I have a small server program that accepts connections on a TCP or local UNIX socket, reads a simple command and (depending on the command) sends a reply.
The problem is that the client may have no interest in the answer and sometimes exits early. So writing to that socket will cause a SIGPIPE and make my server crash.
What's the best practice to prevent the crash here? Is there a way to check if the other side of the line is still reading? (select() doesn't seem to work here as it always says the socket is writable). Or should I just catch the SIGPIPE with a handler and ignore it?
You generally want to ignore the SIGPIPE and handle the error directly in your code. This is because signal handlers in C have many restrictions on what they can do.
The most portable way to do this is to set the SIGPIPE handler to SIG_IGN. This will prevent any socket or pipe write from causing a SIGPIPE signal.
To ignore the SIGPIPE signal, use the following code:
signal(SIGPIPE, SIG_IGN);
If you're using the send() call, another option is to use the MSG_NOSIGNAL option, which will turn the SIGPIPE behavior off on a per call basis. Note that not all operating systems support the MSG_NOSIGNAL flag.
Lastly, you may also want to consider the SO_SIGNOPIPE socket flag that can be set with setsockopt() on some operating systems. This will prevent SIGPIPE from being caused by writes just to the sockets it is set on.
Another method is to change the socket so it never generates SIGPIPE on write(). This is more convenient in libraries, where you might not want a global signal handler for SIGPIPE.
On most BSD-based (MacOS, FreeBSD...) systems, (assuming you are using C/C++), you can do this with:
int set = 1;
setsockopt(sd, SOL_SOCKET, SO_NOSIGPIPE, (void *)&set, sizeof(int));
With this in effect, instead of the SIGPIPE signal being generated, EPIPE will be returned.
I'm super late to the party, but SO_NOSIGPIPE isn't portable, and might not work on your system (it seems to be a BSD thing).
A nice alternative if you're on, say, a Linux system without SO_NOSIGPIPE would be to set the MSG_NOSIGNAL flag on your send(2) call.
Example replacing write(...) by send(...,MSG_NOSIGNAL) (see nobar's comment)
char buf[888];
//write( sockfd, buf, sizeof(buf) );
send( sockfd, buf, sizeof(buf), MSG_NOSIGNAL );
In this post I described possible solution for Solaris case when neither SO_NOSIGPIPE nor MSG_NOSIGNAL is available.
Instead, we have to temporarily suppress SIGPIPE in the current thread that executes library code. Here's how to do this: to suppress SIGPIPE we first check if it is pending. If it does, this means that it is blocked in this thread, and we have to do nothing. If the library generates additional SIGPIPE, it will be merged with the pending one, and that's a no-op. If SIGPIPE is not pending then we block it in this thread, and also check whether it was already blocked. Then we are free to execute our writes. When we are to restore SIGPIPE to its original state, we do the following: if SIGPIPE was pending originally, we do nothing. Otherwise we check if it is pending now. If it does (which means that out actions have generated one or more SIGPIPEs), then we wait for it in this thread, thus clearing its pending status (to do this we use sigtimedwait() with zero timeout; this is to avoid blocking in a scenario where malicious user sent SIGPIPE manually to a whole process: in this case we will see it pending, but other thread may handle it before we had a change to wait for it). After clearing pending status we unblock SIGPIPE in this thread, but only if it wasn't blocked originally.
Example code at https://github.com/kroki/XProbes/blob/1447f3d93b6dbf273919af15e59f35cca58fcc23/src/libxprobes.c#L156
Handle SIGPIPE Locally
It's usually best to handle the error locally rather than in a global signal event handler since locally you will have more context as to what's going on and what recourse to take.
I have a communication layer in one of my apps that allows my app to communicate with an external accessory. When a write error occurs I throw and exception in the communication layer and let it bubble up to a try catch block to handle it there.
Code:
The code to ignore a SIGPIPE signal so that you can handle it locally is:
// We expect write failures to occur but we want to handle them where
// the error occurs rather than in a SIGPIPE handler.
signal(SIGPIPE, SIG_IGN);
This code will prevent the SIGPIPE signal from being raised, but you will get a read / write error when trying to use the socket, so you will need to check for that.
You cannot prevent the process on the far end of a pipe from exiting, and if it exits before you've finished writing, you will get a SIGPIPE signal. If you SIG_IGN the signal, then your write will return with an error - and you need to note and react to that error. Just catching and ignoring the signal in a handler is not a good idea -- you must note that the pipe is now defunct and modify the program's behaviour so it does not write to the pipe again (because the signal will be generated again, and ignored again, and you'll try again, and the whole process could go on for a long time and waste a lot of CPU power).
Or should I just catch the SIGPIPE with a handler and ignore it?
I believe that is right on. You want to know when the other end has closed their descriptor and that's what SIGPIPE tells you.
Sam
What's the best practice to prevent the crash here?
Either disable sigpipes as per everybody, or catch and ignore the error.
Is there a way to check if the other side of the line is still reading?
Yes, use select().
select() doesn't seem to work here as it always says the socket is writable.
You need to select on the read bits. You can probably ignore the write bits.
When the far end closes its file handle, select will tell you that there is data ready to read. When you go and read that, you will get back 0 bytes, which is how the OS tells you that the file handle has been closed.
The only time you can't ignore the write bits is if you are sending large volumes, and there is a risk of the other end getting backlogged, which can cause your buffers to fill. If that happens, then trying to write to the file handle can cause your program/thread to block or fail. Testing select before writing will protect you from that, but it doesn't guarantee that the other end is healthy or that your data is going to arrive.
Note that you can get a sigpipe from close(), as well as when you write.
Close flushes any buffered data. If the other end has already been closed, then close will fail, and you will receive a sigpipe.
If you are using buffered TCPIP, then a successful write just means your data has been queued to send, it doesn't mean it has been sent. Until you successfully call close, you don't know that your data has been sent.
Sigpipe tells you something has gone wrong, it doesn't tell you what, or what you should do about it.
Under a modern POSIX system (i.e. Linux), you can use the sigprocmask() function.
#include <signal.h>
void block_signal(int signal_to_block /* i.e. SIGPIPE */ )
{
sigset_t set;
sigset_t old_state;
// get the current state
//
sigprocmask(SIG_BLOCK, NULL, &old_state);
// add signal_to_block to that existing state
//
set = old_state;
sigaddset(&set, signal_to_block);
// block that signal also
//
sigprocmask(SIG_BLOCK, &set, NULL);
// ... deal with old_state if required ...
}
If you want to restore the previous state later, make sure to save the old_state somewhere safe. If you call that function multiple times, you need to either use a stack or only save the first or last old_state... or maybe have a function which removes a specific blocked signal.
For more info read the man page.
Linux manual said:
EPIPE The local end has been shut down on a connection oriented
socket. In this case the process will also receive a SIGPIPE
unless MSG_NOSIGNAL is set.
But for Ubuntu 12.04 it isn't right. I wrote a test for that case and I always receive EPIPE withot SIGPIPE. SIGPIPE is genereated if I try to write to the same broken socket second time. So you don't need to ignore SIGPIPE if this signal happens it means logic error in your program.
Related
I have a small server program that accepts connections on a TCP or local UNIX socket, reads a simple command and (depending on the command) sends a reply.
The problem is that the client may have no interest in the answer and sometimes exits early. So writing to that socket will cause a SIGPIPE and make my server crash.
What's the best practice to prevent the crash here? Is there a way to check if the other side of the line is still reading? (select() doesn't seem to work here as it always says the socket is writable). Or should I just catch the SIGPIPE with a handler and ignore it?
You generally want to ignore the SIGPIPE and handle the error directly in your code. This is because signal handlers in C have many restrictions on what they can do.
The most portable way to do this is to set the SIGPIPE handler to SIG_IGN. This will prevent any socket or pipe write from causing a SIGPIPE signal.
