In hci_sock_getsockopt_old(), HCI_DATA_DIR and HCI_TIME_STAMP both store
their value into a local int and then call put_user(opt, optval). Because
optval is the function parameter typed char __user *, put_user sizes the
write from sizeof(*optval), so only the low byte of the int is copied to
userspace.
The matching setsockopt path reads sizeof(int) via copy_safe_from_sockptr,
so userspace passes a 4-byte buffer in both directions but previously got
back only one initialized byte on the read side.
Not sending this through 'net' tree given this bug is mostly invisble,
given opt is 0/1, and the last byte is being properly copied.
With this change, the upcoming translation to .getsockopt_iter becomes
mechanical.
FWIW: This behavior appeared in commit 1da177e4c3f4 ("Linux-2.6.12-rc2").
Signed-off-by: Breno Leitao <[email protected]>
---
net/bluetooth/hci_sock.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/net/bluetooth/hci_sock.c b/net/bluetooth/hci_sock.c
index 0290dea081f62..1823c06ba8940 100644
--- a/net/bluetooth/hci_sock.c
+++ b/net/bluetooth/hci_sock.c
@@ -2088,7 +2088,7 @@ static int hci_sock_getsockopt_old(struct socket *sock,
int level, int optname,
else
opt = 0;
- if (put_user(opt, optval))
+ if (put_user(opt, (int __user *)optval))
err = -EFAULT;
break;
@@ -2098,7 +2098,7 @@ static int hci_sock_getsockopt_old(struct socket *sock,
int level, int optname,
else
opt = 0;
- if (put_user(opt, optval))
+ if (put_user(opt, (int __user *)optval))
err = -EFAULT;
break;
--
2.53.0-Meta
