This is a necessary follow up to the first fix I proposed and we merged
in 2669b8b0c7 ("binder: prevent UAF for binderfs devices"). I have been
overly optimistic that the simple fix I proposed would work. But alas,
ihold() + iput() won't work since the inodes won't survive the
destruction of the superblock.
So all we get with my prior fix is a different race with a tinier
race-window but it doesn't solve the issue. Fwiw, the problem lies with
generic_shutdown_super(). It even has this cozy Al-style comment:
if (!list_empty(&sb->s_inodes)) {
printk("VFS: Busy inodes after unmount of %s. "
"Self-destruct in 5 seconds. Have a nice day...\n",
sb->s_id);
}
On binder_release(), binder_defer_work(proc, BINDER_DEFERRED_RELEASE) is
called which punts the actual cleanup operation to a workqueue. At some
point, binder_deferred_func() will be called which will end up calling
binder_deferred_release() which will retrieve and cleanup the
binder_context attach to this struct binder_proc.
If we trace back where this binder_context is attached to binder_proc we
see that it is set in binder_open() and is taken from the struct
binder_device it is associated with. This obviously assumes that the
struct binder_device that context is attached to is _never_ freed. While
that might be true for devtmpfs binder devices it is most certainly
wrong for binderfs binder devices.
So, assume binder_open() is called on a binderfs binder devices. We now
stash away the struct binder_context associated with that struct
binder_devices:
proc->context = &binder_dev->context;
/* binderfs stashes devices in i_private */
if (is_binderfs_device(nodp)) {
binder_dev = nodp->i_private;
info = nodp->i_sb->s_fs_info;
binder_binderfs_dir_entry_proc = info->proc_log_dir;
} else {
.
.
.
proc->context = &binder_dev->context;
Now let's assume that the binderfs instance for that binder devices is
shutdown via umount() and/or the mount namespace associated with it goes
away. As long as there is still an fd open for that binderfs binder
device things are fine. But let's assume we now close the last fd for
that binderfs binder device. Now binder_release() is called and punts to
the workqueue. Assume that the workqueue has quite a bit of stuff to do
and doesn't get to cleaning up the struct binder_proc and the associated
struct binder_context with it for that binderfs binder device right
away. In the meantime, the VFS is killing the super block and is
ultimately calling sb->evict_inode() which means it will call
binderfs_evict_inode() which does:
static void binderfs_evict_inode(struct inode *inode)
{
struct binder_device *device = inode->i_private;
struct binderfs_info *info = BINDERFS_I(inode);
clear_inode(inode);
if (!S_ISCHR(inode->i_mode) || !device)
return;
mutex_lock(&binderfs_minors_mutex);
--info->device_count;
ida_free(&binderfs_minors, device->miscdev.minor);
mutex_unlock(&binderfs_minors_mutex);
kfree(device->context.name);
kfree(device);
}
thereby freeing the struct binder_device including struct
binder_context.
Now the workqueue finally has time to get around to cleaning up struct
binder_proc and is now trying to access the associate struct
binder_context. Since it's already freed it will OOPs.
Fix this by introducing a refounct on binder devices.
This is an alternative fix to 51d8a7eca6 ("binder: prevent UAF read in
print_binder_transaction_log_entry()").
Fixes: 3ad20fe393 ("binder: implement binderfs")
Fixes: 2669b8b0c7 ("binder: prevent UAF for binderfs devices")
Fixes: 03e2e07e38 ("binder: Make transaction_log available in binderfs")
Related : 51d8a7eca6 ("binder: prevent UAF read in print_binder_transaction_log_entry()")
Cc: stable@vger.kernel.org
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Acked-by: Todd Kjos <tkjos@google.com>
Link: https://lore.kernel.org/r/20200303164340.670054-1-christian.brauner@ubuntu.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently /sys/kernel/debug/binder/proc contains
the debug data for every binder_proc instance.
This patch makes this information also available
in a binderfs instance mounted with a mount option
"stats=global" in addition to debugfs. The patch does
not affect the presence of the file in debugfs.
If a binderfs instance is mounted at path /dev/binderfs,
this file would be present at /dev/binderfs/binder_logs/proc.
This change provides an alternate way to access this file when debugfs
is not mounted.
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Hridya Valsaraju <hridya@google.com>
Link: https://lore.kernel.org/r/20190903161655.107408-5-hridya@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, the binder transaction log files 'transaction_log'
and 'failed_transaction_log' live in debugfs at the following locations:
/sys/kernel/debug/binder/failed_transaction_log
/sys/kernel/debug/binder/transaction_log
This patch makes these files also available in a binderfs instance
mounted with the mount option "stats=global".
It does not affect the presence of these files in debugfs.