To ignore the SIGPIPE signal, use the following code:
signal(SIGPIPE, SIG_IGN);
If you're using the send() call, another option is to use the MSG_NOSIGNAL option, which will turn the SIGPIPE behavior off on a per call basis. Note that not all operating systems support the MSG_NOSIGNAL flag.
Lastly, you may also want to consider the SO_SIGNOPIPE socket flag that can be set with setsockopt() on some operating systems. This will prevent SIGPIPE from being caused by writes just to the sockets it is set on.
Another method is to change the socket so it never generates SIGPIPE on write(). This is more convenient in libraries, where you might not want a global signal handler for SIGPIPE.
On most BSD-based (MacOS, FreeBSD...) systems, (assuming you are using C/C++), you can do this with:
int set = 1;
setsockopt(sd, SOL_SOCKET, SO_NOSIGPIPE, (void *)&set, sizeof(int));
With this in effect, instead of the SIGPIPE signal being generated, EPIPE will be returned.
I'm super late to the party, but SO_NOSIGPIPE isn't portable, and might not work on your system (it seems to be a BSD thing).
A nice alternative if you're on, say, a Linux system without SO_NOSIGPIPE would be to set the MSG_NOSIGNAL flag on your send(2) call.
Example replacing write(...) by send(...,MSG_NOSIGNAL) (see nobar's comment)
char buf[888];
//write( sockfd, buf, sizeof(buf) );
send( sockfd, buf, sizeof(buf), MSG_NOSIGNAL );
In this post I described possible solution for Solaris case when neither SO_NOSIGPIPE nor MSG_NOSIGNAL is available.
Instead, we have to temporarily suppress SIGPIPE in the current thread that executes library code. Here's how to do this: to suppress SIGPIPE we first check if it is pending. If it does, this means that it is blocked in this thread, and we have to do nothing. If the library generates additional SIGPIPE, it will be merged with the pending one, and that's a no-op. If SIGPIPE is not pending then we block it in this thread, and also check whether it was already blocked. Then we are free to execute our writes. When we are to restore SIGPIPE to its original state, we do the following: if SIGPIPE was pending originally, we do nothing. Otherwise we check if it is pending now. If it does (which means that out actions have generated one or more SIGPIPEs), then we wait for it in this thread, thus clearing its pending status (to do this we use sigtimedwait() with zero timeout; this is to avoid blocking in a scenario where malicious user sent SIGPIPE manually to a whole process: in this case we will see it pending, but other thread may handle it before we had a change to wait for it). After clearing pending status we unblock SIGPIPE in this thread, but only if it wasn't blocked originally.
Example code at https://github.com/kroki/XProbes/blob/1447f3d93b6dbf273919af15e59f35cca58fcc23/src/libxprobes.c#L156
Handle SIGPIPE Locally
It's usually best to handle the error locally rather than in a global signal event handler since locally you will have more context as to what's going on and what recourse to take.
I have a communication layer in one of my apps that allows my app to communicate with an external accessory. When a write error occurs I throw and exception in the communication layer and let it bubble up to a try catch block to handle it there.
Code:
The code to ignore a SIGPIPE signal so that you can handle it locally is:
// We expect write failures to occur but we want to handle them where
// the error occurs rather than in a SIGPIPE handler.
signal(SIGPIPE, SIG_IGN);
This code will prevent the SIGPIPE signal from being raised, but you will get a read / write error when trying to use the socket, so you will need to check for that.
You cannot prevent the process on the far end of a pipe from exiting, and if it exits before you've finished writing, you will get a SIGPIPE signal. If you SIG_IGN the signal, then your write will return with an error - and you need to note and react to that error. Just catching and ignoring the signal in a handler is not a good idea -- you must note that the pipe is now defunct and modify the program's behaviour so it does not write to the pipe again (because the signal will be generated again, and ignored again, and you'll try again, and the whole process could go on for a long time and waste a lot of CPU power).
Or should I just catch the SIGPIPE with a handler and ignore it?
