[PATCH] ext4: fix the time handling macros when ext4 is using small inodes

[PATCH] ext4: fix the time handling macros when ext4 is using small inodes

[Jeff Layton]

If ext4 is using small on-disk inodes, then it may not be able to store
fine grained timestamps. It also can't store the i_crtime at all in that
case since that fully lives in the extended part of the inode.

979492850abd got the EXT4_EINODE_{GET,SET}_XTIME macros wrong, and would
still store the tv_sec field of the i_crtime into the raw_inode, even
when they were small, corrupting adjacent memory.

This fixes those macros to skip setting anything in the raw_inode if the
tv_sec field doesn't fit, and to properly return a {0,0} timestamp when
the raw_inode doesn't support it.

Cc: Jan Kara <jack@suse.cz>
Fixes: 979492850abd ("ext4: convert to ctime accessor functions")
Reported-by: Hugh Dickins <hughd@google.com>
Signed-off-by: Jeff Layton <jlayton@kernel.org>
---
 fs/ext4/ext4.h | 17 ++++++++++++-----
 1 file changed, 12 insertions(+), 5 deletions(-)

diff --git a/fs/ext4/ext4.h b/fs/ext4/ext4.h
index 2af347669db7..1e2259d9967d 100644
--- a/fs/ext4/ext4.h
+++ b/fs/ext4/ext4.h
@@ -900,8 +900,10 @@ do {										\
 #define EXT4_INODE_SET_CTIME(inode, raw_inode)					\
 	EXT4_INODE_SET_XTIME_VAL(i_ctime, inode, raw_inode, inode_get_ctime(inode))
 
-#define EXT4_EINODE_SET_XTIME(xtime, einode, raw_inode)			       \
-	EXT4_INODE_SET_XTIME_VAL(xtime, &((einode)->vfs_inode), raw_inode, (einode)->xtime)
+#define EXT4_EINODE_SET_XTIME(xtime, einode, raw_inode)				\
+	if (EXT4_FITS_IN_INODE(raw_inode, einode, xtime))			\
+		EXT4_INODE_SET_XTIME_VAL(xtime, &((einode)->vfs_inode),		\
+					 raw_inode, (einode)->xtime)
 
 #define EXT4_INODE_GET_XTIME_VAL(xtime, inode, raw_inode)			\
 	(EXT4_FITS_IN_INODE(raw_inode, EXT4_I(inode), xtime ## _extra) ?	\
@@ -922,9 +924,14 @@ do {										\
 		EXT4_INODE_GET_XTIME_VAL(i_ctime, inode, raw_inode));		\
 } while (0)
 
-#define EXT4_EINODE_GET_XTIME(xtime, einode, raw_inode)			       \
-do {									       \
-	(einode)->xtime = EXT4_INODE_GET_XTIME_VAL(xtime, &(einode->vfs_inode), raw_inode);	\
+#define EXT4_EINODE_GET_XTIME(xtime, einode, raw_inode)				\
+do {										\
+	if (EXT4_FITS_IN_INODE(raw_inode, einode, xtime)) 			\
+		(einode)->xtime =						\
+			EXT4_INODE_GET_XTIME_VAL(xtime, &(einode->vfs_inode),	\
+						 raw_inode);			\
+	else									\
+		(einode)->xtime = (struct timespec64){0, 0};			\
 } while (0)
 
 #define i_disk_version osd1.linux1.l_i_version

---
base-commit: c62e19541f8bb39f1f340247f651afe4532243df
change-id: 20230718-ctime-f140dae8789d

Best regards,
-- 
Jeff Layton <jlayton@kernel.org>

Re: [PATCH] ext4: fix the time handling macros when ext4 is using small inodes

[Jan Kara]

On Tue 18-07-23 13:31:59, Jeff Layton wrote:
> If ext4 is using small on-disk inodes, then it may not be able to store
> fine grained timestamps. It also can't store the i_crtime at all in that
> case since that fully lives in the extended part of the inode.
> 
> 979492850abd got the EXT4_EINODE_{GET,SET}_XTIME macros wrong, and would
> still store the tv_sec field of the i_crtime into the raw_inode, even
> when they were small, corrupting adjacent memory.
> 
> This fixes those macros to skip setting anything in the raw_inode if the
> tv_sec field doesn't fit, and to properly return a {0,0} timestamp when
> the raw_inode doesn't support it.
> 
> Cc: Jan Kara <jack@suse.cz>
> Fixes: 979492850abd ("ext4: convert to ctime accessor functions")
> Reported-by: Hugh Dickins <hughd@google.com>
> Signed-off-by: Jeff Layton <jlayton@kernel.org>

I've seen Christian has already folded this fixup so I just want to say:
Thanks for fixing this up while I was away!

								Honza
-- 
Jan Kara <jack@suse.com>
SUSE Labs, CR