3 Perf Events and tool security
4 =============================
9 Usage of Performance Counters for Linux (perf_events) [1]_ , [2]_ , [3]_
10 can impose a considerable risk of leaking sensitive data accessed by
11 monitored processes. The data leakage is possible both in scenarios of
12 direct usage of perf_events system call API [2]_ and over data files
13 generated by Perf tool user mode utility (Perf) [3]_ , [4]_ . The risk
14 depends on the nature of data that perf_events performance monitoring
15 units (PMU) [2]_ and Perf collect and expose for performance analysis.
16 Collected system and performance data may be split into several
19 1. System hardware and software configuration data, for example: a CPU
20 model and its cache configuration, an amount of available memory and
21 its topology, used kernel and Perf versions, performance monitoring
22 setup including experiment time, events configuration, Perf command
25 2. User and kernel module paths and their load addresses with sizes,
26 process and thread names with their PIDs and TIDs, timestamps for
27 captured hardware and software events.
29 3. Content of kernel software counters (e.g., for context switches, page
30 faults, CPU migrations), architectural hardware performance counters
31 (PMC) [8]_ and machine specific registers (MSR) [9]_ that provide
32 execution metrics for various monitored parts of the system (e.g.,
33 memory controller (IMC), interconnect (QPI/UPI) or peripheral (PCIe)
34 uncore counters) without direct attribution to any execution context
37 4. Content of architectural execution context registers (e.g., RIP, RSP,
38 RBP on x86_64), process user and kernel space memory addresses and
39 data, content of various architectural MSRs that capture data from
42 Data that belong to the fourth category can potentially contain
43 sensitive process data. If PMUs in some monitoring modes capture values
44 of execution context registers or data from process memory then access
45 to such monitoring capabilities requires to be ordered and secured
46 properly. So, perf_events/Perf performance monitoring is the subject for
47 security access control management [5]_ .
49 perf_events/Perf access control
50 -------------------------------
52 To perform security checks, the Linux implementation splits processes
53 into two categories [6]_ : a) privileged processes (whose effective user
54 ID is 0, referred to as superuser or root), and b) unprivileged
55 processes (whose effective UID is nonzero). Privileged processes bypass
56 all kernel security permission checks so perf_events performance
57 monitoring is fully available to privileged processes without access,
58 scope and resource restrictions.
60 Unprivileged processes are subject to a full security permission check
61 based on the process's credentials [5]_ (usually: effective UID,
62 effective GID, and supplementary group list).
64 Linux divides the privileges traditionally associated with superuser
65 into distinct units, known as capabilities [6]_ , which can be
66 independently enabled and disabled on per-thread basis for processes and
67 files of unprivileged users.
69 Unprivileged processes with enabled CAP_SYS_ADMIN capability are treated
70 as privileged processes with respect to perf_events performance
71 monitoring and bypass *scope* permissions checks in the kernel.
73 Unprivileged processes using perf_events system call API is also subject
74 for PTRACE_MODE_READ_REALCREDS ptrace access mode check [7]_ , whose
75 outcome determines whether monitoring is permitted. So unprivileged
76 processes provided with CAP_SYS_PTRACE capability are effectively
77 permitted to pass the check.
79 Other capabilities being granted to unprivileged processes can
80 effectively enable capturing of additional data required for later
81 performance analysis of monitored processes or a system. For example,
82 CAP_SYSLOG capability permits reading kernel space memory addresses from
85 perf_events/Perf privileged users
86 ---------------------------------
88 Mechanisms of capabilities, privileged capability-dumb files [6]_ and
89 file system ACLs [10]_ can be used to create a dedicated group of
90 perf_events/Perf privileged users who are permitted to execute
91 performance monitoring without scope limits. The following steps can be
92 taken to create such a group of privileged Perf users.
94 1. Create perf_users group of privileged Perf users, assign perf_users
95 group to Perf tool executable and limit access to the executable for
96 other users in the system who are not in the perf_users group:
100 # groupadd perf_users
102 -rwxr-xr-x 2 root root 11M Oct 19 15:12 perf
103 # chgrp perf_users perf
105 -rwxr-xr-x 2 root perf_users 11M Oct 19 15:12 perf
108 -rwxr-x--- 2 root perf_users 11M Oct 19 15:12 perf
110 2. Assign the required capabilities to the Perf tool executable file and
111 enable members of perf_users group with performance monitoring
116 # setcap "cap_sys_admin,cap_sys_ptrace,cap_syslog=ep" perf
117 # setcap -v "cap_sys_admin,cap_sys_ptrace,cap_syslog=ep" perf
120 perf = cap_sys_ptrace,cap_sys_admin,cap_syslog+ep
122 As a result, members of perf_users group are capable of conducting
123 performance monitoring by using functionality of the configured Perf
124 tool executable that, when executes, passes perf_events subsystem scope
127 This specific access control management is only available to superuser
128 or root running processes with CAP_SETPCAP, CAP_SETFCAP [6]_
131 perf_events/Perf unprivileged users
132 -----------------------------------
134 perf_events/Perf *scope* and *access* control for unprivileged processes
135 is governed by perf_event_paranoid [2]_ setting:
138 Impose no *scope* and *access* restrictions on using perf_events
139 performance monitoring. Per-user per-cpu perf_event_mlock_kb [2]_
140 locking limit is ignored when allocating memory buffers for storing
141 performance data. This is the least secure mode since allowed
142 monitored *scope* is maximized and no perf_events specific limits
143 are imposed on *resources* allocated for performance monitoring.
