Found by inspecting results of running this small program:
int main(int argc, const char **argv) {
for (int i = 1; i < argc; i++) {
FILE *f;
char line[1024], prev[1024], *r;
int lineno;
prev[0] = '\0';
lineno = 1;
f = fopen(argv[i], "r");
if (!f)
exit(1);
do {
r = fgets(line, sizeof(line), f);
if (!r)
break;
if (strcmp(line, prev) == 0)
printf("%s:%d: error: dup %s", argv[i], lineno, line);
lineno++;
strcpy(prev, line);
} while (!feof(f));
fclose(f);
}
}
By b1c097af8d (#10239), the receive buffer
size for uevents was set by SO_RCVBUF at first, and fallback to
use SO_RCVBUFFORCE. So, as SO_RCVBUF limits to the buffer size
net.core.rmem_max, which is usually much smaller than 128MB udevd requests,
uevents buffer size was not sufficient.
This fixes the ordering of the request: SO_RCVBUFFORCE first, and
fallback to SO_RCVBUF. Then, udevd's uevent buffer size can be set to
128MB.
This also revert 903893237a.
Fixes#11314 and #10754.
This reverts commit b26c904113.
I don't see anythign wrong, but Ubuntu autopkgtest CI started failing fairly
consistently since this was merged. Let's see if reverting fixes things.
Setting -fPIE globally can lead to miscompilations on certain
architectures.
This is caused by both -fPIE and -fPIC options being added to various
compilation commands. Only -fPIC is being recorded in the LTO options
section of the object. The gcc-8 LTO plugin merges -fPIC + -fPIE to
nothing. So, the compilations done by the plugin are not
position-independent and fail to link with -pie.
The simplest solution is to stop setting -fPIE globally and instead
using meson's b_pie=true option. This requires meson 0.49 or later.
Since we don't set this option in meson.build but leave it up to the
distro maintainer to set this option, do not bump the meson version
requirement.
Fixes: #10548
The message is changed from
Cannot create session: Already running in a session...
to
Not creating session: Already running in a session...
This is more neutral and avoids suggesting a problem.
"Will not create session: ..." was suggested, but it sounds like the action
would have yet to be performed. I think Using present continuous is better.
Fixes#10822 (for good now, I hope).
c4b69e990f effectively moved the initalization of socket.
Before that commit:
run → listen_fds → udev_ctrl_new → udev_ctrl_new_from_fd → socket()
After:
run → main_loop → manager_new → udev_ctrl_new_from_fd → socket()
The problem is that main_loop was called after daemonization. Move manager_new
out of main_loop and before daemonization.
Fixes#11314 (hopefully ;)).
v2: Yu Watanabe
sd_event is initialized in main_loop().
Existing use of E2BIG is replaced with ENOBUFS (entry too long), and E2BIG is
reused for the new error condition (too many fields).
This matches the change done for systemd-journald, hence forming the second
part of the fix for CVE-2018-16865
(https://bugzilla.redhat.com/show_bug.cgi?id=1653861).
Calling mhd_respond(), which ulimately calls MHD_queue_response() is
ineffective at point, becuase MHD_queue_response() immediately returns
MHD_NO signifying an error, because the connection is in state
MHD_CONNECTION_CONTINUE_SENT.
As Christian Grothoff kindly explained:
> You are likely calling MHD_queue_repsonse() too late: once you are
> receiving upload_data, HTTP forces you to process it all. At this time,
> MHD has already sent "100 continue" and cannot take it back (hence you
> get MHD_NO!).
>
> In your request handler, the first time when you are called for a
> connection (and when hence *upload_data_size == 0 and upload_data ==
> NULL) you must check the content-length header and react (with
> MHD_queue_response) based on this (to prevent MHD from automatically
> generating 100 continue).
If we ever encounter this kind of error, print a warning and immediately
abort the connection. (The alternative would be to keep reading the data,
but ignore it, and return an error after we get to the end of data.
That is possible, but of course puts additional load on both the
sender and reciever, and doesn't seem important enough just to return
a good error message.)
Note that sending of the error does not work (the connection is always aborted
when MHD_queue_response is used with MHD_RESPMEM_MUST_FREE, as in this case)
with libµhttpd 0.59, but works with 0.61:
https://src.fedoraproject.org/rpms/libmicrohttpd/pull-request/1
We immediately read the whole contents into memory, making thigs much more
expensive. Sealed fds should be used instead since they are more efficient
on our side.
