Container admin can create new namespaces and force kernel to allocate up
to several pages of memory for the namespaces and its associated
structures.
Net and uts namespaces have enabled accounting for such allocations. It
makes sense to account for rest ones to restrict the host's memory
consumption from inside the memcg-limited container.
Link: https://lkml.kernel.org/r/5525bcbf-533e-da27-79b7-158686c64e13@virtuozzo.com
Signed-off-by: Vasily Averin <vvs@virtuozzo.com>
Acked-by: Serge Hallyn <serge@hallyn.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Acked-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: Shakeel Butt <shakeelb@google.com>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Alexey Dobriyan <adobriyan@gmail.com>
Cc: Andrei Vagin <avagin@gmail.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Borislav Petkov <bp@suse.de>
Cc: Dmitry Safonov <0x7f454c46@gmail.com>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: "J. Bruce Fields" <bfields@fieldses.org>
Cc: Jeff Layton <jlayton@kernel.org>
Cc: Jens Axboe <axboe@kernel.dk>
Cc: Jiri Slaby <jirislaby@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Tejun Heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Yutian Yang <nglaive@gmail.com>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull user namespace rlimit handling update from Eric Biederman:
"This is the work mainly by Alexey Gladkov to limit rlimits to the
rlimits of the user that created a user namespace, and to allow users
to have stricter limits on the resources created within a user
namespace."
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
cred: add missing return error code when set_cred_ucounts() failed
ucounts: Silence warning in dec_rlimit_ucounts
ucounts: Set ucount_max to the largest positive value the type can hold
kselftests: Add test to check for rlimit changes in different user namespaces
Reimplement RLIMIT_MEMLOCK on top of ucounts
Reimplement RLIMIT_SIGPENDING on top of ucounts
Reimplement RLIMIT_MSGQUEUE on top of ucounts
Reimplement RLIMIT_NPROC on top of ucounts
Use atomic_t for ucounts reference counting
Add a reference to ucounts for each cred
Increase size of ucounts to atomic_long_t
The ns->ucount_max[] is signed long which is less than the rlimit size.
We have to protect ucount_max[] from overflow and only use the largest
value that we can hold.
On 32bit using "long" instead of "unsigned long" to hold the counts has
the downside that RLIMIT_MSGQUEUE and RLIMIT_MEMLOCK are limited to 2GiB
instead of 4GiB. I don't think anyone cares but it should be mentioned
in case someone does.
The RLIMIT_NPROC and RLIMIT_SIGPENDING used atomic_t so their maximum
hasn't changed.
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Link: https://lkml.kernel.org/r/1825a5dfa18bc5a570e79feb05e2bd07fd57e7e3.1619094428.git.legion@kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The rlimit counter is tied to uid in the user_namespace. This allows
rlimit values to be specified in userns even if they are already
globally exceeded by the user. However, the value of the previous
user_namespaces cannot be exceeded.
Changelog
v11:
* Fix issue found by lkp robot.
v8:
* Fix issues found by lkp-tests project.
v7:
* Keep only ucounts for RLIMIT_MEMLOCK checks instead of struct cred.
v6:
* Fix bug in hugetlb_file_setup() detected by trinity.
Reported-by: kernel test robot <oliver.sang@intel.com>
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Link: https://lkml.kernel.org/r/970d50c70c71bfd4496e0e8d2a0a32feebebb350.1619094428.git.legion@kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The rlimit counter is tied to uid in the user_namespace. This allows
rlimit values to be specified in userns even if they are already
globally exceeded by the user. However, the value of the previous
user_namespaces cannot be exceeded.
Changelog
v11:
* Revert most of changes to fix performance issues.
v10:
* Fix memory leak on get_ucounts failure.
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Link: https://lkml.kernel.org/r/df9d7764dddd50f28616b7840de74ec0f81711a8.1619094428.git.legion@kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The rlimit counter is tied to uid in the user_namespace. This allows
rlimit values to be specified in userns even if they are already
globally exceeded by the user. However, the value of the previous
user_namespaces cannot be exceeded.
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Link: https://lkml.kernel.org/r/2531f42f7884bbfee56a978040b3e0d25cdf6cde.1619094428.git.legion@kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The rlimit counter is tied to uid in the user_namespace. This allows
rlimit values to be specified in userns even if they are already
globally exceeded by the user. However, the value of the previous
user_namespaces cannot be exceeded.
To illustrate the impact of rlimits, let's say there is a program that
does not fork. Some service-A wants to run this program as user X in
multiple containers. Since the program never fork the service wants to
set RLIMIT_NPROC=1.
service-A
\- program (uid=1000, container1, rlimit_nproc=1)
\- program (uid=1000, container2, rlimit_nproc=1)
The service-A sets RLIMIT_NPROC=1 and runs the program in container1.
When the service-A tries to run a program with RLIMIT_NPROC=1 in
container2 it fails since user X already has one running process.
We cannot use existing inc_ucounts / dec_ucounts because they do not
allow us to exceed the maximum for the counter. Some rlimits can be
overlimited by root or if the user has the appropriate capability.
Changelog
v11:
* Change inc_rlimit_ucounts() which now returns top value of ucounts.
* Drop inc_rlimit_ucounts_and_test() because the return code of
inc_rlimit_ucounts() can be checked.
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Link: https://lkml.kernel.org/r/c5286a8aa16d2d698c222f7532f3d735c82bc6bc.1619094428.git.legion@kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
For RLIMIT_NPROC and some other rlimits the user_struct that holds the
global limit is kept alive for the lifetime of a process by keeping it
in struct cred. Adding a pointer to ucounts in the struct cred will
allow to track RLIMIT_NPROC not only for user in the system, but for
user in the user_namespace.
Updating ucounts may require memory allocation which may fail. So, we
cannot change cred.ucounts in the commit_creds() because this function
cannot fail and it should always return 0. For this reason, we modify
cred.ucounts before calling the commit_creds().
Changelog
v6:
* Fix null-ptr-deref in is_ucounts_overlimit() detected by trinity. This
error was caused by the fact that cred_alloc_blank() left the ucounts
pointer empty.
Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Alexey Gladkov <legion@kernel.org>
Link: https://lkml.kernel.org/r/b37aaef28d8b9b0d757e07ba6dd27281bbe39259.1619094428.git.legion@kernel.org
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
cap_setfcap is required to create file capabilities.
Since commit 8db6c34f1d ("Introduce v3 namespaced file capabilities"),
a process running as uid 0 but without cap_setfcap is able to work
around this as follows: unshare a new user namespace which maps parent
uid 0 into the child namespace.
