1 files changed, 516 insertions, 0 deletions

diff --git a/usr/src/lib/brand/lx/lx_brand/common/capabilities.c b/usr/src/lib/brand/lx/lx_brand/common/capabilities.c
new file mode 100644
index 0000000000..e034560452
--- /dev/null
+++ b/usr/src/lib/brand/lx/lx_brand/common/capabilities.c
@@ -0,0 +1,516 @@
+/*
+ * This file and its contents are supplied under the terms of the
+ * Common Development and Distribution License ("CDDL"), version 1.0.
+ * You may only use this file in accordance with the terms of version
+ * 1.0 of the CDDL.
+ *
+ * A full copy of the text of the CDDL should have accompanied this
+ * source.  A copy of the CDDL is also available via the Internet at
+ * http://www.illumos.org/license/CDDL.
+ */
+
+/*
+ * Copyright 2019 Joyent, Inc.
+ */
+
+/*
+ * LX Brand emulation of capget/capset syscalls
+ */
+
+#include <sys/types.h>
+#include <sys/errno.h>
+#include <sys/debug.h>
+#include <sys/lx_types.h>
+#include <sys/lx_syscall.h>
+#include <sys/syscall.h>
+#include <alloca.h>
+#include <errno.h>
+#include <unistd.h>
+#include <string.h>
+#include <sys/lx_misc.h>
+#include <priv.h>
+
+typedef struct {
+	uint32_t version;
+	int pid;
+} lx_cap_user_header_t;
+
+typedef struct {
+	uint32_t effective;
+	uint32_t permitted;
+	uint32_t inheritable;
+} lx_cap_user_data_t;
+
+typedef struct {
+	priv_set_t *p_effective;
+	priv_set_t *p_permitted;
+	priv_set_t *p_inheritable;
+} lx_cap_privs_t;
+
+#define	LX_CAP_UPDATE_PERMITTED		0x1
+#define	LX_CAP_UPDATE_EFFECTIVE		0x2
+#define	LX_CAP_UPDATE_INHERITABLE	0x4
+
+
+#define	LX_CAP_MAXLEN	2
+
+typedef struct {
+	uint32_t effective[LX_CAP_MAXLEN];
+	uint32_t permitted[LX_CAP_MAXLEN];
+	uint32_t inheritable[LX_CAP_MAXLEN];
+} lx_cap_data_t;
+
+#define	LX_CAP_VERSION_1	0x19980330
+#define	LX_CAP_VERSION_2	0x20071026	/* deprecated by Linux */
+#define	LX_CAP_VERSION_3	0x20080522
+
+#define	LX_CAP_SETPCAP	8
+
+/*
+ * Even though we lack mappings for capabilities higher than LX_CAP_MAX_VALID,
+ * it's valuable to test all the way out to the end of the second field. This
+ * ensures that new capabilities we lack support for are not silently accepted.
+ */
+#define	LX_CAP_MAX_CHECK		63
+
+#define	LX_CAP_CAPISSET(id, cap) \
+	(((id < 32) && (((0x1 << id) & cap[0]) != 0)) || \
+	((id >= 32) && (((0x1 << (id - 32) & cap[1]) != 0))))
+
+#define	LX_CAP_CAPSET(id, cap) \
+	if (id < 32) { cap[0] |= (0x1 << id); } \
+	else { cap[1] |= (0x1 << (id - 32)); }
+
+static const char *lx_cap_map_chown[] = {
+	PRIV_FILE_CHOWN,
+	PRIV_FILE_CHOWN_SELF,
+	NULL
+};
+static const char *lx_cap_map_dac_override[] = {
+	PRIV_FILE_DAC_READ,
+	PRIV_FILE_DAC_WRITE,
+	PRIV_FILE_DAC_EXECUTE,
+	NULL
+};
+static const char *lx_cap_map_dac_read_search[] = {
+	PRIV_FILE_DAC_SEARCH,
+	PRIV_FILE_DAC_READ,
+	NULL
+};
+static const char *lx_cap_map_fowner[] = { PRIV_FILE_OWNER, NULL };
+static const char *lx_cap_map_fsetid[] = { PRIV_FILE_SETID, NULL };
+static const char *lx_cap_map_kill[] = { PRIV_PROC_OWNER, NULL };
+/*
+ * One way that Linux capabilities(7) differs from Illumos privileges(5) is
+ * that it distinguishes between setuid and setgroups rights.  This will be a
+ * problem if an lx-branded process requests to drop only CAP_SETUID but not
+ * CAP_SETGID.
