The current posix-cpu-timer code uses pids when holding persistent
references in timers. However the lookups from clock_id_t still return
tasks that need to be converted into pids for use.
This results in usage being pid->task->pid and that can race with
release_task and de_thread. This can lead to some not wrong but
surprising results. Surprising enough that Oleg and I both thought
there were some bugs in the code for a while.
This set of changes modifies the code to just lookup, verify, and return
pids from the clockid_t lookups to remove those potentialy troublesome
races.
Eric W. Biederman (3):
posix-cpu-timers: Extend rcu_read_lock removing task_struct references
posix-cpu-timers: Replace cpu_timer_pid_type with clock_pid_type
posix-cpu-timers: Replace __get_task_for_clock with pid_for_clock
kernel/time/posix-cpu-timers.c | 102 ++++++++++++++++++-----------------------
1 file changed, 45 insertions(+), 57 deletions(-)
Suggested-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Starting from 2c4704756c ("pids: Move the pgrp and session pid pointers
from task_struct to signal_struct") __task_pid_nr_ns() doesn't dereference
task->group_leader, we can remove the pid_alive() check.
pid_nr_ns() has to check pid != NULL anyway, pid_alive() just adds the
unnecessary confusion.
Signed-off-by: Oleg Nesterov <oleg@redhat.com>
Reviewed-by: "Eric W. Biederman" <ebiederm@xmission.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Now that the codes store references to pids instead of referendes to
tasks. Looking up a task for a clock instead of looking up a struct
pid makes the code more difficult to verify it is correct than
necessary.
In posix_cpu_timers_create get_task_pid can race with release_task for
threads and return a NULL pid. As put_pid and cpu_timer_task_rcu
handle NULL pids just fine the code works without problems but it is
an extra case to consider and keep in mind while verifying and
modifying the code.
There are races with de_thread to consider that only don't apply
because thread clocks are only allowed for threads in the same
thread_group.
So instead of leaving a burden for people making modification to the
code in the future return a rcu protected struct pid for the clock
instead.
The logic for __get_task_for_pid and lookup_task has been folded into
the new function pid_for_clock with the only change being the logic
has been modified from working on a task to working on a pid that
will be returned.
In posix_cpu_clock_get instead of calling pid_for_clock checking the
result and then calling pid_task to get the task. The result of
pid_for_clock is fed directly into pid_task. This is safe because
pid_task handles NULL pids. As such an extra error check was
unnecessary.
Instead of hiding the flag that enables the special clock_gettime
handling, I have made the 3 callers just pass the flag in themselves.
That is less code and seems just as simple to work with as the
wrapper functions.
Historically the clock_gettime special case of allowing a process
clock to be found by the thread id did not even exist [33ab0fec33]
but Thomas Gleixner reports that he has found code that uses that
functionality [55e8c8eb2c].
Link: https://lkml.kernel.org/r/87zhaxqkwa.fsf@nanos.tec.linutronix.de/
Ref: 33ab0fec33 ("posix-timers: Consolidate posix_cpu_clock_get()")
Ref: 55e8c8eb2c ("posix-cpu-timers: Store a reference to a pid not a task")
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Taking a clock and returning a pid_type is a more general and
a superset of taking a timer and returning a pid_type.
Perform this generalization so that future changes may use
this code on clocks as well as timers.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Now that the code stores of pid references it is no longer necessary
or desirable to take a reference on task_struct in __get_task_for_clock.
Instead extend the scope of rcu_read_lock and remove the reference
counting on struct task_struct entirely.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Now that both !thread paths through lookup_task call
thread_group_leader, unify them into the single test at the end of
lookup_task.
This unification just makes it clear what is happening in the gettime
special case of lookup_task.
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Replace has_group_leader_pid with thread_group_leader. Years ago Oleg
suggested changing thread_group_leader to has_group_leader_pid to handle
races. Looking at the code then and now I don't see how it ever helped.
Especially as then the code really did need to be the
thread_group_leader.
Today it doesn't make a difference if thread_group_leader races with
de_thread as the task returned from lookup_task in the non-thread case is
just used to find values in task->signal.
Since the races with de_thread have never been handled revert
has_group_header_pid to thread_group_leader for clarity.
Update the comment in lookup_task to remove implementation details that
are no longer true and to mention task->signal instead of task->sighand,
as the relevant cpu timer details are all in task->signal.
Ref: 55e8c8eb2c ("posix-cpu-timers: Store a reference to a pid not a task")
Ref: c0deae8c95 ("posix-cpu-timers: Rcu_read_lock/unlock protect find_task_by_vpid call")
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
When the thread group leader changes during exec and the old leaders
thread is reaped proc_flush_pid will flush the dentries for the entire
process because the leader still has it's original pid.
Fix this by exchanging the pids in an rcu safe manner,
and wrapping the code to do that up in a helper exchange_tids.
When I removed switch_exec_pids and introduced this behavior
in d73d65293e ("[PATCH] pidhash: kill switch_exec_pids") there
really was nothing that cared as flushing happened with
the cached dentry and de_thread flushed both of them on exec.
