Message ID | b8523c3a-0eb2-3aeb-fdf9-2e9d6d320d5c@virtuozzo.com |
---|---|
State | Rejected |
Headers | show |
Series | ext4: number of long lived errors | expand |
On 10/31/2018 04:30 AM, Andreas Dilger wrote: > Could you please explain your statement below that on-stack > initialization does not zero unspecified fields? According > to documents I found, for example: > > https://gcc.gnu.org/onlinedocs/gcc/Designated-Inits.html > > they *are* initialized to zero. I did not know it, I re-checked it in generated assembler code and found that you are right and I was wrong. Please drop this patch, should I resend of rest of this patch set once again? Thank you, Vasily Averin
On Oct 30, 2018, at 9:39 PM, Vasily Averin <vvs@virtuozzo.com> wrote: > > On 10/31/2018 04:30 AM, Andreas Dilger wrote: >> Could you please explain your statement below that on-stack >> initialization does not zero unspecified fields? According >> to documents I found, for example: >> >> https://gcc.gnu.org/onlinedocs/gcc/Designated-Inits.html >> >> they *are* initialized to zero. > > I did not know it, > I re-checked it in generated assembler code and found that you > are right and I was wrong. > > Please drop this patch, > should I resend of rest of this patch set once again? I don't think it is necessary to resend the whole patch series. Ted should notice these emails on this patch and not apply it. Cheers, Andreas
diff --git a/fs/ext4/xattr.c b/fs/ext4/xattr.c index dc1aeab06dba..aae12425597e 100644 --- a/fs/ext4/xattr.c +++ b/fs/ext4/xattr.c @@ -2303,9 +2303,11 @@ ext4_xattr_set_handle(handle_t *handle, struct inode *inode, int name_index, }; struct ext4_xattr_ibody_find is = { .s = { .not_found = -ENODATA, }, + .iloc = { .bh = NULL, }, }; struct ext4_xattr_block_find bs = { .s = { .not_found = -ENODATA, }, + .bh = NULL, }; int no_expand; int error;
On-stack initialization does not guarantee zeroying of unintialized fields. So is.iloc.bh and bs.bh can be contain garbage of old stack conent. Errors in the beginning of ext4_xattr_set_handle() function lead to jump to "cleanup:" label where brelse(is.iloc.bh) and brelse(bs.bh) can access uninitialized bh fields of on-stack located "is" and "bs" structures. Issue was inherited from ext3 and was present in first ext4 commit. Fixes ac27a0ec112a ("ext4: initial copy of files from ext3") # 2.6.19 Signed-off-by: Vasily Averin <vvs@virtuozzo.com> --- fs/ext4/xattr.c | 2 ++ 1 file changed, 2 insertions(+)