If a binderfs instance is mounted at path /dev/binderfs, the location of
these files will be as follows:
/dev/binderfs/binder_logs/failed_transaction_log
/dev/binderfs/binder_logs/transaction_log
This change provides an alternate option to access these files when
debugfs is not mounted.
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Hridya Valsaraju <hridya@google.com>
Link: https://lore.kernel.org/r/20190903161655.107408-4-hridya@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The following binder stat files currently live in debugfs.
/sys/kernel/debug/binder/state
/sys/kernel/debug/binder/stats
/sys/kernel/debug/binder/transactions
This patch makes these files available in a binderfs instance
mounted with the mount option 'stats=global'. For example, if a binderfs
instance is mounted at path /dev/binderfs, the above files will be
available at the following locations:
/dev/binderfs/binder_logs/state
/dev/binderfs/binder_logs/stats
/dev/binderfs/binder_logs/transactions
This provides a way to access them even when debugfs is not mounted.
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Hridya Valsaraju <hridya@google.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Link: https://lore.kernel.org/r/20190903161655.107408-3-hridya@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, all binder state and statistics live in debugfs.
We need this information even when debugfs is not mounted.
This patch adds the mount option 'stats' to enable a binderfs
instance to have binder debug information present in the same.
'stats=global' will enable the global binder statistics. In
the future, 'stats=local' will enable binder statistics local
to the binderfs instance. The two modes 'global' and 'local'
will be mutually exclusive. 'stats=global' option is only available
for a binderfs instance mounted in the initial user namespace.
An attempt to use the option to mount a binderfs instance in
another user namespace will return an EPERM error.
Signed-off-by: Hridya Valsaraju <hridya@google.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Link: https://lore.kernel.org/r/20190903161655.107408-2-hridya@google.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Currently, since each binderfs instance needs its own
private binder devices, every time a binderfs instance is
mounted, all the default binder devices need to be created
via the BINDER_CTL_ADD IOCTL. This patch aims to
add a solution to automatically create the default binder
devices for each binderfs instance that gets mounted.
To achieve this goal, when CONFIG_ANDROID_BINDERFS is set,
the default binder devices specified by CONFIG_ANDROID_BINDER_DEVICES
are created in each binderfs instance instead of global devices
being created by the binder driver.
Co-developed-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Hridya Valsaraju <hridya@google.com>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Link: https://lore.kernel.org/r/20190808222727.132744-2-hridya@google.com
Link: https://lore.kernel.org/r/20190904110704.8606-2-christian.brauner@ubuntu.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Length of a binderfs device name cannot exceed BINDERFS_MAX_NAME.
This patch adds a check in binderfs_init() to ensure the same
for the default binder devices that will be created in every
binderfs instance.
Co-developed-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Hridya Valsaraju <hridya@google.com>
Reviewed-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Link: https://lore.kernel.org/r/20190808222727.132744-3-hridya@google.com
Link: https://lore.kernel.org/r/20190904110704.8606-3-christian.brauner@ubuntu.com
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
binderfs should not have a separate device_initcall(). When a kernel is
compiled with CONFIG_ANDROID_BINDERFS register the filesystem alongside
CONFIG_ANDROID_IPC. This use-case is especially sensible when users specify
CONFIG_ANDROID_IPC=y, CONFIG_ANDROID_BINDERFS=y and
ANDROID_BINDER_DEVICES="".
When CONFIG_ANDROID_BINDERFS=n then this always succeeds so there's no
regression potential for legacy workloads.
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
We currently adhere to the reserved devices limit when creating new
binderfs devices in binderfs instances not located in the inital ipc
namespace. But it is still possible to rob the host instances of their 4
reserved devices by creating the maximum allowed number of devices in a
single binderfs instance located in a non-initial ipc namespace and then
mounting 4 separate binderfs instances in non-initial ipc namespaces. That
happens because the limit is currently not respected for the creation of
the initial binder-control device node. Block this nonsense by performing
the same check in binderfs_binder_ctl_create() that we perform in
binderfs_binder_device_create().
Fixes: 36bdf3cae0 ("binderfs: reserve devices for initial mount")
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The binderfs_binder_ctl_create() call is a no-op on subsequent calls and
the first call is done before we unlock the suberblock. Hence, there is no
need to take inode_lock() in there. Let's remove it.
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Al pointed out that first calling kill_litter_super() before cleaning up
info is more correct since destroying info doesn't depend on the state of
the dentries and inodes. That the opposite remains true is not guaranteed.
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
- switch from d_alloc_name() + d_lookup() to lookup_one_len():
Instead of using d_alloc_name() and then doing a d_lookup() with the
allocated dentry to find whether a device with the name we're trying to
create already exists switch to using lookup_one_len(). The latter will
either return the existing dentry or a new one.
- switch from kmalloc() + strscpy() to kmemdup():
Use a more idiomatic way to copy the name for the new dentry that
userspace gave us.
