Date: Sun, 21 Jun 1998 15:27:17 +0800 From: Peter Wemm <peter@netplex.com.au> To: Terry Lambert <tlambert@primenet.com> Cc: joelh@gnu.org, fenner@parc.xerox.com, current@FreeBSD.ORG Subject: Re: Bogus errno twiddling by lstat... Message-ID: <199806210727.PAA27276@spinner.netplex.com.au> In-Reply-To: Your message of "Sat, 20 Jun 1998 20:02:49 GMT." <199806202002.NAA17957@usr02.primenet.com>
next in thread | previous in thread | raw e-mail | index | archive | help
Terry Lambert wrote:
> > If, however, you don't like our malloc's current implementation, and
> > think that errno should be saved and restored across a successful
> > call, and that the lost cycles would be worthwhile, then diffs would
> > be perfectly welcome, I'm sure.
>
> Since the malloc.conf file is unnecessarily being looked for in the
> printf case (since ld.so already caused it to be looked for, and didn't
> find it), I think the redundant call to llok for it is certainly worth
> removing.
What redundant call??!?
pwroot@spinner[2:59pm]/tmp-115# cat xx.c
#include <stdio.h>
#include <string.h>
#include <sys/stat.h>
#include <errno.h>
main()
{
struct stat sb;
int saveerrno, saveerrno2;
int ret;
char *e;
write(1, "1\n", 2);
errno = 0;
ret = lstat("foo", &sb);
saveerrno = errno;
write(1, "2\n", 2);
e = strerror(saveerrno);
saveerrno2 = errno;
write(1, "3\n", 2);
printf("lstat returned %d, errno = %d = %s\n", ret, saveerrno, e);
write(1, "4\n", 2);
printf("errno is now: %d, saveerrno2 = %d\n", errno, saveerrno2);
write(1, "5\n", 2);
}
pwroot@spinner[2:59pm]/tmp-116# cc -o xx xx.c
pwroot@spinner[2:59pm]/tmp-117# ktrace ./xx
1
2
3
lstat returned 0, errno = 0 = Undefined error: 0
4
errno is now: 2, saveerrno2 = 0
5
pwroot@spinner[2:59pm]/tmp-118# l foo
0 -rw-r--r-- 1 peter wheel 0 Jun 18 16:52 foo
pwroot@spinner[2:59pm]/tmp-119# kdump | grep conf
27160 xx NAMI "/etc/malloc.conf"
Quite clearly there is only one lookup..
pwroot@spinner[DING!]/tmp-120# kdump | more
27160 ktrace RET ktrace 0
27160 ktrace CALL execve(0xefbfd8e3,0xefbfd870,0xefbfd878)
27160 ktrace NAMI "./xx"
27160 xx RET execve 0
27160 xx CALL open(0x10a0,0,0)
27160 xx NAMI "/usr/libexec/ld.so"
27160 xx RET open 3
[... trimmed ...]
27160 xx CALL write(0x1,0x156a,0x2)
27160 xx GIO fd 1 wrote 2 bytes
"3
"
27160 xx RET write 2
27160 xx CALL fstat(0x1,0xefbfd458)
27160 xx RET fstat 0
27160 xx CALL readlink(0x2006e8e2,0xefbfd454,0x3f)
27160 xx NAMI "/etc/malloc.conf"
27160 xx RET readlink -1 errno 2 No such file or directory
27160 xx CALL mmap(0,0x1000,0x3,0x1002,0xffffffff,0,0,0)
27160 xx RET mmap 536952832/0x20014000
27160 xx CALL break(0x7000)
27160 xx RET break 0
27160 xx CALL break(0x17000)
27160 xx RET break 0
27160 xx CALL ioctl(0x1,TIOCGETA,0xefbfd494)
27160 xx RET ioctl 0
27160 xx CALL write(0x1,0x7000,0x31)
27160 xx GIO fd 1 wrote 49 bytes
"lstat returned 0, errno = 0 = Undefined error: 0
"
27160 xx RET write 49/0x31
27160 xx CALL write(0x1,0x1591,0x2)
27160 xx GIO fd 1 wrote 2 bytes
"4
"
27160 xx RET write 2
[..]
Now, I don't see any redundant calls to readlink("/etc/malloc.conf"),
and no calls to malloc() by ld.so (ld.so has it's own malloc that it uses
during startup and switches to the libc malloc after it's got hold of
libc.so.*).
Also, since this is freebsd-current, so I assume you're talking about
-current.
> I would also not save and restore state, except in anticipation of
> an error. If the cached value were being used (as it should be),
> then the malloc.conf reference shouldn't have been tried (and not being
> tried, the errno would not be in danger of being blown).
malloc_init() and the readlink are only called once. I think you're
imagining this.
So, first you blame lstat() with a badly flawed test program as "evidence".
When challenged on it, you say "I knew that" and then claim that ld.so is
calling malloc. At the same time you claim that this is therefore violating
some ansi requirement for functions to not change errno, which somebody
else has pointed out that the specs say the opposite. When confronted with
that, you then come up with malloc not caching it's readlink() of
malloc.conf, which is clearly false and verifiable.
What next?
malloc() has been adjusted to preserve errno across the readlink(), so I
suggest you leave it at that rather than digging youself an even deeper
hole. :-)
Cheers,
-Peter
To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message
Want to link to this message? Use this URL: <https://mail-archive.FreeBSD.org/cgi/mid.cgi?199806210727.PAA27276>