[PATCH v3 1/2] perf: Fix required permissions if sigtrap is requested

Subsystems: performance events subsystem, the rest

STALE1889d LANDED

Landed in mainline as 9d7a6c95f62b on 2021-07-16.

6 messages, 3 authors, 2021-07-13 · open the first message on its own page

[PATCH v3 1/2] perf: Fix required permissions if sigtrap is requested

From: Marco Elver <elver@google.com>
Date: 2021-07-05 08:45:09

If perf_event_open() is called with another task as target and
perf_event_attr::sigtrap is set, and the target task's user does not
match the calling user, also require the CAP_KILL capability or
PTRACE_MODE_ATTACH permissions.

Otherwise, with the CAP_PERFMON capability alone it would be possible
for a user to send SIGTRAP signals via perf events to another user's
tasks. This could potentially result in those tasks being terminated if
they cannot handle SIGTRAP signals.

Note: The check complements the existing capability check, but is not
supposed to supersede the ptrace_may_access() check. At a high level we
now have:

	capable of CAP_PERFMON and (CAP_KILL if sigtrap)
		OR
	ptrace_may_access(...) // also checks for same thread-group and uid

Fixes: 97ba62b27867 ("perf: Add support for SIGTRAP on perf events")
Cc: <redacted> # 5.13+
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Marco Elver <elver@google.com>
---
v3:
* Upgrade ptrace mode check to ATTACH if attr.sigtrap, otherwise it's
  possible to change the target task (send signal) even if only read
  ptrace permissions were granted (reported by Eric W. Biederman).

v2: https://lkml.kernel.org/r/20210701083842.580466-1-elver@google.com
* Drop kill_capable() and just check CAP_KILL (reported by Ondrej Mosnacek).
* Use ns_capable(__task_cred(task)->user_ns, CAP_KILL) to check for
  capability in target task's ns (reported by Ondrej Mosnacek).

v1: https://lkml.kernel.org/r/20210630093709.3612997-1-elver@google.com
---
 kernel/events/core.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index fe88d6eea3c2..f79ee82e644a 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -12152,10 +12152,33 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (task) {
+		unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
+		bool is_capable;
+
 		err = down_read_interruptible(&task->signal->exec_update_lock);
 		if (err)
 			goto err_file;
 
+		is_capable = perfmon_capable();
+		if (attr.sigtrap) {
+			/*
+			 * perf_event_attr::sigtrap sends signals to the other
+			 * task. Require the current task to also have
+			 * CAP_KILL.
+			 */
+			rcu_read_lock();
+			is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
+			rcu_read_unlock();
+
+			/*
+			 * If the required capabilities aren't available, checks
+			 * for ptrace permissions: upgrade to ATTACH, since
+			 * sending signals can effectively change the target
+			 * task.
+			 */
+			ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
+		}
+
 		/*
 		 * Preserve ptrace permission check for backwards compatibility.
 		 *
@@ -12165,7 +12188,7 @@ SYSCALL_DEFINE5(perf_event_open,
 		 * perf_event_exit_task() that could imply).
 		 */
 		err = -EACCES;
-		if (!perfmon_capable() && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
+		if (!is_capable && !ptrace_may_access(task, ptrace_mode))
 			goto err_cred;
 	}
 
-- 
2.32.0.93.g670b81a890-goog

[PATCH v3 2/2] perf: Refactor permissions check into perf_check_permission()

From: Marco Elver <elver@google.com>
Date: 2021-07-05 08:45:13

Refactor the permission check in perf_event_open() into a helper
perf_check_permission(). This makes the permission check logic more
readable (because we no longer have a negated disjunction). Add a
comment mentioning the ptrace check also checks the uid.

No functional change intended.

Signed-off-by: Marco Elver <elver@google.com>
---
v3:
* Introduce this patch to refactor the permissions checking logic to
  make it more readable (reported by Eric W. Biederman).
---
 kernel/events/core.c | 58 ++++++++++++++++++++++++--------------------
 1 file changed, 32 insertions(+), 26 deletions(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f79ee82e644a..3008b986994b 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -11912,6 +11912,37 @@ __perf_event_ctx_lock_double(struct perf_event *group_leader,
 	return gctx;
 }
 
