Skip to contentSkip to conversation

SIGCHLD signal sometimes sent with si_pid==0 (Linux 5.6.5)

PATCH In review Christof Meerwald 4 participants 22 messages Apr 19, 2020 – Apr 24, 2020

lore
Share & tools
lore
#0 lore raw

Hi,

this is probably related to commit 7a0cf094944e2540758b7f957eb6846d5126f535 (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?)

Christof

Signature · 2 lines
http://cmeerw.org                              sip:cmeerw at cmeerw.org
mailto:cmeerw at cmeerw.org                   xmpp:cmeerw at cmeerw.org
#1 lore raw

PATCH 1/1 signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 83 lines · Christof Meerwald

Hi,

this is probably related to commit 7a0cf094944e2540758b7f957eb6846d5126f535 (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

Quoted 8 lines

/* 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:

Quoted 21 lines
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
Quoted 8 lines
 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 7a0cf094944e ("signal: Correct namespace fixups of si_pid and si_uid") where the problem logic was moved out of __send_signal and into send_signal.

Fixes 6588c1e3ff01 ("signals: SI_USER: Masquerade si_pid when crossing pid ns boundary

Ref: 921cf9f63089 ("signals: protect cinit from unblocked SIG_DFL signals")

---

Unless someone has an objection I will apply this one and send it to Linus.

 kernel/signal.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
kernel/signal.c +5 -1
index 9899c5f91ee1..a88a89422227 100644
@@ -1993,8 +1993,12 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = 0;
 	}
+	/*
+	 * Bypass send_signal as the si_pid and si_uid values have
+	 * been generated in the parent's namespaces.
+	 */
 	if (valid_signal(sig) && sig)
-		__group_send_sig_info(sig, &info, tsk->parent);
+		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
 	__wake_up_parent(tsk, tsk->parent);
 	spin_unlock_irqrestore(&psig->siglock, flags);
 
Signature · 1 line
2.20.1
#2 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 160 lines · Eric W. Biederman
Quoted 83 lines · Christof Meerwald

Hi,

this is probably related to commit 7a0cf094944e2540758b7f957eb6846d5126f535 (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

Quoted 8 lines

/* 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:

Quoted 21 lines
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
Quoted 8 lines
 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 7a0cf094944e ("signal: Correct namespace fixups of si_pid and si_uid") where the problem logic was moved out of __send_signal and into send_signal.

Fixes 6588c1e3ff01 ("signals: SI_USER: Masquerade si_pid when crossing pid ns boundary

Ref: 921cf9f63089 ("signals: protect cinit from unblocked SIG_DFL signals")

---

Unless someone has an objection I will apply this one and send it to Linus.

 kernel/signal.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
kernel/signal.c +0 -0
index 9899c5f91ee1..a88a89422227 100644

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 1999

Quoted 7 lines
kernel/signal.c +4 -0
@@ -1993,8 +1993,12 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = 0;
 	}
+	/*
+	 * Bypass send_signal as the si_pid and si_uid values have
+	 * been generated in the parent's namespaces.
+	 */

At first I misread that comment as saying that we're skipping sending a signal not that it relates to a specific function (and I won't admit that I wrote a whole long paragraph on why I'm confused we're skipping sending signals on invalid si_pid and si_uid...).

I think it would be worth to say something that simply states the facts such as:
"If we're not autoreaping, send a signal with si_pid and si_uid set
as generated in the parent's namespaces."
which imho is way clearer then pointing to out that we're skipping send_signal(). The logic here and the comment in its current form are hard to correlate, especially since send_signal() was never called directly here but is rather called by __group_send_sig_info(). The details of why we switched from __group_send_sign_info() to __send_signal() could just go into the commit message. It's more confusing in the code.

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 2000

Quoted 1 line
kernel/signal.c +0 -0
 	if (valid_signal(sig) && sig)

(Not related to you patch but odd ordering of the if-branch here. I would've expected this to read

if (sig && valid_signal(sig))

especially since "sig" can be set to 0 right before.