While this task will not have new capabilities against the parent
namespace, there is a loophole due to the way namespaced file
capabilities are represented as xattrs. File capabilities valid in
userns 1 are distinguished from file capabilities valid in userns 2 by
the kuid which underlies uid 0. Therefore the restricted root process
can unshare a new self-mapping namespace, add a namespaced file
capability onto a file, then use that file capability in the parent
namespace.
To prevent that, do not allow mapping parent uid 0 if the process which
opened the uid_map file does not have CAP_SETFCAP, which is the
capability for setting file capabilities.
As a further wrinkle: a task can unshare its user namespace, then open
its uid_map file itself, and map (only) its own uid. In this case we do
not have the credential from before unshare, which was potentially more
restricted. So, when creating a user namespace, we record whether the
creator had CAP_SETFCAP. Then we can use that during map_write().
With this patch:
1. Unprivileged user can still unshare -Ur
ubuntu@caps:~$ unshare -Ur
root@caps:~# logout
2. Root user can still unshare -Ur
ubuntu@caps:~$ sudo bash
root@caps:/home/ubuntu# unshare -Ur
root@caps:/home/ubuntu# logout
3. Root user without CAP_SETFCAP cannot unshare -Ur:
root@caps:/home/ubuntu# /sbin/capsh --drop=cap_setfcap --
root@caps:/home/ubuntu# /sbin/setcap cap_setfcap=p /sbin/setcap
unable to set CAP_SETFCAP effective capability: Operation not permitted
root@caps:/home/ubuntu# unshare -Ur
unshare: write failed /proc/self/uid_map: Operation not permitted
Note: an alternative solution would be to allow uid 0 mappings by
processes without CAP_SETFCAP, but to prevent such a namespace from
writing any file capabilities. This approach can be seen at [1].
Background history: commit 95ebabde38 ("capabilities: Don't allow
writing ambiguous v3 file capabilities") tried to fix the issue by
preventing v3 fscaps to be written to disk when the root uid would map
to the same uid in nested user namespaces. This led to regressions for
various workloads. For example, see [2]. Ultimately this is a valid
use-case we have to support meaning we had to revert this change in
3b0c2d3eaa ("Revert 95ebabde38 ("capabilities: Don't allow writing
ambiguous v3 file capabilities")").
Link: https://git.kernel.org/pub/scm/linux/kernel/git/sergeh/linux.git/log/?h=2021-04-15/setfcap-nsfscaps-v4 [1]
Link: https://github.com/containers/buildah/issues/3071 [2]
Signed-off-by: Serge Hallyn <serge@hallyn.com>
Reviewed-by: Andrew G. Morgan <morgan@kernel.org>
Tested-by: Christian Brauner <christian.brauner@ubuntu.com>
Reviewed-by: Christian Brauner <christian.brauner@ubuntu.com>
Tested-by: Giuseppe Scrivano <gscrivan@redhat.com>
Cc: Eric Biederman <ebiederm@xmission.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQRAhzRXHqcMeLMyaSiRxhvAZXjcogUCX9daOgAKCRCRxhvAZXjc
ohPkAQChXUB2BAjtIzXlCkZoDBbzHHblm5DZ37oy/4xYFmAcEwEA5sw6dQqyGHnF
GEP9def51HvXLpBV2BzNUGggo1SoGgQ=
=w/cO
-----END PGP SIGNATURE-----
Merge tag 'fixes-v5.11' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux
Pull misc fixes from Christian Brauner:
"This contains several fixes which felt worth being combined into a
single branch:
- Use put_nsproxy() instead of open-coding it switch_task_namespaces()
- Kirill's work to unify lifecycle management for all namespaces. The
lifetime counters are used identically for all namespaces types.
Namespaces may of course have additional unrelated counters and
these are not altered. This work allows us to unify the type of the
counters and reduces maintenance cost by moving the counter in one
place and indicating that basic lifetime management is identical
for all namespaces.
- Peilin's fix adding three byte padding to Dmitry's
PTRACE_GET_SYSCALL_INFO uapi struct to prevent an info leak.
- Two smal patches to convert from the /* fall through */ comment
annotation to the fallthrough keyword annotation which I had taken
into my branch and into -next before df561f6688 ("treewide: Use
fallthrough pseudo-keyword") made it upstream which fixed this
tree-wide.
Since I didn't want to invalidate all testing for other commits I
didn't rebase and kept them"
* tag 'fixes-v5.11' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux:
nsproxy: use put_nsproxy() in switch_task_namespaces()
sys: Convert to the new fallthrough notation
signal: Convert to the new fallthrough notation
time: Use generic ns_common::count
cgroup: Use generic ns_common::count
mnt: Use generic ns_common::count
user: Use generic ns_common::count
pid: Use generic ns_common::count
ipc: Use generic ns_common::count
uts: Use generic ns_common::count
net: Use generic ns_common::count
ns: Add a common refcount into ns_common
ptrace: Prevent kernel-infoleak in ptrace_get_syscall_info()
Fix multiple occurrences of duplicated words in kernel/.
Fix one typo/spello on the same line as a duplicate word. Change one
instance of "the the" to "that the". Otherwise just drop one of the
repeated words.
Signed-off-by: Randy Dunlap <rdunlap@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Link: https://lkml.kernel.org/r/98202fa6-8919-ef63-9efe-c0fad5ca7af1@infradead.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Switch over user namespaces to use the newly introduced common lifetime
counter.
Currently every namespace type has its own lifetime counter which is stored
in the specific namespace struct. The lifetime counters are used
identically for all namespaces types. Namespaces may of course have
additional unrelated counters and these are not altered.
This introduces a common lifetime counter into struct ns_common. The
ns_common struct encompasses information that all namespaces share. That
should include the lifetime counter since its common for all of them.
It also allows us to unify the type of the counters across all namespaces.
Most of them use refcount_t but one uses atomic_t and at least one uses
kref. Especially the last one doesn't make much sense since it's just a
wrapper around refcount_t since 2016 and actually complicates cleanup
operations by having to use container_of() to cast the correct namespace
struct out of struct ns_common.
Having the lifetime counter for the namespaces in one place reduces
maintenance cost. Not just because after switching all namespaces over we
will have removed more code than we added but also because the logic is
more easily understandable and we indicate to the user that the basic
lifetime requirements for all namespaces are currently identical.
