Skip to contentSkip to conversation

Introduce CAP_SYS_PERFMON to secure system performance monitoring and observability

PATCH Superseded Alexey Budankov 12 participants 48 messages Dec 18, 2019 – Apr 3, 2020

Subsystems (13)
DRM drivers PowerPC Intel i915 BPF +9 more subsystems

v4 of 8 Superseded Latest is v8 →

lore
Share & tools
lore
#0 lore raw

v4 cover Introduce CAP_SYS_PERFMON to secure system performance monitoring and observability

Currently access to perf_events, i915_perf and other performance monitoring and observability subsystems of the kernel is open for a privileged process [1] with CAP_SYS_ADMIN capability enabled in the process effective set [2].

This patch set introduces CAP_SYS_PERFMON capability devoted to secure system performance monitoring and observability operations so that CAP_SYS_PERFMON would assist CAP_SYS_ADMIN capability in its governing role for perf_events, i915_perf and other performance monitoring and observability subsystems of the kernel.

CAP_SYS_PERFMON intends to meet the demand to secure system performance monitoring and observability operations in security sensitive, restricted, production environments (e.g. HPC clusters, cloud and virtual compute environments) where root or CAP_SYS_ADMIN credentials are not available to mass users of a system because of security considerations.

CAP_SYS_PERFMON intends to harden system security and integrity during system performance monitoring and observability operations by decreasing attack surface that is available to CAP_SYS_ADMIN privileged processes [2].

CAP_SYS_PERFMON intends to take over CAP_SYS_ADMIN credentials related to system performance monitoring and observability operations and balance amount of CAP_SYS_ADMIN credentials following the recommendations in the capabilities man page [2] for CAP_SYS_ADMIN: "Note: this capability is overloaded; see Notes to kernel developers, below."

For backward compatibility reasons access to system performance monitoring and observability subsystems of the kernel remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN capability usage for secure system performance monitoring and observability operations is discouraged with respect to the introduced CAP_SYS_PERFMON capability.

The patch set is for tip perf/core repository:
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip perf/core
sha1: ceb9e77324fa661b1001a0ae66f061b5fcb4e4e6

---
Changes in v4:
- converted perfmon_capable() into an inline function
- made perf_events kprobes, uprobes, hw breakpoints and namespaces data available to CAP_SYS_PERFMON privileged processes
- applied perfmon_capable() to drivers/perf and drivers/oprofile
- extended __cmd_ftrace() with support of CAP_SYS_PERFMON
Changes in v3:
- implemented perfmon_capable() macros aggregating required capabilities checks Changes in v2:
- made perf_events trace points available to CAP_SYS_PERFMON privileged processes
- made perf_event_paranoid_check() treat CAP_SYS_PERFMON equally to CAP_SYS_ADMIN
- applied CAP_SYS_PERFMON to i915_perf, bpf_trace, powerpc and parisc system performance monitoring and observability related subsystems

---
Alexey Budankov (9):
capabilities: introduce CAP_SYS_PERFMON to kernel and user space perf/core: open access for CAP_SYS_PERFMON privileged process perf tool: extend Perf tool with CAP_SYS_PERFMON capability support drm/i915/perf: open access for CAP_SYS_PERFMON privileged process trace/bpf_trace: open access for CAP_SYS_PERFMON privileged process powerpc/perf: open access for CAP_SYS_PERFMON privileged process parisc/perf: open access for CAP_SYS_PERFMON privileged process drivers/perf: open access for CAP_SYS_PERFMON privileged process drivers/oprofile: open access for CAP_SYS_PERFMON privileged process

 arch/parisc/kernel/perf.c           |  2 +-
 arch/powerpc/perf/imc-pmu.c         |  4 ++--
 drivers/gpu/drm/i915/i915_perf.c    | 13 ++++++-------
 drivers/oprofile/event_buffer.c     |  2 +-
 drivers/perf/arm_spe_pmu.c          |  4 ++--
 include/linux/capability.h          |  4 ++++
 include/linux/perf_event.h          |  6 +++---
 include/uapi/linux/capability.h     |  8 +++++++-
 kernel/events/core.c                |  6 +++---
 kernel/trace/bpf_trace.c            |  2 +-
 security/selinux/include/classmap.h |  4 ++--
 tools/perf/builtin-ftrace.c         |  5 +++--
 tools/perf/design.txt               |  3 ++-
 tools/perf/util/cap.h               |  4 ++++
 tools/perf/util/evsel.c             | 10 +++++-----
 tools/perf/util/util.c              |  1 +
 16 files changed, 47 insertions(+), 31 deletions(-)

---
Testing and validation (Intel Skylake, 8 cores, Fedora 29, 5.4.0-rc8+, x86_64):

libcap library [3], [4] and Perf tool can be used to apply CAP_SYS_PERFMON capability for secure system performance monitoring and observability beyond the scope permitted by the system wide perf_event_paranoid kernel setting [5] and below are the steps for evaluation:

- patch, build and boot the kernel
- patch, build Perf tool e.g. to /home/user/perf
...

  # git clone git://git.kernel.org/pub/scm/libs/libcap/libcap.git libcap
  # pushd libcap
  # patch libcap/include/uapi/linux/capabilities.h with [PATCH 1]
  # make
  # pushd progs
  # ./setcap "cap_sys_perfmon,cap_sys_ptrace,cap_syslog=ep" /home/user/perf
  # ./setcap -v "cap_sys_perfmon,cap_sys_ptrace,cap_syslog=ep" /home/user/perf
  /home/user/perf: OK
  # ./getcap /home/user/perf
  /home/user/perf = cap_sys_ptrace,cap_syslog,cap_sys_perfmon+ep
  # echo 2 > /proc/sys/kernel/perf_event_paranoid
  # cat /proc/sys/kernel/perf_event_paranoid 
  2
  ...
  $ /home/user/perf top
    ... works as expected ...
  $ cat /proc/`pidof perf`/status
  Name:	perf
  Umask:	0002
  State:	S (sleeping)
  Tgid:	2958
  Ngid:	0
  Pid:	2958
  PPid:	9847
  TracerPid:	0
  Uid:	500	500	500	500
  Gid:	500	500	500	500
  FDSize:	256
  ...
  CapInh:	0000000000000000
  CapPrm:	0000004400080000
  CapEff:	0000004400080000 => 01000100 00000000 00001000 00000000 00000000
                                     cap_sys_perfmon,cap_sys_ptrace,cap_syslog
  CapBnd:	0000007fffffffff
  CapAmb:	0000000000000000
  NoNewPrivs:	0
  Seccomp:	0
  Speculation_Store_Bypass:	thread vulnerable
  Cpus_allowed:	ff
  Cpus_allowed_list:	0-7
  ...

Usage of cap_sys_perfmon effectively avoids unused credentials excess:

- with cap_sys_admin:
CapEff: 0000007fffffffff => 01111111 11111111 11111111 11111111 11111111

- with cap_sys_perfmon:
CapEff: 0000004400080000 => 01000100 00000000 00001000 00000000 00000000

                                    38   34               19
                           sys_perfmon   syslog           sys_ptrace

---

[1] https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html
[2] http://man7.org/linux/man-pages/man7/capabilities.7.html
[3] http://man7.org/linux/man-pages/man8/setcap.8.html
[4] https://git.kernel.org/pub/scm/libs/libcap/libcap.git
[5] http://man7.org/linux/man-pages/man2/perf_event_open.2.html
[6] https://sites.google.com/site/fullycapable/, posix_1003.1e-990310.pdf

Signature · 1 line
2.20.1
#1 lore raw

PATCH v4 1/9 capabilities: introduce CAP_SYS_PERFMON to kernel and user space

Introduce CAP_SYS_PERFMON capability devoted to secure system performance monitoring and observability operations so that CAP_SYS_PERFMON would assist CAP_SYS_ADMIN capability in its governing role for perf_events, i915_perf and other subsystems of the kernel.

CAP_SYS_PERFMON intends to harden system security and integrity during system performance monitoring and observability operations by decreasing attack surface that is available to CAP_SYS_ADMIN privileged processes.

CAP_SYS_PERFMON intends to take over CAP_SYS_ADMIN credentials related to system performance monitoring and observability operations and balance amount of CAP_SYS_ADMIN credentials in accordance with the recommendations provided in the man page for CAP_SYS_ADMIN [1]: "Note: this capability is overloaded; see Notes to kernel developers, below."

