Thomas Gleixner | 457c899 | 2019-05-19 13:08:55 +0100 | [diff] [blame] | 1 | // SPDX-License-Identifier: GPL-2.0-only |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2 | /* |
| 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 22 | #include <linux/kernel.h> |
Ingo Molnar | f361bf4 | 2017-02-03 23:47:37 +0100 | [diff] [blame] | 23 | #include <linux/sched/signal.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 24 | #include <linux/syscalls.h> |
| 25 | #include <linux/fs.h> |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 26 | #include <linux/iomap.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 27 | #include <linux/mm.h> |
| 28 | #include <linux/percpu.h> |
| 29 | #include <linux/slab.h> |
Randy Dunlap | 16f7e0f | 2006-01-11 12:17:46 -0800 | [diff] [blame] | 30 | #include <linux/capability.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 31 | #include <linux/blkdev.h> |
| 32 | #include <linux/file.h> |
| 33 | #include <linux/quotaops.h> |
| 34 | #include <linux/highmem.h> |
Paul Gortmaker | 630d9c4 | 2011-11-16 23:57:37 -0500 | [diff] [blame] | 35 | #include <linux/export.h> |
Tejun Heo | bafc0db | 2015-06-02 08:37:23 -0600 | [diff] [blame] | 36 | #include <linux/backing-dev.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 37 | #include <linux/writeback.h> |
| 38 | #include <linux/hash.h> |
| 39 | #include <linux/suspend.h> |
| 40 | #include <linux/buffer_head.h> |
Andrew Morton | 55e829a | 2006-12-10 02:19:27 -0800 | [diff] [blame] | 41 | #include <linux/task_io_accounting_ops.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 42 | #include <linux/bio.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 43 | #include <linux/cpu.h> |
| 44 | #include <linux/bitops.h> |
| 45 | #include <linux/mpage.h> |
Ingo Molnar | fb1c8f9 | 2005-09-10 00:25:56 -0700 | [diff] [blame] | 46 | #include <linux/bit_spinlock.h> |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 47 | #include <linux/pagevec.h> |
Shakeel Butt | f745c6f | 2018-08-17 15:46:44 -0700 | [diff] [blame] | 48 | #include <linux/sched/mm.h> |
Tejun Heo | 5305cb8 | 2013-01-11 13:06:36 -0800 | [diff] [blame] | 49 | #include <trace/events/block.h> |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 50 | #include <linux/fscrypt.h> |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 51 | #include <linux/fsverity.h> |
Marcelo Tosatti | 8a237ad | 2023-06-27 17:08:15 -0300 | [diff] [blame] | 52 | #include <linux/sched/isolation.h> |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 53 | |
Ben Dooks | 2b211dc | 2019-11-30 17:49:18 -0800 | [diff] [blame] | 54 | #include "internal.h" |
| 55 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 56 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list); |
Ritesh Harjani (IBM) | 5bdf402 | 2022-08-18 10:34:40 +0530 | [diff] [blame] | 57 | static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, |
Bart Van Assche | 4498135 | 2024-02-02 12:39:25 -0800 | [diff] [blame] | 58 | enum rw_hint hint, struct writeback_control *wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 59 | |
| 60 | #define BH_ENTRY(list) list_entry((list), struct buffer_head, b_assoc_buffers) |
| 61 | |
Tejun Heo | f0059af | 2013-01-11 13:06:35 -0800 | [diff] [blame] | 62 | inline void touch_buffer(struct buffer_head *bh) |
| 63 | { |
Tejun Heo | 5305cb8 | 2013-01-11 13:06:36 -0800 | [diff] [blame] | 64 | trace_block_touch_buffer(bh); |
Matthew Wilcox (Oracle) | 03c5f33 | 2022-12-15 21:43:53 +0000 | [diff] [blame] | 65 | folio_mark_accessed(bh->b_folio); |
Tejun Heo | f0059af | 2013-01-11 13:06:35 -0800 | [diff] [blame] | 66 | } |
| 67 | EXPORT_SYMBOL(touch_buffer); |
| 68 | |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 69 | void __lock_buffer(struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 70 | { |
NeilBrown | 7431620 | 2014-07-07 15:16:04 +1000 | [diff] [blame] | 71 | wait_on_bit_lock_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 72 | } |
| 73 | EXPORT_SYMBOL(__lock_buffer); |
| 74 | |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 75 | void unlock_buffer(struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 76 | { |
Nick Piggin | 51b07fc | 2008-10-18 20:27:00 -0700 | [diff] [blame] | 77 | clear_bit_unlock(BH_Lock, &bh->b_state); |
Peter Zijlstra | 4e857c5 | 2014-03-17 18:06:10 +0100 | [diff] [blame] | 78 | smp_mb__after_atomic(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 79 | wake_up_bit(&bh->b_state, BH_Lock); |
| 80 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 81 | EXPORT_SYMBOL(unlock_buffer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 82 | |
| 83 | /* |
Matthew Wilcox (Oracle) | 520f301 | 2022-01-17 14:35:22 -0500 | [diff] [blame] | 84 | * Returns if the folio has dirty or writeback buffers. If all the buffers |
| 85 | * are unlocked and clean then the folio_test_dirty information is stale. If |
| 86 | * any of the buffers are locked, it is assumed they are locked for IO. |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 87 | */ |
Matthew Wilcox (Oracle) | 520f301 | 2022-01-17 14:35:22 -0500 | [diff] [blame] | 88 | void buffer_check_dirty_writeback(struct folio *folio, |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 89 | bool *dirty, bool *writeback) |
| 90 | { |
| 91 | struct buffer_head *head, *bh; |
| 92 | *dirty = false; |
| 93 | *writeback = false; |
| 94 | |
Matthew Wilcox (Oracle) | 520f301 | 2022-01-17 14:35:22 -0500 | [diff] [blame] | 95 | BUG_ON(!folio_test_locked(folio)); |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 96 | |
Matthew Wilcox (Oracle) | 520f301 | 2022-01-17 14:35:22 -0500 | [diff] [blame] | 97 | head = folio_buffers(folio); |
| 98 | if (!head) |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 99 | return; |
| 100 | |
Matthew Wilcox (Oracle) | 520f301 | 2022-01-17 14:35:22 -0500 | [diff] [blame] | 101 | if (folio_test_writeback(folio)) |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 102 | *writeback = true; |
| 103 | |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 104 | bh = head; |
| 105 | do { |
| 106 | if (buffer_locked(bh)) |
| 107 | *writeback = true; |
| 108 | |
| 109 | if (buffer_dirty(bh)) |
| 110 | *dirty = true; |
| 111 | |
| 112 | bh = bh->b_this_page; |
| 113 | } while (bh != head); |
| 114 | } |
Mel Gorman | b459722 | 2013-07-03 15:02:05 -0700 | [diff] [blame] | 115 | |
| 116 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 117 | * 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 | */ |
| 121 | void __wait_on_buffer(struct buffer_head * bh) |
| 122 | { |
NeilBrown | 7431620 | 2014-07-07 15:16:04 +1000 | [diff] [blame] | 123 | wait_on_bit_io(&bh->b_state, BH_Lock, TASK_UNINTERRUPTIBLE); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 124 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 125 | EXPORT_SYMBOL(__wait_on_buffer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 126 | |
Robert Elliott | b744c2a | 2014-10-21 13:55:09 -0600 | [diff] [blame] | 127 | static void buffer_io_error(struct buffer_head *bh, char *msg) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 128 | { |
Robert Elliott | 432f16e | 2014-10-21 13:55:11 -0600 | [diff] [blame] | 129 | if (!test_bit(BH_Quiet, &bh->b_state)) |
| 130 | printk_ratelimited(KERN_ERR |
Dmitry Monakhov | a1c6f057 | 2015-04-13 16:31:37 +0400 | [diff] [blame] | 131 | "Buffer I/O error on dev %pg, logical block %llu%s\n", |
| 132 | bh->b_bdev, (unsigned long long)bh->b_blocknr, msg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 133 | } |
| 134 | |
| 135 | /* |
Dmitry Monakhov | 68671f3 | 2007-10-16 01:24:47 -0700 | [diff] [blame] | 136 | * 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 Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 142 | */ |
Dmitry Monakhov | 68671f3 | 2007-10-16 01:24:47 -0700 | [diff] [blame] | 143 | static void __end_buffer_read_notouch(struct buffer_head *bh, int uptodate) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 144 | { |
| 145 | if (uptodate) { |
| 146 | set_buffer_uptodate(bh); |
| 147 | } else { |
Christoph Hellwig | 7024628 | 2016-07-19 11:28:41 +0200 | [diff] [blame] | 148 | /* This happens, due to failed read-ahead attempts. */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 149 | clear_buffer_uptodate(bh); |
| 150 | } |
| 151 | unlock_buffer(bh); |
Dmitry Monakhov | 68671f3 | 2007-10-16 01:24:47 -0700 | [diff] [blame] | 152 | } |
| 153 | |
| 154 | /* |
| 155 | * Default synchronous end-of-IO handler.. Just mark it up-to-date and |
Zhang Yi | 79f5978 | 2022-09-01 21:35:03 +0800 | [diff] [blame] | 156 | * unlock the buffer. |
Dmitry Monakhov | 68671f3 | 2007-10-16 01:24:47 -0700 | [diff] [blame] | 157 | */ |
| 158 | void end_buffer_read_sync(struct buffer_head *bh, int uptodate) |
| 159 | { |
| 160 | __end_buffer_read_notouch(bh, uptodate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 161 | put_bh(bh); |
| 162 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 163 | EXPORT_SYMBOL(end_buffer_read_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 164 | |
| 165 | void end_buffer_write_sync(struct buffer_head *bh, int uptodate) |
| 166 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 167 | if (uptodate) { |
| 168 | set_buffer_uptodate(bh); |
| 169 | } else { |
Robert Elliott | 432f16e | 2014-10-21 13:55:11 -0600 | [diff] [blame] | 170 | buffer_io_error(bh, ", lost sync page write"); |
Jeff Layton | 87354e5 | 2017-07-06 07:02:21 -0400 | [diff] [blame] | 171 | mark_buffer_write_io_error(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 172 | clear_buffer_uptodate(bh); |
| 173 | } |
| 174 | unlock_buffer(bh); |
| 175 | put_bh(bh); |
| 176 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 177 | EXPORT_SYMBOL(end_buffer_write_sync); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 178 | |
| 179 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 180 | * 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 |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 183 | * i_private_lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 184 | * |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 185 | * Hack idea: for the blockdev mapping, i_private_lock contention |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 186 | * may be quite high. This code could TryLock the page, and if that |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 187 | * succeeds, there is no need to take i_private_lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 188 | */ |
| 189 | static struct buffer_head * |
Coywolf Qi Hunt | 385fd4c | 2005-11-07 00:59:39 -0800 | [diff] [blame] | 190 | __find_get_block_slow(struct block_device *bdev, sector_t block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 191 | { |
Al Viro | 53cd4cd | 2024-04-28 19:41:13 -0400 | [diff] [blame] | 192 | struct address_space *bd_mapping = bdev->bd_mapping; |
| 193 | const int blkbits = bd_mapping->host->i_blkbits; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 194 | struct buffer_head *ret = NULL; |
| 195 | pgoff_t index; |
| 196 | struct buffer_head *bh; |
| 197 | struct buffer_head *head; |
Matthew Wilcox (Oracle) | eee25182 | 2023-06-12 22:01:40 +0100 | [diff] [blame] | 198 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 199 | int all_mapped = 1; |
Tetsuo Handa | 43636c8 | 2019-01-21 22:49:37 +0900 | [diff] [blame] | 200 | static DEFINE_RATELIMIT_STATE(last_warned, HZ, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 201 | |
Al Viro | 53cd4cd | 2024-04-28 19:41:13 -0400 | [diff] [blame] | 202 | index = ((loff_t)block << blkbits) / PAGE_SIZE; |
Matthew Wilcox (Oracle) | eee25182 | 2023-06-12 22:01:40 +0100 | [diff] [blame] | 203 | folio = __filemap_get_folio(bd_mapping, index, FGP_ACCESSED, 0); |
| 204 | if (IS_ERR(folio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 205 | goto out; |
| 206 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 207 | spin_lock(&bd_mapping->i_private_lock); |
Matthew Wilcox (Oracle) | eee25182 | 2023-06-12 22:01:40 +0100 | [diff] [blame] | 208 | head = folio_buffers(folio); |
| 209 | if (!head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 210 | goto out_unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 211 | bh = head; |
| 212 | do { |
Nikanth Karthikesan | 97f76d3 | 2009-04-02 16:56:46 -0700 | [diff] [blame] | 213 | if (!buffer_mapped(bh)) |
| 214 | all_mapped = 0; |
| 215 | else if (bh->b_blocknr == block) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 216 | ret = bh; |
| 217 | get_bh(bh); |
| 218 | goto out_unlock; |
| 219 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 220 | 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 Handa | 43636c8 | 2019-01-21 22:49:37 +0900 | [diff] [blame] | 228 | 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, |
Al Viro | 53cd4cd | 2024-04-28 19:41:13 -0400 | [diff] [blame] | 236 | 1 << blkbits); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 237 | } |
| 238 | out_unlock: |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 239 | spin_unlock(&bd_mapping->i_private_lock); |
Matthew Wilcox (Oracle) | eee25182 | 2023-06-12 22:01:40 +0100 | [diff] [blame] | 240 | folio_put(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 241 | out: |
| 242 | return ret; |
| 243 | } |
| 244 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 245 | static void end_buffer_async_read(struct buffer_head *bh, int uptodate) |
| 246 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 247 | unsigned long flags; |
Nick Piggin | a397220 | 2005-07-07 17:56:56 -0700 | [diff] [blame] | 248 | struct buffer_head *first; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 249 | struct buffer_head *tmp; |
Matthew Wilcox (Oracle) | 2e2dba1 | 2022-12-15 21:43:54 +0000 | [diff] [blame] | 250 | struct folio *folio; |
| 251 | int folio_uptodate = 1; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 252 | |
| 253 | BUG_ON(!buffer_async_read(bh)); |
| 254 | |
Matthew Wilcox (Oracle) | 2e2dba1 | 2022-12-15 21:43:54 +0000 | [diff] [blame] | 255 | folio = bh->b_folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 256 | if (uptodate) { |
| 257 | set_buffer_uptodate(bh); |
| 258 | } else { |
| 259 | clear_buffer_uptodate(bh); |
Robert Elliott | 432f16e | 2014-10-21 13:55:11 -0600 | [diff] [blame] | 260 | buffer_io_error(bh, ", async page read"); |
Matthew Wilcox (Oracle) | 2e2dba1 | 2022-12-15 21:43:54 +0000 | [diff] [blame] | 261 | folio_set_error(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 | */ |
Matthew Wilcox (Oracle) | 2e2dba1 | 2022-12-15 21:43:54 +0000 | [diff] [blame] | 269 | first = folio_buffers(folio); |
Thomas Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 270 | spin_lock_irqsave(&first->b_uptodate_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 271 | clear_buffer_async_read(bh); |
| 272 | unlock_buffer(bh); |
| 273 | tmp = bh; |
| 274 | do { |
| 275 | if (!buffer_uptodate(tmp)) |
Matthew Wilcox (Oracle) | 2e2dba1 | 2022-12-15 21:43:54 +0000 | [diff] [blame] | 276 | folio_uptodate = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 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 Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 283 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 284 | |
Matthew Wilcox (Oracle) | 6ba924d | 2023-10-04 17:53:05 +0100 | [diff] [blame] | 285 | folio_end_read(folio, folio_uptodate); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 286 | return; |
| 287 | |
| 288 | still_busy: |
Thomas Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 289 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 290 | return; |
| 291 | } |
| 292 | |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 293 | struct postprocess_bh_ctx { |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 294 | struct work_struct work; |
| 295 | struct buffer_head *bh; |
| 296 | }; |
| 297 | |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 298 | static void verify_bh(struct work_struct *work) |
| 299 | { |
| 300 | struct postprocess_bh_ctx *ctx = |
| 301 | container_of(work, struct postprocess_bh_ctx, work); |
| 302 | struct buffer_head *bh = ctx->bh; |
| 303 | bool valid; |
| 304 | |
Eric Biggers | 8b7d3fe | 2023-02-24 15:25:30 -0800 | [diff] [blame] | 305 | valid = fsverity_verify_blocks(bh->b_folio, bh->b_size, bh_offset(bh)); |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 306 | end_buffer_async_read(bh, valid); |
| 307 | kfree(ctx); |
| 308 | } |
| 309 | |
| 310 | static bool need_fsverity(struct buffer_head *bh) |
| 311 | { |
Eric Biggers | 8b7d3fe | 2023-02-24 15:25:30 -0800 | [diff] [blame] | 312 | struct folio *folio = bh->b_folio; |
| 313 | struct inode *inode = folio->mapping->host; |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 314 | |
| 315 | return fsverity_active(inode) && |
| 316 | /* needed by ext4 */ |
Eric Biggers | 8b7d3fe | 2023-02-24 15:25:30 -0800 | [diff] [blame] | 317 | folio->index < DIV_ROUND_UP(inode->i_size, PAGE_SIZE); |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 318 | } |
| 319 | |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 320 | static void decrypt_bh(struct work_struct *work) |
| 321 | { |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 322 | struct postprocess_bh_ctx *ctx = |
| 323 | container_of(work, struct postprocess_bh_ctx, work); |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 324 | struct buffer_head *bh = ctx->bh; |
| 325 | int err; |
| 326 | |
Eric Biggers | 9c7fb7f | 2023-02-24 15:25:03 -0800 | [diff] [blame] | 327 | err = fscrypt_decrypt_pagecache_blocks(bh->b_folio, bh->b_size, |
| 328 | bh_offset(bh)); |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 329 | if (err == 0 && need_fsverity(bh)) { |
| 330 | /* |
| 331 | * We use different work queues for decryption and for verity |
| 332 | * because verity may require reading metadata pages that need |
| 333 | * decryption, and we shouldn't recurse to the same workqueue. |
| 334 | */ |
| 335 | INIT_WORK(&ctx->work, verify_bh); |
| 336 | fsverity_enqueue_verify_work(&ctx->work); |
| 337 | return; |
| 338 | } |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 339 | end_buffer_async_read(bh, err == 0); |
| 340 | kfree(ctx); |
| 341 | } |
| 342 | |
| 343 | /* |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 344 | * I/O completion handler for block_read_full_folio() - pages |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 345 | * which come unlocked at the end of I/O. |
| 346 | */ |
| 347 | static void end_buffer_async_read_io(struct buffer_head *bh, int uptodate) |
| 348 | { |
Linus Torvalds | 3822a7c | 2023-02-23 17:09:35 -0800 | [diff] [blame] | 349 | struct inode *inode = bh->b_folio->mapping->host; |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 350 | bool decrypt = fscrypt_inode_uses_fs_layer_crypto(inode); |
| 351 | bool verify = need_fsverity(bh); |
| 352 | |
| 353 | /* Decrypt (with fscrypt) and/or verify (with fsverity) if needed. */ |
| 354 | if (uptodate && (decrypt || verify)) { |
| 355 | struct postprocess_bh_ctx *ctx = |
| 356 | kmalloc(sizeof(*ctx), GFP_ATOMIC); |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 357 | |
| 358 | if (ctx) { |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 359 | ctx->bh = bh; |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 360 | if (decrypt) { |
| 361 | INIT_WORK(&ctx->work, decrypt_bh); |
| 362 | fscrypt_enqueue_decrypt_work(&ctx->work); |
| 363 | } else { |
| 364 | INIT_WORK(&ctx->work, verify_bh); |
| 365 | fsverity_enqueue_verify_work(&ctx->work); |
| 366 | } |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 367 | return; |
| 368 | } |
| 369 | uptodate = 0; |
| 370 | } |
| 371 | end_buffer_async_read(bh, uptodate); |
| 372 | } |
| 373 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 374 | /* |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 375 | * Completion handler for block_write_full_folio() - folios which are unlocked |
| 376 | * during I/O, and which have the writeback flag cleared upon I/O completion. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 377 | */ |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 378 | static void end_buffer_async_write(struct buffer_head *bh, int uptodate) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 379 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 380 | unsigned long flags; |
Nick Piggin | a397220 | 2005-07-07 17:56:56 -0700 | [diff] [blame] | 381 | struct buffer_head *first; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 382 | struct buffer_head *tmp; |
Matthew Wilcox (Oracle) | 743ed81 | 2022-12-15 21:43:55 +0000 | [diff] [blame] | 383 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 384 | |
| 385 | BUG_ON(!buffer_async_write(bh)); |
| 386 | |
Matthew Wilcox (Oracle) | 743ed81 | 2022-12-15 21:43:55 +0000 | [diff] [blame] | 387 | folio = bh->b_folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 388 | if (uptodate) { |
| 389 | set_buffer_uptodate(bh); |
| 390 | } else { |
Robert Elliott | 432f16e | 2014-10-21 13:55:11 -0600 | [diff] [blame] | 391 | buffer_io_error(bh, ", lost async page write"); |
Jeff Layton | 87354e5 | 2017-07-06 07:02:21 -0400 | [diff] [blame] | 392 | mark_buffer_write_io_error(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 393 | clear_buffer_uptodate(bh); |
Matthew Wilcox (Oracle) | 743ed81 | 2022-12-15 21:43:55 +0000 | [diff] [blame] | 394 | folio_set_error(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 395 | } |
| 396 | |
Matthew Wilcox (Oracle) | 743ed81 | 2022-12-15 21:43:55 +0000 | [diff] [blame] | 397 | first = folio_buffers(folio); |
Thomas Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 398 | spin_lock_irqsave(&first->b_uptodate_lock, flags); |
Nick Piggin | a397220 | 2005-07-07 17:56:56 -0700 | [diff] [blame] | 399 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 400 | clear_buffer_async_write(bh); |
| 401 | unlock_buffer(bh); |
| 402 | tmp = bh->b_this_page; |
| 403 | while (tmp != bh) { |
| 404 | if (buffer_async_write(tmp)) { |
| 405 | BUG_ON(!buffer_locked(tmp)); |
| 406 | goto still_busy; |
| 407 | } |
| 408 | tmp = tmp->b_this_page; |
| 409 | } |
Thomas Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 410 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
Matthew Wilcox (Oracle) | 743ed81 | 2022-12-15 21:43:55 +0000 | [diff] [blame] | 411 | folio_end_writeback(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 412 | return; |
| 413 | |
| 414 | still_busy: |
Thomas Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 415 | spin_unlock_irqrestore(&first->b_uptodate_lock, flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 416 | return; |
| 417 | } |
| 418 | |
| 419 | /* |
| 420 | * If a page's buffers are under async readin (end_buffer_async_read |
| 421 | * completion) then there is a possibility that another thread of |
| 422 | * control could lock one of the buffers after it has completed |
| 423 | * but while some of the other buffers have not completed. This |
| 424 | * locked buffer would confuse end_buffer_async_read() into not unlocking |
| 425 | * the page. So the absence of BH_Async_Read tells end_buffer_async_read() |
| 426 | * that this buffer is not under async I/O. |
| 427 | * |
| 428 | * The page comes unlocked when it has no locked buffer_async buffers |
| 429 | * left. |
| 430 | * |
| 431 | * PageLocked prevents anyone starting new async I/O reads any of |
| 432 | * the buffers. |
| 433 | * |
| 434 | * PageWriteback is used to prevent simultaneous writeout of the same |
| 435 | * page. |
| 436 | * |
| 437 | * PageLocked prevents anyone from starting writeback of a page which is |
| 438 | * under read I/O (PageWriteback is only ever set against a locked page). |
| 439 | */ |
