[PATCH] extensions: libxt_bpf: support ebpf pinned objects

[Willem de Bruijn <willemdebruijn.kernel@gmail.com>]

From: Willem de Bruijn <willemb@google.com>

Exercise the new kernel feature introduced in commit 2c16d6033264
("netfilter: xt_bpf: support ebpf") to load pinned eBPF programs.

The new interface allows instantiating a bpf match using

  -m bpf --object-pinned ${PATH}

where ${PATH} points to a node in a bpf virtual filesystem. See
also the revised man page.

Signed-off-by: Willem de Bruijn <willemb@google.com>
---
 configure.ac                     |   2 +-
 extensions/libxt_bpf.c           | 220 +++++++++++++++++++++++++++++++--------
 extensions/libxt_bpf.man         |  19 +++-
 include/linux/netfilter/xt_bpf.h |  25 ++++-
 4 files changed, 216 insertions(+), 50 deletions(-)

diff --git a/configure.ac b/configure.ac
index d165d52..33b1b88 100644
--- a/configure.ac
+++ b/configure.ac
@@ -73,7 +73,7 @@ LDFLAGS="$saved_LDFLAGS";
 
 blacklist_modules="";
 
-AC_CHECK_HEADERS([linux/dccp.h linux/ip_vs.h linux/magic.h linux/proc_fs.h])
+AC_CHECK_HEADERS([linux/dccp.h linux/ip_vs.h linux/magic.h linux/proc_fs.h linux/bpf.h])
 if test "$ac_cv_header_linux_dccp_h" != "yes"; then
 	blacklist_modules="$blacklist_modules dccp";
 fi;
diff --git a/extensions/libxt_bpf.c b/extensions/libxt_bpf.c
index dca97d7..92c445e 100644
--- a/extensions/libxt_bpf.c
+++ b/extensions/libxt_bpf.c
@@ -16,11 +16,17 @@
 #include <sys/types.h>
 #include <unistd.h>
 #include <xtables.h>
+#include "config.h"
+
+#ifdef HAVE_LINUX_BPF_H
+#include <linux/bpf.h>
+#endif
 
 #define BCODE_FILE_MAX_LEN_B	1024
 
 enum {
 	O_BCODE_STDIN = 0,
+	O_OBJ_PINNED = 1,
 };
 
 static void bpf_help(void)
@@ -28,7 +34,16 @@ static void bpf_help(void)
 	printf(
 "bpf match options:\n"
 "--bytecode <program>	: a bpf program as generated by\n"
-"  `nfbpf_compiler RAW <filter>`\n");
+"                         $(nfbpf_compile RAW '<filter>')\n");
+}
+
+static void bpf_help_v1(void)
+{
+	printf(
+"bpf match options:\n"
+"--bytecode <program>	        : a bpf program as generated by\n"
+"                                 $(nfbpf_compile RAW '<filter>')\n"
+"--object-pinned <bpf object>	: a path to a pinned BPF object in bpf fs\n");
 }
 
 static const struct xt_option_entry bpf_opts[] = {
@@ -36,23 +51,47 @@ static const struct xt_option_entry bpf_opts[] = {
 	XTOPT_TABLEEND,
 };
 
-static void bpf_parse_string(struct xt_option_call *cb, const char *bpf_program,
-			     const char separator)
+static const struct xt_option_entry bpf_opts_v1[] = {
+	{.name = "bytecode", .id = O_BCODE_STDIN, .type = XTTYPE_STRING},
+	{.name = "object-pinned" , .id = O_OBJ_PINNED, .type = XTTYPE_STRING,
+	 .flags = XTOPT_PUT, XTOPT_POINTER(struct xt_bpf_info_v1, path)},
+	XTOPT_TABLEEND,
+};
+
+static int bpf_obj_get(const char *filepath)
 {
-	struct xt_bpf_info *bi = (void *) cb->data;
+#if defined HAVE_LINUX_BPF_H && defined __NR_bpf
+	union bpf_attr attr;
+
+	memset(&attr, 0, sizeof(attr));
+	attr.pathname = (__u64) filepath;
+
+	return syscall(__NR_bpf, BPF_OBJ_GET, &attr, sizeof(attr));
+#else
+	xtables_error(OTHER_PROBLEM,
+		      "No bpf header, kernel headers too old?\n");
+	return -EINVAL;
+#endif
+}
+
+static void bpf_parse_string(struct sock_filter *pc, __u16 *lenp, __u16 len_max,
+			     const char *bpf_program)
+{
+	const char separator = ',';
 	const char *token;
 	char sp;
 	int i;
+	__u16 len;
 