+ *
+ * In that case, CAP_SETUID will be maintained.
+ */
+static const char *lx_cap_map_setgid[] = { PRIV_PROC_SETID, NULL };
+static const char *lx_cap_map_setuid[] = { PRIV_PROC_SETID, NULL };
+static const char *lx_cap_map_linux_immutable[] = { PRIV_FILE_FLAG_SET, NULL };
+static const char *lx_cap_map_bind_service[] = { PRIV_NET_PRIVADDR, NULL };
+static const char *lx_cap_map_net_admin[] = {
+	PRIV_SYS_IP_CONFIG,
+	NULL
+	/*
+	 * It would probably make sense to include PRIV_SYS_DL_CONFIG, but that
+	 * privilege is not extended to non-global zones by default.  A more
+	 * sophisticated capabilities translation layer could make it optional.
+	 */
+};
+static const char *lx_cap_map_net_raw[] = {
+	PRIV_NET_RAWACCESS,
+	PRIV_NET_ICMPACCESS,
+	NULL
+};
+static const char *lx_cap_map_ipc_lock[] = { PRIV_PROC_LOCK_MEMORY, NULL };
+static const char *lx_cap_map_ipc_owner[] = {
+	PRIV_IPC_DAC_READ,
+	PRIV_IPC_DAC_WRITE,
+	PRIV_IPC_OWNER,
+	NULL
+};
+static const char *lx_cap_map_sys_chroot[] = { PRIV_PROC_CHROOT, NULL };
+static const char *lx_cap_map_sys_admin[] = {
+	PRIV_SYS_MOUNT,
+	PRIV_SYS_ADMIN,
+	NULL
+};
+static const char *lx_cap_map_sys_nice[] = { PRIV_PROC_PRIOUP, NULL };
+static const char *lx_cap_map_sys_resource[] = { PRIV_SYS_RESOURCE, NULL };
+static const char *lx_cap_map_audit_write[] = { PRIV_PROC_AUDIT, NULL };
+static const char *lx_cap_map_audit_control[] = { PRIV_SYS_AUDIT, NULL };
+
+/*
+ * Mapping of Linux capabilities -> Illumos privileges
+ * The ID definitions can be found in the Linux sources here:
+ * include/uapi/linux/capability.h
+ *
+ * Order is critical.
+ */
+static const char ** lx_cap_mapping[LX_CAP_MAX_VALID + 1] = {
+	lx_cap_map_chown,		/* CAP_CHOWN */
+	lx_cap_map_dac_override,	/* CAP_DAC_OVERRIDE */
+	lx_cap_map_dac_read_search,	/* CAP_DAC_READ_SEARCH */
+	lx_cap_map_fowner,		/* CAP_FOWNER */
+	lx_cap_map_fsetid,		/* CAP_FSETID */
+	lx_cap_map_kill,		/* CAP_KILL */
+	lx_cap_map_setgid,		/* CAP_SETGID */
+	lx_cap_map_setuid,		/* CAP_SETUID */
+	NULL,				/* CAP_SETPCAP */
+	lx_cap_map_linux_immutable,	/* CAP_LINUX_IMMUTABLE */
+	lx_cap_map_bind_service,	/* CAP_BIND_SERVICE */
+	NULL,				/* CAP_BROADCAST */
+	lx_cap_map_net_admin,		/* CAP_NET_ADMIN */
+	lx_cap_map_net_raw,		/* CAP_NET_RAW */
+	lx_cap_map_ipc_lock,		/* CAP_IPC_LOCK */
+	lx_cap_map_ipc_owner,		/* CAP_IPC_OWNER */
+	NULL,				/* CAP_MODULE */
+	NULL,				/* CAP_RAWIO */
+	lx_cap_map_sys_chroot,		/* CAP_SYS_CHROOT */
+	NULL,				/* CAP_PTRACE */
+	NULL,				/* CAP_PACCT */
+	lx_cap_map_sys_admin,		/* CAP_SYS_ADMIN */
+	NULL,				/* CAP_BOOT */
+	lx_cap_map_sys_nice,		/* CAP_SYS_NICE */
+	lx_cap_map_sys_resource,	/* CAP_SYS_RESOURCE */
+	NULL,				/* CAP_SYS_TIME */
+	NULL,				/* CAP_SYS_TTY_CONFIG */