+static bool
+perf_check_permission(struct perf_event_attr *attr, struct task_struct *task)
+{
+	unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
+	bool is_capable = perfmon_capable();
+
+	if (attr->sigtrap) {
+		/*
+		 * perf_event_attr::sigtrap sends signals to the other task.
+		 * Require the current task to also have CAP_KILL.
+		 */
+		rcu_read_lock();
+		is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
+		rcu_read_unlock();
+
+		/*
+		 * If the required capabilities aren't available, checks for
+		 * ptrace permissions: upgrade to ATTACH, since sending signals
+		 * can effectively change the target task.
+		 */
+		ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
+	}
+
+	/*
+	 * Preserve ptrace permission check for backwards compatibility. The
+	 * ptrace check also includes checks that the current task and other
+	 * task have matching uids, and is therefore not done here explicitly.
+	 */
+	return is_capable || ptrace_may_access(task, ptrace_mode);
+}
+
 /**
  * sys_perf_event_open - open a performance event, associate it to a task/cpu
  *
@@ -12152,43 +12183,18 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (task) {
-		unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
-		bool is_capable;
-
 		err = down_read_interruptible(&task->signal->exec_update_lock);
 		if (err)
 			goto err_file;
 
-		is_capable = perfmon_capable();
-		if (attr.sigtrap) {
-			/*
-			 * perf_event_attr::sigtrap sends signals to the other
-			 * task. Require the current task to also have
-			 * CAP_KILL.
-			 */
-			rcu_read_lock();
-			is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
-			rcu_read_unlock();
-
-			/*
-			 * If the required capabilities aren't available, checks
-			 * for ptrace permissions: upgrade to ATTACH, since
-			 * sending signals can effectively change the target
-			 * task.
-			 */
-			ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
-		}
-
 		/*
-		 * Preserve ptrace permission check for backwards compatibility.
-		 *
 		 * We must hold exec_update_lock across this and any potential
 		 * perf_install_in_context() call for this new event to
 		 * serialize against exec() altering our credentials (and the
 		 * perf_event_exit_task() that could imply).
 		 */
 		err = -EACCES;
-		if (!is_capable && !ptrace_may_access(task, ptrace_mode))
+		if (!perf_check_permission(&attr, task))
 			goto err_cred;
 	}
 
-- 
2.32.0.93.g670b81a890-goog

Re: [PATCH v3 1/2] perf: Fix required permissions if sigtrap is requested

From: Dmitry Vyukov <dvyukov@google.com>
Date: 2021-07-06 06:16:26

On Mon, Jul 5, 2021 at 10:45 AM Marco Elver [off-list ref] wrote:
If perf_event_open() is called with another task as target and
perf_event_attr::sigtrap is set, and the target task's user does not
match the calling user, also require the CAP_KILL capability or
PTRACE_MODE_ATTACH permissions.

Otherwise, with the CAP_PERFMON capability alone it would be possible
for a user to send SIGTRAP signals via perf events to another user's
tasks. This could potentially result in those tasks being terminated if
they cannot handle SIGTRAP signals.

Note: The check complements the existing capability check, but is not
supposed to supersede the ptrace_may_access() check. At a high level we
now have:

        capable of CAP_PERFMON and (CAP_KILL if sigtrap)
                OR
        ptrace_may_access(...) // also checks for same thread-group and uid

Fixes: 97ba62b27867 ("perf: Add support for SIGTRAP on perf events")
Cc: <redacted> # 5.13+
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Marco Elver <elver@google.com>
Acked-by: Dmitry Vyukov <dvyukov@google.com>
quoted hunk
---
v3:
* Upgrade ptrace mode check to ATTACH if attr.sigtrap, otherwise it's
  possible to change the target task (send signal) even if only read
  ptrace permissions were granted (reported by Eric W. Biederman).

v2: https://lkml.kernel.org/r/20210701083842.580466-1-elver@google.com
* Drop kill_capable() and just check CAP_KILL (reported by Ondrej Mosnacek).
* Use ns_capable(__task_cred(task)->user_ns, CAP_KILL) to check for
  capability in target task's ns (reported by Ondrej Mosnacek).

v1: https://lkml.kernel.org/r/20210630093709.3612997-1-elver@google.com
---
 kernel/events/core.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index fe88d6eea3c2..f79ee82e644a 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -12152,10 +12152,33 @@ SYSCALL_DEFINE5(perf_event_open,
        }

        if (task) {
+               unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
+               bool is_capable;
+
                err = down_read_interruptible(&task->signal->exec_update_lock);
                if (err)
                        goto err_file;