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 2001

Quoted 2 lines
kernel/signal.c +1 -1
-		__group_send_sig_info(sig, &info, tsk->parent);
+		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);

So below you switch to __send_signal() but set the "force" argument to to "false". Before that, if the signal was generated from another pid namespace and we fixed up si_pid and si_uid the "force" argument was set to "true", i.e. __send_signal(..., ..., ..., ..., true) Even though from reading through the code I think that change is safe to make let me verify that this was an intentional change?

Thanks!
Christian

#3 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 6 lines · Christian Brauner

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 2001

Quoted 3 lines
kernel/signal.c +1 -1
-		__group_send_sig_info(sig, &info, tsk->parent);
+		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);

So below you switch to __send_signal() but set the "force" argument to to "false".

it must be false, the signal is generated from the parent's namespace or its descendant

Quoted 3 lines

Before that, if the signal was generated from another pid namespace and we fixed up si_pid and si_uid the "force" argument was set to "true",

before that the "force" argument could be falsely true by the same reason, task_pid_nr_ns(tsk, tsk->parent) can return 0 because "tsk" no longer have pids after __unhash_process().

Oleg.

#4 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 11 lines · Oleg Nesterov
Quoted 6 lines · Christian Brauner

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 2001

Quoted 3 lines
kernel/signal.c +1 -1
-		__group_send_sig_info(sig, &info, tsk->parent);
+		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);

So below you switch to __send_signal() but set the "force" argument to to "false".

it must be false, the signal is generated from the parent's namespace or its descendant

Quoted 7 lines
Quoted 3 lines

Before that, if the signal was generated from another pid namespace and we fixed up si_pid and si_uid the "force" argument was set to "true",

before that the "force" argument could be falsely true by the same reason, task_pid_nr_ns(tsk, tsk->parent) can return 0 because "tsk" no longer have pids after __unhash_process().

As I said in my mail, looking at the codepath it seems safe. My question was whether it is _always_ incorrectly false which I do think it is because child subreapers can't come from outside the pid namespace. If they could you could create a scenario where the signal is generated from a sibling pid namespace in which case it would be correctly set to true.

Christian

#5 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Sorry Christian, I don't understand...

Quoted 17 lines · Christian Brauner
Quoted 15 lines · Oleg Nesterov
Quoted 6 lines · Christian Brauner

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 2001

Quoted 3 lines
kernel/signal.c +1 -1
-		__group_send_sig_info(sig, &info, tsk->parent);
+		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);

So below you switch to __send_signal() but set the "force" argument to to "false".

it must be false, the signal is generated from the parent's namespace or its descendant

Quoted 3 lines

Before that, if the signal was generated from another pid namespace and we fixed up si_pid and si_uid the "force" argument was set to "true",

Quoted 6 lines
Quoted 3 lines

before that the "force" argument could be falsely true by the same reason, task_pid_nr_ns(tsk, tsk->parent) can return 0 because "tsk" no longer have pids after __unhash_process().

As I said in my mail, looking at the codepath it seems safe. My question was whether it is _always_ incorrectly false which I do think it is

Again, it must be always "false", it can be incorrectly "true" and this too is fixed by Eric's patch.

Quoted 4 lines

because child subreapers can't come from outside the pid namespace. If they could you could create a scenario where the signal is generated from a sibling pid namespace in which case it would be correctly set to true.

not sure I understand, but probably the answer is "yes"...

task and task->parent either live in the same namespace or the child's namespace is the descendant of task->parent's namespace. In both cases task_pid_nr_ns(tsk, tsk->parent) should return the valid pid_nr and "force" must be false.

The corner case is release_task() when the last exiting sub-thread sends a signal on behalf of its ->group_leader, and at this point all the tsk's pid pointers are NULL, that is why "force" can be falsely "true".

Oleg.

#6 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 1 line · Oleg Nesterov

Sorry Christian, I don't understand...

In my original mail, it was really just a clarification question. I said the patch is correct from looking at the codepaths. :) I was just trying to see whether there was a potential corner-case we're missing where "force" could be _validly_ true.

