Received-SPF: pass (google.com: domain of linux-bluetooth-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
MIME-Version: 1.0
References: <20210128063901.1362-1-sonnysasaka@chromium.org>
In-Reply-To: <20210128063901.1362-1-sonnysasaka@chromium.org>
From:   Luiz Augusto von Dentz <luiz.dentz@gmail.com>
Date:   Wed, 27 Jan 2021 22:54:01 -0800
Message-ID: <CABBYNZJNq9k0vwKVeJrPcK9t43zYE33Ct9SF0OGAJKMfXq3mPw@mail.gmail.com>
Subject: Re: [PATCH BlueZ v2] input/hog: Fix crashes of UAF of hog->attr
To:     Sonny Sasaka <sonnysasaka@chromium.org>
Cc:     "linux-bluetooth@vger.kernel.org" <linux-bluetooth@vger.kernel.org>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Sonny,

On Wed, Jan 27, 2021 at 10:43 PM Sonny Sasaka <sonnysasaka@chromium.org> wrote:
>
> hog->attr does not own pointer, so it may be invalid when hog object
> gets re-attached at bt_hog_attach(). To solve this, this patch
> intentionally clears hog->attr at bt_hog_detach() to mark it as invalid
> so that it can be repopulated with the valid pointer at bt_hog_attach().
> The same applies to all sub-instances.
>
> Sample stack trace:
> * thread #1, stop reason = signal SIGSEGV
>   * frame #0: 0x05ad49f2 bluetoothd`<name omitted> at gatt-db.c:1428
>     frame #1: 0x05a91922 bluetoothd`bt_hog_attach at hog-lib.c:1694
>     frame #2: 0x05a9160e bluetoothd`hog_accept at hog.c:212
>     frame #3: 0x05ab4784 bluetoothd`service_accept at service.c:203
>     frame #4: 0x05aba1e6 bluetoothd`device_attach_att at device.c:4542
>     frame #5: 0x05a9c4a2 bluetoothd`connect_cb at gatt-database.c:656
>     frame #6: 0x05a98e8c bluetoothd`server_cb at btio.c:264
>     frame #7: 0xec8e6a1a libglib-2.0.so.0`g_main_context_dispatch at gmain.c:3325
>     frame #8: 0xec8e6c58 libglib-2.0.so.0`g_main_context_iterate at gmain.c:4119
>     frame #9: 0xec8e6e52 libglib-2.0.so.0`g_main_loop_run at gmain.c:4317
>     frame #10: 0x05ad582e bluetoothd`mainloop_run at mainloop-glib.c:79
>     frame #11: 0x05ad5a64 bluetoothd`mainloop_run_with_signal at mainloop-notify.c:201
>     frame #12: 0x05ac35ac bluetoothd`main at main.c:1103
>     frame #13: 0xec6ed0a2 libc.so.6`__libc_start_main at libc-start.c:308
>
> ---
>  profiles/input/hog-lib.c | 16 ++++++++++++++++
>  1 file changed, 16 insertions(+)
>
> diff --git a/profiles/input/hog-lib.c b/profiles/input/hog-lib.c
> index 1f132aa4c..711bda73c 100644
> --- a/profiles/input/hog-lib.c
> +++ b/profiles/input/hog-lib.c
> @@ -1651,12 +1651,19 @@ static void primary_cb(uint8_t status, GSList *services, void *user_data)
>  bool bt_hog_attach(struct bt_hog *hog, void *gatt)
>  {
>         GSList *l;
> +       bt_uuid_t uuid;
>
>         if (hog->attrib)
>                 return false;
>
>         hog->attrib = g_attrib_ref(gatt);
>
> +       if (!hog->attr && hog->gatt_db) {
> +               bt_uuid16_create(&uuid, HOG_UUID16);
> +               gatt_db_foreach_service(hog->gatt_db, &uuid,
> +                                       foreach_hog_service, hog);
> +       }
> +
>         if (!hog->attr && !hog->primary) {
>                 discover_primary(hog, hog->attrib, NULL, primary_cb, hog);
>                 return true;
> @@ -1744,6 +1751,15 @@ void bt_hog_detach(struct bt_hog *hog)
>                 bt_hog_detach(instance);
>         }
>
> +       /* hog->attr doesn't own pointer, so it may be invalid when this hog
> +        * object gets re-attached with bt_hog_attach(). So intentionally mark
> +        * it as invalid and remove all instances so that the instances can be
> +        * re-attached at bt_hog_attach().
> +        */
> +       hog->attr = NULL;
> +       g_slist_free_full(hog->instances, hog_free);
> +       hog->instances = NULL;

We can actually what watches instead of cleanup always like this, see:

https://git.kernel.org/pub/scm/bluetooth/bluez.git/tree/src/shared/gatt-db.h#n275

> +
>         for (l = hog->reports; l; l = l->next) {
>                 struct report *r = l->data;
>
> --
> 2.29.2
>


-- 
Luiz Augusto von Dentz