[1] http://man7.org/linux/man-pages/man7/capabilities.7.html

---
 include/linux/capability.h          | 4 ++++
 include/uapi/linux/capability.h     | 8 +++++++-
 security/selinux/include/classmap.h | 4 ++--
 3 files changed, 13 insertions(+), 3 deletions(-)
include/linux/capability.h +4 -0
index ecce0f43c73a..883c879baa4b 100644
@@ -251,6 +251,10 @@ extern bool privileged_wrt_inode_uidgid(struct user_namespace *ns, const struct
 extern bool capable_wrt_inode_uidgid(const struct inode *inode, int cap);
 extern bool file_ns_capable(const struct file *file, struct user_namespace *ns, int cap);
 extern bool ptracer_capable(struct task_struct *tsk, struct user_namespace *ns);
+static inline bool perfmon_capable(void)
+{
+	return capable(CAP_SYS_PERFMON) || capable(CAP_SYS_ADMIN);
+}
 
 /* audit system wants to get cap info from files as well */
 extern int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps);
include/uapi/linux/capability.h +7 -1
index 240fdb9a60f6..98e03cc76c7c 100644
@@ -366,8 +366,14 @@ struct vfs_ns_cap_data {
 
 #define CAP_AUDIT_READ		37
 
+/*
+ * Allow system performance and observability privileged operations
+ * using perf_events, i915_perf and other kernel subsystems
+ */
+
+#define CAP_SYS_PERFMON		38
 
-#define CAP_LAST_CAP         CAP_AUDIT_READ
+#define CAP_LAST_CAP         CAP_SYS_PERFMON
 
 #define cap_valid(x) ((x) >= 0 && (x) <= CAP_LAST_CAP)
 
security/selinux/include/classmap.h +2 -2
index 7db24855e12d..bae602c623b0 100644
@@ -27,9 +27,9 @@
 	    "audit_control", "setfcap"
 
 #define COMMON_CAP2_PERMS  "mac_override", "mac_admin", "syslog", \
-		"wake_alarm", "block_suspend", "audit_read"
+		"wake_alarm", "block_suspend", "audit_read", "sys_perfmon"
 
-#if CAP_LAST_CAP > CAP_AUDIT_READ
+#if CAP_LAST_CAP > CAP_SYS_PERFMON
 #error New capability defined, please update COMMON_CAP2_PERMS.
 #endif
 
Signature · 1 line
2.20.1
Acked-by on v4 1/9
#2 lore raw

Re: [PATCH v4 1/9] capabilities: introduce CAP_SYS_PERFMON to kernel and user space

Quoted 16 lines · Alexey Budankov

Introduce CAP_SYS_PERFMON capability devoted to secure system performance monitoring and observability operations so that CAP_SYS_PERFMON would assist CAP_SYS_ADMIN capability in its governing role for perf_events, i915_perf and other subsystems of the kernel.

CAP_SYS_PERFMON intends to harden system security and integrity during system performance monitoring and observability operations by decreasing attack surface that is available to CAP_SYS_ADMIN privileged processes.

CAP_SYS_PERFMON intends to take over CAP_SYS_ADMIN credentials related to system performance monitoring and observability operations and balance amount of CAP_SYS_ADMIN credentials in accordance with the recommendations provided in the man page for CAP_SYS_ADMIN [1]: "Note: this capability is overloaded; see Notes to kernel developers, below."

Quoted 3 lines

[1] http://man7.org/linux/man-pages/man7/capabilities.7.html

Note for selinux developers: we will need to update the selinux-testsuite tests for perf_event when/if this change lands upstream.

Quoted 58 lines
---
  include/linux/capability.h          | 4 ++++
  include/uapi/linux/capability.h     | 8 +++++++-
  security/selinux/include/classmap.h | 4 ++--
  3 files changed, 13 insertions(+), 3 deletions(-)
include/linux/capability.h +4 -0
index ecce0f43c73a..883c879baa4b 100644
@@ -251,6 +251,10 @@ extern bool privileged_wrt_inode_uidgid(struct user_namespace *ns, const struct
  extern bool capable_wrt_inode_uidgid(const struct inode *inode, int cap);
  extern bool file_ns_capable(const struct file *file, struct user_namespace *ns, int cap);
  extern bool ptracer_capable(struct task_struct *tsk, struct user_namespace *ns);
+static inline bool perfmon_capable(void)
+{
+	return capable(CAP_SYS_PERFMON) || capable(CAP_SYS_ADMIN);
+}
 
  /* audit system wants to get cap info from files as well */
  extern int get_vfs_caps_from_disk(const struct dentry *dentry, struct cpu_vfs_cap_data *cpu_caps);
include/uapi/linux/capability.h +7 -1
index 240fdb9a60f6..98e03cc76c7c 100644
@@ -366,8 +366,14 @@ struct vfs_ns_cap_data {
 
  #define CAP_AUDIT_READ		37
 
+/*
+ * Allow system performance and observability privileged operations
+ * using perf_events, i915_perf and other kernel subsystems
+ */
+
+#define CAP_SYS_PERFMON		38
 
-#define CAP_LAST_CAP         CAP_AUDIT_READ
+#define CAP_LAST_CAP         CAP_SYS_PERFMON
 
  #define cap_valid(x) ((x) >= 0 && (x) <= CAP_LAST_CAP)
 
security/selinux/include/classmap.h +2 -2
index 7db24855e12d..bae602c623b0 100644
@@ -27,9 +27,9 @@
  	    "audit_control", "setfcap"
 
  #define COMMON_CAP2_PERMS  "mac_override", "mac_admin", "syslog", \
-		"wake_alarm", "block_suspend", "audit_read"
+		"wake_alarm", "block_suspend", "audit_read", "sys_perfmon"
 
-#if CAP_LAST_CAP > CAP_AUDIT_READ
+#if CAP_LAST_CAP > CAP_SYS_PERFMON
  #error New capability defined, please update COMMON_CAP2_PERMS.
  #endif
 
maintainer Acked-by on v4 1/9
#3 lore raw

Re: [Intel-gfx] [PATCH v4 1/9] capabilities: introduce CAP_SYS_PERFMON to kernel and user space

In reply to Alexey Budankov
Quoted 16 lines · Alexey Budankov

Introduce CAP_SYS_PERFMON capability devoted to secure system performance monitoring and observability operations so that CAP_SYS_PERFMON would assist CAP_SYS_ADMIN capability in its governing role for perf_events, i915_perf and other subsystems of the kernel.

CAP_SYS_PERFMON intends to harden system security and integrity during system performance monitoring and observability operations by decreasing attack surface that is available to CAP_SYS_ADMIN privileged processes.

CAP_SYS_PERFMON intends to take over CAP_SYS_ADMIN credentials related to system performance monitoring and observability operations and balance amount of CAP_SYS_ADMIN credentials in accordance with the recommendations provided in the man page for CAP_SYS_ADMIN [1]: "Note: this capability is overloaded; see Notes to kernel developers, below."

Quoted 3 lines

[1] http://man7.org/linux/man-pages/man7/capabilities.7.html

Acked-by on v4 1/9
#4 lore raw

Re: [Intel-gfx] [PATCH v4 1/9] capabilities: introduce CAP_SYS_PERFMON to kernel and user space

In reply to Alexey Budankov
Quoted 20 lines · Alexey Budankov

Introduce CAP_SYS_PERFMON capability devoted to secure system performance monitoring and observability operations so that CAP_SYS_PERFMON would assist CAP_SYS_ADMIN capability in its governing role for perf_events, i915_perf and other subsystems of the kernel.

CAP_SYS_PERFMON intends to harden system security and integrity during system performance monitoring and observability operations by decreasing attack surface that is available to CAP_SYS_ADMIN privileged processes.

CAP_SYS_PERFMON intends to take over CAP_SYS_ADMIN credentials related to system performance monitoring and observability operations and balance amount of CAP_SYS_ADMIN credentials in accordance with the recommendations provided in the man page for CAP_SYS_ADMIN [1]: "Note: this capability is overloaded; see Notes to kernel developers, below."