+	NULL,				/* CAP_MKNOD */
+	NULL,				/* CAP_LEASE */
+	lx_cap_map_audit_write,		/* CAP_AUDIT_WRITE */
+	lx_cap_map_audit_control,	/* CAP_AUDIT_CONTROL */
+	NULL,				/* CAP_SETFCAP */
+	NULL,				/* CAP_MAC_OVERRIDE */
+	NULL,				/* CAP_MAC_ADMIN */
+	NULL,				/* CAP_SYSLOG */
+	NULL,				/* CAP_WAKE_ALARM */
+	NULL				/* CAP_BLOCK_SUSPEND */
+};
+
+/* track priv_set_t size, set on entry to lx_capset/lx_capget */
+static unsigned int lx_cap_priv_size = 0;
+
+/* safely allocate priv_set_t triplet on the stack */
+#define	LX_CAP_ALLOC_PRIVS(ptr)	\
+	{ \
+		ptr = SAFE_ALLOCA(sizeof (lx_cap_privs_t) + \
+		    (3 * lx_cap_priv_size)); \
+		if (ptr != NULL) { \
+			ptr->p_effective = (priv_set_t *) \
+			    ((caddr_t)ptr + sizeof (lx_cap_privs_t)); \
+			ptr->p_permitted = (priv_set_t *) \
+			    ((caddr_t)ptr + sizeof (lx_cap_privs_t) + \
+			    lx_cap_priv_size);\
+			ptr->p_inheritable = (priv_set_t *) \
+			    ((caddr_t)ptr + sizeof (lx_cap_privs_t) + \
+			    2 * lx_cap_priv_size); \
+		} \
+	}
+
+static long
+lx_cap_update_priv(priv_set_t *priv, const uint32_t cap[])
+{
+	int i, j;
+	boolean_t cap_set;
+	boolean_t priv_set;
+	boolean_t updated = B_FALSE;
+	for (i = 0; i <= LX_CAP_MAX_CHECK; i++) {
+		cap_set = LX_CAP_CAPISSET(i, cap);
+		if (i > LX_CAP_MAX_VALID || lx_cap_mapping[i] == NULL) {
+			/* don't allow setting unsupported caps */
+			if (cap_set) {
+				/*
+				 * CAP_SETPCAP is a special capability, with
+				 * varying behavior, that can be used to
+				 * control if the process can change other
+				 * process's capabilities, or to control moving
+				 * capabilities between sets. For now we ignore
+				 * this if its passed in.
+				 */
+				if (i == LX_CAP_SETPCAP) {
+					continue;
+				}
+				lx_unsupported("set unsupported capability %d",
+				    i);
+				return (-1);
+			} else {
+				continue;
+			}
+		}
+		for (j = 0; lx_cap_mapping[i][j] != NULL; j++) {
+			priv_set = priv_ismember(priv, lx_cap_mapping[i][j]);
+			if (priv_set && !cap_set) {
+				VERIFY0(priv_delset(priv,
+				    lx_cap_mapping[i][j]));
+				updated = B_TRUE;
+			} else if (!priv_set && cap_set) {
+				VERIFY0(priv_addset(priv,
+				    lx_cap_mapping[i][j]));
+				updated = B_TRUE;
+			}
+		}
+	}
+	if (updated)
+		return (1);
+	else
+		return (0);
+}
+
+static long
+lx_cap_to_priv(lx_cap_data_t *cap, lx_cap_privs_t *priv)
+{
+	long changes = 0;
+	long result;
+
+	result = lx_cap_update_priv(priv->p_permitted, cap->permitted);
+	if (result < 0)
+		return (-1);
+	else if (result > 0)
+		changes |= LX_CAP_UPDATE_PERMITTED;
+
+	result = lx_cap_update_priv(priv->p_effective, cap->effective);
+	if (result < 0)
+		return (-1);
+	else if (result > 0)
+		changes |= LX_CAP_UPDATE_EFFECTIVE;
+
