This syscall allows for the retrieval of file descriptors from other
processes, based on their pidfd. This is possible using ptrace, and
injection of parasitic code to inject code which leverages SCM_RIGHTS
to move file descriptors between a tracee and a tracer. Unfortunately,
ptrace comes with a high cost of requiring the process to be stopped,
and breaks debuggers. This does not require stopping the process under
manipulation.
One reason to use this is to allow sandboxers to take actions on file
descriptors on the behalf of another process. For example, this can be
combined with seccomp-bpf's user notification to do on-demand fd
extraction and take privileged actions. One such privileged action
is binding a socket to a privileged port.
/* prototype */
/* flags is currently reserved and should be set to 0 */
int sys_pidfd_getfd(int pidfd, int fd, unsigned int flags);
/* testing */
Ran self-test suite on x86_64
Signed-off-by: Sargun Dhillon <sargun@sargun.me>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Reviewed-by: Arnd Bergmann <arnd@arndb.de>
Link: https://lore.kernel.org/r/20200107175927.4558-3-sargun@sargun.me
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
The main motivation to add set_tid to clone3() is CRIU.
To restore a process with the same PID/TID CRIU currently uses
/proc/sys/kernel/ns_last_pid. It writes the desired (PID - 1) to
ns_last_pid and then (quickly) does a clone(). This works most of the
time, but it is racy. It is also slow as it requires multiple syscalls.
Extending clone3() to support *set_tid makes it possible restore a
process using CRIU without accessing /proc/sys/kernel/ns_last_pid and
race free (as long as the desired PID/TID is available).
This clone3() extension places the same restrictions (CAP_SYS_ADMIN)
on clone3() with *set_tid as they are currently in place for ns_last_pid.
The original version of this change was using a single value for
set_tid. At the 2019 LPC, after presenting set_tid, it was, however,
decided to change set_tid to an array to enable setting the PID of a
process in multiple PID namespaces at the same time. If a process is
created in a PID namespace it is possible to influence the PID inside
and outside of the PID namespace. Details also in the corresponding
selftest.
To create a process with the following PIDs:
PID NS level Requested PID
0 (host) 31496
1 42
2 1
For that example the two newly introduced parameters to struct
clone_args (set_tid and set_tid_size) would need to be:
set_tid[0] = 1;
set_tid[1] = 42;
set_tid[2] = 31496;
set_tid_size = 3;
If only the PIDs of the two innermost nested PID namespaces should be
defined it would look like this:
set_tid[0] = 1;
set_tid[1] = 42;
set_tid_size = 2;
The PID of the newly created process would then be the next available
free PID in the PID namespace level 0 (host) and 42 in the PID namespace
at level 1 and the PID of the process in the innermost PID namespace
would be 1.
The set_tid array is used to specify the PID of a process starting
from the innermost nested PID namespaces up to set_tid_size PID namespaces.
set_tid_size cannot be larger then the current PID namespace level.
Signed-off-by: Adrian Reber <areber@redhat.com>
Reviewed-by: Christian Brauner <christian.brauner@ubuntu.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: Dmitry Safonov <0x7f454c46@gmail.com>
Acked-by: Andrei Vagin <avagin@gmail.com>
Link: https://lore.kernel.org/r/20191115123621.142252-1-areber@redhat.com
Signed-off-by: Christian Brauner <christian.brauner@ubuntu.com>
Pull pidfd updates from Christian Brauner:
"This adds two main features.
- First, it adds polling support for pidfds. This allows process
managers to know when a (non-parent) process dies in a race-free
way.
The notification mechanism used follows the same logic that is
currently used when the parent of a task is notified of a child's
death. With this patchset it is possible to put pidfds in an
{e}poll loop and get reliable notifications for process (i.e.
thread-group) exit.
- The second feature compliments the first one by making it possible
to retrieve pollable pidfds for processes that were not created
using CLONE_PIDFD.
A lot of processes get created with traditional PID-based calls
such as fork() or clone() (without CLONE_PIDFD). For these
processes a caller can currently not create a pollable pidfd. This
is a problem for Android's low memory killer (LMK) and service
managers such as systemd.
Both patchsets are accompanied by selftests.
It's perhaps worth noting that the work done so far and the work done
in this branch for pidfd_open() and polling support do already see
some adoption:
- Android is in the process of backporting this work to all their LTS
kernels [1]
- Service managers make use of pidfd_send_signal but will need to
wait until we enable waiting on pidfds for full adoption.
