cvs commit: src/lib/libc/gen lockf.c src/lib/libc/sys
Symbol.map fcntl.2 src/sys/compat/freebsd32 syscalls.master
src/sys/compat/linux linux_file.c src/sys/compat/svr4
svr4_fcntl.c src/sys/conf NOTES files options ...
dfr at rabson.org
Thu Mar 27 01:41:35 PDT 2008
On 27 Mar 2008, at 03:56, Xin LI wrote:
> Great work! I have not dig into the code yet but just curious -
> will it be feasible to have the userland RPC client to share some
> code to make it MTSAFE?
I took the userland RPC code as a starting point but I have not tried
to keep them compatible. The main issue for an MTSAFE RPC client is
dealing with replies. The existing locking in the userland RPC code
copes with sending the request messages safely but would have problems
handling the replies if two threads were trying to read from the socket.
It would be possible to take a similar approach to fixing this in
userland. Basically you would build a list of pending requests and
arrange for only one thread to read replies at any given time. That
thread would examine each reply message and match it to a pending
request by examining the XID field. The code is likely to be a bit
different from the kernel version but the concept should be similar.
More information about the cvs-src