Re: Adding file descriptor attribute(s) to gcc and glibc

[David Malcolm <dmalcolm@redhat.com>]

On Wed, 2022-07-13 at 09:37 +0100, Szabolcs Nagy wrote:
> The 07/12/2022 18:25, David Malcolm via Libc-alpha wrote:
> > On Tue, 2022-07-12 at 18:16 -0400, David Malcolm wrote:
> > > On Tue, 2022-07-12 at 23:03 +0530, Mir Immad wrote:
> > > GCC's attribute syntax here:
> > >   https://gcc.gnu.org/onlinedocs/gcc/Attribute-Syntax.html
> > > allows for a parenthesized list of parameters for the attribute,
> > > which
> > > can be:
> > >  (a) An identifier
> > >  (b) An identifier followed by a comma and a non-empty comma-
> > > separated
> > > list of expressions
> > >  (c) A possibly empty comma-separated list of expressions
> > > 
> > > I'd hoped to have an argument number, with an optional extra param
> > > describing the direction of the access, but syntax (b) puts the
> > > identifier first, alas.
> > > 
> > > Here's one possible way of doing it with a single attribute, via
> > > syntax
> > > (b):
> > > e.g.
> > >    __attribute__((fd_argument (access, 1))
> > >    __attribute__((fd_argument (read, 1))
> > >    __attribute__((fd_argument (write, 1))
> > > 
> > > meaning that argument 1 of the function is expected to be an open
> > > file-
> > > descriptor, and that it must be possible to read from/write to that
> > > fd
> > > for cases 2 and 3.
> > > 
> > > Here are some possible examples of how glibc might use this syntax:
> > > 
> > >     int dup (int oldfd)
> > >       __attribute((fd_argument (access, 1)); 
> > > 
> > >     int ftruncate (int fd, off_t length)
> > >       __attribute((fd_argument (access, 1)); 
> > > 
> > >     ssize_t pread(int fd, void *buf, size_t count, off_t offset)
> > >       __attribute((fd_argument (read, 1));
> > > 
> > >     ssize_t pwrite(int fd, const void *buf, size_t count, 
> > >                    off_t offset);
> > >       __attribute((fd_argument (write, 1));
> > > 
> > > ...but as I said, I'm most interested in input from glibc
> > > developers on
> > > this.
> 
> note that glibc headers have to be namespace clean so it
> would be more like
> 
>   __attribute__((__fd_argument (__access, 1)))
>   __attribute__((__fd_argument (__read, 1)))
>   __attribute__((__fd_argument (__write, 1)))
> 
> so it would be even shorter to write
> 
>   __attribute__((__fd_argument_access (1)))
>   __attribute__((__fd_argument_read (1)))
>   __attribute__((__fd_argument_write (1)))

As I understand it, you'd use a macro for this, but this made me think
of the following attributes that GCC could provide:

__attribute__ ((fd_arg(N)))
__attribute__ ((fd_arg_read(N)))
__attribute__ ((fd_arg_write(N)))

(since GCC already has "__attribute__((format_arg(N)))")

It looks like you define your attribute macros in:
  https://sourceware.org/git/?p=glibc.git;a=blob_plain;f=misc/sys/cdefs.h;hb=HEAD

which presumably could be extended to add something like:

#if __GNUC_PREREQ (13, 0)
# define __attr_fd_arg(argno) __attribute__ ((fd_arg(argno)))
# define __attr_fd_arg_read(argno)  __attribute__ ((fd_arg_read(argno)))
# define __attr_fd_arg_write(argno) __attribute__ ((fd_arg_write(argno)))
#else
# define __attr_fd_arg(argno)
# define __attr_fd_arg_read(argno)
# define __attr_fd_arg_write(argno)
#endif

if I've got my syntax correct.

(Or maybe "readable" and "writable"?)


> 
> I just realized that the attribute could accept both the single integer
> argument number (syntax (c)) for the "don't care about access
> direction" case, or the ({read|write}, N) of syntax (b) above, giving
> e.g.:
> 
>     int dup (int oldfd)
>       __attribute((fd_argument (1)); 
> 
>     int ftruncate (int fd, off_t length)
>       __attribute((fd_argument (1)); 
> 
>     ssize_t pread(int fd, void *buf, size_t count, off_t offset)
>       __attribute((fd_argument (read, 1));
> 
>     ssize_t pwrite(int fd, const void *buf, size_t count, 
>                    off_t offset);
>       __attribute((fd_argument (write, 1));
> 
> for the above examples.
> 
> How does that look?
> Dave

i think fd in ftruncate should be open for writing.

Agreed.

So with the above macros, this might look like:

    int dup (int oldfd)
       __attr_fd_arg(1);

    int ftruncate (int fd, off_t length)
       __attr_fd_arg_write(1);

    ssize_t pread(int fd, void *buf, size_t count, off_t offset)
       __attr_fd_arg_read(1);

    ssize_t pwrite(int fd, const void *buf, size_t count, 
                   off_t offset);
       __attr_fd_arg_write(1);


to be honest, i'd expect interesting fd bugs to be
dynamic and not easy to statically analyze.
the use-after-unchecked-open maybe useful. i would
not expect the access direction to catch many bugs.

One goal of -fanalyzer is to help detect problems as code is written,
before it ever leaves the developer's workstation.  So for instance it
might save a few seconds helping catch silly bugs where a developer is
working with two different FDs and gets the read and write FDs the
wrong way around.

Dave