobh_buffers(struct page *page, struct buffer_head *head)
2601{
2602 struct buffer_head *bh;
2603
2604 BUG_ON(!PageLocked(page));
2605
2606 spin_lock(&page->mapping->private_lock);
2607 bh = head;
2608 do {
2609 if (PageDirty(page))
2610 set_buffer_dirty(bh);
2611 if (!bh->b_this_page)
2612 bh->b_this_page = head;
2613 bh = bh->b_this_page;
2614 } while (bh != head);
Guoqing Jiang45dcfc22020-06-01 21:47:48 -07002615 attach_page_private(page, head);
Nick Piggin03158cd2007-10-16 01:25:25 -07002616 spin_unlock(&page->mapping->private_lock);
2617}
2618
2619/*
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002620 * On entry, the page is fully not uptodate.
2621 * On exit the page is fully uptodate in the areas outside (from,to)
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002622 * The filesystem needs to handle block truncation upon failure.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 */
Christoph Hellwigea0f04e2010-06-04 11:29:54 +02002624int nobh_write_begin(struct address_space *mapping,
Nick Piggin03158cd2007-10-16 01:25:25 -07002625 loff_t pos, unsigned len, unsigned flags,
2626 struct page **pagep, void **fsdata,
Linus Torvalds1da177e2005-04-16 15:20:36 -07002627 get_block_t *get_block)
2628{
Nick Piggin03158cd2007-10-16 01:25:25 -07002629 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 const unsigned blkbits = inode->i_blkbits;
2631 const unsigned blocksize = 1 << blkbits;
Nick Piggina4b06722007-10-16 01:24:48 -07002632 struct buffer_head *head, *bh;
Nick Piggin03158cd2007-10-16 01:25:25 -07002633 struct page *page;
2634 pgoff_t index;
2635 unsigned from, to;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002636 unsigned block_in_page;
Nick Piggina4b06722007-10-16 01:24:48 -07002637 unsigned block_start, block_end;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638 sector_t block_in_file;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 int nr_reads = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640 int ret = 0;
2641 int is_mapped_to_disk = 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002642
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002643 index = pos >> PAGE_SHIFT;
2644 from = pos & (PAGE_SIZE - 1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002645 to = from + len;
2646
Nick Piggin54566b22009-01-04 12:00:53 -08002647 page = grab_cache_page_write_begin(mapping, index, flags);
Nick Piggin03158cd2007-10-16 01:25:25 -07002648 if (!page)
2649 return -ENOMEM;
2650 *pagep = page;
2651 *fsdata = NULL;
2652
2653 if (page_has_buffers(page)) {
Namhyung Kim309f77a2010-10-25 15:01:12 +09002654 ret = __block_write_begin(page, pos, len, get_block);
2655 if (unlikely(ret))
2656 goto out_release;
2657 return ret;
Nick Piggin03158cd2007-10-16 01:25:25 -07002658 }
Nick Piggina4b06722007-10-16 01:24:48 -07002659
Linus Torvalds1da177e2005-04-16 15:20:36 -07002660 if (PageMappedToDisk(page))
2661 return 0;
2662
Nick Piggina4b06722007-10-16 01:24:48 -07002663 /*
2664 * Allocate buffers so that we can keep track of state, and potentially
2665 * attach them to the page if an error occurs. In the common case of
2666 * no error, they will just be freed again without ever being attached
2667 * to the page (which is all OK, because we're under the page lock).
2668 *
2669 * Be careful: the buffer linked list is a NULL terminated one, rather
2670 * than the circular one we're used to.
2671 */
Jens Axboe640ab982017-09-27 05:40:16 -06002672 head = alloc_page_buffers(page, blocksize, false);
Nick Piggin03158cd2007-10-16 01:25:25 -07002673 if (!head) {
2674 ret = -ENOMEM;
2675 goto out_release;
2676 }
Nick Piggina4b06722007-10-16 01:24:48 -07002677
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002678 block_in_file = (sector_t)page->index << (PAGE_SHIFT - blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679
2680 /*
2681 * We loop across all blocks in the page, whether or not they are
2682 * part of the affected region. This is so we can discover if the
2683 * page is fully mapped-to-disk.