This lack of fully exchanging pids became a problem a few months later
when I introduced 48e6484d49 ("[PATCH] proc: Rewrite the proc dentry
flush on exit optimization"). Which overlooked the de_thread case
was no longer swapping pids, and I was looking up proc dentries
by task->pid.
The current behavior isn't properly a bug as everything in proc will
continue to work correctly just a little bit less efficiently. Fix
this just so there are no little surprise corner cases waiting to bite
people.
-- Oleg points out this could be an issue in next_tgid in proc where
has_group_leader_pid is called, and reording some of the assignments
should fix that.
-- Oleg points out this will break the 10 year old hack in __exit_signal.c
> /*
> * This can only happen if the caller is de_thread().
> * FIXME: this is the temporary hack, we should teach
> * posix-cpu-timers to handle this case correctly.
> */
> if (unlikely(has_group_leader_pid(tsk)))
> posix_cpu_timers_exit_group(tsk);
The code in next_tgid has been changed to use PIDTYPE_TGID,
and the posix cpu timers code has been fixed so it does not
need the 10 year old hack, so this should be safe to merge
now.
Link: https://lore.kernel.org/lkml/87h7x3ajll.fsf_-_@x220.int.ebiederm.org/
Acked-by: Linus Torvalds <torvalds@linux-foundation.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Fixes: 48e6484d49 ("[PATCH] proc: Rewrite the proc dentry flush on exit optimization").
Signed-off-by: Eric W. Biederman <ebiederm@xmission.com>
Oleg pointed out that in the unlikely event the kernel is compiled
with CONFIG_PROC_FS unset that release_task will now leak the pid.
Move the put_pid out of proc_flush_pid into release_task to fix this
and to guarantee I don't make that mistake again.
When possible it makes sense to keep get and put in the same function
so it can easily been seen how they pair up.
Fixes: 7bc3e6e55a ("proc: Use a list of inodes to flush from proc")
Reported-by: Oleg Nesterov <oleg@redhat.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Christof Meerwald <cmeerw@cmeerw.org> writes:
> Hi,
>
> this is probably related to commit
> 7a0cf09494 (signal: Correct namespace
> fixups of si_pid and si_uid).
>
> With a 5.6.5 kernel I am seeing SIGCHLD signals that don't include a
> properly set si_pid field - this seems to happen for multi-threaded
> child processes.
>
> A simple test program (based on the sample from the signalfd man page):
>
> #include <sys/signalfd.h>
> #include <signal.h>
> #include <unistd.h>
> #include <spawn.h>
> #include <stdlib.h>
> #include <stdio.h>
>
> #define handle_error(msg) \
> do { perror(msg); exit(EXIT_FAILURE); } while (0)
>
> int main(int argc, char *argv[])
> {
> sigset_t mask;
> int sfd;
> struct signalfd_siginfo fdsi;
> ssize_t s;
>
> sigemptyset(&mask);
> sigaddset(&mask, SIGCHLD);
>
> if (sigprocmask(SIG_BLOCK, &mask, NULL) == -1)
> handle_error("sigprocmask");
>
> pid_t chldpid;
> char *chldargv[] = { "./sfdclient", NULL };
> posix_spawn(&chldpid, "./sfdclient", NULL, NULL, chldargv, NULL);
>
> sfd = signalfd(-1, &mask, 0);
> if (sfd == -1)
> handle_error("signalfd");
>
> for (;;) {
> s = read(sfd, &fdsi, sizeof(struct signalfd_siginfo));
> if (s != sizeof(struct signalfd_siginfo))
> handle_error("read");
>
> if (fdsi.ssi_signo == SIGCHLD) {
> printf("Got SIGCHLD %d %d %d %d\n",
> fdsi.ssi_status, fdsi.ssi_code,
> fdsi.ssi_uid, fdsi.ssi_pid);
> return 0;
> } else {
> printf("Read unexpected signal\n");
> }
> }
> }
>
>
> and a multi-threaded client to test with:
>
> #include <unistd.h>
> #include <pthread.h>
>
> void *f(void *arg)
> {
> sleep(100);
> }
>
> int main()
> {
> pthread_t t[8];
>
> for (int i = 0; i != 8; ++i)
> {
> pthread_create(&t[i], NULL, f, NULL);
> }
> }
>
> I tried to do a bit of debugging and what seems to be happening is
> that
>
> /* From an ancestor pid namespace? */
> if (!task_pid_nr_ns(current, task_active_pid_ns(t))) {
>
> fails inside task_pid_nr_ns because the check for "pid_alive" fails.
>
> This code seems to be called from do_notify_parent and there we
> actually have "tsk != current" (I am assuming both are threads of the
> current process?)