Signed-off-by: Kirill Tkhai <ktkhai@virtuozzo.com>
Reviewed-by: Kees Cook <keescook@chromium.org>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Link: https://lore.kernel.org/r/159644979754.604812.601625186726406922.stgit@localhost.localdomain
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Add a simple struct nsset. It holds all necessary pieces to switch to a new
set of namespaces without leaving a task in a half-switched state which we
will make use of in the next patch. This patch switches the existing setns
logic over without causing a change in setns() behavior. This brings
setns() closer to how unshare() works(). The prepare_ns() function is
responsible to prepare all necessary information. This has two reasons.
First it minimizes dependencies between individual namespaces, i.e. all
install handler can expect that all fields are properly initialized
independent in what order they are called in. Second, this makes the code
easier to maintain and easier to follow if it needs to be changed.
The prepare_ns() helper will only be switched over to use a flags argument
in the next patch. Here it will still use nstype as a simple integer
argument which was argued would be clearer. I'm not particularly
opinionated about this if it really helps or not. The struct nsset itself
already contains the flags field since its name already indicates that it
can contain information required by different namespaces. None of this
should have functional consequences.
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Reviewed-by: Serge Hallyn <serge@hallyn.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Serge Hallyn <serge@hallyn.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Aleksa Sarai <cyphar@cyphar.com>
Link: https://lore.kernel.org/r/20200505140432.181565-2-christian.brauner@ubuntu.com
-----BEGIN PGP SIGNATURE-----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=Push
-----END PGP SIGNATURE-----
Merge tag 'keys-namespace-20190627' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs
Pull keyring namespacing from David Howells:
"These patches help make keys and keyrings more namespace aware.
Firstly some miscellaneous patches to make the process easier:
- Simplify key index_key handling so that the word-sized chunks
assoc_array requires don't have to be shifted about, making it
easier to add more bits into the key.
- Cache the hash value in the key so that we don't have to calculate
on every key we examine during a search (it involves a bunch of
multiplications).
- Allow keying_search() to search non-recursively.
Then the main patches:
- Make it so that keyring names are per-user_namespace from the point
of view of KEYCTL_JOIN_SESSION_KEYRING so that they're not
accessible cross-user_namespace.
keyctl_capabilities() shows KEYCTL_CAPS1_NS_KEYRING_NAME for this.
- Move the user and user-session keyrings to the user_namespace
rather than the user_struct. This prevents them propagating
directly across user_namespaces boundaries (ie. the KEY_SPEC_*
flags will only pick from the current user_namespace).
- Make it possible to include the target namespace in which the key
shall operate in the index_key. This will allow the possibility of
multiple keys with the same description, but different target
domains to be held in the same keyring.
keyctl_capabilities() shows KEYCTL_CAPS1_NS_KEY_TAG for this.
- Make it so that keys are implicitly invalidated by removal of a
domain tag, causing them to be garbage collected.
- Institute a network namespace domain tag that allows keys to be
differentiated by the network namespace in which they operate. New
keys that are of a type marked 'KEY_TYPE_NET_DOMAIN' are assigned
the network domain in force when they are created.
- Make it so that the desired network namespace can be handed down
into the request_key() mechanism. This allows AFS, NFS, etc. to
request keys specific to the network namespace of the superblock.
This also means that the keys in the DNS record cache are
thenceforth namespaced, provided network filesystems pass the
appropriate network namespace down into dns_query().
For DNS, AFS and NFS are good, whilst CIFS and Ceph are not. Other
cache keyrings, such as idmapper keyrings, also need to set the
domain tag - for which they need access to the network namespace of
the superblock"
* tag 'keys-namespace-20190627' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
keys: Pass the network namespace into request_key mechanism
keys: Network namespace domain tag
keys: Garbage collect keys for which the domain has been removed
keys: Include target namespace in match criteria
keys: Move the user and user-session keyrings to the user_namespace
keys: Namespace keyring names
keys: Add a 'recurse' flag for keyring searches
keys: Cache the hash value to avoid lots of recalculation
keys: Simplify key description management
Move the user and user-session keyrings to the user_namespace struct rather
than pinning them from the user_struct struct. This prevents these
keyrings from propagating across user-namespaces boundaries with regard to
the KEY_SPEC_* flags, thereby making them more useful in a containerised
environment.
The issue is that a single user_struct may be represent UIDs in several
different namespaces.
The way the patch does this is by attaching a 'register keyring' in each
user_namespace and then sticking the user and user-session keyrings into
that. It can then be searched to retrieve them.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Jann Horn <jannh@google.com>
Keyring names are held in a single global list that any process can pick
from by means of keyctl_join_session_keyring (provided the keyring grants
Search permission). This isn't very container friendly, however.
Make the following changes:
(1) Make default session, process and thread keyring names begin with a
'.' instead of '_'.
(2) Keyrings whose names begin with a '.' aren't added to the list. Such
keyrings are system specials.
(3) Replace the global list with per-user_namespace lists. A keyring adds
its name to the list for the user_namespace that it is currently in.
(4) When a user_namespace is deleted, it just removes itself from the
keyring name list.
The global keyring_name_lock is retained for accessing the name lists.
This allows (4) to work.
This can be tested by:
# keyctl newring foo @s
995906392
# unshare -U
$ keyctl show
...
995906392 --alswrv 65534 65534 \_ keyring: foo
...
$ keyctl session foo
Joined session keyring: 935622349
As can be seen, a new session keyring was created.
The capability bit KEYCTL_CAPS1_NS_KEYRING_NAME is set if the kernel is
employing this feature.
Signed-off-by: David Howells <dhowells@redhat.com>
cc: Eric W. Biederman <ebiederm@xmission.com>
Based on 1 normalized pattern(s):
this program is free software you can redistribute it and or modify
it under the terms of the gnu general public license as published by
the free software foundation version 2 of the license
extracted by the scancode license scanner the SPDX license identifier
GPL-2.0-only
has been chosen to replace the boilerplate/reference in 315 file(s).
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Allison Randal <allison@lohutok.net>
Reviewed-by: Armijn Hemel <armijn@tjaldur.nl>
Cc: linux-spdx@vger.kernel.org
Link: https://lkml.kernel.org/r/20190531190115.503150771@linutronix.de
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The current logic first clones the extent array and sorts both copies, then
maps the lower IDs of the forward mapping into the lower namespace, but
doesn't map the lower IDs of the reverse mapping.
This means that code in a nested user namespace with >5 extents will see
incorrect IDs. It also breaks some access checks, like
inode_owner_or_capable() and privileged_wrt_inode_uidgid(), so a process
can incorrectly appear to be capable relative to an inode.
To fix it, we have to make sure that the "lower_first" members of extents
in both arrays are translated; and we have to make sure that the reverse
map is sorted *after* the translation (since otherwise the translation can
break the sorting).