2684 */
Nick Piggina4b06722007-10-16 01:24:48 -07002685 for (block_start = 0, block_in_page = 0, bh = head;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002686 block_start < PAGE_SIZE;
Nick Piggina4b06722007-10-16 01:24:48 -07002687 block_in_page++, block_start += blocksize, bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002688 int create;
2689
Nick Piggina4b06722007-10-16 01:24:48 -07002690 block_end = block_start + blocksize;
2691 bh->b_state = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002692 create = 1;
2693 if (block_start >= to)
2694 create = 0;
2695 ret = get_block(inode, block_in_file + block_in_page,
Nick Piggina4b06722007-10-16 01:24:48 -07002696 bh, create);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002697 if (ret)
2698 goto failed;
Nick Piggina4b06722007-10-16 01:24:48 -07002699 if (!buffer_mapped(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 is_mapped_to_disk = 0;
Nick Piggina4b06722007-10-16 01:24:48 -07002701 if (buffer_new(bh))
Jan Karae64855c2016-11-04 18:08:15 +01002702 clean_bdev_bh_alias(bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002703 if (PageUptodate(page)) {
2704 set_buffer_uptodate(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002705 continue;
Nick Piggina4b06722007-10-16 01:24:48 -07002706 }
2707 if (buffer_new(bh) || !buffer_mapped(bh)) {
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002708 zero_user_segments(page, block_start, from,
2709 to, block_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710 continue;
2711 }
Nick Piggina4b06722007-10-16 01:24:48 -07002712 if (buffer_uptodate(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 continue; /* reiserfs does this */
2714 if (block_start < from || block_end > to) {
Nick Piggina4b06722007-10-16 01:24:48 -07002715 lock_buffer(bh);
2716 bh->b_end_io = end_buffer_read_nobh;
Mike Christie2a222ca2016-06-05 14:31:43 -05002717 submit_bh(REQ_OP_READ, 0, bh);
Nick Piggina4b06722007-10-16 01:24:48 -07002718 nr_reads++;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719 }
2720 }
2721
2722 if (nr_reads) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002723 /*
2724 * The page is locked, so these buffers are protected from
2725 * any VM or truncate activity. Hence we don't need to care
2726 * for the buffer_head refcounts.
2727 */
Nick Piggina4b06722007-10-16 01:24:48 -07002728 for (bh = head; bh; bh = bh->b_this_page) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 wait_on_buffer(bh);
2730 if (!buffer_uptodate(bh))
2731 ret = -EIO;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 }
2733 if (ret)
2734 goto failed;
2735 }
2736
2737 if (is_mapped_to_disk)
2738 SetPageMappedToDisk(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002739
Nick Piggin03158cd2007-10-16 01:25:25 -07002740 *fsdata = head; /* to be released by nobh_write_end */
Nick Piggina4b06722007-10-16 01:24:48 -07002741
Linus Torvalds1da177e2005-04-16 15:20:36 -07002742 return 0;
2743
2744failed:
Nick Piggin03158cd2007-10-16 01:25:25 -07002745 BUG_ON(!ret);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002746 /*
Nick Piggina4b06722007-10-16 01:24:48 -07002747 * Error recovery is a bit difficult. We need to zero out blocks that
2748 * were newly allocated, and dirty them to ensure they get written out.
2749 * Buffers need to be attached to the page at this point, otherwise
2750 * the handling of potential IO errors during writeout would be hard
2751 * (could try doing synchronous writeout, but what if that fails too?)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752 */
Nick Piggin03158cd2007-10-16 01:25:25 -07002753 attach_nobh_buffers(page, head);
2754 page_zero_new_buffers(page, from, to);
Nick Piggina4b06722007-10-16 01:24:48 -07002755
Nick Piggin03158cd2007-10-16 01:25:25 -07002756out_release:
2757 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002758 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002759 *pagep = NULL;
Nick Piggina4b06722007-10-16 01:24:48 -07002760
npiggin@suse.de7bb46a62010-05-27 01:05:33 +10002761 return ret;
2762}
Nick Piggin03158cd2007-10-16 01:25:25 -07002763EXPORT_SYMBOL(nobh_write_begin);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002764
Nick Piggin03158cd2007-10-16 01:25:25 -07002765int nobh_write_end(struct file *file, struct address_space *mapping,
2766 loff_t pos, unsigned len, unsigned copied,
2767 struct page *page, void *fsdata)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002768{
2769 struct inode *inode = page->mapping->host;
Nick Pigginefdc3132007-10-21 06:57:41 +02002770 struct buffer_head *head = fsdata;
Nick Piggin03158cd2007-10-16 01:25:25 -07002771 struct buffer_head *bh;
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002772 BUG_ON(fsdata != NULL && page_has_buffers(page));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Dave Kleikampd4cf1092009-02-06 14:59:26 -06002774 if (unlikely(copied < len) && head)
Dmitri Monakhov5b41e742008-03-28 14:15:52 -07002775 attach_nobh_buffers(page, head);
2776 if (page_has_buffers(page))
2777 return generic_write_end(file, mapping, pos, len,
2778 copied, page, fsdata);
Nick Piggina4b06722007-10-16 01:24:48 -07002779
Nick Piggin22c8ca72007-02-20 13:58:09 -08002780 SetPageUptodate(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781 set_page_dirty(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002782 if (pos+copied > inode->i_size) {
2783 i_size_write(inode, pos+copied);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784 mark_inode_dirty(inode);
2785 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002786
2787 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002788 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002789
Nick Piggin03158cd2007-10-16 01:25:25 -07002790 while (head) {
2791 bh = head;
2792 head = head->b_this_page;
2793 free_buffer_head(bh);
2794 }
2795
2796 return copied;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797}
Nick Piggin03158cd2007-10-16 01:25:25 -07002798EXPORT_SYMBOL(nobh_write_end);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799
2800/*
2801 * nobh_writepage() - based on block_full_write_page() except
2802 * that it tries to operate without attaching bufferheads to
2803 * the page.
