Pull user namespace updates from Eric Biederman:
"Long ago and far away when user namespaces where young it was realized
that allowing fresh mounts of proc and sysfs with only user namespace
permissions could violate the basic rule that only root gets to decide
if proc or sysfs should be mounted at all.
Some hacks were put in place to reduce the worst of the damage could
be done, and the common sense rule was adopted that fresh mounts of
proc and sysfs should allow no more than bind mounts of proc and
sysfs. Unfortunately that rule has not been fully enforced.
There are two kinds of gaps in that enforcement. Only filesystems
mounted on empty directories of proc and sysfs should be ignored but
the test for empty directories was insufficient. So in my tree
directories on proc, sysctl and sysfs that will always be empty are
created specially. Every other technique is imperfect as an ordinary
directory can have entries added even after a readdir returns and
shows that the directory is empty. Special creation of directories
for mount points makes the code in the kernel a smidge clearer about
it's purpose. I asked container developers from the various container
projects to help test this and no holes were found in the set of mount
points on proc and sysfs that are created specially.
This set of changes also starts enforcing the mount flags of fresh
mounts of proc and sysfs are consistent with the existing mount of
proc and sysfs. I expected this to be the boring part of the work but
unfortunately unprivileged userspace winds up mounting fresh copies of
proc and sysfs with noexec and nosuid clear when root set those flags
on the previous mount of proc and sysfs. So for now only the atime,
read-only and nodev attributes which userspace happens to keep
consistent are enforced. Dealing with the noexec and nosuid
attributes remains for another time.
This set of changes also addresses an issue with how open file
descriptors from /proc/<pid>/ns/* are displayed. Recently readlink of
/proc/<pid>/fd has been triggering a WARN_ON that has not been
meaningful since it was added (as all of the code in the kernel was
converted) and is not now actively wrong.
There is also a short list of issues that have not been fixed yet that
I will mention briefly.
It is possible to rename a directory from below to above a bind mount.
At which point any directory pointers below the renamed directory can
be walked up to the root directory of the filesystem. With user
namespaces enabled a bind mount of the bind mount can be created
allowing the user to pick a directory whose children they can rename
to outside of the bind mount. This is challenging to fix and doubly
so because all obvious solutions must touch code that is in the
performance part of pathname resolution.
As mentioned above there is also a question of how to ensure that
developers by accident or with purpose do not introduce exectuable
files on sysfs and proc and in doing so introduce security regressions
in the current userspace that will not be immediately obvious and as
such are likely to require breaking userspace in painful ways once
they are recognized"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
vfs: Remove incorrect debugging WARN in prepend_path
mnt: Update fs_fully_visible to test for permanently empty directories
sysfs: Create mountpoints with sysfs_create_mount_point
sysfs: Add support for permanently empty directories to serve as mount points.
kernfs: Add support for always empty directories.
proc: Allow creating permanently empty directories that serve as mount points
sysctl: Allow creating permanently empty directories that serve as mountpoints.
fs: Add helper functions for permanently empty directories.
vfs: Ignore unlocked mounts in fs_fully_visible
mnt: Modify fs_fully_visible to deal with locked ro nodev and atime
mnt: Refactor the logic for mounting sysfs and proc in a user namespace
This allows for better documentation in the code and
it allows for a simpler and fully correct version of
fs_fully_visible to be written.
The mount points converted and their filesystems are:
/sys/hypervisor/s390/ s390_hypfs
/sys/kernel/config/ configfs
/sys/kernel/debug/ debugfs
/sys/firmware/efi/efivars/ efivarfs
/sys/fs/fuse/connections/ fusectl
/sys/fs/pstore/ pstore
/sys/kernel/tracing/ tracefs
/sys/fs/cgroup/ cgroup
/sys/kernel/security/ securityfs
/sys/fs/selinux/ selinuxfs
/sys/fs/smackfs/ smackfs
Cc: stable@vger.kernel.org
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
On traditional hierarchies, if a task has write access to "tasks" or
"cgroup.procs" file of a cgroup and its euid agrees with the target,
it can move the target to the cgroup; however, consider the following
scenario. The owner of each cgroup is in the parentheses.