 	/* parse head: length. */
-	if (sscanf(bpf_program, "%hu%c", &bi->bpf_program_num_elem, &sp) != 2 ||
+	if (sscanf(bpf_program, "%hu%c", &len, &sp) != 2 ||
 		   sp != separator)
 		xtables_error(PARAMETER_PROBLEM,
 			      "bpf: error parsing program length");
-	if (!bi->bpf_program_num_elem)
+	if (!len)
 		xtables_error(PARAMETER_PROBLEM,
 			      "bpf: illegal zero length program");
-	if (bi->bpf_program_num_elem > XT_BPF_MAX_NUM_INSTR)
+	if (len > len_max)
 		xtables_error(PARAMETER_PROBLEM,
 			      "bpf: number of instructions exceeds maximum");
 
@@ -60,62 +99,108 @@ static void bpf_parse_string(struct xt_option_call *cb, const char *bpf_program,
 	i = 0;
 	token = bpf_program;
 	while ((token = strchr(token, separator)) && (++token)[0]) {
-		if (i >= bi->bpf_program_num_elem)
+		if (i >= len)
 			xtables_error(PARAMETER_PROBLEM,
 				      "bpf: real program length exceeds"
 				      " the encoded length parameter");
 		if (sscanf(token, "%hu %hhu %hhu %u,",
-			   &bi->bpf_program[i].code,
-			   &bi->bpf_program[i].jt,
-			   &bi->bpf_program[i].jf,
-			   &bi->bpf_program[i].k) != 4)
+			   &pc->code, &pc->jt, &pc->jf, &pc->k) != 4)
 			xtables_error(PARAMETER_PROBLEM,
 				      "bpf: error at instr %d", i);
 		i++;
+		pc++;
 	}
 
-	if (i != bi->bpf_program_num_elem)
+	if (i != len)
 		xtables_error(PARAMETER_PROBLEM,
 			      "bpf: parsed program length is less than the"
 			      " encoded length parameter");
+
+	*lenp = len;
+}
+
+static void bpf_parse_obj_pinned(struct xt_bpf_info_v1 *bi,
+				 const char *filepath)
+{
+	bi->fd = bpf_obj_get(filepath);
+	if (bi->fd < 0)
+		xtables_error(PARAMETER_PROBLEM,
+			      "bpf: failed to get bpf object");
+
+	/* Cannot close bi->fd explicitly. Rely on exit */
+	if (fcntl(bi->fd, F_SETFD, FD_CLOEXEC) == -1) {
+		xtables_error(OTHER_PROBLEM,
+			      "Could not set close on exec: %s\n",
+			      strerror(errno));
+	}
 }
 
 static void bpf_parse(struct xt_option_call *cb)
 {
+	struct xt_bpf_info *bi = (void *) cb->data;
+
 	xtables_option_parse(cb);
 	switch (cb->entry->id) {
 	case O_BCODE_STDIN:
-		bpf_parse_string(cb, cb->arg, ',');
+		bpf_parse_string(bi->bpf_program, &bi->bpf_program_num_elem,
+				 ARRAY_SIZE(bi->bpf_program), cb->arg);
 		break;
 	default:
 		xtables_error(PARAMETER_PROBLEM, "bpf: unknown option");
 	}
 }
 
-static void bpf_print_code(const void *ip, const struct xt_entry_match *match)
+static void bpf_parse_v1(struct xt_option_call *cb)
 {
-	const struct xt_bpf_info *info = (void *) match->data;
-	int i;
+	struct xt_bpf_info_v1 *bi = (void *) cb->data;
+
+	xtables_option_parse(cb);
+	switch (cb->entry->id) {
+	case O_BCODE_STDIN:
+		bpf_parse_string(bi->bpf_program, &bi->bpf_program_num_elem,
+				 ARRAY_SIZE(bi->bpf_program), cb->arg);
+		bi->mode = XT_BPF_MODE_BYTECODE;
+		break;
+	case O_OBJ_PINNED:
+		bpf_parse_obj_pinned(bi, cb->arg);
+		bi->mode = XT_BPF_MODE_FD_PINNED;
+		break;
+	default:
+		xtables_error(PARAMETER_PROBLEM, "bpf: unknown option");
+	}
+}
 