2804 */
2805int nobh_writepage(struct page *page, get_block_t *get_block,
2806 struct writeback_control *wbc)
2807{
2808 struct inode * const inode = page->mapping->host;
2809 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002810 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002811 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002812 int ret;
2813
2814 /* Is the page fully inside i_size? */
2815 if (page->index < end_index)
2816 goto out;
2817
2818 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002819 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002820 if (page->index >= end_index+1 || !offset) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002821 unlock_page(page);
2822 return 0; /* don't care */
2823 }
2824
2825 /*
2826 * The page straddles i_size. It must be zeroed out on each and every
2827 * writepage invocation because it may be mmapped. "A file is mapped
2828 * in multiples of the page size. For a file that is not a multiple of
2829 * the page size, the remaining memory is zeroed when mapped, and
2830 * writes to that region are not written out to the file."
2831 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002832 zero_user_segment(page, offset, PAGE_SIZE);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833out:
2834 ret = mpage_writepage(page, get_block, wbc);
2835 if (ret == -EAGAIN)
Chris Mason35c80d52009-04-15 13:22:38 -04002836 ret = __block_write_full_page(inode, page, get_block, wbc,
2837 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002838 return ret;
2839}
2840EXPORT_SYMBOL(nobh_writepage);
2841
Nick Piggin03158cd2007-10-16 01:25:25 -07002842int nobh_truncate_page(struct address_space *mapping,
2843 loff_t from, get_block_t *get_block)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002844{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002845 pgoff_t index = from >> PAGE_SHIFT;
2846 unsigned offset = from & (PAGE_SIZE-1);
Nick Piggin03158cd2007-10-16 01:25:25 -07002847 unsigned blocksize;
2848 sector_t iblock;
2849 unsigned length, pos;
2850 struct inode *inode = mapping->host;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 struct page *page;
Nick Piggin03158cd2007-10-16 01:25:25 -07002852 struct buffer_head map_bh;
2853 int err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002854
Fabian Frederick93407472017-02-27 14:28:32 -08002855 blocksize = i_blocksize(inode);
Nick Piggin03158cd2007-10-16 01:25:25 -07002856 length = offset & (blocksize - 1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002857
Nick Piggin03158cd2007-10-16 01:25:25 -07002858 /* Block boundary? Nothing to do */
2859 if (!length)
2860 return 0;
2861
2862 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002863 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Nick Piggin03158cd2007-10-16 01:25:25 -07002864
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865 page = grab_cache_page(mapping, index);
Nick Piggin03158cd2007-10-16 01:25:25 -07002866 err = -ENOMEM;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002867 if (!page)
2868 goto out;
2869
Nick Piggin03158cd2007-10-16 01:25:25 -07002870 if (page_has_buffers(page)) {
2871has_buffers:
2872 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002873 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002874 return block_truncate_page(mapping, from, get_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002875 }
Nick Piggin03158cd2007-10-16 01:25:25 -07002876
2877 /* Find the buffer that contains "offset" */
2878 pos = blocksize;
2879 while (offset >= pos) {
2880 iblock++;
2881 pos += blocksize;
2882 }
2883
Theodore Ts'o460bcf52009-05-12 07:37:56 -04002884 map_bh.b_size = blocksize;
2885 map_bh.b_state = 0;
Nick Piggin03158cd2007-10-16 01:25:25 -07002886 err = get_block(inode, iblock, &map_bh, 0);
2887 if (err)
2888 goto unlock;
2889 /* unmapped? It's a hole - nothing to do */
2890 if (!buffer_mapped(&map_bh))
2891 goto unlock;
2892
2893 /* Ok, it's mapped. Make sure it's up-to-date */
2894 if (!PageUptodate(page)) {
2895 err = mapping->a_ops->readpage(NULL, page);
2896 if (err) {
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002897 put_page(page);
Nick Piggin03158cd2007-10-16 01:25:25 -07002898 goto out;
2899 }
2900 lock_page(page);
2901 if (!PageUptodate(page)) {
2902 err = -EIO;
2903 goto unlock;
2904 }
2905 if (page_has_buffers(page))
2906 goto has_buffers;
2907 }
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002908 zero_user(page, offset, length);
Nick Piggin03158cd2007-10-16 01:25:25 -07002909 set_page_dirty(page);
2910 err = 0;
2911
2912unlock:
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002914 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915out:
Nick Piggin03158cd2007-10-16 01:25:25 -07002916 return err;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917}