- And projects I maintain make use of both pidfd_send_signal and
CLONE_PIDFD [2] and will use polling support and pidfd_open() too"
[1] https://android-review.googlesource.com/q/topic:%22pidfd+polling+support+4.9+backport%22https://android-review.googlesource.com/q/topic:%22pidfd+polling+support+4.14+backport%22https://android-review.googlesource.com/q/topic:%22pidfd+polling+support+4.19+backport%22
[2] aab6e3eb73/src/lxc/start.c (L1753)
* tag 'pidfd-updates-v5.3' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux:
tests: add pidfd_open() tests
arch: wire-up pidfd_open()
pid: add pidfd_open()
pidfd: add polling selftests
pidfd: add polling support
This adds the pidfd_open() syscall. It allows a caller to retrieve pollable
pidfds for a process which did not get created via CLONE_PIDFD, i.e. for a
process that is created via traditional fork()/clone() calls that is only
referenced by a PID:
int pidfd = pidfd_open(1234, 0);
ret = pidfd_send_signal(pidfd, SIGSTOP, NULL, 0);
With the introduction of pidfds through CLONE_PIDFD it is possible to
created pidfds at process creation time.
However, a lot of processes get created with traditional PID-based calls
such as fork() or clone() (without CLONE_PIDFD). For these processes a
caller can currently not create a pollable pidfd. This is a problem for
Android's low memory killer (LMK) and service managers such as systemd.
Both are examples of tools that want to make use of pidfds to get reliable
notification of process exit for non-parents (pidfd polling) and race-free
signal sending (pidfd_send_signal()). They intend to switch to this API for
process supervision/management as soon as possible. Having no way to get
pollable pidfds from PID-only processes is one of the biggest blockers for
them in adopting this api. With pidfd_open() making it possible to retrieve
pidfds for PID-based processes we enable them to adopt this api.
In line with Arnd's recent changes to consolidate syscall numbers across
architectures, I have added the pidfd_open() syscall to all architectures
at the same time.
Signed-off-by: Christian Brauner <christian@brauner.io>
Reviewed-by: David Howells <dhowells@redhat.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Cc: "Eric W. Biederman" <ebiederm@xmission.com>
Cc: Kees Cook <keescook@chromium.org>
Cc: Joel Fernandes (Google) <joel@joelfernandes.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Jann Horn <jannh@google.com>
Cc: Andy Lutomirsky <luto@kernel.org>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Aleksa Sarai <cyphar@cyphar.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: linux-api@vger.kernel.org
This patch adds polling support to pidfd.
Android low memory killer (LMK) needs to know when a process dies once
it is sent the kill signal. It does so by checking for the existence of
/proc/pid which is both racy and slow. For example, if a PID is reused
between when LMK sends a kill signal and checks for existence of the
PID, since the wrong PID is now possibly checked for existence.
Using the polling support, LMK will be able to get notified when a process
exists in race-free and fast way, and allows the LMK to do other things
(such as by polling on other fds) while awaiting the process being killed
to die.
For notification to polling processes, we follow the same existing
mechanism in the kernel used when the parent of the task group is to be
notified of a child's death (do_notify_parent). This is precisely when the
tasks waiting on a poll of pidfd are also awakened in this patch.
We have decided to include the waitqueue in struct pid for the following
reasons:
1. The wait queue has to survive for the lifetime of the poll. Including
it in task_struct would not be option in this case because the task can
be reaped and destroyed before the poll returns.
2. By including the struct pid for the waitqueue means that during
de_thread(), the new thread group leader automatically gets the new
waitqueue/pid even though its task_struct is different.
Appropriate test cases are added in the second patch to provide coverage of
all the cases the patch is handling.
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: Steven Rostedt <rostedt@goodmis.org>
Cc: Daniel Colascione <dancol@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Jonathan Kowalski <bl0pbl33p@gmail.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Kees Cook <keescook@chromium.org>
Cc: David Howells <dhowells@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: kernel-team@android.com
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Co-developed-by: Daniel Colascione <dancol@google.com>
Signed-off-by: Daniel Colascione <dancol@google.com>
Signed-off-by: Joel Fernandes (Google) <joel@joelfernandes.org>
Signed-off-by: Christian Brauner <christian@brauner.io>
Add SPDX license identifiers to all files which:
- Have no license information of any form
- Have EXPORT_.*_SYMBOL_GPL inside which was used in the
initial scan/conversion to ignore the file
These files fall under the project license, GPL v2 only. The resulting SPDX
license identifier is:
GPL-2.0-only
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
The failure path removes the allocated PIDs from the wrong namespace.
