Re: [patch] sys_epoll_wait() timeout saga ...

[Davide Libenzi <davidel@xmailserver.org>]

On Sat, 24 Sep 2005, Nishanth Aravamudan wrote:

> Description: The @timeout parameter to ep_poll() is in milliseconds but
> we compare it to (MAX_SCHEDULE_TIMEOUT - 1000 / HZ), which is
> (jiffies/jiffies-per-sec) or seconds. That seems blatantly broken.  This
> led to improper overflow checking for @timeout. As Andrew Morton pointed
> out in a similar fix for sys_poll(), the best solution is to to check
> for potential overflow first, then either select an indefinite value or
> convert @timeout.
>
> To achieve this and clean-up the code, change the prototype of the
> ep_poll() to make it clear that the parameter is in milliseconds and
> rename jtimeout to timeout_jiffies to hold the corresonding jiffies
> value.
>
> fs/eventpoll.c |   46 ++++++++++++++++++++++++++++++++++++++--------
> 1 files changed, 38 insertions(+), 8 deletions(-)

Eh? "To achieve this and clean-up the code"? :)


- Davide


eventpoll.c |    7 +++++--
1 file changed, 5 insertions(+), 2 deletions(-)

--- a/fs/eventpoll.c	2005-09-24 11:07:04.000000000 -0700
+++ b/fs/eventpoll.c	2005-09-24 11:11:06.000000000 -0700
@@ -101,6 +101,10 @@
  /* Maximum number of poll wake up nests we are allowing */
  #define EP_MAX_POLLWAKE_NESTS 4

+/* Maximum msec timeout value storeable in a long int */
+#define EP_MAX_MSTIMEO min(1000ULL * MAX_SCHEDULE_TIMEOUT / HZ, LONG_MAX / HZ - 1000ULL)
+
+
  struct epoll_filefd {
  	struct file *file;
  	int fd;
@@ -1507,8 +1511,7 @@
  	 * and the overflow condition. The passed timeout is in milliseconds,
  	 * that why (t * HZ) / 1000.
  	 */
-	jtimeout = (timeout < 0 ||
-		    (timeout / 1000) >= (MAX_SCHEDULE_TIMEOUT / HZ)) ?
+	jtimeout = (timeout < 0 || timeout >= EP_MAX_MSTIMEO) ?
  		MAX_SCHEDULE_TIMEOUT: (timeout * HZ + 999) / 1000;

  retry: