1CAPABILITIES(7) Linux Programmer's Manual CAPABILITIES(7)
2
3
4
6 capabilities - overview of Linux capabilities
7
9 For the purpose of performing permission checks, traditional Unix
10 implementations distinguish two categories of processes: privileged
11 processes (whose effective user ID is 0, referred to as superuser or
12 root), and unprivileged processes (whose effective UID is non-zero).
13 Privileged processes bypass all kernel permission checks, while unpriv‐
14 ileged processes are subject to full permission checking based on the
15 process's credentials (usually: effective UID, effective GID, and sup‐
16 plementary group list).
17
18 Starting with kernel 2.2, Linux divides the privileges traditionally
19 associated with superuser into distinct units, known as capabilities,
20 which can be independently enabled and disabled. Capabilities are a
21 per-thread attribute.
22
23 Capabilities List
24 The following list shows the capabilities implemented on Linux, and the
25 operations or behaviors that each capability permits:
26
27 CAP_AUDIT_CONTROL (since Linux 2.6.11)
28 Enable and disable kernel auditing; change auditing filter
29 rules; retrieve auditing status and filtering rules.
30
31 CAP_AUDIT_WRITE (since Linux 2.6.11)
32 Write records to kernel auditing log.
33
34 CAP_CHOWN
35 Make arbitrary changes to file UIDs and GIDs (see chown(2)).
36
37 CAP_DAC_OVERRIDE
38 Bypass file read, write, and execute permission checks. (DAC is
39 an abbreviation of "discretionary access control".)
40
41 CAP_DAC_READ_SEARCH
42 Bypass file read permission checks and directory read and exe‐
43 cute permission checks.
44
45 CAP_FOWNER
46 * Bypass permission checks on operations that normally require
47 the file system UID of the process to match the UID of the
48 file (e.g., chmod(2), utime(2)), excluding those operations
49 covered by CAP_DAC_OVERRIDE and CAP_DAC_READ_SEARCH;
50 * set extended file attributes (see chattr(1)) on arbitrary
51 files;
52 * set Access Control Lists (ACLs) on arbitrary files;
53 * ignore directory sticky bit on file deletion;
54 * specify O_NOATIME for arbitrary files in open(2) and fcntl(2).
55
56 CAP_FSETID
57 Don't clear set-user-ID and set-group-ID permission bits when a
58 file is modified; set the set-group-ID bit for a file whose GID
59 does not match the file system or any of the supplementary GIDs
60 of the calling process.
61
62 CAP_IPC_LOCK
63 Lock memory (mlock(2), mlockall(2), mmap(2), shmctl(2)).
64
65 CAP_IPC_OWNER
66 Bypass permission checks for operations on System V IPC objects.
67
68 CAP_KILL
69 Bypass permission checks for sending signals (see kill(2)).
70 This includes use of the ioctl(2) KDSIGACCEPT operation.
71
72 CAP_LEASE (since Linux 2.4)
73 Establish leases on arbitrary files (see fcntl(2)).
74
75 CAP_LINUX_IMMUTABLE
76 Set the FS_APPEND_FL and FS_IMMUTABLE_FL i-node flags (see
77 chattr(1)).
78
79 CAP_MAC_ADMIN (since Linux 2.6.25)
80 Override Mandatory Access Control (MAC). Implemented for the
81 Smack Linux Security Module (LSM).
82
83 CAP_MAC_OVERRIDE (since Linux 2.6.25)
84 Allow MAC configuration or state changes. Implemented for the
85 Smack LSM.
86
87 CAP_MKNOD (since Linux 2.4)
88 Create special files using mknod(2).
89
90 CAP_NET_ADMIN
91 Perform various network-related operations (e.g., setting privi‐
92 leged socket options, enabling multicasting, interface configu‐
93 ration, modifying routing tables).
94
95 CAP_NET_BIND_SERVICE
96 Bind a socket to Internet domain privileged ports (port numbers
97 less than 1024).