I believe that is right on. You want to know when the other end has closed their descriptor and that's what SIGPIPE tells you.
Sam
What's the best practice to prevent the crash here?
Either disable sigpipes as per everybody, or catch and ignore the error.
Is there a way to check if the other side of the line is still reading?
Yes, use select().
select() doesn't seem to work here as it always says the socket is writable.
You need to select on the read bits. You can probably ignore the write bits.
When the far end closes its file handle, select will tell you that there is data ready to read. When you go and read that, you will get back 0 bytes, which is how the OS tells you that the file handle has been closed.
The only time you can't ignore the write bits is if you are sending large volumes, and there is a risk of the other end getting backlogged, which can cause your buffers to fill. If that happens, then trying to write to the file handle can cause your program/thread to block or fail. Testing select before writing will protect you from that, but it doesn't guarantee that the other end is healthy or that your data is going to arrive.
Note that you can get a sigpipe from close(), as well as when you write.
Close flushes any buffered data. If the other end has already been closed, then close will fail, and you will receive a sigpipe.
If you are using buffered TCPIP, then a successful write just means your data has been queued to send, it doesn't mean it has been sent. Until you successfully call close, you don't know that your data has been sent.
Sigpipe tells you something has gone wrong, it doesn't tell you what, or what you should do about it.
Under a modern POSIX system (i.e. Linux), you can use the sigprocmask() function.
#include <signal.h>
void block_signal(int signal_to_block /* i.e. SIGPIPE */ )
{
sigset_t set;
sigset_t old_state;
// get the current state
//
sigprocmask(SIG_BLOCK, NULL, &old_state);
// add signal_to_block to that existing state
//
set = old_state;
sigaddset(&set, signal_to_block);
// block that signal also
//
sigprocmask(SIG_BLOCK, &set, NULL);
// ... deal with old_state if required ...
}
If you want to restore the previous state later, make sure to save the old_state somewhere safe. If you call that function multiple times, you need to either use a stack or only save the first or last old_state... or maybe have a function which removes a specific blocked signal.
For more info read the man page.
Linux manual said:
EPIPE The local end has been shut down on a connection oriented
socket. In this case the process will also receive a SIGPIPE
unless MSG_NOSIGNAL is set.
But for Ubuntu 12.04 it isn't right. I wrote a test for that case and I always receive EPIPE withot SIGPIPE. SIGPIPE is genereated if I try to write to the same broken socket second time. So you don't need to ignore SIGPIPE if this signal happens it means logic error in your program.
In my application, there is a io-thread, that is dedicated for
Wrapping data received from the application in a custom protocol
Sending the data+custom protocol packet over tcp/ip
Receiving data+custom protocol packet over tcp/ip
Unwrapping the custom protocol and handing the data to the application.
Application processes the data over a different thread. Additionally, the requirements dictate that the unacknowledged window size should be 1, i.e. there should be only one pending unacknowledged message at anytime. This implies that if io-thread has dispatched a message over the socket, it will not send any more messages, till it hears an ack from the receiver.
Application's processing thread communicates to io-thread via pipe. Application needs to shut gracefully if someone from linux CLI types ctrl+C.
Thus, given these requirements, i have following options
Use PPoll() on socket and pipe descriptors
Use Select()
Use PSelect()
I have following questions
The decision between select() and poll(). My application only deals with less than 50 file descriptors. Is it okay to assume there would be no difference whether i choose select or poll ?
Decision between select() and pselect(). I read the linux documentation and it states about race condition between signals and select(). I dont have experience with signals, so can someone explain more clearly about the race condition and select() ? Does it have something to do with someone pressing ctrl+C on CLI and application not stopping?
Decision between pselect and ppoll() ? Any thoughts on one vs the other
I'd suggest by starting the comparison with select() vs poll(). Linux also provides both pselect() and ppoll(); and the extra const sigset_t * argument to pselect() and ppoll() (vs select() and poll()) has the same effect on each "p-variant", as it were. If you are not using signals, you have no race to protect against, so the base question is really about efficiency and ease of programming.
