On 29.09.2014 01:35, Kees Cook wrote:
When building with CONFIG_DEBUG_STRICT_USER_COPY_CHECKS, this copy_to_user
wasn't provably correct (the length was in a variable but the structure
wasn't). The compiler wasn't able to check that the length was being
statically assigned, so include a BUG_ON to notice if this ever changes.

Signed-off-by: Kees Cook <[email protected]>
---
  fs/notify/fanotify/fanotify_user.c | 1 +
  1 file changed, 1 insertion(+)

diff --git a/fs/notify/fanotify/fanotify_user.c 
b/fs/notify/fanotify/fanotify_user.c
index b13992a41bd9..7bdb43222875 100644
--- a/fs/notify/fanotify/fanotify_user.c
+++ b/fs/notify/fanotify/fanotify_user.c
@@ -216,6 +216,7 @@ static ssize_t copy_event_to_user(struct fsnotify_group 
*group,

        fd = fanotify_event_metadata.fd;
        ret = -EFAULT;
+       BUG_ON(fanotify_event_metadata.event_len != FAN_EVENT_METADATA_LEN);


Thank you for indicating an inconsistency in fanotify_user.c

In fanotify_user.c a mismatch exists between the assumptions made in get_one_event() and copy_event_to_user().

According to the fanotify.7 man page:
"FAN_EVENT_METADATA_LEN - This is the minimum size (and currently the only size) of any event metadata."

get_one_event() assumes that the event metadata has a length of FAN_EVENT_METADATA_LEN and does not care about the value of event_len.

copy_event_to_user() assumes that event_len specifies the length of the event metadata and does not care about FAN_EVENT_METADATA_LEN.

Due to these different assumptions it cannot be proven inside fanotify_user.c that copy_to_user() is called correctly.

If we want such a prove inside fanotify_user.c, we should change get_one_event() to use the following:

// Get the event without removing it from the group.
fsnotify_event *evt = fsnotify_peek_notify_event(group);
// Check we have enough user memory to copy the event.
if (evt->event_len > count)
    return ERR_PTR(-EINVAL);
// Remove the event from the group.
fsnotify_remove_notify_event(group);
// Return the pointer we tested.
// fanotify_remove_notify_event() should return the same pointer.
return evt;

If after this change you still want a BUG_ON, pass count to copy_event_to_user() and write
BUG_ON(fanotify_event_metadata.event_len > count);

Best regards

Heinrich Schuchardt

        if (copy_to_user(buf, &fanotify_event_metadata,
                         fanotify_event_metadata.event_len))
                goto out_close_fd;


--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to