I instrumented the code with a warning and received the following backtrace:
> WARNING: CPU: 0 PID: 777 at kernel/pid.c:501 __task_pid_nr_ns.cold.6+0xc/0x15
> Modules linked in:
> CPU: 0 PID: 777 Comm: sfdclient Not tainted 5.7.0-rc1userns+ #2924
> Hardware name: Bochs Bochs, BIOS Bochs 01/01/2011
> RIP: 0010:__task_pid_nr_ns.cold.6+0xc/0x15
> Code: ff 66 90 48 83 ec 08 89 7c 24 04 48 8d 7e 08 48 8d 74 24 04 e8 9a b6 44 00 48 83 c4 08 c3 48 c7 c7 59 9f ac 82 e8 c2 c4 04 00 <0f> 0b e9 3fd
> RSP: 0018:ffffc9000042fbf8 EFLAGS: 00010046
> RAX: 000000000000000c RBX: 0000000000000000 RCX: ffffc9000042faf4
> RDX: 0000000000000000 RSI: 0000000000000000 RDI: ffffffff81193d29
> RBP: ffffc9000042fc18 R08: 0000000000000000 R09: 0000000000000001
> R10: 000000100f938416 R11: 0000000000000309 R12: ffff8880b941c140
> R13: 0000000000000000 R14: 0000000000000000 R15: ffff8880b941c140
> FS: 0000000000000000(0000) GS:ffff8880bca00000(0000) knlGS:0000000000000000
> CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
> CR2: 00007f2e8c0a32e0 CR3: 0000000002e10000 CR4: 00000000000006f0
> Call Trace:
> send_signal+0x1c8/0x310
> do_notify_parent+0x50f/0x550
> release_task.part.21+0x4fd/0x620
> do_exit+0x6f6/0xaf0
> do_group_exit+0x42/0xb0
> get_signal+0x13b/0xbb0
> do_signal+0x2b/0x670
> ? __audit_syscall_exit+0x24d/0x2b0
> ? rcu_read_lock_sched_held+0x4d/0x60
> ? kfree+0x24c/0x2b0
> do_syscall_64+0x176/0x640
> ? trace_hardirqs_off_thunk+0x1a/0x1c
> entry_SYSCALL_64_after_hwframe+0x49/0xb3
The immediate problem is as Christof noticed that "pid_alive(current) == false".
This happens because do_notify_parent is called from the last thread to exit
in a process after that thread has been reaped.
The bigger issue is that do_notify_parent can be called from any
process that manages to wait on a thread of a multi-threaded process
from wait_task_zombie. So any logic based upon current for
do_notify_parent is just nonsense, as current can be pretty much
anything.
So change do_notify_parent to call __send_signal directly.
Inspecting the code it appears this problem has existed since the pid
namespace support started handling this case in 2.6.30. This fix only
backports to 7a0cf09494 ("signal: Correct namespace fixups of si_pid and si_uid")
where the problem logic was moved out of __send_signal and into send_signal.
Cc: stable@vger.kernel.org
Fixes: 6588c1e3ff ("signals: SI_USER: Masquerade si_pid when crossing pid ns boundary")
Ref: 921cf9f630 ("signals: protect cinit from unblocked SIG_DFL signals")
Link: https://lore.kernel.org/lkml/20200419201336.GI22017@edge.cmeerw.net/
Reported-by: Christof Meerwald <cmeerw@cmeerw.org>
Acked-by: Oleg Nesterov <oleg@redhat.com>
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
Pull time(keeping) updates from Thomas Gleixner:
- Fix the time_for_children symlink in /proc/$PID/ so it properly
reflects that it part of the 'time' namespace
- Add the missing userns limit for the allowed number of time
namespaces, which was half defined but the actual array member was
not added. This went unnoticed as the array has an exessive empty
member at the end but introduced a user visible regression as the
output was corrupted.
- Prevent further silent ucount corruption by adding a BUILD_BUG_ON()
to catch half updated data.
* tag 'timers-urgent-2020-04-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
ucount: Make sure ucounts in /proc/sys/user don't regress again
time/namespace: Add max_time_namespaces ucount
time/namespace: Fix time_for_children symlink
Pull scheduler fixes/updates from Thomas Gleixner:
- Deduplicate the average computations in the scheduler core and the
fair class code.
- Fix a raise between runtime distribution and assignement which can
cause exceeding the quota by up to 70%.
- Prevent negative results in the imbalanace calculation
- Remove a stale warning in the workqueue code which can be triggered
since the call site was moved out of preempt disabled code. It's a
false positive.
- Deduplicate the print macros for procfs
- Add the ucmap values to the SCHED_DEBUG procfs output for completness
* tag 'sched-urgent-2020-04-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
sched/debug: Add task uclamp values to SCHED_DEBUG procfs
sched/debug: Factor out printing formats into common macros
sched/debug: Remove redundant macro define
sched/core: Remove unused rq::last_load_update_tick
workqueue: Remove the warning in wq_worker_sleeping()
sched/fair: Fix negative imbalance in imbalance calculation
sched/fair: Fix race between runtime distribution and assignment
sched/fair: Align rq->avg_idle and rq->avg_scan_cost
Pull perf fixes from Thomas Gleixner:
"Three fixes/updates for perf:
- Fix the perf event cgroup tracking which tries to track the cgroup
even for disabled events.
