blob: 7c3bc81973b18e0a9e663fdb802b324a37c3e761 [file] [log] [blame]
.TH CAP_GET_PROC 3 "2021-03-06" "" "Linux Programmer's Manual"
cap_get_proc, cap_set_proc, capgetp, cap_get_bound, cap_drop_bound, \
cap_get_ambient, cap_set_ambient, cap_reset_ambient, \
cap_get_secbits, cap_set_secbits, cap_get_mode, cap_set_mode, \
cap_mode_name, cap_get_pid, cap_setuid, cap_prctl, cap_prctlw, cap_setgroups \
\- capability manipulation on processes
#include <sys/capability.h>
cap_t cap_get_proc(void);
int cap_set_proc(cap_t cap_p);
int cap_get_bound(cap_value_t cap);
CAP_IS_SUPPORTED(cap_value_t cap);
int cap_drop_bound(cap_value_t cap);
int cap_get_ambient(cap_value_t cap);
int cap_set_ambient(cap_value_t cap, cap_flag_value_t value);
int cap_reset_ambient(void);
unsigned cap_get_secbits(void);
int cap_set_secbits(unsigned bits);
cap_mode_t cap_get_mode(void);
const char *cap_mode_name(cap_mode_t mode);
int cap_prctl(long int pr_cmd, long int arg1, long int arg2,
long int arg3, long int arg4, long int arg5);
int cap_prctlw(long int pr_cmd, long int arg1, long int arg2,
long int arg3, long int arg4, long int arg5);
int cap_set_mode(cap_mode_t mode);
#include <sys/types.h>
cap_t cap_get_pid(pid_t pid);
int cap_setuid(uid_t uid);
int cap_setgroups(gid_t gid, size_t ngroups, const gid_t groups);
Link with \fI\-lcap\fP.
.BR cap_get_proc ()
allocates a capability state in working storage, sets its state to
that of the calling process, and returns a pointer to this newly
created capability state. The caller should free any releasable
memory, when the capability state in working storage is no longer
required, by calling
.BR cap_free ()
with the
.I cap_t
as an argument.
.BR cap_set_proc ()
sets the values for all capability flags for all capabilities to the
capability state identified by
.IR cap_p .
The new capability state of the process will be completely determined by
the contents of
.I cap_p
upon successful return from this function. If any flag in
.I cap_p
is set for any capability not currently permitted for the calling process,
the function will fail, and the capability state of the process will remain
.BR cap_get_pid ()
.IR cap_t ,
.BR cap_init (3),
with the process capabilities of the process indicated by
.IR pid .
.I pid
is 0, then the calling process's capabilities are returned.)
This information can also be obtained from the
.I /proc/<pid>/status
file. Note, when the caller is operating within a
namespace, the numerical
.I pid
argument is interpreted in the range of that namespace.
.BR cap_get_bound ()
with a
.I cap
as an argument returns the current value of this bounding set
capability flag in effect for the calling process. This operation is
unprivileged. Note, a macro function
.BR "CAP_IS_SUPPORTED(cap_value_t " cap )
is provided that evaluates to true (1) if the system supports the
specified capability,
.IR cap .
If the system does not support the capability, this function returns
0. This macro works by testing for an error condition with
.BR cap_get_bound ().
.BR cap_drop_bound ()
can be used to lower the specified bounding set capability,
.BR cap .
To complete successfully, the prevailing
.I effective
capability set must have a raised
.BR cap_get_ambient ()
returns the prevailing value of the specified ambient capability, or
-1 if the capability is not supported by the running kernel. A macro
uses this function to determine if ambient capabilities are supported
by the kernel.
.BR cap_set_ambient ()
sets the specified ambient capability to a specific value. To complete
successfully, the prevailing
.I effective
capability set must have a raised
Further, to raise a specific ambient capability the
.IR inheritable " and " permitted
sets of the calling process must contain the specified capability, and
raised ambient bits will only be retained as long as this remains true.
.BR cap_reset_ambient ()
resets all of the ambient capabilities for the calling process to
their lowered value. To complete successfully, the prevailing
.I effective
capability set must have a raised
Note, the ambient set is intended to operate in a legacy environment
where the application has limited awareness of capabilities in
general. Executing a file with associated filesystem capabilities, the
kernel will implicitly reset the ambient set of the process. Also,
changes to the inheritable set by the program code without explicitly
fixing up the ambient set can also drop ambient bits.
.BR cap_get_secbits ()
returns the securebits of the calling process. These bits affect the
way in which the calling process implements things like setuid-root
fixup and ambient capabilities.
.BR cap_set_secbits ()
attempts to modify the securebits of the calling process. Note
must be in the effective capability set for this to be effective. Some
settings lock the sub-states of the securebits, so attempts to set values
may be denied by the kernel even when the
capability is raised.
To help manage the complexity of the securebits, libcap provides a
combined securebit and capability set concept called a libcap mode.
.BR cap_get_mode ()
attempts to summarize the prevailing security environment in the form
of a numerical
.B cap_mode_t
value. A text representation of the mode can be obtained via the
.BR cap_mode_name ()
function. The vast majority of combinations of these values are not well
defined in terms of a libcap mode, and for these states
.BR cap_get_mode ()
.RB ( cap_mode_t )0
.BR cap_get_name ()
identifies as
Supported modes are:
.BR cap_prctl ()
can be used to read state via the \fBprctl\fI()\fP system call.
.BR cap_prctlw ()
can be used to write state via the \fBprctl\fI()\fP system call.
.BR cap_set_mode ()
can be used to set the desired mode. The permitted capability
is required for this function to succeed.