2918EXPORT_SYMBOL(nobh_truncate_page);
2919
2920int block_truncate_page(struct address_space *mapping,
2921 loff_t from, get_block_t *get_block)
2922{
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002923 pgoff_t index = from >> PAGE_SHIFT;
2924 unsigned offset = from & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002925 unsigned blocksize;
Andrew Morton54b21a72006-01-08 01:03:05 -08002926 sector_t iblock;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002927 unsigned length, pos;
2928 struct inode *inode = mapping->host;
2929 struct page *page;
2930 struct buffer_head *bh;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 int err;
2932
Fabian Frederick93407472017-02-27 14:28:32 -08002933 blocksize = i_blocksize(inode);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002934 length = offset & (blocksize - 1);
2935
2936 /* Block boundary? Nothing to do */
2937 if (!length)
2938 return 0;
2939
2940 length = blocksize - length;
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002941 iblock = (sector_t)index << (PAGE_SHIFT - inode->i_blkbits);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002942
2943 page = grab_cache_page(mapping, index);
2944 err = -ENOMEM;
2945 if (!page)
2946 goto out;
2947
2948 if (!page_has_buffers(page))
2949 create_empty_buffers(page, blocksize, 0);
2950
2951 /* Find the buffer that contains "offset" */
2952 bh = page_buffers(page);
2953 pos = blocksize;
2954 while (offset >= pos) {
2955 bh = bh->b_this_page;
2956 iblock++;
2957 pos += blocksize;
2958 }
2959
2960 err = 0;
2961 if (!buffer_mapped(bh)) {
Badari Pulavartyb0cf2322006-03-26 01:38:00 -08002962 WARN_ON(bh->b_size != blocksize);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002963 err = get_block(inode, iblock, bh, 0);
2964 if (err)
2965 goto unlock;
2966 /* unmapped? It's a hole - nothing to do */
2967 if (!buffer_mapped(bh))
2968 goto unlock;
2969 }
2970
2971 /* Ok, it's mapped. Make sure it's up-to-date */
2972 if (PageUptodate(page))
2973 set_buffer_uptodate(bh);
2974
David Chinner33a266d2007-02-12 00:51:41 -08002975 if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002976 err = -EIO;
Mike Christiedfec8a12016-06-05 14:31:44 -05002977 ll_rw_block(REQ_OP_READ, 0, 1, &bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 wait_on_buffer(bh);
2979 /* Uhhuh. Read error. Complain and punt. */
2980 if (!buffer_uptodate(bh))
2981 goto unlock;
2982 }
2983
Christoph Lametereebd2aa2008-02-04 22:28:29 -08002984 zero_user(page, offset, length);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002985 mark_buffer_dirty(bh);
2986 err = 0;
2987
2988unlock:
2989 unlock_page(page);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03002990 put_page(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002991out:
2992 return err;
2993}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07002994EXPORT_SYMBOL(block_truncate_page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002995
2996/*
2997 * The generic ->writepage function for buffer-backed address_spaces
2998 */
Matthew Wilcox1b938c02014-06-04 16:07:43 -07002999int block_write_full_page(struct page *page, get_block_t *get_block,
3000 struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001{
3002 struct inode * const inode = page->mapping->host;
3003 loff_t i_size = i_size_read(inode);
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003004 const pgoff_t end_index = i_size >> PAGE_SHIFT;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003005 unsigned offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006
3007 /* Is the page fully inside i_size? */
3008 if (page->index < end_index)
Chris Mason35c80d52009-04-15 13:22:38 -04003009 return __block_write_full_page(inode, page, get_block, wbc,
Matthew Wilcox1b938c02014-06-04 16:07:43 -07003010 end_buffer_async_write);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011
3012 /* Is the page fully outside i_size? (truncate in progress) */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003013 offset = i_size & (PAGE_SIZE-1);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 if (page->index >= end_index+1 || !offset) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015 unlock_page(page);
3016 return 0; /* don't care */
3017 }
3018
3019 /*
3020 * The page straddles i_size. It must be zeroed out on each and every
Adam Buchbinder2a61aa42009-12-11 16:35:40 -05003021 * writepage invocation because it may be mmapped. "A file is mapped
Linus Torvalds1da177e2005-04-16 15:20:36 -07003022 * in multiples of the page size. For a file that is not a multiple of
3023 * the page size, the remaining memory is zeroed when mapped, and
3024 * writes to that region are not written out to the file."
