Commit b234ed6d authored by Rasmus Villemoes's avatar Rasmus Villemoes Committed by Linus Torvalds

init: move usermodehelper_enable() to populate_rootfs()

Currently, usermodehelper is enabled right before PID1 starts going
through the initcalls. However, any call of a usermodehelper from a
pure_, core_, postcore_, arch_, subsys_ or fs_ initcall is futile, as
there is no filesystem contents yet.

Up until commit e7cb072e ("init/initramfs.c: do unpacking
asynchronously"), such calls, whether via some request_module(), a
legacy uevent "/sbin/hotplug" notification or something else, would
just fail silently with (presumably) -ENOENT from
kernel_execve(). However, that commit introduced the
wait_for_initramfs() synchronization hook which must be called from
the usermodehelper exec path right before the kernel_execve, in order
that request_module() et al done from *after* rootfs_initcall()
time (i.e. device_ and late_ initcalls) would continue to find a
populated initramfs as they used to.

Any call of wait_for_initramfs() done before the unpacking has been
scheduled (i.e. before rootfs_initcall time) must just return
immediately [and let the caller find an empty file system] in order
not to deadlock the machine. I mistakenly thought, and my limited
testing confirmed, that there were no such calls, so I added a
pr_warn_once() in wait_for_initramfs(). It turns out that one can
indeed hit request_module() as well as kobject_uevent_env() during
those early init calls, leading to a user-visible warning in the
kernel log emitted consistently for certain configurations.

We could just remove the pr_warn_once(), but I think it's better to
postpone enabling the usermodehelper framework until there is at least
some chance of finding the executable. That is also a little more
efficient in that a lot of work done in umh.c will be elided. However,
it does change the error seen by those early callers from -ENOENT to
-EBUSY, so there is a risk of a regression if any caller care about
the exact error value.

Link: https://lkml.kernel.org/r/20210728134638.329060-1-linux@rasmusvillemoes.dk
Fixes: e7cb072e ("init/initramfs.c: do unpacking asynchronously")
Signed-off-by: default avatarRasmus Villemoes <linux@rasmusvillemoes.dk>
Reported-by: default avatarAlexander Egorenkov <egorenar@linux.ibm.com>
Reported-by: default avatarBruno Goncalves <bgoncalv@redhat.com>
Reported-by: default avatarHeiner Kallweit <hkallweit1@gmail.com>
Cc: Luis Chamberlain <mcgrof@kernel.org>
Signed-off-by: default avatarAndrew Morton <akpm@linux-foundation.org>
Signed-off-by: default avatarLinus Torvalds <torvalds@linux-foundation.org>
parent 1e1c1583
......@@ -15,6 +15,7 @@
#include <linux/mm.h>
#include <linux/namei.h>
#include <linux/init_syscalls.h>
#include <linux/umh.h>
static ssize_t __init xwrite(struct file *file, const char *p, size_t count,
loff_t *pos)
......@@ -727,6 +728,7 @@ static int __init populate_rootfs(void)
{
initramfs_cookie = async_schedule_domain(do_populate_rootfs, NULL,
&initramfs_domain);
usermodehelper_enable();
if (!initramfs_async)
wait_for_initramfs();
return 0;
......
......@@ -1392,7 +1392,6 @@ static void __init do_basic_setup(void)
driver_init();
init_irq_proc();
do_ctors();
usermodehelper_enable();
do_initcalls();
}
......
......@@ -10,6 +10,7 @@
#include <linux/kdev_t.h>
#include <linux/syscalls.h>
#include <linux/init_syscalls.h>
#include <linux/umh.h>
/*
* Create a simple rootfs that is similar to the default initramfs
......@@ -18,6 +19,7 @@ static int __init default_rootfs(void)
{
int err;
usermodehelper_enable();
err = init_mkdir("/dev", 0755);
if (err < 0)
goto out;
......
Markdown is supported
0%
or
You are about to add 0 people to the discussion. Proceed with caution.
Finish editing this message first!
Please register or to comment