Re: [PATCH RFC] ubi: gluebi: Fix NULL pointer dereference caused by ftl notifier
From: Zhihao Cheng
Date: Thu Oct 12 2023 - 07:26:00 EST
在 2023/10/12 17:31, ZhaoLong Wang 写道:
3. P1 P2
gluebi_create -> mtd_device_register -> add_mtd_device:
device_register // dev/mtd1 is visible
fd = open(/dev/mtd1, O_WRONLY)
gluebi_get_device
gluebi->desc = ubi_open_volume
ftl_add_mtd
mtd_read
gluebi_read
gluebi->desc is not ERR_PTR/NULL
close(fd)
gluebi_put_device
ubi_close_volume
kfree(desc)
ubi_read(gluebi->desc) // UAF (×)
Yes, it's also a problem. Perhaps it should be set to NULL after
destroying gluebi->desc.
The key point is that 'gluebi->desc' check & usage is not atomic in
gluebi_read. So following patch still can't handle situation 3.
Setting the desc to NULL works because
mutex_lock "mtd_table_mutex" is held on all paths where
ftl_add_mtd() is called.
Oh, you're right. Just one nit below:
> @@ -154,9 +159,26 @@ static int gluebi_read(struct mtd_info *mtd, loff_t
> from, size_t len,
> size_t *retlen, unsigned char *buf)
> {
> int err = 0, lnum, offs, bytes_left;
> - struct gluebi_device *gluebi;
> + struct gluebi_device *gluebi = container_of(mtd, struct
gluebi_device,
> + mtd);
> + int isnt_get = unlikely(gluebi->desc == NULL) ? 1 : 0;
This 'unlikey' can be removed.
Rename 'isnt_get' as 'has_desc' ?
> + /**
> + * In normal case, the UBI volume desc has been initialized by
> + * ->_get_device(). However, in the ftl notifier process, the
> + * ->_get_device() is not executed in advance and the MTD device
> + * is directly scanned which cause null pointe dereference.
> + * Therefore, try to get the MTD device here.
> + */
> + if (unlikely(isnt_get)) {
> + err = __get_mtd_device(mtd);
> + if (err) {
> + err_msg("cannot get MTD device %d, UBI device %d, volume
> %d, error %d",
> + mtd->index, gluebi->ubi_num, gluebi->vol_id, err);
> + return err;
> + }
> + }