Meanwhile there's already a stackoverflow.com answer here: what are the differences between poll and select.
As for the race: once you start using signals (for whatever reason), you will learn that in general, a signal handler should just set a variable of type volatile sig_atomic_t to indicate that the signal has been detected. The fundamental reason for this is that many library calls are not re-entrant, and a signal can be delivered while you're "in the middle of" such a routine. For instance, simply printing a message to a stream-style data structure such as stdout (C) or cout (C++) can lead to re-entrancy issues.
Suppose you have code that uses a volatile sig_atomic_t flag variable, perhaps to catch SIGINT, something like this (see also http://pubs.opengroup.org/onlinepubs/007904975/functions/sigaction.html):
volatile sig_atomic_t got_interrupted = 0;
void caught_signal(int unused) {
got_interrupted = 1;
}
...
struct sigaction sa;
sa.sa_handler = caught_signal;
sigemptyset(&sa.sa_mask);
sa.sa_flags = SA_RESTART;
if (sigaction(SIGINT, &sa, NULL) == -1) ... handle error ...
...
Now, in the main body of your code, you might want to "run until interrupted":
while (!got_interrupted) {
... do some work ...
}
This is fine up until you start needing to make calls that wait for some input/output, such as select or poll. The "wait" action needs to wait for that I/O—but it also needs to wait for a SIGINT interrupt. If you just write:
while (!got_interrupted) {
... do some work ...
result = select(...); /* or result = poll(...) */
}
then it's possible that the interrupt will happen just before you call select() or poll(), rather than afterward. In this case, you did get interrupted—and the variable got_interrupted gets set—but after that, you start waiting. You should have checked the got_interrupted variable before you started waiting, not after.
You can try writing:
while (!got_interrupted) {
... do some work ...
if (!got_interrupted)
result = select(...); /* or result = poll(...) */
}
This shrinks the "race window", because now you'll detect the interrupt if it happens while you're in the "do some work" code; but there is still a race, because the interrupt can happen right after you test the variable, but right before the select-or-poll.
The solution is to make the "test, then wait" sequence "atomic", using the signal-blocking properties of sigprocmask (or, in POSIX threaded code, pthread_sigmask):
sigset_t mask, omask;
...
while (!got_interrupted) {
... do some work ...
/* begin critical section, test got_interrupted atomically */
sigemptyset(&mask);
sigaddset(&mask, SIGINT);
if (sigprocmask(SIG_BLOCK, &mask, &omask))
... handle error ...
if (got_interrupted) {
sigprocmask(SIG_SETMASK, &omask, NULL); /* restore old signal mask */
break;
}
result = pselect(..., &omask); /* or ppoll() etc */
sigprocmask(SIG_SETMASK, &omask, NULL);
/* end critical section */
}
(the above code is actually not that great, it's structured for illustration rather than efficiency -- it's more efficient to do the signal mask manipulation slightly differently, and place the "got interrupted" tests differently).
Until you actually start needing to catch SIGINT, though, you need only compare select() and poll() (and if you start needing large numbers of descriptors, some of the event-based stuff like epoll() is more efficient than either one).
Between (p)select and (p)poll is a rather subtle difference:
For select, you have to initialize and populate the ugly fd_set bitmaps everytime before you call select because select modifies them in-place in a "destructive" fashion. (poll distinguishes between the .events and .revents members in struct pollfd).
After selecting, the entire bitmap is often scanned (by people/code) for events even if most of the fds are not even watched.
Third, the bitmap can only deal with fds whose number is less than a certain limit (contemporary implementations: somewhere between 1024..4096), which rules it out in programs where high fds can be easibly attained (notwithstanding that such programs are likely to already use epoll instead).
