@Mediaservice.net Security Advisory #2020-07 (last updated on 2020-04-15) Title: Heap-based buffer overflow in Solaris whodo and w commands Application: Setuid root whodo and w binaries distributed with Solaris Platforms: Oracle Solaris 11.x (confirmed on 11.4 X86) Oracle Solaris 10 (confirmed on 10 1/13 X86) Other platforms are potentially affected (see below) Description: A difficult to exploit heap-based buffer overflow in setuid root whodo and w binaries distributed with Solaris allows local users to corrupt memory and potentially execute arbitrary code in order to escalate privileges Author: Marco Ivaldi Vendor Status: notified on 2019-08-23 CVE Name: CVE-2020-2771 CVSS Vector: CVSS:3.0/AV:L/AC:H/PR:L/UI:R/S:C/C:L/I:N/A:N (Base Score: 2.5) References: https://github.com/0xdea/advisories/blob/master/2020-07-solaris-whodo-w.txt https://www.oracle.com/security-alerts/cpuapr2020.html https://www.oracle.com/technetwork/server-storage/solaris11/ https://github.com/illumos/illumos-gate/blob/61aaa916808c601f9ee36d96c05ee9dac211d09e/usr/src/cmd/whodo/whodo.c https://github.com/illumos/illumos-gate/blob/61aaa916808c601f9ee36d96c05ee9dac211d09e/usr/src/cmd/w/w.c https://www.mediaservice.net/ https://0xdeadbeef.info/ 1. Abstract. A difficult to exploit heap-based buffer overflow in setuid root whodo and w binaries distributed with Solaris allows local users to corrupt memory and potentially execute arbitrary code in order to escalate privileges. 2. Example Attack Session. In order to reproduce this bug, the following commands can be used: raptor@stalker:~$ cat /etc/release Oracle Solaris 11.4 X86 Copyright (c) 1983, 2018, Oracle and/or its affiliates. All rights reserved. Assembled 16 August 2018 raptor@stalker:~$ uname -a SunOS stalker 5.11 11.4.0.15.0 i86pc i386 i86pc raptor@stalker:~$ id uid=100(raptor) gid=10(staff) raptor@stalker:~$ cp /usr/bin/sleep AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA raptor@stalker:~$ exec -a '- BBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBBB' ./AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA 256 & [switch to another shell] raptor@stalker:~$ whodo -l # or w 12:43pm up 5 day(s), 20 hr(s), 36 min(s) 5 user(s) User tty login@ idle JCPU PCPU what raptor vt/7 Tue 2pm 6days 1:49 1:49 /usr/lib/tracker-miner-apps Segmentation Fault 3. Discussion. A detailed analysis of the buffer overflow in whodo follows. The w binary is also affected by this bug, because the two programs share a large portion of their codebase. Therefore, similar considerations apply to w. The overflow happens as follows (the Illumos source code available on GitHub has been used as a reference for this analysis, even though it doesn't exactly match the code of the binaries shipped with commercial Solaris versions): * The psinfo structure info is populated by reading /proc//psinfo * The char array info.pr_fname[16] is copied into the char array up->p_comm[80+1] * As a side note, the call to strncpy() at lines 344-345 incorrectly uses the size of the source buffer instead of the size of the destination buffer, but in this case this programming mistake doesn't cause a problem, because the source buffer is always smaller than the destination buffer: (void) strncpy(up->p_comm, info.pr_fname, sizeof (info.pr_fname)); * The char array up->p_args[80+1] is then populated at line 418 based on the char array info.pr_psargs[80] as follows: (void) strcpy(up->p_args, info.pr_psargs); * If up->p_args begins with "?" or "- " (or, more correctly, with "-" followed by any byte <= 0x20), the following code branch at lines 423-425 is taken: (void) strcat(up->p_args, " ("); (void) strcat(up->p_args, up->p_comm); (void) strcat(up->p_args, ")"); * In detail, the following chars are appended to the string: " (" + up->p_comm [maximum size excluding NULL-terminator is 15] + ")" + NULL * Therefore, it is possible to overflow the up->p_args buffer at most as follows: * Buffer is 81 bytes: "- " + "B"x77 + " (" * Overflow is 17 bytes: "A"x15 + ")" + NULL The uproc structure is declared at lines 106-119: struct uproc { pid_t p_upid; /* user process id */ char p_state; /* numeric value of process state */ dev_t p_ttyd; /* controlling tty of process */ time_t p_time; /* ticks of user & system time */ time_t p_ctime; /* ticks of child user & system time */ int p_igintr; /* 1=ignores SIGQUIT and SIGINT */ char p_comm[PRARGSZ+1]; /* command */ char p_args[PRARGSZ+1]; /* command line arguments */ struct uproc *p_child, /* first child pointer */ *p_sibling, /* sibling pointer */ *p_pgrplink, /* pgrp link */ *p_link; /* hash table chain pointer */ }; A 17 bytes overflow past the p_args buffer is not large enough to reach critical control structures and directly take control of the program flow. However, we are able to overflow into the p_child and p_sibling members of the uproc structure up, assuming 64-bit addressing. With 32-bit addressing we should be able to corrupt additional pointers, i.e. p_pgrplink and p_link. A skilled attacker might be able to leverage the corruption of these pointers to obtain arbitrary code execution. However, he or she would face a number of additional challenges: * The target program uses privilege bracketing with the PRIV_PROC_OWNER privilege. This privilege allows a process to send signals to other processes, inspect, and potentially modify (with some additional restrictions) the process state in other processes, regardless of ownership. Therefore, it's theoretically possible to write a shellcode that activates the privilege and dumps the memory of a privileged process (e.g. "passwd") via /proc//mem, without ever executing an actual shell. However, this must be done before privileges are relinquinshed at line 455. This leaves only a limited amount of code paths to leverage our corrupted structure (namely, the main loop through /proc starting at line 315 and ending at line 452). * The char array info.pr_psargs[80] is cleaned up by the clnarglist() function at line 417: non-printable ASCII chars (c < 0x20 and c > 0x7e) get replaced with a "?" and must therefore be considered badchars. Luckily this restriction does not apply to the part of the buffer that causes the actual overflow, but only bytes that are valid in file names can be used in our malicious buffer. * The ")" + NULL chars at the end of the evil buffer might cause unforeseen problems during exploitation. * Additional security measures such as Address Space Layout Randomization (ASLR) might get in the way of reliable exploitation. Based on this analysis, our conclusion is that this bug not exploitable on Solaris 11.x and 10 in order to escalate privileges. That said, as a rule of thumb all memory corruption issues have the potential to become serious security vulnerabilities until otherwise proven. For instance, it might very well be possible to exploit this bug on systems that don't implement privilege bracketing, such as Solaris 9 and earlier. Therefore, we recommend to treat this bug as a potential security vulnerability and to fix it as such. 4. Affected Platforms. This bug was confirmed on the following platforms: * Oracle Solaris 11.x (confirmed on 11.4 X86) * Oracle Solaris 10 (confirmed on 10 1/13 X86) Other Oracle Solaris versions (including those that run on the SPARC architecture) and Illumos distributions are also likely affected. 5. Fix. Oracle has assigned the tracking# S1199548 and has released a fix for all affected and supported versions of Solaris in the Critical Patch Update (CPU) of April 2020. As a temporary workaround, it is possible to remove the setuid bit from whodo and w executables as follows (note that this might prevent them from working properly): bash-3.2# chmod -s /usr/sbin/whodo /usr/bin/w Copyright (c) 2020 Marco Ivaldi and @Mediaservice.net. All rights reserved.