3025 */
Kirill A. Shutemov09cbfea2016-04-01 15:29:47 +03003026 zero_user_segment(page, offset, PAGE_SIZE);
Matthew Wilcox1b938c02014-06-04 16:07:43 -07003027 return __block_write_full_page(inode, page, get_block, wbc,
3028 end_buffer_async_write);
Chris Mason35c80d52009-04-15 13:22:38 -04003029}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003030EXPORT_SYMBOL(block_write_full_page);
Chris Mason35c80d52009-04-15 13:22:38 -04003031
Linus Torvalds1da177e2005-04-16 15:20:36 -07003032sector_t generic_block_bmap(struct address_space *mapping, sector_t block,
3033 get_block_t *get_block)
3034{
Linus Torvalds1da177e2005-04-16 15:20:36 -07003035 struct inode *inode = mapping->host;
Alexander Potapenko2a527d62017-07-05 00:56:21 -04003036 struct buffer_head tmp = {
3037 .b_size = i_blocksize(inode),
3038 };
3039
Linus Torvalds1da177e2005-04-16 15:20:36 -07003040 get_block(inode, block, &tmp, 0);
3041 return tmp.b_blocknr;
3042}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003043EXPORT_SYMBOL(generic_block_bmap);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003044
Christoph Hellwig4246a0b2015-07-20 15:29:37 +02003045static void end_bio_bh_io_sync(struct bio *bio)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003046{
3047 struct buffer_head *bh = bio->bi_private;
3048
Jens Axboeb7c44ed2015-07-24 12:37:59 -06003049 if (unlikely(bio_flagged(bio, BIO_QUIET)))
Keith Mannthey08bafc02008-11-25 10:24:35 +01003050 set_bit(BH_Quiet, &bh->b_state);
3051
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +02003052 bh->b_end_io(bh, !bio->bi_status);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003053 bio_put(bio);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054}
3055
Mike Christie2a222ca2016-06-05 14:31:43 -05003056static int submit_bh_wbc(int op, int op_flags, struct buffer_head *bh,
Jens Axboe8e8f9292017-06-27 09:30:05 -06003057 enum rw_hint write_hint, struct writeback_control *wbc)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003058{
3059 struct bio *bio;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003060
3061 BUG_ON(!buffer_locked(bh));
3062 BUG_ON(!buffer_mapped(bh));
3063 BUG_ON(!bh->b_end_io);
Aneesh Kumar K.V8fb0e342009-05-12 16:22:37 -04003064 BUG_ON(buffer_delay(bh));
3065 BUG_ON(buffer_unwritten(bh));
Linus Torvalds1da177e2005-04-16 15:20:36 -07003066
Jens Axboe48fd4f92008-08-22 10:00:36 +02003067 /*
Jens Axboe48fd4f92008-08-22 10:00:36 +02003068 * Only clear out a write error when rewriting
Linus Torvalds1da177e2005-04-16 15:20:36 -07003069 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003070 if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071 clear_buffer_write_io_error(bh);
3072
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073 bio = bio_alloc(GFP_NOIO, 1);
3074
Eric Biggers4f74d152020-07-02 01:56:07 +00003075 fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO);
3076
Kent Overstreet4f024f32013-10-11 15:44:27 -07003077 bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9);
Christoph Hellwig74d46992017-08-23 19:10:32 +02003078 bio_set_dev(bio, bh->b_bdev);
Jens Axboe8e8f9292017-06-27 09:30:05 -06003079 bio->bi_write_hint = write_hint;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080
Kent Overstreet6cf66b42014-12-22 12:48:42 +01003081 bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh));
3082 BUG_ON(bio->bi_iter.bi_size != bh->b_size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003083
3084 bio->bi_end_io = end_bio_bh_io_sync;
3085 bio->bi_private = bh;
3086
Theodore Ts'o877f9622013-04-20 19:58:37 -04003087 if (buffer_meta(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003088 op_flags |= REQ_META;
Theodore Ts'o877f9622013-04-20 19:58:37 -04003089 if (buffer_prio(bh))
Mike Christie2a222ca2016-06-05 14:31:43 -05003090 op_flags |= REQ_PRIO;
3091 bio_set_op_attrs(bio, op, op_flags);
Theodore Ts'o877f9622013-04-20 19:58:37 -04003092
Ming Lei83c9c542020-01-05 09:41:14 +08003093 /* Take care of bh's that straddle the end of the device */
3094 guard_bio_eod(bio);
3095
Dennis Zhoufd42df32018-12-05 12:10:34 -05003096 if (wbc) {
3097 wbc_init_bio(wbc, bio);
Tejun Heo34e51a52019-06-27 13:39:49 -07003098 wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size);
Dennis Zhoufd42df32018-12-05 12:10:34 -05003099 }
3100
Mike Christie4e49ea42016-06-05 14:31:41 -05003101 submit_bio(bio);
Julia Lawallf6454b02015-05-26 21:59:53 +02003102 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103}
Tejun Heobafc0db2015-06-02 08:37:23 -06003104
Eric Biggers020c28332017-03-25 21:02:18 -07003105int submit_bh(int op, int op_flags, struct buffer_head *bh)
Tejun Heobafc0db2015-06-02 08:37:23 -06003106{
Jens Axboe8e8f9292017-06-27 09:30:05 -06003107 return submit_bh_wbc(op, op_flags, bh, 0, NULL);
Darrick J. Wong7136851112013-04-29 15:07:25 -07003108}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003109EXPORT_SYMBOL(submit_bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003110
3111/**
3112 * ll_rw_block: low-level access to block devices (DEPRECATED)
Mike Christiedfec8a12016-06-05 14:31:44 -05003113 * @op: whether to %READ or %WRITE
Christoph Hellwigef295ec2016-10-28 08:48:16 -06003114 * @op_flags: req_flag_bits
Linus Torvalds1da177e2005-04-16 15:20:36 -07003115 * @nr: number of &struct buffer_heads in the array
3116 * @bhs: array of pointers to &struct buffer_head
3117 *
Jan Karaa7662232005-09-06 15:19:10 -07003118 * ll_rw_block() takes an array of pointers to &struct buffer_heads, and
Christoph Hellwig70246282016-07-19 11:28:41 +02003119 * requests an I/O operation on them, either a %REQ_OP_READ or a %REQ_OP_WRITE.
