IdTCPClient ConnectTimeout broken on linux >6.1.32 (potentially affects higher-ups like TIdHTTP* ? ) about indy HOT 4 CLOSED

I'm not familiar with how the Linux kernel works, but Indy has never relied on OS-level timeouts for its ConnectTimeout handling, especially since there is no standard socket option to specify a timeout for connect(). As Indy uses blocking sockets, it runs connect() in its own worker thread, waiting up to ConnectTimeout for that thread to terminate. If the thread does not terminate in time, shutdown() is then called to abort the connection in progress, and close() to close the socket. So, the only hangup should be if connect() is not respecting shutdown()/close() in a timely manner.

That being said, there is an open ticket to re-design ConnectTimeout handling to use a non-blocking connect() with select()/(e)poll() on platforms which support that option. But that is work for a future release.

from indy.

So, the only hangup should be if connect() is not respecting shutdown()/close() in a timely manner.

Yes, and as I understand, the kernel now prohibits shutting down the connection attempt, and Indy's mechanism is now useless.

https://github.com/torvalds/linux/blob/4faeee0cf8a5d88d63cdbc3bab124fb0e6aed08c/net/ipv4/tcp.c#L3084-L3088

shutdown() does not work... further down TIdStackUnix.WSShutdown returns -1 . So the thread keeps waiting until eventually the OS decides to give up on connect()

from indy.

Yes, and as I understand, the kernel now prohibits shutting down the connection attempt, and Indy's mechanism is now useless.

Oh, well, that would certainly suck :-( That doesn't seem like a very good change for them to make in the kernel, and it certainly isn't documented behavior, AFAICS.

https://github.com/torvalds/linux/blob/4faeee0cf8a5d88d63cdbc3bab124fb0e6aed08c/net/ipv4/tcp.c#L3084-L3088

shutdown() does not work... further down TIdStackUnix.WSShutdown returns -1 . So the thread keeps waiting until eventually the OS decides to give up on connect()

Then I guess I need to bump up the priority on #3.

from indy.

I have verified it working again. Without changing anything in freepascal, it works in linux 6.6.2 and 6.1.63. (timeout in the program is 3000ms, see above).

Most probably fixed by torvalds/linux@419ce13 , as mentioned in #493 .

$ uname -a
Linux 6.6.2-v8+ #1 SMP PREEMPT Tue Nov 21 22:28:24 UTC 2023 
aarch64 GNU/Linux

and 

$ uname -a
Linux 6.1.63-v8+ #1700 SMP PREEMPT Thu Nov 23 14:10:01 GMT 2023 
aarch64 GNU/Linux

/dev/shm $  strace -r -f -e trace=open,socket,connect,bind 
 ./conecttimeout
      0.000000 socket(AF_INET, SOCK_STREAM, IPPROTO_IP) = 3
     0.000889 bind(3, {sa_family=AF_INET, sin_port=htons(0), sin_addr=inet_addr("0.0.0.0")}, 16) = 0
strace: Process 1162 attached
[pid  1162]      0.002929 connect(3, {sa_family=AF_INET, sin_port=htons(8090), sin_addr=inet_addr("192.168.1.184")}, 16) = -1 ECONNRESET 
[pid  1162]      3.008621 +++ exited with 0 +++
An unhandled exception occurred at $00000000004E40A4:
EIdConnectTimeout: Connect timed out.
  $00000000004E40A4  DOCONNECTTIMEOUT,  line 328 of ../../lib/indy10/Core/IdIOHandlerStack.pas
  $00000000004E32CC  OPEN,  line 369 of ../../lib/indy10/Core/IdIOHandlerSocket.pas
  $000000000045DFC8  CONNECT,  line 328 of ../../lib/indy10/Core/IdTCPClient.pas
  $000000000045E90C  CONNECT,  line 514 of ../../lib/indy10/Core/IdTCPClient.pas
  $00000000004009F0  main,  line 22 of ../../../../../dev/shm/conecttimeout
  $0000007F9AF1EE18
  $0000000000400928

     1.503324 +++ exited with 217 +++ 

from indy.