- Add Ice Lake server support for uncore events
- Disable pagefaults when retrieving the physical address in the
sampling code"
* tag 'perf-urgent-2020-04-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
perf/core: Disable page faults when getting phys address
perf/x86/intel/uncore: Add Ice Lake server uncore support
perf/cgroup: Correct indirection in perf_less_group_idx()
perf/core: Fix event cgroup tracking
Pull locking fixes from Thomas Gleixner:
"Three small fixes/updates for the locking core code:
- Plug a task struct reference leak in the percpu rswem
implementation.
- Document the refcount interaction with PID_MAX_LIMIT
- Improve the 'invalid wait context' data dump in lockdep so it
contains all information which is required to decode the problem"
* tag 'locking-urgent-2020-04-12' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
locking/lockdep: Improve 'invalid wait context' splat
locking/refcount: Document interaction with PID_MAX_LIMIT
locking/percpu-rwsem: Fix a task_struct refcount
Pull dma-mapping fixes from Christoph Hellwig:
- fix an integer truncation in dma_direct_get_required_mask
(Kishon Vijay Abraham)
- fix the display of dma mapping types (Grygorii Strashko)
* tag 'dma-mapping-5.7-1' of git://git.infradead.org/users/hch/dma-mapping:
dma-debug: fix displaying of dma allocation type
dma-direct: fix data truncation in dma_direct_get_required_mask()
Merge yet more updates from Andrew Morton:
- Almost all of the rest of MM (memcg, slab-generic, slab, pagealloc,
gup, hugetlb, pagemap, memremap)
- Various other things (hfs, ocfs2, kmod, misc, seqfile)
* akpm: (34 commits)
ipc/util.c: sysvipc_find_ipc() should increase position index
kernel/gcov/fs.c: gcov_seq_next() should increase position index
fs/seq_file.c: seq_read(): add info message about buggy .next functions
drivers/dma/tegra20-apb-dma.c: fix platform_get_irq.cocci warnings
change email address for Pali Rohár
selftests: kmod: test disabling module autoloading
selftests: kmod: fix handling test numbers above 9
docs: admin-guide: document the kernel.modprobe sysctl
fs/filesystems.c: downgrade user-reachable WARN_ONCE() to pr_warn_once()
kmod: make request_module() return an error when autoloading is disabled
mm/memremap: set caching mode for PCI P2PDMA memory to WC
mm/memory_hotplug: add pgprot_t to mhp_params
powerpc/mm: thread pgprot_t through create_section_mapping()
x86/mm: introduce __set_memory_prot()
x86/mm: thread pgprot_t through init_memory_mapping()
mm/memory_hotplug: rename mhp_restrictions to mhp_params
mm/memory_hotplug: drop the flags field from struct mhp_restrictions
mm/special: create generic fallbacks for pte_special() and pte_mkspecial()
mm/vma: introduce VM_ACCESS_FLAGS
mm/vma: define a default value for VM_DATA_DEFAULT_FLAGS
...
Patch series "module autoloading fixes and cleanups", v5.
This series fixes a bug where request_module() was reporting success to
kernel code when module autoloading had been completely disabled via
'echo > /proc/sys/kernel/modprobe'.
It also addresses the issues raised on the original thread
(https://lkml.kernel.org/lkml/20200310223731.126894-1-ebiggers@kernel.org/T/#u)
bydocumenting the modprobe sysctl, adding a self-test for the empty path
case, and downgrading a user-reachable WARN_ONCE().
This patch (of 4):
It's long been possible to disable kernel module autoloading completely
(while still allowing manual module insertion) by setting
/proc/sys/kernel/modprobe to the empty string.
This can be preferable to setting it to a nonexistent file since it
avoids the overhead of an attempted execve(), avoids potential
deadlocks, and avoids the call to security_kernel_module_request() and
thus on SELinux-based systems eliminates the need to write SELinux rules
to dontaudit module_request.
However, when module autoloading is disabled in this way,
request_module() returns 0. This is broken because callers expect 0 to
mean that the module was successfully loaded.
Apparently this was never noticed because this method of disabling
module autoloading isn't used much, and also most callers don't use the
return value of request_module() since it's always necessary to check
whether the module registered its functionality or not anyway.
But improperly returning 0 can indeed confuse a few callers, for example
get_fs_type() in fs/filesystems.c where it causes a WARNING to be hit:
if (!fs && (request_module("fs-%.*s", len, name) == 0)) {
fs = __get_fs_type(name, len);
WARN_ONCE(!fs, "request_module fs-%.*s succeeded, but still no fs?\n", len, name);
}
This is easily reproduced with:
echo > /proc/sys/kernel/modprobe
mount -t NONEXISTENT none /
It causes:
request_module fs-NONEXISTENT succeeded, but still no fs?
WARNING: CPU: 1 PID: 1106 at fs/filesystems.c:275 get_fs_type+0xd6/0xf0
[...]
This should actually use pr_warn_once() rather than WARN_ONCE(), since
it's also user-reachable if userspace immediately unloads the module.
Regardless, request_module() should correctly return an error when it
fails. So let's make it return -ENOENT, which matches the error when
the modprobe binary doesn't exist.
I've also sent patches to document and test this case.
