strace/bpf.c

/*
 * Copyright (c) 2015-2017 Dmitry V. Levin <ldv@altlinux.org>
 * Copyright (c) 2017 Quentin Monnet <quentin.monnet@6wind.com>
 * Copyright (c) 2015-2018 The strace developers.
 * All rights reserved.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 * 3. The name of the author may not be used to endorse or promote products
 *    derived from this software without specific prior written permission.
 *
 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
 * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
 * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
 * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
 * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
 * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
 * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
 * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
 */

#include "defs.h"
#include "print_fields.h"

#ifdef HAVE_LINUX_BPF_H
# include <linux/bpf.h>
#endif

#include "bpf_attr.h"

#include "xlat/bpf_commands.h"
#include "xlat/bpf_file_mode_flags.h"
#include "xlat/bpf_map_types.h"
#include "xlat/bpf_map_flags.h"
#include "xlat/bpf_prog_types.h"
#include "xlat/bpf_prog_flags.h"
#include "xlat/bpf_map_update_elem_flags.h"
#include "xlat/bpf_attach_type.h"
#include "xlat/bpf_attach_flags.h"
#include "xlat/bpf_query_flags.h"
#include "xlat/numa_node.h"

/** Storage for all the data that is needed to be stored on entering. */
struct bpf_priv_data {
	bool     bpf_prog_query_stored;
	uint32_t bpf_prog_query_prog_cnt;
};

#define DECL_BPF_CMD_DECODER(bpf_cmd_decoder)				\
int									\
bpf_cmd_decoder(struct tcb *const tcp,					\
		const kernel_ulong_t addr,				\
		const unsigned int size,				\
		void *const data,					\
		struct bpf_priv_data *priv)				\
/* End of DECL_BPF_CMD_DECODER definition. */

#define BEGIN_BPF_CMD_DECODER(bpf_cmd)					\
	static DECL_BPF_CMD_DECODER(decode_ ## bpf_cmd)			\
	{								\
		struct bpf_cmd ## _struct attr = {};			\
		const size_t attr_size = bpf_cmd ## _struct_size;	\
		const unsigned int len = MIN(size, attr_size);		\
		memcpy(&attr, data, len);				\
		do {							\
/* End of BEGIN_BPF_CMD_DECODER definition. */

#define END_BPF_CMD_DECODER(rval)					\
			decode_attr_extra_data(tcp, data, size, attr_size); \
		} while (0);						\
		tprints("}");						\
		return (rval);						\
	}								\
/* End of END_BPF_CMD_DECODER definition. */

#define BPF_CMD_ENTRY(bpf_cmd)						\
	[bpf_cmd] = decode_ ## bpf_cmd

typedef DECL_BPF_CMD_DECODER((*bpf_cmd_decoder_t));

/*
 * A note about bpf syscall decoder: it doesn't perform any size sanity checks,
 * so even if it leads to partial copying of one of the fields, the command
 * handler will still use the (partially-copied-from-userspace, partially
 * zeroed) field value.  That's why we stop decoding and check for known sizes
 * that correspond to released versions of the structure used by the specific
 * command - it looks like the most sensible way to parse this insanity.
 */

static int
decode_attr_extra_data(struct tcb *const tcp,
		       const char *data,
		       unsigned int size,
		       const size_t attr_size)
{
	if (size <= attr_size)
		return 0;

	data += attr_size;
	size -= attr_size;

	unsigned int i;
	for (i = 0; i < size; ++i) {
		if (data[i]) {
			tprints(", ");
			if (abbrev(tcp)) {
				tprints("...");
			} else {
				tprintf("/* bytes %zu..%zu */ ",
					attr_size, attr_size + size - 1);
				print_quoted_string(data, size,
						    QUOTE_FORCE_HEX);
			}
			return RVAL_DECODED;
		}
	}

	return 0;
}

BEGIN_BPF_CMD_DECODER(BPF_MAP_CREATE)
{
	PRINT_FIELD_XVAL("{", attr, map_type, bpf_map_types,
			 "BPF_MAP_TYPE_???");
	PRINT_FIELD_U(", ", attr, key_size);
	PRINT_FIELD_U(", ", attr, value_size);
	PRINT_FIELD_U(", ", attr, max_entries);

	/* map_flags field was added in Linux commit v4.6-rc1~91^2~108^2~6. */
	if (len <= offsetof(struct BPF_MAP_CREATE_struct, map_flags))
		break;
	PRINT_FIELD_FLAGS(", ", attr, map_flags, bpf_map_flags, "BPF_F_???");