This could lead to us inadvertently reusing PIDs in the leaf namespace
and leaking PIDs in parent namespaces.
Fixes: 95846ecf9d ("pid: replace pid bitmap implementation with IDR API")
Cc: <stable@vger.kernel.org>
Signed-off-by: Matthew Wilcox <willy@infradead.org>
Acked-by: "Eric W. Biederman" <ebiederm@xmission.com>
Reviewed-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Make the clone and fork syscalls return EAGAIN when the limit on the
number of pids /proc/sys/kernel/pid_max is exceeded.
Currently, when the pid_max limit is exceeded, the kernel will return
ENOSPC from the fork and clone syscalls. This is contrary to the
documented behaviour, which explicitly calls out the pid_max case as one
where EAGAIN should be returned. It also leads to really confusing error
messages in userspace programs which will complain about a lack of disk
space when they fail to create processes/threads for this reason.
This error is being returned because alloc_pid() uses the idr api to find
a new pid; when there are none available, idr_alloc_cyclic() returns
-ENOSPC, and this is being propagated back to userspace.
This behaviour has been broken before, and was explicitly fixed in
commit 35f71bc0a0 ("fork: report pid reservation failure properly"),
so I think -EAGAIN is definitely the right thing to return in this case.
The current behaviour change dates from commit 95846ecf9d ("pid:
replace pid bitmap implementation with IDR AIP") and was I believe
unintentional.
This patch has no impact on the case where allocating a pid fails because
the child reaper for the namespace is dead; that case will still return
-ENOMEM.
Link: http://lkml.kernel.org/r/20180903111016.46461-1-ktsanaktsidis@zendesk.com
Fixes: 95846ecf9d ("pid: replace pid bitmap implementation with IDR AIP")
Signed-off-by: KJ Tsanaktsidis <ktsanaktsidis@zendesk.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Cc: Gargi Sharma <gs051095@gmail.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Everywhere except in the pid array we distinguish between a tasks pid and
a tasks tgid (thread group id). Even in the enumeration we want that
distinction sometimes so we have added __PIDTYPE_TGID. With leader_pid
we almost have an implementation of PIDTYPE_TGID in struct signal_struct.
Add PIDTYPE_TGID as a first class member of the pid_type enumeration and
into the pids array. Then remove the __PIDTYPE_TGID special case and the
leader_pid in signal_struct.
The net size increase is just an extra pointer added to struct pid and
an extra pair of pointers of an hlist_node added to task_struct.
The effect on code maintenance is the removal of a number of special
cases today and the potential to remove many more special cases as
PIDTYPE_TGID gets used to it's fullest. The long term potential
is allowing zombie thread group leaders to exit, which will remove
a lot more special cases in the code.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
To access these fields the code always has to go to group leader so
going to signal struct is no loss and is actually a fundamental simplification.
This saves a little bit of memory by only allocating the pid pointer array
once instead of once for every thread, and even better this removes a
few potential races caused by the fact that group_leader can be changed
by de_thread, while signal_struct can not.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
The cost is the the same and this removes the need
to worry about complications that come from de_thread
and group_leader changing.
__task_pid_nr_ns has been updated to take advantage of this change.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
This results in no change in structure size on 64-bit machines as it
fits in the padding between the gfp_t and the void *. 32-bit machines
will grow the structure from 8 to 12 bytes. Almost all radix trees are
protected with (at least) a spinlock, so as they are converted from
radix trees to xarrays, the data structures will shrink again.
Initialising the spinlock requires a name for the benefit of lockdep, so
RADIX_TREE_INIT() now needs to know the name of the radix tree it's
initialising, and so do IDR_INIT() and IDA_INIT().
Also add the xa_lock() and xa_unlock() family of wrappers to make it
easier to use the lock. If we could rely on -fplan9-extensions in the
compiler, we could avoid all of this syntactic sugar, but that wasn't
added until gcc 4.6.
Link: http://lkml.kernel.org/r/20180313132639.17387-8-willy@infradead.org
Signed-off-by: Matthew Wilcox <mawilcox@microsoft.com>
Reviewed-by: Jeff Layton <jlayton@kernel.org>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: Dave Chinner <david@fromorbit.com>
Cc: Ryusuke Konishi <konishi.ryusuke@lab.ntt.co.jp>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull init_task initializer cleanups from David Howells:
"It doesn't seem useful to have the init_task in a header file rather
than in a normal source file. We could consolidate init_task handling
instead and expand out various macros.
Here's a series of patches that consolidate init_task handling:
(1) Make THREAD_SIZE available to vmlinux.lds for cris, hexagon and
openrisc.