This is CVE-2018-18955.
Fixes: 6397fac491 ("userns: bump idmap limits to 340")
Cc: stable@vger.kernel.org
Signed-off-by: Jann Horn <jannh@google.com>
Tested-by: Eric W. Biederman <ebiederm@xmission.com>
Reviewed-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The old code would hold the userns_state_mutex indefinitely if
memdup_user_nul stalled due to e.g. a userfault region. Prevent that by
moving the memdup_user_nul in front of the mutex_lock().
Note: This changes the error precedence of invalid buf/count/*ppos vs
map already written / capabilities missing.
Fixes: 22d917d80e ("userns: Rework the user_namespace adding uid/gid...")
Cc: stable@vger.kernel.org
Signed-off-by: Jann Horn <jannh@google.com>
Acked-by: Christian Brauner <christian@brauner.io>
Acked-by: Serge Hallyn <serge@hallyn.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Unprivileged users are normally restricted from mounting with the
allow_other option by system policy, but this could be bypassed for a mount
done with user namespace root permissions. In such cases allow_other should
not allow users outside the userns to access the mount as doing so would
give the unprivileged user the ability to manipulate processes it would
otherwise be unable to manipulate. Restrict allow_other to apply to users
in the same userns used at mount or a descendant of that namespace. Also
export current_in_userns() for use by fuse when built as a module.
Reviewed-by: Serge Hallyn <serge@hallyn.com>
Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Signed-off-by: Dongsu Park <dongsu@kinvolk.io>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>
Pull user namespace update from Eric Biederman:
"The only change that is production ready this round is the work to
increase the number of uid and gid mappings a user namespace can
support from 5 to 340.
This code was carefully benchmarked and it was confirmed that in the
existing cases the performance remains the same. In the worst case
with 340 mappings an cache cold stat times go from 158ns to 248ns.
That is noticable but still quite small, and only the people who are
doing crazy things pay the cost.
This work uncovered some documentation and cleanup opportunities in
the mapping code, and patches to make those cleanups and improve the
documentation will be coming in the next merge window"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
userns: Simplify insert_extent
userns: Make map_id_down a wrapper for map_id_range_down
userns: Don't read extents twice in m_start
userns: Simplify the user and group mapping functions
userns: Don't special case a count of 0
userns: bump idmap limits to 340
userns: use union in {g,u}idmap struct
Consolidate the code to write to the new mapping at the end of the
function to remove the duplication. Move the increase in the number
of mappings into insert_extent, keeping the logic together.
Just a small increase in readability and maintainability.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
There is no good reason for this code duplication, the number of cache
line accesses not the number of instructions are the bottleneck in
this code.
Therefore simplify maintenance by removing unnecessary code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
This is important so reading /proc/<pid>/{uid_map,gid_map,projid_map} while
the map is being written does not do strange things.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Consolidate reading the number of extents and computing the return
value in the map_id_down, map_id_range_down and map_id_range.
This removal of one read of extents makes one smp_rmb unnecessary
and makes the code safe it is executed during the map write. Reading
the number of extents twice and depending on the result being the same
is not safe, as it could be 0 the first time and > 5 the second time,
which would lead to misinterpreting the union fields.
The consolidation of the return value just removes a duplicate
caluculation which should make it easier to understand and maintain
the code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
There are quite some use cases where users run into the current limit for
{g,u}id mappings. Consider a user requesting us to map everything but 999, and
1001 for a given range of 1000000000 with a sub{g,u}id layout of:
some-user:100000:1000000000
some-user:999:1
some-user:1000:1
some-user:1001:1
some-user:1002:1
This translates to:
MAPPING-TYPE | CONTAINER | HOST | RANGE |
-------------|-----------|---------|-----------|
uid | 999 | 999 | 1 |
uid | 1001 | 1001 | 1 |
uid | 0 | 1000000 | 999 |
uid | 1000 | 1001000 | 1 |
uid | 1002 | 1001002 | 999998998 |
------------------------------------------------
gid | 999 | 999 | 1 |
gid | 1001 | 1001 | 1 |
gid | 0 | 1000000 | 999 |
gid | 1000 | 1001000 | 1 |
gid | 1002 | 1001002 | 999998998 |
which is already the current limit.
As discussed at LPC simply bumping the number of limits is not going to work
since this would mean that struct uid_gid_map won't fit into a single cache-line
anymore thereby regressing performance for the base-cases. The same problem
seems to arise when using a single pointer. So the idea is to use
struct uid_gid_extent {
u32 first;
u32 lower_first;
u32 count;
};
struct uid_gid_map { /* 64 bytes -- 1 cache line */
u32 nr_extents;
union {
struct uid_gid_extent extent[UID_GID_MAP_MAX_BASE_EXTENTS];
struct {
struct uid_gid_extent *forward;
struct uid_gid_extent *reverse;
};
};
};
For the base cases we will only use the struct uid_gid_extent extent member. If
we go over UID_GID_MAP_MAX_BASE_EXTENTS mappings we perform a single 4k
kmalloc() which means we can have a maximum of 340 mappings
(340 * size(struct uid_gid_extent) = 4080). For the latter case we use two
pointers "forward" and "reverse". The forward pointer points to an array sorted
by "first" and the reverse pointer points to an array sorted by "lower_first".
We can then perform binary search on those arrays.
Performance Testing:
When Eric introduced the extent-based struct uid_gid_map approach he measured
the performanc impact of his idmap changes:
> My benchmark consisted of going to single user mode where nothing else was
> running. On an ext4 filesystem opening 1,000,000 files and looping through all
> of the files 1000 times and calling fstat on the individuals files. This was
> to ensure I was benchmarking stat times where the inodes were in the kernels
> cache, but the inode values were not in the processors cache. My results:
> v3.4-rc1: ~= 156ns (unmodified v3.4-rc1 with user namespace support disabled)
> v3.4-rc1-userns-: ~= 155ns (v3.4-rc1 with my user namespace patches and user namespace support disabled)
> v3.4-rc1-userns+: ~= 164ns (v3.4-rc1 with my user namespace patches and user namespace support enabled)
I used an identical approach on my laptop. Here's a thorough description of what
I did. I built a 4.14.0-rc4 mainline kernel with my new idmap patches applied. I
booted into single user mode and used an ext4 filesystem to open/create
1,000,000 files. Then I looped through all of the files calling fstat() on each
of them 1000 times and calculated the mean fstat() time for a single file. (The
test program can be found below.)
