kernel error in system call accept() under kernel 2.6.8
----------------------------------------------------------------------
One line summary of the problem:
linux accept() system call does not set always errno if the returnvalue is negative.
----------------------------------------------------------------------
Full description of the problem/report:
Obviously there are some cases where the accept() system call does not set the errno variable if the accept() system call returns with a value less than zero:
Distribution : Linux Fedora Core 2 - but with kernel 2.6.8 Sample code:
--snip----------
....
//accept may return with a protocol error, simply try again while( (n = accept(m_ListenFd, (struct sockaddr *) cliaddr, &len)) < 0) {
Log.Log("Error accept, fd=%d, addrlen=%d, len=%d, errno=%d, %s",m_ListenFd,m_AddrLen, len, errno, strerror_r(errno, l_strebuf, sizeof(l_strebuf)));
if (errno == EPROTO || errno == ECONNABORTED) //connection already
aborted
{
Log.Log("connection aborted, wait for next");
continue; //next try
} else if (errno == EINTR) { //signal
if (CnThread::TestCancel(m_Log)) {
Log.Log("thread is cancelled");
delete[] cliaddr;
return -1;
} else {
Log.Log("EINTR, try again");
continue; //next try
}
}
else
{
throw CnException(__FILE__, __LINE__, "accept error errno=%d, %s",errno,strerror_r(errno, l_strebuf, sizeof(l_strebuf)));
}
}
...
--snip-----
output:CnTcpServer::WaitClient Error accept, fd=19, addrlen=16, len=16,
errno=0, Success
----------------------------------------------------------------------
Keywords (i.e., modules, networking, kernel):
network accept errno error return value
----------------------------------------------------------------------
Kernel version (from /proc/version):
cat /proc/version
Linux version 2.6.8-1.521 (root@wsa92_D2_FEDTest) (gcc version 3.3.3 20040412 (Red Hat Linux 3.3.3-7)) #3 Fri Jul 8 11:08:56 CEST 2005
----------------------------------------------------------------------
----------------------------------------------------------------------
Question:
Is there any update/information/explanation about this behaviour available??
Quoted 3 lines · Peter Duellings
Obviously there are some cases where the
accept()system call does not set the errno variable if theaccept()system call returns with a value less than zero:
Not actually possible. The kernel returns a positive value, zero or a negative value which is the errno code negated. It has no mechanism to return a negative value and not error.
What does strace show for the failing case ?
Alan,
will try to get a strace.
Thanx,
Peter
Quoted 13 lines · Alan Cox
Quoted 3 lines · Peter Duellings
Obviously there are some cases where the
accept()system call does not set the errno variable if theaccept()system call returns with a value less than zero:Not actually possible. The kernel returns a positive value, zero or a negative value which is the errno code negated. It has no mechanism to return a negative value and not error.
What does strace show for the failing case ?
Alan,
meanwhile we could generate a strace for the problem. However, I guess that the strace does not give the desired information (see parts below).
Additionally we added in the program the output of the return value of the accept() system call . The return value is -512 and the errno value is 0!
Usually the return value should be -1 and the errno should contain the value without the sign of the error.
Any idea or comment on this ?
Thanks,
Peter Düllings
---------<strace>------------
2682 09:25:29.238663 accept(21, <unfinished ...>
2688 09:25:29.263486 accept(33, {sa_family=AF_INET, sin_port=htons(32811), sin_addr=inet_addr("127.0.0.1")}, [16]) = 40 <27.171270>
2688 09:25:56.589969 accept(33, 0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.385453>
2688 09:25:56.975563 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:56.975676 accept(33, 0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.205963>
2688 09:25:57.181770 --- SIGCHLD (Child exited) @ 0 (0) ---
2688 09:25:57.181842 accept(33, <unfinished ...>
2682 09:25:57.231961 <... accept resumed> {sa_family=AF_INET, sin_port=htons(32882), sin_addr=inet_addr("127.0.0.1")}, [16]) = 43 <27.993066> 2682 09:25:57.234320 accept(21, <unfinished ...>
2688 09:25:57.538314 <... accept resumed> 0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.356435>
2688 09:25:57.538429 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.538488 accept(33, 0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.015688>
2688 09:25:57.554315 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.554370 accept(33, 0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.192660>
2688 09:25:57.747151 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.747236 accept(33, 0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.097813>
....