| 440 | static void mark_buffer_async_read(struct buffer_head *bh) |
| 441 | { |
Eric Biggers | 31fb992 | 2019-10-22 20:33:11 -0700 | [diff] [blame] | 442 | bh->b_end_io = end_buffer_async_read_io; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 443 | set_buffer_async_read(bh); |
| 444 | } |
| 445 | |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 446 | static void mark_buffer_async_write_endio(struct buffer_head *bh, |
| 447 | bh_end_io_t *handler) |
Chris Mason | 35c80d5 | 2009-04-15 13:22:38 -0400 | [diff] [blame] | 448 | { |
| 449 | bh->b_end_io = handler; |
| 450 | set_buffer_async_write(bh); |
| 451 | } |
| 452 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 453 | void mark_buffer_async_write(struct buffer_head *bh) |
| 454 | { |
Chris Mason | 35c80d5 | 2009-04-15 13:22:38 -0400 | [diff] [blame] | 455 | mark_buffer_async_write_endio(bh, end_buffer_async_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 456 | } |
| 457 | EXPORT_SYMBOL(mark_buffer_async_write); |
| 458 | |
| 459 | |
| 460 | /* |
| 461 | * fs/buffer.c contains helper functions for buffer-backed address space's |
| 462 | * fsync functions. A common requirement for buffer-based filesystems is |
| 463 | * that certain data from the backing blockdev needs to be written out for |
| 464 | * a successful fsync(). For example, ext2 indirect blocks need to be |
| 465 | * written back and waited upon before fsync() returns. |
| 466 | * |
Andreas Gruenbacher | 73f65b8 | 2024-01-08 18:20:40 +0100 | [diff] [blame] | 467 | * The functions mark_buffer_dirty_inode(), fsync_inode_buffers(), |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 468 | * inode_has_buffers() and invalidate_inode_buffers() are provided for the |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 469 | * management of a list of dependent buffers at ->i_mapping->i_private_list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 470 | * |
| 471 | * Locking is a little subtle: try_to_free_buffers() will remove buffers |
| 472 | * from their controlling inode's queue when they are being freed. But |
| 473 | * try_to_free_buffers() will be operating against the *blockdev* mapping |
| 474 | * at the time, not against the S_ISREG file which depends on those buffers. |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 475 | * So the locking for i_private_list is via the i_private_lock in the address_space |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 476 | * which backs the buffers. Which is different from the address_space |
| 477 | * against which the buffers are listed. So for a particular address_space, |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 478 | * mapping->i_private_lock does *not* protect mapping->i_private_list! In fact, |
| 479 | * mapping->i_private_list will always be protected by the backing blockdev's |
| 480 | * ->i_private_lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 481 | * |
| 482 | * Which introduces a requirement: all buffers on an address_space's |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 483 | * ->i_private_list must be from the same address_space: the blockdev's. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 484 | * |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 485 | * address_spaces which do not place buffers at ->i_private_list via these |
| 486 | * utility functions are free to use i_private_lock and i_private_list for |
| 487 | * whatever they want. The only requirement is that list_empty(i_private_list) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 488 | * be true at clear_inode() time. |
| 489 | * |
| 490 | * FIXME: clear_inode should not call invalidate_inode_buffers(). The |
| 491 | * filesystems should do that. invalidate_inode_buffers() should just go |
| 492 | * BUG_ON(!list_empty). |
| 493 | * |
| 494 | * FIXME: mark_buffer_dirty_inode() is a data-plane operation. It should |
| 495 | * take an address_space, not an inode. And it should be called |
| 496 | * mark_buffer_dirty_fsync() to clearly define why those buffers are being |
| 497 | * queued up. |
| 498 | * |
| 499 | * FIXME: mark_buffer_dirty_inode() doesn't need to add the buffer to the |
| 500 | * list if it is already on a list. Because if the buffer is on a list, |
| 501 | * it *must* already be on the right one. If not, the filesystem is being |
| 502 | * silly. This will save a ton of locking. But first we have to ensure |
| 503 | * that buffers are taken *off* the old inode's list when they are freed |
| 504 | * (presumably in truncate). That requires careful auditing of all |
| 505 | * filesystems (do it inside bforget()). It could also be done by bringing |
| 506 | * b_inode back. |
| 507 | */ |
| 508 | |
| 509 | /* |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 510 | * The buffer's backing address_space's i_private_lock must be held |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 511 | */ |
Thomas Petazzoni | dbacefc | 2008-07-29 22:33:47 -0700 | [diff] [blame] | 512 | static void __remove_assoc_queue(struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 513 | { |
| 514 | list_del_init(&bh->b_assoc_buffers); |
Jan Kara | 58ff407b | 2006-10-17 00:10:19 -0700 | [diff] [blame] | 515 | WARN_ON(!bh->b_assoc_map); |
Jan Kara | 58ff407b | 2006-10-17 00:10:19 -0700 | [diff] [blame] | 516 | bh->b_assoc_map = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 517 | } |
| 518 | |
| 519 | int inode_has_buffers(struct inode *inode) |
| 520 | { |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 521 | return !list_empty(&inode->i_data.i_private_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 522 | } |
| 523 | |
| 524 | /* |
| 525 | * osync is designed to support O_SYNC io. It waits synchronously for |
| 526 | * all already-submitted IO to complete, but does not queue any new |
| 527 | * writes to the disk. |
| 528 | * |
Zhang Yi | 79f5978 | 2022-09-01 21:35:03 +0800 | [diff] [blame] | 529 | * To do O_SYNC writes, just queue the buffer writes with write_dirty_buffer |
| 530 | * as you dirty the buffers, and then use osync_inode_buffers to wait for |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 531 | * completion. Any other dirty buffers which are not yet queued for |
| 532 | * write will not be flushed to disk by the osync. |
| 533 | */ |
| 534 | static int osync_buffers_list(spinlock_t *lock, struct list_head *list) |
| 535 | { |
| 536 | struct buffer_head *bh; |
| 537 | struct list_head *p; |
| 538 | int err = 0; |
| 539 | |
| 540 | spin_lock(lock); |
| 541 | repeat: |
| 542 | list_for_each_prev(p, list) { |
| 543 | bh = BH_ENTRY(p); |
| 544 | if (buffer_locked(bh)) { |
| 545 | get_bh(bh); |
| 546 | spin_unlock(lock); |
| 547 | wait_on_buffer(bh); |
| 548 | if (!buffer_uptodate(bh)) |
| 549 | err = -EIO; |
| 550 | brelse(bh); |
| 551 | spin_lock(lock); |
| 552 | goto repeat; |
| 553 | } |
| 554 | } |
| 555 | spin_unlock(lock); |
| 556 | return err; |
| 557 | } |
| 558 | |
| 559 | /** |
Randy Dunlap | 78a4a50 | 2008-02-29 22:02:31 -0800 | [diff] [blame] | 560 | * sync_mapping_buffers - write out & wait upon a mapping's "associated" buffers |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 561 | * @mapping: the mapping which wants those buffers written |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 562 | * |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 563 | * Starts I/O against the buffers at mapping->i_private_list, and waits upon |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 564 | * that I/O. |
| 565 | * |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 566 | * Basically, this is a convenience function for fsync(). |
| 567 | * @mapping is a file or directory which needs those buffers to be written for |
| 568 | * a successful fsync(). |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 569 | */ |
| 570 | int sync_mapping_buffers(struct address_space *mapping) |
| 571 | { |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 572 | struct address_space *buffer_mapping = mapping->i_private_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 573 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 574 | if (buffer_mapping == NULL || list_empty(&mapping->i_private_list)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 575 | return 0; |
| 576 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 577 | return fsync_buffers_list(&buffer_mapping->i_private_lock, |
| 578 | &mapping->i_private_list); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 579 | } |
| 580 | EXPORT_SYMBOL(sync_mapping_buffers); |
| 581 | |
Ritesh Harjani (IBM) | 31b2ebc | 2023-04-21 15:16:12 +0530 | [diff] [blame] | 582 | /** |
| 583 | * generic_buffers_fsync_noflush - generic buffer fsync implementation |
| 584 | * for simple filesystems with no inode lock |
| 585 | * |
| 586 | * @file: file to synchronize |
| 587 | * @start: start offset in bytes |
| 588 | * @end: end offset in bytes (inclusive) |
| 589 | * @datasync: only synchronize essential metadata if true |
| 590 | * |
| 591 | * This is a generic implementation of the fsync method for simple |
| 592 | * filesystems which track all non-inode metadata in the buffers list |
| 593 | * hanging off the address_space structure. |
| 594 | */ |
| 595 | int generic_buffers_fsync_noflush(struct file *file, loff_t start, loff_t end, |
| 596 | bool datasync) |
| 597 | { |
| 598 | struct inode *inode = file->f_mapping->host; |
| 599 | int err; |
| 600 | int ret; |
| 601 | |
| 602 | err = file_write_and_wait_range(file, start, end); |
| 603 | if (err) |
| 604 | return err; |
| 605 | |
| 606 | ret = sync_mapping_buffers(inode->i_mapping); |
| 607 | if (!(inode->i_state & I_DIRTY_ALL)) |
| 608 | goto out; |
| 609 | if (datasync && !(inode->i_state & I_DIRTY_DATASYNC)) |
| 610 | goto out; |
| 611 | |
| 612 | err = sync_inode_metadata(inode, 1); |
| 613 | if (ret == 0) |
| 614 | ret = err; |
| 615 | |
| 616 | out: |
| 617 | /* check and advance again to catch errors after syncing out buffers */ |
| 618 | err = file_check_and_advance_wb_err(file); |
| 619 | if (ret == 0) |
| 620 | ret = err; |
| 621 | return ret; |
| 622 | } |
| 623 | EXPORT_SYMBOL(generic_buffers_fsync_noflush); |
| 624 | |
| 625 | /** |
| 626 | * generic_buffers_fsync - generic buffer fsync implementation |
| 627 | * for simple filesystems with no inode lock |
| 628 | * |
| 629 | * @file: file to synchronize |
| 630 | * @start: start offset in bytes |
| 631 | * @end: end offset in bytes (inclusive) |
| 632 | * @datasync: only synchronize essential metadata if true |
| 633 | * |
| 634 | * This is a generic implementation of the fsync method for simple |
| 635 | * filesystems which track all non-inode metadata in the buffers list |
| 636 | * hanging off the address_space structure. This also makes sure that |
| 637 | * a device cache flush operation is called at the end. |
| 638 | */ |
| 639 | int generic_buffers_fsync(struct file *file, loff_t start, loff_t end, |
| 640 | bool datasync) |
| 641 | { |
| 642 | struct inode *inode = file->f_mapping->host; |
| 643 | int ret; |
| 644 | |
| 645 | ret = generic_buffers_fsync_noflush(file, start, end, datasync); |
| 646 | if (!ret) |
| 647 | ret = blkdev_issue_flush(inode->i_sb->s_bdev); |
| 648 | return ret; |
| 649 | } |
| 650 | EXPORT_SYMBOL(generic_buffers_fsync); |
| 651 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 652 | /* |
| 653 | * Called when we've recently written block `bblock', and it is known that |
| 654 | * `bblock' was for a buffer_boundary() buffer. This means that the block at |
| 655 | * `bblock + 1' is probably a dirty indirect block. Hunt it down and, if it's |
| 656 | * dirty, schedule it for IO. So that indirects merge nicely with their data. |
| 657 | */ |
| 658 | void write_boundary_block(struct block_device *bdev, |
| 659 | sector_t bblock, unsigned blocksize) |
| 660 | { |
| 661 | struct buffer_head *bh = __find_get_block(bdev, bblock + 1, blocksize); |
| 662 | if (bh) { |
| 663 | if (buffer_dirty(bh)) |
Zhang Yi | e7ea112 | 2022-09-01 21:34:54 +0800 | [diff] [blame] | 664 | write_dirty_buffer(bh, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 665 | put_bh(bh); |
| 666 | } |
| 667 | } |
| 668 | |
| 669 | void mark_buffer_dirty_inode(struct buffer_head *bh, struct inode *inode) |
| 670 | { |
| 671 | struct address_space *mapping = inode->i_mapping; |
Matthew Wilcox (Oracle) | abc8a8a | 2022-12-15 21:43:52 +0000 | [diff] [blame] | 672 | struct address_space *buffer_mapping = bh->b_folio->mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 673 | |
| 674 | mark_buffer_dirty(bh); |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 675 | if (!mapping->i_private_data) { |
| 676 | mapping->i_private_data = buffer_mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 677 | } else { |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 678 | BUG_ON(mapping->i_private_data != buffer_mapping); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 679 | } |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 680 | if (!bh->b_assoc_map) { |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 681 | spin_lock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 682 | list_move_tail(&bh->b_assoc_buffers, |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 683 | &mapping->i_private_list); |
Jan Kara | 58ff407b | 2006-10-17 00:10:19 -0700 | [diff] [blame] | 684 | bh->b_assoc_map = mapping; |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 685 | spin_unlock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 686 | } |
| 687 | } |
| 688 | EXPORT_SYMBOL(mark_buffer_dirty_inode); |
| 689 | |
Matthew Wilcox (Oracle) | 3814ec8 | 2024-04-16 04:17:46 +0100 | [diff] [blame] | 690 | /** |
| 691 | * block_dirty_folio - Mark a folio as dirty. |
| 692 | * @mapping: The address space containing this folio. |
| 693 | * @folio: The folio to mark dirty. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 694 | * |
Matthew Wilcox (Oracle) | 3814ec8 | 2024-04-16 04:17:46 +0100 | [diff] [blame] | 695 | * Filesystems which use buffer_heads can use this function as their |
| 696 | * ->dirty_folio implementation. Some filesystems need to do a little |
| 697 | * work before calling this function. Filesystems which do not use |
| 698 | * buffer_heads should call filemap_dirty_folio() instead. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 699 | * |
Matthew Wilcox (Oracle) | 3814ec8 | 2024-04-16 04:17:46 +0100 | [diff] [blame] | 700 | * If the folio has buffers, the uptodate buffers are set dirty, to |
| 701 | * preserve dirty-state coherency between the folio and the buffers. |
| 702 | * Buffers added to a dirty folio are created dirty. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 703 | * |
Matthew Wilcox (Oracle) | 3814ec8 | 2024-04-16 04:17:46 +0100 | [diff] [blame] | 704 | * The buffers are dirtied before the folio is dirtied. There's a small |
| 705 | * race window in which writeback may see the folio cleanness but not the |
| 706 | * buffer dirtiness. That's fine. If this code were to set the folio |
| 707 | * dirty before the buffers, writeback could clear the folio dirty flag, |
| 708 | * see a bunch of clean buffers and we'd end up with dirty buffers/clean |
| 709 | * folio on the dirty folio list. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 710 | * |
Matthew Wilcox (Oracle) | 3814ec8 | 2024-04-16 04:17:46 +0100 | [diff] [blame] | 711 | * We use i_private_lock to lock against try_to_free_buffers() while |
| 712 | * using the folio's buffer list. This also prevents clean buffers |
| 713 | * being added to the folio after it was set dirty. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 714 | * |
Matthew Wilcox (Oracle) | 3814ec8 | 2024-04-16 04:17:46 +0100 | [diff] [blame] | 715 | * Context: May only be called from process context. Does not sleep. |
| 716 | * Caller must ensure that @folio cannot be truncated during this call, |
| 717 | * typically by holding the folio lock or having a page in the folio |
| 718 | * mapped and holding the page table lock. |
| 719 | * |
| 720 | * Return: True if the folio was dirtied; false if it was already dirtied. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 721 | */ |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 722 | bool block_dirty_folio(struct address_space *mapping, struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 723 | { |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 724 | struct buffer_head *head; |
| 725 | bool newly_dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 726 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 727 | spin_lock(&mapping->i_private_lock); |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 728 | head = folio_buffers(folio); |
| 729 | if (head) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 730 | struct buffer_head *bh = head; |
| 731 | |
| 732 | do { |
| 733 | set_buffer_dirty(bh); |
| 734 | bh = bh->b_this_page; |
| 735 | } while (bh != head); |
| 736 | } |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 737 | /* |
Roman Gushchin | bcfe06b | 2020-12-01 13:58:27 -0800 | [diff] [blame] | 738 | * Lock out page's memcg migration to keep PageDirty |
Johannes Weiner | 81f8c3a | 2016-03-15 14:57:04 -0700 | [diff] [blame] | 739 | * synchronized with per-memcg dirty page counters. |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 740 | */ |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 741 | folio_memcg_lock(folio); |
| 742 | newly_dirty = !folio_test_set_dirty(folio); |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 743 | spin_unlock(&mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 744 | |
Linus Torvalds | a8e7d49 | 2009-03-19 11:32:05 -0700 | [diff] [blame] | 745 | if (newly_dirty) |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 746 | __folio_mark_dirty(folio, mapping, 1); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 747 | |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 748 | folio_memcg_unlock(folio); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 749 | |
| 750 | if (newly_dirty) |
| 751 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
| 752 | |
Linus Torvalds | a8e7d49 | 2009-03-19 11:32:05 -0700 | [diff] [blame] | 753 | return newly_dirty; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 754 | } |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 755 | EXPORT_SYMBOL(block_dirty_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 756 | |
| 757 | /* |
| 758 | * Write out and wait upon a list of buffers. |
| 759 | * |
| 760 | * We have conflicting pressures: we want to make sure that all |
| 761 | * initially dirty buffers get waited on, but that any subsequently |
| 762 | * dirtied buffers don't. After all, we don't want fsync to last |
| 763 | * forever if somebody is actively writing to the file. |
| 764 | * |
| 765 | * Do this in two main stages: first we copy dirty buffers to a |
| 766 | * temporary inode list, queueing the writes as we go. Then we clean |
| 767 | * up, waiting for those writes to complete. |
| 768 | * |
| 769 | * During this second stage, any subsequent updates to the file may end |
| 770 | * up refiling the buffer on the original inode's dirty list again, so |
| 771 | * there is a chance we will end up with a buffer queued for write but |
| 772 | * not yet completed on that list. So, as a final cleanup we go through |
| 773 | * the osync code to catch these locked, dirty buffers without requeuing |
| 774 | * any newly dirty buffers for write. |
| 775 | */ |
| 776 | static int fsync_buffers_list(spinlock_t *lock, struct list_head *list) |
| 777 | { |
| 778 | struct buffer_head *bh; |
| 779 | struct list_head tmp; |
Jens Axboe | 7eaceac | 2011-03-10 08:52:07 +0100 | [diff] [blame] | 780 | struct address_space *mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 781 | int err = 0, err2; |
Jens Axboe | 4ee2491 | 2011-03-17 10:51:40 +0100 | [diff] [blame] | 782 | struct blk_plug plug; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 783 | |
| 784 | INIT_LIST_HEAD(&tmp); |
Jens Axboe | 4ee2491 | 2011-03-17 10:51:40 +0100 | [diff] [blame] | 785 | blk_start_plug(&plug); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 786 | |
| 787 | spin_lock(lock); |
| 788 | while (!list_empty(list)) { |
| 789 | bh = BH_ENTRY(list->next); |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 790 | mapping = bh->b_assoc_map; |
Jan Kara | 58ff407b | 2006-10-17 00:10:19 -0700 | [diff] [blame] | 791 | __remove_assoc_queue(bh); |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 792 | /* Avoid race with mark_buffer_dirty_inode() which does |
| 793 | * a lockless check and we rely on seeing the dirty bit */ |
| 794 | smp_mb(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 795 | if (buffer_dirty(bh) || buffer_locked(bh)) { |
| 796 | list_add(&bh->b_assoc_buffers, &tmp); |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 797 | bh->b_assoc_map = mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 798 | if (buffer_dirty(bh)) { |
| 799 | get_bh(bh); |
| 800 | spin_unlock(lock); |
| 801 | /* |
| 802 | * Ensure any pending I/O completes so that |
Christoph Hellwig | 9cb569d | 2010-08-11 17:06:24 +0200 | [diff] [blame] | 803 | * write_dirty_buffer() actually writes the |
| 804 | * current contents - it is a noop if I/O is |
| 805 | * still in flight on potentially older |
| 806 | * contents. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 807 | */ |
Christoph Hellwig | 70fd761 | 2016-11-01 07:40:10 -0600 | [diff] [blame] | 808 | write_dirty_buffer(bh, REQ_SYNC); |
Jens Axboe | 9cf6b72 | 2009-04-06 14:48:03 +0200 | [diff] [blame] | 809 | |
| 810 | /* |
| 811 | * Kick off IO for the previous mapping. Note |
| 812 | * that we will not run the very last mapping, |
| 813 | * wait_on_buffer() will do that for us |
| 814 | * through sync_buffer(). |
| 815 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 816 | brelse(bh); |
| 817 | spin_lock(lock); |
| 818 | } |
| 819 | } |
| 820 | } |
| 821 | |
Jens Axboe | 4ee2491 | 2011-03-17 10:51:40 +0100 | [diff] [blame] | 822 | spin_unlock(lock); |
| 823 | blk_finish_plug(&plug); |
| 824 | spin_lock(lock); |
| 825 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 826 | while (!list_empty(&tmp)) { |
| 827 | bh = BH_ENTRY(tmp.prev); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 828 | get_bh(bh); |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 829 | mapping = bh->b_assoc_map; |
| 830 | __remove_assoc_queue(bh); |
| 831 | /* Avoid race with mark_buffer_dirty_inode() which does |
| 832 | * a lockless check and we rely on seeing the dirty bit */ |
| 833 | smp_mb(); |
| 834 | if (buffer_dirty(bh)) { |
| 835 | list_add(&bh->b_assoc_buffers, |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 836 | &mapping->i_private_list); |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 837 | bh->b_assoc_map = mapping; |
| 838 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 839 | spin_unlock(lock); |
| 840 | wait_on_buffer(bh); |
| 841 | if (!buffer_uptodate(bh)) |
| 842 | err = -EIO; |
| 843 | brelse(bh); |
| 844 | spin_lock(lock); |
| 845 | } |
| 846 | |
| 847 | spin_unlock(lock); |
| 848 | err2 = osync_buffers_list(lock, list); |
| 849 | if (err) |
| 850 | return err; |
| 851 | else |
| 852 | return err2; |
| 853 | } |
| 854 | |
| 855 | /* |
| 856 | * Invalidate any and all dirty buffers on a given inode. We are |
| 857 | * probably unmounting the fs, but that doesn't mean we have already |
| 858 | * done a sync(). Just drop the buffers from the inode list. |
| 859 | * |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 860 | * NOTE: we take the inode's blockdev's mapping's i_private_lock. Which |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 861 | * assumes that all the buffers are against the blockdev. Not true |
