Skip to contentSkip to conversation

negative timeout can be set up by setsockopt system call

DISCUSSION Ram Gupta 2 participants 2 messages Nov 7, 2005

lore
Share & tools
lore

Ram Gupta

author
#0 lore raw
Quoted 9 lines · Ram Gupta

On 11/4/05, Nish Aravamudan <[email protected]>

Quoted 5 lines

In Ram's specific case, I think, the call path is sys_setsockopt() -> sock_setsockopt() -> sock_set_timeout, which has a definition of:

static int sock_set_timeout(long *timeo_p, char __user *optval, int

optlen)

Quoted 1 line
Quoted 1 line

Exactly right.

Ok.

Quoted 11 lines
Quoted 6 lines

Ram, what is the expected behavior of negative values in the timeval? And what are you seeing happen right now?

As of 2.6.14, looks like we convert any non-zero values into jiffies and store them in sk->sk_{rcv,snd}timeo...

I don't see any problem from the kernel side but the application
times out immediately causing certain failures as the schedule_timeout returns immediately in case of negative values. Shouldn't there be a check for negative values and return error to the application so that it can handle it.

I mean more along the lines of what does a man-page say the kernel should be doing if you request a negative timeout? More explicitly, what made you think negative timeouts should have a specific effect?

Quoted 1 line
Quoted 1 line

The man page is silent about the timeout behaviour in case of its being negative. I believe that negative timeout is a mistake on behalf of an application and hence should be treated as such (i.e should be notified accordingly)

When you say schedule_timeout() returns immediately, I assume your logs are filling up with "schedule_timeout: wrong timeout ..." ? (You may need to bump your loglevel). If not, then schedule_timeout() isn't getting a negative value.

Quoted 1 line
Quoted 1 line

Yes I am getting the "schedule_timeout: wrong timeout ..." messages so I am sure the timeout has negative value.

Thanks,
Nish

regards
Ram Gupta

#1 lore raw
Quoted 26 lines · Ram Gupta
Quoted 9 lines · Ram Gupta

On 11/4/05, Nish Aravamudan <[email protected]>

Quoted 5 lines

In Ram's specific case, I think, the call path is sys_setsockopt() -> sock_setsockopt() -> sock_set_timeout, which has a definition of:

static int sock_set_timeout(long *timeo_p, char __user *optval, int

optlen)

Quoted 1 line
Quoted 1 line

Exactly right.

Ok.

Quoted 11 lines
Quoted 6 lines

Ram, what is the expected behavior of negative values in the timeval? And what are you seeing happen right now?

As of 2.6.14, looks like we convert any non-zero values into jiffies and store them in sk->sk_{rcv,snd}timeo...

I don't see any problem from the kernel side but the application
times out immediately causing certain failures as the schedule_timeout returns immediately in case of negative values. Shouldn't there be a check for negative values and return error to the application so that it can handle it.

Quoted 8 lines

I mean more along the lines of what does a man-page say the kernel should be doing if you request a negative timeout? More explicitly, what made you think negative timeouts should have a specific effect?

Quoted 4 lines

The man page is silent about the timeout behaviour in case of its being negative. I believe that negative timeout is a mistake on behalf of an application and hence should be treated as such (i.e should be notified accordingly)

Well, the problem is that there is no defined behavior for specifying a negative timeout (unlike poll(), for instance). So I'm not sure what the best approach is (beyond complaining to the application developers that their app is busted).

Quoted 7 lines

When you say schedule_timeout() returns immediately, I assume your logs are filling up with "schedule_timeout: wrong timeout ..." ? (You may need to bump your loglevel). If not, then schedule_timeout() isn't getting a negative value.

Quoted 1 line
Quoted 1 line

Yes I am getting the "schedule_timeout: wrong timeout ..."

messages so I am sure the timeout has negative value.

Ok, maybe bring this up with the networking folks, as they may have a better idea of what to do.

Thanks,
Nish