3120 * @op_flags contains flags modifying the detailed I/O behavior, most notably
3121 * %REQ_RAHEAD.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003122 *
3123 * This function drops any buffer that it cannot get a lock on (with the
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003124 * BH_Lock state bit), any buffer that appears to be clean when doing a write
3125 * request, and any buffer that appears to be up-to-date when doing read
3126 * request. Further it marks as clean buffers that are processed for
3127 * writing (the buffer cache won't assume that they are actually clean
3128 * until the buffer gets unlocked).
Linus Torvalds1da177e2005-04-16 15:20:36 -07003129 *
3130 * ll_rw_block sets b_end_io to simple completion handler that marks
Masanari Iidae2278672014-02-18 22:54:36 +09003131 * the buffer up-to-date (if appropriate), unlocks the buffer and wakes
Linus Torvalds1da177e2005-04-16 15:20:36 -07003132 * any waiters.
3133 *
3134 * All of the buffers must be for the same device, and must also be a
3135 * multiple of the current approved size for the device.
3136 */
Mike Christiedfec8a12016-06-05 14:31:44 -05003137void ll_rw_block(int op, int op_flags, int nr, struct buffer_head *bhs[])
Linus Torvalds1da177e2005-04-16 15:20:36 -07003138{
3139 int i;
3140
3141 for (i = 0; i < nr; i++) {
3142 struct buffer_head *bh = bhs[i];
3143
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003144 if (!trylock_buffer(bh))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003145 continue;
Mike Christiedfec8a12016-06-05 14:31:44 -05003146 if (op == WRITE) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003147 if (test_clear_buffer_dirty(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003148 bh->b_end_io = end_buffer_write_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003149 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003150 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003151 continue;
3152 }
3153 } else {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154 if (!buffer_uptodate(bh)) {
akpm@osdl.org76c30732005-04-16 15:24:07 -07003155 bh->b_end_io = end_buffer_read_sync;
OGAWA Hirofumie60e5c52006-02-03 03:04:43 -08003156 get_bh(bh);
Mike Christiedfec8a12016-06-05 14:31:44 -05003157 submit_bh(op, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158 continue;
3159 }
3160 }
3161 unlock_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003162 }
3163}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003164EXPORT_SYMBOL(ll_rw_block);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003165
Mike Christie2a222ca2016-06-05 14:31:43 -05003166void write_dirty_buffer(struct buffer_head *bh, int op_flags)
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003167{
3168 lock_buffer(bh);
3169 if (!test_clear_buffer_dirty(bh)) {
3170 unlock_buffer(bh);
3171 return;
3172 }
3173 bh->b_end_io = end_buffer_write_sync;
3174 get_bh(bh);
Mike Christie2a222ca2016-06-05 14:31:43 -05003175 submit_bh(REQ_OP_WRITE, op_flags, bh);
Christoph Hellwig9cb569d2010-08-11 17:06:24 +02003176}
3177EXPORT_SYMBOL(write_dirty_buffer);
3178
Linus Torvalds1da177e2005-04-16 15:20:36 -07003179/*
3180 * For a data-integrity writeout, we need to wait upon any in-progress I/O
3181 * and then start new I/O and then wait upon it. The caller must have a ref on
3182 * the buffer_head.
3183 */
Mike Christie2a222ca2016-06-05 14:31:43 -05003184int __sync_dirty_buffer(struct buffer_head *bh, int op_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003185{
3186 int ret = 0;
3187
3188 WARN_ON(atomic_read(&bh->b_count) < 1);
3189 lock_buffer(bh);
3190 if (test_clear_buffer_dirty(bh)) {
Xianting Tian377254b2020-07-31 12:10:25 -04003191 /*
3192 * The bh should be mapped, but it might not be if the
3193 * device was hot-removed. Not much we can do but fail the I/O.
3194 */
3195 if (!buffer_mapped(bh)) {
3196 unlock_buffer(bh);
3197 return -EIO;
3198 }
3199
Linus Torvalds1da177e2005-04-16 15:20:36 -07003200 get_bh(bh);
3201 bh->b_end_io = end_buffer_write_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003202 ret = submit_bh(REQ_OP_WRITE, op_flags, bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003203 wait_on_buffer(bh);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003204 if (!ret && !buffer_uptodate(bh))
3205 ret = -EIO;
3206 } else {
3207 unlock_buffer(bh);
3208 }
3209 return ret;
3210}
Christoph Hellwig87e99512010-08-11 17:05:45 +02003211EXPORT_SYMBOL(__sync_dirty_buffer);
3212
3213int sync_dirty_buffer(struct buffer_head *bh)
3214{
Christoph Hellwig70fd7612016-11-01 07:40:10 -06003215 return __sync_dirty_buffer(bh, REQ_SYNC);
Christoph Hellwig87e99512010-08-11 17:05:45 +02003216}
H Hartley Sweeten1fe72ea2009-09-22 16:43:51 -07003217EXPORT_SYMBOL(sync_dirty_buffer);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003218
3219/*
3220 * try_to_free_buffers() checks if all the buffers on this particular page
3221 * are unused, and releases them if so.
