tty-dev-number <-> /dev/pts

[Uwe Wolfram]

Hi, I'm writing a program under Solaris (sparc) that has to know
from what machine a user is logged on who is starting that program.

Reading /proc/self/psinfo only returns a device number for the
controlling terminal (long int) whereas the functions listed
in getutxent(3C) give me the device name of the logged on user as
listed by who as well as the remote host (if any).
Is there a way to link all those informations:
device number from /proc -> /dev/pts -> utmpx entry -> remote host?
Or is there a shortcut/simple function to achieve this?

Uwe

--
anti-spam: Remove "duMMy." from address above to obtain correct one

[Uwe Wolfram]

Richard L. Hamilton wrote:

> > Reading /proc/self/psinfo only returns a device number for the
> > controlling terminal (long int) whereas the functions listed
> > in getutxent(3C) give me the device name of the logged on user as
> > listed by who as well as the remote host (if any).
> > Is there a way to link all those informations:
> > device number from /proc -> /dev/pts -> utmpx entry -> remote host?
> > Or is there a shortcut/simple function to achieve this?

>
> Given a file descriptor to a terminal, ttyname(3c) will (usually)
> give the name of the terminal, which you can then scan utmpx for.

Thanks, that works. Since the app doesn't touch the stdin/stdout file
descriptors, a ttyname (0) returns the pts/# that I can feed to
getutxline(3C).

Uwe

--
anti-spam: Remove "duMMy." from address above to obtain correct one

[Richard L. Hamilton]

In article <20030711203016.37dba3ac.u.wolfram@dummy.freenet.d e>,
Uwe Wolfram <u.wolfram@duMMy.freenet.de> writes:

> Richard L. Hamilton wrote:

>> > Reading /proc/self/psinfo only returns a device number for the
>> > controlling terminal (long int) whereas the functions listed
>> > in getutxent(3C) give me the device name of the logged on user as
>> > listed by who as well as the remote host (if any).
>> > Is there a way to link all those informations:
>> > device number from /proc -> /dev/pts -> utmpx entry -> remote host?
>> > Or is there a shortcut/simple function to achieve this?

>>
>> Given a file descriptor to a terminal, ttyname(3c) will (usually)
>> give the name of the terminal, which you can then scan utmpx for.

>
> Thanks, that works. Since the app doesn't touch the stdin/stdout file
> descriptors, a ttyname (0) returns the pts/# that I can feed to
> getutxline(3C).

That's ok if you're sure nobody will ever run it with stdin redirected.

The following code works as best it can to find the name of the
controlling terminal whether stdin/stdout/stderr has been redirected or
not. First, it gets pr_ttydev, to know that whatever it finds really is
the controlling terminal. Then, it checks fd 2, 1, 0 until one of them
matches; if one does, it uses ttyname() on that. If none do, it uses
ftw() to search /dev for something with a st_rdev that matches pr_ttydev.
Warning: the code is _not_ reentrant - that would be a but more work, and
one should typically only need to run it once in a process anyway.
Disclaimer: just wrote it very quickly (20 minutes or so), only minimally
tested.

Test run:

$ gcc -Wall -pedantic -DTEST_MAIN ctermname.c -o ctermname
$ ./ctermname
/dev/pts/6
$ ./ctermname </dev/null 2>&1 |cat # with all three redirected
/dev/pts/6


code:

#include <stdio.h>
#include <sys/param.h>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <errno.h>
#include <procfs.h>
#include <ftw.h>
#include <unistd.h>
#include <string.h>

static dev_t rdev;
static char name_found[MAXPATHLEN+1];

static int searchfunc(const char *name, const struct stat *statp, int type)
{
if (type!=FTW_F || !S_ISCHR(statp->st_mode) || statp->st_rdev!=rdev)
return 0;

strncpy(name_found,name,MAXPATHLEN);
name_found[MAXPATHLEN]='\0';
return 1;
}

const char *ctermname()
{

struct psinfo p;
struct stat s;
int psinfo_fd;
int errno_save;
int use_fd;


if ((psinfo_fd=open("/proc/self/psinfo",O_RDONLY)) == -1)
return NULL; /* can't obtain controlling terminal device # */

switch(read(psinfo_fd,&p,sizeof p)) {
case sizeof p: /* good */
break;
case -1: /* error */
errno_save=errno;
close(psinfo_fd);
errno=errno_save;
return NULL;
default: /* unexpected short read */
close(psinfo_fd);
errno=EINVAL; /* or whatever you think is most appropriate */
return NULL;
}

close(psinfo_fd);

if (p.pr_ttydev==PRNODEV) { /* no controlling terminal */
errno=ENODEV; /* seems like best choice */
return NULL;
}

/* now try to fstat() 2, 1, 0 until one gives something useful; if
none do, we can fall back to a slower approach */

if ((fstat(use_fd=2,&s) != -1 && S_ISCHR(s.st_mode) && s.st_rdev==p.pr_ttydev) ||
(fstat(use_fd=1,&s) != -1 && S_ISCHR(s.st_mode) && s.st_rdev==p.pr_ttydev) ||
(fstat(use_fd=0,&s) != -1 && S_ISCHR(s.st_mode) && s.st_rdev==p.pr_ttydev)) {
/* use_fd now corresponds to an fd on the controlling tty */
return ttyname(use_fd);
}

/* if we get this far, none of stdin, stdout, stderr fds matched the
controlling device, so we have to do it the hard, slow way */

rdev=p.pr_ttydev; /* what we're searching for */
name_found[0]='\0'; /* just to be safe, we clear this */

if (ftw("/dev",searchfunc,7) == 1) {
return name_found;
}

/* still nothing, set errno to something interesting and return NULL */
errno=ENOENT;
return NULL;
}

#ifdef TEST_MAIN

int main(int argc, char **argv)
{
const char *p;

if ((p=ctermname())!=NULL) {
printf("%s\n",p);
return 0;
}
else {
perror("ctermname() failed");
return 1;
}
}
#endif


--
mailto:rlhamil@mindwarp.smart.net [url]http://www.smart.net/~rlhamil[/url]

[Uwe Wolfram]

Richard L. Hamilton wrote:

> That's ok if you're sure nobody will ever run it with stdin redirected.

Quite sure since it's an X11-app, but of course, I cannot guarantee it.

> The following code works as best it can to find the name of the
> controlling terminal whether stdin/stdout/stderr has been redirected or
> not. First, it gets pr_ttydev, to know that whatever it finds really is
> the controlling terminal. Then, it checks fd 2, 1, 0 until one of them
> matches; if one does, it uses ttyname() on that. If none do, it uses
> ftw() to search /dev for something with a st_rdev that matches
> pr_ttydev. Warning: the code is _not_ reentrant - that would be a but
> more work, and one should typically only need to run it once in a
> process anyway. Disclaimer: just wrote it very quickly (20 minutes or
> so), only minimally tested.

Quite a bunch of lines :-)
I'm going to examine it later, thanks a lot for providing the code.
Reentrance is definitely not a requirement.

Uwe

--
anti-spam: Remove "duMMy." from address above to obtain correct one