98
99 CAP_NET_BROADCAST
100 (Unused) Make socket broadcasts, and listen to multicasts.
101
102 CAP_NET_RAW
103 Use RAW and PACKET sockets.
104
105 CAP_SETGID
106 Make arbitrary manipulations of process GIDs and supplementary
107 GID list; forge GID when passing socket credentials via Unix
108 domain sockets.
109
110 CAP_SETFCAP (since Linux 2.6.24)
111 Set file capabilities.
112
113 CAP_SETPCAP
114 If file capabilities are not supported: grant or remove any
115 capability in the caller's permitted capability set to or from
116 any other process. (This property of CAP_SETPCAP is not avail‐
117 able when the kernel is configured to support file capabilities,
118 since CAP_SETPCAP has entirely different semantics for such ker‐
119 nels.)
120
121 If file capabilities are supported: add any capability from the
122 calling thread's bounding set to its inheritable set; drop capa‐
123 bilities from the bounding set (via prctl(2) PR_CAPBSET_DROP);
124 make changes to the securebits flags.
125
126 CAP_SETUID
127 Make arbitrary manipulations of process UIDs (setuid(2),
128 setreuid(2), setresuid(2), setfsuid(2)); make forged UID when
129 passing socket credentials via Unix domain sockets.
130
131 CAP_SYS_ADMIN
132 * Perform a range of system administration operations including:
133 quotactl(2), mount(2), umount(2), swapon(2), swapoff(2),
134 sethostname(2), and setdomainname(2);
135 * perform IPC_SET and IPC_RMID operations on arbitrary System V
136 IPC objects;
137 * perform operations on trusted and security Extended Attributes
138 (see attr(5));
139 * use lookup_dcookie(2);
140 * use ioprio_set(2) to assign IOPRIO_CLASS_RT and (before Linux
141 2.6.25) IOPRIO_CLASS_IDLE I/O scheduling classes;
142 * forge UID when passing socket credentials;
143 * exceed /proc/sys/fs/file-max, the system-wide limit on the
144 number of open files, in system calls that open files (e.g.,
145 accept(2), execve(2), open(2), pipe(2));
146 * employ CLONE_NEWNS flag with clone(2) and unshare(2);
147 * perform KEYCTL_CHOWN and KEYCTL_SETPERM keyctl(2) operations.
148
149 CAP_SYS_BOOT
150 Use reboot(2) and kexec_load(2).
151
152 CAP_SYS_CHROOT
153 Use chroot(2).
154
155 CAP_SYS_MODULE
156 Load and unload kernel modules (see init_module(2) and
157 delete_module(2)); in kernels before 2.6.25: drop capabilities
158 from the system-wide capability bounding set.
159
160 CAP_SYS_NICE
161 * Raise process nice value (nice(2), setpriority(2)) and change
162 the nice value for arbitrary processes;
163 * set real-time scheduling policies for calling process, and set
164 scheduling policies and priorities for arbitrary processes
165 (sched_setscheduler(2), sched_setparam(2));
166 * set CPU affinity for arbitrary processes (sched_setaffin‐
167 ity(2));
168 * set I/O scheduling class and priority for arbitrary processes
169 (ioprio_set(2));
170 * apply migrate_pages(2) to arbitrary processes and allow pro‐
171 cesses to be migrated to arbitrary nodes;
172 * apply move_pages(2) to arbitrary processes;
173 * use the MPOL_MF_MOVE_ALL flag with mbind(2) and move_pages(2).
174
175 CAP_SYS_PACCT
176 Use acct(2).
177
178 CAP_SYS_PTRACE
179 Trace arbitrary processes using ptrace(2)
180
181 CAP_SYS_RAWIO
182 Perform I/O port operations (iopl(2) and ioperm(2)); access
183 /proc/kcore.
184
185 CAP_SYS_RESOURCE
186 * Use reserved space on ext2 file systems;
187 * make ioctl(2) calls controlling ext3 journaling;
188 * override disk quota limits;
189 * increase resource limits (see setrlimit(2));
190 * override RLIMIT_NPROC resource limit;
191 * raise msg_qbytes limit for a System V message queue above the
192 limit in /proc/sys/kernel/msgmnb (see msgop(2) and msgctl(2)).
