A disgruntled user poses a question about YP

[Daniel R. Ehrlich]

One of our more knowledgable users has made a few observations about YP
running under SunOS 4 and has posed a number of questions that some
sun-spotters may be able to answer.  If someone has any insights we would
appreciate hearing from you.  Thanks in advance.

--Dan Ehrlich

------- Forwarded Message from dae at shire.cs.psu.edu:

Imagine the following tiny, innocent little program.  All it wants
to do is get information out of the password file.

#include <pwd.h>
main()
{
	struct passwd *pw;

	while (pw = getpwent())
		close(255);
	endpwent();
	exit(0);
}

Now imagine that poor, tiny little program meeting the dreaded monster
we all loathe, YP (as of SunOS 4.0).  The close(255)'s are so we can
see what's happening in a system call trace.

I've deleted stuff from the beginning relating to the shared library
loader (that's for another piece of mail).

**********************************************************************
* YP wakes up and starts mumbling ************************************
**********************************************************************
getdomainname ("", 256) = 0
open ("/etc/passwd", 0, 0666) = 3
ioctl (3, 0x40125401, 0xf7fff366) = -1 ENOTTY (Inappropriate ioctl for device)
fstat (3, 0xf7fff3d8) = 0
brk (0xc0b8) = 0
**********************************************************************
* The next lines deal with lines that were in /etc/passwd. ***********
**********************************************************************
read (3, "root:AkiKKWAJsne5I:0:1:Superuser".., 8192) = 1895
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
close (255) = -1 EBADF (Bad file number)
**********************************************************************
* Now it gets really bad. The first YP call is very difficult,
* involving, among other things,
* -> 1 call  to ioctl()
* -> 2 calls to open() (on the same file)
* -> 2 calls to flock() (on the same file, failing both times)
* -> 2 calls to socket()
* -> 2 calls to gettimeofday()
* -> 3 calls to getpid() (in case we fork() in the middle of getpwent())
* -> 3 calls to bind() (two of which fail, for different reasons)
* Maybe this is why SunOS runs like a pig on small machines?
**********************************************************************
getpid () = 1036
open ("/var/yp/binding/cs.psu.edu.2", 0, 036734075701) = 4
flock (4, 06) = -1 EWOULDBLOCK (Operation would block)
read (4, "".., 2) = 2
close (4) = 0
socket (2, 2, 0) = 4
bind (4, "".., 16) = -1 EADDRINUSE (Address already in use)
close (4) = 0
open ("/var/yp/binding/cs.psu.edu.2", 0, 0) = 4
flock (4, 06) = -1 EWOULDBLOCK (Operation would block)
lseek (4, 2, 0) = 2
read (4, "".., 12) = 12
close (4) = 0
gettimeofday (0xf7fff2c0, 0) = 0
getpid () = 1036
socket (2, 2, 17) = 4
getpid () = 1036
bind (4, "".., 16) = -1 EACCES (Permission denied)
ioctl (4, 0x8004667e, 0xf7fff28c) = 0
fcntl (4, 02, 0x1) = 0
bind (4, "".., 16) = 0
getsockname (4, 0xf7fff334, 0xf7fff354) = 0
sendto (4, "".., 88, 0, 0x46b8, 16) = 88
getdtablesize () = 64
select (64, 0xf7fff2e4, 0, 0, 0x46cc) = 1
recvfrom (4, "".., 1600, 0, 0xf7fff2d0, 0xf7fff2e0) = 112
gettimeofday (0xf7fff468, 0xf7fff460) = 0
close (255) = -1 EBADF (Bad file number)
**********************************************************************
* Now things settle down into a pattern.  Not bad, 10 system calls
* per query.
**********************************************************************
getpid () = 1036
open ("/var/yp/binding/cs.psu.edu.2", 0, 036734075701) = 5
flock (5, 06) = -1 EWOULDBLOCK (Operation would block)
read (5, "".., 2) = 2
close (5) = 0
getsockname (4, 0xf7fff344, 0xf7fff354) = 0
sendto (4, "".., 88, 0, 0x46b8, 16) = 88
select (64, 0xf7fff2e4, 0, 0, 0x46cc) = 1
recvfrom (4, "".., 1600, 0, 0xf7fff2d0, 0xf7fff2e0) = 104
gettimeofday (0xf7fff468, 0xf7fff460) = 0
close (255) = -1 EBADF (Bad file number)
**********************************************************************

Ok, question time.

* Does anybody at Sun use the trace program?
* There are 14 characters in /var/yp/binding/cs.psu.edu.2.  Why do
  we only look at the first two?
* Why doesn't it like the first (TCP?) socket?
* Is there nothing we can do to avoid doing one namei() per query?
  It is, after all, probably the single hardest thing for the kernel
  to do...
* What is the flock() supposed to do?  What would we do differently if
  it worked?  Should it succeed?  Who has it locked?
* Does Hesiod work yet?  Are there any other alternatives to YP?  Please?

Daemon