Quoting Andrew Morgan (morgan@kernel.org):
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Serge E. Hallyn wrote:
> > +		case PR_GET_CAPBSET:
> > +			error = put_user(current->cap_bset, (unsigned long __user *)arg2);
> > +			break;
> > +		case PR_SET_CAPBSET:
> > +			if (!capable(CAP_SYS_ADMIN))
> > +				return -EPERM;
> > +			if (!cap_issubset(arg2, current->cap_bset))
> > +				return -EINVAL;
> > +			current->cap_bset = arg2;
> > +			break;
> 
> You need to pass the capability magic value in both get and set
> directions... Otherwise, you'll not be able to tell what vintage of
> cap_bset you are manipulating.

Great point, thanks.  How about this version?

thanks,
-serge


>From 1236d211e2e1d41a8a60b9577e3d4b509a17de92 Mon Sep 17 00:00:00 2001
From: sergeh@us.ibm.com <hallyn@kernel.(none)>
Date: Mon, 24 Sep 2007 13:02:11 -0700
Subject: [PATCH 1/1] capabilities: introduce per-process capability bounding set (v2)

The capability bounding set is a set beyond which capabilities
cannot grow.  Currently cap_bset is per-system.  It can be
manipulated through sysctl, but only init can add capabilities.
Root can remove capabilities.  By default it includes all caps
except CAP_SETPCAP.

This patch makes the bounding set per-process.  It is inherited
at fork from parent.  Noone can add elements, CAP_SYS_ADMIN is
required to remove them.  Perhaps a new capability should be
introduced to control the ability to remove capabilities, in
order to help prevent running a privileged app with enough
privs to be dangerous but not enough to be successful.

One example use of this is to start a safer container.  For
instance, until device namespaces or per-container device
whitelists are introduced, it is best to take CAP_MKNOD away
from a container.

The following hacky test program will get and set the bounding
 set.  For instance

	./bset get
		(lists capabilities in bset)
	./bset strset cap_sys_admin
		(starts shell with new bset)
		(use capset or setuid binary to try to increase caps)

===========================================================
bset.c:
===========================================================

unsigned long newval;
int cmd_getbcap;

char *captable[] = {
	"cap_dac_override",
	"cap_dac_read_search",
	"cap_fowner",
	"cap_fsetid",
	"cap_kill",
	"cap_setgid",
	"cap_setuid",
	"cap_setpcap",
	"cap_linux_immutable",
	"cap_net_bind_service",
	"cap_net_broadcast",
	"cap_net_admin",
	"cap_net_raw",
	"cap_ipc_lock",
	"cap_ipc_owner",
	"cap_sys_module",
	"cap_sys_rawio",
	"cap_sys_chroot",
	"cap_sys_ptrace",
	"cap_sys_pacct",
	"cap_sys_admin",
	"cap_sys_boot",
	"cap_sys_nice",
	"cap_sys_resource",
	"cap_sys_time",
	"cap_sys_tty_config",
	"cap_mknod",
	"cap_lease",
	"cap_audit_write",
	"cap_audit_control",
	"cap_setfcap"};

char *inttocap(unsigned long v)
{
	char *str = NULL;
	int i;

	str = malloc(1);
	str[0] = '\0';
	for (i=0; i<31; i++) {
		if (v & (1 << (i+1))) {
			char *tmp = captable[i];
			str = realloc(str, strlen(str)+2+strlen(tmp));
			sprintf(str+strlen(str), ",%s", tmp);
		}
	}
	return str;
}

int getbcap(void)
{
	unsigned long bcap;
	int ret;
	unsigned long ver;

	ret = prctl(PR_GET_CAPBSET, &ver, &bcap);
	if (ret == -1)
		perror("prctl");
	if (ver != _LINUX_CAPABILITY_VERSION)
		printf("wrong capability version: %lu not %lu\n",
					ver, _LINUX_CAPABILITY_VERSION);
	printf("prctl get_bcap returned %lu (ret %d)\n", bcap, ret);
	printf("that is %s\n", inttocap(bcap));
	return ret;
}

int setbcap(unsigned long val)
{
	int ret;

	ret = prctl(PR_SET_CAPBSET, _LINUX_CAPABILITY_VERSION, val);
	return ret;
}

int usage(char *me)
{
	printf("Usage: %s get\n", me);
	printf("       %s set <newval>\n", me);
	printf("       %s strset capability_string\n", me);
	printf("         capability_string is for instance:\n");
	printf("         cap_sys_admin,cap_mknod,cap_dac_override\n");
	return 1;
}