Quoted 7 lines
Quoted 5 lines

because child subreapers can't come from outside the pid namespace. If they could you could create a scenario where the signal is generated from a sibling pid namespace in which case it would be correctly set to true.

not sure I understand, but probably the answer is "yes"...

(This is really purely academic now since it isn't possible, but for pure amusement assume that a child subreaper could cross namespace boundaries (which they don't). A marks itself as a subreaper and creates a new process B in a new pid namespace <pidnsB>, process B setnses into <pidnsC> which is a sibling pid namespace, B clones a new proces in <pidnsC> which is now a full member of <pidnsC>, B dies and C is reparented to A, B exits and then you'd be getting a sigchld from a pid in a pid namespace in which you have no pid nr.)

#7 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 3 lines · Christian Brauner

process B setnses into
<pidnsC> which is a sibling pid namespace,

please see pidns_install(), it verifies that

* Only allow entering the current active pid namespace
* or a child of the current active pid namespace.

Oleg.

#8 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 9 lines · Oleg Nesterov
Quoted 3 lines · Christian Brauner

process B setnses into
<pidnsC> which is a sibling pid namespace,

please see pidns_install(), it verifies that

* Only allow entering the current active pid namespace
* or a child of the current active pid namespace.

I forgot about that.

Though, don't we have the same problem in:

static void do_notify_parent_cldstop(struct task_struct *tsk,
bool for_ptracer, int why)

at least for the for_ptrace is false case?

Christian

#9 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 11 lines · Christian Brauner
Quoted 9 lines · Oleg Nesterov
Quoted 3 lines · Christian Brauner

process B setnses into
<pidnsC> which is a sibling pid namespace,

please see pidns_install(), it verifies that

* Only allow entering the current active pid namespace
* or a child of the current active pid namespace.

Quoted 8 lines

I forgot about that.

Though, don't we have the same problem in:

static void do_notify_parent_cldstop(struct task_struct *tsk,
bool for_ptracer, int why)

at least for the for_ptrace is false case?

The same problem does not exist with do_notify_parent_cldstop because do_notify_parent_cldstop is always called from current (there is one case of current->group_leader but that is close enough calculations made against current are true).

However because do_notify_parent_cldstop calculates si_pid and si_uid of the target parent process I think we can still get the wrong si_uid.

So it probably makes sense to generalize the fixup code in send_signal and make do_notify_parent_cldstop just generate ids in the current namespace and let the fixup code do it's job.

Eric

#10 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

In reply to Oleg Nesterov
Quoted 4 lines · Oleg Nesterov

The corner case is release_task() when the last exiting sub-thread sends a signal on behalf of its ->group_leader, and at this point all the tsk's pid pointers are NULL, that is why "force" can be falsely "true".

Or do_notify_parent() can be called by debugger from the parent namespace, in this case "force" can be falsely "true" too.

Oleg.

#11 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 8 lines · Oleg Nesterov
Quoted 4 lines · Oleg Nesterov

The corner case is release_task() when the last exiting sub-thread sends a signal on behalf of its ->group_leader, and at this point all the tsk's pid pointers are NULL, that is why "force" can be falsely "true".

Or do_notify_parent() can be called by debugger from the parent namespace, in this case "force" can be falsely "true" too.

That's an interesting scenario to think about as well. Cross-pid-namespace interactions are fun... That's why the cross-pid-namespace-signal sending aspects we discussed a while back on-list though pretty nice to have at some point are somewhat scary.

Christian

#12 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

In reply to Christian Brauner
Quoted 174 lines · Christian Brauner
Quoted 160 lines · Eric W. Biederman
Quoted 83 lines · Christof Meerwald

Hi,

this is probably related to commit 7a0cf094944e2540758b7f957eb6846d5126f535 (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

Quoted 8 lines

/* 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:

Quoted 21 lines
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
Quoted 8 lines
 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 7a0cf094944e ("signal: Correct namespace fixups of si_pid and si_uid") where the problem logic was moved out of __send_signal and into send_signal.

