Re: [dm-devel] [PATCH V7 2/4] multipath-tools: New way to limit the IPC command length.

2016-08-12 Thread Benjamin Marzinski
On Fri, Aug 12, 2016 at 08:12:38PM +0800, Gris Ge wrote: Perhaps a better way would be to keep your v6 code, and just make the client ignore the SIGPIPE signals. Personally, I'm fine with always ignoring them, since the writes will still fail with EPIPE, and the multipathd client code should be

Re: [dm-devel] [PATCH V7 2/4] multipath-tools: New way to limit the IPC command length.

2016-08-12 Thread Bart Van Assche
On 08/12/2016 05:12 AM, Gris Ge wrote: * Moved write_all() to file.h of libmultipath as all usage of write_all() is against file rather than socket. This sounds very wrong. As far as I can see write_all() is used to send data over a Unix socket. A Unix socket is not a file.

[dm-devel] [PATCH V7 2/4] multipath-tools: New way to limit the IPC command length.

2016-08-12 Thread Gris Ge
Problem: mpath_recv_reply() return -EINVAL on command 'show maps json' with 2k paths. Root cause: Commit 174e717d351789a3cb29e1417f8e910baabcdb16 introduced the limitation on max bytes(65535) of reply string from multipathd. With 2k paths(1k mpaths) simulated by scsi_debug, the