193
194 CAP_SYS_TIME
195 Set system clock (settimeofday(2), stime(2), adjtimex(2)); set
196 real-time (hardware) clock.
197
198 CAP_SYS_TTY_CONFIG
199 Use vhangup(2).
200
201 Past and Current Implementation
202 A full implementation of capabilities requires that:
203
204 1. For all privileged operations, the kernel must check whether the
205 thread has the required capability in its effective set.
206
207 2. The kernel must provide system calls allowing a thread's capability
208 sets to be changed and retrieved.
209
210 3. The file system must support attaching capabilities to an executable
211 file, so that a process gains those capabilities when the file is
212 executed.
213
214 Before kernel 2.6.24, only the first two of these requirements are met;
215 since kernel 2.6.24, all three requirements are met.
216
217 Thread Capability Sets
218 Each thread has three capability sets containing zero or more of the
219 above capabilities:
220
221 Permitted:
222 This is a limiting superset for the effective capabilities that
223 the thread may assume. It is also a limiting superset for the
224 capabilities that may be added to the inheritable set by a
225 thread that does not have the CAP_SETPCAP capability in its
226 effective set.
227
228 If a thread drops a capability from its permitted set, it can
229 never re-acquire that capability (unless it execve(2)s either a
230 set-user-ID-root program, or a program whose associated file
231 capabilities grant that capability).
232
233 Inheritable:
234 This is a set of capabilities preserved across an execve(2). It
235 provides a mechanism for a process to assign capabilities to the
236 permitted set of the new program during an execve(2).
237
238 Effective:
239 This is the set of capabilities used by the kernel to perform
240 permission checks for the thread.
241
242 A child created via fork(2) inherits copies of its parent's capability
243 sets. See below for a discussion of the treatment of capabilities dur‐
244 ing execve(2).
245
246 Using capset(2), a thread may manipulate its own capability sets (see
247 below).
248
249 File Capabilities
250 Since kernel 2.6.24, the kernel supports associating capability sets
251 with an executable file using setcap(8). The file capability sets are
252 stored in an extended attribute (see setxattr(2)) named security.capa‐
253 bility. Writing to this extended attribute requires the CAP_SETFCAP
254 capability. The file capability sets, in conjunction with the capabil‐
255 ity sets of the thread, determine the capabilities of a thread after an
256 execve(2).
257
258 The three file capability sets are:
259
260 Permitted (formerly known as forced):
261 These capabilities are automatically permitted to the thread,
262 regardless of the thread's inheritable capabilities.
263
264 Inheritable (formerly known as allowed):
265 This set is ANDed with the thread's inheritable set to determine
266 which inheritable capabilities are enabled in the permitted set
267 of the thread after the execve(2).
268
269 Effective:
270 This is not a set, but rather just a single bit. If this bit is
271 set, then during an execve(2) all of the new permitted capabili‐
272 ties for the thread are also raised in the effective set. If
273 this bit is not set, then after an execve(2), none of the new
274 permitted capabilities is in the new effective set.
275
276 Enabling the file effective capability bit implies that any file
277 permitted or inheritable capability that causes a thread to
278 acquire the corresponding permitted capability during an
279 execve(2) (see the transformation rules described below) will
280 also acquire that capability in its effective set. Therefore,
281 when assigning capabilities to a file (setcap(8),
282 cap_set_file(3), cap_set_fd(3)), if we specify the effective
283 flag as being enabled for any capability, then the effective
284 flag must also be specified as enabled for all other capabili‐
285 ties for which the corresponding permitted or inheritable flags
286 is enabled.