| 862 | * for reiserfs. |
| 863 | */ |
| 864 | void invalidate_inode_buffers(struct inode *inode) |
| 865 | { |
| 866 | if (inode_has_buffers(inode)) { |
| 867 | struct address_space *mapping = &inode->i_data; |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 868 | struct list_head *list = &mapping->i_private_list; |
| 869 | struct address_space *buffer_mapping = mapping->i_private_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 870 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 871 | spin_lock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 872 | while (!list_empty(list)) |
| 873 | __remove_assoc_queue(BH_ENTRY(list->next)); |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 874 | spin_unlock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 875 | } |
| 876 | } |
Jan Kara | 52b19ac | 2008-09-23 18:24:08 +0200 | [diff] [blame] | 877 | EXPORT_SYMBOL(invalidate_inode_buffers); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 878 | |
| 879 | /* |
| 880 | * Remove any clean buffers from the inode's buffer list. This is called |
| 881 | * when we're trying to free the inode itself. Those buffers can pin it. |
| 882 | * |
| 883 | * Returns true if all buffers were removed. |
| 884 | */ |
| 885 | int remove_inode_buffers(struct inode *inode) |
| 886 | { |
| 887 | int ret = 1; |
| 888 | |
| 889 | if (inode_has_buffers(inode)) { |
| 890 | struct address_space *mapping = &inode->i_data; |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 891 | struct list_head *list = &mapping->i_private_list; |
| 892 | struct address_space *buffer_mapping = mapping->i_private_data; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 893 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 894 | spin_lock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 895 | while (!list_empty(list)) { |
| 896 | struct buffer_head *bh = BH_ENTRY(list->next); |
| 897 | if (buffer_dirty(bh)) { |
| 898 | ret = 0; |
| 899 | break; |
| 900 | } |
| 901 | __remove_assoc_queue(bh); |
| 902 | } |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 903 | spin_unlock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 904 | } |
| 905 | return ret; |
| 906 | } |
| 907 | |
| 908 | /* |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 909 | * Create the appropriate buffers when given a folio for data area and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 910 | * the size of each buffer.. Use the bh->b_this_page linked list to |
| 911 | * follow the buffers created. Return NULL if unable to create more |
| 912 | * buffers. |
| 913 | * |
| 914 | * The retry flag is used to differentiate async IO (paging, swapping) |
| 915 | * which may not fail from ordinary buffer allocations. |
| 916 | */ |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 917 | struct buffer_head *folio_alloc_buffers(struct folio *folio, unsigned long size, |
Matthew Wilcox (Oracle) | 2a41815 | 2023-09-14 16:00:04 +0100 | [diff] [blame] | 918 | gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 919 | { |
| 920 | struct buffer_head *bh, *head; |
| 921 | long offset; |
Roman Gushchin | b87d8ce | 2020-10-17 16:13:40 -0700 | [diff] [blame] | 922 | struct mem_cgroup *memcg, *old_memcg; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 923 | |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 924 | /* The folio lock pins the memcg */ |
| 925 | memcg = folio_memcg(folio); |
Roman Gushchin | b87d8ce | 2020-10-17 16:13:40 -0700 | [diff] [blame] | 926 | old_memcg = set_active_memcg(memcg); |
Shakeel Butt | f745c6f | 2018-08-17 15:46:44 -0700 | [diff] [blame] | 927 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 928 | head = NULL; |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 929 | offset = folio_size(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 930 | while ((offset -= size) >= 0) { |
Jens Axboe | 640ab98 | 2017-09-27 05:40:16 -0600 | [diff] [blame] | 931 | bh = alloc_buffer_head(gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 932 | if (!bh) |
| 933 | goto no_grow; |
| 934 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 935 | bh->b_this_page = head; |
| 936 | bh->b_blocknr = -1; |
| 937 | head = bh; |
| 938 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 939 | bh->b_size = size; |
| 940 | |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 941 | /* Link the buffer to its folio */ |
| 942 | folio_set_bh(bh, folio, offset); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 943 | } |
Shakeel Butt | f745c6f | 2018-08-17 15:46:44 -0700 | [diff] [blame] | 944 | out: |
Roman Gushchin | b87d8ce | 2020-10-17 16:13:40 -0700 | [diff] [blame] | 945 | set_active_memcg(old_memcg); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 946 | return head; |
| 947 | /* |
| 948 | * In case anything failed, we just free everything we got. |
| 949 | */ |
| 950 | no_grow: |
| 951 | if (head) { |
| 952 | do { |
| 953 | bh = head; |
| 954 | head = head->b_this_page; |
| 955 | free_buffer_head(bh); |
| 956 | } while (head); |
| 957 | } |
| 958 | |
Shakeel Butt | f745c6f | 2018-08-17 15:46:44 -0700 | [diff] [blame] | 959 | goto out; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 960 | } |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 961 | EXPORT_SYMBOL_GPL(folio_alloc_buffers); |
| 962 | |
| 963 | struct buffer_head *alloc_page_buffers(struct page *page, unsigned long size, |
| 964 | bool retry) |
| 965 | { |
Matthew Wilcox (Oracle) | 2a41815 | 2023-09-14 16:00:04 +0100 | [diff] [blame] | 966 | gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT; |
| 967 | if (retry) |
| 968 | gfp |= __GFP_NOFAIL; |
| 969 | |
| 970 | return folio_alloc_buffers(page_folio(page), size, gfp); |
Pankaj Raghav | c71124a | 2023-04-17 14:36:16 +0200 | [diff] [blame] | 971 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 972 | EXPORT_SYMBOL_GPL(alloc_page_buffers); |
| 973 | |
Matthew Wilcox (Oracle) | 08d84ad | 2023-06-12 22:01:39 +0100 | [diff] [blame] | 974 | static inline void link_dev_buffers(struct folio *folio, |
| 975 | struct buffer_head *head) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 976 | { |
| 977 | struct buffer_head *bh, *tail; |
| 978 | |
| 979 | bh = head; |
| 980 | do { |
| 981 | tail = bh; |
| 982 | bh = bh->b_this_page; |
| 983 | } while (bh); |
| 984 | tail->b_this_page = head; |
Matthew Wilcox (Oracle) | 08d84ad | 2023-06-12 22:01:39 +0100 | [diff] [blame] | 985 | folio_attach_private(folio, head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 986 | } |
| 987 | |
Linus Torvalds | bbec0270 | 2012-11-29 12:31:52 -0800 | [diff] [blame] | 988 | static sector_t blkdev_max_block(struct block_device *bdev, unsigned int size) |
| 989 | { |
| 990 | sector_t retval = ~((sector_t)0); |
Christoph Hellwig | b86058f | 2021-10-18 12:11:09 +0200 | [diff] [blame] | 991 | loff_t sz = bdev_nr_bytes(bdev); |
Linus Torvalds | bbec0270 | 2012-11-29 12:31:52 -0800 | [diff] [blame] | 992 | |
| 993 | if (sz) { |
| 994 | unsigned int sizebits = blksize_bits(size); |
| 995 | retval = (sz >> sizebits); |
| 996 | } |
| 997 | return retval; |
| 998 | } |
| 999 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1000 | /* |
Matthew Wilcox (Oracle) | 6f24ce6 | 2023-06-12 22:01:38 +0100 | [diff] [blame] | 1001 | * Initialise the state of a blockdev folio's buffers. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1002 | */ |
Matthew Wilcox (Oracle) | 6f24ce6 | 2023-06-12 22:01:38 +0100 | [diff] [blame] | 1003 | static sector_t folio_init_buffers(struct folio *folio, |
Matthew Wilcox (Oracle) | 382497a | 2023-11-09 21:06:03 +0000 | [diff] [blame] | 1004 | struct block_device *bdev, unsigned size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1005 | { |
Matthew Wilcox (Oracle) | 6f24ce6 | 2023-06-12 22:01:38 +0100 | [diff] [blame] | 1006 | struct buffer_head *head = folio_buffers(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1007 | struct buffer_head *bh = head; |
Matthew Wilcox (Oracle) | 6f24ce6 | 2023-06-12 22:01:38 +0100 | [diff] [blame] | 1008 | bool uptodate = folio_test_uptodate(folio); |
Matthew Wilcox (Oracle) | 382497a | 2023-11-09 21:06:03 +0000 | [diff] [blame] | 1009 | sector_t block = div_u64(folio_pos(folio), size); |
Christoph Hellwig | bcd1d06 | 2021-10-18 12:11:10 +0200 | [diff] [blame] | 1010 | sector_t end_block = blkdev_max_block(bdev, size); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1011 | |
| 1012 | do { |
| 1013 | if (!buffer_mapped(bh)) { |
Eric Biggers | 01950a3 | 2018-01-16 22:25:12 -0800 | [diff] [blame] | 1014 | bh->b_end_io = NULL; |
| 1015 | bh->b_private = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1016 | bh->b_bdev = bdev; |
| 1017 | bh->b_blocknr = block; |
| 1018 | if (uptodate) |
| 1019 | set_buffer_uptodate(bh); |
Jeff Moyer | 080399a | 2012-05-11 16:34:10 +0200 | [diff] [blame] | 1020 | if (block < end_block) |
| 1021 | set_buffer_mapped(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1022 | } |
| 1023 | block++; |
| 1024 | bh = bh->b_this_page; |
| 1025 | } while (bh != head); |
Hugh Dickins | 676ce6d | 2012-08-23 12:17:36 +0200 | [diff] [blame] | 1026 | |
| 1027 | /* |
| 1028 | * Caller needs to validate requested block against end of device. |
| 1029 | */ |
| 1030 | return end_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1031 | } |
| 1032 | |
| 1033 | /* |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1034 | * Create the page-cache folio that contains the requested block. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1035 | * |
Hugh Dickins | 676ce6d | 2012-08-23 12:17:36 +0200 | [diff] [blame] | 1036 | * This is used purely for blockdev mappings. |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1037 | * |
Matthew Wilcox (Oracle) | bcd30d4 | 2024-01-01 09:38:48 +0000 | [diff] [blame] | 1038 | * Returns false if we have a failure which cannot be cured by retrying |
| 1039 | * without sleeping. Returns true if we succeeded, or the caller should retry. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1040 | */ |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1041 | static bool grow_dev_folio(struct block_device *bdev, sector_t block, |
Matthew Wilcox (Oracle) | 382497a | 2023-11-09 21:06:03 +0000 | [diff] [blame] | 1042 | pgoff_t index, unsigned size, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1043 | { |
Al Viro | 22f89a4 | 2024-04-11 15:53:38 +0100 | [diff] [blame] | 1044 | struct address_space *mapping = bdev->bd_mapping; |
Matthew Wilcox (Oracle) | 3c98a41 | 2023-06-12 22:01:37 +0100 | [diff] [blame] | 1045 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1046 | struct buffer_head *bh; |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1047 | sector_t end_block = 0; |
Johannes Weiner | 84235de | 2013-10-16 13:47:00 -0700 | [diff] [blame] | 1048 | |
Al Viro | 22f89a4 | 2024-04-11 15:53:38 +0100 | [diff] [blame] | 1049 | folio = __filemap_get_folio(mapping, index, |
Matthew Wilcox (Oracle) | 3ed65f0 | 2023-09-14 16:00:05 +0100 | [diff] [blame] | 1050 | FGP_LOCK | FGP_ACCESSED | FGP_CREAT, gfp); |
| 1051 | if (IS_ERR(folio)) |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1052 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1053 | |
Matthew Wilcox (Oracle) | 3c98a41 | 2023-06-12 22:01:37 +0100 | [diff] [blame] | 1054 | bh = folio_buffers(folio); |
| 1055 | if (bh) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1056 | if (bh->b_size == size) { |
Matthew Wilcox (Oracle) | 382497a | 2023-11-09 21:06:03 +0000 | [diff] [blame] | 1057 | end_block = folio_init_buffers(folio, bdev, size); |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1058 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1059 | } |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1060 | |
Matthew Wilcox (Oracle) | bcd30d4 | 2024-01-01 09:38:48 +0000 | [diff] [blame] | 1061 | /* |
| 1062 | * Retrying may succeed; for example the folio may finish |
| 1063 | * writeback, or buffers may be cleaned. This should not |
| 1064 | * happen very often; maybe we have old buffers attached to |
| 1065 | * this blockdev's page cache and we're trying to change |
| 1066 | * the block size? |
| 1067 | */ |
| 1068 | if (!try_to_free_buffers(folio)) { |
| 1069 | end_block = ~0ULL; |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1070 | goto unlock; |
Matthew Wilcox (Oracle) | bcd30d4 | 2024-01-01 09:38:48 +0000 | [diff] [blame] | 1071 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1072 | } |
| 1073 | |
Matthew Wilcox (Oracle) | 3ed65f0 | 2023-09-14 16:00:05 +0100 | [diff] [blame] | 1074 | bh = folio_alloc_buffers(folio, size, gfp | __GFP_ACCOUNT); |
| 1075 | if (!bh) |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1076 | goto unlock; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1077 | |
| 1078 | /* |
Matthew Wilcox (Oracle) | 3c98a41 | 2023-06-12 22:01:37 +0100 | [diff] [blame] | 1079 | * Link the folio to the buffers and initialise them. Take the |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1080 | * lock to be atomic wrt __find_get_block(), which does not |
Matthew Wilcox (Oracle) | 3c98a41 | 2023-06-12 22:01:37 +0100 | [diff] [blame] | 1081 | * run under the folio lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1082 | */ |
Al Viro | 22f89a4 | 2024-04-11 15:53:38 +0100 | [diff] [blame] | 1083 | spin_lock(&mapping->i_private_lock); |
Matthew Wilcox (Oracle) | 08d84ad | 2023-06-12 22:01:39 +0100 | [diff] [blame] | 1084 | link_dev_buffers(folio, bh); |
Matthew Wilcox (Oracle) | 382497a | 2023-11-09 21:06:03 +0000 | [diff] [blame] | 1085 | end_block = folio_init_buffers(folio, bdev, size); |
Al Viro | 22f89a4 | 2024-04-11 15:53:38 +0100 | [diff] [blame] | 1086 | spin_unlock(&mapping->i_private_lock); |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1087 | unlock: |
Matthew Wilcox (Oracle) | 3c98a41 | 2023-06-12 22:01:37 +0100 | [diff] [blame] | 1088 | folio_unlock(folio); |
| 1089 | folio_put(folio); |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1090 | return block < end_block; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1091 | } |
| 1092 | |
| 1093 | /* |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1094 | * Create buffers for the specified block device block's folio. If |
| 1095 | * that folio was dirty, the buffers are set dirty also. Returns false |
| 1096 | * if we've hit a permanent error. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1097 | */ |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1098 | static bool grow_buffers(struct block_device *bdev, sector_t block, |
| 1099 | unsigned size, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1100 | { |
Matthew Wilcox (Oracle) | 5f3bd90 | 2023-11-09 21:06:04 +0000 | [diff] [blame] | 1101 | loff_t pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1102 | |
Andrew Morton | e565793 | 2006-10-11 01:21:46 -0700 | [diff] [blame] | 1103 | /* |
Matthew Wilcox (Oracle) | 5f3bd90 | 2023-11-09 21:06:04 +0000 | [diff] [blame] | 1104 | * Check for a block which lies outside our maximum possible |
| 1105 | * pagecache index. |
Andrew Morton | e565793 | 2006-10-11 01:21:46 -0700 | [diff] [blame] | 1106 | */ |
Matthew Wilcox (Oracle) | 5f3bd90 | 2023-11-09 21:06:04 +0000 | [diff] [blame] | 1107 | if (check_mul_overflow(block, (sector_t)size, &pos) || pos > MAX_LFS_FILESIZE) { |
| 1108 | printk(KERN_ERR "%s: requested out-of-range block %llu for device %pg\n", |
Harvey Harrison | 8e24eea | 2008-04-30 00:55:09 -0700 | [diff] [blame] | 1109 | __func__, (unsigned long long)block, |
Dmitry Monakhov | a1c6f057 | 2015-04-13 16:31:37 +0400 | [diff] [blame] | 1110 | bdev); |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1111 | return false; |
Andrew Morton | e565793 | 2006-10-11 01:21:46 -0700 | [diff] [blame] | 1112 | } |
Hugh Dickins | 676ce6d | 2012-08-23 12:17:36 +0200 | [diff] [blame] | 1113 | |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1114 | /* Create a folio with the proper size buffers */ |
Matthew Wilcox (Oracle) | 5f3bd90 | 2023-11-09 21:06:04 +0000 | [diff] [blame] | 1115 | return grow_dev_folio(bdev, block, pos / PAGE_SIZE, size, gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1116 | } |
| 1117 | |
Eric Biggers | 0026ba4 | 2016-09-12 13:30:41 -0700 | [diff] [blame] | 1118 | static struct buffer_head * |
Gioh Kim | 3b5e645 | 2014-09-04 22:04:42 -0400 | [diff] [blame] | 1119 | __getblk_slow(struct block_device *bdev, sector_t block, |
| 1120 | unsigned size, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1121 | { |
| 1122 | /* Size must be multiple of hard sectorsize */ |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 1123 | if (unlikely(size & (bdev_logical_block_size(bdev)-1) || |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1124 | (size < 512 || size > PAGE_SIZE))) { |
| 1125 | printk(KERN_ERR "getblk(): invalid block size %d requested\n", |
| 1126 | size); |
Martin K. Petersen | e1defc4 | 2009-05-22 17:17:49 -0400 | [diff] [blame] | 1127 | printk(KERN_ERR "logical block size: %d\n", |
| 1128 | bdev_logical_block_size(bdev)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1129 | |
| 1130 | dump_stack(); |
| 1131 | return NULL; |
| 1132 | } |
| 1133 | |
Hugh Dickins | 676ce6d | 2012-08-23 12:17:36 +0200 | [diff] [blame] | 1134 | for (;;) { |
| 1135 | struct buffer_head *bh; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1136 | |
| 1137 | bh = __find_get_block(bdev, block, size); |
| 1138 | if (bh) |
| 1139 | return bh; |
Hugh Dickins | 676ce6d | 2012-08-23 12:17:36 +0200 | [diff] [blame] | 1140 | |
Matthew Wilcox (Oracle) | 6d840a1 | 2023-11-09 21:06:02 +0000 | [diff] [blame] | 1141 | if (!grow_buffers(bdev, block, size, gfp)) |
Hugh Dickins | 676ce6d | 2012-08-23 12:17:36 +0200 | [diff] [blame] | 1142 | return NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1143 | } |
| 1144 | } |
| 1145 | |
| 1146 | /* |
| 1147 | * The relationship between dirty buffers and dirty pages: |
| 1148 | * |
| 1149 | * Whenever a page has any dirty buffers, the page's dirty bit is set, and |
Matthew Wilcox | ec82e1c | 2017-12-04 10:40:41 -0500 | [diff] [blame] | 1150 | * the page is tagged dirty in the page cache. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1151 | * |
| 1152 | * At all times, the dirtiness of the buffers represents the dirtiness of |
| 1153 | * subsections of the page. If the page has buffers, the page dirty bit is |
| 1154 | * merely a hint about the true dirty state. |
| 1155 | * |
| 1156 | * When a page is set dirty in its entirety, all its buffers are marked dirty |
| 1157 | * (if the page has buffers). |
| 1158 | * |
| 1159 | * When a buffer is marked dirty, its page is dirtied, but the page's other |
| 1160 | * buffers are not. |
| 1161 | * |
| 1162 | * Also. When blockdev buffers are explicitly read with bread(), they |
| 1163 | * individually become uptodate. But their backing page remains not |
| 1164 | * uptodate - even if all of its buffers are uptodate. A subsequent |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 1165 | * block_read_full_folio() against that folio will discover all the uptodate |
| 1166 | * buffers, will set the folio uptodate and will perform no I/O. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1167 | */ |
| 1168 | |
| 1169 | /** |
| 1170 | * mark_buffer_dirty - mark a buffer_head as needing writeout |
Martin Waitz | 67be2dd | 2005-05-01 08:59:26 -0700 | [diff] [blame] | 1171 | * @bh: the buffer_head to mark dirty |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1172 | * |
Matthew Wilcox | ec82e1c | 2017-12-04 10:40:41 -0500 | [diff] [blame] | 1173 | * mark_buffer_dirty() will set the dirty bit against the buffer, then set |
| 1174 | * its backing page dirty, then tag the page as dirty in the page cache |
| 1175 | * and then attach the address_space's inode to its superblock's dirty |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1176 | * inode list. |
| 1177 | * |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1178 | * mark_buffer_dirty() is atomic. It takes bh->b_folio->mapping->i_private_lock, |
Matthew Wilcox | b93b016 | 2018-04-10 16:36:56 -0700 | [diff] [blame] | 1179 | * i_pages lock and mapping->host->i_lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1180 | */ |
Harvey Harrison | fc9b52c | 2008-02-08 04:19:52 -0800 | [diff] [blame] | 1181 | void mark_buffer_dirty(struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1182 | { |
Nick Piggin | 787d2214 | 2007-07-17 04:03:34 -0700 | [diff] [blame] | 1183 | WARN_ON_ONCE(!buffer_uptodate(bh)); |
Linus Torvalds | 1be62dc | 2008-04-04 14:38:17 -0700 | [diff] [blame] | 1184 | |
Tejun Heo | 5305cb8 | 2013-01-11 13:06:36 -0800 | [diff] [blame] | 1185 | trace_block_dirty_buffer(bh); |
| 1186 | |
Linus Torvalds | 1be62dc | 2008-04-04 14:38:17 -0700 | [diff] [blame] | 1187 | /* |
| 1188 | * Very *carefully* optimize the it-is-already-dirty case. |
| 1189 | * |
| 1190 | * Don't let the final "is it dirty" escape to before we |
| 1191 | * perhaps modified the buffer. |
| 1192 | */ |
| 1193 | if (buffer_dirty(bh)) { |
| 1194 | smp_mb(); |
| 1195 | if (buffer_dirty(bh)) |
| 1196 | return; |
| 1197 | } |
| 1198 | |
Linus Torvalds | a8e7d49 | 2009-03-19 11:32:05 -0700 | [diff] [blame] | 1199 | if (!test_set_buffer_dirty(bh)) { |
Matthew Wilcox (Oracle) | cf1d341 | 2022-12-15 21:43:57 +0000 | [diff] [blame] | 1200 | struct folio *folio = bh->b_folio; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1201 | struct address_space *mapping = NULL; |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1202 | |
Matthew Wilcox (Oracle) | cf1d341 | 2022-12-15 21:43:57 +0000 | [diff] [blame] | 1203 | folio_memcg_lock(folio); |
| 1204 | if (!folio_test_set_dirty(folio)) { |
| 1205 | mapping = folio->mapping; |
Linus Torvalds | 8e9d78e | 2009-08-21 17:40:08 -0700 | [diff] [blame] | 1206 | if (mapping) |
Matthew Wilcox (Oracle) | cf1d341 | 2022-12-15 21:43:57 +0000 | [diff] [blame] | 1207 | __folio_mark_dirty(folio, mapping, 0); |
Linus Torvalds | 8e9d78e | 2009-08-21 17:40:08 -0700 | [diff] [blame] | 1208 | } |
Matthew Wilcox (Oracle) | cf1d341 | 2022-12-15 21:43:57 +0000 | [diff] [blame] | 1209 | folio_memcg_unlock(folio); |
Greg Thelen | c4843a7 | 2015-05-22 17:13:16 -0400 | [diff] [blame] | 1210 | if (mapping) |
| 1211 | __mark_inode_dirty(mapping->host, I_DIRTY_PAGES); |
Linus Torvalds | a8e7d49 | 2009-03-19 11:32:05 -0700 | [diff] [blame] | 1212 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1213 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 1214 | EXPORT_SYMBOL(mark_buffer_dirty); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1215 | |
Jeff Layton | 87354e5 | 2017-07-06 07:02:21 -0400 | [diff] [blame] | 1216 | void mark_buffer_write_io_error(struct buffer_head *bh) |
| 1217 | { |
| 1218 | set_buffer_write_io_error(bh); |
| 1219 | /* FIXME: do we need to set this in both places? */ |
Matthew Wilcox (Oracle) | abc8a8a | 2022-12-15 21:43:52 +0000 | [diff] [blame] | 1220 | if (bh->b_folio && bh->b_folio->mapping) |
| 1221 | mapping_set_error(bh->b_folio->mapping, -EIO); |
Christoph Hellwig | 4b2201d | 2023-08-07 12:26:22 +0100 | [diff] [blame] | 1222 | if (bh->b_assoc_map) { |
Jeff Layton | 87354e5 | 2017-07-06 07:02:21 -0400 | [diff] [blame] | 1223 | mapping_set_error(bh->b_assoc_map, -EIO); |
Christoph Hellwig | 4b2201d | 2023-08-07 12:26:22 +0100 | [diff] [blame] | 1224 | errseq_set(&bh->b_assoc_map->host->i_sb->s_wb_err, -EIO); |
| 1225 | } |
Jeff Layton | 87354e5 | 2017-07-06 07:02:21 -0400 | [diff] [blame] | 1226 | } |
| 1227 | EXPORT_SYMBOL(mark_buffer_write_io_error); |
| 1228 | |
