journal: refresh cached credentials of stdout streams

journald assumes that getsockopt(SO_PEERCRED) correctly identifies the
process on the remote end of the socket. However, this is incorrect
according to man 7 socket:

    The returned  credentials  are  those that were in effect at the
    time of the call to connect(2) or socketpair(2).

This becomes a problem when a new process inherits the stdout stream
from a parent. First, log messages from the child process will
be attributed to the parent. Second, the struct ucred used by journald
becomes invalid as soon as the parent exits. Further sendmsg calls then
fail with ENOENT. Logs for the child process then vanish from the journal.

Fix this by using recvmsg on the stdout stream, and refreshing the cached
struct ucred if SCM_CREDENTIALS indicate a new process.

Fixes #13708

hwdb.d/60-keyboard.hwdb

@@ -75,7 +75,13 @@
 #   XKB_FIXED_VARIANT=""
 # Examples of such devices: the Yubikey or other key-code generating
 # devices.
-#
+
+# A device where the scan code to key code mapping is insufficient and
+# requires a special key code to symbol configuration may specify that with:
+#   XKB_FIXED_MODEL="xkbmodel"
+# Examples of such devices: Chromebooks where the top row is used for both
+# media and F1-F10 keys.
+
 # To update this file, create a new file
 #   /etc/udev/hwdb.d/70-keyboard.hwdb
 # and add your rules there. To load the new rules execute (as root):
@@ -569,21 +575,6 @@ evdev:atkbd:dmi:bvn*:bvr*:bd*:svnHewlett-Packard*:pnHDX9494NR:pvr*
  KEYBOARD_KEY_d8=!f23                                   # touchpad off
  KEYBOARD_KEY_d9=!f22                                   # touchpad on
 
-# Chromebook 14
-# Top row keys (between ESC and power button)
-evdev:atkbd:dmi:bvn*:bvr*:bd*:svnHewlett-Packard*:pnFalco:pvr*
- KEYBOARD_KEY_3b=back
- KEYBOARD_KEY_3c=forward
- KEYBOARD_KEY_3d=refresh
- KEYBOARD_KEY_3f=switchvideomode
- KEYBOARD_KEY_40=brightnessdown
- KEYBOARD_KEY_41=brightnessup
- KEYBOARD_KEY_42=mute
- KEYBOARD_KEY_43=volumedown
- KEYBOARD_KEY_44=volumeup
- KEYBOARD_KEY_db=search # Same position as caps lock key on most keyboards
- # KEYBOARD_KEY_3e=fullscreen, no defined key sym
-
 # HP EliteBook 725 G2
 evdev:atkbd:dmi:bvn*:bvr*:bd*:svnHewlett-Packard*:pnHPLicrice:pvr*
 # HP ProBook 440 G2
@@ -1733,3 +1724,13 @@ evdev:atkbd:dmi:bvn*:bvr*:bd*:svnLENOVO:pn*:pvrThinkPadT560s
 # Lenovo ThinkPad X1 Carbon 3rd Gen
 evdev:atkbd:dmi:bvn*:bvr*:bd*:svnLENOVO:pn*:pvrThinkPadX1Carbon3rd
  KEYBOARD_LED_CAPSLOCK=0
+
+######################### FIXED MODEL DEVICES #############################
+# This section lists devices which require special handling in their key
+# code to keysym mapping by setting the xkb model.
+# The model must be an xkb compatible model (defined with XKB_FIXED_MODEL).
+
+# Chromebooks
+evdev:atkbd:dmi:bvn*:bvr*:bd*:svnHewlett-Packard*:pnFalco:pvr*
+evdev:atkbd:dmi:bvn*:bvr*:bd*:svnAcer*:pnPeppy:pvr*
+ XKB_FIXED_MODEL="chromebook"

hwdb.d/parse_hwdb.py

@@ -123,6 +123,7 @@ def property_grammar():
              ('ID_INPUT_TOUCHPAD_INTEGRATION', Or(('internal', 'external'))),
              ('XKB_FIXED_LAYOUT', STRING),
              ('XKB_FIXED_VARIANT', STRING),
+             ('XKB_FIXED_MODEL', STRING),
              ('KEYBOARD_LED_NUMLOCK', Literal('0')),
              ('KEYBOARD_LED_CAPSLOCK', Literal('0')),
              ('ACCEL_MOUNT_MATRIX', mount_matrix),

src/boot/bootctl.c

@@ -27,6 +27,7 @@
 #include "fs-util.h"
 #include "locale-util.h"
 #include "main-func.h"
+#include "mkdir.h"
 #include "pager.h"
 #include "parse-util.h"
 #include "pretty-print.h"
@@ -1364,6 +1365,13 @@ static int install_random_seed(const char *esp) {
         if (r < 0)
                 return log_error_errno(r, "Failed to acquire random seed: %m");
 
