Thread: [BUG/PATCH] backend crashes during authentication if data/global/pg_pwd is empty

[BUG/PATCH] backend crashes during authentication if data/global/pg_pwd is empty

From
Michael Wildpaner
Date:
Hi,

on Solaris 9 with PostgreSQL 7.4:

when you
 - 'initdb' a fresh database, - _don't_ set a password for user 'postgres', - convert the 'trust' lines in
data/pg_hba.confto 'md5' or 'password'
 

and then try to connect as user 'postgres', the backend crashes in
backend/libpq/hba.c:372:
   368 static int   369 user_group_bsearch_cmp(const void *user, const void *list)   370 {   371         /* first node
isline number */   372         char       *user2 = lfirst(lnext(*(List **) list));
 

due to 'list' being NULL, which might mean that 'user_sorted' was never
allocated, due to user_length being zero for an missing or empty
global/pg_pwd:
   916         /* create sorted lines for binary searching */   917         user_length = length(user_lines);   918
   if (user_length)   919         {   920                 int                     i = 0;   921   922
user_sorted= palloc(user_length * sizeof(List *));
 

I know this is looks like a case of "don't do it, then", but since it's a
backend crash, I would suggest the following fix:

--- postgresql-7.4.orig/src/backend/libpq/hba.c    2003-10-25 05:48:46.000001000 +0200
+++ postgresql-7.4/src/backend/libpq/hba.c    2003-12-05 15:21:54.000003000 +0100
@@ -62,7 +62,7 @@static List **user_sorted = NULL;        /* sorted user list, for bsearch() */static List
**group_sorted= NULL;        /* sorted group list, for                                         * bsearch() */
 
-static int    user_length;
+static int    user_length = 0;static int    group_length;
static List *tokenize_file(FILE *file);
@@ -395,6 +395,10 @@List      **get_user_line(const char *user){
+    /* fail if there is nothing to search in */
+    if ((user_sorted == NULL) || (user_length == 0))
+        return NULL;
+    return (List **) bsearch((void *) user,                             (void *) user_sorted,
  user_length,
 

The initialization of user_length might not be necessary.

Best wishes, Mike

PS: This might be related to   http://archives.postgresql.org/pgsql-admin/2003-03/msg00413.php

-- 
Life is like a fire.                            DI Michael Wildpaner
Flames which the passer-by forgets.                    Ph.D. Student
Ashes which the wind scatters.
A man lived.       -- Omar Khayyam


Michael Wildpaner <mike@rainbow.studorg.tuwien.ac.at> writes:
> +    /* fail if there is nothing to search in */
> +    if ((user_sorted == NULL) || (user_length == 0))
> +        return NULL;

Hm, Solaris' bsearch() fails on empty input?  How bizarre.
Easily worked around though --- thanks for the report!

I suspect we'd better put a defense in get_group_line as well.
It looks like there are no other places at risk in the backend.
        regards, tom lane


> Hm, Solaris' bsearch() fails on empty input?  How bizarre.

I was skeptical but apparently this is a known bug ...
googling turned up a couple of references, eg
http://www.opencm.org/pipermail/opencm-dev/2002-July/001077.html
        regards, tom lane


Re: [BUG/PATCH] backend crashes during authentication if

From
Michael Wildpaner
Date:
Hi,

On Fri, 5 Dec 2003, Tom Lane wrote:
> > Hm, Solaris' bsearch() fails on empty input?  How bizarre.
>
> I was skeptical but apparently this is a known bug ...
> googling turned up a couple of references, eg
> http://www.opencm.org/pipermail/opencm-dev/2002-July/001077.html

in defense of Solaris' bsearch it should be said that it only breaks if
one passes NULL as the array base, a decidedly undefined (and unfriendly)
case. Passing an array with zero elements works as advertised. Btw, the
same happens on IRIX.

Best wishes, Mike

PS: A little program to demonstrate this: array with elements, empty   array, NULL pointer as base:


#include <stdio.h>
#include <stdlib.h>
#include <string.h>

char strings[][4] = { "abc", "efg", "hij", "klm" };

typedef int (*cmp_t)(const void*, const void*);

int main(int argc, char**argv) { char *s, *term = "hij";
 s = bsearch(term, strings, sizeof(strings)/sizeof(char[4]),               sizeof(char*), (cmp_t) strcmp);
fprintf(stderr,"1: %s\n", (s != NULL) ? s : "<not found>");
 
 s = bsearch(term, strings, 0, sizeof(char*), (cmp_t) strcmp); fprintf(stderr, "2: %s\n", (s != NULL) ? s : "<not
found>");
 s = bsearch(term, NULL, 0, sizeof(char*), (cmp_t) strcmp); fprintf(stderr, "3: %s\n", (s != NULL) ? s : "<not
found>");
 return 0;
}

Results:

$ ./a.out      # Solaris 9
1: hij
2: <not found>
Segmentation Fault (core dumped)


$ ./a.out      # IRIX 6.5
1: hij
2: <not found>
Segmentation fault (core dumped)


$ ./a.out      # Linux with glibc 2.2.5
1: hij
2: <not found>
3: <not found>


$ ./a.out      # OpenBSD 3.2
1: hij
2: <not found>
3: <not found>

-- 
Life is like a fire.                            DI Michael Wildpaner
Flames which the passer-by forgets.                    Ph.D. Student
Ashes which the wind scatters.
A man lived.       -- Omar Khayyam