287
288 Transformation of Capabilities During execve()
289 During an execve(2), the kernel calculates the new capabilities of the
290 process using the following algorithm:
291
292 P'(permitted) = (P(inheritable) & F(inheritable)) |
293 (F(permitted) & cap_bset)
294
295 P'(effective) = F(effective) ? P'(permitted) : 0
296
297 P'(inheritable) = P(inheritable) [i.e., unchanged]
298
299 where:
300
301 P denotes the value of a thread capability set before the
302 execve(2)
303
304 P' denotes the value of a capability set after the execve(2)
305
306 F denotes a file capability set
307
308 cap_bset is the value of the capability bounding set (described
309 below).
310
311 Capabilities and execution of programs by root
312 In order to provide an all-powerful root using capability sets, during
313 an execve(2):
314
315 1. If a set-user-ID-root program is being executed, or the real user ID
316 of the process is 0 (root) then the file inheritable and permitted
317 sets are defined to be all ones (i.e., all capabilities enabled).
318
319 2. If a set-user-ID-root program is being executed, then the file
320 effective bit is defined to be one (enabled).
321
322 The upshot of the above rules, combined with the capabilities transfor‐
323 mations described above, is that when a process execve(2)s a set-user-
324 ID-root program, or when a process with an effective UID of 0
325 execve(2)s a program, it gains all capabilities in its permitted and
326 effective capability sets, except those masked out by the capability
327 bounding set. This provides semantics that are the same as those pro‐
328 vided by traditional Unix systems.
329
330 Capability bounding set
331 The capability bounding set is a security mechanism that can be used to
332 limit the capabilities that can be gained during an execve(2). The
333 bounding set is used in the following ways:
334
335 * During an execve(2), the capability bounding set is ANDed with the
336 file permitted capability set, and the result of this operation is
337 assigned to the thread's permitted capability set. The capability
338 bounding set thus places a limit on the permitted capabilities that
339 may be granted by an executable file.
340
341 * (Since Linux 2.6.25) The capability bounding set acts as a limiting
342 superset for the capabilities that a thread can add to its inherita‐
343 ble set using capset(2). This means that if the capability is not in
344 the bounding set, then a thread can't add one of its permitted capa‐
345 bilities to its inheritable set and thereby have that capability pre‐
346 served in its permitted set when it execve(2)s a file that has the
347 capability in its inheritable set.
348
349 Note that the bounding set masks the file permitted capabilities, but
350 not the inherited capabilities. If a thread maintains a capability in
351 its inherited set that is not in its bounding set, then it can still
352 gain that capability in its permitted set by executing a file that has
353 the capability in its inherited set.
354
355 Depending on the kernel version, the capability bounding set is either
356 a system-wide attribute, or a per-process attribute.
357
358 Capability bounding set prior to Linux 2.6.25
359
360 In kernels before 2.6.25, the capability bounding set is a system-wide
361 attribute that affects all threads on the system. The bounding set is
362 accessible via the file /proc/sys/kernel/cap-bound. (Confusingly, this
363 bit mask parameter is expressed as a signed decimal number in
364 /proc/sys/kernel/cap-bound.)
365
366 Only the init process may set capabilities in the capability bounding
367 set; other than that, the superuser (more precisely: programs with the
368 CAP_SYS_MODULE capability) may only clear capabilities from this set.
369
370 On a standard system the capability bounding set always masks out the
371 CAP_SETPCAP capability. To remove this restriction (dangerous!), mod‐
372 ify the definition of CAP_INIT_EFF_SET in include/linux/capability.h
373 and rebuild the kernel.
374
375 The system-wide capability bounding set feature was added to Linux
376 starting with kernel version 2.2.11.
377
378 Capability bounding set from Linux 2.6.25 onwards
379
380 From Linux 2.6.25, the capability bounding set is a per-thread
381 attribute. (There is no longer a system-wide capability bounding set.)
382
383 The bounding set is inherited at fork(2) from the thread's parent, and
384 is preserved across an execve(2).
385
386 A thread may remove capabilities from its capability bounding set using
387 the prctl(2) PR_CAPBSET_DROP operation, provided it has the CAP_SETPCAP
388 capability. Once a capability has been dropped from the bounding set,
389 it cannot be restored to that set. A thread can determine if a capa‐
390 bility is in its bounding set using the prctl(2) PR_CAPBSET_READ opera‐
391 tion.