Here are the results. For fun, I compared the first version of my patch which
scaled linearly with the new version of the patch:
| # MAPPINGS | PATCH-V1 | PATCH-NEW |
|--------------|------------|-----------|
| 0 mappings | 158 ns | 158 ns |
| 1 mappings | 164 ns | 157 ns |
| 2 mappings | 170 ns | 158 ns |
| 3 mappings | 175 ns | 161 ns |
| 5 mappings | 187 ns | 165 ns |
| 10 mappings | 218 ns | 199 ns |
| 50 mappings | 528 ns | 218 ns |
| 100 mappings | 980 ns | 229 ns |
| 200 mappings | 1880 ns | 239 ns |
| 300 mappings | 2760 ns | 240 ns |
| 340 mappings | not tested | 248 ns |
Here's the test program I used. I asked Eric what he did and this is a more
"advanced" implementation of the idea. It's pretty straight-forward:
#define __GNU_SOURCE
#define __STDC_FORMAT_MACROS
#include <errno.h>
#include <dirent.h>
#include <fcntl.h>
#include <inttypes.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <unistd.h>
#include <sys/stat.h>
#include <sys/time.h>
#include <sys/types.h>
int main(int argc, char *argv[])
{
int ret;
size_t i, k;
int fd[1000000];
int times[1000];
char pathname[4096];
struct stat st;
struct timeval t1, t2;
uint64_t time_in_mcs;
uint64_t sum = 0;
if (argc != 2) {
fprintf(stderr, "Please specify a directory where to create "
"the test files\n");
exit(EXIT_FAILURE);
}
for (i = 0; i < sizeof(fd) / sizeof(fd[0]); i++) {
sprintf(pathname, "%s/idmap_test_%zu", argv[1], i);
fd[i]= open(pathname, O_RDWR | O_CREAT, S_IXUSR | S_IXGRP | S_IXOTH);
if (fd[i] < 0) {
ssize_t j;
for (j = i; j >= 0; j--)
close(fd[j]);
exit(EXIT_FAILURE);
}
}
for (k = 0; k < 1000; k++) {
ret = gettimeofday(&t1, NULL);
if (ret < 0)
goto close_all;
for (i = 0; i < sizeof(fd) / sizeof(fd[0]); i++) {
ret = fstat(fd[i], &st);
if (ret < 0)
goto close_all;
}
ret = gettimeofday(&t2, NULL);
if (ret < 0)
goto close_all;
time_in_mcs = (1000000 * t2.tv_sec + t2.tv_usec) -
(1000000 * t1.tv_sec + t1.tv_usec);
printf("Total time in micro seconds: %" PRIu64 "\n",
time_in_mcs);
printf("Total time in nanoseconds: %" PRIu64 "\n",
time_in_mcs * 1000);
printf("Time per file in nanoseconds: %" PRIu64 "\n",
(time_in_mcs * 1000) / 1000000);
times[k] = (time_in_mcs * 1000) / 1000000;
}
close_all:
for (i = 0; i < sizeof(fd) / sizeof(fd[0]); i++)
close(fd[i]);
if (ret < 0)
exit(EXIT_FAILURE);
for (k = 0; k < 1000; k++) {
sum += times[k];
}
printf("Mean time per file in nanoseconds: %" PRIu64 "\n", sum / 1000);
exit(EXIT_SUCCESS);;
}
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
CC: Serge Hallyn <serge@hallyn.com>
CC: Eric Biederman <ebiederm@xmission.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Please do not apply this to mainline directly, instead please re-run the
coccinelle script shown below and apply its output.
For several reasons, it is desirable to use {READ,WRITE}_ONCE() in
preference to ACCESS_ONCE(), and new code is expected to use one of the
former. So far, there's been no reason to change most existing uses of
ACCESS_ONCE(), as these aren't harmful, and changing them results in
churn.
However, for some features, the read/write distinction is critical to
correct operation. To distinguish these cases, separate read/write
accessors must be used. This patch migrates (most) remaining
ACCESS_ONCE() instances to {READ,WRITE}_ONCE(), using the following
coccinelle script:
----
// Convert trivial ACCESS_ONCE() uses to equivalent READ_ONCE() and
// WRITE_ONCE()
// $ make coccicheck COCCI=/home/mark/once.cocci SPFLAGS="--include-headers" MODE=patch
virtual patch
@ depends on patch @
expression E1, E2;
@@
- ACCESS_ONCE(E1) = E2
+ WRITE_ONCE(E1, E2)
@ depends on patch @
expression E;
@@
- ACCESS_ONCE(E)
+ READ_ONCE(E)
----
Signed-off-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Paul E. McKenney <paulmck@linux.vnet.ibm.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: davem@davemloft.net
Cc: linux-arch@vger.kernel.org
Cc: mpe@ellerman.id.au
Cc: shuah@kernel.org
Cc: snitzer@redhat.com
Cc: thor.thayer@linux.intel.com
Cc: tj@kernel.org
Cc: viro@zeniv.linux.org.uk
Cc: will.deacon@arm.com
Link: http://lkml.kernel.org/r/1508792849-3115-19-git-send-email-paulmck@linux.vnet.ibm.com
Signed-off-by: Ingo Molnar <mingo@kernel.org>
It is pointless and confusing to allow a pid namespace hierarchy and
the user namespace hierarchy to get out of sync. The owner of a child
pid namespace should be the owner of the parent pid namespace or
a descendant of the owner of the parent pid namespace.
Otherwise it is possible to construct scenarios where a process has a
capability over a parent pid namespace but does not have the
capability over a child pid namespace. Which confusingly makes
permission checks non-transitive.
It requires use of setns into a pid namespace (but not into a user
namespace) to create such a scenario.
Add the function in_userns to help in making this determination.
v2: Optimized in_userns by using level as suggested
by: Kirill Tkhai <ktkhai@virtuozzo.com>
Ref: 49f4d8b93c ("pidns: Capture the user namespace and filter ns_last_pid")
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
We are going to split <linux/sched/signal.h> out of <linux/sched.h>, which
will have to be picked up from other headers and a couple of .c files.
Create a trivial placeholder <linux/sched/signal.h> file that just
maps to <linux/sched.h> to make this patch obviously correct and
bisectable.
Include the new header in the files that are going to need it.
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: linux-kernel@vger.kernel.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
From: Andrey Vagin <avagin@openvz.org>
Each namespace has an owning user namespace and now there is not way
to discover these relationships.
Pid and user namepaces are hierarchical. There is no way to discover
parent-child relationships too.
Why we may want to know relationships between namespaces?
One use would be visualization, in order to understand the running
system. Another would be to answer the question: what capability does
process X have to perform operations on a resource governed by namespace
Y?