Quoted 3 lines

[1] https://urldefense.proofpoint.com/v2/url?u=http-3A__man7.org_linux_man-2Dpages_man7_capabilities.7.html&d=DwICaQ&c=5VD0RTtNlTh3ycd41b3MUw&r=dR8692q0_uaizy0jkrBJQM5k2hfm4CiFxYT8KaysFrg&m=L5qCuMRrTvYhyjR1rpgE9vEv4HppVlOXDIzKzoGL30c&s=FNJpET4buKFRuqktVHQphaY1qE7IsdFpU4iYwpCn4tY&e=

#5 lore raw

PATCH v4 2/9 perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to perf_events monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to perf_events subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure perf_events monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 include/linux/perf_event.h | 6 +++---
 kernel/events/core.c       | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/perf_event.h +3 -3
index 34c7c6910026..f46acd69425f 100644
@@ -1285,7 +1285,7 @@ static inline int perf_is_paranoid(void)
 
 static inline int perf_allow_kernel(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
@@ -1293,7 +1293,7 @@ static inline int perf_allow_kernel(struct perf_event_attr *attr)
 
 static inline int perf_allow_cpu(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 0 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_CPU);
@@ -1301,7 +1301,7 @@ static inline int perf_allow_cpu(struct perf_event_attr *attr)
 
 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > -1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
 		return -EPERM;
 
 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);
kernel/events/core.c +3 -3
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*
@@ -9116,7 +9116,7 @@ static int perf_uprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_uprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*
@@ -11157,7 +11157,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (attr.namespaces) {
-		if (!capable(CAP_SYS_ADMIN))
+		if (!perfmon_capable())
 			return -EACCES;
 	}
 
Signature · 1 line
2.20.1
maintainer
#6 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 36 lines · Alexey Budankov

Open access to perf_events monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to perf_events subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure perf_events monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 include/linux/perf_event.h | 6 +++---
 kernel/events/core.c       | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/perf_event.h +2 -2
index 34c7c6910026..f46acd69425f 100644
@@ -1285,7 +1285,7 @@ static inline int perf_is_paranoid(void)
 
 static inline int perf_allow_kernel(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
@@ -1293,7 +1293,7 @@ static inline int perf_allow_kernel(struct perf_event_attr *attr)
 
 static inline int perf_allow_cpu(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 0 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_CPU);

include/linux/perf_event.h static inline int perf_allow_cpu(struct perf_event_attr *attr) · line 1307

Quoted 8 lines
include/linux/perf_event.h +1 -1
@@ -1301,7 +1301,7 @@ static inline int perf_allow_cpu(struct perf_event_attr *attr)
 
 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > -1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
 		return -EPERM;
 
 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);

These are OK I suppose.

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 13 lines
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

kernel/events/core.c static int perf_uprobe_event_init(struct perf_event *event) · line 9122

Quoted 9 lines
kernel/events/core.c +1 -1
@@ -9116,7 +9116,7 @@ static int perf_uprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_uprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

Idem, I presume.

kernel/events/core.c SYSCALL_DEFINE5(perf_event_open, · line 11162

Quoted 8 lines
kernel/events/core.c +1 -1
@@ -11157,7 +11157,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (attr.namespaces) {
-		if (!capable(CAP_SYS_ADMIN))
+		if (!perfmon_capable())
 			return -EACCES;
 	}

And given we basically make the entire kernel observable with this CAP, busting namespaces shoulnd't be a problem either.

So yeah, I suppose that works.

#7 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 56 lines · Peter Zijlstra
Quoted 36 lines · Alexey Budankov

Open access to perf_events monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to perf_events subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure perf_events monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 include/linux/perf_event.h | 6 +++---
 kernel/events/core.c       | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/perf_event.h +2 -2
index 34c7c6910026..f46acd69425f 100644
@@ -1285,7 +1285,7 @@ static inline int perf_is_paranoid(void)
 
 static inline int perf_allow_kernel(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
@@ -1293,7 +1293,7 @@ static inline int perf_allow_kernel(struct perf_event_attr *attr)
 
 static inline int perf_allow_cpu(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 0 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_CPU);

include/linux/perf_event.h static inline int perf_allow_cpu(struct perf_event_attr *attr) · line 1307

Quoted 8 lines
include/linux/perf_event.h +1 -1
@@ -1301,7 +1301,7 @@ static inline int perf_allow_cpu(struct perf_event_attr *attr)
 
 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > -1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
 		return -EPERM;
 
 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);

These are OK I suppose.

Quoted 8 lines
kernel/events/core.c +0 -0
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
Quoted 8 lines

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 5 lines
kernel/events/core.c +1 -1
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

This unblocks creation of local trace kprobes and uprobes by CAP_SYS_PERFMON privileged process, exactly the same as for CAP_SYS_ADMIN privileged process.

Quoted 27 lines

kernel/events/core.c static int perf_uprobe_event_init(struct perf_event *event) · line 9122

Quoted 10 lines
kernel/events/core.c +1 -1
@@ -9116,7 +9116,7 @@ static int perf_uprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_uprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

Idem, I presume.

kernel/events/core.c SYSCALL_DEFINE5(perf_event_open, · line 11162

Quoted 8 lines
kernel/events/core.c +1 -1
@@ -11157,7 +11157,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (attr.namespaces) {
-		if (!capable(CAP_SYS_ADMIN))
+		if (!perfmon_capable())
 			return -EACCES;
 	}

And given we basically make the entire kernel observable with this CAP, busting namespaces shoulnd't be a problem either.

So yeah, I suppose that works.

maintainer
#8 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 15 lines · Alexey Budankov
Quoted 14 lines · Peter Zijlstra
Quoted 12 lines · Alexey Budankov
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
Quoted 7 lines
Quoted 4 lines

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 1 line
kernel/events/core.c +0 -0
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

This unblocks creation of local trace kprobes and uprobes by CAP_SYS_PERFMON privileged process, exactly the same as for CAP_SYS_ADMIN privileged process.

I've no idea what you just said; it's just words.

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

maintainer
#9 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 26 lines · On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra <[email protected]> wrote:
Quoted 15 lines · Alexey Budankov
Quoted 14 lines · Peter Zijlstra
Quoted 12 lines · Alexey Budankov
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
Quoted 7 lines
Quoted 4 lines

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 1 line
kernel/events/core.c +0 -0
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

This unblocks creation of local trace kprobes and uprobes by CAP_SYS_PERFMON privileged process, exactly the same as for CAP_SYS_ADMIN privileged process.

I've no idea what you just said; it's just words.

Quoted 8 lines

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace interface to define new kprobe events, and those events are treated as completely same as tracepoint events. On the other hand, ebpf tries to define new probe event via perf_event interface. Above one is that interface. IOW, it creates new kprobe.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>
#10 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 7 lines · On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra <[email protected]> wrote:
Quoted 5 lines

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

Quoted 7 lines
Quoted 2 lines

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace interface to define new kprobe events, and those events are treated as completely same as tracepoint events. On the other hand, ebpf tries to define new probe event via perf_event interface. Above one is that interface. IOW, it creates new kprobe.

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

- Arnaldo

maintainer
#11 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 11 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 9 lines · On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra <[email protected]> wrote:
Quoted 8 lines

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

Quoted 7 lines
Quoted 4 lines

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace interface to define new kprobe events, and those events are treated as completely same as tracepoint events. On the other hand, ebpf tries to define new probe event via perf_event interface. Above one is that interface. IOW, it creates new kprobe.

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()?
No, perf probe is for setting up the ftrace probe events. I think we can add an option to use perf_event_open(). But current kprobe creation from perf_event_open() is separated from ftrace by design.

I think the reason why ebpf uses perf_event_open() interface is to avoid conflict with ftrace users. Those probes are temporally used by ebpf, but if it is appeared on ftrace, it is easy to be used by ftrace. In that case, it can not be removed when the ebpf exits.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>
#12 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 23 lines · Masami Hiramatsu
Quoted 16 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 14 lines · On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra <[email protected]> wrote:
Quoted 8 lines

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace interface to define new kprobe events, and those events are treated as completely same as tracepoint events. On the other hand, ebpf tries to define new probe event via perf_event interface. Above one is that interface. IOW, it creates new kprobe.

Quoted 7 lines
Quoted 2 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()?
No, perf probe is for setting up the ftrace probe events. I think we can add an option to use perf_event_open(). But current kprobe creation from perf_event_open() is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

perf stat/record -e kprobe:kernel_func ...

Thanks,
Song

#13 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

<[email protected]>,Jann Horn <[email protected]>,Thomas Gleixner <[email protected]>,Tvrtko Ursulin <[email protected]>,Lionel Landwerlin <[email protected]>,linux-kernel <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>
From: Arnaldo Carvalho de Melo <[email protected]>
Message-ID: <[email protected]>

Quoted 43 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

Quoted 6 lines

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

perf stat/record -e kprobe:kernel_func ...

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Regards,

- Arnaldo

Quoted 3 lines

Thanks,
Song

#14 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 51 lines · On 11.01.2020 3:35, [email protected] wrote:

<[email protected]>,Jann Horn <[email protected]>,Thomas Gleixner <[email protected]>,Tvrtko Ursulin <[email protected]>,Lionel Landwerlin <[email protected]>,linux-kernel <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>
From: Arnaldo Carvalho de Melo <[email protected]>
Message-ID: <[email protected]>

Quoted 46 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

Quoted 6 lines
Quoted 3 lines

We can just do

perf stat/record -e kprobe:kernel_func ...

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Arnaldo, Masami, Song,

What do you think about making this also open to CAP_SYS_PERFMON privileged processes? Could you please also review and comment on patch 5/9 for bpf_trace.c?

