fix task_work interation with freezing
v2 cover fix task_work interation with freezing
It's reported [1] that a task_work queued at a wrong time can prevent freezing and make the tasks to spin in get_signal() taking 100% of CPU. Patch 1 is a preparation. Patch 2 addresses the issue.
[1] https://github.com/systemd/systemd/issues/33626
v2: move task_work_run() into do_freezer_trap()
change Fixes tag
Pavel Begunkov (2):
io_uring/io-wq: limit retrying worker initialisation
kernel: rerun task_work while freezing in get_signal()
io_uring/io-wq.c | 10 +++++++--- kernel/signal.c | 8 ++++++++ 2 files changed, 15 insertions(+), 3 deletions(-)
Signature · 1 line
2.44.0
PATCH v2 1/2 io_uring/io-wq: limit retrying worker initialisation
If io-wq worker creation fails, we retry it by queueing up a task_work. tasK_work is needed because it should be done from the user process context. The problem is that retries are not limited, and if queueing a task_work is the reason for the failure, we might get into an infinite loop.
It doesn't seem to happen now but it will with the following patch executing task_work in the freezing loop. For now, arbitrarily limit the number of attempts to create a worker.
--- io_uring/io-wq.c | 10 +++++++--- 1 file changed, 7 insertions(+), 3 deletions(-)
io_uring/io-wq.c +7 -3
@@ -23,6 +23,7 @@#include "io_uring.h" #define WORKER_IDLE_TIMEOUT (5 * HZ) +#define WORKER_INIT_LIMIT 3enum {IO_WORKER_F_UP =0 ,/* up and active */
@@ -58,6 +59,7 @@ struct io_worker {unsigned long create_state;struct callback_head create_work; +int init_retries;union {struct rcu_head rcu;
@@ -745,7 +747,7 @@ static bool io_wq_work_match_all(struct io_wq_work *work, void *data)return true; } -static inline bool io_should_retry_thread (long err) +static inline bool io_should_retry_thread (struct io_worker *worker,long err) {/* * Prevent perpetual task_work retry, if the task (or its group) is
@@ -753,6 +755,8 @@ static inline bool io_should_retry_thread(long err)*/ if (fatal_signal_pending (current))return false; +if (worker->init_retries++ >=WORKER_INIT_LIMIT ) +return false;switch (err) {case -EAGAIN :
@@ -779,7 +783,7 @@ static void create_worker_cont(struct callback_head *cb)io_init_new_worker (wq, worker, tsk);io_worker_release (worker);return ; - }else if (!io_should_retry_thread (PTR_ERR (tsk))) { + }else if (!io_should_retry_thread (worker,PTR_ERR (tsk))) {struct io_wq_acct *acct =io_wq_get_acct (worker);atomic_dec (&acct->nr_running);
@@ -846,7 +850,7 @@ static bool create_io_worker(struct io_wq *wq, int index) tsk =create_io_thread (io_wq_worker, worker,NUMA_NO_NODE );if (!IS_ERR (tsk)) {io_init_new_worker (wq, worker, tsk); - }else if (!io_should_retry_thread (PTR_ERR (tsk))) { + }else if (!io_should_retry_thread (worker,PTR_ERR (tsk))) {kfree (worker); goto fail; }else {
Signature · 1 line
2.44.0
PATCH v2 2/2 kernel: rerun task_work while freezing in get_signal()
In reply to Pavel Begunkovio_uring can asynchronously add a task_work while the task is getting freezed. TIF_NOTIFY_SIGNAL will prevent the task from sleeping in do_freezer_trap(), and since the get_signal()'s relock loop doesn't retry task_work, the task will spin there not being able to sleep until the freezing is cancelled / the task is killed / etc.
--- kernel/signal.c | 8 ++++++++ 1 file changed, 8 insertions(+)
kernel/signal.c +8 -0
@@ -2600,6 +2600,14 @@ static void do_freezer_trap(void)spin_unlock_irq (¤t->sighand->siglock);cgroup_enter_frozen ();schedule (); + +/* +* We could've been woken by task_work, run it to clear +* TIF_NOTIFY_SIGNAL. The caller will retry if necessary. +*/ +clear_notify_signal (); +if (unlikely (task_work_pending (current))) +task_work_run (); }static int ptrace_signal (int signr,kernel_siginfo_t *info,enum pid_type type)
Signature · 1 line
2.44.0
Re: [PATCH v2 2/2] kernel: rerun task_work while freezing in get_signal()
Quoted 6 lines · Pavel Begunkov
kernel/signal.c +0
-0@@ -2600,6 +2600,14 @@ static void do_freezer_trap(void)spin_unlock_irq (¤t->sighand->siglock);cgroup_enter_frozen ();
kernel/signal.c static void do_freezer_trap(void) · line 2611
Quoted 10 lines
kernel/signal.c +8
-0schedule (); + +/* +* We could've been woken by task_work, run it to clear +* TIF_NOTIFY_SIGNAL. The caller will retry if necessary. +*/ +clear_notify_signal (); +if (unlikely (task_work_pending (current))) +task_work_run (); }
Re: [PATCH v2 2/2] kernel: rerun task_work while freezing in get_signal()
In reply to Pavel BegunkovQuoted 11 lines · Pavel Begunkov
io_uringcan asynchronously add atask_workwhile the task is getting freezed.TIF_NOTIFY_SIGNALwill prevent the task from sleeping indo_freezer_trap(), and since the get_signal()'s relock loop doesn't retrytask_work, the task will spin there not being able to sleep until the freezing is cancelled / the task is killed / etc.Fixes 12db8b690010c ("entry: Add support for TIF_NOTIFY_SIGNAL")Reported-by Julian Orth <[email protected]>Signed-off-by Pavel Begunkov <[email protected]>
I haven't looked at the signal code for too long to be all that useful but the problem described and the patch does make sense to me. FWIW,
Maybe note that this is structured specifically to ease backport and we need further cleanups? It's not great that this is special cased in do_freezer_trap() instead of being integrated into the outer loop.
Thanks.
Signature · 1 line
tejun
Re: [PATCH v2 2/2] kernel: rerun task_work while freezing in get_signal()
Quoted 13 lines · Tejun Heo
Quoted 11 lines · Pavel Begunkov
io_uringcan asynchronously add atask_workwhile the task is getting freezed.TIF_NOTIFY_SIGNALwill prevent the task from sleeping indo_freezer_trap(), and since the get_signal()'s relock loop doesn't retrytask_work, the task will spin there not being able to sleep until the freezing is cancelled / the task is killed / etc.Fixes 12db8b690010c ("entry: Add support for TIF_NOTIFY_SIGNAL")Reported-by Julian Orth <[email protected]>Signed-off-by Pavel Begunkov <[email protected]>
Quoted 7 lines
I haven't looked at the signal code for too long to be all that useful but the problem described and the patch does make sense to me. FWIW,
Acked-by Tejun Heo <[email protected]>Maybe note that this is structured specifically to ease backport and we need further cleanups? It's not great that this is special cased in
I'll add a couple of words
Quoted 1 line
do_freezer_trap()instead of being integrated into the outer loop.
v1 had it in the common loop, but might actually be good it's limited to freezing, need more digging.
Signature · 1 line
Pavel Begunkov