+	result = lx_cap_update_priv(priv->p_inheritable, cap->inheritable);
+	if (result < 0)
+		return (-1);
+	else if (result > 0)
+		changes |= LX_CAP_UPDATE_INHERITABLE;
+
+	return (changes);
+}
+
+static void
+lx_cap_from_priv(const priv_set_t *priv, uint32_t cap[])
+{
+	int i, j;
+	boolean_t valid;
+	(void) memset(cap, '\0', sizeof (uint32_t) * LX_CAP_MAXLEN);
+	for (i = 0; i <= LX_CAP_MAX_VALID; i++) {
+		if (lx_cap_mapping[i] == NULL) {
+			continue;
+		}
+		valid = B_TRUE;
+		for (j = 0; lx_cap_mapping[i][j] != NULL; j++) {
+			if (!priv_ismember(priv,
+			    lx_cap_mapping[i][j])) {
+				valid = B_FALSE;
+			}
+		}
+		if (valid) {
+			LX_CAP_CAPSET(i, cap);
+		}
+	}
+}
+
+static long
+lx_cap_read_cap(const lx_cap_user_header_t *uhp, const lx_cap_user_data_t *udp,
+    lx_cap_data_t *cd)
+{
+	lx_cap_user_header_t uh;
+	lx_cap_user_data_t ud_buf;
+	int cap_count;
+	int i;
+
+	if (uucopy(uhp, &uh, sizeof (uh)) != 0)
+		return (-errno);
+
+	switch (uh.version) {
+	case LX_CAP_VERSION_1:
+		cap_count = 1;
+		break;
+	case LX_CAP_VERSION_2:
+	case LX_CAP_VERSION_3:
+		cap_count = 2;
+		break;
+	default:
+		return (-EINVAL);
+	}
+
+	/* Only allow capset on calling process */
+	if (uh.pid != 0 && uh.pid != getpid())
+		return (-EPERM);
+
+	/* zero the struct in case cap_count < 2 */
+	(void) memset(cd, '\0', sizeof (lx_cap_data_t));
+
+	for (i = 0; i < cap_count; i++) {
+		if (uucopy(udp + i, &ud_buf, sizeof (ud_buf)) != 0)
+			return (-errno);
+		cd->permitted[i] = ud_buf.permitted;
+		cd->effective[i] = ud_buf.effective;
+		cd->inheritable[i] = ud_buf.inheritable;
+	}
+	return (0);
+}
+
+long
+lx_capget(uintptr_t p1, uintptr_t p2)
+{
+	const priv_impl_info_t *impl;
+	lx_cap_user_header_t *uhp = (lx_cap_user_header_t *)p1;
+	lx_cap_user_data_t *udp = (lx_cap_user_data_t *)p2;
+	lx_cap_user_header_t uh;
+	lx_cap_privs_t *privs;
+	lx_cap_data_t cd_result;
+	lx_cap_user_data_t cd_buf;
+	int cap_count;
+	int i;
+
+	if (lx_cap_priv_size == 0) {
+		impl = getprivimplinfo();
+		lx_cap_priv_size = sizeof (priv_chunk_t) * impl->priv_setsize;
+	}
+
+	if (uucopy(uhp, &uh, sizeof (uh)) != 0)
+		return (-errno);
+
+	switch (uh.version) {
+	case LX_CAP_VERSION_1:
+		cap_count = 1;
+		break;
+	case LX_CAP_VERSION_2:
+	case LX_CAP_VERSION_3:
+		cap_count = 2;
+		break;
+	default:
+		/*
+		 * As per the man page: call will fail with EINVAL and set the
+		 * version field of the header to the kernel preferred version
+		 * when an unsupported version value is provided.
+		 */
+		uh.version = LX_CAP_VERSION_3;
+		if (uucopy(&uh, uhp, sizeof (uh)) != 0)
+			return (-errno);
+		return (-EINVAL);
+	}
+
+	/*
+	 * Only allow capget on the calling process.
+	 * If a pid is specified, lie about being able to locate it.