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Al pointed out that on binderfs_fill_super() error
deactivate_locked_super() will call binderfs_kill_super() so all of the
freeing and putting we currently do in binderfs_fill_super() is unnecessary
and buggy. Let's simply return errors and let binderfs_fill_super() take
care of cleaning up on error.
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
- make binderfs control dentry immutable:
We don't allow to unlink it since it is crucial for binderfs to be
useable but if we allow to rename it we make the unlink trivial to
bypass. So prevent renaming too and simply treat the control dentry as
immutable.
- add is_binderfs_control_device() helper:
Take the opportunity and turn the check for the control dentry into a
separate helper is_binderfs_control_device() since it's now used in two
places.
- simplify binderfs_rename():
Instead of hand-rolling our custom version of simple_rename() just dumb
the whole function down to first check whether we're trying to rename the
control dentry. If we do EPERM the caller and if not call simple_rename().
Suggested-by: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The comment stems from an early version of that patchset and is just
confusing now.
Cc: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Fix to return a negative error code -ENOMEM from the new_inode() and
d_make_root() error handling cases instead of 0, as done elsewhere in
this function.
Fixes: 849d540ddf ("binderfs: implement "max" mount option")
Signed-off-by: Wei Yongjun <weiyongjun1@huawei.com>
Reviewed-by: Christian Brauner <christian@brauner.io>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
kbuild reported a build faile in [1]. This is triggered when CONFIG_IPC_NS
is not set. So let's make the use of init_ipc_ns conditional on
CONFIG_IPC_NS being set.
[1]: https://lists.01.org/pipermail/kbuild-all/2019-January/056903.html
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The binderfs instance in the initial ipc namespace will always have a
reserve of 4 binder devices unless explicitly capped by specifying a lower
value via the "max" mount option.
This ensures when binder devices are removed (on accident or on purpose)
they can always be recreated without risking that all minor numbers have
already been used up.
Cc: Todd Kjos <tkjos@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
It doesn't make sense to call the header binder_ctl.h when its sole
existence is tied to binderfs. So give it a sensible name. Users will far
more easily remember binderfs.h than binder_ctl.h.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Since binderfs can be mounted by userns root in non-initial user namespaces
some precautions are in order. First, a way to set a maximum on the number
of binder devices that can be allocated per binderfs instance and second, a
way to reserve a reasonable chunk of binderfs devices for the initial ipc
namespace.
A first approach as seen in [1] used sysctls similiar to devpts but was
shown to be flawed (cf. [2] and [3]) since some aspects were unneeded. This
is an alternative approach which avoids sysctls completely and instead
switches to a single mount option.
Starting with this commit binderfs instances can be mounted with a limit on
the number of binder devices that can be allocated. The max=<count> mount
option serves as a per-instance limit. If max=<count> is set then only
<count> number of binder devices can be allocated in this binderfs
instance.
This allows to safely bind-mount binderfs instances into unprivileged user
namespaces since userns root in a non-initial user namespace cannot change
the mount option as long as it does not own the mount namespace the
binderfs mount was created in and hence cannot drain the host of minor
device numbers
[1]: https://lore.kernel.org/lkml/20181221133909.18794-1-christian@brauner.io/
[2]; https://lore.kernel.org/lkml/20181221163316.GA8517@kroah.com/
[3]: https://lore.kernel.org/lkml/CAHRSSEx+gDVW4fKKK8oZNAir9G5icJLyodO8hykv3O0O1jt2FQ@mail.gmail.com/
[4]: https://lore.kernel.org/lkml/20181221192044.5yvfnuri7gdop4rs@brauner.io/
Cc: Todd Kjos <tkjos@google.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
When currently mounting binderfs in the same ipc namespace twice:
mount -t binder binder /A
mount -t binder binder /B
then the binderfs instances mounted on /A and /B will be the same, i.e.
they will have the same superblock. This was the first approach that seemed
reasonable. However, this leads to some problems and inconsistencies:
/* private binderfs instance in same ipc namespace */
There is no way for a user to request a private binderfs instance in the
same ipc namespace.
This request has been made in a private mail to me by two independent
people.
/* bind-mounts */
If users want the same binderfs instance to appear in multiple places they
can use bind mounts. So there is no value in having a request for a new
binderfs mount giving them the same instance.
/* unexpected behavior */
It's surprising that request to mount binderfs is not giving the user a new
instance like tmpfs, devpts, ramfs, and others do.
/* past mistakes */
Other pseudo-filesystems once made the same mistakes of giving back the
same superblock when actually requesting a new mount (cf. devpts's
deprecated "newinstance" option).
We should not make the same mistake. Once we've committed to always giving
back the same superblock in the same IPC namespace with the next kernel
release we will not be able to make that change so better to do it now.