+        /* Normally create_subdirs() should already have created everything we need, but in case "bootctl
+         * random-seed" is called we want to just create the minimum we need for it, and not the full
+         * list. */
+        r = mkdir_parents(path, 0755);
+        if (r < 0)
+                return log_error_errno(r, "Failed to create parent directory for %s: %m", path);
+
         r = tempfn_random(path, "bootctl", &tmp);
         if (r < 0)
                 return log_oom();

src/journal/journald-stream.c

@@ -487,11 +487,22 @@ static int stdout_stream_scan(StdoutStream *s, bool force_flush) {
 }
 
 static int stdout_stream_process(sd_event_source *es, int fd, uint32_t revents, void *userdata) {
+        uint8_t buf[CMSG_SPACE(sizeof(struct ucred))];
         StdoutStream *s = userdata;
+        struct ucred *ucred = NULL;
+        struct cmsghdr *cmsg;
+        struct iovec iovec;
         size_t limit;
         ssize_t l;
         int r;
 
+        struct msghdr msghdr = {
+                .msg_iov = &iovec,
+                .msg_iovlen = 1,
+                .msg_control = buf,
+                .msg_controllen = sizeof(buf),
+        };
+
         assert(s);
 
         if ((revents|EPOLLIN|EPOLLHUP) != (EPOLLIN|EPOLLHUP)) {
@@ -511,20 +522,50 @@ static int stdout_stream_process(sd_event_source *es, int fd, uint32_t revents,
          * always leave room for a terminating NUL we might need to add. */
         limit = MIN(s->allocated - 1, s->server->line_max);
 
-        l = read(s->fd, s->buffer + s->length, limit - s->length);
+        iovec = IOVEC_MAKE(s->buffer + s->length, limit - s->length);
+
+        l = recvmsg(s->fd, &msghdr, MSG_DONTWAIT|MSG_CMSG_CLOEXEC);
         if (l < 0) {
-                if (errno == EAGAIN)
+                if (IN_SET(errno, EINTR, EAGAIN))
                         return 0;
 
                 log_warning_errno(errno, "Failed to read from stream: %m");
                 goto terminate;
         }
+        cmsg_close_all(&msghdr);
 
         if (l == 0) {
                 stdout_stream_scan(s, true);
                 goto terminate;
         }
 
+        CMSG_FOREACH(cmsg, &msghdr)
+                if (cmsg->cmsg_level == SOL_SOCKET &&
+                    cmsg->cmsg_type == SCM_CREDENTIALS &&
+                    cmsg->cmsg_len == CMSG_LEN(sizeof(struct ucred))) {
+                        ucred = (struct ucred *)CMSG_DATA(cmsg);
+                        break;
+                }
+
+        /* Invalidate the context if the pid of the sender changed.
+         * This happens when a forked process inherits stdout / stderr
+         * from a parent. In this case getpeercred returns the ucred
+         * of the parent, which can be invalid if the parent has exited
+         * in the meantime.
+         */
+        if (ucred && ucred->pid != s->ucred.pid) {
+                /* force out any previously half-written lines from a
+                 * different process, before we switch to the new ucred
+                 * structure for everything we just added */
+                r = stdout_stream_scan(s, true);
+                if (r < 0)
+                        goto terminate;
+
+                s->ucred = *ucred;
+                client_context_release(s->server, s->context);
+                s->context = NULL;
+        }
+
         s->length += l;
         r = stdout_stream_scan(s, false);
         if (r < 0)
@@ -562,6 +603,10 @@ int stdout_stream_install(Server *s, int fd, StdoutStream **ret) {
         if (r < 0)
                 return log_error_errno(r, "Failed to determine peer credentials: %m");
 
+        r = setsockopt_int(fd, SOL_SOCKET, SO_PASSCRED, true);
+        if (r < 0)
+                return log_error_errno(r, "SO_PASSCRED failed: %m");
+
         if (mac_selinux_use()) {
                 r = getpeersec(fd, &stream->label);
                 if (r < 0 && r != -EOPNOTSUPP)

test/TEST-04-JOURNAL/test-journal.sh

@@ -74,6 +74,22 @@ cmp /expected /output
 { journalctl -ball -b  -m 2>&1 || :; } | head -1 > /output
 cmp /expected /output
 
+# https://github.com/systemd/systemd/issues/13708
+ID=$(systemd-id128 new)
+systemd-cat -t "$ID" bash -c 'echo parent; (echo child) & wait' &
+PID=$!
+wait %%
+journalctl --sync
+# We can drop this grep when https://github.com/systemd/systemd/issues/13937
+# has a fix.
+journalctl -b -o export -t "$ID" --output-fields=_PID | grep '^_PID=' >/output
+[[ `grep -c . /output` -eq 2 ]]
+grep -q "^_PID=$PID" /output
+grep -vq "^_PID=$PID" /output
+
+# Add new tests before here, the journald restarts below
+# may make tests flappy.
+
 # Don't lose streams on restart
 systemctl start forever-print-hola
 sleep 3