3222 *
3223 * Exclusion against try_to_free_buffers may be obtained by either
3224 * locking the page or by holding its mapping's private_lock.
3225 *
3226 * If the page is dirty but all the buffers are clean then we need to
3227 * be sure to mark the page clean as well. This is because the page
3228 * may be against a block device, and a later reattachment of buffers
3229 * to a dirty page will set *all* buffers dirty. Which would corrupt
3230 * filesystem data on the same device.
3231 *
3232 * The same applies to regular filesystem pages: if all the buffers are
3233 * clean then we set the page clean and proceed. To do that, we require
3234 * total exclusion from __set_page_dirty_buffers(). That is obtained with
3235 * private_lock.
3236 *
3237 * try_to_free_buffers() is non-blocking.
3238 */
3239static inline int buffer_busy(struct buffer_head *bh)
3240{
3241 return atomic_read(&bh->b_count) |
3242 (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock)));
3243}
3244
3245static int
3246drop_buffers(struct page *page, struct buffer_head **buffers_to_free)
3247{
3248 struct buffer_head *head = page_buffers(page);
3249 struct buffer_head *bh;
3250
3251 bh = head;
3252 do {
Linus Torvalds1da177e2005-04-16 15:20:36 -07003253 if (buffer_busy(bh))
3254 goto failed;
3255 bh = bh->b_this_page;
3256 } while (bh != head);
3257
3258 do {
3259 struct buffer_head *next = bh->b_this_page;
3260
Jan Kara535ee2f2008-02-08 04:21:59 -08003261 if (bh->b_assoc_map)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003262 __remove_assoc_queue(bh);
3263 bh = next;
3264 } while (bh != head);
3265 *buffers_to_free = head;
Guoqing Jiang45dcfc22020-06-01 21:47:48 -07003266 detach_page_private(page);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003267 return 1;
3268failed:
3269 return 0;
3270}
3271
3272int try_to_free_buffers(struct page *page)
3273{
3274 struct address_space * const mapping = page->mapping;
3275 struct buffer_head *buffers_to_free = NULL;
3276 int ret = 0;
3277
3278 BUG_ON(!PageLocked(page));
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003279 if (PageWriteback(page))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280 return 0;
3281
3282 if (mapping == NULL) { /* can this still happen? */
3283 ret = drop_buffers(page, &buffers_to_free);
3284 goto out;
3285 }
3286
3287 spin_lock(&mapping->private_lock);
3288 ret = drop_buffers(page, &buffers_to_free);
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003289
3290 /*
3291 * If the filesystem writes its buffers by hand (eg ext3)
3292 * then we can have clean buffers against a dirty page. We
3293 * clean the page here; otherwise the VM will never notice
3294 * that the filesystem did any IO at all.
3295 *
3296 * Also, during truncate, discard_buffer will have marked all
3297 * the page's buffers clean. We discover that here and clean
3298 * the page also.
Nick Piggin87df7242007-01-30 14:36:27 +11003299 *
3300 * private_lock must be held over this entire operation in order
3301 * to synchronise against __set_page_dirty_buffers and prevent the
3302 * dirty bit from being lost.
Linus Torvaldsecdfc972007-01-26 12:47:06 -08003303 */
Tejun Heo11f81be2015-05-22 17:13:15 -04003304 if (ret)
3305 cancel_dirty_page(page);
Nick Piggin87df7242007-01-30 14:36:27 +11003306 spin_unlock(&mapping->private_lock);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307out:
3308 if (buffers_to_free) {
3309 struct buffer_head *bh = buffers_to_free;
3310
3311 do {
3312 struct buffer_head *next = bh->b_this_page;
3313 free_buffer_head(bh);
3314 bh = next;
3315 } while (bh != buffers_to_free);
3316 }
3317 return ret;
3318}
3319EXPORT_SYMBOL(try_to_free_buffers);
3320
Linus Torvalds1da177e2005-04-16 15:20:36 -07003321/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07003322 * Buffer-head allocation
3323 */
Shai Fultheima0a9b042012-05-15 12:29:52 +03003324static struct kmem_cache *bh_cachep __read_mostly;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003325
3326/*
3327 * Once the number of bh's in the machine exceeds this level, we start
3328 * stripping them in writeback.
