samples/bpf: add a samples/bpf test for BPF_TASK_FD_QUERY
authorYonghong Song <yhs@fb.com>
Thu, 24 May 2018 18:21:56 +0000 (11:21 -0700)
committerAlexei Starovoitov <ast@kernel.org>
Fri, 25 May 2018 01:18:20 +0000 (18:18 -0700)
This is mostly to test kprobe/uprobe which needs kernel headers.

Signed-off-by: Yonghong Song <yhs@fb.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
samples/bpf/Makefile
samples/bpf/task_fd_query_kern.c [new file with mode: 0644]
samples/bpf/task_fd_query_user.c [new file with mode: 0644]

index 62d1aa1a4cf3786da0d82a20e2937b96e22f2b49..7dc85ed0ce4bfb72630e9f5ffc2d4a1c8086039d 100644 (file)
@@ -51,6 +51,7 @@ hostprogs-y += cpustat
 hostprogs-y += xdp_adjust_tail
 hostprogs-y += xdpsock
 hostprogs-y += xdp_fwd
+hostprogs-y += task_fd_query
 
 # Libbpf dependencies
 LIBBPF = $(TOOLS_PATH)/lib/bpf/libbpf.a
@@ -105,6 +106,7 @@ cpustat-objs := bpf_load.o cpustat_user.o
 xdp_adjust_tail-objs := xdp_adjust_tail_user.o
 xdpsock-objs := bpf_load.o xdpsock_user.o
 xdp_fwd-objs := bpf_load.o xdp_fwd_user.o
+task_fd_query-objs := bpf_load.o task_fd_query_user.o $(TRACE_HELPERS)
 
 # Tell kbuild to always build the programs
 always := $(hostprogs-y)
@@ -160,6 +162,7 @@ always += cpustat_kern.o
 always += xdp_adjust_tail_kern.o
 always += xdpsock_kern.o
 always += xdp_fwd_kern.o
+always += task_fd_query_kern.o
 
 HOSTCFLAGS += -I$(objtree)/usr/include
 HOSTCFLAGS += -I$(srctree)/tools/lib/
@@ -175,6 +178,7 @@ HOSTCFLAGS_offwaketime_user.o += -I$(srctree)/tools/lib/bpf/
 HOSTCFLAGS_spintest_user.o += -I$(srctree)/tools/lib/bpf/
 HOSTCFLAGS_trace_event_user.o += -I$(srctree)/tools/lib/bpf/
 HOSTCFLAGS_sampleip_user.o += -I$(srctree)/tools/lib/bpf/
+HOSTCFLAGS_task_fd_query_user.o += -I$(srctree)/tools/lib/bpf/
 
 HOST_LOADLIBES         += $(LIBBPF) -lelf
 HOSTLOADLIBES_tracex4          += -lrt
diff --git a/samples/bpf/task_fd_query_kern.c b/samples/bpf/task_fd_query_kern.c
new file mode 100644 (file)
index 0000000..f4b0a9e
--- /dev/null
@@ -0,0 +1,19 @@
+// SPDX-License-Identifier: GPL-2.0
+#include <linux/version.h>
+#include <linux/ptrace.h>
+#include <uapi/linux/bpf.h>
+#include "bpf_helpers.h"
+
+SEC("kprobe/blk_start_request")
+int bpf_prog1(struct pt_regs *ctx)
+{
+       return 0;
+}
+
+SEC("kretprobe/blk_account_io_completion")
+int bpf_prog2(struct pt_regs *ctx)
+{
+       return 0;
+}
+char _license[] SEC("license") = "GPL";
+u32 _version SEC("version") = LINUX_VERSION_CODE;
diff --git a/samples/bpf/task_fd_query_user.c b/samples/bpf/task_fd_query_user.c
new file mode 100644 (file)
index 0000000..8381d79
--- /dev/null
@@ -0,0 +1,382 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <signal.h>
+#include <unistd.h>
+#include <stdbool.h>
+#include <string.h>
+#include <stdint.h>
+#include <fcntl.h>
+#include <linux/bpf.h>
+#include <sys/ioctl.h>
+#include <sys/resource.h>
+#include <sys/types.h>
+#include <sys/stat.h>
+
+#include "libbpf.h"
+#include "bpf_load.h"
+#include "bpf_util.h"
+#include "perf-sys.h"
+#include "trace_helpers.h"
+
+#define CHECK_PERROR_RET(condition) ({                 \
+       int __ret = !!(condition);                      \
+       if (__ret) {                                    \
+               printf("FAIL: %s:\n", __func__);        \
+               perror("    ");                 \
+               return -1;                              \
+       }                                               \
+})
+
+#define CHECK_AND_RET(condition) ({                    \
+       int __ret = !!(condition);                      \
+       if (__ret)                                      \
+               return -1;                              \
+})
+
+static __u64 ptr_to_u64(void *ptr)
+{
+       return (__u64) (unsigned long) ptr;
+}
+
+#define PMU_TYPE_FILE "/sys/bus/event_source/devices/%s/type"
+static int bpf_find_probe_type(const char *event_type)
+{
+       char buf[256];
+       int fd, ret;
+
+       ret = snprintf(buf, sizeof(buf), PMU_TYPE_FILE, event_type);
+       CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
+
+       fd = open(buf, O_RDONLY);
+       CHECK_PERROR_RET(fd < 0);
+
+       ret = read(fd, buf, sizeof(buf));
+       close(fd);
+       CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
+
+       errno = 0;