.
---------</strace>------------
Quoted 14 lines · Alan Cox
Quoted 3 lines · Peter Duellings
Obviously there are some cases where the
accept()system call does not set the errno variable if theaccept()system call returns with a value less than zero:Not actually possible. The kernel returns a positive value, zero or a negative value which is the errno code negated. It has no mechanism to return a negative value and not error.
What does strace show for the failing case ?
Quoted 13 lines · Peter Duellings
Alan,
meanwhile we could generate a strace for the problem. However, I guess that the strace does not give the desired information (see parts below).
Additionally we added in the program the output of the return value of the
accept()system call . The return value is -512 and the errno value is 0!
Usually the return value should be -1 and the errno should contain the value without the sign of the error.
Quoted 4 lines
Any idea or comment on this ?
Thanks,
If you run an ordinary 'C' runtime library, not something that you wrote to interface with the kernel, then a return value of -512 with a 0 errno value is not possible unless you have code that is trashing something in user-space.
The interface, whether a sys-call or an interrupt, takes the return value, negates it, and puts it into errno if it was negative, then it sets the return value to -1. This is common code for all system-calls, therefore nothing you can turn off for a particular system call.
A possible problem may be that you are not using the proper 'C' runtime library headers to define what errno actually is. On many (most) runtime libraries 'errno' is NOT 'extern int errno'. Instead it is actually a de-reference of the return-value of a function that located the proper variable for your particular thread! You can't create your own global errno and have it magically updated.
Quoted 35 lines
Peter Düllings
---------<strace>------------
2682 09:25:29.238663 accept(21, <unfinished ...>
2688 09:25:29.263486accept(33, {sa_family=AF_INET, sin_port=htons(32811), sin_addr=inet_addr("127.0.0.1")}, [16])= 40 <27.171270>
2688 09:25:56.589969accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.385453>
2688 09:25:56.975563 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:56.975676accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.205963>2688 09:25:57.181770 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.181842 accept(33, <unfinished ...> 2682 09:25:57.231961 <... accept resumed> {sa_family=AF_INET,sin_port=htons
(32882), sin_addr=inet_addr("127.0.0.1")}, [16]) = 43 <27.993066> 2682 09:25:57.234320 accept(21, <unfinished ...>
2688 09:25:57.538314 <... accept resumed>0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.356435>
2688 09:25:57.538429 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.538488accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.015688>
2688 09:25:57.554315 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.554370accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.192660>
2688 09:25:57.747151 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.747236accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.097813>
....
.
---------</strace>------------
Quoted 15 lines
Quoted 13 lines · Alan Cox
Quoted 3 lines · Peter Duellings
Obviously there are some cases where the
accept()system call does not set the errno variable if theaccept()system call returns with a value less than zero:Not actually possible. The kernel returns a positive value, zero or a negative value which is the errno code negated. It has no mechanism to return a negative value and not error.
What does strace show for the failing case ?
Cheers,
Dick Johnson
Penguin : Linux version 2.6.13 on an i686 machine (5589.55 BogoMips). Warning : 98.36% of all statistics are fiction.
****************************************************************
The information transmitted in this message is confidential and may be privileged. Any review, retransmission, dissemination, or other use of this information by persons or entities other than the intended recipient is prohibited. If you are not the intended recipient, please notify Analogic Corporation immediately - by replying to this message or by sending an email to [email protected] - and destroy all copies of this information, including any attachments, without reading or disclosing them.
Thank you.
On Wed, 2005-10-05 at 17:41 +0200, Peter Duellings wrote:
[....]
Quoted 3 lines
meanwhile we could generate a strace for the problem. However, I guess that the strace does not give the desired information (see parts below).
Yup. Where exactly does the kernel return -512 in the strace output? I can see only 40 and 43 which are valid file descriptors.