+	 */
+	if (uh.pid > 0 && uh.pid != getpid())
+		return (-ESRCH);
+	if (uh.pid < 0)
+		return (-EINVAL);
+
+	LX_CAP_ALLOC_PRIVS(privs)
+
+	if (privs == NULL)
+		return (-ENOMEM);
+
+	if (getppriv(PRIV_PERMITTED, privs->p_permitted) != 0)
+		return (-errno);
+	if (getppriv(PRIV_EFFECTIVE, privs->p_effective) != 0)
+		return (-errno);
+	if (getppriv(PRIV_INHERITABLE, privs->p_inheritable) != 0)
+		return (-errno);
+
+	lx_cap_from_priv(privs->p_permitted, cd_result.permitted);
+	lx_cap_from_priv(privs->p_effective, cd_result.effective);
+	lx_cap_from_priv(privs->p_inheritable, cd_result.inheritable);
+
+	/* convert to output format */
+	for (i = 0; i < cap_count; i++) {
+		cd_buf.effective = cd_result.effective[i];
+		cd_buf.permitted = cd_result.permitted[i];
+		cd_buf.inheritable = cd_result.inheritable[i];
+		if (uucopy(&cd_buf, udp + i, sizeof (cd_buf)) != 0)
+			return (-errno);
+	}
+
+	return (0);
+}
+
+long
+lx_capset(uintptr_t p1, uintptr_t p2)
+{
+	const priv_impl_info_t *impl;
+	lx_cap_data_t cd;
+	lx_cap_privs_t *privs;
+	long result;
+
+	if (lx_cap_priv_size == 0) {
+		impl = getprivimplinfo();
+		lx_cap_priv_size = sizeof (priv_chunk_t) * impl->priv_setsize;
+	}
+
+	/* verify header and read in desired capabilities */
+	result = lx_cap_read_cap((lx_cap_user_header_t *)p1,
+	    (lx_cap_user_data_t *)p2, &cd);
+	if (result != 0)
+		return (result);
+
+	LX_CAP_ALLOC_PRIVS(privs)
+
+	if (privs == NULL)
+		return (-ENOMEM);
+
+	/* fetch current privs to compare against */
+	if (getppriv(PRIV_PERMITTED, privs->p_permitted) != 0)
+		return (-errno);
+	if (getppriv(PRIV_EFFECTIVE, privs->p_effective) != 0)
+		return (-errno);
+	if (getppriv(PRIV_INHERITABLE, privs->p_inheritable) != 0)
+		return (-errno);
+
+
+	result = lx_cap_to_priv(&cd, privs);
+	if (result < 0)
+		return (-EPERM);
+
+	/* report success if no changes needed */
+	if (result == 0)
+		return (0);
+
+	/* Ensure the effective/inheritable caps aren't > permitted */
+	if (!priv_issubset(privs->p_effective, privs->p_permitted) ||
+	    !priv_issubset(privs->p_inheritable, privs->p_permitted))
+		return (-EPERM);
+
+	/*
+	 * Here is where things become racy.  Linux updates all three
+	 * capability sets simultaneously in the capset syscall.  In order to
+	 * emulate capabilities via privileges, three setppriv operations are
+	 * required in sequence.  If one or two should fail, there is not a
+	 * mechanism to convey the incomplete operation to the caller.
+	 *
+	 * We do two things to make this less risky:
+	 * 1. Verify that both the desired effective and inheritable
+	 *    sets are subsets of the desired permitted set.
+	 * 2. Perform the setppriv of the permitted set first.
+	 *
+	 * Should the setppriv(permitted) fail, we can safely bail out with an
+	 * error.  If it succeeds, the setppriv of effective and inheritable
+	 * are likely to succeed given that they've been verified legal.
+	 *
+	 * If the partial error does happen, we'll be forced to report failure
+	 * even though the privileges were altered.
+	 */
+
+	if ((result & LX_CAP_UPDATE_PERMITTED) != 0) {
+		/* failure here is totally safe */
+		if (setppriv(PRIV_SET, PRIV_PERMITTED, privs->p_permitted) != 0)
+			return (-errno);
+	}
+	if ((result & LX_CAP_UPDATE_EFFECTIVE) != 0) {
+		/* failure here is a bummer */
+		if (setppriv(PRIV_SET, PRIV_EFFECTIVE, privs->p_effective) != 0)
+			return (-errno);
+	}
+	if ((result & LX_CAP_UPDATE_EFFECTIVE) != 0) {
+		/* failure here is a major bummer */
+		if (setppriv(PRIV_SET, PRIV_INHERITABLE,
+		    privs->p_inheritable) != 0)
+			return (-errno);
+	}
+
+	return (0);
+}