	/*
	 * inner_map_fd field was added in Linux commit
	 * v4.12-rc1~64^3~373^2~2.
	 */
	if (len <= offsetof(struct BPF_MAP_CREATE_struct, inner_map_fd))
		break;
	PRINT_FIELD_FD(", ", attr, inner_map_fd, tcp);

	/* numa_node field was added in Linux commit v4.14-rc1~130^2~196^2~1. */
	if (len <= offsetof(struct BPF_MAP_CREATE_struct, numa_node))
		break;
	if (attr.map_flags & BPF_F_NUMA_NODE) {
		/*
		 * Kernel uses the value of -1 as a designation for "no NUMA
		 * node specified", and even uses NUMA_NO_NODE constant;
		 * however, the constant definition is not a part of UAPI
		 * headers, thus we can't simply print this named constant
		 * instead of the value. Let's force verbose xlat style instead
		 * in order to provide the information for the user while
		 * not hampering the availability to derive the actual value
		 * without the access to the kernel headers.
		 */
		tprints(", numa_node=");
		printxvals_ex(attr.numa_node, NULL,
			      XLAT_STYLE_FMT_U | XLAT_STYLE_VERBOSE,
			      numa_node, NULL);
	}
}
END_BPF_CMD_DECODER(RVAL_DECODED | RVAL_FD)

BEGIN_BPF_CMD_DECODER(BPF_MAP_LOOKUP_ELEM)
{
	PRINT_FIELD_FD("{", attr, map_fd, tcp);
	PRINT_FIELD_X(", ", attr, key);
	PRINT_FIELD_X(", ", attr, value);
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_MAP_UPDATE_ELEM)
{
	PRINT_FIELD_FD("{", attr, map_fd, tcp);
	PRINT_FIELD_X(", ", attr, key);
	PRINT_FIELD_X(", ", attr, value);
	PRINT_FIELD_XVAL(", ", attr, flags, bpf_map_update_elem_flags,
			 "BPF_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_MAP_DELETE_ELEM)
{
	PRINT_FIELD_FD("{", attr, map_fd, tcp);
	PRINT_FIELD_X(", ", attr, key);
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_MAP_GET_NEXT_KEY)
{
	PRINT_FIELD_FD("{", attr, map_fd, tcp);
	PRINT_FIELD_X(", ", attr, key);
	PRINT_FIELD_X(", ", attr, next_key);
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_PROG_LOAD)
{
	PRINT_FIELD_XVAL("{", attr, prog_type, bpf_prog_types,
			 "BPF_PROG_TYPE_???");
	PRINT_FIELD_U(", ", attr, insn_cnt);
	PRINT_FIELD_X(", ", attr, insns);

	tprintf(", license=");
	print_big_u64_addr(attr.license);
	printstr(tcp, attr.license);

	/* log_* fields were added in Linux commit v3.18-rc1~52^2~1^2~4.  */
	if (len <= offsetof(struct BPF_PROG_LOAD_struct, log_level))
		break;
	PRINT_FIELD_U(", ", attr, log_level);
	PRINT_FIELD_U(", ", attr, log_size);
	PRINT_FIELD_X(", ", attr, log_buf);

	/* kern_version field was added in Linux commit v4.1-rc1~84^2~50.  */
	if (len <= offsetof(struct BPF_PROG_LOAD_struct, kern_version))
		break;
	tprintf(", kern_version=KERNEL_VERSION(%u, %u, %u)",
		attr.kern_version >> 16,
		(attr.kern_version >> 8) & 0xFF,
		attr.kern_version & 0xFF);

	/* prog_flags field was added in Linux commit v4.12-rc2~34^2~29^2~2.  */
	if (len <= offsetof(struct BPF_PROG_LOAD_struct, prog_flags))
		break;
	PRINT_FIELD_FLAGS(", ", attr, prog_flags, bpf_prog_flags, "BPF_F_???");

	/* prog_name field was added in Linux commit v4.15-rc1~84^2~605^2~4. */
	if (len <= offsetof(struct BPF_PROG_LOAD_struct, prog_name))
		break;
	PRINT_FIELD_CSTRING_SZ(", ", attr, prog_name,
			       MIN(sizeof(attr.prog_name),
				   len - offsetof(struct BPF_PROG_LOAD_struct,
						   prog_name)));