Matthew Wilcox (Oracle) | 66924fd | 2024-04-16 04:17:49 +0100 | [diff] [blame] | 1229 | /** |
| 1230 | * __brelse - Release a buffer. |
| 1231 | * @bh: The buffer to release. |
| 1232 | * |
| 1233 | * This variant of brelse() can be called if @bh is guaranteed to not be NULL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1234 | */ |
Matthew Wilcox (Oracle) | 66924fd | 2024-04-16 04:17:49 +0100 | [diff] [blame] | 1235 | void __brelse(struct buffer_head *bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1236 | { |
Matthew Wilcox (Oracle) | 66924fd | 2024-04-16 04:17:49 +0100 | [diff] [blame] | 1237 | if (atomic_read(&bh->b_count)) { |
| 1238 | put_bh(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1239 | return; |
| 1240 | } |
Arjan van de Ven | 5c752ad | 2008-07-25 19:45:40 -0700 | [diff] [blame] | 1241 | WARN(1, KERN_ERR "VFS: brelse: Trying to free free buffer\n"); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1242 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 1243 | EXPORT_SYMBOL(__brelse); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1244 | |
Matthew Wilcox (Oracle) | b73a936 | 2024-04-16 04:17:50 +0100 | [diff] [blame] | 1245 | /** |
| 1246 | * __bforget - Discard any dirty data in a buffer. |
| 1247 | * @bh: The buffer to forget. |
| 1248 | * |
| 1249 | * This variant of bforget() can be called if @bh is guaranteed to not |
| 1250 | * be NULL. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1251 | */ |
| 1252 | void __bforget(struct buffer_head *bh) |
| 1253 | { |
| 1254 | clear_buffer_dirty(bh); |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 1255 | if (bh->b_assoc_map) { |
Matthew Wilcox (Oracle) | abc8a8a | 2022-12-15 21:43:52 +0000 | [diff] [blame] | 1256 | struct address_space *buffer_mapping = bh->b_folio->mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1257 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1258 | spin_lock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1259 | list_del_init(&bh->b_assoc_buffers); |
Jan Kara | 58ff407b | 2006-10-17 00:10:19 -0700 | [diff] [blame] | 1260 | bh->b_assoc_map = NULL; |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1261 | spin_unlock(&buffer_mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1262 | } |
| 1263 | __brelse(bh); |
| 1264 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 1265 | EXPORT_SYMBOL(__bforget); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1266 | |
| 1267 | static struct buffer_head *__bread_slow(struct buffer_head *bh) |
| 1268 | { |
| 1269 | lock_buffer(bh); |
| 1270 | if (buffer_uptodate(bh)) { |
| 1271 | unlock_buffer(bh); |
| 1272 | return bh; |
| 1273 | } else { |
| 1274 | get_bh(bh); |
| 1275 | bh->b_end_io = end_buffer_read_sync; |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 1276 | submit_bh(REQ_OP_READ, bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1277 | wait_on_buffer(bh); |
| 1278 | if (buffer_uptodate(bh)) |
| 1279 | return bh; |
| 1280 | } |
| 1281 | brelse(bh); |
| 1282 | return NULL; |
| 1283 | } |
| 1284 | |
| 1285 | /* |
| 1286 | * Per-cpu buffer LRU implementation. To reduce the cost of __find_get_block(). |
| 1287 | * The bhs[] array is sorted - newest buffer is at bhs[0]. Buffers have their |
| 1288 | * refcount elevated by one when they're in an LRU. A buffer can only appear |
| 1289 | * once in a particular CPU's LRU. A single buffer can be present in multiple |
| 1290 | * CPU's LRUs at the same time. |
| 1291 | * |
| 1292 | * This is a transparent caching front-end to sb_bread(), sb_getblk() and |
| 1293 | * sb_find_get_block(). |
| 1294 | * |
| 1295 | * The LRUs themselves only need locking against invalidate_bh_lrus. We use |
| 1296 | * a local interrupt disable for that. |
| 1297 | */ |
| 1298 | |
Sebastien Buisson | 86cf78d | 2014-10-09 15:29:38 -0700 | [diff] [blame] | 1299 | #define BH_LRU_SIZE 16 |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1300 | |
| 1301 | struct bh_lru { |
| 1302 | struct buffer_head *bhs[BH_LRU_SIZE]; |
| 1303 | }; |
| 1304 | |
| 1305 | static DEFINE_PER_CPU(struct bh_lru, bh_lrus) = {{ NULL }}; |
| 1306 | |
| 1307 | #ifdef CONFIG_SMP |
| 1308 | #define bh_lru_lock() local_irq_disable() |
| 1309 | #define bh_lru_unlock() local_irq_enable() |
| 1310 | #else |
| 1311 | #define bh_lru_lock() preempt_disable() |
| 1312 | #define bh_lru_unlock() preempt_enable() |
| 1313 | #endif |
| 1314 | |
| 1315 | static inline void check_irqs_on(void) |
| 1316 | { |
| 1317 | #ifdef irqs_disabled |
| 1318 | BUG_ON(irqs_disabled()); |
| 1319 | #endif |
| 1320 | } |
| 1321 | |
| 1322 | /* |
Eric Biggers | 241f01f | 2017-07-10 15:47:29 -0700 | [diff] [blame] | 1323 | * Install a buffer_head into this cpu's LRU. If not already in the LRU, it is |
| 1324 | * inserted at the front, and the buffer_head at the back if any is evicted. |
| 1325 | * Or, if already in the LRU it is moved to the front. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1326 | */ |
| 1327 | static void bh_lru_install(struct buffer_head *bh) |
| 1328 | { |
Eric Biggers | 241f01f | 2017-07-10 15:47:29 -0700 | [diff] [blame] | 1329 | struct buffer_head *evictee = bh; |
| 1330 | struct bh_lru *b; |
| 1331 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1332 | |
| 1333 | check_irqs_on(); |
Minchan Kim | c0226eb | 2022-03-22 14:39:34 -0700 | [diff] [blame] | 1334 | bh_lru_lock(); |
| 1335 | |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1336 | /* |
| 1337 | * the refcount of buffer_head in bh_lru prevents dropping the |
| 1338 | * attached page(i.e., try_to_free_buffers) so it could cause |
| 1339 | * failing page migration. |
| 1340 | * Skip putting upcoming bh into bh_lru until migration is done. |
| 1341 | */ |
Marcelo Tosatti | 8a237ad | 2023-06-27 17:08:15 -0300 | [diff] [blame] | 1342 | if (lru_cache_disabled() || cpu_is_isolated(smp_processor_id())) { |
Minchan Kim | c0226eb | 2022-03-22 14:39:34 -0700 | [diff] [blame] | 1343 | bh_lru_unlock(); |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1344 | return; |
Minchan Kim | c0226eb | 2022-03-22 14:39:34 -0700 | [diff] [blame] | 1345 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1346 | |
Eric Biggers | 241f01f | 2017-07-10 15:47:29 -0700 | [diff] [blame] | 1347 | b = this_cpu_ptr(&bh_lrus); |
| 1348 | for (i = 0; i < BH_LRU_SIZE; i++) { |
| 1349 | swap(evictee, b->bhs[i]); |
| 1350 | if (evictee == bh) { |
| 1351 | bh_lru_unlock(); |
| 1352 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1353 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1354 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1355 | |
Eric Biggers | 241f01f | 2017-07-10 15:47:29 -0700 | [diff] [blame] | 1356 | get_bh(bh); |
| 1357 | bh_lru_unlock(); |
| 1358 | brelse(evictee); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1359 | } |
| 1360 | |
| 1361 | /* |
| 1362 | * Look up the bh in this cpu's LRU. If it's there, move it to the head. |
| 1363 | */ |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 1364 | static struct buffer_head * |
Tomasz Kvarsin | 3991d3b | 2007-02-12 00:52:14 -0800 | [diff] [blame] | 1365 | lookup_bh_lru(struct block_device *bdev, sector_t block, unsigned size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1366 | { |
| 1367 | struct buffer_head *ret = NULL; |
Tomasz Kvarsin | 3991d3b | 2007-02-12 00:52:14 -0800 | [diff] [blame] | 1368 | unsigned int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1369 | |
| 1370 | check_irqs_on(); |
| 1371 | bh_lru_lock(); |
Marcelo Tosatti | 8a237ad | 2023-06-27 17:08:15 -0300 | [diff] [blame] | 1372 | if (cpu_is_isolated(smp_processor_id())) { |
| 1373 | bh_lru_unlock(); |
| 1374 | return NULL; |
| 1375 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1376 | for (i = 0; i < BH_LRU_SIZE; i++) { |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 1377 | struct buffer_head *bh = __this_cpu_read(bh_lrus.bhs[i]); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1378 | |
Zach Brown | 9470dd5 | 2014-10-13 15:55:05 -0700 | [diff] [blame] | 1379 | if (bh && bh->b_blocknr == block && bh->b_bdev == bdev && |
| 1380 | bh->b_size == size) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1381 | if (i) { |
| 1382 | while (i) { |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 1383 | __this_cpu_write(bh_lrus.bhs[i], |
| 1384 | __this_cpu_read(bh_lrus.bhs[i - 1])); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1385 | i--; |
| 1386 | } |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 1387 | __this_cpu_write(bh_lrus.bhs[0], bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1388 | } |
| 1389 | get_bh(bh); |
| 1390 | ret = bh; |
| 1391 | break; |
| 1392 | } |
| 1393 | } |
| 1394 | bh_lru_unlock(); |
| 1395 | return ret; |
| 1396 | } |
| 1397 | |
| 1398 | /* |
| 1399 | * Perform a pagecache lookup for the matching buffer. If it's there, refresh |
| 1400 | * it in the LRU and mark it as accessed. If it is not present then return |
| 1401 | * NULL |
| 1402 | */ |
| 1403 | struct buffer_head * |
Tomasz Kvarsin | 3991d3b | 2007-02-12 00:52:14 -0800 | [diff] [blame] | 1404 | __find_get_block(struct block_device *bdev, sector_t block, unsigned size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1405 | { |
| 1406 | struct buffer_head *bh = lookup_bh_lru(bdev, block, size); |
| 1407 | |
| 1408 | if (bh == NULL) { |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1409 | /* __find_get_block_slow will mark the page accessed */ |
Coywolf Qi Hunt | 385fd4c | 2005-11-07 00:59:39 -0800 | [diff] [blame] | 1410 | bh = __find_get_block_slow(bdev, block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1411 | if (bh) |
| 1412 | bh_lru_install(bh); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1413 | } else |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1414 | touch_buffer(bh); |
Mel Gorman | 2457aec | 2014-06-04 16:10:31 -0700 | [diff] [blame] | 1415 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1416 | return bh; |
| 1417 | } |
| 1418 | EXPORT_SYMBOL(__find_get_block); |
| 1419 | |
Matthew Wilcox (Oracle) | 3ed65f0 | 2023-09-14 16:00:05 +0100 | [diff] [blame] | 1420 | /** |
| 1421 | * bdev_getblk - Get a buffer_head in a block device's buffer cache. |
| 1422 | * @bdev: The block device. |
| 1423 | * @block: The block number. |
| 1424 | * @size: The size of buffer_heads for this @bdev. |
| 1425 | * @gfp: The memory allocation flags to use. |
| 1426 | * |
Matthew Wilcox (Oracle) | 0b116ff | 2024-04-16 04:17:51 +0100 | [diff] [blame] | 1427 | * The returned buffer head has its reference count incremented, but is |
| 1428 | * not locked. The caller should call brelse() when it has finished |
| 1429 | * with the buffer. The buffer may not be uptodate. If needed, the |
| 1430 | * caller can bring it uptodate either by reading it or overwriting it. |
| 1431 | * |
Matthew Wilcox (Oracle) | 3ed65f0 | 2023-09-14 16:00:05 +0100 | [diff] [blame] | 1432 | * Return: The buffer head, or NULL if memory could not be allocated. |
| 1433 | */ |
| 1434 | struct buffer_head *bdev_getblk(struct block_device *bdev, sector_t block, |
| 1435 | unsigned size, gfp_t gfp) |
| 1436 | { |
| 1437 | struct buffer_head *bh = __find_get_block(bdev, block, size); |
| 1438 | |
| 1439 | might_alloc(gfp); |
| 1440 | if (bh) |
| 1441 | return bh; |
| 1442 | |
| 1443 | return __getblk_slow(bdev, block, size, gfp); |
| 1444 | } |
| 1445 | EXPORT_SYMBOL(bdev_getblk); |
| 1446 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1447 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1448 | * Do async read-ahead on a buffer.. |
| 1449 | */ |
Tomasz Kvarsin | 3991d3b | 2007-02-12 00:52:14 -0800 | [diff] [blame] | 1450 | void __breadahead(struct block_device *bdev, sector_t block, unsigned size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1451 | { |
Matthew Wilcox (Oracle) | 775d9b1 | 2023-09-14 16:00:07 +0100 | [diff] [blame] | 1452 | struct buffer_head *bh = bdev_getblk(bdev, block, size, |
| 1453 | GFP_NOWAIT | __GFP_MOVABLE); |
| 1454 | |
Andrew Morton | a3e713b | 2005-10-30 15:03:15 -0800 | [diff] [blame] | 1455 | if (likely(bh)) { |
Zhang Yi | e7ea112 | 2022-09-01 21:34:54 +0800 | [diff] [blame] | 1456 | bh_readahead(bh, REQ_RAHEAD); |
Andrew Morton | a3e713b | 2005-10-30 15:03:15 -0800 | [diff] [blame] | 1457 | brelse(bh); |
| 1458 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1459 | } |
| 1460 | EXPORT_SYMBOL(__breadahead); |
| 1461 | |
| 1462 | /** |
Matthew Wilcox (Oracle) | 324ecae | 2024-04-16 04:17:48 +0100 | [diff] [blame] | 1463 | * __bread_gfp() - Read a block. |
| 1464 | * @bdev: The block device to read from. |
| 1465 | * @block: Block number in units of block size. |
| 1466 | * @size: The block size of this device in bytes. |
| 1467 | * @gfp: Not page allocation flags; see below. |
Gioh Kim | 3b5e645 | 2014-09-04 22:04:42 -0400 | [diff] [blame] | 1468 | * |
Matthew Wilcox (Oracle) | 324ecae | 2024-04-16 04:17:48 +0100 | [diff] [blame] | 1469 | * You are not expected to call this function. You should use one of |
| 1470 | * sb_bread(), sb_bread_unmovable() or __bread(). |
| 1471 | * |
| 1472 | * Read a specified block, and return the buffer head that refers to it. |
| 1473 | * If @gfp is 0, the memory will be allocated using the block device's |
| 1474 | * default GFP flags. If @gfp is __GFP_MOVABLE, the memory may be |
| 1475 | * allocated from a movable area. Do not pass in a complete set of |
| 1476 | * GFP flags. |
| 1477 | * |
| 1478 | * The returned buffer head has its refcount increased. The caller should |
| 1479 | * call brelse() when it has finished with the buffer. |
| 1480 | * |
| 1481 | * Context: May sleep waiting for I/O. |
| 1482 | * Return: NULL if the block was unreadable. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1483 | */ |
Matthew Wilcox (Oracle) | 324ecae | 2024-04-16 04:17:48 +0100 | [diff] [blame] | 1484 | struct buffer_head *__bread_gfp(struct block_device *bdev, sector_t block, |
| 1485 | unsigned size, gfp_t gfp) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1486 | { |
Matthew Wilcox (Oracle) | 93b13ec | 2023-09-14 16:00:11 +0100 | [diff] [blame] | 1487 | struct buffer_head *bh; |
| 1488 | |
Al Viro | 224941e | 2024-04-11 15:53:37 +0100 | [diff] [blame] | 1489 | gfp |= mapping_gfp_constraint(bdev->bd_mapping, ~__GFP_FS); |
Matthew Wilcox (Oracle) | 93b13ec | 2023-09-14 16:00:11 +0100 | [diff] [blame] | 1490 | |
| 1491 | /* |
| 1492 | * Prefer looping in the allocator rather than here, at least that |
| 1493 | * code knows what it's doing. |
| 1494 | */ |
| 1495 | gfp |= __GFP_NOFAIL; |
| 1496 | |
| 1497 | bh = bdev_getblk(bdev, block, size, gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1498 | |
Andrew Morton | a3e713b | 2005-10-30 15:03:15 -0800 | [diff] [blame] | 1499 | if (likely(bh) && !buffer_uptodate(bh)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1500 | bh = __bread_slow(bh); |
| 1501 | return bh; |
| 1502 | } |
Gioh Kim | 3b5e645 | 2014-09-04 22:04:42 -0400 | [diff] [blame] | 1503 | EXPORT_SYMBOL(__bread_gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1504 | |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1505 | static void __invalidate_bh_lrus(struct bh_lru *b) |
| 1506 | { |
| 1507 | int i; |
| 1508 | |
| 1509 | for (i = 0; i < BH_LRU_SIZE; i++) { |
| 1510 | brelse(b->bhs[i]); |
| 1511 | b->bhs[i] = NULL; |
| 1512 | } |
| 1513 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1514 | /* |
| 1515 | * invalidate_bh_lrus() is called rarely - but not only at unmount. |
| 1516 | * This doesn't race because it runs in each cpu either in irq |
| 1517 | * or with preempt disabled. |
| 1518 | */ |
| 1519 | static void invalidate_bh_lru(void *arg) |
| 1520 | { |
| 1521 | struct bh_lru *b = &get_cpu_var(bh_lrus); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1522 | |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1523 | __invalidate_bh_lrus(b); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1524 | put_cpu_var(bh_lrus); |
| 1525 | } |
Gilad Ben-Yossef | 42be35d | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 1526 | |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1527 | bool has_bh_in_lru(int cpu, void *dummy) |
Gilad Ben-Yossef | 42be35d | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 1528 | { |
| 1529 | struct bh_lru *b = per_cpu_ptr(&bh_lrus, cpu); |
| 1530 | int i; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1531 | |
Gilad Ben-Yossef | 42be35d | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 1532 | for (i = 0; i < BH_LRU_SIZE; i++) { |
| 1533 | if (b->bhs[i]) |
Saurav Girepunje | 1d70667 | 2019-11-30 17:49:15 -0800 | [diff] [blame] | 1534 | return true; |
Gilad Ben-Yossef | 42be35d | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 1535 | } |
| 1536 | |
Saurav Girepunje | 1d70667 | 2019-11-30 17:49:15 -0800 | [diff] [blame] | 1537 | return false; |
Gilad Ben-Yossef | 42be35d | 2012-03-28 14:42:45 -0700 | [diff] [blame] | 1538 | } |
| 1539 | |
Peter Zijlstra | f9a1439 | 2007-05-06 14:49:55 -0700 | [diff] [blame] | 1540 | void invalidate_bh_lrus(void) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1541 | { |
Sebastian Andrzej Siewior | cb92315 | 2020-01-17 10:01:37 +0100 | [diff] [blame] | 1542 | on_each_cpu_cond(has_bh_in_lru, invalidate_bh_lru, NULL, 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1543 | } |
Nick Piggin | 9db5579 | 2008-02-08 04:19:49 -0800 | [diff] [blame] | 1544 | EXPORT_SYMBOL_GPL(invalidate_bh_lrus); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1545 | |
Minchan Kim | 243418e | 2021-09-24 15:43:47 -0700 | [diff] [blame] | 1546 | /* |
| 1547 | * It's called from workqueue context so we need a bh_lru_lock to close |
| 1548 | * the race with preemption/irq. |
| 1549 | */ |
| 1550 | void invalidate_bh_lrus_cpu(void) |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1551 | { |
| 1552 | struct bh_lru *b; |
| 1553 | |
| 1554 | bh_lru_lock(); |
Minchan Kim | 243418e | 2021-09-24 15:43:47 -0700 | [diff] [blame] | 1555 | b = this_cpu_ptr(&bh_lrus); |
Minchan Kim | 8cc621d | 2021-05-04 18:37:00 -0700 | [diff] [blame] | 1556 | __invalidate_bh_lrus(b); |
| 1557 | bh_lru_unlock(); |
| 1558 | } |
| 1559 | |
Pankaj Raghav | 465e5e6 | 2023-04-17 14:36:15 +0200 | [diff] [blame] | 1560 | void folio_set_bh(struct buffer_head *bh, struct folio *folio, |
| 1561 | unsigned long offset) |
| 1562 | { |
| 1563 | bh->b_folio = folio; |
| 1564 | BUG_ON(offset >= folio_size(folio)); |
| 1565 | if (folio_test_highmem(folio)) |
| 1566 | /* |
| 1567 | * This catches illegal uses and preserves the offset: |
| 1568 | */ |
| 1569 | bh->b_data = (char *)(0 + offset); |
| 1570 | else |
| 1571 | bh->b_data = folio_address(folio) + offset; |
| 1572 | } |
| 1573 | EXPORT_SYMBOL(folio_set_bh); |
| 1574 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1575 | /* |
| 1576 | * Called when truncating a buffer on a page completely. |
| 1577 | */ |
Mel Gorman | e7470ee | 2014-06-04 16:10:29 -0700 | [diff] [blame] | 1578 | |
| 1579 | /* Bits that are cleared during an invalidate */ |
| 1580 | #define BUFFER_FLAGS_DISCARD \ |
| 1581 | (1 << BH_Mapped | 1 << BH_New | 1 << BH_Req | \ |
| 1582 | 1 << BH_Delay | 1 << BH_Unwritten) |
| 1583 | |
Arjan van de Ven | 858119e | 2006-01-14 13:20:43 -0800 | [diff] [blame] | 1584 | static void discard_buffer(struct buffer_head * bh) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1585 | { |
Uros Bizjak | b019229 | 2022-07-14 19:16:53 +0200 | [diff] [blame] | 1586 | unsigned long b_state; |
Mel Gorman | e7470ee | 2014-06-04 16:10:29 -0700 | [diff] [blame] | 1587 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1588 | lock_buffer(bh); |
| 1589 | clear_buffer_dirty(bh); |
| 1590 | bh->b_bdev = NULL; |
Uros Bizjak | b019229 | 2022-07-14 19:16:53 +0200 | [diff] [blame] | 1591 | b_state = READ_ONCE(bh->b_state); |
| 1592 | do { |
| 1593 | } while (!try_cmpxchg(&bh->b_state, &b_state, |
| 1594 | b_state & ~BUFFER_FLAGS_DISCARD)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1595 | unlock_buffer(bh); |
| 1596 | } |
| 1597 | |
| 1598 | /** |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1599 | * block_invalidate_folio - Invalidate part or all of a buffer-backed folio. |
| 1600 | * @folio: The folio which is affected. |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 1601 | * @offset: start of the range to invalidate |
| 1602 | * @length: length of the range to invalidate |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1603 | * |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1604 | * block_invalidate_folio() is called when all or part of the folio has been |
Wang Sheng-Hui | 814e1d2 | 2011-09-01 08:22:57 +0800 | [diff] [blame] | 1605 | * invalidated by a truncate operation. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1606 | * |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1607 | * block_invalidate_folio() does not have to release all buffers, but it must |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1608 | * ensure that no dirty buffer is left outside @offset and that no I/O |
| 1609 | * is underway against any of the blocks which are outside the truncation |
| 1610 | * point. Because the caller is about to free (and possibly reuse) those |
| 1611 | * blocks on-disk. |
| 1612 | */ |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1613 | void block_invalidate_folio(struct folio *folio, size_t offset, size_t length) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1614 | { |
| 1615 | struct buffer_head *head, *bh, *next; |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1616 | size_t curr_off = 0; |
| 1617 | size_t stop = length + offset; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1618 | |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1619 | BUG_ON(!folio_test_locked(folio)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1620 | |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 1621 | /* |
| 1622 | * Check for overflow |
| 1623 | */ |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1624 | BUG_ON(stop > folio_size(folio) || stop < length); |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 1625 | |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1626 | head = folio_buffers(folio); |
| 1627 | if (!head) |
| 1628 | return; |
| 1629 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1630 | bh = head; |
| 1631 | do { |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1632 | size_t next_off = curr_off + bh->b_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1633 | next = bh->b_this_page; |
| 1634 | |
| 1635 | /* |
Lukas Czerner | d47992f | 2013-05-21 23:17:23 -0400 | [diff] [blame] | 1636 | * Are we still fully in range ? |
| 1637 | */ |
| 1638 | if (next_off > stop) |
| 1639 | goto out; |
| 1640 | |
| 1641 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1642 | * is this block fully invalidated? |
| 1643 | */ |
| 1644 | if (offset <= curr_off) |
| 1645 | discard_buffer(bh); |
| 1646 | curr_off = next_off; |
| 1647 | bh = next; |
| 1648 | } while (bh != head); |
| 1649 | |
| 1650 | /* |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1651 | * We release buffers only if the entire folio is being invalidated. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1652 | * The get_block cached value has been unconditionally invalidated, |
| 1653 | * so real IO is not possible anymore. |
| 1654 | */ |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1655 | if (length == folio_size(folio)) |
| 1656 | filemap_release_folio(folio, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1657 | out: |
NeilBrown | 2ff28e2 | 2006-03-26 01:37:18 -0800 | [diff] [blame] | 1658 | return; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1659 | } |