392
393 Removing capabilities from the bounding set is only supported if file
394 capabilities are compiled into the kernel (CONFIG_SECURITY_FILE_CAPA‐
395 BILITIES). In that case, the init process (the ancestor of all pro‐
396 cesses) begins with a full bounding set. If file capabilities are not
397 compiled into the kernel, then init begins with a full bounding set
398 minus CAP_SETPCAP, because this capability has a different meaning when
399 there are no file capabilities.
400
401 Removing a capability from the bounding set does not remove it from the
402 thread's inherited set. However it does prevent the capability from
403 being added back into the thread's inherited set in the future.
404
405 Effect of User ID Changes on Capabilities
406 To preserve the traditional semantics for transitions between 0 and
407 non-zero user IDs, the kernel makes the following changes to a thread's
408 capability sets on changes to the thread's real, effective, saved set,
409 and file system user IDs (using setuid(2), setresuid(2), or similar):
410
411 1. If one or more of the real, effective or saved set user IDs was pre‐
412 viously 0, and as a result of the UID changes all of these IDs have
413 a non-zero value, then all capabilities are cleared from the permit‐
414 ted and effective capability sets.
415
416 2. If the effective user ID is changed from 0 to non-zero, then all
417 capabilities are cleared from the effective set.
418
419 3. If the effective user ID is changed from non-zero to 0, then the
420 permitted set is copied to the effective set.
421
422 4. If the file system user ID is changed from 0 to non-zero (see setf‐
423 suid(2)) then the following capabilities are cleared from the effec‐
424 tive set: CAP_CHOWN, CAP_DAC_OVERRIDE, CAP_DAC_READ_SEARCH,
425 CAP_FOWNER, CAP_FSETID, and CAP_MAC_OVERRIDE. If the file system
426 UID is changed from non-zero to 0, then any of these capabilities
427 that are enabled in the permitted set are enabled in the effective
428 set.
429
430 If a thread that has a 0 value for one or more of its user IDs wants to
431 prevent its permitted capability set being cleared when it resets all
432 of its user IDs to non-zero values, it can do so using the prctl(2)
433 PR_SET_KEEPCAPS operation.
434
435 Programmatically adjusting capability sets
436 A thread can retrieve and change its capability sets using the
437 capget(2) and capset(2) system calls. However, the use of
438 cap_get_proc(3) and cap_set_proc(3), both provided in the libcap pack‐
439 age, is preferred for this purpose. The following rules govern changes
440 to the thread capability sets:
441
442 1. If the caller does not have the CAP_SETPCAP capability, the new
443 inheritable set must be a subset of the combination of the existing
444 inheritable and permitted sets.
445
446 2. (Since kernel 2.6.25) The new inheritable set must be a subset of
447 the combination of the existing inheritable set and the capability
448 bounding set.
449
450 3. The new permitted set must be a subset of the existing permitted set
451 (i.e., it is not possible to acquire permitted capabilities that the
452 thread does not currently have).
453
454 4. The new effective set must be a subset of the new permitted set.
455
456 The "securebits" flags: establishing a capabilities-only environment
457 Starting with kernel 2.6.26, and with a kernel in which file capabili‐
458 ties are enabled, Linux implements a set of per-thread securebits flags
459 that can be used to disable special handling of capabilities for UID 0
460 (root). These flags are as follows:
461
462 SECURE_KEEP_CAPS
463 Setting this flag allows a thread that has one or more 0 UIDs to
464 retain its capabilities when it switches all of its UIDs to a
465 non-zero value. If this flag is not set, then such a UID switch
466 causes the thread to lose all capabilities. This flag is always
467 cleared on an execve(2). (This flag provides the same function‐
468 ality as the older prctl(2) PR_SET_KEEPCAPS operation.)