(2) Alter the INIT_TASK_DATA linker script macro to set
init_thread_union and init_stack rather than defining these in C.
Insert init_task and init_thread_into into the init_stack area in
the linker script as appropriate to the configuration, with
different section markers so that they end up correctly ordered.
We can then get merge ia64's init_task.c into the main one.
We then have a bunch of single-use INIT_*() macros that seem only
to be macros because they used to be used per-arch. We can then
expand these in place of the user and get rid of a few lines and
a lot of backslashes.
(3) Expand INIT_TASK() in place.
(4) Expand in place various small INIT_*() macros that are defined
conditionally. Expand them and surround them by #if[n]def/#endif
in the .c file as it takes fewer lines.
(5) Expand INIT_SIGNALS() and INIT_SIGHAND() in place.
(6) Expand INIT_STRUCT_PID in place.
These macros can then be discarded"
* tag 'init_task-20180117' of git://git.kernel.org/pub/scm/linux/kernel/git/dhowells/linux-fs:
Expand INIT_STRUCT_PID and remove
Expand the INIT_SIGNALS and INIT_SIGHAND macros and remove
Expand various INIT_* macros and remove
Expand INIT_TASK() in init/init_task.c and remove
Construct init thread stack in the linker script rather than by union
openrisc: Make THREAD_SIZE available to vmlinux.lds
hexagon: Make THREAD_SIZE available to vmlinux.lds
cris: Make THREAD_SIZE available to vmlinux.lds
Patch series "Replacing PID bitmap implementation with IDR API", v4.
This series replaces kernel bitmap implementation of PID allocation with
IDR API. These patches are written to simplify the kernel by replacing
custom code with calls to generic code.
The following are the stats for pid and pid_namespace object files
before and after the replacement. There is a noteworthy change between
the IDR and bitmap implementation.
Before
text data bss dec hex filename
8447 3894 64 12405 3075 kernel/pid.o
After
text data bss dec hex filename
3397 304 0 3701 e75 kernel/pid.o
Before
text data bss dec hex filename
5692 1842 192 7726 1e2e kernel/pid_namespace.o
After
text data bss dec hex filename
2854 216 16 3086 c0e kernel/pid_namespace.o
The following are the stats for ps, pstree and calling readdir on /proc
for 10,000 processes.
ps:
With IDR API With bitmap
real 0m1.479s 0m2.319s
user 0m0.070s 0m0.060s
sys 0m0.289s 0m0.516s
pstree:
With IDR API With bitmap
real 0m1.024s 0m1.794s
user 0m0.348s 0m0.612s
sys 0m0.184s 0m0.264s
proc:
With IDR API With bitmap
real 0m0.059s 0m0.074s
user 0m0.000s 0m0.004s
sys 0m0.016s 0m0.016s
This patch (of 2):
Replace the current bitmap implementation for Process ID allocation.
Functions that are no longer required, for example, free_pidmap(),
alloc_pidmap(), etc. are removed. The rest of the functions are
modified to use the IDR API. The change was made to make the PID
allocation less complex by replacing custom code with calls to generic
API.
[gs051095@gmail.com: v6]
Link: http://lkml.kernel.org/r/1507760379-21662-2-git-send-email-gs051095@gmail.com
[avagin@openvz.org: restore the old behaviour of the ns_last_pid sysctl]
Link: http://lkml.kernel.org/r/20171106183144.16368-1-avagin@openvz.org
Link: http://lkml.kernel.org/r/1507583624-22146-2-git-send-email-gs051095@gmail.com
Signed-off-by: Gargi Sharma <gs051095@gmail.com>
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Cc: Julia Lawall <julia.lawall@lip6.fr>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Pavel Tatashin <pasha.tatashin@oracle.com>
Cc: Kirill Tkhai <ktkhai@virtuozzo.com>
Cc: Eric W. Biederman <ebiederm@xmission.com>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This was reported many times, and this was even mentioned in commit
52ee2dfdd4 ("pids: refactor vnr/nr_ns helpers to make them safe") but
somehow nobody bothered to fix the obvious problem: task_tgid_nr_ns() is
not safe because task->group_leader points to nowhere after the exiting
task passes exit_notify(), rcu_read_lock() can not help.
We really need to change __unhash_process() to nullify group_leader,
parent, and real_parent, but this needs some cleanups. Until then we
can turn task_tgid_nr_ns() into another user of __task_pid_nr_ns() and
fix the problem.
Reported-by: Troy Kensinger <tkensinger@google.com>
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