Matthew Wilcox (Oracle) | 7ba13ab | 2022-02-09 20:21:34 +0000 | [diff] [blame] | 1660 | EXPORT_SYMBOL(block_invalidate_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1661 | |
| 1662 | /* |
| 1663 | * We attach and possibly dirty the buffers atomically wrt |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1664 | * block_dirty_folio() via i_private_lock. try_to_free_buffers |
Pankaj Raghav | 8e2e175 | 2023-04-17 14:36:17 +0200 | [diff] [blame] | 1665 | * is already excluded via the folio lock. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1666 | */ |
Matthew Wilcox (Oracle) | 0a88810 | 2023-10-16 21:11:14 +0100 | [diff] [blame] | 1667 | struct buffer_head *create_empty_buffers(struct folio *folio, |
Matthew Wilcox (Oracle) | 3decb85 | 2023-10-16 21:10:49 +0100 | [diff] [blame] | 1668 | unsigned long blocksize, unsigned long b_state) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1669 | { |
| 1670 | struct buffer_head *bh, *head, *tail; |
Matthew Wilcox (Oracle) | 2a41815 | 2023-09-14 16:00:04 +0100 | [diff] [blame] | 1671 | gfp_t gfp = GFP_NOFS | __GFP_ACCOUNT | __GFP_NOFAIL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1672 | |
Matthew Wilcox (Oracle) | 2a41815 | 2023-09-14 16:00:04 +0100 | [diff] [blame] | 1673 | head = folio_alloc_buffers(folio, blocksize, gfp); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1674 | bh = head; |
| 1675 | do { |
| 1676 | bh->b_state |= b_state; |
| 1677 | tail = bh; |
| 1678 | bh = bh->b_this_page; |
| 1679 | } while (bh); |
| 1680 | tail->b_this_page = head; |
| 1681 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1682 | spin_lock(&folio->mapping->i_private_lock); |
Pankaj Raghav | 8e2e175 | 2023-04-17 14:36:17 +0200 | [diff] [blame] | 1683 | if (folio_test_uptodate(folio) || folio_test_dirty(folio)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1684 | bh = head; |
| 1685 | do { |
Pankaj Raghav | 8e2e175 | 2023-04-17 14:36:17 +0200 | [diff] [blame] | 1686 | if (folio_test_dirty(folio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1687 | set_buffer_dirty(bh); |
Pankaj Raghav | 8e2e175 | 2023-04-17 14:36:17 +0200 | [diff] [blame] | 1688 | if (folio_test_uptodate(folio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1689 | set_buffer_uptodate(bh); |
| 1690 | bh = bh->b_this_page; |
| 1691 | } while (bh != head); |
| 1692 | } |
Pankaj Raghav | 8e2e175 | 2023-04-17 14:36:17 +0200 | [diff] [blame] | 1693 | folio_attach_private(folio, head); |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1694 | spin_unlock(&folio->mapping->i_private_lock); |
Matthew Wilcox (Oracle) | 3decb85 | 2023-10-16 21:10:49 +0100 | [diff] [blame] | 1695 | |
| 1696 | return head; |
Pankaj Raghav | 8e2e175 | 2023-04-17 14:36:17 +0200 | [diff] [blame] | 1697 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1698 | EXPORT_SYMBOL(create_empty_buffers); |
| 1699 | |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1700 | /** |
| 1701 | * clean_bdev_aliases: clean a range of buffers in block device |
| 1702 | * @bdev: Block device to clean buffers in |
| 1703 | * @block: Start of a range of blocks to clean |
| 1704 | * @len: Number of blocks to clean |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1705 | * |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1706 | * We are taking a range of blocks for data and we don't want writeback of any |
| 1707 | * buffer-cache aliases starting from return from this function and until the |
| 1708 | * moment when something will explicitly mark the buffer dirty (hopefully that |
| 1709 | * will not happen until we will free that block ;-) We don't even need to mark |
| 1710 | * it not-uptodate - nobody can expect anything from a newly allocated buffer |
| 1711 | * anyway. We used to use unmap_buffer() for such invalidation, but that was |
| 1712 | * wrong. We definitely don't want to mark the alias unmapped, for example - it |
| 1713 | * would confuse anyone who might pick it with bread() afterwards... |
| 1714 | * |
| 1715 | * Also.. Note that bforget() doesn't lock the buffer. So there can be |
| 1716 | * writeout I/O going on against recently-freed buffers. We don't wait on that |
| 1717 | * I/O in bforget() - it's more efficient to wait on the I/O only if we really |
| 1718 | * need to. That happens here. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1719 | */ |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1720 | void clean_bdev_aliases(struct block_device *bdev, sector_t block, sector_t len) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1721 | { |
Al Viro | 53cd4cd | 2024-04-28 19:41:13 -0400 | [diff] [blame] | 1722 | struct address_space *bd_mapping = bdev->bd_mapping; |
| 1723 | const int blkbits = bd_mapping->host->i_blkbits; |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1724 | struct folio_batch fbatch; |
Al Viro | 53cd4cd | 2024-04-28 19:41:13 -0400 | [diff] [blame] | 1725 | pgoff_t index = ((loff_t)block << blkbits) / PAGE_SIZE; |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1726 | pgoff_t end; |
Jan Kara | c10f778 | 2017-09-06 16:21:24 -0700 | [diff] [blame] | 1727 | int i, count; |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1728 | struct buffer_head *bh; |
| 1729 | struct buffer_head *head; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1730 | |
Al Viro | 53cd4cd | 2024-04-28 19:41:13 -0400 | [diff] [blame] | 1731 | end = ((loff_t)(block + len - 1) << blkbits) / PAGE_SIZE; |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1732 | folio_batch_init(&fbatch); |
| 1733 | while (filemap_get_folios(bd_mapping, &index, end, &fbatch)) { |
| 1734 | count = folio_batch_count(&fbatch); |
Jan Kara | c10f778 | 2017-09-06 16:21:24 -0700 | [diff] [blame] | 1735 | for (i = 0; i < count; i++) { |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1736 | struct folio *folio = fbatch.folios[i]; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1737 | |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1738 | if (!folio_buffers(folio)) |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1739 | continue; |
| 1740 | /* |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 1741 | * We use folio lock instead of bd_mapping->i_private_lock |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1742 | * to pin buffers here since we can afford to sleep and |
| 1743 | * it scales better than a global spinlock lock. |
| 1744 | */ |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1745 | folio_lock(folio); |
| 1746 | /* Recheck when the folio is locked which pins bhs */ |
| 1747 | head = folio_buffers(folio); |
| 1748 | if (!head) |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1749 | goto unlock_page; |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1750 | bh = head; |
| 1751 | do { |
Chandan Rajendra | 6c006a9 | 2016-12-25 19:01:03 +0530 | [diff] [blame] | 1752 | if (!buffer_mapped(bh) || (bh->b_blocknr < block)) |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1753 | goto next; |
| 1754 | if (bh->b_blocknr >= block + len) |
| 1755 | break; |
| 1756 | clear_buffer_dirty(bh); |
| 1757 | wait_on_buffer(bh); |
| 1758 | clear_buffer_req(bh); |
| 1759 | next: |
| 1760 | bh = bh->b_this_page; |
| 1761 | } while (bh != head); |
| 1762 | unlock_page: |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1763 | folio_unlock(folio); |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1764 | } |
Matthew Wilcox (Oracle) | 9e0b6f3 | 2022-06-04 16:24:22 -0400 | [diff] [blame] | 1765 | folio_batch_release(&fbatch); |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1766 | cond_resched(); |
Jan Kara | c10f778 | 2017-09-06 16:21:24 -0700 | [diff] [blame] | 1767 | /* End of range already reached? */ |
| 1768 | if (index > end || !index) |
| 1769 | break; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1770 | } |
| 1771 | } |
Jan Kara | 29f3ad7 | 2016-11-04 18:08:11 +0100 | [diff] [blame] | 1772 | EXPORT_SYMBOL(clean_bdev_aliases); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1773 | |
Pankaj Raghav | c6c8c3e | 2023-04-17 14:36:18 +0200 | [diff] [blame] | 1774 | static struct buffer_head *folio_create_buffers(struct folio *folio, |
| 1775 | struct inode *inode, |
| 1776 | unsigned int b_state) |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 1777 | { |
Matthew Wilcox (Oracle) | 3decb85 | 2023-10-16 21:10:49 +0100 | [diff] [blame] | 1778 | struct buffer_head *bh; |
| 1779 | |
Pankaj Raghav | c6c8c3e | 2023-04-17 14:36:18 +0200 | [diff] [blame] | 1780 | BUG_ON(!folio_test_locked(folio)); |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 1781 | |
Matthew Wilcox (Oracle) | 3decb85 | 2023-10-16 21:10:49 +0100 | [diff] [blame] | 1782 | bh = folio_buffers(folio); |
| 1783 | if (!bh) |
Matthew Wilcox (Oracle) | 0a88810 | 2023-10-16 21:11:14 +0100 | [diff] [blame] | 1784 | bh = create_empty_buffers(folio, |
Matthew Wilcox (Oracle) | 3decb85 | 2023-10-16 21:10:49 +0100 | [diff] [blame] | 1785 | 1 << READ_ONCE(inode->i_blkbits), b_state); |
| 1786 | return bh; |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 1787 | } |
| 1788 | |
| 1789 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1790 | * NOTE! All mapped/uptodate combinations are valid: |
| 1791 | * |
| 1792 | * Mapped Uptodate Meaning |
| 1793 | * |
| 1794 | * No No "unknown" - must do get_block() |
| 1795 | * No Yes "hole" - zero-filled |
| 1796 | * Yes No "allocated" - allocated on disk, not read in |
| 1797 | * Yes Yes "valid" - allocated and up-to-date in memory. |
| 1798 | * |
| 1799 | * "Dirty" is valid only with the last case (mapped+uptodate). |
| 1800 | */ |
| 1801 | |
| 1802 | /* |
Matthew Wilcox (Oracle) | 17bf23a | 2023-12-15 20:02:44 +0000 | [diff] [blame] | 1803 | * While block_write_full_folio is writing back the dirty buffers under |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1804 | * the page lock, whoever dirtied the buffers may decide to clean them |
| 1805 | * again at any time. We handle that by only looking at the buffer |
| 1806 | * state inside lock_buffer(). |
| 1807 | * |
Matthew Wilcox (Oracle) | 17bf23a | 2023-12-15 20:02:44 +0000 | [diff] [blame] | 1808 | * If block_write_full_folio() is called for regular writeback |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1809 | * (wbc->sync_mode == WB_SYNC_NONE) then it will redirty a page which has a |
| 1810 | * locked buffer. This only can happen if someone has written the buffer |
| 1811 | * directly, with submit_bh(). At the address_space level PageWriteback |
| 1812 | * prevents this contention from occurring. |
Theodore Ts'o | 6e34eedd | 2009-04-07 18:12:43 -0400 | [diff] [blame] | 1813 | * |
Matthew Wilcox (Oracle) | 17bf23a | 2023-12-15 20:02:44 +0000 | [diff] [blame] | 1814 | * If block_write_full_folio() is called with wbc->sync_mode == |
Christoph Hellwig | 70fd761 | 2016-11-01 07:40:10 -0600 | [diff] [blame] | 1815 | * WB_SYNC_ALL, the writes are posted using REQ_SYNC; this |
Jens Axboe | 721a960 | 2011-03-09 11:56:30 +0100 | [diff] [blame] | 1816 | * causes the writes to be flagged as synchronous writes. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1817 | */ |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1818 | int __block_write_full_folio(struct inode *inode, struct folio *folio, |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 1819 | get_block_t *get_block, struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1820 | { |
| 1821 | int err; |
| 1822 | sector_t block; |
| 1823 | sector_t last_block; |
Andrew Morton | f0fbd5f | 2005-05-05 16:15:48 -0700 | [diff] [blame] | 1824 | struct buffer_head *bh, *head; |
Matthew Wilcox (Oracle) | fa399c3 | 2023-11-09 21:06:08 +0000 | [diff] [blame] | 1825 | size_t blocksize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1826 | int nr_underway = 0; |
Bart Van Assche | 3ae7286 | 2022-07-14 11:07:12 -0700 | [diff] [blame] | 1827 | blk_opf_t write_flags = wbc_to_write_flags(wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1828 | |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1829 | head = folio_create_buffers(folio, inode, |
Pankaj Raghav | c6c8c3e | 2023-04-17 14:36:18 +0200 | [diff] [blame] | 1830 | (1 << BH_Dirty) | (1 << BH_Uptodate)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1831 | |
| 1832 | /* |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 1833 | * Be very careful. We have no exclusion from block_dirty_folio |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1834 | * here, and the (potentially unmapped) buffers may become dirty at |
| 1835 | * any time. If a buffer becomes dirty here after we've inspected it |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1836 | * then we just miss that fact, and the folio stays dirty. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1837 | * |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 1838 | * Buffers outside i_size may be dirtied by block_dirty_folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1839 | * handle that here by just cleaning them. |
| 1840 | */ |
| 1841 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1842 | bh = head; |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 1843 | blocksize = bh->b_size; |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 1844 | |
Matthew Wilcox (Oracle) | fa399c3 | 2023-11-09 21:06:08 +0000 | [diff] [blame] | 1845 | block = div_u64(folio_pos(folio), blocksize); |
| 1846 | last_block = div_u64(i_size_read(inode) - 1, blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1847 | |
| 1848 | /* |
| 1849 | * Get all the dirty buffers mapped to disk addresses and |
| 1850 | * handle any aliases from the underlying blockdev's mapping. |
| 1851 | */ |
| 1852 | do { |
| 1853 | if (block > last_block) { |
| 1854 | /* |
| 1855 | * mapped buffers outside i_size will occur, because |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1856 | * this folio can be outside i_size when there is a |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1857 | * truncate in progress. |
| 1858 | */ |
| 1859 | /* |
Matthew Wilcox (Oracle) | 17bf23a | 2023-12-15 20:02:44 +0000 | [diff] [blame] | 1860 | * The buffer was zeroed by block_write_full_folio() |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1861 | */ |
| 1862 | clear_buffer_dirty(bh); |
| 1863 | set_buffer_uptodate(bh); |
Alex Tomas | 29a814d | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1864 | } else if ((!buffer_mapped(bh) || buffer_delay(bh)) && |
| 1865 | buffer_dirty(bh)) { |
Badari Pulavarty | b0cf232 | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 1866 | WARN_ON(bh->b_size != blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1867 | err = get_block(inode, block, bh, 1); |
| 1868 | if (err) |
| 1869 | goto recover; |
Alex Tomas | 29a814d | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1870 | clear_buffer_delay(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1871 | if (buffer_new(bh)) { |
| 1872 | /* blockdev mappings never come here */ |
| 1873 | clear_buffer_new(bh); |
Jan Kara | e64855c | 2016-11-04 18:08:15 +0100 | [diff] [blame] | 1874 | clean_bdev_bh_alias(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1875 | } |
| 1876 | } |
| 1877 | bh = bh->b_this_page; |
| 1878 | block++; |
| 1879 | } while (bh != head); |
| 1880 | |
| 1881 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1882 | if (!buffer_mapped(bh)) |
| 1883 | continue; |
| 1884 | /* |
| 1885 | * If it's a fully non-blocking write attempt and we cannot |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1886 | * lock the buffer then redirty the folio. Note that this can |
Jens Axboe | 5b0830c | 2009-09-23 19:37:09 +0200 | [diff] [blame] | 1887 | * potentially cause a busy-wait loop from writeback threads |
| 1888 | * and kswapd activity, but those code paths have their own |
| 1889 | * higher-level throttling. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1890 | */ |
Wu Fengguang | 1b430be | 2010-10-26 14:21:26 -0700 | [diff] [blame] | 1891 | if (wbc->sync_mode != WB_SYNC_NONE) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1892 | lock_buffer(bh); |
Nick Piggin | ca5de40 | 2008-08-02 12:02:13 +0200 | [diff] [blame] | 1893 | } else if (!trylock_buffer(bh)) { |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1894 | folio_redirty_for_writepage(wbc, folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1895 | continue; |
| 1896 | } |
| 1897 | if (test_clear_buffer_dirty(bh)) { |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 1898 | mark_buffer_async_write_endio(bh, |
| 1899 | end_buffer_async_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1900 | } else { |
| 1901 | unlock_buffer(bh); |
| 1902 | } |
| 1903 | } while ((bh = bh->b_this_page) != head); |
| 1904 | |
| 1905 | /* |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1906 | * The folio and its buffers are protected by the writeback flag, |
| 1907 | * so we can drop the bh refcounts early. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1908 | */ |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1909 | BUG_ON(folio_test_writeback(folio)); |
| 1910 | folio_start_writeback(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1911 | |
| 1912 | do { |
| 1913 | struct buffer_head *next = bh->b_this_page; |
| 1914 | if (buffer_async_write(bh)) { |
Bart Van Assche | 4498135 | 2024-02-02 12:39:25 -0800 | [diff] [blame] | 1915 | submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, |
| 1916 | inode->i_write_hint, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1917 | nr_underway++; |
| 1918 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1919 | bh = next; |
| 1920 | } while (bh != head); |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1921 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1922 | |
| 1923 | err = 0; |
| 1924 | done: |
| 1925 | if (nr_underway == 0) { |
| 1926 | /* |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1927 | * The folio was marked dirty, but the buffers were |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1928 | * clean. Someone wrote them back by hand with |
Zhang Yi | 79f5978 | 2022-09-01 21:35:03 +0800 | [diff] [blame] | 1929 | * write_dirty_buffer/submit_bh. A rare case. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1930 | */ |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1931 | folio_end_writeback(folio); |
Nick Piggin | 3d67f2d | 2007-05-06 14:49:05 -0700 | [diff] [blame] | 1932 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1933 | /* |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1934 | * The folio and buffer_heads can be released at any time from |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1935 | * here on. |
| 1936 | */ |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1937 | } |
| 1938 | return err; |
| 1939 | |
| 1940 | recover: |
| 1941 | /* |
| 1942 | * ENOSPC, or some other error. We may already have added some |
| 1943 | * blocks to the file, so we need to write these out to avoid |
| 1944 | * exposing stale data. |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1945 | * The folio is currently locked and not marked for writeback |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1946 | */ |
| 1947 | bh = head; |
| 1948 | /* Recovery: lock and submit the mapped buffers */ |
| 1949 | do { |
Alex Tomas | 29a814d | 2008-07-11 19:27:31 -0400 | [diff] [blame] | 1950 | if (buffer_mapped(bh) && buffer_dirty(bh) && |
| 1951 | !buffer_delay(bh)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1952 | lock_buffer(bh); |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 1953 | mark_buffer_async_write_endio(bh, |
| 1954 | end_buffer_async_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1955 | } else { |
| 1956 | /* |
| 1957 | * The buffer may have been set dirty during |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1958 | * attachment to a dirty folio. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1959 | */ |
| 1960 | clear_buffer_dirty(bh); |
| 1961 | } |
| 1962 | } while ((bh = bh->b_this_page) != head); |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1963 | folio_set_error(folio); |
| 1964 | BUG_ON(folio_test_writeback(folio)); |
| 1965 | mapping_set_error(folio->mapping, err); |
| 1966 | folio_start_writeback(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1967 | do { |
| 1968 | struct buffer_head *next = bh->b_this_page; |
| 1969 | if (buffer_async_write(bh)) { |
| 1970 | clear_buffer_dirty(bh); |
Bart Van Assche | 4498135 | 2024-02-02 12:39:25 -0800 | [diff] [blame] | 1971 | submit_bh_wbc(REQ_OP_WRITE | write_flags, bh, |
| 1972 | inode->i_write_hint, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1973 | nr_underway++; |
| 1974 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1975 | bh = next; |
| 1976 | } while (bh != head); |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1977 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1978 | goto done; |
| 1979 | } |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 1980 | EXPORT_SYMBOL(__block_write_full_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 1981 | |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 1982 | /* |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 1983 | * If a folio has any new buffers, zero them out here, and mark them uptodate |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 1984 | * and dirty so they'll be written out (in order to prevent uninitialised |
| 1985 | * block data from leaking). And clear the new bit. |
| 1986 | */ |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 1987 | void folio_zero_new_buffers(struct folio *folio, size_t from, size_t to) |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 1988 | { |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 1989 | size_t block_start, block_end; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 1990 | struct buffer_head *head, *bh; |
| 1991 | |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 1992 | BUG_ON(!folio_test_locked(folio)); |
| 1993 | head = folio_buffers(folio); |
| 1994 | if (!head) |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 1995 | return; |
| 1996 | |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 1997 | bh = head; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 1998 | block_start = 0; |
| 1999 | do { |
| 2000 | block_end = block_start + bh->b_size; |
| 2001 | |
| 2002 | if (buffer_new(bh)) { |
| 2003 | if (block_end > from && block_start < to) { |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 2004 | if (!folio_test_uptodate(folio)) { |
| 2005 | size_t start, xend; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2006 | |
| 2007 | start = max(from, block_start); |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 2008 | xend = min(to, block_end); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2009 | |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 2010 | folio_zero_segment(folio, start, xend); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2011 | set_buffer_uptodate(bh); |
| 2012 | } |
| 2013 | |
| 2014 | clear_buffer_new(bh); |
| 2015 | mark_buffer_dirty(bh); |
| 2016 | } |
| 2017 | } |
| 2018 | |
| 2019 | block_start = block_end; |
| 2020 | bh = bh->b_this_page; |
| 2021 | } while (bh != head); |
| 2022 | } |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 2023 | EXPORT_SYMBOL(folio_zero_new_buffers); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2024 | |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2025 | static int |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2026 | iomap_to_bh(struct inode *inode, sector_t block, struct buffer_head *bh, |