Thanks,
Alexey

Quoted 9 lines

Regards,

- Arnaldo

Quoted 3 lines

Thanks,
Song

#15 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 59 lines · Alexey Budankov
Quoted 53 lines · On 11.01.2020 3:35, [email protected] wrote:

<[email protected]>,Jann Horn <[email protected]>,Thomas Gleixner <[email protected]>,Tvrtko Ursulin <[email protected]>,Lionel Landwerlin <[email protected]>,linux-kernel <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>
From: Arnaldo Carvalho de Melo <[email protected]>
Message-ID: <[email protected]>

Quoted 48 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

Quoted 8 lines
Quoted 4 lines
Quoted 1 line

perf stat/record -e kprobe:kernel_func ...

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Arnaldo, Masami, Song,

What do you think about making this also open to CAP_SYS_PERFMON privileged processes?

I think we should at least allow CAP_SYS_PERFMON to create some kprobes. Maybe we can limited that to per-task kprobes, and the task should be owned by the user?

Thanks,
Song

maintainer
#16 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov
Quoted 57 lines · On Sat, 11 Jan 2020 12:57:18 +0300 Alexey Budankov…
Quoted 54 lines · On 11.01.2020 3:35, [email protected] wrote:

Message-ID: <[email protected]>

Quoted 49 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

perf stat/record -e kprobe:kernel_func ...

Quoted 6 lines
Quoted 1 line

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Arnaldo, Masami, Song,

What do you think about making this also open to CAP_SYS_PERFMON privileged processes? Could you please also review and comment on patch 5/9 for bpf_trace.c?

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>
#17 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 71 lines · Masami Hiramatsu
Quoted 57 lines · On Sat, 11 Jan 2020 12:57:18 +0300 Alexey Budankov…
Quoted 54 lines · On 11.01.2020 3:35, [email protected] wrote:

Message-ID: <[email protected]>

Quoted 49 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

perf stat/record -e kprobe:kernel_func ...

Quoted 6 lines
Quoted 1 line

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Arnaldo, Masami, Song,

What do you think about making this also open to CAP_SYS_PERFMON privileged processes? Could you please also review and comment on patch 5/9 for bpf_trace.c?

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

Quoted 7 lines

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

#18 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 80 lines · Alexei Starovoitov
Quoted 71 lines · Masami Hiramatsu
Quoted 57 lines · On Sat, 11 Jan 2020 12:57:18 +0300 Alexey Budankov…
Quoted 54 lines · On 11.01.2020 3:35, [email protected] wrote:

Message-ID: <[email protected]>

Quoted 49 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

perf stat/record -e kprobe:kernel_func ...

Quoted 6 lines
Quoted 1 line

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Arnaldo, Masami, Song,

What do you think about making this also open to CAP_SYS_PERFMON privileged processes? Could you please also review and comment on patch 5/9 for bpf_trace.c?

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

Quoted 7 lines

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

Thanks,
Alexey

#19 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 22 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Quoted 8 lines

Alexei, Masami,

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw. imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

#20 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

Thanks,
Alexey

#21 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 47 lines · Alexey Budankov
Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Quoted 3 lines

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

yes. the current patches 2 and 5 look good to me.
I would only change patch 1 to what Andy was proposing earlier:

static inline bool perfmon_capable(void)
{
if (capable_noaudit(CAP_PERFMON))
return capable(CAP_PERFMON);
if (capable_noaudit(CAP_SYS_ADMIN))
return capable(CAP_SYS_ADMIN);

return capable(CAP_PERFMON);
}
I think Andy was trying to preserve the order of audit events.

I'm also suggesting to drop SYS from the cap name. It doesn't add any value to the name.

#22 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 49 lines · Alexei Starovoitov
Quoted 47 lines · Alexey Budankov
Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Quoted 5 lines
Quoted 3 lines

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

yes. the current patches 2 and 5 look good to me.

Thanks. I appreciate your cooperation.

Quoted 1 line

I would only change patch 1 to what Andy was proposing earlier:

Could you please share the link to the proposal to get more details? In this patch set discussion there was only this [1] on more generic naming of PERFMON cap from Andi Kleen.

Quoted 10 lines

static inline bool perfmon_capable(void)
{
if (capable_noaudit(CAP_PERFMON))
return capable(CAP_PERFMON);
if (capable_noaudit(CAP_SYS_ADMIN))
return capable(CAP_SYS_ADMIN);

return capable(CAP_PERFMON);
}

Yes, this makes sense and adds up.

Quoted 4 lines

I think Andy was trying to preserve the order of audit events.

I'm also suggesting to drop SYS from the cap name. It doesn't add any value to the name.

Agreed, CAP_PERFMON sounds more generic, as it actually is.

Gratefully,
Alexey

[1] https://lore.kernel.org/lkml/[email protected]/

#23 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Alexei Starovoitov

Hi Alexei,

Quoted 49 lines · Alexei Starovoitov
Quoted 47 lines · Alexey Budankov
Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Quoted 5 lines
Quoted 3 lines

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

yes. the current patches 2 and 5 look good to me.

Could this have you explicit Reviewed-by or Acked-by tag so
the changes could be driven into the kernel?
Latest v7 is here: https://lore.kernel.org/lkml/[email protected]/

Thanks,
Alexey

#24 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 56 lines · Alexey Budankov

Hi Alexei,

Quoted 53 lines · Alexei Starovoitov
Quoted 50 lines · Alexey Budankov
Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

Quoted 5 lines
Quoted 1 line

yes. the current patches 2 and 5 look good to me.

Could this have you explicit Reviewed-by or Acked-by tag so
the changes could be driven into the kernel?
Latest v7 is here: https://lore.kernel.org/lkml/[email protected]/

Posted v8 with all acquired tags so far:
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Alexey

#25 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov
Quoted 56 lines · Alexey Budankov

Hi Alexei,

Quoted 53 lines · Alexei Starovoitov
Quoted 50 lines · Alexey Budankov
Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

Quoted 5 lines
Quoted 1 line

yes. the current patches 2 and 5 look good to me.

Could this have you explicit Reviewed-by or Acked-by tag so
the changes could be driven into the kernel?
Latest v7 is here: https://lore.kernel.org/lkml/[email protected]/

Posted v8 with all acquired tags so far:
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Alexey

maintainer
#26 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov
Quoted 28 lines · On Tue, 14 Jan 2020 21:50:33 +0300 Alexey Budankov…
Quoted 26 lines · Alexei Starovoitov
Quoted 24 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Quoted 9 lines
Quoted 8 lines
Quoted 6 lines

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

OK, anyway, for higher security, admin may not give CAP_SYS_PERFMON to unpriviledged users, since it might allows users to analyze kernel, which can lead security concerns.

Quoted 2 lines

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Agreed, bugs must be fixed anyway.

Thank you,

Quoted 12 lines
Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

Thanks,
Alexey

Signature · 1 line
Masami Hiramatsu <[email protected]>
#27 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 38 lines · Masami Hiramatsu
Quoted 35 lines · On Tue, 14 Jan 2020 21:50:33 +0300 Alexey Budankov…
Quoted 33 lines · Alexei Starovoitov
Quoted 30 lines · Alexey Budankov
Quoted 12 lines
Quoted 11 lines

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 9 lines

I don't see how enable/disable is any safer than creation.
If there are kernel bugs in kprobes the kernel will crash anyway. I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

Alexei, Masami,

Thanks for your meaningful input.
If we know in advance that it still can crash the system in some cases and on some archs, even though root fully controls delegation thru CAP_SYS_PERFMON, such delegation looks premature until the crashes are avoided. So it looks like access to eBPF for CAP_SYS_PERFMON privileged processes is the subject for a separate patch set.

Quoted 6 lines
Quoted 2 lines
Quoted 1 line

perf_event_open is always dangerous. sw cannot guarantee non-bugginess of hw.

OK, anyway, for higher security, admin may not give CAP_SYS_PERFMON to unpriviledged users, since it might allows users to analyze kernel, which can lead security concerns.

FWIW,
Discovered security related hardware issues could be mitigated in software and here [1] is the official procedure documented on how to follow up, so this could be a draft plan to approach eBPF perf_events related hardware issues, if required.

[1] https://www.kernel.org/doc/html/latest/process/embargoed-hardware-issues.html

Quoted 22 lines
Quoted 3 lines

Sure, software cannot guarantee, but known software bugs could still be fixed, that's what I meant.

Agreed, bugs must be fixed anyway.

Thank you,

Quoted 12 lines
Quoted 5 lines

imo adding a cap just for pmc is pointless.
if you add a new cap it should cover all of sys_perf_event_open syscall. subdividing it into sw vs hw counters, kprobe create vs enable, etc will be the source of ongoing confusion. nack to such cap.

Well, as this patch set already covers complete perf_event_open functionality, and also eBPF related parts too, could you please review and comment on it? Does the patches 2/9 and 5/9 already bring all required extentions?

Thanks,
Alexey

maintainer
#28 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Alexei Starovoitov
Quoted 80 lines · On Mon, 13 Jan 2020 21:17:49 -0800 Alexei Starovoitov…
Quoted 71 lines · Masami Hiramatsu
Quoted 57 lines · On Sat, 11 Jan 2020 12:57:18 +0300 Alexey Budankov…
Quoted 54 lines · On 11.01.2020 3:35, [email protected] wrote:

Message-ID: <[email protected]>

Quoted 49 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

We can just do

perf stat/record -e kprobe:kernel_func ...