R (root) - 0 (root) - 00 (user1) - 000 (user1)
| \ 001 (user1)
\ 1 (root) - 10 (user1)
The subtrees of 00 and 10 are delegated to user1; however, while both
subtrees may belong to the same user, it is clear that the two
subtrees are to be isolated - they're under completely separate
resource limits imposed by 0 and 1, respectively. Note that 0 and 1
aren't strictly necessary but added to ease illustrating the issue.
If user1 is allowed to move processes between the two subtrees, the
intention of the hierarchy - keeping a given group of processes under
a subtree with certain resource restrictions while delegating
management of the subtree - can be circumvented by user1.
This happens because migration permission check doesn't consider the
hierarchical nature of cgroups. To fix the issue, this patch adds an
extra permission requirement when userland tries to migrate a process
in the default hierarchy - the issuing task must have write access to
the common ancestor of "cgroup.procs" file of the ancestor in addition
to the destination's.
Conceptually, the issuer must be able to move the target process from
the source cgroup to the common ancestor of source and destination
cgroups and then to the destination. As long as delegation is done in
a proper top-down way, this guarantees that a delegatee can't smuggle
processes across disjoint delegation domains.
The next patch will add documentation on the delegation model on the
default hierarchy.
v2: Fixed missing !ret test. Spotted by Li Zefan.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Li Zefan <lizefan@huawei.com>
Separate out task / process migration permission check from
__cgroup_procs_write() into cgroup_procs_write_permission().
* Permission check is moved right above the actual migration and no
longer performed while holding rcu_read_lock().
cgroup_procs_write_permission() uses get_task_cred() / put_cred()
instead of __task_cred(). Also, !root trying to migrate kthreadd or
PF_NO_SETAFFINITY tasks will now fail with -EINVAL rather than
-EACCES which should be fine.
* The same permission check is now performed even when moving self by
specifying 0 as pid. This always succeeds so there's no functional
difference. We'll add more permission checks later and the benefits
of keeping both cases consistent outweigh the minute overhead of
doing perm checks on pid 0 case.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fix the fact that @ssid is uninitialised in the case where
CGROUP_SUBSYS_COUNT = 0 by setting ssid to 0.
Fixes: cb4a316752 ("cgroup: use bitmask to filter for_each_subsys")
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Replace the explicit checking against ss_masks inside a for_each_subsys
block with for_each_subsys_which(..., ss_mask), to take advantage of the
more readable (and more efficient) macro.
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Add a new macro for_each_subsys_which that allows all enabled cgroup
subsystems to be filtered by a bitmask, such that mask & (1 << ssid)
determines if the subsystem is to be processed in the loop body (where
ssid is the unique id of the subsystem).
Also replace the need_forkexit_callback with two separate bitmasks for
each callback to make (ss->{fork,exit}) checks unnecessary.
tj: add a short comment for "if (!CGROUP_SUBSYS_COUNT)".
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Now that threadgroup locking is made global, code paths around it can
be simplified.
* lock-verify-unlock-retry dancing removed from __cgroup_procs_write().
* Race protection against de_thread() removed from
cgroup_update_dfl_csses().
Signed-off-by: Tejun Heo <tj@kernel.org>
The cgroup side of threadgroup locking uses signal_struct->group_rwsem
to synchronize against threadgroup changes. This per-process rwsem
adds small overhead to thread creation, exit and exec paths, forces
cgroup code paths to do lock-verify-unlock-retry dance in a couple
places and makes it impossible to atomically perform operations across
multiple processes.
This patch replaces signal_struct->group_rwsem with a global
percpu_rwsem cgroup_threadgroup_rwsem which is cheaper on the reader
side and contained in cgroups proper. This patch converts one-to-one.
This does make writer side heavier and lower the granularity; however,
cgroup process migration is a fairly cold path, we do want to optimize
thread operations over it and cgroup migration operations don't take
enough time for the lower granularity to matter.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
threadgroup_change_begin/end() are used to mark the beginning and end
of threadgroup modifying operations to allow code paths which require
a threadgroup to stay stable across blocking operations to synchronize
against those sections using threadgroup_lock/unlock().
