projects / thirdparty / kernel / stable-queue.git / blame
| Commit | Line | Data |
|---|---|---|
| d992de84 GKH |
1 | From 51e3ae81ec58e95f10a98ef3dd6d7bce5d8e35a2 Mon Sep 17 00:00:00 2001 |
| 2 | From: Theodore Ts'o <tytso@mit.edu> | |
| 3 | Date: Fri, 6 Oct 2017 23:09:55 -0400 | |
| 4 | Subject: ext4: fix interaction between i_size, fallocate, and delalloc after a crash | |
| 5 | ||
| 6 | From: Theodore Ts'o <tytso@mit.edu> | |
| 7 | ||
| 8 | commit 51e3ae81ec58e95f10a98ef3dd6d7bce5d8e35a2 upstream. | |
| 9 | ||
| 10 | If there are pending writes subject to delayed allocation, then i_size | |
| 11 | will show size after the writes have completed, while i_disksize | |
| 12 | contains the value of i_size on the disk (since the writes have not | |
| 13 | been persisted to disk). | |
| 14 | ||
| 15 | If fallocate(2) is called with the FALLOC_FL_KEEP_SIZE flag, either | |
| 16 | with or without the FALLOC_FL_ZERO_RANGE flag set, and the new size | |
| 17 | after the fallocate(2) is between i_size and i_disksize, then after a | |
| 18 | crash, if a journal commit has resulted in the changes made by the | |
| 19 | fallocate() call to be persisted after a crash, but the delayed | |
| 20 | allocation write has not resolved itself, i_size would not be updated, | |
| 21 | and this would cause the following e2fsck complaint: | |
| 22 | ||
| 23 | Inode 12, end of extent exceeds allowed value | |
| 24 | (logical block 33, physical block 33441, len 7) | |
| 25 | ||
| 26 | This can only take place on a sparse file, where the fallocate(2) call | |
| 27 | is allocating blocks in a range which is before a pending delayed | |
| 28 | allocation write which is extending i_size. Since this situation is | |
| 29 | quite rare, and the window in which the crash must take place is | |
| 30 | typically < 30 seconds, in practice this condition will rarely happen. | |
| 31 | ||
| 32 | Nevertheless, it can be triggered in testing, and in particular by | |
| 33 | xfstests generic/456. | |
| 34 | ||
| 35 | Signed-off-by: Theodore Ts'o <tytso@mit.edu> | |
| 36 | Reported-by: Amir Goldstein <amir73il@gmail.com> | |
| 37 | Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org> | |
| 38 | ||
| 39 | --- | |
| 40 | fs/ext4/extents.c | 6 ++++-- | |
| 41 | 1 file changed, 4 insertions(+), 2 deletions(-) | |
| 42 | ||
| 43 | --- a/fs/ext4/extents.c | |
| 44 | +++ b/fs/ext4/extents.c | |
| 45 | @@ -4807,7 +4807,8 @@ static long ext4_zero_range(struct file | |
| 46 | } | |
| 47 | ||
| 48 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
| 49 | - offset + len > i_size_read(inode)) { | |
| 50 | + (offset + len > i_size_read(inode) || | |
| 51 | + offset + len > EXT4_I(inode)->i_disksize)) { | |
| 52 | new_size = offset + len; | |
| 53 | ret = inode_newsize_ok(inode, new_size); | |
| 54 | if (ret) | |
| 55 | @@ -4951,7 +4952,8 @@ long ext4_fallocate(struct file *file, i | |
| 56 | } | |
| 57 | ||
| 58 | if (!(mode & FALLOC_FL_KEEP_SIZE) && | |
| 59 | - offset + len > i_size_read(inode)) { | |
| 60 | + (offset + len > i_size_read(inode) || | |
| 61 | + offset + len > EXT4_I(inode)->i_disksize)) { | |
| 62 | new_size = offset + len; | |
| 63 | ret = inode_newsize_ok(inode, new_size); | |
| 64 | if (ret) |