Quoted 6 lines
Quoted 1 line

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Arnaldo, Masami, Song,

What do you think about making this also open to CAP_SYS_PERFMON privileged processes? Could you please also review and comment on patch 5/9 for bpf_trace.c?

As we talked at RFC series of CAP_SYS_TRACING last year, I just expected to open it for enabling/disabling kprobes, not for creation.

Quoted 7 lines

If we can accept user who has no admin priviledge but the CAP_SYS_PERFMON, to shoot their foot by their own risk, I'm OK to allow it. (Even though, it should check the max number of probes to be created by something like ulimit)
I think nowadays we have fixed all such kernel crash problems on x86, but not sure for other archs, especially on the devices I can not reach. I need more help to stabilize it.

Quoted 2 lines

I don't see how enable/disable is any safer than creation. If there are kernel bugs in kprobes the kernel will crash anyway.

Why? admin can test the probes before using it via bpf.

My point was only admin can make a dicision to allow (or delegate) the priviledge to a user, and if it is OK, I don't mind it. (Maybe it is better to give a knob to allow this CAP only for admin.)

Quoted 6 lines

I think such partial CAP_SYS_PERFMON would be very confusing to the users. CAP_* is about delegation of root privileges to non-root. Delegating some of it is ok, but disallowing creation makes it useless for bpf tracing, so we would need to add another CAP later. Hence I suggest to do it right away instead of breaking sys_perf_even_open() access into two CAPs.

I understand that the single strong CAP will useful anyway (even if it is CAP_SYS_ADMIN). I just concern that causes any issue and when someone wants to mitigate it, it is sad if there is only way to disable all tracing facilities.

What about providing a sysctl to control the power of the CAP? maybe it is also good from the viewpoint of system security.

Thank you,

Signature · 1 line
Masami Hiramatsu <[email protected]>
maintainer
#29 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to [email protected]
Quoted 1 line · On Fri, 10 Jan 2020 21:35:12 -0300 [email protected] wrote:

<[email protected]>,Jann Horn <[email protected]>,Thomas Gleixner <[email protected]>,Tvrtko Ursulin <[email protected]>,Lionel Landwerlin <[email protected]>,linux-kernel <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <[email protected]>,"[email protected]" <oprofile-list@list

s.sf.net>

Quoted 50 lines

From: Arnaldo Carvalho de Melo <[email protected]>
Message-ID: <[email protected]>

Quoted 46 lines · Song Liu
Quoted 3 lines

On Jan 10, 2020, at 3:47 PM, Masami Hiramatsu <[email protected]>

Quoted 6 lines · wrote:
Quoted 2 lines · On Fri, 10 Jan 2020 13:45:31 -0300 Arnaldo Carvalho de Melo <[email protected]>…

Em Sat, Jan 11, 2020 at 12:52:13AM +0900, Masami Hiramatsu escreveu:

Quoted 1 line

On Fri, 10 Jan 2020 15:02:34 +0100 Peter Zijlstra

Quoted 1 line · <[email protected]> wrote:
Quoted 1 line
Quoted 1 line
Quoted 1 line

Again, this only allows attaching to previously created kprobes,

it does

Quoted 9 lines
Quoted 9 lines
Quoted 9 lines
Quoted 7 lines

not allow creating kprobes, right?

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

There are 2 ABIs nowadays, ftrace and ebpf. perf-probe uses ftrace

interface to

Quoted 1 line
Quoted 1 line
Quoted 1 line

define new kprobe events, and those events are treated as

completely same as

Quoted 1 line
Quoted 1 line
Quoted 1 line

tracepoint events. On the other hand, ebpf tries to define new

probe event

Quoted 1 line
Quoted 1 line
Quoted 1 line

via perf_event interface. Above one is that interface. IOW, it

creates new kprobe.

Quoted 6 lines
Quoted 3 lines

Masami, any plans to make 'perf probe' use the perf_event_open() interface for creating kprobes/uprobes?

Would you mean perf probe to switch to perf_event_open()? No, perf probe is for setting up the ftrace probe events. I think we

can add an

Quoted 1 line

option to use perf_event_open(). But current kprobe creation from

perf_event_open()

Quoted 1 line

is separated from ftrace by design.

I guess we can extend event parser to understand kprobe directly. Instead of

perf probe kernel_func
perf stat/record -e probe:kernel_func ...

Quoted 6 lines
Quoted 3 lines

We can just do

perf stat/record -e kprobe:kernel_func ...

You took the words from my mouth, exactly, that is a perfect use case, an alternative to the 'perf probe' one of making a disabled event that then gets activated via record/stat/trace, in many cases it's better, removes the explicit probe setup case.

Ah, I got it. If the perf event parser just kicks perf's kprobe creation interface, it will be easy. In that case, there should be following differences.

- perf * -e "kprobe":kernel_func will put a local (hidden) kprobe events. So ftrace user can not access it.
- perf * -e "kprobe":kernel_func may not support inline/function-body nor trace local variables etc.

Hm, if we support inline function via -e "kprobe" interface, we have to expand perf_event_open() to support multi-probe event.

Thanks,

Quoted 9 lines

Regards,

- Arnaldo

Quoted 3 lines

Thanks,
Song

Signature · 1 line
Masami Hiramatsu <[email protected]>
#30 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Peter Zijlstra
Quoted 21 lines · Peter Zijlstra
Quoted 20 lines · Alexey Budankov
Quoted 18 lines · Peter Zijlstra

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 13 lines · Alexey Budankov
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

Quoted 7 lines
Quoted 2 lines

This unblocks creation of local trace kprobes and uprobes by CAP_SYS_PERFMON privileged process, exactly the same as for CAP_SYS_ADMIN privileged process.

I've no idea what you just said; it's just words.

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

Not really, this allows creating a kprobe using perf_event_open syscall that associates file descriptor with the kprobe [1].

Lifetime of that kprobe is equal to the lifetime of the file descriptor and the kprobe is not visible in tracefs: /sys/kernel/debug/tracing/kprobe_events

Quoted 6 lines

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

~Alexey

---

[1] https://lore.kernel.org/lkml/[email protected]/

#31 lore raw

Re: [Intel-gfx] [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Peter Zijlstra
Quoted 21 lines · Peter Zijlstra
Quoted 20 lines · Alexey Budankov
Quoted 18 lines · Peter Zijlstra

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 13 lines · Alexey Budankov
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

Quoted 7 lines
Quoted 2 lines

This unblocks creation of local trace kprobes and uprobes by CAP_SYS_PERFMON privileged process, exactly the same as for CAP_SYS_ADMIN privileged process.

I've no idea what you just said; it's just words.

Again, this only allows attaching to previously created kprobes, it does not allow creating kprobes, right?

Not really, this allows creating a kprobe using perf_event_open syscall that associates file descriptor with the kprobe [1].

Lifetime of that kprobe is equal to the lifetime of the file descriptor and the kprobe is not visible in tracefs: /sys/kernel/debug/tracing/kprobe_events

Quoted 7 lines

That is; I don't think CAP_SYS_PERFMON should be allowed to create kprobes.

As might be clear; I don't actually know what the user-ABI is for creating kprobes.

~Alexey

[1] https://lore.kernel.org/lkml/[email protected]/

#32 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

In reply to Peter Zijlstra

Hi Peter,

Quoted 89 lines · Peter Zijlstra
Quoted 36 lines · Alexey Budankov

Open access to perf_events monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to perf_events subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure perf_events monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 include/linux/perf_event.h | 6 +++---
 kernel/events/core.c       | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/perf_event.h +2 -2
index 34c7c6910026..f46acd69425f 100644
@@ -1285,7 +1285,7 @@ static inline int perf_is_paranoid(void)
 
 static inline int perf_allow_kernel(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
@@ -1293,7 +1293,7 @@ static inline int perf_allow_kernel(struct perf_event_attr *attr)
 
 static inline int perf_allow_cpu(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 0 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_CPU);

include/linux/perf_event.h static inline int perf_allow_cpu(struct perf_event_attr *attr) · line 1307

Quoted 8 lines
include/linux/perf_event.h +1 -1
@@ -1301,7 +1301,7 @@ static inline int perf_allow_cpu(struct perf_event_attr *attr)
 
 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > -1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
 		return -EPERM;
 
 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);

These are OK I suppose.