Signed-off-by: Eric Biggers <ebiggers@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Reviewed-by: Kees Cook <keescook@chromium.org>
Reviewed-by: Jessica Yu <jeyu@kernel.org>
Acked-by: Luis Chamberlain <mcgrof@kernel.org>
Cc: Alexei Starovoitov <ast@kernel.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Jeff Vander Stoep <jeffv@google.com>
Cc: Ben Hutchings <benh@debian.org>
Cc: Josh Triplett <josh@joshtriplett.org>
Cc: <stable@vger.kernel.org>
Link: http://lkml.kernel.org/r/20200310223731.126894-1-ebiggers@kernel.org
Link: http://lkml.kernel.org/r/20200312202552.241885-1-ebiggers@kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
printk_deferred(), similarly to printk_safe/printk_nmi, does not
immediately attempt to print a new message on the consoles, avoiding
calls into non-reentrant kernel paths, e.g. scheduler or timekeeping,
which potentially can deadlock the system.
Those printk() flavors, instead, rely on per-CPU flush irq_work to print
messages from safer contexts. For same reasons (recursive scheduler or
timekeeping calls) printk() uses per-CPU irq_work in order to wake up
user space syslog/kmsg readers.
However, only printk_safe/printk_nmi do make sure that per-CPU areas
have been initialised and that it's safe to modify per-CPU irq_work.
This means that, for instance, should printk_deferred() be invoked "too
early", that is before per-CPU areas are initialised, printk_deferred()
will perform illegal per-CPU access.
Lech Perczak [0] reports that after commit 1b710b1b10 ("char/random:
silence a lockdep splat with printk()") user-space syslog/kmsg readers
are not able to read new kernel messages.
The reason is printk_deferred() being called too early (as was pointed
out by Petr and John).
Fix printk_deferred() and do not queue per-CPU irq_work before per-CPU
areas are initialized.
Link: https://lore.kernel.org/lkml/aa0732c6-5c4e-8a8b-a1c1-75ebe3dca05b@camlintechnologies.com/
Reported-by: Lech Perczak <l.perczak@camlintechnologies.com>
Signed-off-by: Sergey Senozhatsky <sergey.senozhatsky@gmail.com>
Tested-by: Jann Horn <jannh@google.com>
Reviewed-by: Petr Mladek <pmladek@suse.com>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Cc: Theodore Ts'o <tytso@mit.edu>
Cc: John Ogness <john.ogness@linutronix.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull proc fix from Eric Biederman:
"A brown paper bag slipped through my proc changes, and syzcaller
caught it when the code ended up in your tree.
I have opted to fix it the simplest cleanest way I know how, so there
is no reasonable chance for the bug to repeat"
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/ebiederm/user-namespace:
proc: Use a dedicated lock in struct pid
Pull more power management updates from Rafael Wysocki:
"Rework compat ioctl handling in the user space hibernation interface
(Christoph Hellwig) and fix a typo in a function name in the cpuidle
haltpoll driver (Yihao Wu)"
* tag 'pm-5.7-rc1-3' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
cpuidle-haltpoll: Fix small typo
PM / sleep: handle the compat case in snapshot_set_swap_area()
PM / sleep: move SNAPSHOT_SET_SWAP_AREA handling into a helper
Pull module updates from Jessica Yu:
"Only a small cleanup this time around: a trivial conversion of
zero-length arrays to flexible arrays"
* tag 'modules-for-v5.7' of git://git.kernel.org/pub/scm/linux/kernel/git/jeyu/linux:
kernel: module: Replace zero-length array with flexible-array member
syzbot wrote:
> ========================================================
> WARNING: possible irq lock inversion dependency detected
> 5.6.0-syzkaller #0 Not tainted
> --------------------------------------------------------
> swapper/1/0 just changed the state of lock:
> ffffffff898090d8 (tasklist_lock){.+.?}-{2:2}, at: send_sigurg+0x9f/0x320 fs/fcntl.c:840
> but this lock took another, SOFTIRQ-unsafe lock in the past:
> (&pid->wait_pidfd){+.+.}-{2:2}
>
>
> and interrupts could create inverse lock ordering between them.
>
>
> other info that might help us debug this:
> Possible interrupt unsafe locking scenario:
>
> CPU0 CPU1
> ---- ----
> lock(&pid->wait_pidfd);
> local_irq_disable();
> lock(tasklist_lock);
> lock(&pid->wait_pidfd);
> <Interrupt>
> lock(tasklist_lock);
>
> *** DEADLOCK ***
>
> 4 locks held by swapper/1/0:
The problem is that because wait_pidfd.lock is taken under the tasklist
lock. It must always be taken with irqs disabled as tasklist_lock can be
taken from interrupt context and if wait_pidfd.lock was already taken this
would create a lock order inversion.
Oleg suggested just disabling irqs where I have added extra calls to
wait_pidfd.lock. That should be safe and I think the code will eventually
do that. It was rightly pointed out by Christian that sharing the
wait_pidfd.lock was a premature optimization.
It is also true that my pre-merge window testing was insufficient. So
remove the premature optimization and give struct pid a dedicated lock of
it's own for struct pid things. I have verified that lockdep sees all 3
paths where we take the new pid->lock and lockdep does not complain.
