Greetings. I have an application (a FTP server) that is behaving oddly on a 2.2.19 kernel. In response to a retrieval command, it does: socket connect or accept, depending on mode fcntl +O_NONBLOCK poll write ... poll write fcntl -O_NONBLOCK close write(1, "226 Transfer Complete...") Standard output (fd 1) is also connected to a socket to the same client. What I'm seeing is that packet containing the "226" message is frequently sent to the client (according to tcpdump on the server) *before* the data from the final write on the new socket, even though that socket is closed (and the return from the close checked) before the "226" write. I am currently unable to reproduce this effect with a 2.4 kernel. Has anybody else seen this behavior? Any ideas what might cause it? I can produce complete strace and tcpdump logs if that is of use. -- Bruce Guenter <bruceg@em.ca> http://em.ca/~bruceg/ http://untroubled.org/ OpenPGP key: 699980E8 / D0B7 C8DD 365D A395 29DA 2E2A E96F B2DC 6999 80E8
Attachment:
pgp00029.pgp
Description: PGP signature