+       ret = (int)strtol(buf, NULL, 10);
+       CHECK_PERROR_RET(errno);
+       return ret;
+}
+
+#define PMU_RETPROBE_FILE "/sys/bus/event_source/devices/%s/format/retprobe"
+static int bpf_get_retprobe_bit(const char *event_type)
+{
+       char buf[256];
+       int fd, ret;
+
+       ret = snprintf(buf, sizeof(buf), PMU_RETPROBE_FILE, event_type);
+       CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
+
+       fd = open(buf, O_RDONLY);
+       CHECK_PERROR_RET(fd < 0);
+
+       ret = read(fd, buf, sizeof(buf));
+       close(fd);
+       CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
+       CHECK_PERROR_RET(strlen(buf) < strlen("config:"));
+
+       errno = 0;
+       ret = (int)strtol(buf + strlen("config:"), NULL, 10);
+       CHECK_PERROR_RET(errno);
+       return ret;
+}
+
+static int test_debug_fs_kprobe(int prog_fd_idx, const char *fn_name,
+                               __u32 expected_fd_type)
+{
+       __u64 probe_offset, probe_addr;
+       __u32 len, prog_id, fd_type;
+       char buf[256];
+       int err;
+
+       len = sizeof(buf);
+       err = bpf_task_fd_query(getpid(), event_fd[prog_fd_idx], 0, buf, &len,
+                               &prog_id, &fd_type, &probe_offset,
+                               &probe_addr);
+       if (err < 0) {
+               printf("FAIL: %s, for event_fd idx %d, fn_name %s\n",
+                      __func__, prog_fd_idx, fn_name);
+               perror("    :");
+               return -1;
+       }
+       if (strcmp(buf, fn_name) != 0 ||
+           fd_type != expected_fd_type ||
+           probe_offset != 0x0 || probe_addr != 0x0) {
+               printf("FAIL: bpf_trace_event_query(event_fd[%d]):\n",
+                      prog_fd_idx);
+               printf("buf: %s, fd_type: %u, probe_offset: 0x%llx,"
+                      " probe_addr: 0x%llx\n",
+                      buf, fd_type, probe_offset, probe_addr);
+               return -1;
+       }
+       return 0;
+}
+
+static int test_nondebug_fs_kuprobe_common(const char *event_type,
+       const char *name, __u64 offset, __u64 addr, bool is_return,
+       char *buf, __u32 *buf_len, __u32 *prog_id, __u32 *fd_type,
+       __u64 *probe_offset, __u64 *probe_addr)
+{
+       int is_return_bit = bpf_get_retprobe_bit(event_type);
+       int type = bpf_find_probe_type(event_type);
+       struct perf_event_attr attr = {};
+       int fd;
+
+       if (type < 0 || is_return_bit < 0) {
+               printf("FAIL: %s incorrect type (%d) or is_return_bit (%d)\n",
+                       __func__, type, is_return_bit);
+               return -1;
+       }
+
+       attr.sample_period = 1;
+       attr.wakeup_events = 1;
+       if (is_return)
+               attr.config |= 1 << is_return_bit;
+
+       if (name) {
+               attr.config1 = ptr_to_u64((void *)name);
+               attr.config2 = offset;
+       } else {
+               attr.config1 = 0;
+               attr.config2 = addr;
+       }
+       attr.size = sizeof(attr);
+       attr.type = type;
+
+       fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
+       CHECK_PERROR_RET(fd < 0);
+
+       CHECK_PERROR_RET(ioctl(fd, PERF_EVENT_IOC_ENABLE, 0) < 0);
+       CHECK_PERROR_RET(ioctl(fd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) < 0);
+       CHECK_PERROR_RET(bpf_task_fd_query(getpid(), fd, 0, buf, buf_len,
+                        prog_id, fd_type, probe_offset, probe_addr) < 0);
+
+       return 0;
+}
+
+static int test_nondebug_fs_probe(const char *event_type, const char *name,
+                                 __u64 offset, __u64 addr, bool is_return,
+                                 __u32 expected_fd_type,
+                                 __u32 expected_ret_fd_type,
+                                 char *buf, __u32 buf_len)
+{
+       __u64 probe_offset, probe_addr;
+       __u32 prog_id, fd_type;
+       int err;
+
+       err = test_nondebug_fs_kuprobe_common(event_type, name,
+                                             offset, addr, is_return,
+                                             buf, &buf_len, &prog_id,
+                                             &fd_type, &probe_offset,
+                                             &probe_addr);
+       if (err < 0) {