It is my current day dream that one day pid->lock can be used to guard the
task lists as well and then the tasklist_lock won't need to be held to
deliver signals. That will require taking pid->lock with irqs disabled.
Acked-by: Christian Brauner <christian.brauner@ubuntu.com>
Link: https://lore.kernel.org/lkml/00000000000011d66805a25cd73f@google.com/
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Christian Brauner <christian.brauner@ubuntu.com>
Reported-by: syzbot+343f75cdeea091340956@syzkaller.appspotmail.com
Reported-by: syzbot+832aabf700bc3ec920b9@syzkaller.appspotmail.com
Reported-by: syzbot+f675f964019f884dbd0f@syzkaller.appspotmail.com
Reported-by: syzbot+a9fb1457d720a55d6dc5@syzkaller.appspotmail.com
Fixes: 7bc3e6e55a ("proc: Use a list of inodes to flush from proc")
Signed-off-by: "Eric W. Biederman" <ebiederm@xmission.com>
The upper 32-bit physical address gets truncated inadvertently
when dma_direct_get_required_mask() invokes phys_to_dma_direct().
This results in dma_addressing_limited() return incorrect value
when used in platforms with LPAE enabled.
Fix it here by explicitly type casting 'max_pfn' to phys_addr_t
in order to prevent overflow of intermediate value while evaluating
'(max_pfn - 1) << PAGE_SHIFT'.
Signed-off-by: Kishon Vijay Abraham I <kishon@ti.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
The 'invalid wait context' splat doesn't print all the information
required to reconstruct / validate the error, specifically the
irq-context state is missing.
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
The following commit:
7f26482a87 ("locking/percpu-rwsem: Remove the embedded rwsem")
introduced task_struct memory leaks due to messing up the task_struct
refcount.
At the beginning of percpu_rwsem_wake_function(), it calls get_task_struct(),
but if the trylock failed, it will remain in the waitqueue. However, it
will run percpu_rwsem_wake_function() again with get_task_struct() to
increase the refcount but then only call put_task_struct() once the trylock
succeeded.
Fix it by adjusting percpu_rwsem_wake_function() a bit to guard against
when percpu_rwsem_wait() observing !private, terminating the wait and
doing a quick exit() while percpu_rwsem_wake_function() then doing
wake_up_process(p) as a use-after-free.
Fixes: 7f26482a87 ("locking/percpu-rwsem: Remove the embedded rwsem")
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Qian Cai <cai@lca.pw>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200330213002.2374-1-cai@lca.pw
The kernel test robot triggered a warning with the following race:
task-ctx A interrupt-ctx B
worker
-> process_one_work()
-> work_item()
-> schedule();
-> sched_submit_work()
-> wq_worker_sleeping()
-> ->sleeping = 1
atomic_dec_and_test(nr_running)
__schedule(); *interrupt*
async_page_fault()
-> local_irq_enable();
-> schedule();
-> sched_submit_work()
-> wq_worker_sleeping()
-> if (WARN_ON(->sleeping)) return
-> __schedule()
-> sched_update_worker()
-> wq_worker_running()
-> atomic_inc(nr_running);
-> ->sleeping = 0;
-> sched_update_worker()
-> wq_worker_running()
if (!->sleeping) return
In this context the warning is pointless everything is fine.
An interrupt before wq_worker_sleeping() will perform the ->sleeping
assignment (0 -> 1 > 0) twice.
An interrupt after wq_worker_sleeping() will trigger the warning and
nr_running will be decremented (by A) and incremented once (only by B, A
will skip it). This is the case until the ->sleeping is zeroed again in
wq_worker_running().
Remove the WARN statement because this condition may happen. Document
that preemption around wq_worker_sleeping() needs to be disabled to
protect ->sleeping and not just as an optimisation.
Fixes: 6d25be5782 ("sched/core, workqueues: Distangle worker accounting from rq lock")
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Cc: Tejun Heo <tj@kernel.org>
Link: https://lkml.kernel.org/r/20200327074308.GY11705@shao2-debian
Currently, there is a potential race between distribute_cfs_runtime()
and assign_cfs_rq_runtime(). Race happens when cfs_b->runtime is read,
distributes without holding lock and finds out there is not enough
runtime to charge against after distribution. Because
assign_cfs_rq_runtime() might be called during distribution, and use
cfs_b->runtime at the same time.
Fibtest is the tool to test this race. Assume all gcfs_rq is throttled
and cfs period timer runs, slow threads might run and sleep, returning
unused cfs_rq runtime and keeping min_cfs_rq_runtime in their local
pool. If all this happens sufficiently quickly, cfs_b->runtime will drop
a lot. If runtime distributed is large too, over-use of runtime happens.
A runtime over-using by about 70 percent of quota is seen when we
test fibtest on a 96-core machine. We run fibtest with 1 fast thread and
95 slow threads in test group, configure 10ms quota for this group and
see the CPU usage of fibtest is 17.0%, which is far more than the
expected 10%.
On a smaller machine with 32 cores, we also run fibtest with 96
threads. CPU usage is more than 12%, which is also more than expected
10%. This shows that on similar workloads, this race do affect CPU
bandwidth control.