One more use-case (which usually called abnormal) is checkpoint/restart.
In CRIU we are going to dump and restore nested namespaces.
There [1] was a discussion about which interface to choose to determing
relationships between namespaces.
Eric suggested to add two ioctl-s [2]:
> Grumble, Grumble. I think this may actually a case for creating ioctls
> for these two cases. Now that random nsfs file descriptors are bind
> mountable the original reason for using proc files is not as pressing.
>
> One ioctl for the user namespace that owns a file descriptor.
> One ioctl for the parent namespace of a namespace file descriptor.
Here is an implementaions of these ioctl-s.
$ man man7/namespaces.7
...
Since Linux 4.X, the following ioctl(2) calls are supported for
namespace file descriptors. The correct syntax is:
fd = ioctl(ns_fd, ioctl_type);
where ioctl_type is one of the following:
NS_GET_USERNS
Returns a file descriptor that refers to an owning user names‐
pace.
NS_GET_PARENT
Returns a file descriptor that refers to a parent namespace.
This ioctl(2) can be used for pid and user namespaces. For
user namespaces, NS_GET_PARENT and NS_GET_USERNS have the same
meaning.
In addition to generic ioctl(2) errors, the following specific ones
can occur:
EINVAL NS_GET_PARENT was called for a nonhierarchical namespace.
EPERM The requested namespace is outside of the current namespace
scope.
[1] https://lkml.org/lkml/2016/7/6/158
[2] https://lkml.org/lkml/2016/7/9/101
Changes for v2:
* don't return ENOENT for init_user_ns and init_pid_ns. There is nothing
outside of the init namespace, so we can return EPERM in this case too.
> The fewer special cases the easier the code is to get
> correct, and the easier it is to read. // Eric
Changes for v3:
* rename ns->get_owner() to ns->owner(). get_* usually means that it
grabs a reference.
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: "Michael Kerrisk (man-pages)" <mtk.manpages@gmail.com>
Cc: "W. Trevor King" <wking@tremily.us>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>
Cc: Serge Hallyn <serge.hallyn@canonical.com>
Pid and user namepaces are hierarchical. There is no way to discover
parent-child relationships.
In a future we will use this interface to dump and restore nested
namespaces.
Acked-by: Serge Hallyn <serge@hallyn.com>
Signed-off-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Return -EPERM if an owning user namespace is outside of a process
current user namespace.
v2: In a first version ns_get_owner returned ENOENT for init_user_ns.
This special cases was removed from this version. There is nothing
outside of init_user_ns, so we can return EPERM.
v3: rename ns->get_owner() to ns->owner(). get_* usually means that it
grabs a reference.
Acked-by: Serge Hallyn <serge@hallyn.com>
Signed-off-by: Andrei Vagin <avagin@openvz.org>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
The current error codes returned when a the per user per user
namespace limit are hit (EINVAL, EUSERS, and ENFILE) are wrong. I
asked for advice on linux-api and it we made clear that those were
the wrong error code, but a correct effor code was not suggested.
The best general error code I have found for hitting a resource limit
is ENOSPC. It is not perfect but as it is unambiguous it will serve
until someone comes up with a better error code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
The same kind of recursive sane default limit and policy
countrol that has been implemented for the user namespace
is desirable for the other namespaces, so generalize
the user namespace refernce count into a ucount.
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Add a structure that is per user and per user ns and use it to hold
the count of user namespaces. This makes prevents one user from
creating denying service to another user by creating the maximum
number of user namespaces.
Rename the sysctl export of the maximum count from
/proc/sys/userns/max_user_namespaces to /proc/sys/user/max_user_namespaces
to reflect that the count is now per user.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Export the export the maximum number of user namespaces as
/proc/sys/userns/max_user_namespaces.
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Limit per userns sysctls to only be opened for write by a holder
of CAP_SYS_RESOURCE.
Add all of the necessary boilerplate for having per user namespace
sysctls.
Acked-by: Kees Cook <keescook@chromium.org>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Add the necessary boiler plate to move freeing of user namespaces into
work queue and thus into process context where things can sleep.
This is a necessary precursor to per user namespace sysctls.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Capability sets attached to files must be ignored except in the
user namespaces where the mounter is privileged, i.e. s_user_ns
and its descendants. Otherwise a vector exists for gaining
privileges in namespaces where a user is not already privileged.
Add a new helper function, current_in_user_ns(), to test whether a user
namespace is the same as or a descendant of another namespace.
Use this helper to determine whether a file's capability set
should be applied to the caps constructed during exec.
--EWB Replaced in_userns with the simpler current_in_userns.
Acked-by: Serge Hallyn <serge.hallyn@canonical.com>
Signed-off-by: Seth Forshee <seth.forshee@canonical.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Credit where credit is due: this idea comes from Christoph Lameter with
a lot of valuable input from Serge Hallyn. This patch is heavily based
on Christoph's patch.
===== The status quo =====
On Linux, there are a number of capabilities defined by the kernel. To
perform various privileged tasks, processes can wield capabilities that
they hold.
Each task has four capability masks: effective (pE), permitted (pP),
inheritable (pI), and a bounding set (X). When the kernel checks for a
capability, it checks pE. The other capability masks serve to modify
what capabilities can be in pE.
Any task can remove capabilities from pE, pP, or pI at any time. If a
task has a capability in pP, it can add that capability to pE and/or pI.
If a task has CAP_SETPCAP, then it can add any capability to pI, and it
can remove capabilities from X.
Tasks are not the only things that can have capabilities; files can also
have capabilities. A file can have no capabilty information at all [1].
If a file has capability information, then it has a permitted mask (fP)
and an inheritable mask (fI) as well as a single effective bit (fE) [2].
File capabilities modify the capabilities of tasks that execve(2) them.
A task that successfully calls execve has its capabilities modified for
the file ultimately being excecuted (i.e. the binary itself if that
binary is ELF or for the interpreter if the binary is a script.) [3] In
the capability evolution rules, for each mask Z, pZ represents the old
value and pZ' represents the new value. The rules are:
pP' = (X & fP) | (pI & fI)
pI' = pI
pE' = (fE ? pP' : 0)
X is unchanged
For setuid binaries, fP, fI, and fE are modified by a moderately
complicated set of rules that emulate POSIX behavior. Similarly, if
euid == 0 or ruid == 0, then fP, fI, and fE are modified differently
(primary, fP and fI usually end up being the full set). For nonroot
users executing binaries with neither setuid nor file caps, fI and fP
are empty and fE is false.