-	for (i = 0; i < info->bpf_program_num_elem-1; i++)
-		printf("%hu %hhu %hhu %u,", info->bpf_program[i].code,
-					    info->bpf_program[i].jt,
-					    info->bpf_program[i].jf,
-					    info->bpf_program[i].k);
+static void bpf_print_code(const struct sock_filter *pc, __u16 len, char tail)
+{
+	for (; len; len--, pc++)
+		printf("%hu %hhu %hhu %u%c",
+		       pc->code, pc->jt, pc->jf, pc->k,
+		       len > 1 ? ',' : tail);
+}
 
-	printf("%hu %hhu %hhu %u", info->bpf_program[i].code,
-				    info->bpf_program[i].jt,
-				    info->bpf_program[i].jf,
-				    info->bpf_program[i].k);
+static void bpf_save_code(const struct sock_filter *pc, __u16 len)
+{
+	printf(" --bytecode \"%hu,", len);
+	bpf_print_code(pc, len, '\"');
 }
 
 static void bpf_save(const void *ip, const struct xt_entry_match *match)
 {
 	const struct xt_bpf_info *info = (void *) match->data;
 
-	printf(" --bytecode \"%hu,", info->bpf_program_num_elem);
-	bpf_print_code(ip, match);
-	printf("\"");
+	bpf_save_code(info->bpf_program, info->bpf_program_num_elem);
+}
+
+static void bpf_save_v1(const void *ip, const struct xt_entry_match *match)
+{
+	const struct xt_bpf_info_v1 *info = (void *) match->data;
+
+	if (info->mode == XT_BPF_MODE_BYTECODE)
+		bpf_save_code(info->bpf_program, info->bpf_program_num_elem);
+	else if (info->mode == XT_BPF_MODE_FD_PINNED)
+		printf(" --object-pinned %s", info->path);
+	else
+		xtables_error(OTHER_PROBLEM, "unknown bpf mode");
 }
 
 static void bpf_fcheck(struct xt_fcheck_call *cb)
@@ -125,28 +210,73 @@ static void bpf_fcheck(struct xt_fcheck_call *cb)
 			      "bpf: missing --bytecode parameter");
 }
 
+static void bpf_fcheck_v1(struct xt_fcheck_call *cb)
+{
+	const unsigned int bit_bcode = 1 << O_BCODE_STDIN;
+	const unsigned int bit_pinned = 1 << O_OBJ_PINNED;
+	unsigned int flags;
+
+	flags = cb->xflags & (bit_bcode | bit_pinned);
+	if (flags != bit_bcode && flags != bit_pinned)
+		xtables_error(PARAMETER_PROBLEM,
+			      "bpf: one of --bytecode or --pinned is required");
+}
+
 static void bpf_print(const void *ip, const struct xt_entry_match *match,
 		      int numeric)
 {
+	const struct xt_bpf_info *info = (void *) match->data;
+
+	printf("match bpf ");
+	bpf_print_code(info->bpf_program, info->bpf_program_num_elem, '\0');
+}
+
+static void bpf_print_v1(const void *ip, const struct xt_entry_match *match,
+			 int numeric)
+{
+	const struct xt_bpf_info_v1 *info = (void *) match->data;
+
 	printf("match bpf ");
-	return bpf_print_code(ip, match);
-}
-
-static struct xtables_match bpf_match = {
-	.family		= NFPROTO_UNSPEC,
-	.name		= "bpf",
-	.version	= XTABLES_VERSION,
-	.size		= XT_ALIGN(sizeof(struct xt_bpf_info)),
-	.userspacesize	= XT_ALIGN(offsetof(struct xt_bpf_info, filter)),
-	.help		= bpf_help,
-	.print		= bpf_print,
-	.save		= bpf_save,
-	.x6_parse	= bpf_parse,
-	.x6_fcheck	= bpf_fcheck,
-	.x6_options	= bpf_opts,
+	if (info->mode == XT_BPF_MODE_BYTECODE)
+		bpf_print_code(info->bpf_program, info->bpf_program_num_elem, '\0');
+	else if (info->mode == XT_BPF_MODE_FD_PINNED)
+		printf("pinned %s", info->path);
+	else
+		printf("unknown");
+}
+
+static struct xtables_match bpf_matches[] = {
+	{
+		.family		= NFPROTO_UNSPEC,
+		.name		= "bpf",
+		.version	= XTABLES_VERSION,
+		.revision	= 0,
+		.size		= XT_ALIGN(sizeof(struct xt_bpf_info)),
+		.userspacesize	= XT_ALIGN(offsetof(struct xt_bpf_info, filter)),
+		.help		= bpf_help,
+		.print		= bpf_print,
+		.save		= bpf_save,
+		.x6_parse	= bpf_parse,
+		.x6_fcheck	= bpf_fcheck,
+		.x6_options	= bpf_opts,
+	},
+	{
+		.family		= NFPROTO_UNSPEC,
+		.name		= "bpf",
+		.version	= XTABLES_VERSION,
+		.revision	= 1,
+		.size		= XT_ALIGN(sizeof(struct xt_bpf_info_v1)),
+		.userspacesize	= XT_ALIGN(offsetof(struct xt_bpf_info_v1, filter)),
+		.help		= bpf_help_v1,
+		.print		= bpf_print_v1,
+		.save		= bpf_save_v1,
+		.x6_parse	= bpf_parse_v1,
+		.x6_fcheck	= bpf_fcheck_v1,
+		.x6_options	= bpf_opts_v1,
+	},
 };
 