Christoph Hellwig | 6d49cc8 | 2021-08-10 18:33:05 -0700 | [diff] [blame] | 2027 | const struct iomap *iomap) |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2028 | { |
Matthew Wilcox (Oracle) | 8084419 | 2023-11-09 21:06:05 +0000 | [diff] [blame] | 2029 | loff_t offset = (loff_t)block << inode->i_blkbits; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2030 | |
| 2031 | bh->b_bdev = iomap->bdev; |
| 2032 | |
| 2033 | /* |
| 2034 | * Block points to offset in file we need to map, iomap contains |
| 2035 | * the offset at which the map starts. If the map ends before the |
| 2036 | * current block, then do not map the buffer and let the caller |
| 2037 | * handle it. |
| 2038 | */ |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2039 | if (offset >= iomap->offset + iomap->length) |
| 2040 | return -EIO; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2041 | |
| 2042 | switch (iomap->type) { |
| 2043 | case IOMAP_HOLE: |
| 2044 | /* |
| 2045 | * If the buffer is not up to date or beyond the current EOF, |
| 2046 | * we need to mark it as new to ensure sub-block zeroing is |
| 2047 | * executed if necessary. |
| 2048 | */ |
| 2049 | if (!buffer_uptodate(bh) || |
| 2050 | (offset >= i_size_read(inode))) |
| 2051 | set_buffer_new(bh); |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2052 | return 0; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2053 | case IOMAP_DELALLOC: |
| 2054 | if (!buffer_uptodate(bh) || |
| 2055 | (offset >= i_size_read(inode))) |
| 2056 | set_buffer_new(bh); |
| 2057 | set_buffer_uptodate(bh); |
| 2058 | set_buffer_mapped(bh); |
| 2059 | set_buffer_delay(bh); |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2060 | return 0; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2061 | case IOMAP_UNWRITTEN: |
| 2062 | /* |
Andreas Gruenbacher | 3d7b6b21 | 2018-06-19 15:10:55 -0700 | [diff] [blame] | 2063 | * For unwritten regions, we always need to ensure that regions |
| 2064 | * in the block we are not writing to are zeroed. Mark the |
| 2065 | * buffer as new to ensure this. |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2066 | */ |
| 2067 | set_buffer_new(bh); |
| 2068 | set_buffer_unwritten(bh); |
Gustavo A. R. Silva | df561f66 | 2020-08-23 17:36:59 -0500 | [diff] [blame] | 2069 | fallthrough; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2070 | case IOMAP_MAPPED: |
Andreas Gruenbacher | 3d7b6b21 | 2018-06-19 15:10:55 -0700 | [diff] [blame] | 2071 | if ((iomap->flags & IOMAP_F_NEW) || |
Christoph Hellwig | 381c043 | 2023-09-25 08:54:45 -0700 | [diff] [blame] | 2072 | offset >= i_size_read(inode)) { |
| 2073 | /* |
| 2074 | * This can happen if truncating the block device races |
| 2075 | * with the check in the caller as i_size updates on |
| 2076 | * block devices aren't synchronized by i_rwsem for |
| 2077 | * block devices. |
| 2078 | */ |
| 2079 | if (S_ISBLK(inode->i_mode)) |
| 2080 | return -EIO; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2081 | set_buffer_new(bh); |
Christoph Hellwig | 381c043 | 2023-09-25 08:54:45 -0700 | [diff] [blame] | 2082 | } |
Andreas Gruenbacher | 19fe5f6 | 2017-10-01 17:55:54 -0400 | [diff] [blame] | 2083 | bh->b_blocknr = (iomap->addr + offset - iomap->offset) >> |
| 2084 | inode->i_blkbits; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2085 | set_buffer_mapped(bh); |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2086 | return 0; |
| 2087 | default: |
| 2088 | WARN_ON_ONCE(1); |
| 2089 | return -EIO; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2090 | } |
| 2091 | } |
| 2092 | |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2093 | int __block_write_begin_int(struct folio *folio, loff_t pos, unsigned len, |
Christoph Hellwig | 6d49cc8 | 2021-08-10 18:33:05 -0700 | [diff] [blame] | 2094 | get_block_t *get_block, const struct iomap *iomap) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2095 | { |
Matthew Wilcox (Oracle) | b061940 | 2023-11-09 21:06:07 +0000 | [diff] [blame] | 2096 | size_t from = offset_in_folio(folio, pos); |
| 2097 | size_t to = from + len; |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2098 | struct inode *inode = folio->mapping->host; |
Matthew Wilcox (Oracle) | b061940 | 2023-11-09 21:06:07 +0000 | [diff] [blame] | 2099 | size_t block_start, block_end; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2100 | sector_t block; |
| 2101 | int err = 0; |
Matthew Wilcox (Oracle) | b061940 | 2023-11-09 21:06:07 +0000 | [diff] [blame] | 2102 | size_t blocksize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2103 | struct buffer_head *bh, *head, *wait[2], **wait_bh=wait; |
| 2104 | |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2105 | BUG_ON(!folio_test_locked(folio)); |
Matthew Wilcox (Oracle) | b061940 | 2023-11-09 21:06:07 +0000 | [diff] [blame] | 2106 | BUG_ON(to > folio_size(folio)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2107 | BUG_ON(from > to); |
| 2108 | |
Pankaj Raghav | c6c8c3e | 2023-04-17 14:36:18 +0200 | [diff] [blame] | 2109 | head = folio_create_buffers(folio, inode, 0); |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 2110 | blocksize = head->b_size; |
Matthew Wilcox (Oracle) | b061940 | 2023-11-09 21:06:07 +0000 | [diff] [blame] | 2111 | block = div_u64(folio_pos(folio), blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2112 | |
Matthew Wilcox (Oracle) | b061940 | 2023-11-09 21:06:07 +0000 | [diff] [blame] | 2113 | for (bh = head, block_start = 0; bh != head || !block_start; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2114 | block++, block_start=block_end, bh = bh->b_this_page) { |
| 2115 | block_end = block_start + blocksize; |
| 2116 | if (block_end <= from || block_start >= to) { |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2117 | if (folio_test_uptodate(folio)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2118 | if (!buffer_uptodate(bh)) |
| 2119 | set_buffer_uptodate(bh); |
| 2120 | } |
| 2121 | continue; |
| 2122 | } |
| 2123 | if (buffer_new(bh)) |
| 2124 | clear_buffer_new(bh); |
| 2125 | if (!buffer_mapped(bh)) { |
Badari Pulavarty | b0cf232 | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 2126 | WARN_ON(bh->b_size != blocksize); |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2127 | if (get_block) |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2128 | err = get_block(inode, block, bh, 1); |
Christoph Hellwig | 4aa8cdd | 2023-09-12 10:05:48 -0700 | [diff] [blame] | 2129 | else |
| 2130 | err = iomap_to_bh(inode, block, bh, iomap); |
| 2131 | if (err) |
| 2132 | break; |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2133 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2134 | if (buffer_new(bh)) { |
Jan Kara | e64855c | 2016-11-04 18:08:15 +0100 | [diff] [blame] | 2135 | clean_bdev_bh_alias(bh); |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2136 | if (folio_test_uptodate(folio)) { |
Nick Piggin | 637aff4 | 2007-10-16 01:25:00 -0700 | [diff] [blame] | 2137 | clear_buffer_new(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2138 | set_buffer_uptodate(bh); |
Nick Piggin | 637aff4 | 2007-10-16 01:25:00 -0700 | [diff] [blame] | 2139 | mark_buffer_dirty(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2140 | continue; |
| 2141 | } |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 2142 | if (block_end > to || block_start < from) |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2143 | folio_zero_segments(folio, |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 2144 | to, block_end, |
| 2145 | block_start, from); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2146 | continue; |
| 2147 | } |
| 2148 | } |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2149 | if (folio_test_uptodate(folio)) { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2150 | if (!buffer_uptodate(bh)) |
| 2151 | set_buffer_uptodate(bh); |
| 2152 | continue; |
| 2153 | } |
| 2154 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && |
David Chinner | 33a266d | 2007-02-12 00:51:41 -0800 | [diff] [blame] | 2155 | !buffer_unwritten(bh) && |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2156 | (block_start < from || block_end > to)) { |
Zhang Yi | e7ea112 | 2022-09-01 21:34:54 +0800 | [diff] [blame] | 2157 | bh_read_nowait(bh, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2158 | *wait_bh++=bh; |
| 2159 | } |
| 2160 | } |
| 2161 | /* |
| 2162 | * If we issued read requests - let them complete. |
| 2163 | */ |
| 2164 | while(wait_bh > wait) { |
| 2165 | wait_on_buffer(*--wait_bh); |
| 2166 | if (!buffer_uptodate(*wait_bh)) |
Nick Piggin | f3ddbdc | 2005-05-05 16:15:45 -0700 | [diff] [blame] | 2167 | err = -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2168 | } |
Jan Kara | f9f07b6 | 2011-06-14 00:58:27 +0200 | [diff] [blame] | 2169 | if (unlikely(err)) |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 2170 | folio_zero_new_buffers(folio, from, to); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2171 | return err; |
| 2172 | } |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2173 | |
| 2174 | int __block_write_begin(struct page *page, loff_t pos, unsigned len, |
| 2175 | get_block_t *get_block) |
| 2176 | { |
Matthew Wilcox (Oracle) | d1bd0b4 | 2021-11-03 14:05:47 -0400 | [diff] [blame] | 2177 | return __block_write_begin_int(page_folio(page), pos, len, get_block, |
| 2178 | NULL); |
Christoph Hellwig | ae259a9 | 2016-06-21 09:23:11 +1000 | [diff] [blame] | 2179 | } |
Christoph Hellwig | ebdec24 | 2010-10-06 10:47:23 +0200 | [diff] [blame] | 2180 | EXPORT_SYMBOL(__block_write_begin); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2181 | |
Bean Huo | a524fcf | 2023-06-26 07:55:18 +0200 | [diff] [blame] | 2182 | static void __block_commit_write(struct folio *folio, size_t from, size_t to) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2183 | { |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2184 | size_t block_start, block_end; |
| 2185 | bool partial = false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2186 | unsigned blocksize; |
| 2187 | struct buffer_head *bh, *head; |
| 2188 | |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2189 | bh = head = folio_buffers(folio); |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 2190 | blocksize = bh->b_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2191 | |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 2192 | block_start = 0; |
| 2193 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2194 | block_end = block_start + blocksize; |
| 2195 | if (block_end <= from || block_start >= to) { |
| 2196 | if (!buffer_uptodate(bh)) |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2197 | partial = true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2198 | } else { |
| 2199 | set_buffer_uptodate(bh); |
| 2200 | mark_buffer_dirty(bh); |
| 2201 | } |
Yang Guo | 4ebd3ae | 2021-02-24 12:02:48 -0800 | [diff] [blame] | 2202 | if (buffer_new(bh)) |
| 2203 | clear_buffer_new(bh); |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 2204 | |
| 2205 | block_start = block_end; |
| 2206 | bh = bh->b_this_page; |
| 2207 | } while (bh != head); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2208 | |
| 2209 | /* |
| 2210 | * If this is a partial write which happened to make all buffers |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2211 | * uptodate then we can optimize away a bogus read_folio() for |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2212 | * the next read(). Here we 'discover' whether the folio went |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2213 | * uptodate as a result of this (potentially partial) write. |
| 2214 | */ |
| 2215 | if (!partial) |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2216 | folio_mark_uptodate(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2217 | } |
| 2218 | |
| 2219 | /* |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 2220 | * block_write_begin takes care of the basic task of block allocation and |
| 2221 | * bringing partial write blocks uptodate first. |
| 2222 | * |
npiggin@suse.de | 7bb46a6 | 2010-05-27 01:05:33 +1000 | [diff] [blame] | 2223 | * The filesystem needs to handle block truncation upon failure. |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2224 | */ |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 2225 | int block_write_begin(struct address_space *mapping, loff_t pos, unsigned len, |
Matthew Wilcox (Oracle) | b3992d1 | 2022-02-22 11:25:12 -0500 | [diff] [blame] | 2226 | struct page **pagep, get_block_t *get_block) |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2227 | { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2228 | pgoff_t index = pos >> PAGE_SHIFT; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2229 | struct page *page; |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2230 | int status; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2231 | |
Matthew Wilcox (Oracle) | b7446e7 | 2022-02-22 11:25:12 -0500 | [diff] [blame] | 2232 | page = grab_cache_page_write_begin(mapping, index); |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2233 | if (!page) |
| 2234 | return -ENOMEM; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2235 | |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2236 | status = __block_write_begin(page, pos, len, get_block); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2237 | if (unlikely(status)) { |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2238 | unlock_page(page); |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2239 | put_page(page); |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2240 | page = NULL; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2241 | } |
| 2242 | |
Christoph Hellwig | 6e1db88 | 2010-06-04 11:29:57 +0200 | [diff] [blame] | 2243 | *pagep = page; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2244 | return status; |
| 2245 | } |
| 2246 | EXPORT_SYMBOL(block_write_begin); |
| 2247 | |
| 2248 | int block_write_end(struct file *file, struct address_space *mapping, |
| 2249 | loff_t pos, unsigned len, unsigned copied, |
| 2250 | struct page *page, void *fsdata) |
| 2251 | { |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2252 | struct folio *folio = page_folio(page); |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2253 | size_t start = pos - folio_pos(folio); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2254 | |
| 2255 | if (unlikely(copied < len)) { |
| 2256 | /* |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2257 | * The buffers that were written will now be uptodate, so |
| 2258 | * we don't have to worry about a read_folio reading them |
| 2259 | * and overwriting a partial write. However if we have |
| 2260 | * encountered a short write and only partially written |
| 2261 | * into a buffer, it will not be marked uptodate, so a |
| 2262 | * read_folio might come in and destroy our partial write. |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2263 | * |
| 2264 | * Do the simplest thing, and just treat any short write to a |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2265 | * non uptodate folio as a zero-length write, and force the |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2266 | * caller to redo the whole thing. |
| 2267 | */ |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2268 | if (!folio_test_uptodate(folio)) |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2269 | copied = 0; |
| 2270 | |
Matthew Wilcox (Oracle) | 4a9622f | 2023-06-12 22:01:36 +0100 | [diff] [blame] | 2271 | folio_zero_new_buffers(folio, start+copied, start+len); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2272 | } |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2273 | flush_dcache_folio(folio); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2274 | |
| 2275 | /* This could be a short (even 0-length) commit */ |
Bean Huo | 489b7e7 | 2023-06-26 07:55:17 +0200 | [diff] [blame] | 2276 | __block_commit_write(folio, start, start + copied); |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2277 | |
| 2278 | return copied; |
| 2279 | } |
| 2280 | EXPORT_SYMBOL(block_write_end); |
| 2281 | |
| 2282 | int generic_write_end(struct file *file, struct address_space *mapping, |
| 2283 | loff_t pos, unsigned len, unsigned copied, |
| 2284 | struct page *page, void *fsdata) |
| 2285 | { |
Christoph Hellwig | 8af54f2 | 2019-06-27 17:28:40 -0700 | [diff] [blame] | 2286 | struct inode *inode = mapping->host; |
| 2287 | loff_t old_size = inode->i_size; |
| 2288 | bool i_size_changed = false; |
| 2289 | |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2290 | copied = block_write_end(file, mapping, pos, len, copied, page, fsdata); |
Christoph Hellwig | 8af54f2 | 2019-06-27 17:28:40 -0700 | [diff] [blame] | 2291 | |
| 2292 | /* |
| 2293 | * No need to use i_size_read() here, the i_size cannot change under us |
| 2294 | * because we hold i_rwsem. |
| 2295 | * |
| 2296 | * But it's important to update i_size while still holding page lock: |
| 2297 | * page writeout could otherwise come in and zero beyond i_size. |
| 2298 | */ |
| 2299 | if (pos + copied > inode->i_size) { |
| 2300 | i_size_write(inode, pos + copied); |
| 2301 | i_size_changed = true; |
| 2302 | } |
| 2303 | |
| 2304 | unlock_page(page); |
Andreas Gruenbacher | 7a77dad | 2019-04-30 08:45:34 -0700 | [diff] [blame] | 2305 | put_page(page); |
Christoph Hellwig | 8af54f2 | 2019-06-27 17:28:40 -0700 | [diff] [blame] | 2306 | |
| 2307 | if (old_size < pos) |
| 2308 | pagecache_isize_extended(inode, old_size, pos); |
| 2309 | /* |
| 2310 | * Don't mark the inode dirty under page lock. First, it unnecessarily |
| 2311 | * makes the holding time of page lock longer. Second, it forces lock |
| 2312 | * ordering of page lock and transaction start for journaling |
| 2313 | * filesystems. |
| 2314 | */ |
| 2315 | if (i_size_changed) |
| 2316 | mark_inode_dirty(inode); |
Andreas Gruenbacher | 26ddb1f | 2019-04-30 08:45:33 -0700 | [diff] [blame] | 2317 | return copied; |
Nick Piggin | afddba4 | 2007-10-16 01:25:01 -0700 | [diff] [blame] | 2318 | } |
| 2319 | EXPORT_SYMBOL(generic_write_end); |
| 2320 | |
| 2321 | /* |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2322 | * block_is_partially_uptodate checks whether buffers within a folio are |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2323 | * uptodate or not. |
| 2324 | * |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2325 | * Returns true if all buffers which correspond to the specified part |
| 2326 | * of the folio are uptodate. |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2327 | */ |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2328 | bool block_is_partially_uptodate(struct folio *folio, size_t from, size_t count) |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2329 | { |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2330 | unsigned block_start, block_end, blocksize; |
| 2331 | unsigned to; |
| 2332 | struct buffer_head *bh, *head; |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2333 | bool ret = true; |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2334 | |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2335 | head = folio_buffers(folio); |
| 2336 | if (!head) |
| 2337 | return false; |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 2338 | blocksize = head->b_size; |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2339 | to = min_t(unsigned, folio_size(folio) - from, count); |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2340 | to = from + to; |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2341 | if (from < blocksize && to > folio_size(folio) - blocksize) |
| 2342 | return false; |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2343 | |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2344 | bh = head; |
| 2345 | block_start = 0; |
| 2346 | do { |
| 2347 | block_end = block_start + blocksize; |
| 2348 | if (block_end > from && block_start < to) { |
| 2349 | if (!buffer_uptodate(bh)) { |
Matthew Wilcox (Oracle) | 2e7e80f | 2022-02-09 20:21:27 +0000 | [diff] [blame] | 2350 | ret = false; |
Hisashi Hifumi | 8ab22b9 | 2008-07-28 15:46:36 -0700 | [diff] [blame] | 2351 | break; |
| 2352 | } |
| 2353 | if (block_end >= to) |
| 2354 | break; |
| 2355 | } |
| 2356 | block_start = block_end; |
| 2357 | bh = bh->b_this_page; |
| 2358 | } while (bh != head); |
| 2359 | |
| 2360 | return ret; |
| 2361 | } |
| 2362 | EXPORT_SYMBOL(block_is_partially_uptodate); |
| 2363 | |
| 2364 | /* |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2365 | * Generic "read_folio" function for block devices that have the normal |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2366 | * get_block functionality. This is most of the block device filesystems. |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2367 | * Reads the folio asynchronously --- the unlock_buffer() and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2368 | * set/clear_buffer_uptodate() functions propagate buffer state into the |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2369 | * folio once IO has completed. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2370 | */ |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2371 | int block_read_full_folio(struct folio *folio, get_block_t *get_block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2372 | { |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2373 | struct inode *inode = folio->mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2374 | sector_t iblock, lblock; |
| 2375 | struct buffer_head *bh, *head, *arr[MAX_BUF_PER_PAGE]; |
Matthew Wilcox (Oracle) | fa399c3 | 2023-11-09 21:06:08 +0000 | [diff] [blame] | 2376 | size_t blocksize; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2377 | int nr, i; |
| 2378 | int fully_mapped = 1; |
Matthew Wilcox (Oracle) | b7a6eb2 | 2022-05-26 23:19:49 -0400 | [diff] [blame] | 2379 | bool page_error = false; |
Eric Biggers | 4fa512c | 2022-12-23 12:36:37 -0800 | [diff] [blame] | 2380 | loff_t limit = i_size_read(inode); |
| 2381 | |
| 2382 | /* This is needed for ext4. */ |
| 2383 | if (IS_ENABLED(CONFIG_FS_VERITY) && IS_VERITY(inode)) |
| 2384 | limit = inode->i_sb->s_maxbytes; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2385 | |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2386 | VM_BUG_ON_FOLIO(folio_test_large(folio), folio); |
| 2387 | |
Pankaj Raghav | c6c8c3e | 2023-04-17 14:36:18 +0200 | [diff] [blame] | 2388 | head = folio_create_buffers(folio, inode, 0); |
Linus Torvalds | 45bce8f | 2012-11-29 10:21:43 -0800 | [diff] [blame] | 2389 | blocksize = head->b_size; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2390 | |
Matthew Wilcox (Oracle) | fa399c3 | 2023-11-09 21:06:08 +0000 | [diff] [blame] | 2391 | iblock = div_u64(folio_pos(folio), blocksize); |
| 2392 | lblock = div_u64(limit + blocksize - 1, blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2393 | bh = head; |
| 2394 | nr = 0; |
| 2395 | i = 0; |
| 2396 | |
| 2397 | do { |
| 2398 | if (buffer_uptodate(bh)) |
| 2399 | continue; |
| 2400 | |
| 2401 | if (!buffer_mapped(bh)) { |
Andrew Morton | c64610b | 2005-05-16 21:53:49 -0700 | [diff] [blame] | 2402 | int err = 0; |
| 2403 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2404 | fully_mapped = 0; |
| 2405 | if (iblock < lblock) { |
Badari Pulavarty | b0cf232 | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 2406 | WARN_ON(bh->b_size != blocksize); |
Andrew Morton | c64610b | 2005-05-16 21:53:49 -0700 | [diff] [blame] | 2407 | err = get_block(inode, iblock, bh, 0); |
Matthew Wilcox (Oracle) | b7a6eb2 | 2022-05-26 23:19:49 -0400 | [diff] [blame] | 2408 | if (err) { |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2409 | folio_set_error(folio); |