As an extra complication, if you execute a process as nonroot and fE is
set, then the "secure exec" rules are in effect: AT_SECURE gets set,
LD_PRELOAD doesn't work, etc.
This is rather messy. We've learned that making any changes is
dangerous, though: if a new kernel version allows an unprivileged
program to change its security state in a way that persists cross
execution of a setuid program or a program with file caps, this
persistent state is surprisingly likely to allow setuid or file-capped
programs to be exploited for privilege escalation.
===== The problem =====
Capability inheritance is basically useless.
If you aren't root and you execute an ordinary binary, fI is zero, so
your capabilities have no effect whatsoever on pP'. This means that you
can't usefully execute a helper process or a shell command with elevated
capabilities if you aren't root.
On current kernels, you can sort of work around this by setting fI to
the full set for most or all non-setuid executable files. This causes
pP' = pI for nonroot, and inheritance works. No one does this because
it's a PITA and it isn't even supported on most filesystems.
If you try this, you'll discover that every nonroot program ends up with
secure exec rules, breaking many things.
This is a problem that has bitten many people who have tried to use
capabilities for anything useful.
===== The proposed change =====
This patch adds a fifth capability mask called the ambient mask (pA).
pA does what most people expect pI to do.
pA obeys the invariant that no bit can ever be set in pA if it is not
set in both pP and pI. Dropping a bit from pP or pI drops that bit from
pA. This ensures that existing programs that try to drop capabilities
still do so, with a complication. Because capability inheritance is so
broken, setting KEEPCAPS, using setresuid to switch to nonroot uids, and
then calling execve effectively drops capabilities. Therefore,
setresuid from root to nonroot conditionally clears pA unless
SECBIT_NO_SETUID_FIXUP is set. Processes that don't like this can
re-add bits to pA afterwards.
The capability evolution rules are changed:
pA' = (file caps or setuid or setgid ? 0 : pA)
pP' = (X & fP) | (pI & fI) | pA'
pI' = pI
pE' = (fE ? pP' : pA')
X is unchanged
If you are nonroot but you have a capability, you can add it to pA. If
you do so, your children get that capability in pA, pP, and pE. For
example, you can set pA = CAP_NET_BIND_SERVICE, and your children can
automatically bind low-numbered ports. Hallelujah!
Unprivileged users can create user namespaces, map themselves to a
nonzero uid, and create both privileged (relative to their namespace)
and unprivileged process trees. This is currently more or less
impossible. Hallelujah!
You cannot use pA to try to subvert a setuid, setgid, or file-capped
program: if you execute any such program, pA gets cleared and the
resulting evolution rules are unchanged by this patch.
Users with nonzero pA are unlikely to unintentionally leak that
capability. If they run programs that try to drop privileges, dropping
privileges will still work.
It's worth noting that the degree of paranoia in this patch could
possibly be reduced without causing serious problems. Specifically, if
we allowed pA to persist across executing non-pA-aware setuid binaries
and across setresuid, then, naively, the only capabilities that could
leak as a result would be the capabilities in pA, and any attacker
*already* has those capabilities. This would make me nervous, though --
setuid binaries that tried to privilege-separate might fail to do so,
and putting CAP_DAC_READ_SEARCH or CAP_DAC_OVERRIDE into pA could have
unexpected side effects. (Whether these unexpected side effects would
be exploitable is an open question.) I've therefore taken the more
paranoid route. We can revisit this later.
An alternative would be to require PR_SET_NO_NEW_PRIVS before setting
ambient capabilities. I think that this would be annoying and would
make granting otherwise unprivileged users minor ambient capabilities
(CAP_NET_BIND_SERVICE or CAP_NET_RAW for example) much less useful than
it is with this patch.
===== Footnotes =====
[1] Files that are missing the "security.capability" xattr or that have
unrecognized values for that xattr end up with has_cap set to false.
The code that does that appears to be complicated for no good reason.
[2] The libcap capability mask parsers and formatters are dangerously
misleading and the documentation is flat-out wrong. fE is *not* a mask;
it's a single bit. This has probably confused every single person who
has tried to use file capabilities.
[3] Linux very confusingly processes both the script and the interpreter
if applicable, for reasons that elude me. The results from thinking
about a script's file capabilities and/or setuid bits are mostly
discarded.
Preliminary userspace code is here, but it needs updating:
https://git.kernel.org/cgit/linux/kernel/git/luto/util-linux-playground.git/commit/?h=cap_ambient&id=7f5afbd175d2
Here is a test program that can be used to verify the functionality
(from Christoph):
/*
* Test program for the ambient capabilities. This program spawns a shell
* that allows running processes with a defined set of capabilities.
*
* (C) 2015 Christoph Lameter <cl@linux.com>
* Released under: GPL v3 or later.
*
*
* Compile using:
*
* gcc -o ambient_test ambient_test.o -lcap-ng
*
* This program must have the following capabilities to run properly:
* Permissions for CAP_NET_RAW, CAP_NET_ADMIN, CAP_SYS_NICE
*
* A command to equip the binary with the right caps is:
*
* setcap cap_net_raw,cap_net_admin,cap_sys_nice+p ambient_test
*
*
* To get a shell with additional caps that can be inherited by other processes:
*
* ./ambient_test /bin/bash
*
*
* Verifying that it works:
*
* From the bash spawed by ambient_test run
*
* cat /proc/$$/status
*
* and have a look at the capabilities.
*/
#include <stdlib.h>
#include <stdio.h>
#include <errno.h>
#include <cap-ng.h>
#include <sys/prctl.h>
#include <linux/capability.h>
/*
* Definitions from the kernel header files. These are going to be removed
* when the /usr/include files have these defined.