+               is_capable = perfmon_capable();
+               if (attr.sigtrap) {
+                       /*
+                        * perf_event_attr::sigtrap sends signals to the other
+                        * task. Require the current task to also have
+                        * CAP_KILL.
+                        */
+                       rcu_read_lock();
+                       is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
+                       rcu_read_unlock();
+
+                       /*
+                        * If the required capabilities aren't available, checks
+                        * for ptrace permissions: upgrade to ATTACH, since
+                        * sending signals can effectively change the target
+                        * task.
+                        */
+                       ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
+               }
+
                /*
                 * Preserve ptrace permission check for backwards compatibility.
                 *
@@ -12165,7 +12188,7 @@ SYSCALL_DEFINE5(perf_event_open,
                 * perf_event_exit_task() that could imply).
                 */
                err = -EACCES;
-               if (!perfmon_capable() && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
+               if (!is_capable && !ptrace_may_access(task, ptrace_mode))
                        goto err_cred;
        }

--
2.32.0.93.g670b81a890-goog

Re: [PATCH v3 2/2] perf: Refactor permissions check into perf_check_permission()

From: Dmitry Vyukov <dvyukov@google.com>
Date: 2021-07-06 06:16:56

On Mon, Jul 5, 2021 at 10:45 AM Marco Elver [off-list ref] wrote:
Refactor the permission check in perf_event_open() into a helper
perf_check_permission(). This makes the permission check logic more
readable (because we no longer have a negated disjunction). Add a
comment mentioning the ptrace check also checks the uid.

No functional change intended.

Signed-off-by: Marco Elver <elver@google.com>
Reviewed-by: Dmitry Vyukov <dvyukov@google.com>
quoted hunk
---
v3:
* Introduce this patch to refactor the permissions checking logic to
  make it more readable (reported by Eric W. Biederman).
---
 kernel/events/core.c | 58 ++++++++++++++++++++++++--------------------
 1 file changed, 32 insertions(+), 26 deletions(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index f79ee82e644a..3008b986994b 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -11912,6 +11912,37 @@ __perf_event_ctx_lock_double(struct perf_event *group_leader,
        return gctx;
 }

+static bool
+perf_check_permission(struct perf_event_attr *attr, struct task_struct *task)
+{
+       unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
+       bool is_capable = perfmon_capable();
+
+       if (attr->sigtrap) {
+               /*
+                * perf_event_attr::sigtrap sends signals to the other task.
+                * Require the current task to also have CAP_KILL.
+                */
+               rcu_read_lock();
+               is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
+               rcu_read_unlock();
+
+               /*
+                * If the required capabilities aren't available, checks for
+                * ptrace permissions: upgrade to ATTACH, since sending signals
+                * can effectively change the target task.
+                */
+               ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
+       }
+
+       /*
+        * Preserve ptrace permission check for backwards compatibility. The
+        * ptrace check also includes checks that the current task and other
+        * task have matching uids, and is therefore not done here explicitly.
+        */
+       return is_capable || ptrace_may_access(task, ptrace_mode);
+}
+
 /**
  * sys_perf_event_open - open a performance event, associate it to a task/cpu
  *
@@ -12152,43 +12183,18 @@ SYSCALL_DEFINE5(perf_event_open,
        }

        if (task) {
-               unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
-               bool is_capable;
-
                err = down_read_interruptible(&task->signal->exec_update_lock);
                if (err)
                        goto err_file;

-               is_capable = perfmon_capable();
-               if (attr.sigtrap) {
-                       /*
-                        * perf_event_attr::sigtrap sends signals to the other
-                        * task. Require the current task to also have
-                        * CAP_KILL.
-                        */
-                       rcu_read_lock();
-                       is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
-                       rcu_read_unlock();
-
-                       /*
-                        * If the required capabilities aren't available, checks
-                        * for ptrace permissions: upgrade to ATTACH, since
-                        * sending signals can effectively change the target
-                        * task.
-                        */
-                       ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
-               }
-
                /*
-                * Preserve ptrace permission check for backwards compatibility.
-                *
                 * We must hold exec_update_lock across this and any potential
                 * perf_install_in_context() call for this new event to
                 * serialize against exec() altering our credentials (and the
                 * perf_event_exit_task() that could imply).
                 */
                err = -EACCES;
-               if (!is_capable && !ptrace_may_access(task, ptrace_mode))
+               if (!perf_check_permission(&attr, task))
                        goto err_cred;
        }

--
2.32.0.93.g670b81a890-goog

Re: [PATCH v3 1/2] perf: Fix required permissions if sigtrap is requested

From: Marco Elver <elver@google.com>
Date: 2021-07-12 10:33:28

It'd be good to get this sorted -- please take another look.

