I'm having a problem with EPIPE in my iOS app, and it's not being caught in the @try/@catch/@finally block. How can I catch this signal (SIGPIPE, likely)...
I've built a "web proxy" into my app that will handle certain kinds of URLs - in this error case, it seems that the remote end (also in my app, but hiding in the iOS libraries) closes its end of the socket. I don't get a notification (should I? Is there something I should register for with the NSFileHandle that might help here?).
I've based this proxy on HTTPServer that Matt Gallagher put together (available here), and the problem is in a subclass of the HTTPRequestHandler class he put together. Here's the code (this code is the equivalent of the startResponse method in the base class):
-(void)proxyTS:(SSProxyTSResource *)proxyTS didReceiveResource:(NSData *)resource
{
NSLog(@"[%@ %@]", NSStringFromClass([self class]), NSStringFromSelector(_cmd));
CFHTTPMessageRef response =
CFHTTPMessageCreateResponse(kCFAllocatorDefault, 200, NULL, kCFHTTPVersion1_1);
CFHTTPMessageSetHeaderFieldValue(response,
(CFStringRef)@"Content-Type",
(__bridge CFStringRef)s_MIMEtype);
CFHTTPMessageSetHeaderFieldValue(response,
(CFStringRef)@"Connection",
(CFStringRef)@"close");
CFHTTPMessageSetBody(response,
(__bridge CFDataRef)resource);
CFDataRef headerData = CFHTTPMessageCopySerializedMessage(response);
@try
{
NSLog(@" -> writing %u bytes to filehandle...",[((__bridge NSData *)headerData) length]);
[self.fileHandle writeData:(__bridge NSData *)headerData];
}
@catch (NSException *exception)
{
// Ignore the exception, it normally just means the client
// closed the connection from the other end.
}
@finally
{
NSLog(@" *ding*");
CFRelease(headerData);
CFRelease(response);
[self.server closeHandler:self];
}
}
And here's what shows up in the console log when it crashes:
Jan 15 14:55:10 AWT-NoTouch-iPhone-1 Streamer[1788] <Warning>: [SSProxyTSResponseHandler proxyTS:didReceiveResource:]
Jan 15 14:55:10 iPhone-1 Streamer[1788] <Warning>: -> writing 261760 bytes to filehandle...
Jan 15 14:55:11 iPhone-1 com.apple.launchd[1] (UIKitApplication:com.XXX.Streamer[0xf58][1788]) <Warning>: (UIKitApplication:com.XXX.Streamer[0xf58]) Exited abnormally: Broken pipe: 13
It seems that because the other end closed the pipe the write() fails, so if someone can point me at how I can either discover that it's already closed and not try to write data to it OR whatever will make it not crash my program that would be very helpful.
The immediate problem of crashing with SIGPIPE is solved. I'm not entirely giggly about this solution, but at least the app doesn't crash. It's not clear that it's working 100% correctly, but it does seem to be behaving quite a bit better.
I've resolved this issue by examining further what's going on. In doing some research, I found that perhaps I should be using NSFileHandle's writeabilityHandler property to install a block to do the writing. I'm not fully sold on that approach (it felt convoluted to me), but it might help.
Writability-handler solution:
In doing some web searching on writeabilityHandler, I stumbled on Bert Leung's blog entry on some issues he was having in a similar area. I took his code and modified it as follows, replacing the @try/@catch/@finally block above with this code:
self.pendingData = [NSMutableData dataWithData:(__bridge NSData *)(headerData)];
CFRelease(headerData);
CFRelease(response);
self.fileHandle.writeabilityHandler = ^(NSFileHandle* thisFileHandle)
{
int amountSent = send([thisFileHandle fileDescriptor],
[self.pendingData bytes],
[self.pendingData length],
MSG_DONTWAIT);
if (amountSent < 0) {
// errno is provided by system
NSLog(@"[%@ %@] Error while sending response: %d", NSStringFromClass([self class]), NSStringFromSelector(_cmd), errno);
// Setting the length to 0 will cause this handler to complete processing.
self.pendingData.length = 0;
} else {
[self.pendingData replaceBytesInRange:NSMakeRange(0, amountSent)
withBytes:NULL
length:0];
}
if ([self.pendingData length] == 0) {
thisFileHandle.writeabilityHandler = nil;
// Hack to avoid ARC cycle with self. I don't like this, but...
[[NSNotificationCenter defaultCenter] postNotification:self.myNotification];
}
};
That worked fine but it didn't solve the problem. I was still getting SIGPIPE/EPIPE.
SIGPIPE be gone!
This wasn't a surprise, exactly, as this does pretty much the same thing as the former writeData: did but does it using send() instead. The key difference though is that using send() allows errno to be set. This was quite helpful, actually - I was getting a couple of error codes (in errno), such as 54 (Connection reset by peer) and 32 (Broken pipe). The 54's were fine, but the 32's resulted in the SIGPIPE/EPIPE. Then it dawned on me - perhaps I should just ignore SIGPIPE.
Given that thought, I added a couple of hooks into my UIApplicationDelegate in application:didFinishLaunchingWithOptions:
- (BOOL)application:(UIApplication *)application didFinishLaunchingWithOptions:(NSDictionary *)launchOptions
{
// Override point for customization after application launch.
[self installSignalHandlers];
if ([[UIDevice currentDevice] userInterfaceIdiom] == UIUserInterfaceIdiomPad) {
...
and applicationWillTerminate::
- (void)applicationWillTerminate:(UIApplication *)application
{
// Saves changes in the application's managed object context before the application terminates.
[self removeSignalHandlers];
[self saveContext];
}
-(void)installSignalHandlers
{
signal(SIGPIPE,SIG_IGN);
}
-(void)removeSignalHandlers
{
signal(SIGPIPE, SIG_DFL);
}
Now at least the app doesn't crash. It's not clear that it's working 100% correctly, but it does seem to be behaving.
I also switched back to the @try/@catch/@finally structure because it's more direct. Further, after ignoring SIGPIPE, the @catch block does get triggered. Right now, I'm logging the exception, but only so I can see that it's working. In the released code, that log will be disabled.