*/
#define PR_CAP_AMBIENT 47
#define PR_CAP_AMBIENT_IS_SET 1
#define PR_CAP_AMBIENT_RAISE 2
#define PR_CAP_AMBIENT_LOWER 3
#define PR_CAP_AMBIENT_CLEAR_ALL 4
static void set_ambient_cap(int cap)
{
int rc;
capng_get_caps_process();
rc = capng_update(CAPNG_ADD, CAPNG_INHERITABLE, cap);
if (rc) {
printf("Cannot add inheritable cap\n");
exit(2);
}
capng_apply(CAPNG_SELECT_CAPS);
/* Note the two 0s at the end. Kernel checks for these */
if (prctl(PR_CAP_AMBIENT, PR_CAP_AMBIENT_RAISE, cap, 0, 0)) {
perror("Cannot set cap");
exit(1);
}
}
int main(int argc, char **argv)
{
int rc;
set_ambient_cap(CAP_NET_RAW);
set_ambient_cap(CAP_NET_ADMIN);
set_ambient_cap(CAP_SYS_NICE);
printf("Ambient_test forking shell\n");
if (execv(argv[1], argv + 1))
perror("Cannot exec");
return 0;
}
Signed-off-by: Christoph Lameter <cl@linux.com> # Original author
Signed-off-by: Andy Lutomirski <luto@kernel.org>
Acked-by: Serge E. Hallyn <serge.hallyn@ubuntu.com>
Acked-by: Kees Cook <keescook@chromium.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Aaron Jones <aaronmdjones@gmail.com>
Cc: Ted Ts'o <tytso@mit.edu>
Cc: Andrew G. Morgan <morgan@kernel.org>
Cc: Mimi Zohar <zohar@linux.vnet.ibm.com>
Cc: Austin S Hemmelgarn <ahferroin7@gmail.com>
Cc: Markku Savela <msa@moth.iki.fi>
Cc: Jarkko Sakkinen <jarkko.sakkinen@linux.intel.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: James Morris <james.l.morris@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The code that places signals in signal queues computes the uids, gids,
and pids at the time the signals are enqueued. Which means that tasks
that share signal queues must be in the same pid and user namespaces.
Sharing signal handlers is fine, but bizarre.
So make the code in fork and userns_install clearer by only testing
for what is functionally necessary.
Also update the comment in unshare about unsharing a user namespace to
be a little more explicit and make a little more sense.
Acked-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Pull user namespace related fixes from Eric Biederman:
"As these are bug fixes almost all of thes changes are marked for
backporting to stable.
The first change (implicitly adding MNT_NODEV on remount) addresses a
regression that was created when security issues with unprivileged
remount were closed. I go on to update the remount test to make it
easy to detect if this issue reoccurs.
Then there are a handful of mount and umount related fixes.
Then half of the changes deal with the a recently discovered design
bug in the permission checks of gid_map. Unix since the beginning has
allowed setting group permissions on files to less than the user and
other permissions (aka ---rwx---rwx). As the unix permission checks
stop as soon as a group matches, and setgroups allows setting groups
that can not later be dropped, results in a situtation where it is
possible to legitimately use a group to assign fewer privileges to a
process. Which means dropping a group can increase a processes
privileges.
The fix I have adopted is that gid_map is now no longer writable
without privilege unless the new file /proc/self/setgroups has been
set to permanently disable setgroups.
The bulk of user namespace using applications even the applications
using applications using user namespaces without privilege remain
unaffected by this change. Unfortunately this ix breaks a couple user
space applications, that were relying on the problematic behavior (one
of which was tools/selftests/mount/unprivileged-remount-test.c).
To hopefully prevent needing a regression fix on top of my security
fix I rounded folks who work with the container implementations mostly
like to be affected and encouraged them to test the changes.
> So far nothing broke on my libvirt-lxc test bed. :-)
> Tested with openSUSE 13.2 and libvirt 1.2.9.
> Tested-by: Richard Weinberger <richard@nod.at>
> Tested on Fedora20 with libvirt 1.2.11, works fine.
> Tested-by: Chen Hanxiao <chenhanxiao@cn.fujitsu.com>
> Ok, thanks - yes, unprivileged lxc is working fine with your kernels.
> Just to be sure I was testing the right thing I also tested using
> my unprivileged nsexec testcases, and they failed on setgroup/setgid
> as now expected, and succeeded there without your patches.
> Tested-by: Serge Hallyn <serge.hallyn@ubuntu.com>
> I tested this with Sandstorm. It breaks as is and it works if I add
> the setgroups thing.
> Tested-by: Andy Lutomirski <luto@amacapital.net> # breaks things as designed :("
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
userns: Unbreak the unprivileged remount tests
userns; Correct the comment in map_write
userns: Allow setting gid_maps without privilege when setgroups is disabled
userns: Add a knob to disable setgroups on a per user namespace basis
userns: Rename id_map_mutex to userns_state_mutex
userns: Only allow the creator of the userns unprivileged mappings
userns: Check euid no fsuid when establishing an unprivileged uid mapping
userns: Don't allow unprivileged creation of gid mappings
userns: Don't allow setgroups until a gid mapping has been setablished
userns: Document what the invariant required for safe unprivileged mappings.
groups: Consolidate the setgroups permission checks
mnt: Clear mnt_expire during pivot_root
mnt: Carefully set CL_UNPRIVILEGED in clone_mnt
mnt: Move the clear of MNT_LOCKED from copy_tree to it's callers.
umount: Do not allow unmounting rootfs.
umount: Disallow unprivileged mount force
mnt: Update unprivileged remount test
mnt: Implicitly add MNT_NODEV on remount when it was implicitly added by mount
It is important that all maps are less than PAGE_SIZE
or else setting the last byte of the buffer to '0'
could write off the end of the allocated storage.
Correct the misleading comment.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Now that setgroups can be disabled and not reenabled, setting gid_map
without privielge can now be enabled when setgroups is disabled.
This restores most of the functionality that was lost when unprivileged
setting of gid_map was removed. Applications that use this functionality
will need to check to see if they use setgroups or init_groups, and if they
don't they can be fixed by simply disabling setgroups before writing to
gid_map.
Cc: stable@vger.kernel.org
Reviewed-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
- Expose the knob to user space through a proc file /proc/<pid>/setgroups
A value of "deny" means the setgroups system call is disabled in the
current processes user namespace and can not be enabled in the
future in this user namespace.
A value of "allow" means the segtoups system call is enabled.
- Descendant user namespaces inherit the value of setgroups from
their parents.
- A proc file is used (instead of a sysctl) as sysctls currently do
not allow checking the permissions at open time.
- Writing to the proc file is restricted to before the gid_map
for the user namespace is set.
This ensures that disabling setgroups at a user namespace
level will never remove the ability to call setgroups
from a process that already has that ability.
A process may opt in to the setgroups disable for itself by
creating, entering and configuring a user namespace or by calling
setns on an existing user namespace with setgroups disabled.
Processes without privileges already can not call setgroups so this
is a noop. Prodcess with privilege become processes without
privilege when entering a user namespace and as with any other path
to dropping privilege they would not have the ability to call
setgroups. So this remains within the bounds of what is possible
without a knob to disable setgroups permanently in a user namespace.
Cc: stable@vger.kernel.org
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Generalize id_map_mutex so it can be used for more state of a user namespace.
Cc: stable@vger.kernel.org
Reviewed-by: Andy Lutomirski <luto@amacapital.net>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Vasily Averin