Many thanks,
-- Marco

On Mon, 5 Jul 2021 at 10:45, Marco Elver [off-list ref] wrote:
quoted hunk
If perf_event_open() is called with another task as target and
perf_event_attr::sigtrap is set, and the target task's user does not
match the calling user, also require the CAP_KILL capability or
PTRACE_MODE_ATTACH permissions.

Otherwise, with the CAP_PERFMON capability alone it would be possible
for a user to send SIGTRAP signals via perf events to another user's
tasks. This could potentially result in those tasks being terminated if
they cannot handle SIGTRAP signals.

Note: The check complements the existing capability check, but is not
supposed to supersede the ptrace_may_access() check. At a high level we
now have:

        capable of CAP_PERFMON and (CAP_KILL if sigtrap)
                OR
        ptrace_may_access(...) // also checks for same thread-group and uid

Fixes: 97ba62b27867 ("perf: Add support for SIGTRAP on perf events")
Cc: <redacted> # 5.13+
Reported-by: Dmitry Vyukov <dvyukov@google.com>
Signed-off-by: Marco Elver <elver@google.com>
---
v3:
* Upgrade ptrace mode check to ATTACH if attr.sigtrap, otherwise it's
  possible to change the target task (send signal) even if only read
  ptrace permissions were granted (reported by Eric W. Biederman).

v2: https://lkml.kernel.org/r/20210701083842.580466-1-elver@google.com
* Drop kill_capable() and just check CAP_KILL (reported by Ondrej Mosnacek).
* Use ns_capable(__task_cred(task)->user_ns, CAP_KILL) to check for
  capability in target task's ns (reported by Ondrej Mosnacek).

v1: https://lkml.kernel.org/r/20210630093709.3612997-1-elver@google.com
---
 kernel/events/core.c | 25 ++++++++++++++++++++++++-
 1 file changed, 24 insertions(+), 1 deletion(-)
diff --git a/kernel/events/core.c b/kernel/events/core.c
index fe88d6eea3c2..f79ee82e644a 100644
--- a/kernel/events/core.c
+++ b/kernel/events/core.c
@@ -12152,10 +12152,33 @@ SYSCALL_DEFINE5(perf_event_open,
        }

        if (task) {
+               unsigned int ptrace_mode = PTRACE_MODE_READ_REALCREDS;
+               bool is_capable;
+
                err = down_read_interruptible(&task->signal->exec_update_lock);
                if (err)
                        goto err_file;

+               is_capable = perfmon_capable();
+               if (attr.sigtrap) {
+                       /*
+                        * perf_event_attr::sigtrap sends signals to the other
+                        * task. Require the current task to also have
+                        * CAP_KILL.
+                        */
+                       rcu_read_lock();
+                       is_capable &= ns_capable(__task_cred(task)->user_ns, CAP_KILL);
+                       rcu_read_unlock();
+
+                       /*
+                        * If the required capabilities aren't available, checks
+                        * for ptrace permissions: upgrade to ATTACH, since
+                        * sending signals can effectively change the target
+                        * task.
+                        */
+                       ptrace_mode = PTRACE_MODE_ATTACH_REALCREDS;
+               }
+
                /*
                 * Preserve ptrace permission check for backwards compatibility.
                 *
@@ -12165,7 +12188,7 @@ SYSCALL_DEFINE5(perf_event_open,
                 * perf_event_exit_task() that could imply).
                 */
                err = -EACCES;
-               if (!perfmon_capable() && !ptrace_may_access(task, PTRACE_MODE_READ_REALCREDS))
+               if (!is_capable && !ptrace_may_access(task, ptrace_mode))
                        goto err_cred;
        }

--
2.32.0.93.g670b81a890-goog

Re: [PATCH v3 1/2] perf: Fix required permissions if sigtrap is requested

From: Peter Zijlstra <peterz@infradead.org>
Date: 2021-07-13 09:49:33

On Mon, Jul 12, 2021 at 12:32:33PM +0200, Marco Elver wrote:
It'd be good to get this sorted -- please take another look.
Thanks!

I'll queue them into perf/urgent.
Keyboard shortcuts
hback out one level
jnext message in thread
kprevious message in thread
ldrill in
Escclose help / fold thread tree
?toggle this help