Matthew Wilcox (Oracle) | b7a6eb2 | 2022-05-26 23:19:49 -0400 | [diff] [blame] | 2410 | page_error = true; |
| 2411 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2412 | } |
| 2413 | if (!buffer_mapped(bh)) { |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2414 | folio_zero_range(folio, i * blocksize, |
| 2415 | blocksize); |
Andrew Morton | c64610b | 2005-05-16 21:53:49 -0700 | [diff] [blame] | 2416 | if (!err) |
| 2417 | set_buffer_uptodate(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2418 | continue; |
| 2419 | } |
| 2420 | /* |
| 2421 | * get_block() might have updated the buffer |
| 2422 | * synchronously |
| 2423 | */ |
| 2424 | if (buffer_uptodate(bh)) |
| 2425 | continue; |
| 2426 | } |
| 2427 | arr[nr++] = bh; |
| 2428 | } while (i++, iblock++, (bh = bh->b_this_page) != head); |
| 2429 | |
| 2430 | if (fully_mapped) |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2431 | folio_set_mappedtodisk(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2432 | |
| 2433 | if (!nr) { |
| 2434 | /* |
Matthew Wilcox (Oracle) | 6ba924d | 2023-10-04 17:53:05 +0100 | [diff] [blame] | 2435 | * All buffers are uptodate or get_block() returned an |
| 2436 | * error when trying to map them - we can finish the read. |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2437 | */ |
Matthew Wilcox (Oracle) | 6ba924d | 2023-10-04 17:53:05 +0100 | [diff] [blame] | 2438 | folio_end_read(folio, !page_error); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2439 | return 0; |
| 2440 | } |
| 2441 | |
| 2442 | /* Stage two: lock the buffers */ |
| 2443 | for (i = 0; i < nr; i++) { |
| 2444 | bh = arr[i]; |
| 2445 | lock_buffer(bh); |
| 2446 | mark_buffer_async_read(bh); |
| 2447 | } |
| 2448 | |
| 2449 | /* |
| 2450 | * Stage 3: start the IO. Check for uptodateness |
| 2451 | * inside the buffer lock in case another process reading |
| 2452 | * the underlying blockdev brought it uptodate (the sct fix). |
| 2453 | */ |
| 2454 | for (i = 0; i < nr; i++) { |
| 2455 | bh = arr[i]; |
| 2456 | if (buffer_uptodate(bh)) |
| 2457 | end_buffer_async_read(bh, 1); |
| 2458 | else |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 2459 | submit_bh(REQ_OP_READ, bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2460 | } |
| 2461 | return 0; |
| 2462 | } |
Matthew Wilcox (Oracle) | 2c69e20 | 2022-04-29 10:40:40 -0400 | [diff] [blame] | 2463 | EXPORT_SYMBOL(block_read_full_folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2464 | |
| 2465 | /* utility function for filesystems that need to do work on expanding |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2466 | * truncates. Uses filesystem pagecache writes to allow the filesystem to |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2467 | * deal with the hole. |
| 2468 | */ |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2469 | int generic_cont_expand_simple(struct inode *inode, loff_t size) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2470 | { |
| 2471 | struct address_space *mapping = inode->i_mapping; |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2472 | const struct address_space_operations *aops = mapping->a_ops; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2473 | struct page *page; |
Alexander Potapenko | 1468c6f | 2022-09-15 17:04:16 +0200 | [diff] [blame] | 2474 | void *fsdata = NULL; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2475 | int err; |
| 2476 | |
npiggin@suse.de | c08d3b0 | 2009-08-21 02:35:06 +1000 | [diff] [blame] | 2477 | err = inode_newsize_ok(inode, size); |
| 2478 | if (err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2479 | goto out; |
| 2480 | |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2481 | err = aops->write_begin(NULL, mapping, size, 0, &page, &fsdata); |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2482 | if (err) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2483 | goto out; |
OGAWA Hirofumi | 05eb0b5 | 2006-01-08 01:02:13 -0800 | [diff] [blame] | 2484 | |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2485 | err = aops->write_end(NULL, mapping, size, 0, 0, page, fsdata); |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2486 | BUG_ON(err > 0); |
OGAWA Hirofumi | 05eb0b5 | 2006-01-08 01:02:13 -0800 | [diff] [blame] | 2487 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2488 | out: |
| 2489 | return err; |
| 2490 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2491 | EXPORT_SYMBOL(generic_cont_expand_simple); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2492 | |
Adrian Bunk | f1e3af7 | 2008-04-29 00:59:01 -0700 | [diff] [blame] | 2493 | static int cont_expand_zero(struct file *file, struct address_space *mapping, |
| 2494 | loff_t pos, loff_t *bytes) |
OGAWA Hirofumi | 05eb0b5 | 2006-01-08 01:02:13 -0800 | [diff] [blame] | 2495 | { |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2496 | struct inode *inode = mapping->host; |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2497 | const struct address_space_operations *aops = mapping->a_ops; |
Fabian Frederick | 9340747 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 2498 | unsigned int blocksize = i_blocksize(inode); |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2499 | struct page *page; |
Alexander Potapenko | 1468c6f | 2022-09-15 17:04:16 +0200 | [diff] [blame] | 2500 | void *fsdata = NULL; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2501 | pgoff_t index, curidx; |
| 2502 | loff_t curpos; |
| 2503 | unsigned zerofrom, offset, len; |
| 2504 | int err = 0; |
OGAWA Hirofumi | 05eb0b5 | 2006-01-08 01:02:13 -0800 | [diff] [blame] | 2505 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2506 | index = pos >> PAGE_SHIFT; |
| 2507 | offset = pos & ~PAGE_MASK; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2508 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2509 | while (index > (curidx = (curpos = *bytes)>>PAGE_SHIFT)) { |
| 2510 | zerofrom = curpos & ~PAGE_MASK; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2511 | if (zerofrom & (blocksize-1)) { |
| 2512 | *bytes |= (blocksize-1); |
| 2513 | (*bytes)++; |
| 2514 | } |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2515 | len = PAGE_SIZE - zerofrom; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2516 | |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2517 | err = aops->write_begin(file, mapping, curpos, len, |
Tetsuo Handa | c718a97 | 2017-05-08 15:58:59 -0700 | [diff] [blame] | 2518 | &page, &fsdata); |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2519 | if (err) |
| 2520 | goto out; |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 2521 | zero_user(page, zerofrom, len); |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2522 | err = aops->write_end(file, mapping, curpos, len, len, |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2523 | page, fsdata); |
| 2524 | if (err < 0) |
| 2525 | goto out; |
| 2526 | BUG_ON(err != len); |
| 2527 | err = 0; |
OGAWA Hirofumi | 061e974 | 2008-04-28 02:16:28 -0700 | [diff] [blame] | 2528 | |
| 2529 | balance_dirty_pages_ratelimited(mapping); |
Mikulas Patocka | c2ca0fc | 2014-07-27 13:00:41 -0400 | [diff] [blame] | 2530 | |
Davidlohr Bueso | 08d405c | 2019-01-03 15:28:58 -0800 | [diff] [blame] | 2531 | if (fatal_signal_pending(current)) { |
Mikulas Patocka | c2ca0fc | 2014-07-27 13:00:41 -0400 | [diff] [blame] | 2532 | err = -EINTR; |
| 2533 | goto out; |
| 2534 | } |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2535 | } |
| 2536 | |
| 2537 | /* page covers the boundary, find the boundary offset */ |
| 2538 | if (index == curidx) { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2539 | zerofrom = curpos & ~PAGE_MASK; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2540 | /* if we will expand the thing last block will be filled */ |
| 2541 | if (offset <= zerofrom) { |
| 2542 | goto out; |
| 2543 | } |
| 2544 | if (zerofrom & (blocksize-1)) { |
| 2545 | *bytes |= (blocksize-1); |
| 2546 | (*bytes)++; |
| 2547 | } |
| 2548 | len = offset - zerofrom; |
| 2549 | |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2550 | err = aops->write_begin(file, mapping, curpos, len, |
Tetsuo Handa | c718a97 | 2017-05-08 15:58:59 -0700 | [diff] [blame] | 2551 | &page, &fsdata); |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2552 | if (err) |
| 2553 | goto out; |
Christoph Lameter | eebd2aa | 2008-02-04 22:28:29 -0800 | [diff] [blame] | 2554 | zero_user(page, zerofrom, len); |
Matthew Wilcox (Oracle) | 53b524b | 2022-03-03 13:35:20 -0500 | [diff] [blame] | 2555 | err = aops->write_end(file, mapping, curpos, len, len, |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2556 | page, fsdata); |
| 2557 | if (err < 0) |
| 2558 | goto out; |
| 2559 | BUG_ON(err != len); |
| 2560 | err = 0; |
| 2561 | } |
| 2562 | out: |
| 2563 | return err; |
OGAWA Hirofumi | 05eb0b5 | 2006-01-08 01:02:13 -0800 | [diff] [blame] | 2564 | } |
| 2565 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2566 | /* |
| 2567 | * For moronic filesystems that do not allow holes in file. |
| 2568 | * We may have to extend the file. |
| 2569 | */ |
Christoph Hellwig | 282dc17 | 2010-06-04 11:29:55 +0200 | [diff] [blame] | 2570 | int cont_write_begin(struct file *file, struct address_space *mapping, |
Matthew Wilcox (Oracle) | be3bbbc | 2022-02-22 11:25:12 -0500 | [diff] [blame] | 2571 | loff_t pos, unsigned len, |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2572 | struct page **pagep, void **fsdata, |
| 2573 | get_block_t *get_block, loff_t *bytes) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2574 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2575 | struct inode *inode = mapping->host; |
Fabian Frederick | 9340747 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 2576 | unsigned int blocksize = i_blocksize(inode); |
| 2577 | unsigned int zerofrom; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2578 | int err; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2579 | |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2580 | err = cont_expand_zero(file, mapping, pos, bytes); |
| 2581 | if (err) |
Christoph Hellwig | 155130a | 2010-06-04 11:29:58 +0200 | [diff] [blame] | 2582 | return err; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2583 | |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2584 | zerofrom = *bytes & ~PAGE_MASK; |
Nick Piggin | 89e1078 | 2007-10-16 01:25:07 -0700 | [diff] [blame] | 2585 | if (pos+len > *bytes && zerofrom & (blocksize-1)) { |
| 2586 | *bytes |= (blocksize-1); |
| 2587 | (*bytes)++; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2588 | } |
| 2589 | |
Matthew Wilcox (Oracle) | b3992d1 | 2022-02-22 11:25:12 -0500 | [diff] [blame] | 2590 | return block_write_begin(mapping, pos, len, pagep, get_block); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2591 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2592 | EXPORT_SYMBOL(cont_write_begin); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2593 | |
Bean Huo | a524fcf | 2023-06-26 07:55:18 +0200 | [diff] [blame] | 2594 | void block_commit_write(struct page *page, unsigned from, unsigned to) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2595 | { |
Matthew Wilcox (Oracle) | 8c6cb3e | 2023-06-12 22:01:35 +0100 | [diff] [blame] | 2596 | struct folio *folio = page_folio(page); |
Bean Huo | 489b7e7 | 2023-06-26 07:55:17 +0200 | [diff] [blame] | 2597 | __block_commit_write(folio, from, to); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2598 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2599 | EXPORT_SYMBOL(block_commit_write); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2600 | |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2601 | /* |
| 2602 | * block_page_mkwrite() is not allowed to change the file size as it gets |
| 2603 | * called from a page fault handler when a page is first dirtied. Hence we must |
| 2604 | * be careful to check for EOF conditions here. We set the page up correctly |
| 2605 | * for a written page which means we get ENOSPC checking when writing into |
| 2606 | * holes and correct delalloc and unwritten extent mapping on filesystems that |
| 2607 | * support these features. |
| 2608 | * |
| 2609 | * We are not allowed to take the i_mutex here so we have to play games to |
| 2610 | * protect against truncate races as the page could now be beyond EOF. Because |
npiggin@suse.de | 7bb46a6 | 2010-05-27 01:05:33 +1000 | [diff] [blame] | 2611 | * truncate writes the inode size before removing pages, once we have the |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2612 | * page lock we can determine safely if the page is beyond EOF. If it is not |
| 2613 | * beyond EOF, then the page is guaranteed safe against truncation until we |
| 2614 | * unlock the page. |
Jan Kara | ea13a86 | 2011-05-24 00:23:35 +0200 | [diff] [blame] | 2615 | * |
Jan Kara | 14da920 | 2012-06-12 16:20:37 +0200 | [diff] [blame] | 2616 | * Direct callers of this function should protect against filesystem freezing |
Ross Zwisler | 5c50002 | 2015-10-13 16:51:02 -0600 | [diff] [blame] | 2617 | * using sb_start_pagefault() - sb_end_pagefault() functions. |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2618 | */ |
Ross Zwisler | 5c50002 | 2015-10-13 16:51:02 -0600 | [diff] [blame] | 2619 | int block_page_mkwrite(struct vm_area_struct *vma, struct vm_fault *vmf, |
Jan Kara | 24da4fa | 2011-05-24 00:23:34 +0200 | [diff] [blame] | 2620 | get_block_t get_block) |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2621 | { |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2622 | struct folio *folio = page_folio(vmf->page); |
Al Viro | 496ad9a | 2013-01-23 17:07:38 -0500 | [diff] [blame] | 2623 | struct inode *inode = file_inode(vma->vm_file); |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2624 | unsigned long end; |
| 2625 | loff_t size; |
Jan Kara | 24da4fa | 2011-05-24 00:23:34 +0200 | [diff] [blame] | 2626 | int ret; |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2627 | |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2628 | folio_lock(folio); |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2629 | size = i_size_read(inode); |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2630 | if ((folio->mapping != inode->i_mapping) || |
| 2631 | (folio_pos(folio) >= size)) { |
Jan Kara | 24da4fa | 2011-05-24 00:23:34 +0200 | [diff] [blame] | 2632 | /* We overload EFAULT to mean page got truncated */ |
| 2633 | ret = -EFAULT; |
| 2634 | goto out_unlock; |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2635 | } |
| 2636 | |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2637 | end = folio_size(folio); |
| 2638 | /* folio is wholly or partially inside EOF */ |
| 2639 | if (folio_pos(folio) + end > size) |
| 2640 | end = size - folio_pos(folio); |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2641 | |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2642 | ret = __block_write_begin_int(folio, 0, end, get_block, NULL); |
Bean Huo | a524fcf | 2023-06-26 07:55:18 +0200 | [diff] [blame] | 2643 | if (unlikely(ret)) |
Jan Kara | 24da4fa | 2011-05-24 00:23:34 +0200 | [diff] [blame] | 2644 | goto out_unlock; |
Bean Huo | a524fcf | 2023-06-26 07:55:18 +0200 | [diff] [blame] | 2645 | |
| 2646 | __block_commit_write(folio, 0, end); |
| 2647 | |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2648 | folio_mark_dirty(folio); |
| 2649 | folio_wait_stable(folio); |
Jan Kara | 24da4fa | 2011-05-24 00:23:34 +0200 | [diff] [blame] | 2650 | return 0; |
| 2651 | out_unlock: |
Matthew Wilcox (Oracle) | fe18137 | 2023-06-12 22:01:34 +0100 | [diff] [blame] | 2652 | folio_unlock(folio); |
David Chinner | 5417169 | 2007-07-19 17:39:55 +1000 | [diff] [blame] | 2653 | return ret; |
| 2654 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2655 | EXPORT_SYMBOL(block_page_mkwrite); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2656 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2657 | int block_truncate_page(struct address_space *mapping, |
| 2658 | loff_t from, get_block_t *get_block) |
| 2659 | { |
Kirill A. Shutemov | 09cbfea | 2016-04-01 15:29:47 +0300 | [diff] [blame] | 2660 | pgoff_t index = from >> PAGE_SHIFT; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2661 | unsigned blocksize; |
Andrew Morton | 54b21a7 | 2006-01-08 01:03:05 -0800 | [diff] [blame] | 2662 | sector_t iblock; |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2663 | size_t offset, length, pos; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2664 | struct inode *inode = mapping->host; |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2665 | struct folio *folio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2666 | struct buffer_head *bh; |
Jiapeng Chong | dc7cb2d | 2023-03-23 10:32:59 +0800 | [diff] [blame] | 2667 | int err = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2668 | |
Fabian Frederick | 9340747 | 2017-02-27 14:28:32 -0800 | [diff] [blame] | 2669 | blocksize = i_blocksize(inode); |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2670 | length = from & (blocksize - 1); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2671 | |
| 2672 | /* Block boundary? Nothing to do */ |
| 2673 | if (!length) |
| 2674 | return 0; |
| 2675 | |
| 2676 | length = blocksize - length; |
Matthew Wilcox (Oracle) | 4b04646 | 2023-11-09 21:06:06 +0000 | [diff] [blame] | 2677 | iblock = ((loff_t)index * PAGE_SIZE) >> inode->i_blkbits; |
| 2678 | |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2679 | folio = filemap_grab_folio(mapping, index); |
| 2680 | if (IS_ERR(folio)) |
| 2681 | return PTR_ERR(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2682 | |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2683 | bh = folio_buffers(folio); |
Matthew Wilcox (Oracle) | 3decb85 | 2023-10-16 21:10:49 +0100 | [diff] [blame] | 2684 | if (!bh) |
Matthew Wilcox (Oracle) | 0a88810 | 2023-10-16 21:11:14 +0100 | [diff] [blame] | 2685 | bh = create_empty_buffers(folio, blocksize, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2686 | |
| 2687 | /* Find the buffer that contains "offset" */ |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2688 | offset = offset_in_folio(folio, from); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2689 | pos = blocksize; |
| 2690 | while (offset >= pos) { |
| 2691 | bh = bh->b_this_page; |
| 2692 | iblock++; |
| 2693 | pos += blocksize; |
| 2694 | } |
| 2695 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2696 | if (!buffer_mapped(bh)) { |
Badari Pulavarty | b0cf232 | 2006-03-26 01:38:00 -0800 | [diff] [blame] | 2697 | WARN_ON(bh->b_size != blocksize); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2698 | err = get_block(inode, iblock, bh, 0); |
| 2699 | if (err) |
| 2700 | goto unlock; |
| 2701 | /* unmapped? It's a hole - nothing to do */ |
| 2702 | if (!buffer_mapped(bh)) |
| 2703 | goto unlock; |
| 2704 | } |
| 2705 | |
| 2706 | /* Ok, it's mapped. Make sure it's up-to-date */ |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2707 | if (folio_test_uptodate(folio)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2708 | set_buffer_uptodate(bh); |
| 2709 | |
David Chinner | 33a266d | 2007-02-12 00:51:41 -0800 | [diff] [blame] | 2710 | if (!buffer_uptodate(bh) && !buffer_delay(bh) && !buffer_unwritten(bh)) { |
Zhang Yi | e7ea112 | 2022-09-01 21:34:54 +0800 | [diff] [blame] | 2711 | err = bh_read(bh, 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2712 | /* Uhhuh. Read error. Complain and punt. */ |
Zhang Yi | e7ea112 | 2022-09-01 21:34:54 +0800 | [diff] [blame] | 2713 | if (err < 0) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2714 | goto unlock; |
| 2715 | } |
| 2716 | |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2717 | folio_zero_range(folio, offset, length); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2718 | mark_buffer_dirty(bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2719 | |
| 2720 | unlock: |
Matthew Wilcox (Oracle) | 6d68f64 | 2023-06-12 22:01:41 +0100 | [diff] [blame] | 2721 | folio_unlock(folio); |
| 2722 | folio_put(folio); |
Jiapeng Chong | dc7cb2d | 2023-03-23 10:32:59 +0800 | [diff] [blame] | 2723 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2724 | return err; |
| 2725 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2726 | EXPORT_SYMBOL(block_truncate_page); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2727 | |
| 2728 | /* |
| 2729 | * The generic ->writepage function for buffer-backed address_spaces |
| 2730 | */ |
Matthew Wilcox (Oracle) | 17bf23a | 2023-12-15 20:02:44 +0000 | [diff] [blame] | 2731 | int block_write_full_folio(struct folio *folio, struct writeback_control *wbc, |
| 2732 | void *get_block) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2733 | { |
Matthew Wilcox (Oracle) | bb0ea59 | 2023-06-12 22:01:33 +0100 | [diff] [blame] | 2734 | struct inode * const inode = folio->mapping->host; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2735 | loff_t i_size = i_size_read(inode); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2736 | |
Matthew Wilcox (Oracle) | bb0ea59 | 2023-06-12 22:01:33 +0100 | [diff] [blame] | 2737 | /* Is the folio fully inside i_size? */ |
| 2738 | if (folio_pos(folio) + folio_size(folio) <= i_size) |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 2739 | return __block_write_full_folio(inode, folio, get_block, wbc); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2740 | |
Matthew Wilcox (Oracle) | bb0ea59 | 2023-06-12 22:01:33 +0100 | [diff] [blame] | 2741 | /* Is the folio fully outside i_size? (truncate in progress) */ |
| 2742 | if (folio_pos(folio) >= i_size) { |
Matthew Wilcox (Oracle) | 53418a1 | 2023-06-12 22:01:31 +0100 | [diff] [blame] | 2743 | folio_unlock(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2744 | return 0; /* don't care */ |
| 2745 | } |
| 2746 | |
| 2747 | /* |
Matthew Wilcox (Oracle) | bb0ea59 | 2023-06-12 22:01:33 +0100 | [diff] [blame] | 2748 | * The folio straddles i_size. It must be zeroed out on each and every |
Adam Buchbinder | 2a61aa4 | 2009-12-11 16:35:40 -0500 | [diff] [blame] | 2749 | * writepage invocation because it may be mmapped. "A file is mapped |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2750 | * in multiples of the page size. For a file that is not a multiple of |
Matthew Wilcox (Oracle) | bb0ea59 | 2023-06-12 22:01:33 +0100 | [diff] [blame] | 2751 | * the page size, the remaining memory is zeroed when mapped, and |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2752 | * writes to that region are not written out to the file." |
| 2753 | */ |
Matthew Wilcox (Oracle) | bb0ea59 | 2023-06-12 22:01:33 +0100 | [diff] [blame] | 2754 | folio_zero_segment(folio, offset_in_folio(folio, i_size), |
| 2755 | folio_size(folio)); |
Matthew Wilcox (Oracle) | 14059f6 | 2023-12-15 20:02:45 +0000 | [diff] [blame] | 2756 | return __block_write_full_folio(inode, folio, get_block, wbc); |
Chris Mason | 35c80d5 | 2009-04-15 13:22:38 -0400 | [diff] [blame] | 2757 | } |
Chris Mason | 35c80d5 | 2009-04-15 13:22:38 -0400 | [diff] [blame] | 2758 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2759 | sector_t generic_block_bmap(struct address_space *mapping, sector_t block, |
| 2760 | get_block_t *get_block) |
| 2761 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2762 | struct inode *inode = mapping->host; |
Alexander Potapenko | 2a527d6 | 2017-07-05 00:56:21 -0400 | [diff] [blame] | 2763 | struct buffer_head tmp = { |
| 2764 | .b_size = i_blocksize(inode), |
| 2765 | }; |
| 2766 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2767 | get_block(inode, block, &tmp, 0); |
| 2768 | return tmp.b_blocknr; |