Quoted 9 lines
Additionally we added in the program the output of the return value of the
accept()system call . The return value is -512 and the errno value is 0!Usually the return value should be -1 and the errno should contain the value without the sign of the error.
Any idea or comment on this ?
You are doing somewhere else something wrong so post a minimalistic piece of source code of how you get to it.
Or show the *failing* case in the strace output.
Quoted 1 line
Thanks,
[...]
Quoted 21 lines
---------<strace>------------
2682 09:25:29.238663 accept(21, <unfinished ...>
2688 09:25:29.263486accept(33, {sa_family=AF_INET, sin_port=htons(32811), sin_addr=inet_addr("127.0.0.1")}, [16])= 40 <27.171270>
2688 09:25:56.589969accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.385453>
2688 09:25:56.975563 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:56.975676accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.205963>2688 09:25:57.181770 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.181842 accept(33, <unfinished ...> 2682 09:25:57.231961 <... accept resumed> {sa_family=AF_INET,sin_port=htons
(32882), sin_addr=inet_addr("127.0.0.1")}, [16]) = 43 <27.993066> 2682 09:25:57.234320 accept(21, <unfinished ...>
2688 09:25:57.538314 <... accept resumed>0x82fa7e0, [16]) = ? ERESTARTSYS (To be restarted) <0.356435>
2688 09:25:57.538429 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.538488accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.015688>
2688 09:25:57.554315 --- SIGCHLD (Child exited) @ 0 (0) ---
Quoted 8 lines
2688 09:25:57.554370
accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.192660>
2688 09:25:57.747151 --- SIGCHLD (Child exited) @ 0 (0) --- 2688 09:25:57.747236accept(33, 0x82fa7e0, [16])= ? ERESTARTSYS (To be restarted) <0.097813>
....
.
---------</strace>------------
[...]
Bernd
Signature · 3 lines
Firmix Software GmbH http://www.firmix.at/
mobil: +43 664 4416156 fax: +43 1 7890849-55
Embedded Linux Development and ServicesA message that is not in the archive (sent off-list or not yet received): [email protected]
Hi Ben,
if Log.Log would modify errno the Log.Log debug output should not be affected since the value of errno - from my understanding - is copied on the stack *before* Log.Log is called.
Or did I forget something?
Thanx,
Peter
Quoted 14 lines · Benjamin LaHaise
Quoted 7 lines · Peter Duellings
//accept may return with a protocol error, simply try again
while( (n = accept(m_ListenFd, (struct sockaddr *) cliaddr, &len)) < 0){
Log.Log("Error accept, fd=%d, addrlen=%d, len=%d, errno=%d, %s",m_ListenFd,m_AddrLen, len, errno,strerror_r(errno, l_strebuf, sizeof(l_strebuf)));
if (errno == EPROTO || errno == ECONNABORTED) //connection alreadyLet's see here: what happens if
Log.Log()performs a syscall to, say, write out the log message to a buffer?-ben
Quoted 6 lines · Peter Duellings
Hi Ben,
if Log.Log would modify errno the Log.Log debug output should not be affected since the value of errno - from my understanding - is copied on the stack *before* Log.Log is called.
Or did I forget something?
errno does not reside on the stack.
-ben
Ben,
Right. But before Log.Log is called arguments of methods are copied on the stack. That means, also the current content of errno is copied. And "current" means in that case before the call to Log.Log is performed (errno is transferred by value - not by reference).
-Peter
Quoted 13 lines · Benjamin LaHaise
Quoted 6 lines · Peter Duellings
Hi Ben,
if Log.Log would modify errno the Log.Log debug output should not be affected since the value of errno - from my understanding - is copied on the stack *before* Log.Log is called.
Or did I forget something?errno does not reside on the stack.
-ben
Quoted 5 lines · Peter Duellings
Right. But before Log.Log is called arguments of methods are copied on the stack. That means, also the current content of errno is copied. And "current" means in that case before the call to Log.Log is performed (errno is transferred by value - not by reference).
And strerror_r() modifies errno. You can't rely on the order arguments to functions are evaluated in.
-ben