Fixes 6588c1e3ff01 ("signals: SI_USER: Masquerade si_pid when crossing pid ns boundary

Ref: 921cf9f63089 ("signals: protect cinit from unblocked SIG_DFL signals")

---

Unless someone has an objection I will apply this one and send it to Linus.

 kernel/signal.c | 6 +++++-
 1 file changed, 5 insertions(+), 1 deletion(-)
kernel/signal.c +0 -0
index 9899c5f91ee1..a88a89422227 100644

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 1999

Quoted 7 lines
kernel/signal.c +4 -0
@@ -1993,8 +1993,12 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = 0;
 	}
+	/*
+	 * Bypass send_signal as the si_pid and si_uid values have
+	 * been generated in the parent's namespaces.
+	 */

At first I misread that comment as saying that we're skipping sending a signal not that it relates to a specific function (and I won't admit that I wrote a whole long paragraph on why I'm confused we're skipping sending signals on invalid si_pid and si_uid...).

Quoted 11 lines

I think it would be worth to say something that simply states the facts such as:
"If we're not autoreaping, send a signal with si_pid and si_uid set
as generated in the parent's namespaces."
which imho is way clearer then pointing to out that we're skipping send_signal(). The logic here and the comment in its current form are hard to correlate, especially since send_signal() was never called directly here but is rather called by __group_send_sig_info(). The details of why we switched from __group_send_sign_info() to __send_signal() could just go into the commit message. It's more confusing in the code.

Forgot before, otherwise:

Thanks!
Christian

#13 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

In reply to Christian Brauner
Quoted 10 lines · Christian Brauner
Quoted 9 lines · Eric W. Biederman
kernel/signal.c +1 -0
index 9899c5f91ee1..a88a89422227 100644
@@ -1993,8 +1993,12 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = 0;
 	}
+	/*
Quoted 8 lines

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 1999

Quoted 3 lines
kernel/signal.c +3 -0
+	 * Bypass send_signal as the si_pid and si_uid values have
+	 * been generated in the parent's namespaces.
+	 */

At first I misread that comment as saying that we're skipping sending a signal not that it relates to a specific function (and I won't admit that I wrote a whole long paragraph on why I'm confused we're skipping sending signals on invalid si_pid and si_uid...).

I have updated the comment to read:
+ /*

+        * Send with __send_signal as si_pid and si_uid are in the
+        * parent's namespaces.
+        */

That should be enough of a hint for someone to read the code and figure out what is going on.

Eric

#14 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

Quoted 27 lines · Eric W. Biederman
Quoted 10 lines · Christian Brauner
Quoted 9 lines · Eric W. Biederman
kernel/signal.c +1 -0
index 9899c5f91ee1..a88a89422227 100644
@@ -1993,8 +1993,12 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = 0;
 	}
+	/*
Quoted 8 lines

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 1999

Quoted 3 lines
kernel/signal.c +3 -0
+	 * Bypass send_signal as the si_pid and si_uid values have
+	 * been generated in the parent's namespaces.
+	 */

At first I misread that comment as saying that we're skipping sending a signal not that it relates to a specific function (and I won't admit that I wrote a whole long paragraph on why I'm confused we're skipping sending signals on invalid si_pid and si_uid...).

I have updated the comment to read:

kernel/signal.c +4 -0
+       /*
+        * Send with __send_signal as si_pid and si_uid are in the
+        * parent's namespaces.
+        */
Quoted 2 lines

That should be enough of a hint for someone to read the code and figure out what is going on.

Perfect, thanks!

Christian

Acked-by on v1 1/1
#15 lore raw

Re: [PATCH] signal: Avoid corrupting si_pid and si_uid in do_notify_parent

In reply to Eric W. Biederman
Quoted 2 lines · Eric W. Biederman

The immediate problem is as Christof noticed that "pid_alive(current) == false".

this is slightly offtopic, but we can probably remove this "pid_alive" check, pid_nr_ns() checks pid != NULL anyway.

