Re: fstat and similar methods in fhandler_socket_local and fhandler_socket_unix

[Ken Brown <kbrown@cornell.edu>]

On 2/24/2021 4:06 AM, Corinna Vinschen via Cygwin-developers wrote:
> On Feb 22 15:01, Ken Brown via Cygwin-developers wrote:
>> OK, I've got patches ready to go for fstat, fstatvfs, fchmod, fchown, facl,
>> and link.  I have to test them, and then I'll send them to cygwin-patches.
>> In the case of link, I think the only way an fhandler_socket's link method
>> can be called is through link(2) being called on a socket file.  So the
>> following should do the job:
>>
>> diff --git a/winsup/cygwin/fhandler.h b/winsup/cygwin/fhandler.h
>> index 21e1df172..f4aa12956 100644
>> --- a/winsup/cygwin/fhandler.h
>> +++ b/winsup/cygwin/fhandler.h
>> @@ -611,7 +611,6 @@ class fhandler_socket: public fhandler_base
>>     int __reg1 fchmod (mode_t newmode);
>>     int __reg2 fchown (uid_t newuid, gid_t newgid);
>>     int __reg3 facl (int, int, struct acl *);
>> -  int __reg2 link (const char *);
>>     off_t lseek (off_t, int)
>>     {
>>       set_errno (ESPIPE);
>> diff --git a/winsup/cygwin/fhandler_socket.cc b/winsup/cygwin/fhandler_socket.cc
>> index f22412650..08870cc6b 100644
>> --- a/winsup/cygwin/fhandler_socket.cc
>> +++ b/winsup/cygwin/fhandler_socket.cc
>> @@ -357,9 +357,3 @@ fhandler_socket::facl (int cmd, int nentries, aclent_t *aclbufp)
>>     set_errno (EOPNOTSUPP);
>>     return -1;
>>   }
>> -
>> -int
>> -fhandler_socket::link (const char *newpath)
>> -{
>> -  return fhandler_base::link (newpath);
>> -}
> 
> Hmm.  What about linkat (AT_EMPTY_PATH, socket, ...)?
> 
> This could be called for inet sockets as well as unnamed or abstract
> unix sockets, in theory...

Thanks, I missed that case.  So we do need fhandler_socket::link.  The only 
question in my mind is what errno it should set.  I just tried the following:

$ cat socket_linkat.c
#define _GNU_SOURCE
#include <sys/socket.h>
#include <sys/un.h>
#include <errno.h>
#include <stdio.h>
#include <stdlib.h>
#include <unistd.h>
#include <fcntl.h>

#define NEW_PATH "/tmp/newpath"

int
main ()
{
   struct sockaddr_un addr;
   int fd;

   if (unlink (NEW_PATH) == -1  && errno != ENOENT)
     {
       perror ("unlink");
       exit (1);
     }
   fd = socket (AF_UNIX, SOCK_STREAM, 0);
   if (fd == -1)
     {
       perror ("socket");
       exit (1);
     }
  if (linkat (fd, "", AT_FDCWD, NEW_PATH, AT_EMPTY_PATH) == -1)
    perror ("linkat");
}

On Linux I get

   linkat: Invalid cross-device link

So I think fhandler_socket::link should fail with EXDEV.  Probably it should 
continue to call fhandler_base::link, and the latter should fail with EXDEV 
instead of EPERM.  Or is there some case where EPERM is more appropriate?

Ken