With program:
char buf[32];
int main(int argc, char **argv)
{
int fd;
ssize_t cnt;
int err;
unsigned int flags;
fd = open("hi", O_RDWR | O_CREAT | O_TRUNC, 0644);
assert(fd >= 0);
cnt = write(fd, buf, sizeof(buf));
assert(cnt == sizeof(buf));
/* EXT4_JOURNAL_DATA_FL */
flags = 0x00004000;
/* FS_IOC_SETFLAGS */
err = ioctl(fd, _IOW('f', 2, long), &flags);
assert(err == 0);
/* EXT4_IOC_SWAP_BOOT */
err = ioctl(fd, _IO('f', 17), 0);
assert(err == 0);
close(fd);
fd = open("hi", O_RDWR);
assert(fd >= 0);
cnt = write(fd, buf, sizeof(buf));
close(fd);
return 0;
}
It will trigger a warning for the case of ext3fs or ext4fs with
nodealloc as follow:
[ 123.644524] EXT4-fs (vdb): mounting ext3 file system using the ext4 subsystem
[ 123.647408] EXT4-fs (vdb): mounted filesystem with ordered data mode. Opts: (null)
[ 138.323196] WARNING: CPU: 1 PID: 1130 at fs/ext4/ext4_jbd2.c:271 __ext4_handle_dirty_metadata+0x103/0x1a0
[ 138.323198] Modules linked in:
[ 138.323203] CPU: 1 PID: 1130 Comm: a.out Not tainted 5.0.0-rc2opt+ #62
[ 138.323205] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.12.0-0-ga698c8995f-prebuilt.qemu.org 04/01/2014
[ 138.323208] RIP: 0010:__ext4_handle_dirty_metadata+0x103/0x1a0
[ 138.323210] Code: 00 48 8b 40 68 48 89 90 d8 01 00 00 48 8b 4b 18 44 89 fa e8 ff c3 04 00 eb 84 48 89 df 45 31 ed e8 52 40 f9 ff e9 74 ff ff ff <0f> 0b 48 c7 c2 c0 88 e4 b1 45 89 e8 48 89 e9 44 89 fe 4c 89 f7 e8
[ 138.323211] RSP: 0018:ffffb997422bfc00 EFLAGS: 00010286
[ 138.323212] RAX: ffff9f0ab10ef800 RBX: ffff9f0a8cc74208 RCX: 0000000000000000
[ 138.323213] RDX: ffff9f0a8cc64000 RSI: ffff9f0a8cc74208 RDI: ffff9f0a8cc64000
[ 138.323214] RBP: ffff9f0a8cc64000 R08: ffff9f0a8cc74208 R09: ffffffffb1375300
[ 138.323215] R10: 0000000000000020 R11: ffff9f0a8cc74208 R12: 0000000000000000
[ 138.323216] R13: 00000000ffffff8b R14: ffffffffb1e496e8 R15: 0000000000000559
[ 138.323217] FS: 00007f878e152440(0000) GS:ffff9f0ab3a40000(0000) knlGS:0000000000000000
[ 138.323218] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 138.323219] CR2: 00007f878dcec395 CR3: 000000041bbc5000 CR4: 00000000000006e0
[ 138.323223] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 138.323223] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[ 138.323224] Call Trace:
[ 138.323282] write_end_fn+0x42/0x50
[ 138.323303] ext4_walk_page_buffers+0x72/0xa0
[ 138.323320] ? __ext4_expand_extra_isize+0x90/0x90
[ 138.323322] ext4_journalled_write_end+0xdb/0x510
[ 138.323335] ? copyin+0x22/0x30
[ 138.323355] generic_perform_write+0xfd/0x1b0
[ 138.323385] __generic_file_write_iter+0x196/0x1e0
[ 138.323402] ? generic_write_checks+0x4c/0xb0
[ 138.323404] ext4_file_write_iter+0xc7/0x400
[ 138.323439] ? tty_write+0x1bf/0x2e0
[ 138.323441] ? n_tty_open+0xa0/0xa0
[ 138.323453] __vfs_write+0x11e/0x1b0
[ 138.323479] vfs_write+0xb3/0x1b0
[ 138.323481] ksys_write+0x52/0xc0
[ 138.323487] do_syscall_64+0x55/0x170
[ 138.323523] entry_SYSCALL_64_after_hwframe+0x44/0xa9
[ 138.323555] RIP: 0033:0x7f878dc6b130
[ 138.323556] Code: 73 01 c3 48 8b 0d 58 ed 2c 00 f7 d8 64 89 01 48 83 c8 ff c3 66 0f 1f 44 00 00 83 3d b9 45 2d 00 00 75 10 b8 01 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 31 c3 48 83 ec 08 e8 3e f3 01 00 48 89 04 24