/* kdbusfs */
It was pointed out to me that kdbusfs - which is conceptually closely
related to binderfs - also allowed users to get a private kdbusfs instance
in the same IPC namespace by making each mount of kdbusfs a separate
instance. I think that makes a lot of sense.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The binderfs filesystem never needs to be mounted by the kernel itself.
This is conceptually wrong and should never have been done in the first
place.
Fixes: 3ad20fe393 ("binder: implement binderfs")
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
As discussed at Linux Plumbers Conference 2018 in Vancouver [1] this is the
implementation of binderfs.
/* Abstract */
binderfs is a backwards-compatible filesystem for Android's binder ipc
mechanism. Each ipc namespace will mount a new binderfs instance. Mounting
binderfs multiple times at different locations in the same ipc namespace
will not cause a new super block to be allocated and hence it will be the
same filesystem instance.
Each new binderfs mount will have its own set of binder devices only
visible in the ipc namespace it has been mounted in. All devices in a new
binderfs mount will follow the scheme binder%d and numbering will always
start at 0.
/* Backwards compatibility */
Devices requested in the Kconfig via CONFIG_ANDROID_BINDER_DEVICES for the
initial ipc namespace will work as before. They will be registered via
misc_register() and appear in the devtmpfs mount. Specifically, the
standard devices binder, hwbinder, and vndbinder will all appear in their
standard locations in /dev. Mounting or unmounting the binderfs mount in
the initial ipc namespace will have no effect on these devices, i.e. they
will neither show up in the binderfs mount nor will they disappear when the
binderfs mount is gone.
/* binder-control */
Each new binderfs instance comes with a binder-control device. No other
devices will be present at first. The binder-control device can be used to
dynamically allocate binder devices. All requests operate on the binderfs
mount the binder-control device resides in.
Assuming a new instance of binderfs has been mounted at /dev/binderfs
via mount -t binderfs binderfs /dev/binderfs. Then a request to create a
new binder device can be made as illustrated in [2].
Binderfs devices can simply be removed via unlink().
/* Implementation details */
- dynamic major number allocation:
When binderfs is registered as a new filesystem it will dynamically
allocate a new major number. The allocated major number will be returned
in struct binderfs_device when a new binder device is allocated.
- global minor number tracking:
Minor are tracked in a global idr struct that is capped at
BINDERFS_MAX_MINOR. The minor number tracker is protected by a global
mutex. This is the only point of contention between binderfs mounts.
- struct binderfs_info:
Each binderfs super block has its own struct binderfs_info that tracks
specific details about a binderfs instance:
- ipc namespace
- dentry of the binder-control device
- root uid and root gid of the user namespace the binderfs instance
was mounted in
- mountable by user namespace root:
binderfs can be mounted by user namespace root in a non-initial user
namespace. The devices will be owned by user namespace root.
- binderfs binder devices without misc infrastructure:
New binder devices associated with a binderfs mount do not use the
full misc_register() infrastructure.
The misc_register() infrastructure can only create new devices in the
host's devtmpfs mount. binderfs does however only make devices appear
under its own mountpoint and thus allocates new character device nodes
from the inode of the root dentry of the super block. This will have
the side-effect that binderfs specific device nodes do not appear in
sysfs. This behavior is similar to devpts allocated pts devices and
has no effect on the functionality of the ipc mechanism itself.
[1]: https://goo.gl/JL2tfX
[2]: program to allocate a new binderfs binder device:
#define _GNU_SOURCE
#include <errno.h>
#include <fcntl.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sys/ioctl.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <unistd.h>
#include <linux/android/binder_ctl.h>
int main(int argc, char *argv[])
{
int fd, ret, saved_errno;
size_t len;
struct binderfs_device device = { 0 };
if (argc < 2)
exit(EXIT_FAILURE);
len = strlen(argv[1]);
if (len > BINDERFS_MAX_NAME)
exit(EXIT_FAILURE);
memcpy(device.name, argv[1], len);
fd = open("/dev/binderfs/binder-control", O_RDONLY | O_CLOEXEC);
if (fd < 0) {
printf("%s - Failed to open binder-control device\n",
strerror(errno));
exit(EXIT_FAILURE);
}
ret = ioctl(fd, BINDER_CTL_ADD, &device);
saved_errno = errno;
close(fd);
errno = saved_errno;
if (ret < 0) {
printf("%s - Failed to allocate new binder device\n",
strerror(errno));
exit(EXIT_FAILURE);
}
printf("Allocated new binder device with major %d, minor %d, and "
"name %s\n", device.major, device.minor,
device.name);
exit(EXIT_SUCCESS);
}
Cc: Martijn Coenen <maco@android.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Acked-by: Todd Kjos <tkjos@google.com>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Christian Brauner