search

Icinga / icinga-core

Icinga 1.x, the old core (EOL 31.12.2018)
GNU General Public License v2.0
45 stars 27 forks source link

[dev.icinga.com #5434] CVE-2014-1878: Possible segfault in cmd.cgi #1418

Closed icinga-migration closed 10 years ago

icinga-migration commented 10 years ago

This issue has been migrated from Redmine: https://dev.icinga.com/issues/5434

Created by mfrosch on 2014-01-10 11:41:01 +00:00

Assignee: mfriedrich Status: Resolved (closed on 2014-02-11 15:33:40 +00:00) Target Version: 1.10.3 Last Update: 2014-12-08 09:21:46 +00:00 (in Redmine)

Icinga Version: 1.10.2
OS Version: any

Here is a potential security issue I've got in private.

Props to the GitHub security team and Dirkjan Bussink <dirkjan.bussink@github.com>.

Please keep it private for now! We also try to share information mit Naemon Team...

So we first encountered this problem as a segfault of our Nagios setup. The problem was with the second file patched in the patch, lib/XXX.c. The problem was that due to a large message coming it, the process would segfault since the message didn’t fit in the buffer that was allocated.

This is due wrong usage of snprintf. In case snprintf doesn’t have enough space in the buffer, it will return the number of needed bytes, not the number of bytes written. This is a fairly classic issue that happens quite often. What it means in this case is that it will read memory outside the buffer, in our case resulting in a segfault.

I’ve audited the nagios core code for more cases and found two more. Especially the case found in cgi/cmd.c worries me since I think it might potentially be used to DoS a Nagios install if it’s publicly accessible. It might even be used to overwrite other memory because the return value from the first call to snprintf is used in the second one.

This patch should fix it.

diff --git i/cgi/cmd.c w/cgi/cmd.c
index 50504eb..426d944 100644
--- i/cgi/cmd.c
+++ w/cgi/cmd.c
@@ -1903,14 +1903,14 @@ static int cmd_submitf(int id, const char *fmt, ...) {
                return ERROR;

        len = snprintf(cmd, sizeof(cmd) - 1, "[%lu] %s;", time(NULL), command_name);
-       if(len < 0)
+       if(len < 0 || len >= sizeof(cmd))
                return ERROR;

        if(fmt) {
                va_start(ap, fmt);
                len2 = vsnprintf(&cmd[len], sizeof(cmd) - len - 1, fmt, ap);
                va_end(ap);
-               if(len2 < 0)
+               if(len2 < 0 || len2 >= sizeof(cmd) - len)
                        return ERROR;
                }

Changesets

2014-01-10 12:40:09 +00:00 by (unknown) 37fc7668f35945d8064a1151c6235b0df333300a

Aggressively check for possible buffer overflows in cmd.cgi

Refs #5434

2014-01-10 12:40:37 +00:00 by (unknown) cf0d20ab882cfa1e425c7c229676f2a284401460

Merge branch 'fix/cmd-5434' into support/1.10

Refs #5434

2014-01-10 12:41:35 +00:00 by (unknown) f1c4f364d70117ea5d72117bf4fe0cd1b5f8b580

Aggressively check for possible buffer overflows in cmd.cgi

Refs #5434

2014-01-10 12:42:12 +00:00 by (unknown) eedf4f7d88cdc50843572224eb38a2f5c78a2dc5

Aggressively check for possible buffer overflows in cmd.cgi

Refs #5434

2014-01-10 12:42:40 +00:00 by (unknown) 10aeffc786fb7e192d9264124802bfa8c880a1f1

Merge branch 'fix/cmd-5434' into next

Refs #5434

2014-01-23 15:15:33 +00:00 by (unknown) 144a0b7557ce208421c34113e376169315d8b440

Update Changelog.