The accepted answer is not correct vis a vis difference between select and pselect. It does describe well how a race condition between sig-handler and select can arise, but it is incorrect in how it uses pselect to solve the problem. It misses the main point about pselect which is that it waits for EITHER the file-descriptor or the signal to become ready. pselect returns when either of these are ready.Select ONLY waits on the file-descriptor. Select ignores signals. See this blog post for a good working example:
https://www.linuxprogrammingblog.com/code-examples/using-pselect-to-avoid-a-signal-race
To make the picture presented by the accepted answer complete following basic fact should be mentioned: both select() and pselect() may return EINTR as stated in their man pages:
EINTR A signal was caught; see signal(7).
This "caught" means that the signal should be recognized as "occurred during the system call execution":
1. If non-masked signal occurs during select/pselect execution then select/pselect will exit.
2. If non-masked signal occurs before select/pselect has been called this will not have any effect and select/pselect will continue waiting, potentially forever.
So if a signal occurs during select/pselect execution we are ok - the execution of select/pselect will be interrupted and then we can test the reason for the exit and discover that is was EINTR and then we can exit the loop.
The real threat that we face is a possibility of signal occurrence outside of select/pselect execution, then we may hang in the system call forever. Any attempt to discover this "outsider" signal by naive means:
if (was_a_signal) {
...
}
will fail since no matter how close this test will be to the call of select/pselect there is always a possibility that the signal will occur just after the test and before the call to select/pselect.
Then, if the only place to catch the signal is during select/pselect execution we should invent some kind of "wine funnel" so all "wine splashes" (signals), even outside of "bottle neck" (select/pselect execution period) will eventually come to the "bottle neck".
But how can you deceive system call and make it "think" that the signal has occurred during this system call execution when in reality it has occurred before?
Easy. Here is our "wine funnel": you just block the signal of interest and by that cause it (if it has occurred at all) waiting outside of the process "for the door to be opened" and you "open the door" (unmask the signal) only when you're prepared "to welcome the guest" (select/pselect is running). Then the "arrived" signal will be recognized as "just occurred" and will interrupt the execution of the system call.
Of course, "opening the door" is the most critical part of the plan - it cannot be done by the usual means (first unmask, then call to select/pselect), the only possibility is to do the both actions (unmask and system call) at once (atomically) - this is what pselect() is capable of but select() is not.
This is a question similar to Proper way to close a blocking UDP socket. I have a thread in C which is reading from a UDP socket. The read is blocking. I would like to know if it is possible to be able to exit the thread, without relying on the recv() returning? For example can I close the socket from another thread and safely expect the socket read thread to exit? Didn't see any high voted answer on that thread, thats why I am asking it again.
This really depends on what system you're running under. For example, if you're running under a POSIX-compliant system and your thread is cancelable, the recv() call will be interrupted when you cancel the thread since it's a cancel point.
If you're using an older socket implementation, you could set a signal handler for your thread for something like SIGUSR1 and hope nobody else wanted it and signal, since recv() will interrupt on a signal. Your best option is not to block, if at all possible.
I don't think closing a socket involved in a blocking operation is a safe guaranteed way of terminating the operation. For instance, kernel.org warns darkly:
It is probably unwise to close file descriptors while they may be in
use by system calls in other threads in the same process. Since a
file descriptor may be reused, there are some obscure race conditions
that may cause unintended side effects.
Instead you could use a signal and make recv fail with EINTR
(make sure SA_RESTART is not enabled). You can send a signal to a
specific thread with pthread_kill
You could enable SO_RCVTIMEO on the socket before starting the recv
call
Personally I usually try to stay clear of all the signal nastiness but it's a viable option.
You've got a couple of options for that. A signal will interrupt the read operation, so all you need to do is make sure a signal goes off. The recv operation should fail with error number EINTR.
The simplest option is to set up a timer to interrupt your own process after some timeout e.g. 30 seconds:
itimerval timer
timeval time;
time.tv_sec = 30;
time.tv_usec = 0;
timer.it_value = time;
if( setitimer( ITIMER_REAL, &timer, NULL ) != 0 )
printf( "failed to start timer\n" );
You'll get a SIGALRM after the specified time, which will interrupt your blocking operation, and give you the chance to repeat the operation or quit.