+               printf("FAIL: %s, "
+                      "for name %s, offset 0x%llx, addr 0x%llx, is_return %d\n",
+                      __func__, name ? name : "", offset, addr, is_return);
+               perror("    :");
+               return -1;
+       }
+       if ((is_return && fd_type != expected_ret_fd_type) ||
+           (!is_return && fd_type != expected_fd_type)) {
+               printf("FAIL: %s, incorrect fd_type %u\n",
+                      __func__, fd_type);
+               return -1;
+       }
+       if (name) {
+               if (strcmp(name, buf) != 0) {
+                       printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
+                       return -1;
+               }
+               if (probe_offset != offset) {
+                       printf("FAIL: %s, incorrect probe_offset 0x%llx\n",
+                              __func__, probe_offset);
+                       return -1;
+               }
+       } else {
+               if (buf_len != 0) {
+                       printf("FAIL: %s, incorrect buf %p\n",
+                              __func__, buf);
+                       return -1;
+               }
+
+               if (probe_addr != addr) {
+                       printf("FAIL: %s, incorrect probe_addr 0x%llx\n",
+                              __func__, probe_addr);
+                       return -1;
+               }
+       }
+       return 0;
+}
+
+static int test_debug_fs_uprobe(char *binary_path, long offset, bool is_return)
+{
+       const char *event_type = "uprobe";
+       struct perf_event_attr attr = {};
+       char buf[256], event_alias[256];
+       __u64 probe_offset, probe_addr;
+       __u32 len, prog_id, fd_type;
+       int err, res, kfd, efd;
+       ssize_t bytes;
+
+       snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/%s_events",
+                event_type);
+       kfd = open(buf, O_WRONLY | O_APPEND, 0);
+       CHECK_PERROR_RET(kfd < 0);
+
+       res = snprintf(event_alias, sizeof(event_alias), "test_%d", getpid());
+       CHECK_PERROR_RET(res < 0 || res >= sizeof(event_alias));
+
+       res = snprintf(buf, sizeof(buf), "%c:%ss/%s %s:0x%lx",
+                      is_return ? 'r' : 'p', event_type, event_alias,
+                      binary_path, offset);
+       CHECK_PERROR_RET(res < 0 || res >= sizeof(buf));
+       CHECK_PERROR_RET(write(kfd, buf, strlen(buf)) < 0);
+
+       close(kfd);
+       kfd = -1;
+
+       snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%ss/%s/id",
+                event_type, event_alias);
+       efd = open(buf, O_RDONLY, 0);
+       CHECK_PERROR_RET(efd < 0);
+
+       bytes = read(efd, buf, sizeof(buf));
+       CHECK_PERROR_RET(bytes <= 0 || bytes >= sizeof(buf));
+       close(efd);
+       buf[bytes] = '\0';
+
+       attr.config = strtol(buf, NULL, 0);
+       attr.type = PERF_TYPE_TRACEPOINT;
+       attr.sample_period = 1;
+       attr.wakeup_events = 1;
+       kfd = sys_perf_event_open(&attr, -1, 0, -1, PERF_FLAG_FD_CLOEXEC);
+       CHECK_PERROR_RET(kfd < 0);
+       CHECK_PERROR_RET(ioctl(kfd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) < 0);
+       CHECK_PERROR_RET(ioctl(kfd, PERF_EVENT_IOC_ENABLE, 0) < 0);
+
+       len = sizeof(buf);
+       err = bpf_task_fd_query(getpid(), kfd, 0, buf, &len,
+                               &prog_id, &fd_type, &probe_offset,
+                               &probe_addr);
+       if (err < 0) {
+               printf("FAIL: %s, binary_path %s\n", __func__, binary_path);
+               perror("    :");
+               return -1;
+       }
+       if ((is_return && fd_type != BPF_FD_TYPE_URETPROBE) ||
+           (!is_return && fd_type != BPF_FD_TYPE_UPROBE)) {
+               printf("FAIL: %s, incorrect fd_type %u\n", __func__,
+                      fd_type);
+               return -1;
+       }
+       if (strcmp(binary_path, buf) != 0) {
+               printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
+               return -1;
+       }
+       if (probe_offset != offset) {
+               printf("FAIL: %s, incorrect probe_offset 0x%llx\n", __func__,
+                      probe_offset);
+               return -1;
+       }
+
+       close(kfd);
+       return 0;
+}
+