It's currently implemented as a general mechanism in sched.h using
per-signal_struct rwsem; however, this never grew non-cgroup use cases
and becomes noop if !CONFIG_CGROUPS. It turns out that cgroups is
gonna be better served with a different sycnrhonization scheme and is
a bit silly to keep cgroups specific details as a general mechanism.
What's general here is identifying the places where threadgroups are
modified. This patch restructures threadgroup locking so that
threadgroup_change_begin/end() become a place where subsystems which
need to sycnhronize against threadgroup changes can hook into.
cgroup_threadgroup_change_begin/end() which operate on the
per-signal_struct rwsem are created and threadgroup_lock/unlock() are
moved to cgroup.c and made static.
This is pure reorganization which doesn't cause any functional
changes.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Switch the type of all internal cgroup masks to (unsigned long), which
is the correct type for bitmasks. This is in preparation for the
for_each_subsys_which patch.
Signed-off-by: Aleksa Sarai <cyphar@cyphar.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
The seq_printf return value, because it's frequently misused,
will eventually be converted to void.
See: commit 1f33c41c03 ("seq_file: Rename seq_overflow() to
seq_has_overflowed() and make public")
Signed-off-by: Joe Perches <joe@perches.com>
Acked-by: Tejun Heo <tj@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
mem_cgroup_lookup() is a wrapper around mem_cgroup_from_id(), which
checks that id != 0 before issuing the function call. Today, there is
no point in this additional check apart from optimization, because there
is no css with id <= 0, so that css_from_id, called by
mem_cgroup_from_id, will return NULL for any id <= 0.
Since mem_cgroup_from_id is only called from mem_cgroup_lookup, let us
zap mem_cgroup_lookup, substituting calls to it with mem_cgroup_from_id
and moving the check if id > 0 to css_from_id.
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Acked-by: Michal Hocko <mhocko@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The wrapper already calls the appropriate free
function, use it instead of spinning our own.
Signed-off-by: Bandan Das <bsd@redhat.com>
Acked-by: Zefan Li <lizefan@huawei.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Currently, we call cgroup_subsys->bind only on unmount, remount, and
when creating a new root on mount. Since the default hierarchy root is
created in cgroup_init, we will not call cgroup_subsys->bind if the
default hierarchy is freshly mounted. As a result, some controllers will
behave incorrectly (most notably, the "memory" controller will not
enable hierarchy support). Fix this by calling cgroup_subsys->bind right
after initializing a cgroup subsystem.
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
When a new kernfs node is created, KERNFS_STATIC_NAME is used to avoid
making a separate copy of its name. It's currently only used for sysfs
attributes whose filenames are required to stay accessible and unchanged.
There are rare exceptions where these names are allocated and formatted
dynamically but for the vast majority of cases they're consts in the
rodata section.
Now that kernfs is converted to use kstrdup_const() and kfree_const(),
there's little point in keeping KERNFS_STATIC_NAME around. Remove it.
Signed-off-by: Tejun Heo <tj@kernel.org>
Cc: Andrzej Hajda <a.hajda@samsung.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Currently, we release css->id in css_release_work_fn, right before calling
css_free callback, so that when css_free is called, the id may have
already been reused for a new cgroup.
I am going to use css->id to create unique names for per memcg kmem
caches. Since kmem caches are destroyed only on css_free, I need css->id
to be freed after css_free was called to avoid name clashes. This patch
therefore moves css->id removal to css_free_work_fn. To prevent
css_from_id from returning a pointer to a stale css, it makes
css_release_work_fn replace the css ptr at css_idr:css->id with NULL.
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Acked-by: Tejun Heo <tj@kernel.org>
Cc: Christoph Lameter <cl@linux.com>
Cc: Pekka Enberg <penberg@kernel.org>
Cc: David Rientjes <rientjes@google.com>
Cc: Joonsoo Kim <iamjoonsoo.kim@lge.com>
Cc: Dave Chinner <david@fromorbit.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Since b2052564e6 ("mm: memcontrol: continue cache reclaim from
offlined groups"), re-mounting the memory controller after using it is
very likely to hang.