Solve this by holding lock inside distribute_cfs_runtime().
Fixes: c06f04c704 ("sched: Fix potential near-infinite distribute_cfs_runtime() loop")
Reviewed-by: Ben Segall <bsegall@google.com>
Signed-off-by: Huaixin Chang <changhuaixin@linux.alibaba.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lore.kernel.org/lkml/20200325092602.22471-1-changhuaixin@linux.alibaba.com/
sched/core.c uses update_avg() for rq->avg_idle and sched/fair.c uses an
open-coded version (with the exact same decay factor) for
rq->avg_scan_cost. On top of that, select_idle_cpu() expects to be able to
compare these two fields.
The only difference between the two is that rq->avg_scan_cost is computed
using a pure division rather than a shift. Turns out it actually matters,
first of all because the shifted value can be negative, and the standard
has this to say about it:
"""
The result of E1 >> E2 is E1 right-shifted E2 bit positions. [...] If E1
has a signed type and a negative value, the resulting value is
implementation-defined.
"""
Not only this, but (arithmetic) right shifting a negative value (using 2's
complement) is *not* equivalent to dividing it by the corresponding power
of 2. Let's look at a few examples:
-4 -> 0xF..FC
-4 >> 3 -> 0xF..FF == -1 != -4 / 8
-8 -> 0xF..F8
-8 >> 3 -> 0xF..FF == -1 == -8 / 8
-9 -> 0xF..F7
-9 >> 3 -> 0xF..FE == -2 != -9 / 8
Make update_avg() use a division, and export it to the private scheduler
header to reuse it where relevant. Note that this still lets compilers use
a shift here, but should prevent any unwanted surprise. The disassembly of
select_idle_cpu() remains unchanged on arm64, and ttwu_do_wakeup() gains 2
instructions; the diff sort of looks like this:
- sub x1, x1, x0
+ subs x1, x1, x0 // set condition codes
+ add x0, x1, #0x7
+ csel x0, x0, x1, mi // x0 = x1 < 0 ? x0 : x1
add x0, x3, x0, asr #3
which does the right thing (i.e. gives us the expected result while still
using an arithmetic shift)
Signed-off-by: Valentin Schneider <valentin.schneider@arm.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200330090127.16294-1-valentin.schneider@arm.com
We hit following warning when running tests on kernel
compiled with CONFIG_DEBUG_ATOMIC_SLEEP=y:
WARNING: CPU: 19 PID: 4472 at mm/gup.c:2381 __get_user_pages_fast+0x1a4/0x200
CPU: 19 PID: 4472 Comm: dummy Not tainted 5.6.0-rc6+ #3
RIP: 0010:__get_user_pages_fast+0x1a4/0x200
...
Call Trace:
perf_prepare_sample+0xff1/0x1d90
perf_event_output_forward+0xe8/0x210
__perf_event_overflow+0x11a/0x310
__intel_pmu_pebs_event+0x657/0x850
intel_pmu_drain_pebs_nhm+0x7de/0x11d0
handle_pmi_common+0x1b2/0x650
intel_pmu_handle_irq+0x17b/0x370
perf_event_nmi_handler+0x40/0x60
nmi_handle+0x192/0x590
default_do_nmi+0x6d/0x150
do_nmi+0x2f9/0x3c0
nmi+0x8e/0xd7
While __get_user_pages_fast() is IRQ-safe, it calls access_ok(),
which warns on:
WARN_ON_ONCE(!in_task() && !pagefault_disabled())
Peter suggested disabling page faults around __get_user_pages_fast(),
which gets rid of the warning in access_ok() call.
Suggested-by: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200407141427.3184722-1-jolsa@kernel.org
Song reports that installing cgroup events is broken since:
db0503e4f6 ("perf/core: Optimize perf_install_in_event()")
The problem being that cgroup events try to track cpuctx->cgrp even
for disabled events, which is pointless and actively harmful since the
above commit. Rework the code to have explicit enable/disable hooks
for cgroup events, such that we can limit cgroup tracking to active
events.
More specifically, since the above commit disabled events are no
longer added to their context from the 'right' CPU, and we can't
access things like the current cgroup for a remote CPU.
Cc: <stable@vger.kernel.org> # v5.5+
Fixes: db0503e4f6 ("perf/core: Optimize perf_install_in_event()")
Reported-by: Song Liu <songliubraving@fb.com>
Tested-by: Song Liu <songliubraving@fb.com>
Reviewed-by: Song Liu <songliubraving@fb.com>
Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Signed-off-by: Ingo Molnar <mingo@kernel.org>
Link: https://lkml.kernel.org/r/20200318193337.GB20760@hirez.programming.kicks-ass.net
Commit 769071ac9f "ns: Introduce Time Namespace" broke reporting of
inotify ucounts (max_inotify_instances, max_inotify_watches) in
/proc/sys/user because it has added UCOUNT_TIME_NAMESPACES into enum
ucount_type but didn't properly update reporting in
kernel/ucount.c:setup_userns_sysctls(). This problem got fixed in commit
eeec26d5da "time/namespace: Add max_time_namespaces ucount".
