Bug 31255
| Summary: | select with zero timeout returns 0 even when data pending | ||
|---|---|---|---|
| Product: | Base System | Reporter: | root <dozen> |
| Component: | kern | Assignee: | freebsd-bugs (Nobody) <bugs> |
| Status: | Closed FIXED | ||
| Severity: | Affects Only Me | ||
| Priority: | Normal | ||
| Version: | 4.3-RELEASE | ||
| Hardware: | Any | ||
| OS: | Any | ||
On 13 Oct 2001 root@do-labs.spb.ru wrote: > >Description: > struct timeval tv = { 0,0 }; > rc = select(...,&tv); > > This code will always return zero even if data is pending > one any descriptor or any descriptor is ready to write. > This is because check for timed out performed before > asking kernel about descriptor state. I can't see this. The descriptors are checked before the timeout. The timeout is just copyin()'ed and (bogusly) validated early, so select() may return an error for certain problems with the timeout even when there is a ready descriptor so there is no need to use the timeout. > Single Unix Spec do not states exactly that zero timeout shall > lead to immediate return with correct descriptor number as result, > but it looks like it is supposed to be as such. POSIX.1-200x-draft7 seems to be exact enough. It says that if none of the descriptors are ready, then select() shall block until one becomes ready or the timeout expires, etc. It doesn't say anything about using the timeout if a descriptor is ready. Bruce State Changed From-To: open->closed Using select with a zero timeout to poll for ready descriptors works fine for me, and many things depend on it working. Without example code that demonstrates it not working, I can only assume it was a bug in your code that calls select. |
struct timeval tv = { 0,0 }; rc = select(...,&tv); This code will always return zero even if data is pending one any descriptor or any descriptor is ready to write. This is because check for timed out performed before asking kernel about descriptor state. Single Unix Spec do not states exactly that zero timeout shall lead to immediate return with correct descriptor number as result, but it looks like it is supposed to be as such. Fix: The problem code is in /usr/src/sys/kern/sys_generic.c