* Simon Josefsson:
>> Suppose that a package wants to create a UNIX domain socket as part of
>> its test suite. If the socket is created within the package build
>> directory, this might fail because of the quite low path name length
>> limit. What is the correct way of dealing with this? "mktemp -d" uses
>> TMPDIR, which is potentially affected by the same issue.
>>
>> (My personal opinion is "fix the buildd", especially since none of the
>> official buildds seems to use long path names, but there is
>> disagreement.)
>
> Can't the low path name length limit be fixed? What restricts the
> length? Kernel, libc, filesystem, ...?
It's part of the API:
/* Structure describing the address of an AF_LOCAL (aka AF_UNIX)
socket. */
struct sockaddr_un
{
__SOCKADDR_COMMON (sun_);
char sun_path[108]; /* Path name. */
};
I haven't checked if the Linux kernel actually cares about the actually
length of this data structure (there are explicit length fields in all
cases), but I guess it should work. However, this might not be portable
to the Hurd or kfreebsd.
In a larger context, there is sockaddr_storage, which indirectly imposes
a limit (because getpeername must complete when passed a
sockaddr_storage argument, for instance). Overlong UNIX domain sockets
might break some libraries that rely on that functionality.
--
To UNSUBSCRIBE, email to [EMAIL PROTECTED]
with a subject of "unsubscribe". Trouble? Contact [EMAIL PROTECTED]