Add BUILD_BUG_ON to catch a similar problem in the future.
Signed-off-by: Jan Kara <jack@suse.cz>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Andrei Vagin <avagin@gmail.com>
Link: https://lkml.kernel.org/r/20200407154643.10102-1-jack@suse.cz
The current codebase makes use of the zero-length array language extension
to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning in
case the flexible array does not occur last in the structure, which will
help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by this
change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Oberparleiter <oberpar@linux.ibm.com>
Link: http://lkml.kernel.org/r/20200302224851.GA26467@embeddedor
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The current codebase makes use of the zero-length array language extension
to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning in
case the flexible array does not occur last in the structure, which will
help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by this
change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Peter Oberparleiter <oberpar@linux.ibm.com>
Link: http://lkml.kernel.org/r/20200302224501.GA14175@embeddedor
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The current codebase makes use of the zero-length array language extension
to the C90 standard, but the preferred mechanism to declare
variable-length types such as these ones is a flexible array member[1][2],
introduced in C99:
struct foo {
int stuff;
struct boo array[];
};
By making use of the mechanism above, we will get a compiler warning in
case the flexible array does not occur last in the structure, which will
help us prevent some kind of undefined behavior bugs from being
inadvertently introduced[3] to the codebase from now on.
Also, notice that, dynamic memory allocations won't be affected by this
change:
"Flexible array members have incomplete type, and so the sizeof operator
may not be applied. As a quirk of the original implementation of
zero-length arrays, sizeof evaluates to zero."[1]
This issue was found with the help of Coccinelle.
[1] https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
[2] https://github.com/KSPP/linux/issues/21
[3] commit 7649773293 ("cxgb3/l2t: Fix undefined behaviour")
Signed-off-by: Gustavo A. R. Silva <gustavo@embeddedor.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Acked-by: Peter Oberparleiter <oberpar@linux.ibm.com>
Link: http://lkml.kernel.org/r/20200213152241.GA877@embeddedor
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Now that "struct proc_ops" exist we can start putting there stuff which
could not fly with VFS "struct file_operations"...
Most of fs/proc/inode.c file is dedicated to make open/read/.../close
reliable in the event of disappearing /proc entries which usually happens
if module is getting removed. Files like /proc/cpuinfo which never
disappear simply do not need such protection.
Save 2 atomic ops, 1 allocation, 1 free per open/read/close sequence for such
"permanent" files.
Enable "permanent" flag for
/proc/cpuinfo
/proc/kmsg
/proc/modules
/proc/slabinfo
/proc/stat
/proc/sysvipc/*
/proc/swaps
More will come once I figure out foolproof way to prevent out module
authors from marking their stuff "permanent" for performance reasons
when it is not.
This should help with scalability: benchmark is "read /proc/cpuinfo R times
by N threads scattered over the system".
N R t, s (before) t, s (after)
-----------------------------------------------------
64 4096 1.582458 1.530502 -3.2%
256 4096 6.371926 6.125168 -3.9%
1024 4096 25.64888 24.47528 -4.6%
Benchmark source:
#include <chrono>
#include <iostream>
#include <thread>
#include <vector>
#include <sys/types.h>
#include <sys/stat.h>
#include <fcntl.h>
#include <unistd.h>
const int NR_CPUS = sysconf(_SC_NPROCESSORS_ONLN);
int N;
const char *filename;
int R;
int xxx = 0;
int glue(int n)
{
cpu_set_t m;
CPU_ZERO(&m);
CPU_SET(n, &m);
return sched_setaffinity(0, sizeof(cpu_set_t), &m);
}
void f(int n)
{
glue(n % NR_CPUS);
while (*(volatile int *)&xxx == 0) {
}
for (int i = 0; i < R; i++) {
int fd = open(filename, O_RDONLY);
char buf[4096];
ssize_t rv = read(fd, buf, sizeof(buf));
asm volatile ("" :: "g" (rv));
close(fd);
}
}
int main(int argc, char *argv[])
{
if (argc < 4) {
std::cerr << "usage: " << argv[0] << ' ' << "N /proc/filename R
";
return 1;
}
N = atoi(argv[1]);
filename = argv[2];
R = atoi(argv[3]);
for (int i = 0; i < NR_CPUS; i++) {
if (glue(i) == 0)
break;
}
std::vector<std::thread> T;
T.reserve(N);
for (int i = 0; i < N; i++) {
T.emplace_back(f, i);
}
auto t0 = std::chrono::system_clock::now();
{
*(volatile int *)&xxx = 1;
for (auto& t: T) {
t.join();
}
}
auto t1 = std::chrono::system_clock::now();
std::chrono::duration<double> dt = t1 - t0;
std::cout << dt.count() << '
';
return 0;
}
P.S.:
Explicit randomization marker is added because adding non-function pointer
will silently disable structure layout randomization.
[akpm@linux-foundation.org: coding style fixes]
Reported-by: kbuild test robot <lkp@intel.com>
Reported-by: Dan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Joe Perches <joe@perches.com>
Link: http://lkml.kernel.org/r/20200222201539.GA22576@avx2
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
