[SSSD] [PATCH] Do not change attr pointer with realloc

Jakub Hrozek jhrozek at redhat.com
Thu Apr 18 11:52:26 UTC 2013


On Thu, Apr 18, 2013 at 10:50:55AM +0200, Pavel Březina wrote:
> On 04/18/2013 10:48 AM, Pavel Březina wrote:
> >On 04/12/2013 12:25 PM, Jakub Hrozek wrote:
> >>https://fedorahosted.org/sssd/ticket/1799
> >>
> >>One peculiarity of the sysdb_attrs_get_el interface is that if the
> >>attribute does not exist, then the attrs array is reallocated and the
> >>element is created. But in case other pointers are already pointing
> >>into the array, the realloc might invalidate them.
> >>
> >>One such case was in the sdap_process_ghost_members function where if
> >>the group had no members, the "gh" pointer requested earlier might have
> >>been invalidated by the realloc in order to create the member element
> >>
> >>I'm actually pondering the idea of renaming the interface to make it
> >>clear that if you request a nonexistent attr, the pointer already
> >>pointing there might become invalid.
> >
> >Nack.
> >
> >Please, add a comment explaining that you just create a dummy element
> >that have zero values. Or set memberel->value and num_values explicitly.
> >
> >Also, you don't really have to set memberel->name - it is not used
> >anywhere and it doesn't add any semantic value. However, if you want to
> >create a complete element, I'm ok with it.
> >

That was the intent, yes, but the function is so small we don't have to care,
really. I added a comment and explicitly set memberel->value and num_values.

> >Thanks.
> 
> Also the commit title is a little misleading. I thought that you
> changed one of the sysdb functions.

Fixed the title.
-------------- next part --------------
>From 86e07effe8c2f061882132f230b6ac3b105d407c Mon Sep 17 00:00:00 2001
From: Jakub Hrozek <jhrozek at redhat.com>
Date: Fri, 12 Apr 2013 12:01:01 +0200
Subject: [PATCH] LDAP: do not invalidate pointer with realloc while processing
 ghost users

https://fedorahosted.org/sssd/ticket/1799

One peculiarity of the sysdb_attrs_get_el interface is that if the
attribute does not exist, then the attrs array is reallocated and the
element is created. But in case other pointers are already pointing
into the array, the realloc might invalidate them.

Such case was in the sdap_process_ghost_members function where if
the group had no members, the "gh" pointer requested earlier might have
been invalidated by the realloc in order to create the member element.
---
 src/providers/ldap/sdap_async_groups.c | 16 +++++++++++++---
 1 file changed, 13 insertions(+), 3 deletions(-)

diff --git a/src/providers/ldap/sdap_async_groups.c b/src/providers/ldap/sdap_async_groups.c
index 9d4a84871793b3af9b6b57a06c358f91fcaf684d..bb88d6c4ebc12978c83ec4f0dd78af1f2c77ca28 100644
--- a/src/providers/ldap/sdap_async_groups.c
+++ b/src/providers/ldap/sdap_async_groups.c
@@ -323,10 +323,20 @@ sdap_process_ghost_members(struct sysdb_attrs *attrs,
         return ret;
     }
 
-    ret = sysdb_attrs_get_el(attrs,
+    ret = sysdb_attrs_get_el_ext(attrs,
                              opts->group_map[SDAP_AT_GROUP_MEMBER].sys_name,
-                             &memberel);
-    if (ret != EOK) {
+                             false, &memberel);
+    if (ret == ENOENT) {
+        /* Create a dummy element with no values in order for the loop to just
+         * fall through and make sure the attrs array is not reallocated.
+         */
+        memberel = talloc(attrs, struct ldb_message_element);
+        if (memberel == NULL) {
+            return ENOMEM;
+        }
+        memberel->num_values = 0;
+        memberel->values = NULL;
+    } else if (ret != EOK) {
         DEBUG(SSSDBG_MINOR_FAILURE,
                 ("Error reading members: [%s]\n", strerror(ret)));
         return ret;
-- 
1.8.1.4



More information about the sssd-devel mailing list