Stray nulls when reading from AF_UNIX sockets
Hi,
With recent kernels (2.6.22.x), we are experiencing that random null characters are read by our mySQL server (i.e. the number of bytes read by the server is larger than the number of bytes written). I'm currently investigating the problem (that occurs every 1-35 hours on a loaded system). A current suspect is the following code in net/unix/af_unix.c:
static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
struct msghdr *msg, size_t size, int flags)
{
...
return copied ? : err;}
Shouldn't this read:
return copied ? copied : err;
Or am I missing something?
Please CC me personally.
Regards
Anders Blomdell
Signature · 5 lines
Anders Blomdell Email: [email protected] Department of Automatic Control Lund University Phone: +46 46 222 4625 P.O. Box 118 Fax: +46 46 138118 SE-221 00 Lund, Sweden
Quoted 16 lines · On Thu, 06 Sep 2007 17:48:07 +0200 Anders Blomdell wrote:
Hi,
With recent kernels (2.6.22.x), we are experiencing that random null characters are read by our mySQL server (i.e. the number of bytes read by the server is larger than the number of bytes written). I'm currently investigating the problem (that occurs every 1-35 hours on a loaded system). A current suspect is the following code in
net/unix/af_unix.c:static int unix_stream_recvmsg(struct kiocb *iocb, struct socket *sock,
struct msghdr *msg,size_tsize, int flags){ ... return copied ? : err;}
Quoted 7 lines
Shouldn't this read:
return copied ? copied : err;Or am I missing something?
Please CC me personally.
It's (?:) a gcc extension. See
http://gcc.gnu.org/onlinedocs/gcc-4.2.1/gcc/Conditionals.html#Conditionals
---
~Randy
*** Remember to use Documentation/SubmitChecklist when testing your code ***
From: Anders Blomdell <[email protected]>
Date: Thu, 06 Sep 2007 17:48:07 +0200
Quoted 8 lines
return copied ? : err;}
Shouldn't this read:
return copied ? copied : err;Or am I missing something?
These two statements are equivalent, the first version is a shorthand allowed by gcc.
Quoted 11 lines · David Miller
Quoted 8 lines
return copied ? : err;}
Shouldn't this read:
return copied ? copied : err;Or am I missing something?
These two statements are equivalent, the first version is a shorthand allowed by gcc.
Not only that. With x?x:z, x is evaluated twice,
while with x?:z, x is only evaluated once. That's for stuff when you want to, say [dumb example follows],
size_t my_read(..) {
return read(..) ? : -1
}and the only other way would be to use a temporary,
size_t my_read(..) {
size_t x = read(..);
return x ? x : -1;
}gcc should be smart enough to also do optimization in the second case..
Jan