469
470 SECURE_NO_SETUID_FIXUP
471 Setting this flag stops the kernel from adjusting capability
472 sets when the threads's effective and file system UIDs are
473 switched between zero and non-zero values. (See the subsection
474 Effect of User ID Changes on Capabilities.)
475
476 SECURE_NOROOT
477 If this bit is set, then the kernel does not grant capabilities
478 when a set-user-ID-root program is executed, or when a process
479 with an effective or real UID of 0 calls execve(2). (See the
480 subsection Capabilities and execution of programs by root.)
481
482 Each of the above "base" flags has a companion "locked" flag. Setting
483 any of the "locked" flags is irreversible, and has the effect of pre‐
484 venting further changes to the corresponding "base" flag. The locked
485 flags are: SECURE_KEEP_CAPS_LOCKED, SECURE_NO_SETUID_FIXUP_LOCKED, and
486 SECURE_NOROOT_LOCKED.
487
488 The securebits flags can be modified and retrieved using the prctl(2)
489 PR_SET_SECUREBITS and PR_GET_SECUREBITS operations. The CAP_SETPCAP
490 capability is required to modify the flags.
491
492 The securebits flags are inherited by child processes. During an
493 execve(2), all of the flags are preserved, except SECURE_KEEP_CAPS
494 which is always cleared.
495
496 An application can use the following call to lock itself, and all of
497 its descendants, into an environment where the only way of gaining
498 capabilities is by executing a program with associated file capabili‐
499 ties:
500
501 prctl(PR_SET_SECUREBITS,
502 1 << SECURE_KEEP_CAPS_LOCKED |
503 1 << SECURE_NO_SETUID_FIXUP |
504 1 << SECURE_NO_SETUID_FIXUP_LOCKED |
505 1 << SECURE_NOROOT |
506 1 << SECURE_NOROOT_LOCKED);
507
509 No standards govern capabilities, but the Linux capability implementa‐
510 tion is based on the withdrawn POSIX.1e draft standard; see
511 http://wt.xpilot.org/publications/posix.1e/.
512
514 Since kernel 2.5.27, capabilities are an optional kernel component, and
515 can be enabled/disabled via the CONFIG_SECURITY_CAPABILITIES kernel
516 configuration option.
517
518 The /proc/PID/task/TID/status file can be used to view the capability
519 sets of a thread. The /proc/PID/status file shows the capability sets
520 of a process's main thread.
521
522 The libcap package provides a suite of routines for setting and getting
523 capabilities that is more comfortable and less likely to change than
524 the interface provided by capset(2) and capget(2). This package also
525 provides the setcap(8) and getcap(8) programs. It can be found at
526 http://www.kernel.org/pub/linux/libs/security/linux-privs.
527
528 Before kernel 2.6.24, and since kernel 2.6.24 if file capabilities are
529 not enabled, a thread with the CAP_SETPCAP capability can manipulate
530 the capabilities of threads other than itself. However, this is only
531 theoretically possible, since no thread ever has CAP_SETPCAP in either
532 of these cases:
533
534 * In the pre-2.6.25 implementation the system-wide capability bounding
535 set, /proc/sys/kernel/cap-bound, always masks out this capability,
536 and this can not be changed without modifying the kernel source and
537 rebuilding.
538
539 * If file capabilities are disabled in the current implementation, then
540 init starts out with this capability removed from its per-process
541 bounding set, and that bounding set is inherited by all other pro‐
542 cesses created on the system.
543
545 capget(2), prctl(2), setfsuid(2), cap_clear(3), cap_copy_ext(3),
546 cap_from_text(3), cap_get_file(3), cap_get_proc(3), cap_init(3),
547 capgetp(3), capsetp(3), credentials(7), pthreads(7), getcap(8), set‐
548 cap(8)
549
550 include/linux/capability.h in the kernel source
551
553 This page is part of release 3.22 of the Linux man-pages project. A
554 description of the project, and information about reporting bugs, can
555 be found at http://www.kernel.org/doc/man-pages/.
556
557
558
559Linux 2008-11-27 CAPABILITIES(7)