.BR cap_setuid ()
is a convenience function for the
.BR setuid (2)
system call. Where
.BR cap_setuid ()
arranges for the right effective capability to be raised in order to
perform the system call, and also arranges to preserve the
availability of permitted capabilities after the uid has
changed. Following this call all effective capabilities are lowered.
.BR cap_setgroups ()
is a convenience function for performing both
.BR setgid (2)
.BR setgroups (2)
calls in one call. The
.BR cap_setgroups ()
call raises the right effective capability for the duration of the
call, and empties the effective capability set before returning.
The functions
.BR cap_get_proc ()
.BR cap_get_pid ()
return a non-NULL value on success, and NULL on failure.
The function
.BR cap_get_bound ()
returns \-1 if the requested capability is unknown, otherwise the
return value reflects the current state of that capability in the
prevailing bounding set. Note, a macro function,
The all of the setting functions such as
.BR cap_set_proc ()
.BR cap_drop_bound ()
return zero for success, and \-1 on failure.
On failure,
.I errno
is set to
.BR cap_set_proc ()
.BR cap_get_proc ()
are specified in the withdrawn POSIX.1e draft specification.
.BR cap_get_pid ()
is a Linux extension.
Neither glibc, nor the Linux kernel honors POSIX semantics for setting
capabilities and securebits in the presence of pthreads. That is,
changing capability sets, by default, only affect the running
thread. To be meaningfully secure, however, the capability sets should
be mirrored by all threads within a common program because threads are
not memory isolated. As a workaround for this,
.B libcap
is packaged with a separate POSIX semantics system call library:
.BR libpsx .
If your program uses POSIX threads, to achieve meaningful POSIX
semantics capability manipulation, you should link your program with:
.B ld ... \-lcap \-lpsx \-lpthread \-\-wrap=pthread_create
.B gcc ... \-lcap \-lpsx \-lpthread \-Wl,\-wrap,pthread_create
When linked this way, due to linker magic, libcap uses
.BR psx_syscall "(3) and " psx_syscall6 (3)
to perform state setting system calls. Notably, this also ensures that
.BI cap_prctlw ()
can be used to ensure process control bits are shared over all threads
of a single process.
.SS capgetp() and capsetp()
The library also supports the deprecated functions:
.BI "int capgetp(pid_t " pid ", cap_t " cap_d );
.BI "int capsetp(pid_t " pid ", cap_t " cap_d );
.BR capgetp ()
attempts to obtain the capabilities of some other process; storing the
capabilities in a pre-allocated
.IR cap_d .
.BR cap_init ()
for information on allocating an empty capability set. This function
is deprecated; you should use
.BR cap_get_pid ().
.BR capsetp ()
attempts to set the capabilities of the calling process or of
some other process(es),
.IR pid .
Note that setting capabilities of another process is only possible on older
kernels that do not provide VFS support for setting file capabilities.
.BR capset (2)
for information on which kernels provide such support.
.I pid
is positive it refers to a specific process; if it is zero, it refers
to the calling process; \-1 refers to all processes other than the
calling process and process '1' (typically
.BR init (8));
other negative values refer to the
.I \-pid
process group.
In order to use this function, the kernel must support
it and the calling process must have
raised in its Effective capability set. The capabilities set in the
target process(es) are those contained in
.IR cap_d .
Kernels that support filesystem capabilities redefine the semantics of
and on such systems,
.BR capsetp ()
will always fail for any target not
equal to the calling process.
.BR capsetp ()
returns zero for success, and \-1 on failure.
On kernels where it is (was) supported,
.BR capsetp ()
should be used with care. It existed, primarily, to overcome an early
lack of support for capabilities in the filesystems supported by
Linux. Note that on older kernels where
.BR capsetp ()
could be used to set the capabilities of another process,
the only processes that had
available to them by default were processes started as kernel threads.
(Typically this includes
.BR init (8),
kflushd and kswapd.) A kernel recompilation was needed to modify
this default.
The code segment below raises the
effective capabilities for the caller:
cap_t caps;
const cap_value_t cap_list[2] = {CAP_FOWNER, CAP_SETFCAP};
/* handle error */
caps = cap_get_proc();
if (caps == NULL)
/* handle error */;
if (cap_set_flag(caps, CAP_EFFECTIVE, 2, cap_list, CAP_SET) == \-1)
/* handle error */;
if (cap_set_proc(caps) == \-1)
/* handle error */;
if (cap_free(caps) == \-1)
/* handle error */;
Alternatively, to completely drop privilege in a program launched
setuid-root but wanting to run as a specific user ID etc. in such a
way that neither it, nor any of its children can acquire privilege
uid_t nobody = 65534;
const gid_t groups[] = {65534};
if (cap_setgroups(groups[0], 1, groups) != 0)
/* handle error */;
if (cap_setuid(nobody) != 0)
/* handle error */;
* privilege is still available here
if (cap_set_mode(CAP_MODE_NOPRIV) != 0)
/* handle error */
Note, the above sequence can be performed by the
.B capsh
tool as follows:
.B sudo /sbin/capsh \-\-user=nobody \-\-mode=NOPRIV \-\-print
.B \-\-print
displays the resulting privilege state.
.BR libcap (3),
.BR libpsx (3),
.BR capsh (1),
.BR cap_clear (3),
.BR cap_copy_ext (3),
.BR cap_from_text (3),
.BR cap_get_file (3),
.BR cap_init (3),
.BR namespaces (7),
.BR psx_syscall (3),
.BR capabilities (7).