Further insights after several kernel recompilations. | I checked 2.6.19.x (which works), 2.6.20, 2.6.20.1 and 2.6.21-rc1 | (which all have the bug). 2.6.21-rc2 wouldn't compile on my system | due to some apic isssue. Since I couldn't reproduce the bug on the current kernel (2.6.21-rc2), I reverted to 2.6.21-rc1 and tried the paraslash application as described. I got a "main: dccp connect error" when connecting to the paraserver (which was listening on port 5001, as `ss' reported). With wireshark, I could see that the handshake and the close request were both handled correctly, so it seems that the connect error was rather down to a user error. After recompiling to the most recent 2.6.20-rc2, same error. I also changed configuration to PREEMPT but need to leave further testing for now. | | > If it is a very recent | > one, can you please check whether you get the following message in your | > syslog: | > "[...] listen_overflow!" | > | > If yes (dccp_debug should be turned on), then very likely setting listen(fd, 1) | > instead of listen(fd,0) may remove any strange effects. | > The reason is a recent change in sk_accept_queue_is_full which causes a different | > treatment of zero-sized listen-accept queues. | | Will test this evening and report tomorrow. This is indeed a(nother) bug - zero-sized listen queues currently do not work. Patch submitted to netdev. | | > 2. With the most recent davem-2.6 kernel I was not able to reproduce this bug. It | > should, after some more thought, really make no difference whether you are using | > loopback (127.0.0.1) or not. | | I can try this kernel as well. I'm currently downloading | | git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6.git | | Hope, that's the right one. Yes that should work. There is also a net-2.6.22.git but I think it is only there for the more recent changes which not yet have made it into 2.6. | > (a) an update of this patch | > (b) a second patch to do with timer initialisation of child sockets. | > In particular (a) may help. | | OK. Are these patches against the current linus-tree? They should work on any recent 2.6.19/20 tree (maybe with some offset). | > I managed to get the paraslash application built, but could not figure out how to | > populate the user lists and required configuration files. | | That's explained in the INSTALL file. But you don't need any Thanks - it was not spite not to rtfm, rather since busy with a lot of other stuff at the moment. | > I don't understand your code fully yet, but with the more recent stack trace I | > was wondering whether this has to do with setting the listen socket non-blocking | > (mark_fd_nonblock), which is done both in sender and receiver. | | IMHO it's considered good practice to set all fds which are used for | select() to non-blocking mode. AFAIR the reason is the situation | where a network packet arrives but is discarded because of a checksum | error. In this case it might happen that select() indicates readability | of an fd, but a subsequent read() blocks nevertheless. Maybe it's | unneccesssary to set an fd to non-blocking mode if it is only used | for writing. But it won't hurt either, so.. Oh sorry - it was not my intention to talk about programming style. The code is fine and you are right with what you are saying. It is just that I wanted to pin down the exact cause of the bug as in `it happens on non-blocking socket when X and Y and Z hold' &) - To unsubscribe from this list: send the line "unsubscribe dccp" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