Refs #4968
Refs #5434
Refs #4427
Refs #4825
Refs #5263
Refs #5545
icinga-migration commented 10 years ago

Updated by mfrosch on 2014-01-10 11:50:32 +00:00

icinga-migration commented 10 years ago

Updated by mfrosch on 2014-01-10 12:32:03 +00:00

This is the patch against current git master:

diff --git a/cgi/cmd.c b/cgi/cmd.c
index e8b3e41..2fa61bc 100644
--- a/cgi/cmd.c
+++ b/cgi/cmd.c
@@ -2691,14 +2691,14 @@ static int cmd_submitf(int id, const char *fmt, ...) {

        len = snprintf(cmd, sizeof(cmd) - 1, "[%lu] %s;", time(NULL), command);

-       if (len < 0)
+       if (len < 0 || len >= sizeof(cmd))
                return ERROR;

        if (fmt) {
                va_start(ap, fmt);
                len2 = vsnprintf(&cmd[len], sizeof(cmd) - len - 1, fmt, ap);
                va_end(ap);
-               if (len2 < 0)
+        if (len2 < 0 || len2 >= sizeof(cmd))
                        return ERROR;
        }
icinga-migration commented 10 years ago

Updated by mfriedrich on 2014-01-10 12:34:13 +00:00

icinga-migration commented 10 years ago

Updated by crfriend on 2014-01-10 23:42:24 +00:00

Do we believe that the problem is isolated to this one particular case, or do we need to dig in and look for others?

Is it an historical issue that may require back-ports?

icinga-migration commented 10 years ago

Updated by mfrosch on 2014-01-11 11:10:13 +00:00

crfriend wrote:

Is it an historical issue that may require back-ports?

At least it's not Icinga specific, so it would require a backport.

Though, the risk might be very small.

icinga-migration commented 10 years ago

Updated by crfriend on 2014-01-11 17:55:46 +00:00

lazyfrosch wrote:

At least it's not Icinga specific, so it would require a backport.

Though, the risk might be very small.

In looking at the documentation for the snprintf() call -- at least for Solaris -- it is stated that the default behavior for snprintf() when trying to copy more data than allowed by the second argument is to copy one octet less than that value and then to null-terminate the resulting string. So, I'm not sure where the original reporter's segfault comes into play unless somebody then tries to append to it (which would overflow the space). This behavior may be different in Linux.

Ultimately, we're paying the price for the entire UNIX-like "null-terminated string" concept. If we're to defend against string-related overflows we'll either be paying a huge price in CPU cycles with never-ending strlen() calls or we'll have to come up with another mechanism for storing strings. I recall a number of years ago faced with overflow problems finally abandoning the null-terminated string almost completely and instead used a structure which led off with an int that described the current length of the string in the area immediately following the int which made length-checking very inexpensive indeed for calculating whether a new allocation (and copying the "string-struct" thereto, and freeing the old) was required (along with changing the int to reflect the new length).

In any event, and in digging through the code found that the patch in question addresses the single instance where the return-value from snprintf() is used in the CGIs. Testing for an oversize return-value makes good sense, however, in all cases as it'll catch instances where data gets corrupted (by virtue of the truncation) by the program and probably should get flagged as an error.

icinga-migration commented 10 years ago

Updated by mfriedrich on 2014-02-11 10:03:53 +00:00

Will be released as 1.8.6, 1.9.5 and 1.10.3 while the latter includes more bug fixes, 1.9 and 1.8 support branches only get the important fixes.

Will ask for a CVE number once the release is tagged and available for users.

icinga-migration commented 10 years ago

Updated by mfriedrich on 2014-02-11 10:30:43 +00:00

icinga-migration commented 10 years ago

Updated by mfriedrich on 2014-02-11 15:33:40 +00:00

Assigned CVE number: CVE-2014-1878 Fixed in: 1.10.3, 1.9.5, 1.8.6

https://www.icinga.org/2014/02/11/bugfix-releases-1-10-3-1-9-5-1-8-6/

icinga-migration commented 9 years ago

Updated by mfriedrich on 2014-12-08 09:21:47 +00:00