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 13 lines
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

kernel/events/core.c static int perf_uprobe_event_init(struct perf_event *event) · line 9122

Quoted 9 lines
kernel/events/core.c +1 -1
@@ -9116,7 +9116,7 @@ static int perf_uprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_uprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

Idem, I presume.

kernel/events/core.c SYSCALL_DEFINE5(perf_event_open, · line 11162

Quoted 8 lines
kernel/events/core.c +1 -1
@@ -11157,7 +11157,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (attr.namespaces) {
-		if (!capable(CAP_SYS_ADMIN))
+		if (!perfmon_capable())
 			return -EACCES;
 	}

And given we basically make the entire kernel observable with this CAP, busting namespaces shoulnd't be a problem either.

Quoted 1 line

So yeah, I suppose that works.

Could this have you explicit Reviewed-by or Acked-by tag so the changes could be driven into the kernel? Latest v7 is here:
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Alexey

#33 lore raw

Re: [PATCH v4 2/9] perf/core: open access for CAP_SYS_PERFMON privileged process

Quoted 92 lines · Alexey Budankov

Hi Peter,

Quoted 89 lines · Peter Zijlstra
Quoted 36 lines · Alexey Budankov

Open access to perf_events monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to perf_events subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure perf_events monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 include/linux/perf_event.h | 6 +++---
 kernel/events/core.c       | 6 +++---
 2 files changed, 6 insertions(+), 6 deletions(-)
include/linux/perf_event.h +2 -2
index 34c7c6910026..f46acd69425f 100644
@@ -1285,7 +1285,7 @@ static inline int perf_is_paranoid(void)
 
 static inline int perf_allow_kernel(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 1 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_KERNEL);
@@ -1293,7 +1293,7 @@ static inline int perf_allow_kernel(struct perf_event_attr *attr)
 
 static inline int perf_allow_cpu(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > 0 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > 0 && !perfmon_capable())
 		return -EACCES;
 
 	return security_perf_event_open(attr, PERF_SECURITY_CPU);

include/linux/perf_event.h static inline int perf_allow_cpu(struct perf_event_attr *attr) · line 1307

Quoted 8 lines
include/linux/perf_event.h +1 -1
@@ -1301,7 +1301,7 @@ static inline int perf_allow_cpu(struct perf_event_attr *attr)
 
 static inline int perf_allow_tracepoint(struct perf_event_attr *attr)
 {
-	if (sysctl_perf_event_paranoid > -1 && !capable(CAP_SYS_ADMIN))
+	if (sysctl_perf_event_paranoid > -1 && !perfmon_capable())
 		return -EPERM;
 
 	return security_perf_event_open(attr, PERF_SECURITY_TRACEPOINT);

These are OK I suppose.

kernel/events/core.c static int perf_kprobe_event_init(struct perf_event *event) · line 9062

Quoted 13 lines
kernel/events/core.c +1 -1
index 059ee7116008..d9db414f2197 100644
@@ -9056,7 +9056,7 @@ static int perf_kprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_kprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

This one only allows attaching to already extant kprobes, right? It does not allow creation of kprobes.

kernel/events/core.c static int perf_uprobe_event_init(struct perf_event *event) · line 9122

Quoted 9 lines
kernel/events/core.c +1 -1
@@ -9116,7 +9116,7 @@ static int perf_uprobe_event_init(struct perf_event *event)
 	if (event->attr.type != perf_uprobe.type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/*

Idem, I presume.

kernel/events/core.c SYSCALL_DEFINE5(perf_event_open, · line 11162

Quoted 8 lines
kernel/events/core.c +1 -1
@@ -11157,7 +11157,7 @@ SYSCALL_DEFINE5(perf_event_open,
 	}
 
 	if (attr.namespaces) {
-		if (!capable(CAP_SYS_ADMIN))
+		if (!perfmon_capable())
 			return -EACCES;
 	}

And given we basically make the entire kernel observable with this CAP, busting namespaces shoulnd't be a problem either.

Quoted 6 lines
Quoted 1 line

So yeah, I suppose that works.

Could this have you explicit Reviewed-by or Acked-by tag so the changes could be driven into the kernel? Latest v7 is here:
https://lore.kernel.org/lkml/[email protected]/

Posted v8 with all acquired tags so far:
https://lore.kernel.org/lkml/[email protected]/

Thanks,
Alexey

#34 lore raw

PATCH v4 3/9 perf tool: extend Perf tool with CAP_SYS_PERFMON capability support

In reply to Alexey Budankov

Extend error messages to mention CAP_SYS_PERFMON capability as an option to substitute CAP_SYS_ADMIN capability for secure system performance monitoring and observability operations. Make perf_event_paranoid_check() and __cmd_ftrace() to be aware of CAP_SYS_PERFMON capability.

---
 tools/perf/builtin-ftrace.c |  5 +++--
 tools/perf/design.txt       |  3 ++-
 tools/perf/util/cap.h       |  4 ++++
 tools/perf/util/evsel.c     | 10 +++++-----
 tools/perf/util/util.c      |  1 +
 5 files changed, 15 insertions(+), 8 deletions(-)
tools/perf/builtin-ftrace.c +3 -2
index d5adc417a4ca..8096e9b5f4f9 100644
@@ -284,10 +284,11 @@ static int __cmd_ftrace(struct perf_ftrace *ftrace, int argc, const char **argv)
 		.events = POLLIN,
 	};
 
-	if (!perf_cap__capable(CAP_SYS_ADMIN)) {
+	if (!(perf_cap__capable(CAP_SYS_PERFMON) ||
+	      perf_cap__capable(CAP_SYS_ADMIN))) {
 		pr_err("ftrace only works for %s!\n",
 #ifdef HAVE_LIBCAP_SUPPORT
-		"users with the SYS_ADMIN capability"
+		"users with the CAP_SYS_PERFMON or CAP_SYS_ADMIN capability"
 #else
 		"root"
 #endif
tools/perf/design.txt +2 -1
index 0453ba26cdbd..71755b3e1303 100644
@@ -258,7 +258,8 @@ gets schedule to. Per task counters can be created by any user, for
 their own tasks.
 
 A 'pid == -1' and 'cpu == x' counter is a per CPU counter that counts
-all events on CPU-x. Per CPU counters need CAP_SYS_ADMIN privilege.
+all events on CPU-x. Per CPU counters need CAP_SYS_PERFMON or
+CAP_SYS_ADMIN privilege.
 
 The 'flags' parameter is currently unused and must be zero.
 
tools/perf/util/cap.h +4 -0
index 051dc590ceee..0f79fbf6638b 100644
@@ -29,4 +29,8 @@ static inline bool perf_cap__capable(int cap __maybe_unused)
 #define CAP_SYSLOG	34
 #endif
 
+#ifndef CAP_SYS_PERFMON
+#define CAP_SYS_PERFMON 38
+#endif
+
 #endif /* __PERF_CAP_H */
tools/perf/util/evsel.c +5 -5
index f4dea055b080..3a46325e3702 100644
@@ -2468,14 +2468,14 @@ int perf_evsel__open_strerror(struct evsel *evsel, struct target *target,
 		 "You may not have permission to collect %sstats.\n\n"
 		 "Consider tweaking /proc/sys/kernel/perf_event_paranoid,\n"
 		 "which controls use of the performance events system by\n"
-		 "unprivileged users (without CAP_SYS_ADMIN).\n\n"
+		 "unprivileged users (without CAP_SYS_PERFMON or CAP_SYS_ADMIN).\n\n"
 		 "The current value is %d:\n\n"
 		 "  -1: Allow use of (almost) all events by all users\n"
 		 "      Ignore mlock limit after perf_event_mlock_kb without CAP_IPC_LOCK\n"
-		 ">= 0: Disallow ftrace function tracepoint by users without CAP_SYS_ADMIN\n"
-		 "      Disallow raw tracepoint access by users without CAP_SYS_ADMIN\n"
-		 ">= 1: Disallow CPU event access by users without CAP_SYS_ADMIN\n"
-		 ">= 2: Disallow kernel profiling by users without CAP_SYS_ADMIN\n\n"
+		 ">= 0: Disallow ftrace function tracepoint by users without CAP_SYS_PERFMON or CAP_SYS_ADMIN\n"
+		 "      Disallow raw tracepoint access by users without CAP_SYS_PERFMON or CAP_SYS_ADMIN\n"
+		 ">= 1: Disallow CPU event access by users without CAP_SYS_PERFMON or CAP_SYS_ADMIN\n"
+		 ">= 2: Disallow kernel profiling by users without CAP_SYS_PERFMON or CAP_SYS_ADMIN\n\n"
 		 "To make this setting permanent, edit /etc/sysctl.conf too, e.g.:\n\n"
 		 "	kernel.perf_event_paranoid = -1\n" ,
 				 target->system_wide ? "system-wide " : "",
tools/perf/util/util.c +1 -0
index 969ae560dad9..9981db0d8d09 100644
@@ -272,6 +272,7 @@ int perf_event_paranoid(void)
 bool perf_event_paranoid_check(int max_level)
 {
 	return perf_cap__capable(CAP_SYS_ADMIN) ||
+			perf_cap__capable(CAP_SYS_PERFMON) ||
 			perf_event_paranoid() <= max_level;
 }
 
