core: Do not bind a mount unit to a device, if it was from mountinfo

If a mount unit is bound to a device, systemd tries to umount the
mount point, if it thinks the device has gone away.

Due to the uevent queue and inotify of /proc/self/mountinfo being two
different sources, systemd can never get the ordering reliably correct.

It can happen, that in the uevent queue ADD,REMOVE,ADD is queued
and an inotify of mountinfo (or libmount event) happend with the
device in question.

systemd cannot know, at which point of time the mount happend in the
ADD,REMOVE,ADD sequence.

The real ordering might have been ADD,REMOVE,ADD,mount
and systemd might think ADD,mount,REMOVE,ADD and would umount the
mountpoint.

A test script which triggered this behaviour is:
rm -f test-efi-disk.img
dd if=/dev/null of=test-efi-disk.img bs=1M seek=512 count=1
parted --script test-efi-disk.img \
  "mklabel gpt" \
  "mkpart ESP fat32 1MiB 511MiB" \
  "set 1 boot on"
LOOP=$(losetup --show -f -P test-efi-disk.img)
udevadm settle
mkfs.vfat -F32 ${LOOP}p1
mkdir -p mnt
mount ${LOOP}p1 mnt
... <dostuffwith mnt>

Without the "udevadm settle" systemd unmounted mnt while the script was
operating on mnt.

Of course the question is, why there was a REMOVE in the first place,
but this is not part of this patch.

src/core/mount.c

@@ -335,7 +335,7 @@ static int mount_add_device_links(Mount *m) {
         if (mount_is_auto(p) && UNIT(m)->manager->running_as == MANAGER_SYSTEM)
                 device_wants_mount = true;
 
-        r = unit_add_node_link(UNIT(m), p->what, device_wants_mount);
+        r = unit_add_node_link(UNIT(m), p->what, device_wants_mount, m->from_fragment ? UNIT_BINDS_TO : UNIT_REQUIRES);
         if (r < 0)
                 return r;
 

src/core/socket.c

@@ -289,7 +289,7 @@ static int socket_add_device_link(Socket *s) {
                 return 0;
 
         t = strjoina("/sys/subsystem/net/devices/", s->bind_to_device);
-        return unit_add_node_link(UNIT(s), t, false);
+        return unit_add_node_link(UNIT(s), t, false, UNIT_BINDS_TO);
 }
 
 static int socket_add_default_dependencies(Socket *s) {

src/core/swap.c

@@ -202,7 +202,7 @@ static int swap_add_device_links(Swap *s) {
                 return 0;
 
         if (is_device_path(s->what))
-                return unit_add_node_link(UNIT(s), s->what, UNIT(s)->manager->running_as == MANAGER_SYSTEM);
+                return unit_add_node_link(UNIT(s), s->what, UNIT(s)->manager->running_as == MANAGER_SYSTEM, UNIT_BINDS_TO);
         else
                 /* File based swap devices need to be ordered after
                  * systemd-remount-fs.service, since they might need a

src/core/unit.c

@@ -2840,7 +2840,7 @@ int unit_deserialize(Unit *u, FILE *f, FDSet *fds) {
         }
 }
 
-int unit_add_node_link(Unit *u, const char *what, bool wants) {
+int unit_add_node_link(Unit *u, const char *what, bool wants, UnitDependency dep) {
         Unit *device;
         _cleanup_free_ char *e = NULL;
         int r;
@@ -2867,7 +2867,9 @@ int unit_add_node_link(Unit *u, const char *what, bool wants) {
         if (r < 0)
                 return r;
 
-        r = unit_add_two_dependencies(u, UNIT_AFTER, u->manager->running_as == MANAGER_SYSTEM ? UNIT_BINDS_TO : UNIT_WANTS, device, true);
+        r = unit_add_two_dependencies(u, UNIT_AFTER,
+                                      u->manager->running_as == MANAGER_SYSTEM ? dep : UNIT_WANTS,
+                                      device, true);
         if (r < 0)
                 return r;
 

src/core/unit.h

@@ -528,7 +528,7 @@ int unit_serialize_item_escaped(Unit *u, FILE *f, const char *key, const char *v
 int unit_serialize_item_fd(Unit *u, FILE *f, FDSet *fds, const char *key, int fd);
 void unit_serialize_item_format(Unit *u, FILE *f, const char *key, const char *value, ...) _printf_(4,5);
 
-int unit_add_node_link(Unit *u, const char *what, bool wants);
+int unit_add_node_link(Unit *u, const char *what, bool wants, UnitDependency d);
 
 int unit_coldplug(Unit *u);