You cannot deallocate a shared resource while another thread is or might be using it. In practice, you will find that you cannot even write code to do what you suggest.
Think about it. When you go to call close, how can you possibly know that the other thread is actually blocked in recv? What if it's about to call recv, but then another thread calls socket and gets the descriptor you just closed? Now, not only will that thread not detect any error, but it will be calling recv on the wrong socket!
There is probably a good way to solve your outer problem, the reason you need to exit from a blocking UDP socket read. There are also several ugly hacks available. The basic approach is to make the socket non-blocking and instead of making a blocking UDP socket read, fake a blocking read with select or poll. You can then abort this loop several ways:
One way is to have select time out and check an 'abort' flag when select returns.
Another way is to also select on the read end of a pipe. Send a single byte to the pipe to abort the select.
If posix complient system, you can try to monitor your thread:
pthread_create with a function that makes your recv and pthread_cond_signal just after, then returns.
The calling thread makes a pthread_cond_timedwait with the desired timeout and terminates the called thread if timed_out.
While writing a simple server-client application, this question came in my mind. When someone tries to write to a broken pipe, a SIGPIPE would be generated. Let's say I handle the signal in my code.
Now what error does the write call returns - EPIPE or EINTR ( as it was interrupted by a signal). I tried with a sample program and I seem to be getting EPIPE always. Is this a guaranteed behavior or it could be any of the two error values?
POSIX says that EPIPE should be returned and SIGPIPE sent:
For write()s or pwrite()s to pipes or FIFOs not open for reading by any process, or with only one end open.
For write()s to sockets that are no longer connected or shut down for writing.
You can have a look at the POSIX standard here
The write(2) call returns -1 on error, so I guess you are asking about the value of errno(3).
You'll get EPIPE if you handle, block, or ignore the signal. Otherwise the process is terminated by default, see signal(7).
In general, "interrupted by a signal" (EINTR) refers to the utterly ridiculous Unix System V signal handling, whereby ANY system call could fail if your process received (and handled) a signal during the system call. This required wrapping every system call with do ... while (ret==-1 && errno==EINTR); or similar. While POSIX still allows either this or the good ("BSD") behavior, sane systems like GNU/Linux have the BSD behavior by default. You can always obtain the BSD behavior by calling sigaction with the right arguments, or even make a wrapper function to do so for you.
As such, EINTR is unrelated to the SIGPIPE caused by write errors.
A situation I have under Windows XP (SP3) has been driving me nuts, and I'm reaching the end of my tether, so maybe someone can provide some inspiration.
I have a C++ networking program (non-GUI). This program is built to compile and run under Windows, MacOS/X, and Linux, so it uses select() and non-blocking I/O as the basis for its event loop.
In addition to its networking duties, this program needs to read text commands from stdin, and exit gracefully when stdin is closed. Under Linux and MacOS/X, that's easy enough -- I just include STDIN_FILENO in my read fd_set to select(), and select() returns when stdin is closed. I check to see that FD_ISSET(STDIN_FILENO, &readSet) is true, try to read some data from stdin, recv() returns 0/EOF, and so I exit the process.
Under Windows, on the other hand, you can't select on STDIN_FILE_HANDLE, because it's not a real socket. You can't do non-blocking reads on STDIN_FILE_HANDLE, either. That means there is no way to read stdin from the main thread, since ReadFile() might block indefinitely, causing the main thread to stop serving its network function.
No problem, says I, I'll just spawn a thread to handle stdin for me. This thread will run in an infinite loop, blocking in ReadFile(stdinHandle), and whenever ReadFile() returns data, the stdin-thread will write that data to a TCP socket. That socket's connection's other end will be select()'d on by the main thread, so the main thread will see the stdin data coming in over the connection, and handle "stdin" the same way it would under any other OS. And if ReadFile() returns false to indicate that stdin has closed, the stdin-thread just closes its end of the socket-pair so that the main thread will be notified via select(), as described above.