Signature · 1 line
2.20.1
#35 lore raw

PATCH v4 4/9 drm/i915/perf: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to i915_perf monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to i915_perf subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure i915_perf monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 drivers/gpu/drm/i915/i915_perf.c | 13 ++++++-------
 1 file changed, 6 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_perf.c +6 -7
index e42b86827d6b..e2697f8d04de 100644
@@ -2748,10 +2748,10 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv,
 	/* Similar to perf's kernel.perf_paranoid_cpu sysctl option
 	 * we check a dev.i915.perf_stream_paranoid sysctl option
 	 * to determine if it's ok to access system wide OA counters
-	 * without CAP_SYS_ADMIN privileges.
+	 * without CAP_SYS_PERFMON or CAP_SYS_ADMIN privileges.
 	 */
 	if (privileged_op &&
-	    i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
+	    i915_perf_stream_paranoid && !perfmon_capable()) {
 		DRM_DEBUG("Insufficient privileges to open system-wide i915 perf stream\n");
 		ret = -EACCES;
 		goto err_ctx;
@@ -2939,9 +2939,8 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
 			} else
 				oa_freq_hz = 0;
 
-			if (oa_freq_hz > i915_oa_max_sample_rate &&
-			    !capable(CAP_SYS_ADMIN)) {
-				DRM_DEBUG("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
+			if (oa_freq_hz > i915_oa_max_sample_rate && !perfmon_capable()) {
+				DRM_DEBUG("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without CAP_SYS_PERFMON or CAP_SYS_ADMIN privileges\n",
 					  i915_oa_max_sample_rate);
 				return -EACCES;
 			}
@@ -3328,7 +3327,7 @@ int i915_perf_add_config_ioctl(struct drm_device *dev, void *data,
 		return -EINVAL;
 	}
 
-	if (i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
+	if (i915_perf_stream_paranoid && !perfmon_capable()) {
 		DRM_DEBUG("Insufficient privileges to add i915 OA config\n");
 		return -EACCES;
 	}
@@ -3474,7 +3473,7 @@ int i915_perf_remove_config_ioctl(struct drm_device *dev, void *data,
 		return -ENOTSUPP;
 	}
 
-	if (i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
+	if (i915_perf_stream_paranoid && !perfmon_capable()) {
 		DRM_DEBUG("Insufficient privileges to remove i915 OA config\n");
 		return -EACCES;
 	}
Signature · 1 line
2.20.1
Acked-by on v4 4/9
#36 lore raw

Re: [PATCH v4 4/9] drm/i915/perf: open access for CAP_SYS_PERFMON privileged process

Quoted 7 lines · Alexey Budankov

Open access to i915_perf monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to i915_perf subsystem remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure i915_perf monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

Quoted 51 lines
---
  drivers/gpu/drm/i915/i915_perf.c | 13 ++++++-------
  1 file changed, 6 insertions(+), 7 deletions(-)
drivers/gpu/drm/i915/i915_perf.c +6 -7
index e42b86827d6b..e2697f8d04de 100644
@@ -2748,10 +2748,10 @@ i915_perf_open_ioctl_locked(struct drm_i915_private *dev_priv,
  	/* Similar to perf's kernel.perf_paranoid_cpu sysctl option
  	 * we check a dev.i915.perf_stream_paranoid sysctl option
  	 * to determine if it's ok to access system wide OA counters
-	 * without CAP_SYS_ADMIN privileges.
+	 * without CAP_SYS_PERFMON or CAP_SYS_ADMIN privileges.
  	 */
  	if (privileged_op &&
-	    i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
+	    i915_perf_stream_paranoid && !perfmon_capable()) {
  		DRM_DEBUG("Insufficient privileges to open system-wide i915 perf stream\n");
  		ret = -EACCES;
  		goto err_ctx;
@@ -2939,9 +2939,8 @@ static int read_properties_unlocked(struct drm_i915_private *dev_priv,
  			} else
  				oa_freq_hz = 0;
 
-			if (oa_freq_hz > i915_oa_max_sample_rate &&
-			    !capable(CAP_SYS_ADMIN)) {
-				DRM_DEBUG("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without root privileges\n",
+			if (oa_freq_hz > i915_oa_max_sample_rate && !perfmon_capable()) {
+				DRM_DEBUG("OA exponent would exceed the max sampling frequency (sysctl dev.i915.oa_max_sample_rate) %uHz without CAP_SYS_PERFMON or CAP_SYS_ADMIN privileges\n",
  					  i915_oa_max_sample_rate);
  				return -EACCES;
  			}
@@ -3328,7 +3327,7 @@ int i915_perf_add_config_ioctl(struct drm_device *dev, void *data,
  		return -EINVAL;
  	}
 
-	if (i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
+	if (i915_perf_stream_paranoid && !perfmon_capable()) {
  		DRM_DEBUG("Insufficient privileges to add i915 OA config\n");
  		return -EACCES;
  	}
@@ -3474,7 +3473,7 @@ int i915_perf_remove_config_ioctl(struct drm_device *dev, void *data,
  		return -ENOTSUPP;
  	}
 
-	if (i915_perf_stream_paranoid && !capable(CAP_SYS_ADMIN)) {
+	if (i915_perf_stream_paranoid && !perfmon_capable()) {
  		DRM_DEBUG("Insufficient privileges to remove i915 OA config\n");
  		return -EACCES;
  	}
#37 lore raw

PATCH v4 5/9 trace/bpf_trace: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to bpf_trace monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to bpf_trace monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure bpf_trace monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 kernel/trace/bpf_trace.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
kernel/trace/bpf_trace.c +1 -1
index 44bd08f2443b..bafe21ac6d92 100644
@@ -1272,7 +1272,7 @@ int perf_event_query_prog_array(struct perf_event *event, void __user *info)
 	u32 *ids, prog_cnt, ids_len;
 	int ret;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EPERM;
 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
 		return -EINVAL;
Signature · 1 line
2.20.1
Acked-by on v4 5/9
#38 lore raw

Re: [Intel-gfx] [PATCH v4 5/9] trace/bpf_trace: open access for CAP_SYS_PERFMON privileged process

Quoted 12 lines · Alexey Budankov

Open access to bpf_trace monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to bpf_trace monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure bpf_trace monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

Quoted 9 lines

---
kernel/trace/bpf_trace.c | 2 +-

1 file changed, 1 insertion(+), 1 deletion(-)
kernel/trace/bpf_trace.c +0 -0
index 44bd08f2443b..bafe21ac6d92 100644

kernel/trace/bpf_trace.c int perf_event_query_prog_array(struct perf_event *event, void __user *info) · line 1278

Quoted 8 lines
kernel/trace/bpf_trace.c +1 -1
@@ -1272,7 +1272,7 @@ int perf_event_query_prog_array(struct perf_event *event, void __user *info)
 	u32 *ids, prog_cnt, ids_len;
 	int ret;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EPERM;
 	if (event->attr.type != PERF_TYPE_TRACEPOINT)
 		return -EINVAL;

I guess we need to fix this check for kprobe/uprobe created with perf_event_open()...

Thanks,
Song

#39 lore raw

PATCH v4 6/9 powerpc/perf: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 arch/powerpc/perf/imc-pmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
arch/powerpc/perf/imc-pmu.c +2 -2
index cb50a9e1fd2d..e837717492e4 100644
@@ -898,7 +898,7 @@ static int thread_imc_event_init(struct perf_event *event)
 	if (event->attr.type != event->pmu->type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/* Sampling not supported */
@@ -1307,7 +1307,7 @@ static int trace_imc_event_init(struct perf_event *event)
 	if (event->attr.type != event->pmu->type)
 		return -ENOENT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	/* Return if this is a couting event */
Signature · 1 line
2.20.1
#40 lore raw

PATCH v4 7/9 parisc/perf: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 arch/parisc/kernel/perf.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/parisc/kernel/perf.c +1 -1
index 676683641d00..c4208d027794 100644
@@ -300,7 +300,7 @@ static ssize_t perf_write(struct file *file, const char __user *buf,
 	else
 		return -EFAULT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	if (count != sizeof(uint32_t))
Signature · 1 line
2.20.1
maintainer Acked-by on v4 7/9
#41 lore raw

Re: [PATCH v4 7/9] parisc/perf: open access for CAP_SYS_PERFMON privileged process