| 2769 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2770 | EXPORT_SYMBOL(generic_block_bmap); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2771 | |
Christoph Hellwig | 4246a0b | 2015-07-20 15:29:37 +0200 | [diff] [blame] | 2772 | static void end_bio_bh_io_sync(struct bio *bio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2773 | { |
| 2774 | struct buffer_head *bh = bio->bi_private; |
| 2775 | |
Jens Axboe | b7c44ed | 2015-07-24 12:37:59 -0600 | [diff] [blame] | 2776 | if (unlikely(bio_flagged(bio, BIO_QUIET))) |
Keith Mannthey | 08bafc0 | 2008-11-25 10:24:35 +0100 | [diff] [blame] | 2777 | set_bit(BH_Quiet, &bh->b_state); |
| 2778 | |
Christoph Hellwig | 4e4cbee | 2017-06-03 09:38:06 +0200 | [diff] [blame] | 2779 | bh->b_end_io(bh, !bio->bi_status); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2780 | bio_put(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2781 | } |
| 2782 | |
Ritesh Harjani (IBM) | 5bdf402 | 2022-08-18 10:34:40 +0530 | [diff] [blame] | 2783 | static void submit_bh_wbc(blk_opf_t opf, struct buffer_head *bh, |
Bart Van Assche | 4498135 | 2024-02-02 12:39:25 -0800 | [diff] [blame] | 2784 | enum rw_hint write_hint, |
Ritesh Harjani (IBM) | 5bdf402 | 2022-08-18 10:34:40 +0530 | [diff] [blame] | 2785 | struct writeback_control *wbc) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2786 | { |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 2787 | const enum req_op op = opf & REQ_OP_MASK; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2788 | struct bio *bio; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2789 | |
| 2790 | BUG_ON(!buffer_locked(bh)); |
| 2791 | BUG_ON(!buffer_mapped(bh)); |
| 2792 | BUG_ON(!bh->b_end_io); |
Aneesh Kumar K.V | 8fb0e34 | 2009-05-12 16:22:37 -0400 | [diff] [blame] | 2793 | BUG_ON(buffer_delay(bh)); |
| 2794 | BUG_ON(buffer_unwritten(bh)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2795 | |
Jens Axboe | 48fd4f9 | 2008-08-22 10:00:36 +0200 | [diff] [blame] | 2796 | /* |
Jens Axboe | 48fd4f9 | 2008-08-22 10:00:36 +0200 | [diff] [blame] | 2797 | * Only clear out a write error when rewriting |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2798 | */ |
Mike Christie | 2a222ca | 2016-06-05 14:31:43 -0500 | [diff] [blame] | 2799 | if (test_set_buffer_req(bh) && (op == REQ_OP_WRITE)) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2800 | clear_buffer_write_io_error(bh); |
| 2801 | |
Christoph Hellwig | 07888c66 | 2022-01-24 10:11:05 +0100 | [diff] [blame] | 2802 | if (buffer_meta(bh)) |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 2803 | opf |= REQ_META; |
Christoph Hellwig | 07888c66 | 2022-01-24 10:11:05 +0100 | [diff] [blame] | 2804 | if (buffer_prio(bh)) |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 2805 | opf |= REQ_PRIO; |
Christoph Hellwig | 07888c66 | 2022-01-24 10:11:05 +0100 | [diff] [blame] | 2806 | |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 2807 | bio = bio_alloc(bh->b_bdev, 1, opf, GFP_NOIO); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2808 | |
Eric Biggers | 4f74d15 | 2020-07-02 01:56:07 +0000 | [diff] [blame] | 2809 | fscrypt_set_bio_crypt_ctx_bh(bio, bh, GFP_NOIO); |
| 2810 | |
Kent Overstreet | 4f024f3 | 2013-10-11 15:44:27 -0700 | [diff] [blame] | 2811 | bio->bi_iter.bi_sector = bh->b_blocknr * (bh->b_size >> 9); |
Bart Van Assche | 4498135 | 2024-02-02 12:39:25 -0800 | [diff] [blame] | 2812 | bio->bi_write_hint = write_hint; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2813 | |
Johannes Thumshirn | 741af75 | 2023-05-31 04:50:27 -0700 | [diff] [blame] | 2814 | __bio_add_page(bio, bh->b_page, bh->b_size, bh_offset(bh)); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2815 | |
| 2816 | bio->bi_end_io = end_bio_bh_io_sync; |
| 2817 | bio->bi_private = bh; |
| 2818 | |
Ming Lei | 83c9c54 | 2020-01-05 09:41:14 +0800 | [diff] [blame] | 2819 | /* Take care of bh's that straddle the end of the device */ |
| 2820 | guard_bio_eod(bio); |
| 2821 | |
Dennis Zhou | fd42df3 | 2018-12-05 12:10:34 -0500 | [diff] [blame] | 2822 | if (wbc) { |
| 2823 | wbc_init_bio(wbc, bio); |
Tejun Heo | 34e51a5 | 2019-06-27 13:39:49 -0700 | [diff] [blame] | 2824 | wbc_account_cgroup_owner(wbc, bh->b_page, bh->b_size); |
Dennis Zhou | fd42df3 | 2018-12-05 12:10:34 -0500 | [diff] [blame] | 2825 | } |
| 2826 | |
Mike Christie | 4e49ea4 | 2016-06-05 14:31:41 -0500 | [diff] [blame] | 2827 | submit_bio(bio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2828 | } |
Tejun Heo | bafc0db | 2015-06-02 08:37:23 -0600 | [diff] [blame] | 2829 | |
Ritesh Harjani (IBM) | 5bdf402 | 2022-08-18 10:34:40 +0530 | [diff] [blame] | 2830 | void submit_bh(blk_opf_t opf, struct buffer_head *bh) |
Tejun Heo | bafc0db | 2015-06-02 08:37:23 -0600 | [diff] [blame] | 2831 | { |
Bart Van Assche | 4498135 | 2024-02-02 12:39:25 -0800 | [diff] [blame] | 2832 | submit_bh_wbc(opf, bh, WRITE_LIFE_NOT_SET, NULL); |
Darrick J. Wong | 713685111 | 2013-04-29 15:07:25 -0700 | [diff] [blame] | 2833 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2834 | EXPORT_SYMBOL(submit_bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2835 | |
Bart Van Assche | 3ae7286 | 2022-07-14 11:07:12 -0700 | [diff] [blame] | 2836 | void write_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) |
Christoph Hellwig | 9cb569d | 2010-08-11 17:06:24 +0200 | [diff] [blame] | 2837 | { |
| 2838 | lock_buffer(bh); |
| 2839 | if (!test_clear_buffer_dirty(bh)) { |
| 2840 | unlock_buffer(bh); |
| 2841 | return; |
| 2842 | } |
| 2843 | bh->b_end_io = end_buffer_write_sync; |
| 2844 | get_bh(bh); |
Bart Van Assche | 1420c4a5 | 2022-07-14 11:07:13 -0700 | [diff] [blame] | 2845 | submit_bh(REQ_OP_WRITE | op_flags, bh); |
Christoph Hellwig | 9cb569d | 2010-08-11 17:06:24 +0200 | [diff] [blame] | 2846 | } |
| 2847 | EXPORT_SYMBOL(write_dirty_buffer); |
| 2848 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2849 | /* |
| 2850 | * For a data-integrity writeout, we need to wait upon any in-progress I/O |
| 2851 | * and then start new I/O and then wait upon it. The caller must have a ref on |
| 2852 | * the buffer_head. |
| 2853 | */ |
Bart Van Assche | 3ae7286 | 2022-07-14 11:07:12 -0700 | [diff] [blame] | 2854 | int __sync_dirty_buffer(struct buffer_head *bh, blk_opf_t op_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2855 | { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2856 | WARN_ON(atomic_read(&bh->b_count) < 1); |
| 2857 | lock_buffer(bh); |
| 2858 | if (test_clear_buffer_dirty(bh)) { |
Xianting Tian | 377254b | 2020-07-31 12:10:25 -0400 | [diff] [blame] | 2859 | /* |
| 2860 | * The bh should be mapped, but it might not be if the |
| 2861 | * device was hot-removed. Not much we can do but fail the I/O. |
| 2862 | */ |
| 2863 | if (!buffer_mapped(bh)) { |
| 2864 | unlock_buffer(bh); |
| 2865 | return -EIO; |
| 2866 | } |
| 2867 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2868 | get_bh(bh); |
| 2869 | bh->b_end_io = end_buffer_write_sync; |
Ritesh Harjani (IBM) | ab62062 | 2022-08-18 10:34:39 +0530 | [diff] [blame] | 2870 | submit_bh(REQ_OP_WRITE | op_flags, bh); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2871 | wait_on_buffer(bh); |
Ritesh Harjani (IBM) | ab62062 | 2022-08-18 10:34:39 +0530 | [diff] [blame] | 2872 | if (!buffer_uptodate(bh)) |
| 2873 | return -EIO; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2874 | } else { |
| 2875 | unlock_buffer(bh); |
| 2876 | } |
Ritesh Harjani (IBM) | ab62062 | 2022-08-18 10:34:39 +0530 | [diff] [blame] | 2877 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2878 | } |
Christoph Hellwig | 87e9951 | 2010-08-11 17:05:45 +0200 | [diff] [blame] | 2879 | EXPORT_SYMBOL(__sync_dirty_buffer); |
| 2880 | |
| 2881 | int sync_dirty_buffer(struct buffer_head *bh) |
| 2882 | { |
Christoph Hellwig | 70fd761 | 2016-11-01 07:40:10 -0600 | [diff] [blame] | 2883 | return __sync_dirty_buffer(bh, REQ_SYNC); |
Christoph Hellwig | 87e9951 | 2010-08-11 17:05:45 +0200 | [diff] [blame] | 2884 | } |
H Hartley Sweeten | 1fe72ea | 2009-09-22 16:43:51 -0700 | [diff] [blame] | 2885 | EXPORT_SYMBOL(sync_dirty_buffer); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2886 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2887 | static inline int buffer_busy(struct buffer_head *bh) |
| 2888 | { |
| 2889 | return atomic_read(&bh->b_count) | |
| 2890 | (bh->b_state & ((1 << BH_Dirty) | (1 << BH_Lock))); |
| 2891 | } |
| 2892 | |
Matthew Wilcox (Oracle) | 6439476 | 2022-05-01 06:39:50 -0400 | [diff] [blame] | 2893 | static bool |
| 2894 | drop_buffers(struct folio *folio, struct buffer_head **buffers_to_free) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2895 | { |
Matthew Wilcox (Oracle) | 6439476 | 2022-05-01 06:39:50 -0400 | [diff] [blame] | 2896 | struct buffer_head *head = folio_buffers(folio); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2897 | struct buffer_head *bh; |
| 2898 | |
| 2899 | bh = head; |
| 2900 | do { |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2901 | if (buffer_busy(bh)) |
| 2902 | goto failed; |
| 2903 | bh = bh->b_this_page; |
| 2904 | } while (bh != head); |
| 2905 | |
| 2906 | do { |
| 2907 | struct buffer_head *next = bh->b_this_page; |
| 2908 | |
Jan Kara | 535ee2f | 2008-02-08 04:21:59 -0800 | [diff] [blame] | 2909 | if (bh->b_assoc_map) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2910 | __remove_assoc_queue(bh); |
| 2911 | bh = next; |
| 2912 | } while (bh != head); |
| 2913 | *buffers_to_free = head; |
Matthew Wilcox (Oracle) | 6439476 | 2022-05-01 06:39:50 -0400 | [diff] [blame] | 2914 | folio_detach_private(folio); |
| 2915 | return true; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2916 | failed: |
Matthew Wilcox (Oracle) | 6439476 | 2022-05-01 06:39:50 -0400 | [diff] [blame] | 2917 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2918 | } |
| 2919 | |
Matthew Wilcox (Oracle) | b1888d1 | 2024-04-16 04:17:47 +0100 | [diff] [blame] | 2920 | /** |
| 2921 | * try_to_free_buffers - Release buffers attached to this folio. |
| 2922 | * @folio: The folio. |
| 2923 | * |
| 2924 | * If any buffers are in use (dirty, under writeback, elevated refcount), |
| 2925 | * no buffers will be freed. |
| 2926 | * |
| 2927 | * If the folio is dirty but all the buffers are clean then we need to |
| 2928 | * be sure to mark the folio clean as well. This is because the folio |
| 2929 | * may be against a block device, and a later reattachment of buffers |
| 2930 | * to a dirty folio will set *all* buffers dirty. Which would corrupt |
| 2931 | * filesystem data on the same device. |
| 2932 | * |
| 2933 | * The same applies to regular filesystem folios: if all the buffers are |
| 2934 | * clean then we set the folio clean and proceed. To do that, we require |
| 2935 | * total exclusion from block_dirty_folio(). That is obtained with |
| 2936 | * i_private_lock. |
| 2937 | * |
| 2938 | * Exclusion against try_to_free_buffers may be obtained by either |
| 2939 | * locking the folio or by holding its mapping's i_private_lock. |
| 2940 | * |
| 2941 | * Context: Process context. @folio must be locked. Will not sleep. |
| 2942 | * Return: true if all buffers attached to this folio were freed. |
| 2943 | */ |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2944 | bool try_to_free_buffers(struct folio *folio) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2945 | { |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2946 | struct address_space * const mapping = folio->mapping; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2947 | struct buffer_head *buffers_to_free = NULL; |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2948 | bool ret = 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2949 | |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2950 | BUG_ON(!folio_test_locked(folio)); |
| 2951 | if (folio_test_writeback(folio)) |
| 2952 | return false; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2953 | |
| 2954 | if (mapping == NULL) { /* can this still happen? */ |
Matthew Wilcox (Oracle) | 6439476 | 2022-05-01 06:39:50 -0400 | [diff] [blame] | 2955 | ret = drop_buffers(folio, &buffers_to_free); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2956 | goto out; |
| 2957 | } |
| 2958 | |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 2959 | spin_lock(&mapping->i_private_lock); |
Matthew Wilcox (Oracle) | 6439476 | 2022-05-01 06:39:50 -0400 | [diff] [blame] | 2960 | ret = drop_buffers(folio, &buffers_to_free); |
Linus Torvalds | ecdfc97 | 2007-01-26 12:47:06 -0800 | [diff] [blame] | 2961 | |
| 2962 | /* |
| 2963 | * If the filesystem writes its buffers by hand (eg ext3) |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2964 | * then we can have clean buffers against a dirty folio. We |
| 2965 | * clean the folio here; otherwise the VM will never notice |
Linus Torvalds | ecdfc97 | 2007-01-26 12:47:06 -0800 | [diff] [blame] | 2966 | * that the filesystem did any IO at all. |
| 2967 | * |
| 2968 | * Also, during truncate, discard_buffer will have marked all |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2969 | * the folio's buffers clean. We discover that here and clean |
| 2970 | * the folio also. |
Nick Piggin | 87df724 | 2007-01-30 14:36:27 +1100 | [diff] [blame] | 2971 | * |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 2972 | * i_private_lock must be held over this entire operation in order |
Matthew Wilcox (Oracle) | e621900 | 2022-02-09 20:22:12 +0000 | [diff] [blame] | 2973 | * to synchronise against block_dirty_folio and prevent the |
Nick Piggin | 87df724 | 2007-01-30 14:36:27 +1100 | [diff] [blame] | 2974 | * dirty bit from being lost. |
Linus Torvalds | ecdfc97 | 2007-01-26 12:47:06 -0800 | [diff] [blame] | 2975 | */ |
Tejun Heo | 11f81be | 2015-05-22 17:13:15 -0400 | [diff] [blame] | 2976 | if (ret) |
Matthew Wilcox (Oracle) | 68189fe | 2022-05-01 01:08:08 -0400 | [diff] [blame] | 2977 | folio_cancel_dirty(folio); |
Matthew Wilcox (Oracle) | 600f111 | 2023-11-17 21:58:23 +0000 | [diff] [blame] | 2978 | spin_unlock(&mapping->i_private_lock); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2979 | out: |
| 2980 | if (buffers_to_free) { |
| 2981 | struct buffer_head *bh = buffers_to_free; |
| 2982 | |
| 2983 | do { |
| 2984 | struct buffer_head *next = bh->b_this_page; |
| 2985 | free_buffer_head(bh); |
| 2986 | bh = next; |
| 2987 | } while (bh != buffers_to_free); |
| 2988 | } |
| 2989 | return ret; |
| 2990 | } |
| 2991 | EXPORT_SYMBOL(try_to_free_buffers); |
| 2992 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2993 | /* |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2994 | * Buffer-head allocation |
| 2995 | */ |
Alexey Dobriyan | 68279f9 | 2023-10-11 19:55:00 +0300 | [diff] [blame] | 2996 | static struct kmem_cache *bh_cachep __ro_after_init; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 2997 | |
| 2998 | /* |
| 2999 | * Once the number of bh's in the machine exceeds this level, we start |
| 3000 | * stripping them in writeback. |
| 3001 | */ |
Alexey Dobriyan | 68279f9 | 2023-10-11 19:55:00 +0300 | [diff] [blame] | 3002 | static unsigned long max_buffer_heads __ro_after_init; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3003 | |
| 3004 | int buffer_heads_over_limit; |
| 3005 | |
| 3006 | struct bh_accounting { |
| 3007 | int nr; /* Number of live bh's */ |
| 3008 | int ratelimit; /* Limit cacheline bouncing */ |
| 3009 | }; |
| 3010 | |
| 3011 | static DEFINE_PER_CPU(struct bh_accounting, bh_accounting) = {0, 0}; |
| 3012 | |
| 3013 | static void recalc_bh_state(void) |
| 3014 | { |
| 3015 | int i; |
| 3016 | int tot = 0; |
| 3017 | |
Christoph Lameter | ee1be86 | 2010-12-06 11:40:05 -0600 | [diff] [blame] | 3018 | if (__this_cpu_inc_return(bh_accounting.ratelimit) - 1 < 4096) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3019 | return; |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3020 | __this_cpu_write(bh_accounting.ratelimit, 0); |
Eric Dumazet | 8a14342 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 3021 | for_each_online_cpu(i) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3022 | tot += per_cpu(bh_accounting, i).nr; |
| 3023 | buffer_heads_over_limit = (tot > max_buffer_heads); |
| 3024 | } |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3025 | |
Al Viro | dd0fc66 | 2005-10-07 07:46:04 +0100 | [diff] [blame] | 3026 | struct buffer_head *alloc_buffer_head(gfp_t gfp_flags) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3027 | { |
Richard Kennedy | 019b4d1 | 2010-03-10 15:20:33 -0800 | [diff] [blame] | 3028 | struct buffer_head *ret = kmem_cache_zalloc(bh_cachep, gfp_flags); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3029 | if (ret) { |
Christoph Lameter | a35afb8 | 2007-05-16 22:10:57 -0700 | [diff] [blame] | 3030 | INIT_LIST_HEAD(&ret->b_assoc_buffers); |
Thomas Gleixner | f1e67e3 | 2019-11-18 14:28:24 +0100 | [diff] [blame] | 3031 | spin_lock_init(&ret->b_uptodate_lock); |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3032 | preempt_disable(); |
| 3033 | __this_cpu_inc(bh_accounting.nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3034 | recalc_bh_state(); |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3035 | preempt_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3036 | } |
| 3037 | return ret; |
| 3038 | } |
| 3039 | EXPORT_SYMBOL(alloc_buffer_head); |
| 3040 | |
| 3041 | void free_buffer_head(struct buffer_head *bh) |
| 3042 | { |
| 3043 | BUG_ON(!list_empty(&bh->b_assoc_buffers)); |
| 3044 | kmem_cache_free(bh_cachep, bh); |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3045 | preempt_disable(); |
| 3046 | __this_cpu_dec(bh_accounting.nr); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3047 | recalc_bh_state(); |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3048 | preempt_enable(); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3049 | } |
| 3050 | EXPORT_SYMBOL(free_buffer_head); |
| 3051 | |
Sebastian Andrzej Siewior | fc4d24c | 2016-11-03 15:49:57 +0100 | [diff] [blame] | 3052 | static int buffer_exit_cpu_dead(unsigned int cpu) |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3053 | { |
| 3054 | int i; |
| 3055 | struct bh_lru *b = &per_cpu(bh_lrus, cpu); |
| 3056 | |
| 3057 | for (i = 0; i < BH_LRU_SIZE; i++) { |
| 3058 | brelse(b->bhs[i]); |
| 3059 | b->bhs[i] = NULL; |
| 3060 | } |
Christoph Lameter | c7b9251 | 2010-12-06 11:16:28 -0600 | [diff] [blame] | 3061 | this_cpu_add(bh_accounting.nr, per_cpu(bh_accounting, cpu).nr); |
Eric Dumazet | 8a14342 | 2006-03-24 03:18:10 -0800 | [diff] [blame] | 3062 | per_cpu(bh_accounting, cpu).nr = 0; |
Sebastian Andrzej Siewior | fc4d24c | 2016-11-03 15:49:57 +0100 | [diff] [blame] | 3063 | return 0; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3064 | } |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3065 | |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3066 | /** |
Randy Dunlap | a6b9191 | 2008-03-19 17:01:00 -0700 | [diff] [blame] | 3067 | * bh_uptodate_or_lock - Test whether the buffer is uptodate |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3068 | * @bh: struct buffer_head |
| 3069 | * |
| 3070 | * Return true if the buffer is up-to-date and false, |
| 3071 | * with the buffer locked, if not. |
| 3072 | */ |
| 3073 | int bh_uptodate_or_lock(struct buffer_head *bh) |
| 3074 | { |
| 3075 | if (!buffer_uptodate(bh)) { |
| 3076 | lock_buffer(bh); |
| 3077 | if (!buffer_uptodate(bh)) |
| 3078 | return 0; |
| 3079 | unlock_buffer(bh); |
| 3080 | } |
| 3081 | return 1; |
| 3082 | } |
| 3083 | EXPORT_SYMBOL(bh_uptodate_or_lock); |
| 3084 | |
| 3085 | /** |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3086 | * __bh_read - Submit read for a locked buffer |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3087 | * @bh: struct buffer_head |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3088 | * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ |
| 3089 | * @wait: wait until reading finish |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3090 | * |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3091 | * Returns zero on success or don't wait, and -EIO on error. |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3092 | */ |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3093 | int __bh_read(struct buffer_head *bh, blk_opf_t op_flags, bool wait) |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3094 | { |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3095 | int ret = 0; |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3096 | |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3097 | BUG_ON(!buffer_locked(bh)); |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3098 | |
| 3099 | get_bh(bh); |
| 3100 | bh->b_end_io = end_buffer_read_sync; |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3101 | submit_bh(REQ_OP_READ | op_flags, bh); |
| 3102 | if (wait) { |
| 3103 | wait_on_buffer(bh); |
| 3104 | if (!buffer_uptodate(bh)) |
| 3105 | ret = -EIO; |
| 3106 | } |
| 3107 | return ret; |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3108 | } |
Zhang Yi | fdee117 | 2022-09-01 21:34:53 +0800 | [diff] [blame] | 3109 | EXPORT_SYMBOL(__bh_read); |
| 3110 | |
| 3111 | /** |
| 3112 | * __bh_read_batch - Submit read for a batch of unlocked buffers |
| 3113 | * @nr: entry number of the buffer batch |
| 3114 | * @bhs: a batch of struct buffer_head |
| 3115 | * @op_flags: appending REQ_OP_* flags besides REQ_OP_READ |
| 3116 | * @force_lock: force to get a lock on the buffer if set, otherwise drops any |
| 3117 | * buffer that cannot lock. |
| 3118 | * |
| 3119 | * Returns zero on success or don't wait, and -EIO on error. |
| 3120 | */ |
| 3121 | void __bh_read_batch(int nr, struct buffer_head *bhs[], |
| 3122 | blk_opf_t op_flags, bool force_lock) |
| 3123 | { |
| 3124 | int i; |
| 3125 | |
| 3126 | for (i = 0; i < nr; i++) { |
| 3127 | struct buffer_head *bh = bhs[i]; |
| 3128 | |
| 3129 | if (buffer_uptodate(bh)) |
| 3130 | continue; |
| 3131 | |
| 3132 | if (force_lock) |
| 3133 | lock_buffer(bh); |
| 3134 | else |
| 3135 | if (!trylock_buffer(bh)) |
| 3136 | continue; |
| 3137 | |
| 3138 | if (buffer_uptodate(bh)) { |
| 3139 | unlock_buffer(bh); |
| 3140 | continue; |
| 3141 | } |
| 3142 | |
| 3143 | bh->b_end_io = end_buffer_read_sync; |
| 3144 | get_bh(bh); |
| 3145 | submit_bh(REQ_OP_READ | op_flags, bh); |
| 3146 | } |
| 3147 | } |
| 3148 | EXPORT_SYMBOL(__bh_read_batch); |
Aneesh Kumar K.V | 389d1b0 | 2008-01-28 23:58:26 -0500 | [diff] [blame] | 3149 | |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3150 | void __init buffer_init(void) |
| 3151 | { |
Zhang Yanfei | 43be594 | 2013-02-22 16:35:46 -0800 | [diff] [blame] | 3152 | unsigned long nrpages; |
Sebastian Andrzej Siewior | fc4d24c | 2016-11-03 15:49:57 +0100 | [diff] [blame] | 3153 | int ret; |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3154 | |
Kunwu Chan | de8a320 | 2024-01-16 17:11:37 +0800 | [diff] [blame] | 3155 | bh_cachep = KMEM_CACHE(buffer_head, |
Chengming Zhou | c997d68 | 2024-02-24 13:53:15 +0000 | [diff] [blame] | 3156 | SLAB_RECLAIM_ACCOUNT|SLAB_PANIC); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3157 | /* |
| 3158 | * Limit the bh occupancy to 10% of ZONE_NORMAL |
| 3159 | */ |
| 3160 | nrpages = (nr_free_buffer_pages() * 10) / 100; |
| 3161 | max_buffer_heads = nrpages * (PAGE_SIZE / sizeof(struct buffer_head)); |
Sebastian Andrzej Siewior | fc4d24c | 2016-11-03 15:49:57 +0100 | [diff] [blame] | 3162 | ret = cpuhp_setup_state_nocalls(CPUHP_FS_BUFF_DEAD, "fs/buffer:dead", |
| 3163 | NULL, buffer_exit_cpu_dead); |
| 3164 | WARN_ON(ret < 0); |
Linus Torvalds | 1da177e | 2005-04-16 15:20:36 -0700 | [diff] [blame] | 3165 | } |