[ 138.323557] RSP: 002b:00007ffe8104ecc8 EFLAGS: 00000246 ORIG_RAX: 0000000000000001
[ 138.323559] RAX: ffffffffffffffda RBX: 00007f878dd035b0 RCX: 00007f878dc6b130
[ 138.323560] RDX: 0000000000000020 RSI: 0000000000601080 RDI: 0000000000000003
[ 138.323560] RBP: 00007ffe8104ed10 R08: 0000000000000000 R09: 0000000000000000
[ 138.323561] R10: 0000000000000000 R11: 0000000000000246 R12: 0000000000400610
[ 138.323562] R13: 00007ffe8104edf0 R14: 0000000000000000 R15: 0000000000000000
[ 138.323564] ---[ end trace 8c5d15ab55f9bea9 ]---
[ 138.323586] EXT4-fs: write_end_fn:1369: aborting transaction: Corrupt filesystem in __ext4_handle_dirty_metadata
[ 138.326177] EXT4: jbd2_journal_dirty_metadata failed: handle type 2 started at line 1289, credits 19/17, errcode -117
[ 138.326231] EXT4-fs error (device vdb) in ext4_do_update_inode:5362: Readonly filesystem
[ 138.329147] EXT4-fs error (device vdb) in ext4_journalled_write_end:1550: Corrupt filesystem
The reason is that while swapping two inode, we swap the flags too. In this
program, file 'hi' has change to journal_data mode after the first ioctl, but
with flag swap after the second ioctl, ext4_should_journal_data in
ext4_write_begin will return false, so there won't be any journal_head append
to this buffer_head, and while do ext4_journalled_write_end, it will trigger
the warning since the buffer_jbd check fail in jbd2_journal_dirty_metadata.
We can fix this by reset aops of the address_space, but it's prefer to
set a mask to distinguish the flags which should be swap or not.
Signed-off-by: yangerkun <yangerkun@xxxxxxxxxx>
---
fs/ext4/ext4.h | 4 ++++
fs/ext4/ioctl.c | 6 +++++-
2 files changed, 9 insertions(+), 1 deletion(-)
diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 185a05d..30f782e 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -426,6 +426,10 @@ struct flex_groups {
/* Flags that are appropriate for non-directories/regular files. */
#define EXT4_OTHER_FLMASK (EXT4_NODUMP_FL | EXT4_NOATIME_FL)
+/* Flags that should be swap */
+#define EXT4_FL_SHOULD_SWAP (EXT4_COMPR_FL | EXT4_COMPRBLK_FL | EXT4_NOCOMPR_FL |\
+ EXT4_HUGE_FILE_FL | EXT4_EXTENTS_FL)
+
/* Mask out flags that are inappropriate for the given type of inode. */
static inline __u32 ext4_mask_flags(umode_t mode, __u32 flags)
{
diff --git a/fs/ext4/ioctl.c b/fs/ext4/ioctl.c
index eff6835..2e76fb5 100644
--- a/fs/ext4/ioctl.c
+++ b/fs/ext4/ioctl.c
@@ -63,6 +63,7 @@ static void swap_inode_data(struct inode *inode1, struct inode *inode2)
loff_t isize;
struct ext4_inode_info *ei1;
struct ext4_inode_info *ei2;
+ unsigned long tmp;
ei1 = EXT4_I(inode1);
ei2 = EXT4_I(inode2);
@@ -72,7 +73,10 @@ static void swap_inode_data(struct inode *inode1, struct inode *inode2)
swap(inode1->i_mtime, inode2->i_mtime);
memswap(ei1->i_data, ei2->i_data, sizeof(ei1->i_data));
- swap(ei1->i_flags, ei2->i_flags);
+ tmp = ei1->i_flags & EXT4_FL_SHOULD_SWAP;
+ ei1->i_flags = (ei2->i_flags & EXT4_FL_SHOULD_SWAP) |
+ (ei1->i_flags & ~EXT4_FL_SHOULD_SWAP);
+ ei2->i_flags = tmp | (ei2->i_flags & ~EXT4_FL_SHOULD_SWAP);
swap(ei1->i_disksize, ei2->i_disksize);
ext4_es_remove_extent(inode1, 0, EXT_MAX_BLOCKS);
ext4_es_remove_extent(inode2, 0, EXT_MAX_BLOCKS);
--
2.9.5