The cgroup core assumes that any remaining references after deleting a
cgroup are temporary in nature, and synchroneously waits for them, but
the above-mentioned commit has left-over page cache pin its css until
it is reclaimed naturally. That being said, swap entries and charged
kernel memory have been doing the same indefinite pinning forever, the
bug is just more likely to trigger with left-over page cache.
Reparenting kernel memory is highly impractical, which leaves changing
the cgroup assumptions to reflect this: once a controller has been
mounted and used, it has internal state that is independent from mount
and cgroup lifetime. It can be unmounted and remounted, but it can't
be reconfigured during subsequent mounts.
Don't offline the controller root as long as there are any children,
dead or alive. A remount will no longer wait for these old references
to drain, it will simply mount the persistent controller state again.
Reported-by: "Suzuki K. Poulose" <Suzuki.Poulose@arm.com>
Reported-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Johannes Weiner <hannes@cmpxchg.org>
Signed-off-by: Tejun Heo <tj@kernel.org>
Implement cgroup_get_e_css() which finds and gets the effective css
for the specified cgroup and subsystem combination. This function
always returns a valid pinned css. This will be used by cgroup
writeback support.
While at it, add comment to cgroup_e_css() to explain why that
function is different from cgroup_get_e_css() and has to test
cgrp->child_subsys_mask instead of cgroup_css(cgrp, ss).
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Zefan Li <lizefan@huawei.com>
Add a new cgroup_subsys operatoin ->css_e_css_changed(). This is
invoked if any of the effective csses seen from the css's cgroup may
have changed. This will be used to implement cgroup writeback
support.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Zefan Li <lizefan@huawei.com>
Add a new cgroup subsys callback css_released(). This is called when
the reference count of the css (cgroup_subsys_state) reaches zero
before RCU scheduling free.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Zefan Li <lizefan@huawei.com>
When a subsystem is offlined, its entry on @cgrp->subsys[] is cleared
asynchronously. If cgroup_subtree_control_write() is requested to
enable the subsystem again before the entry is cleared, it has to wait
for the previous offlining to finish and clear the @cgrp->subsys[]
entry before trying to enable the subsystem again.
This is currently done while verifying the input enable / disable
parameters. This used to be correct but f63070d350 ("cgroup: make
interface files visible iff enabled on cgroup->subtree_control")
breaks it. The commit is one of the commits implementing subsystem
dependency.
Through subsystem dependency, some subsystems may be enabled and
disabled implicitly in addition to the explicitly requested ones. The
actual subsystems to be enabled and disabled are determined during
@css_enable/disable calculation. The current offline wait logic skips
the ones which are already implicitly enabled and then waits for
subsystems in @enable; however, this misses the subsystems which may
be implicitly enabled through dependency from @enable. If such
implicitly subsystem hasn't yet finished offlining yet, the function
ends up trying to create a css when its @cgrp->subsys[] slot is
already occupied triggering BUG_ON() in init_and_link_css().
Fix it by moving the wait logic after @css_enable is calculated and
waiting for all the subsystems in @css_enable. This fixes the above
bug as the mask contains all subsystems which are to be enabled
including the ones enabled through dependencies.
Signed-off-by: Tejun Heo <tj@kernel.org>
Fixes: f63070d350 ("cgroup: make interface files visible iff enabled on cgroup->subtree_control")
Acked-by: Zefan Li <lizefan@huawei.com>
Make cgroup_subtree_control_write() first calculate new
subtree_control (new_sc), child_subsys_mask (new_ss) and
css_enable/disable masks before applying them to the cgroup. Also,
store the original subtree_control (old_sc) and child_subsys_mask
(old_ss) and use them to restore the orignal state after failure.
This patch shouldn't cause any behavior changes. This prepares for a
fix for a bug in the async css offline wait logic.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Zefan Li <lizefan@huawei.com>
cgroup_refresh_child_subsys_mask() calculates and updates the
effective @cgrp->child_subsys_maks according to the current
@cgrp->subtree_control. Separate out the calculation part into
cgroup_calc_child_subsys_mask(). This will be used to fix a bug in
the async css offline wait logic.
Signed-off-by: Tejun Heo <tj@kernel.org>
Acked-by: Zefan Li <lizefan@huawei.com>