Quoted 7 lines · Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

Quoted 18 lines
---
 arch/parisc/kernel/perf.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
arch/parisc/kernel/perf.c +1 -1
index 676683641d00..c4208d027794 100644
@@ -300,7 +300,7 @@ static ssize_t perf_write(struct file *file, const char __user *buf,
 	else
 		return -EFAULT;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EACCES;
 
 	if (count != sizeof(uint32_t))
#42 lore raw

PATCH v4 8/9 drivers/perf: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 drivers/perf/arm_spe_pmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/perf/arm_spe_pmu.c +2 -2
index 4e4984a55cd1..5dff81bc3324 100644
@@ -274,7 +274,7 @@ static u64 arm_spe_event_to_pmscr(struct perf_event *event)
 	if (!attr->exclude_kernel)
 		reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT);
 
-	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && capable(CAP_SYS_ADMIN))
+	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && perfmon_capable())
 		reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT);
 
 	return reg;
@@ -700,7 +700,7 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
 		return -EOPNOTSUPP;
 
 	reg = arm_spe_event_to_pmscr(event);
-	if (!capable(CAP_SYS_ADMIN) &&
+	if (!perfmon_capable() &&
 	    (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_CX_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_PCT_SHIFT))))
Signature · 1 line
2.20.1
maintainer Acked-by on v4 8/9
#43 lore raw

Re: [Intel-gfx] [PATCH v4 8/9] drivers/perf: open access for CAP_SYS_PERFMON privileged process

Quoted 25 lines · Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 drivers/perf/arm_spe_pmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/perf/arm_spe_pmu.c +1 -1
index 4e4984a55cd1..5dff81bc3324 100644
@@ -274,7 +274,7 @@ static u64 arm_spe_event_to_pmscr(struct perf_event *event)
 	if (!attr->exclude_kernel)
 		reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT);
 
-	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && capable(CAP_SYS_ADMIN))
+	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && perfmon_capable())
 		reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT);
 
 	return reg;

drivers/perf/arm_spe_pmu.c static int arm_spe_pmu_event_init(struct perf_event *event) · line 706

Quoted 8 lines
drivers/perf/arm_spe_pmu.c +1 -1
@@ -700,7 +700,7 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
 		return -EOPNOTSUPP;
 
 	reg = arm_spe_event_to_pmscr(event);
-	if (!capable(CAP_SYS_ADMIN) &&
+	if (!perfmon_capable() &&
 	    (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_CX_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_PCT_SHIFT))))

Worth noting that this allows profiling of *physical* addresses used by memory access instructions and so probably has some security implications beyond the usual "but perf is buggy" line of reasoning.

Will

#44 lore raw

Re: [Intel-gfx] [PATCH v4 8/9] drivers/perf: open access for CAP_SYS_PERFMON privileged process

Quoted 37 lines · Will Deacon
Quoted 25 lines · Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 drivers/perf/arm_spe_pmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/perf/arm_spe_pmu.c +1 -1
index 4e4984a55cd1..5dff81bc3324 100644
@@ -274,7 +274,7 @@ static u64 arm_spe_event_to_pmscr(struct perf_event *event)
 	if (!attr->exclude_kernel)
 		reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT);
 
-	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && capable(CAP_SYS_ADMIN))
+	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && perfmon_capable())
 		reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT);
 
 	return reg;

drivers/perf/arm_spe_pmu.c static int arm_spe_pmu_event_init(struct perf_event *event) · line 706

Quoted 8 lines
drivers/perf/arm_spe_pmu.c +1 -1
@@ -700,7 +700,7 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
 		return -EOPNOTSUPP;
 
 	reg = arm_spe_event_to_pmscr(event);
-	if (!capable(CAP_SYS_ADMIN) &&
+	if (!perfmon_capable() &&
 	    (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_CX_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_PCT_SHIFT))))
Quoted 3 lines

Worth noting that this allows profiling of *physical* addresses used by memory access instructions and so probably has some security implications beyond the usual "but perf is buggy" line of reasoning.

Good to know. Thank you!
The data on physical addresses used by memory access instructions can already be provided under CAP_SYS_ADMIN privileges [1] thus, I suppose, any implications you have mentioned are already in place. I believe providing the data under CAP_PERFMON alone without the rest of CAP_SYS_ADMIN credentials decreases chances to misuse the data for harm and makes the monitoring more secure.

~Alexey

[1] https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html

Quoted 3 lines

Will

#45 lore raw

Re: [PATCH v4 8/9] drivers/perf: open access for CAP_SYS_PERFMON privileged process

In reply to Will Deacon
Quoted 37 lines · Will Deacon
Quoted 25 lines · Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 drivers/perf/arm_spe_pmu.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
drivers/perf/arm_spe_pmu.c +1 -1
index 4e4984a55cd1..5dff81bc3324 100644
@@ -274,7 +274,7 @@ static u64 arm_spe_event_to_pmscr(struct perf_event *event)
 	if (!attr->exclude_kernel)
 		reg |= BIT(SYS_PMSCR_EL1_E1SPE_SHIFT);
 
-	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && capable(CAP_SYS_ADMIN))
+	if (IS_ENABLED(CONFIG_PID_IN_CONTEXTIDR) && perfmon_capable())
 		reg |= BIT(SYS_PMSCR_EL1_CX_SHIFT);
 
 	return reg;

drivers/perf/arm_spe_pmu.c static int arm_spe_pmu_event_init(struct perf_event *event) · line 706

Quoted 8 lines
drivers/perf/arm_spe_pmu.c +1 -1
@@ -700,7 +700,7 @@ static int arm_spe_pmu_event_init(struct perf_event *event)
 		return -EOPNOTSUPP;
 
 	reg = arm_spe_event_to_pmscr(event);
-	if (!capable(CAP_SYS_ADMIN) &&
+	if (!perfmon_capable() &&
 	    (reg & (BIT(SYS_PMSCR_EL1_PA_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_CX_SHIFT) |
 		    BIT(SYS_PMSCR_EL1_PCT_SHIFT))))
Quoted 3 lines

Worth noting that this allows profiling of *physical* addresses used by memory access instructions and so probably has some security implications beyond the usual "but perf is buggy" line of reasoning.

Good to know. Thank you!
The data on physical addresses used by memory access instructions can already be provided under CAP_SYS_ADMIN privileges [1] thus, I suppose, any implications you have mentioned are already in place. I believe providing the data under CAP_PERFMON alone without the rest of CAP_SYS_ADMIN credentials decreases chances to misuse the data for harm and makes the monitoring more secure.

~Alexey

[1] https://www.kernel.org/doc/html/latest/admin-guide/perf-security.html

Quoted 3 lines

Will

#46 lore raw

PATCH v4 9/9 drivers/oprofile: open access for CAP_SYS_PERFMON privileged process

In reply to Alexey Budankov

Open access to monitoring for CAP_SYS_PERFMON privileged processes. For backward compatibility reasons access to the monitoring remains open for CAP_SYS_ADMIN privileged processes but CAP_SYS_ADMIN usage for secure monitoring is discouraged with respect to CAP_SYS_PERFMON capability.

---
 drivers/oprofile/event_buffer.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
drivers/oprofile/event_buffer.c +1 -1
index 12ea4a4ad607..6c9edc8bbc95 100644
@@ -113,7 +113,7 @@ static int event_buffer_open(struct inode *inode, struct file *file)
 {
 	int err = -EPERM;
 
-	if (!capable(CAP_SYS_ADMIN))
+	if (!perfmon_capable())
 		return -EPERM;
 
 	if (test_and_set_bit_lock(0, &buffer_opened))
Signature · 1 line
2.20.1
#47 lore raw

[Intel-gfx] ✗ Fi.CI.BUILD: failure for Introduce CAP_SYS_PERFMON to secure system performance monitoring and observability (rev3)

In reply to Alexey Budankov
== Series Details ==

== Series Details ==

Series: Introduce CAP_SYS_PERFMON to secure system performance monitoring and observability (rev3)
URL : https://patchwork.freedesktop.org/series/70961/
State : failure

== Summary ==

Applying: capabilities: introduce CAP_SYS_PERFMON to kernel and user space
Applying: perf/core: open access for CAP_SYS_PERFMON privileged process
Applying: perf tool: extend Perf tool with CAP_SYS_PERFMON capability support
Applying: drm/i915/perf: open access for CAP_SYS_PERFMON privileged process Using index info to reconstruct a base tree...
M drivers/gpu/drm/i915/i915_perf.c
Falling back to patching base and 3-way merge...
Auto-merging drivers/gpu/drm/i915/i915_perf.c
CONFLICT (content): Merge conflict in drivers/gpu/drm/i915/i915_perf.c
error: Failed to merge in the changes.
hint: Use 'git am --show-current-patch' to see the failed patch
Patch failed at 0004 drm/i915/perf: open access for CAP_SYS_PERFMON privileged process When you have resolved this problem, run "git am --continue".
If you prefer to skip this patch, run "git am --skip" instead.
To restore the original branch and stop patching, run "git am --abort".

_______________________________________________
Intel-gfx mailing list
[email protected]
https://lists.freedesktop.org/mailman/listinfo/intel-gfx