 void _init(void)
 {
-	xtables_register_match(&bpf_match);
+	xtables_register_matches(bpf_matches, ARRAY_SIZE(bpf_matches));
 }
diff --git a/extensions/libxt_bpf.man b/extensions/libxt_bpf.man
index 5b1d042..8a0eb0c 100644
--- a/extensions/libxt_bpf.man
+++ b/extensions/libxt_bpf.man
@@ -1,8 +1,21 @@
-Match using Linux Socket Filter. Expects a BPF program in decimal format. This
-is the format generated by the \fBnfbpf_compile\fP utility.
+Match using Linux Socket Filter. Expects a path to an eBPF object or a cBPF
+program in decimal format.
+.TP
+\fB\-\-object\-pinned\fP \fIpath\fP
+Pass a path to a pinned eBPF object.
+.PP
+Applications load eBPF programs into the kernel with the bpf() system call and
+BPF_PROG_LOAD command and can pin them in a virtual filesystem with BPF_OBJ_PIN.
+To use a pinned object in iptables, mount the bpf filesystem using
+.IP
+mount \-t bpf bpf ${BPF_MOUNT}
+.PP
+then insert the filter in iptables by path:
+.IP
+iptables \-A OUTPUT \-m bpf \-\-object\-pinned ${BPF_MOUNT}/{PINNED_PATH} \-j ACCEPT
 .TP
 \fB\-\-bytecode\fP \fIcode\fP
-Pass the BPF byte code format (described in the example below).
+Pass the BPF byte code format as generated by the \fBnfbpf_compile\fP utility.
 .PP
 The code format is similar to the output of the tcpdump -ddd command: one line
 that stores the number of instructions, followed by one line for each
diff --git a/include/linux/netfilter/xt_bpf.h b/include/linux/netfilter/xt_bpf.h
index 5dda450..b97725a 100644
--- a/include/linux/netfilter/xt_bpf.h
+++ b/include/linux/netfilter/xt_bpf.h
@@ -2,16 +2,39 @@
 #define _XT_BPF_H
 
 #include <linux/filter.h>
+#include <linux/limits.h>
 #include <linux/types.h>
 
 #define XT_BPF_MAX_NUM_INSTR	64
+#define XT_BPF_PATH_MAX		(XT_BPF_MAX_NUM_INSTR * sizeof(struct sock_filter))
+
+struct bpf_prog;
 
 struct xt_bpf_info {
 	__u16 bpf_program_num_elem;
 	struct sock_filter bpf_program[XT_BPF_MAX_NUM_INSTR];
 
 	/* only used in the kernel */
-	struct sk_filter *filter __attribute__((aligned(8)));
+	struct bpf_prog *filter __attribute__((aligned(8)));
+};
+
+enum xt_bpf_modes {
+	XT_BPF_MODE_BYTECODE,
+	XT_BPF_MODE_FD_PINNED,
+	XT_BPF_MODE_FD_ELF,
+};
+
+struct xt_bpf_info_v1 {
+	__u16 mode;
+	__u16 bpf_program_num_elem;
+	__s32 fd;
+	union {
+		struct sock_filter bpf_program[XT_BPF_MAX_NUM_INSTR];
+		char path[XT_BPF_PATH_MAX];
+	};
+
+	/* only used in the kernel */
+	struct bpf_prog *filter __attribute__((aligned(8)));
 };
 
 #endif /*_XT_BPF_H */
-- 
2.8.0.rc3.226.g39d4020