We'd first parse all or most of the message, and only then consider if it
is not too large. Also, when encountering a single field over the limit,
we'd still process the preceding part of the message. Let's be stricter,
and check size limits early, and let's refuse the whole message if it fails
any of the size limits.
We allocate a iovec entry for each field, so with many short entries,
our memory usage and processing time can be large, even with a relatively
small message size. Let's refuse overly long entries.
CVE-2018-16865
https://bugzilla.redhat.com/show_bug.cgi?id=1653861
What from I can see, the problem is not from an alloca, despite what the CVE
description says, but from the attack multiplication that comes from creating
many very small iovecs: (void* + size_t) for each three bytes of input message.
If creation of the message failed, we'd write a bogus entry:
systemd-coredump[1400]: Cannot store coredump of 416 (systemd-journal): No space left on device
systemd-coredump[1400]: MESSAGE=Process 416 (systemd-journal) of user 0 dumped core.
systemd-coredump[1400]: Coredump diverted to
This affects systemd-journald and systemd-coredump.
Example entry:
$ journalctl -o export -n1 'MESSAGE=Something logged'
__CURSOR=s=976542d120c649f494471be317829ef9;i=34e;b=4871e4c474574ce4a462dfe3f1c37f06;m=c7d0c37dd2;t=57c4ac58f3b98;x=67598e942bd23dc0
__REALTIME_TIMESTAMP=1544035467475864
__MONOTONIC_TIMESTAMP=858200964562
_BOOT_ID=4871e4c474574ce4a462dfe3f1c37f06
PRIORITY=6
_UID=1000
_GID=1000
_CAP_EFFECTIVE=0
_SELINUX_CONTEXT=unconfined_u:unconfined_r:unconfined_t:s0-s0:c0.c1023
_AUDIT_SESSION=1
_AUDIT_LOGINUID=1000
_SYSTEMD_OWNER_UID=1000
_SYSTEMD_UNIT=user@1000.service
_SYSTEMD_SLICE=user-1000.slice
_SYSTEMD_USER_SLICE=-.slice
_SYSTEMD_INVOCATION_ID=1c4a469986d448719cb0f9141a10810e
_MACHINE_ID=08a5690a2eed47cf92ac0a5d2e3cf6b0
_HOSTNAME=krowka
_TRANSPORT=syslog
SYSLOG_FACILITY=17
SYSLOG_IDENTIFIER=syslog-caller
MESSAGE=Something logged
_COMM=poc
_EXE=/home/zbyszek/src/systemd-work3/poc
_SYSTEMD_CGROUP=/user.slice/user-1000.slice/user@1000.service/gnome-terminal-server.service
_SYSTEMD_USER_UNIT=gnome-terminal-server.service
SYSLOG_PID=4108
SYSLOG_TIMESTAMP=Dec 5 19:44:27
_PID=4108
_CMDLINE=./poc AAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAAA>
_SOURCE_REALTIME_TIMESTAMP=1544035467475848
$ journalctl -o export -n1 'MESSAGE=Something logged' --output-fields=_CMDLINE|wc
6 2053 2097410
2MB might be hard for some clients to use meaningfully, but OTOH, it is
important to log the full commandline sometimes. For example, when the program
is crashing, the exact argument list is useful.
This fixes a crash where we would read the commandline, whose length is under
control of the sending program, and then crash when trying to create a stack
allocation for it.
CVE-2018-16864
https://bugzilla.redhat.com/show_bug.cgi?id=1653855
The message actually doesn't get written to disk, because
journal_file_append_entry() returns -E2BIG.
systemd-coredump[9982]: MESSAGE=Process 771 (systemd-journal) of user 0 dumped core.
systemd-coredump[9982]: Coredump diverted to /var/lib/systemd/coredump/core...
log_dispatch() calls log_dispatch_internal() which calls write_to_journal()
which appends MESSAGE= on its own.
In all three cases, sd_event_loop() will return the exit code anyway.
If sd_event_loop() returns negative, failure is logged and results in an
immediate return. Otherwise, we don't care if sd_event_loop() returns 0
or positive, because the return value feeds into DEFINE_MAIN_FUNCTION(), which
doesn't make the distinction.
Empty line between setting the output parameter and return is removed. I like
to think about both steps as part of returning from the function, and there's
no need to separate them.
Similarly, if we need to unset a pointer after successfully passing ownership,
use TAKE_PTR and do it immediately after the ownership change, without an empty
line inbetween.