146 *scope* includes per-process and system wide performance monitoring
147 but excludes raw tracepoints and ftrace function tracepoints
148 monitoring. CPU and system events happened when executing either in
149 user or in kernel space can be monitored and captured for later
150 analysis. Per-user per-cpu perf_event_mlock_kb locking limit is
151 imposed but ignored for unprivileged processes with CAP_IPC_LOCK
155 *scope* includes per-process performance monitoring only and
156 excludes system wide performance monitoring. CPU and system events
157 happened when executing either in user or in kernel space can be
158 monitored and captured for later analysis. Per-user per-cpu
159 perf_event_mlock_kb locking limit is imposed but ignored for
160 unprivileged processes with CAP_IPC_LOCK capability.
163 *scope* includes per-process performance monitoring only. CPU and
164 system events happened when executing in user space only can be
165 monitored and captured for later analysis. Per-user per-cpu
166 perf_event_mlock_kb locking limit is imposed but ignored for
167 unprivileged processes with CAP_IPC_LOCK capability.
169 perf_events/Perf resource control
170 ---------------------------------
172 Open file descriptors
173 +++++++++++++++++++++
175 The perf_events system call API [2]_ allocates file descriptors for
176 every configured PMU event. Open file descriptors are a per-process
177 accountable resource governed by the RLIMIT_NOFILE [11]_ limit
178 (ulimit -n), which is usually derived from the login shell process. When
179 configuring Perf collection for a long list of events on a large server
180 system, this limit can be easily hit preventing required monitoring
181 configuration. RLIMIT_NOFILE limit can be increased on per-user basis
182 modifying content of the limits.conf file [12]_ . Ordinarily, a Perf
183 sampling session (perf record) requires an amount of open perf_event
184 file descriptors that is not less than the number of monitored events
185 multiplied by the number of monitored CPUs.
190 The amount of memory available to user processes for capturing
191 performance monitoring data is governed by the perf_event_mlock_kb [2]_
192 setting. This perf_event specific resource setting defines overall
193 per-cpu limits of memory allowed for mapping by the user processes to
194 execute performance monitoring. The setting essentially extends the
195 RLIMIT_MEMLOCK [11]_ limit, but only for memory regions mapped
196 specifically for capturing monitored performance events and related data.
198 For example, if a machine has eight cores and perf_event_mlock_kb limit
199 is set to 516 KiB, then a user process is provided with 516 KiB * 8 =
200 4128 KiB of memory above the RLIMIT_MEMLOCK limit (ulimit -l) for
201 perf_event mmap buffers. In particular, this means that, if the user
202 wants to start two or more performance monitoring processes, the user is
203 required to manually distribute the available 4128 KiB between the
204 monitoring processes, for example, using the --mmap-pages Perf record
205 mode option. Otherwise, the first started performance monitoring process
206 allocates all available 4128 KiB and the other processes will fail to
207 proceed due to the lack of memory.
209 RLIMIT_MEMLOCK and perf_event_mlock_kb resource constraints are ignored
210 for processes with the CAP_IPC_LOCK capability. Thus, perf_events/Perf
211 privileged users can be provided with memory above the constraints for
212 perf_events/Perf performance monitoring purpose by providing the Perf
213 executable with CAP_IPC_LOCK capability.
218 .. [1] `<https://lwn.net/Articles/337493/>`_
219 .. [2] `<http://man7.org/linux/man-pages/man2/perf_event_open.2.html>`_
220 .. [3] `<http://web.eece.maine.edu/~vweaver/projects/perf_events/>`_
221 .. [4] `<https://perf.wiki.kernel.org/index.php/Main_Page>`_
222 .. [5] `<https://www.kernel.org/doc/html/latest/security/credentials.html>`_
223 .. [6] `<http://man7.org/linux/man-pages/man7/capabilities.7.html>`_
224 .. [7] `<http://man7.org/linux/man-pages/man2/ptrace.2.html>`_
225 .. [8] `<https://en.wikipedia.org/wiki/Hardware_performance_counter>`_
226 .. [9] `<https://en.wikipedia.org/wiki/Model-specific_register>`_
227 .. [10] `<http://man7.org/linux/man-pages/man5/acl.5.html>`_
228 .. [11] `<http://man7.org/linux/man-pages/man2/getrlimit.2.html>`_
229 .. [12] `<http://man7.org/linux/man-pages/man5/limits.conf.5.html>`_