Quoted 2 lines

Inspecting the code it appears this problem has existed since the pid namespace support started handling this case

Agreed...

kernel/signal.c bool do_notify_parent(struct task_struct *tsk, int sig) · line 2001

Quoted 11 lines
kernel/signal.c +5 -1
@@ -1993,8 +1993,12 @@ bool do_notify_parent(struct task_struct *tsk, int sig)
 		if (psig->action[SIGCHLD-1].sa.sa_handler == SIG_IGN)
 			sig = 0;
 	}
+	/*
+	 * Bypass send_signal as the si_pid and si_uid values have
+	 * been generated in the parent's namespaces.
+	 */
 	if (valid_signal(sig) && sig)
-		__group_send_sig_info(sig, &info, tsk->parent);
+		__send_signal(sig, &info, tsk->parent, PIDTYPE_TGID, false);
#16 lore raw

PATCH 1/1 remove the no longer needed pid_alive() check in __task_pid_nr_ns()

Starting from 2c4704756cab ("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.

---
 kernel/pid.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
kernel/pid.c +1 -2
index bc21c0fb26d8..47221db038e2 100644
@@ -475,8 +475,7 @@ pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
 	rcu_read_lock();
 	if (!ns)
 		ns = task_active_pid_ns(current);
-	if (likely(pid_alive(task)))
-		nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
+	nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
 	rcu_read_unlock();
 
 	return nr;
Signature · 1 line
2.25.1.362.g51ebf55
Acked-by on v1 1/1
#17 lore raw

Re: [PATCH] remove the no longer needed pid_alive() check in __task_pid_nr_ns()

Quoted 9 lines · Oleg Nesterov

Starting from 2c4704756cab ("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.

---

Reviewed-by on v1 1/1
#18 lore raw

Re: [PATCH] remove the no longer needed pid_alive() check in __task_pid_nr_ns()

In reply to Oleg Nesterov
Quoted 8 lines · Oleg Nesterov

Starting from 2c4704756cab ("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.

Quoted 1 line

---

Good catch that does simplify things.

Eric

Quoted 17 lines
 kernel/pid.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
kernel/pid.c +1 -2
index bc21c0fb26d8..47221db038e2 100644
@@ -475,8 +475,7 @@ pid_t __task_pid_nr_ns(struct task_struct *task, enum pid_type type,
 	rcu_read_lock();
 	if (!ns)
 		ns = task_active_pid_ns(current);
-	if (likely(pid_alive(task)))
-		nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
+	nr = pid_nr_ns(rcu_dereference(*task_pid_ptr(task, type)), ns);
 	rcu_read_unlock();
 
 	return nr;
#19 lore raw

Re: [PATCH] remove the no longer needed pid_alive() check in __task_pid_nr_ns()

Quoted 7 lines · Eric W. Biederman
Quoted 4 lines · Oleg Nesterov

Starting from 2c4704756cab ("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.

Quoted 5 lines
Quoted 4 lines

pid_nr_ns() has to check pid != NULL anyway, pid_alive() just adds the unnecessary confusion.

Thanks, can you take this patch?

Oleg.

#20 lore raw

Re: [PATCH] remove the no longer needed pid_alive() check in __task_pid_nr_ns()

Quoted 8 lines · Oleg Nesterov
Quoted 7 lines · Eric W. Biederman
Quoted 4 lines · Oleg Nesterov

Starting from 2c4704756cab ("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.

Quoted 7 lines
Quoted 5 lines
Quoted 4 lines

pid_nr_ns() has to check pid != NULL anyway, pid_alive() just adds the unnecessary confusion.

Thanks, can you take this patch?

I plan to. Probably on a topic branch for signals. I am sorting that out now.

Eric

#21 lore raw
In reply to Christof Meerwald
Quoted 9 lines · Christof Meerwald

Hi,

this is probably related to commit 7a0cf094944e2540758b7f957eb6846d5126f535 (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.

Christof I want to say very good spotting and reporting of this issue.

Thank you.
Eric