Of course, Windows doesn't have a nice socketpair() function, so I had to roll my own using listen(), connect(), and accept() (as seen in the CreateConnectedSocketPair() function here. But I did that, and it seems to work, in general.
The problem is that it doesn't work 100%. In particular, if stdin is closed within a few hundred milliseconds of when the program starts up, about half the time the main thread doesn't get any notification that the stdin-end of the socket-pair has been closed. What I mean by that is, I can see (by my printf()-debugging) that the stdin-thread has called closesocket() on its socket, and I can see that the main thread is select()-ing on the associated socket (i.e. the other end of the socket-pair), but select() never returns as it should... and if it does return, due to some other socket selecting ready-for-whatever, FD_ISSET(main_thread_socket_for_socket_pair, &readSet) returns 0, as if the connection wasn't closed.
At this point, the only hypothesis I have is that there is a bug in Windows' select() implementation that causes the main thread's select() not to notice that the other end of the socket-pair has closed by the stdin-thread. Is there another explanation? (Note that this problem has been reported under Windows 7 as well, although I haven't looked at it personally on that platform)
Just for the record, this problem turned out to be a different issue entirely, unrelated to threading, Windows, or stdin. The actual problem was an inter-process deadlock, where the parent process was blocked, waiting for the child processes to quit, but sometimes the child processes would be simultaneously blocked, waiting on the parent to supply them with some data, and so nothing would move forward.
Apologies to all for wasting your time on a red herring; if there's a standard way to close this case as unwarranted, let me know and I'll do it.
-Jeremy
Is it possible you have a race condition? Eg. Do you ensure that the CreateConnectedSocketPair() function has definitely returned before the stdin-thread has a chance to try closing its socket?
I am studying in your code. In the CreateConnectedSocketPair(), socket1 is used for listen(), and newfd is used for send/recv data. So, why does "socket1 = newfd"? How to close the listenfd then?
Not a solution, but as a workaround, couldn't you send some magic "stdin has closed" message across the TCP socket and have your receiving end disconnect its socket when it sees that and run whatever 'stdin has closed' handler?
Honestly your code is too long and I don't have time right now to spend on it.
Most likely the problem is in some cases closing the socket doesn't cause a graceful (FIN) shutdown.
Checking for exceptions returning from your select may catch the remainder of cases. There is also the (slim) possibility that no notification is actually being sent to the socket that the other end has closed. In that case, there is no way other than timeouts or "keep alive"/ping messages between the endpoints to know that the socket has closed.
If you want to figure out exactly what is happening, break out wireshark and look for FINs and RSTs (and the absence of anything). If you see the proper FIN sequence going across when your socket is closed, then the problem must be in your code. if you see RST, it may be caught by exceptions, and if you don't see anything you'll need to devise a way in your protocol to 'ping' each side of the connection to make sure they are still alive, or set a sufficiently short timeout for more data.
Rather than chasing perceived bugs in select(), I'm going to address your original fallacy that drove you away from simple, reliable, single-threaded design.
You said "You can't do non-blocking reads on STDIN_FILE_HANDLE, either. That means there is no way to read stdin from the main thread, since ReadFile() might block indefinitely" but this simply isn't the whole story. Look at ReadConsoleInput, WSAEventSelect, and WaitForMultipleObjects. The stdin handle will be signalled only when there is input and ReadConsoleInput will return immediately (pretty much the same idea behind select() in Unix).
Or, use ReadFileEx and WaitForMultipleObjectsEx to have the console reads fire off an APC (which isn't all that asynchronous, it runs on the main thread and only during WaitForMultipleObjectsEx or another explicit wait function).
If you want to stick with using a second thread to get async I/O on stdin, then you might try closing the handle being passed to select instead of doing a socket shutdown (via closesocket on the other end). In my experience select() tends to return really quickly when one of the fds it is waiting on gets closed.
Or, maybe your problem is the other way around. The select docs say "For connection-oriented sockets, readability can also indicate that a request to close the socket has been received from the peer". Typically you'd send that "request to close the socket" by calling shutdown(), not closesocket().