+int main(int argc, char **argv)
+{
+       struct rlimit r = {1024*1024, RLIM_INFINITY};
+       extern char __executable_start;
+       char filename[256], buf[256];
+       __u64 uprobe_file_offset;
+
+       snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
+       if (setrlimit(RLIMIT_MEMLOCK, &r)) {
+               perror("setrlimit(RLIMIT_MEMLOCK)");
+               return 1;
+       }
+
+       if (load_kallsyms()) {
+               printf("failed to process /proc/kallsyms\n");
+               return 1;
+       }
+
+       if (load_bpf_file(filename)) {
+               printf("%s", bpf_log_buf);
+               return 1;
+       }
+
+       /* test two functions in the corresponding *_kern.c file */
+       CHECK_AND_RET(test_debug_fs_kprobe(0, "blk_start_request",
+                                          BPF_FD_TYPE_KPROBE));
+       CHECK_AND_RET(test_debug_fs_kprobe(1, "blk_account_io_completion",
+                                          BPF_FD_TYPE_KRETPROBE));
+
+       /* test nondebug fs kprobe */
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
+                                            false, BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            buf, sizeof(buf)));
+#ifdef __x86_64__
+       /* set a kprobe on "bpf_check + 0x5", which is x64 specific */
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x5, 0x0,
+                                            false, BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            buf, sizeof(buf)));
+#endif
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
+                                            true, BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            buf, sizeof(buf)));
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
+                                            ksym_get_addr("bpf_check"), false,
+                                            BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            buf, sizeof(buf)));
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
+                                            ksym_get_addr("bpf_check"), false,
+                                            BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            NULL, 0));
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
+                                            ksym_get_addr("bpf_check"), true,
+                                            BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            buf, sizeof(buf)));
+       CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
+                                            ksym_get_addr("bpf_check"), true,
+                                            BPF_FD_TYPE_KPROBE,
+                                            BPF_FD_TYPE_KRETPROBE,
+                                            0, 0));
+
+       /* test nondebug fs uprobe */
+       /* the calculation of uprobe file offset is based on gcc 7.3.1 on x64
+        * and the default linker script, which defines __executable_start as
+        * the start of the .text section. The calculation could be different
+        * on different systems with different compilers. The right way is
+        * to parse the ELF file. We took a shortcut here.
+        */
+       uprobe_file_offset = (__u64)main - (__u64)&__executable_start;
+       CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
+                                            uprobe_file_offset, 0x0, false,
+                                            BPF_FD_TYPE_UPROBE,
+                                            BPF_FD_TYPE_URETPROBE,
+                                            buf, sizeof(buf)));
+       CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
+                                            uprobe_file_offset, 0x0, true,
+                                            BPF_FD_TYPE_UPROBE,
+                                            BPF_FD_TYPE_URETPROBE,
+                                            buf, sizeof(buf)));
+
+       /* test debug fs uprobe */
+       CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
+                                          false));
+       CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
+                                          true));
+
+       return 0;
+}