3329 */
Zhang Yanfei43be5942013-02-22 16:35:46 -08003330static unsigned long max_buffer_heads;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003331
3332int buffer_heads_over_limit;
3333
3334struct bh_accounting {
3335 int nr; /* Number of live bh's */
3336 int ratelimit; /* Limit cacheline bouncing */
3337};
3338
3339static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0};
3340
3341static void recalc_bh_state(void)
3342{
3343 int i;
3344 int tot = 0;
3345
Christoph Lameteree1be862010-12-06 11:40:05 -06003346 if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003347 return;
Christoph Lameterc7b92512010-12-06 11:16:28 -06003348 __this_cpu_write(bh_accounting.ratelimit, 0);
Eric Dumazet8a143422006-03-24 03:18:10 -08003349 for_each_online_cpu(i)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003350 tot += per_cpu(bh_accounting, i).nr;
3351 buffer_heads_over_limit = (tot > max_buffer_heads);
3352}
Christoph Lameterc7b92512010-12-06 11:16:28 -06003353
Al Virodd0fc662005-10-07 07:46:04 +01003354struct buffer_head *alloc_buffer_head(gfp_t gfp_flags)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003355{
Richard Kennedy019b4d12010-03-10 15:20:33 -08003356 struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003357 if (ret) {
Christoph Lametera35afb82007-05-16 22:10:57 -07003358 INIT_LIST_HEAD(&ret->b_assoc_buffers);
Thomas Gleixnerf1e67e32019-11-18 14:28:24 +01003359 spin_lock_init(&ret->b_uptodate_lock);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003360 preempt_disable();
3361 __this_cpu_inc(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003362 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003363 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003364 }
3365 return ret;
3366}
3367EXPORT_SYMBOL(alloc_buffer_head);
3368
3369void free_buffer_head(struct buffer_head *bh)
3370{
3371 BUG_ON(!list_empty(&bh->b_assoc_buffers));
3372 kmem_cache_free(bh_cachep, bh);
Christoph Lameterc7b92512010-12-06 11:16:28 -06003373 preempt_disable();
3374 __this_cpu_dec(bh_accounting.nr);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003375 recalc_bh_state();
Christoph Lameterc7b92512010-12-06 11:16:28 -06003376 preempt_enable();
Linus Torvalds1da177e2005-04-16 15:20:36 -07003377}
3378EXPORT_SYMBOL(free_buffer_head);
3379
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003380static int buffer_exit_cpu_dead(unsigned int cpu)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003381{
3382 int i;
3383 struct bh_lru *b = &per_cpu(bh_lrus, cpu);
3384
3385 for (i = 0; i < BH_LRU_SIZE; i++) {
3386 brelse(b->bhs[i]);
3387 b->bhs[i] = NULL;
3388 }
Christoph Lameterc7b92512010-12-06 11:16:28 -06003389 this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr);
Eric Dumazet8a143422006-03-24 03:18:10 -08003390 per_cpu(bh_accounting, cpu).nr = 0;
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003391 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003392}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003393
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003394/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003395 * bh_uptodate_or_lock - Test whether the buffer is uptodate
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003396 * @bh: struct buffer_head
3397 *
3398 * Return true if the buffer is up-to-date and false,
3399 * with the buffer locked, if not.
3400 */
3401int bh_uptodate_or_lock(struct buffer_head *bh)
3402{
3403 if (!buffer_uptodate(bh)) {
3404 lock_buffer(bh);
3405 if (!buffer_uptodate(bh))
3406 return 0;
3407 unlock_buffer(bh);
3408 }
3409 return 1;
3410}
3411EXPORT_SYMBOL(bh_uptodate_or_lock);
3412
3413/**
Randy Dunlapa6b91912008-03-19 17:01:00 -07003414 * bh_submit_read - Submit a locked buffer for reading
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003415 * @bh: struct buffer_head
3416 *
3417 * Returns zero on success and -EIO on error.
3418 */
3419int bh_submit_read(struct buffer_head *bh)
3420{
3421 BUG_ON(!buffer_locked(bh));
3422
3423 if (buffer_uptodate(bh)) {
3424 unlock_buffer(bh);
3425 return 0;
3426 }
3427
3428 get_bh(bh);
3429 bh->b_end_io = end_buffer_read_sync;
Mike Christie2a222ca2016-06-05 14:31:43 -05003430 submit_bh(REQ_OP_READ, 0, bh);
Aneesh Kumar K.V389d1b02008-01-28 23:58:26 -05003431 wait_on_buffer(bh);
3432 if (buffer_uptodate(bh))
3433 return 0;
3434 return -EIO;
3435}
3436EXPORT_SYMBOL(bh_submit_read);
3437
Linus Torvalds1da177e2005-04-16 15:20:36 -07003438void __init buffer_init(void)
3439{
Zhang Yanfei43be5942013-02-22 16:35:46 -08003440 unsigned long nrpages;
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003441 int ret;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003442
Christoph Lameterb98938c2008-02-04 22:28:36 -08003443 bh_cachep = kmem_cache_create("buffer_head",
3444 sizeof(struct buffer_head), 0,
3445 (SLAB_RECLAIM_ACCOUNT|SLAB_PANIC|
3446 SLAB_MEM_SPREAD),
Richard Kennedy019b4d12010-03-10 15:20:33 -08003447 NULL);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448
3449 /*
3450 * Limit the bh occupancy to 10% of ZONE_NORMAL
3451 */
3452 nrpages = (nr_free_buffer_pages() * 10) / 100;
3453 max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head));
Sebastian Andrzej Siewiorfc4d24c2016-11-03 15:49:57 +01003454 ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead",
3455 NULL, buffer_exit_cpu_dead);
3456 WARN_ON(ret < 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003457}