	/*
	 * prog_ifindex field was added as prog_target_ifindex in Linux commit
	 * v4.15-rc1~84^2~127^2~13 and renamed to its current name in
	 * v4.15-rc1~15^2~5^2~3^2~7.
	 */
	if (len <= offsetof(struct BPF_PROG_LOAD_struct, prog_ifindex))
		break;
	PRINT_FIELD_IFINDEX(", ", attr, prog_ifindex);
}
END_BPF_CMD_DECODER(RVAL_DECODED | RVAL_FD)

BEGIN_BPF_CMD_DECODER(BPF_OBJ_PIN)
{
	tprintf("{pathname=");
	print_big_u64_addr(attr.pathname);
	printpath(tcp, attr.pathname);

	PRINT_FIELD_FD(", ", attr, bpf_fd, tcp);
	if (len <= offsetofend(struct BPF_OBJ_PIN_struct, bpf_fd))
		break;

	/* file_flags field was added in Linux v4.15-rc1~84^2~384^2~4 */
	PRINT_FIELD_FLAGS(", ", attr, file_flags, bpf_file_mode_flags,
			  "BPF_F_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED | RVAL_FD)

#define decode_BPF_OBJ_GET decode_BPF_OBJ_PIN

BEGIN_BPF_CMD_DECODER(BPF_PROG_ATTACH)
{
	PRINT_FIELD_FD("{", attr, target_fd, tcp);
	PRINT_FIELD_FD(", ", attr, attach_bpf_fd, tcp);
	PRINT_FIELD_XVAL(", ", attr, attach_type, bpf_attach_type, "BPF_???");
	PRINT_FIELD_FLAGS(", ", attr, attach_flags, bpf_attach_flags,
			  "BPF_F_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_PROG_DETACH)
{
	PRINT_FIELD_FD("{", attr, target_fd, tcp);
	PRINT_FIELD_XVAL(", ", attr, attach_type, bpf_attach_type, "BPF_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_PROG_TEST_RUN)
{
	PRINT_FIELD_FD("{test={", attr, prog_fd, tcp);
	PRINT_FIELD_U(", ", attr, retval);
	PRINT_FIELD_U(", ", attr, data_size_in);
	PRINT_FIELD_U(", ", attr, data_size_out);
	PRINT_FIELD_X(", ", attr, data_in);
	PRINT_FIELD_X(", ", attr, data_out);
	PRINT_FIELD_U(", ", attr, repeat);
	PRINT_FIELD_U(", ", attr, duration);
	tprints("}");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_PROG_GET_NEXT_ID)
{
	PRINT_FIELD_U("{", attr, start_id);
	PRINT_FIELD_U(", ", attr, next_id);
	if (len <= offsetofend(struct BPF_PROG_GET_NEXT_ID_struct, next_id))
		break;

	/* open_flags field has been added in Linux v4.15-rc1~84^2~384^2~4 */
	PRINT_FIELD_FLAGS(", ", attr, open_flags, bpf_file_mode_flags,
			  "BPF_F_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

#define decode_BPF_MAP_GET_NEXT_ID decode_BPF_PROG_GET_NEXT_ID

BEGIN_BPF_CMD_DECODER(BPF_PROG_GET_FD_BY_ID)
{
	PRINT_FIELD_U("{", attr, prog_id);
	PRINT_FIELD_U(", ", attr, next_id);
	if (len <= offsetofend(struct BPF_PROG_GET_FD_BY_ID_struct, next_id))
		break;

	/* open_flags field has been added in Linux v4.15-rc1~84^2~384^2~4 */
	PRINT_FIELD_FLAGS(", ", attr, open_flags, bpf_file_mode_flags,
			  "BPF_F_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_MAP_GET_FD_BY_ID)
{
	PRINT_FIELD_U("{", attr, map_id);
	PRINT_FIELD_U(", ", attr, next_id);
	if (len <= offsetofend(struct BPF_MAP_GET_FD_BY_ID_struct, next_id))
		break;

	/* open_flags field has been added in Linux v4.15-rc1~84^2~384^2~4 */
	PRINT_FIELD_FLAGS(", ", attr, open_flags, bpf_file_mode_flags,
			  "BPF_F_???");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

BEGIN_BPF_CMD_DECODER(BPF_OBJ_GET_INFO_BY_FD)
{
	PRINT_FIELD_FD("{info={", attr, bpf_fd, tcp);
	PRINT_FIELD_U(", ", attr, info_len);
	PRINT_FIELD_X(", ", attr, info);
	tprints("}");
}
END_BPF_CMD_DECODER(RVAL_DECODED | RVAL_FD)