unsigned long captoint(char *cap)
{
	if (strcmp(cap, "cap_dac_override") == 0)
		return 1;
	else if (strcmp(cap, "cap_dac_read_search") == 0)
		return 2;
	else if (strcmp(cap, "cap_fowner") == 0)
		return 3;
	else if (strcmp(cap, "cap_fsetid") == 0)
		return 4;
	else if (strcmp(cap, "cap_kill") == 0)
		return 5;
	else if (strcmp(cap, "cap_setgid") == 0)
		return 6;
	else if (strcmp(cap, "cap_setuid") == 0)
		return 7;
	else if (strcmp(cap, "cap_setpcap") == 0)
		return 8;
	else if (strcmp(cap, "cap_linux_immutable") == 0)
		return 9;
	else if (strcmp(cap, "cap_net_bind_service") == 0)
		return 10;
	else if (strcmp(cap, "cap_net_broadcast") == 0)
		return 11;
	else if (strcmp(cap, "cap_net_admin") == 0)
		return 12;
	else if (strcmp(cap, "cap_net_raw") == 0)
		return 13;
	else if (strcmp(cap, "cap_ipc_lock") == 0)
		return 14;
	else if (strcmp(cap, "cap_ipc_owner") == 0)
		return 15;
	else if (strcmp(cap, "cap_sys_module") == 0)
		return 16;
	else if (strcmp(cap, "cap_sys_rawio") == 0)
		return 17;
	else if (strcmp(cap, "cap_sys_chroot") == 0)
		return 18;
	else if (strcmp(cap, "cap_sys_ptrace") == 0)
		return 19;
	else if (strcmp(cap, "cap_sys_pacct") == 0)
		return 20;
	else if (strcmp(cap, "cap_sys_admin") == 0)
		return 21;
	else if (strcmp(cap, "cap_sys_boot") == 0)
		return 22;
	else if (strcmp(cap, "cap_sys_nice") == 0)
		return 23;
	else if (strcmp(cap, "cap_sys_resource") == 0)
		return 24;
	else if (strcmp(cap, "cap_sys_time") == 0)
		return 25;
	else if (strcmp(cap, "cap_sys_tty_config") == 0)
		return 26;
	else if (strcmp(cap, "cap_mknod") == 0)
		return 27;
	else if (strcmp(cap, "cap_lease") == 0)
		return 28;
	else if (strcmp(cap, "cap_audit_write") == 0)
		return 29;
	else if (strcmp(cap, "cap_audit_control") == 0)
		return 30;
	else if (strcmp(cap, "cap_setfcap") == 0)
		return 31;
}

unsigned long parse_cap_string(char *capstring)
{
	unsigned long tmp, newval = 0;
	char *token = strtok(capstring, ",");

	while (token) {
		tmp = captoint(token);
		if (tmp < 0)
			return -1;
		newval |= 1<<tmp;
		token = strtok(NULL, ",");
	}
	return newval;
}

int read_args(int argc, char *argv[])
{
	if (strcmp(argv[1], "get") == 0) {
		cmd_getbcap = 1;
		return 0;
	}
	if (strcmp(argv[1], "strset") == 0) {
		newval = parse_cap_string(argv[2]);
		if (newval < 0)
			return newval;
		return 0;
	}
	if (strcmp(argv[1], "set") != 0)
		return 1;
	if (argc != 3)
		return 1;
	newval = strtoul(argv[2], NULL, 10);
	return 0;
}

int main(int argc, char *argv[])
{
	int ret;

	if (argc<2)
		return usage(argv[0]);

	if ((ret=read_args(argc, argv)))
		return ret;

	if (cmd_getbcap)
		return getbcap();

	ret = setbcap(newval);
	if (ret != 0)
		return ret;
	return execl("/bin/bash", "/bin/bash", NULL);
}
===========================================================

Changelog:
	As suggested by Andrew Morgan, send the capability
	version along with the bset for prctl(PR_SET_CAPBSET)
	and PR_GET_CAPBSET)

Signed-off-by: Serge Hallyn <serue@us.ibm.com>
---
 include/linux/capability.h |   12 +++++++++++-
 include/linux/init_task.h  |    1 +
 include/linux/prctl.h      |    4 ++++
 include/linux/sched.h      |    2 +-
 include/linux/security.h   |    5 -----
 include/linux/sysctl.h     |    3 ---
 kernel/fork.c              |    1 +
 kernel/sys.c               |   17 +++++++++++++++++
 kernel/sysctl.c            |   35 -----------------------------------
 kernel/sysctl_check.c      |    7 -------
 security/commoncap.c       |    6 ++----
 11 files changed, 37 insertions(+), 56 deletions(-)

diff --git a/include/linux/capability.h b/include/linux/capability.h
index 7a8d7ad..25b83bb 100644
--- a/include/linux/capability.h
+++ b/include/linux/capability.h
@@ -197,7 +197,6 @@ typedef __u32 kernel_cap_t;
 #define CAP_IPC_OWNER        15
 
 /* Insert and remove kernel modules - modify kernel without limit */
-/* Modify cap_bset */
 #define CAP_SYS_MODULE       16
 
 /* Allow ioperm/iopl access */
@@ -332,6 +331,17 @@ typedef __u32 kernel_cap_t;
 #define CAP_INIT_EFF_SET    to_cap_t(~0 & ~CAP_TO_MASK(CAP_SETPCAP))
 #define CAP_INIT_INH_SET    to_cap_t(0)
 
+#ifdef CONFIG_SECURITY_FILE_CAPABILITIES
+/*
+ * Because of the reduced scope of CAP_SETPCAP when filesystem
+ * capabilities are in effect, it is safe to allow this capability to
+ * be available in the default configuration.
+ */
+# define CAP_INIT_BSET  CAP_FULL_SET
+#else
+# define CAP_INIT_BSET  CAP_INIT_EFF_SET
+#endif
+
 #define CAP_TO_MASK(x) (1 << (x))
 #define cap_raise(c, flag)   (cap_t(c) |=  CAP_TO_MASK(flag))
 #define cap_lower(c, flag)   (cap_t(c) &= ~CAP_TO_MASK(flag))
diff --git a/include/linux/init_task.h b/include/linux/init_task.h
index a3f2541..c2f4285 100644
--- a/include/linux/init_task.h
+++ b/include/linux/init_task.h
@@ -148,6 +148,7 @@ extern struct group_info init_groups;
 	.cap_effective	= CAP_INIT_EFF_SET,				\
 	.cap_inheritable = CAP_INIT_INH_SET,				\
 	.cap_permitted	= CAP_FULL_SET,					\
+	.cap_bset 	= CAP_INIT_BSET,				\
 	.keep_capabilities = 0,						\
 	.user		= INIT_USER,					\
 	.comm		= "swapper",					\
diff --git a/include/linux/prctl.h b/include/linux/prctl.h
index e2eff90..a7de023 100644
--- a/include/linux/prctl.h