BEGIN_BPF_CMD_DECODER(BPF_PROG_QUERY)
{
	uint64_t prog_id_buf;

	if (entering(tcp)) {
		PRINT_FIELD_FD("{query={", attr, target_fd, tcp);
		PRINT_FIELD_XVAL(", ", attr, attach_type, bpf_attach_type,
				 "BPF_???");
		PRINT_FIELD_FLAGS(", ", attr, query_flags, bpf_query_flags,
				  "BPF_F_QUERY_???");
		PRINT_FIELD_FLAGS(", ", attr, attach_flags, bpf_attach_flags,
				  "BPF_F_???");

		tprints(", prog_ids=");

		if (!priv)
			priv = xcalloc(1, sizeof(*priv));

		priv->bpf_prog_query_stored = true;
		priv->bpf_prog_query_prog_cnt = attr.prog_cnt;

		set_tcb_priv_data(tcp, priv, free);

		return 0;
	}

	/*
	 * The issue here is that we can't pass pointers bigger than
	 * (our) kernel long ti print_array, so we opt out from decoding
	 * the array.
	 */
	if (syserror(tcp) || attr.prog_ids > max_kaddr())
		printaddr64(attr.prog_ids);
	else
		print_array(tcp, attr.prog_ids, attr.prog_cnt, &prog_id_buf,
			    sizeof(prog_id_buf), umoven_or_printaddr,
			    print_uint64_array_member, 0);

	tprints(", prog_cnt=");
	if (priv && priv->bpf_prog_query_stored
	    && priv->bpf_prog_query_prog_cnt != attr.prog_cnt)
		tprintf("%" PRIu32 " => ", priv->bpf_prog_query_prog_cnt);
	tprintf("%" PRIu32, attr.prog_cnt);
	tprints("}");
}
END_BPF_CMD_DECODER(RVAL_DECODED)

SYS_FUNC(bpf)
{
	static const bpf_cmd_decoder_t bpf_cmd_decoders[] = {
		BPF_CMD_ENTRY(BPF_MAP_CREATE),
		BPF_CMD_ENTRY(BPF_MAP_LOOKUP_ELEM),
		BPF_CMD_ENTRY(BPF_MAP_UPDATE_ELEM),
		BPF_CMD_ENTRY(BPF_MAP_DELETE_ELEM),
		BPF_CMD_ENTRY(BPF_MAP_GET_NEXT_KEY),
		BPF_CMD_ENTRY(BPF_PROG_LOAD),
		BPF_CMD_ENTRY(BPF_OBJ_PIN),
		BPF_CMD_ENTRY(BPF_OBJ_GET),
		BPF_CMD_ENTRY(BPF_PROG_ATTACH),
		BPF_CMD_ENTRY(BPF_PROG_DETACH),
		BPF_CMD_ENTRY(BPF_PROG_TEST_RUN),
		BPF_CMD_ENTRY(BPF_PROG_GET_NEXT_ID),
		BPF_CMD_ENTRY(BPF_MAP_GET_NEXT_ID),
		BPF_CMD_ENTRY(BPF_PROG_GET_FD_BY_ID),
		BPF_CMD_ENTRY(BPF_MAP_GET_FD_BY_ID),
		BPF_CMD_ENTRY(BPF_OBJ_GET_INFO_BY_FD),
		BPF_CMD_ENTRY(BPF_PROG_QUERY),
	};

	static char *buf;
	struct bpf_priv_data *priv = NULL;

	const unsigned int cmd = tcp->u_arg[0];
	const kernel_ulong_t addr = tcp->u_arg[1];
	const unsigned int size = tcp->u_arg[2];
	int rc = RVAL_DECODED;

	if (!buf)
		buf = xmalloc(get_pagesize());

	if (entering(tcp)) {
		printxval(bpf_commands, cmd, "BPF_???");
		tprints(", ");
	} else {
		priv = get_tcb_priv_data(tcp);
	}

	if (size > 0
	    && size <= get_pagesize()
	    && cmd < ARRAY_SIZE(bpf_cmd_decoders)
	    && bpf_cmd_decoders[cmd]) {
		if (!umoven_or_printaddr_ignore_syserror(tcp, addr, size, buf))
			rc = bpf_cmd_decoders[cmd](tcp, addr, size, buf, priv);
	} else {
		printaddr(addr);
	}

	if (exiting(tcp) || (rc & RVAL_DECODED))
		tprintf(", %u", size);

	return rc;
}