samples/bpf/task_fd_query_user.c

   1 // SPDX-License-Identifier: GPL-2.0
   2
   3 #include <stdio.h>
   4 #include <stdlib.h>
   5 #include <signal.h>
   6 #include <unistd.h>
   7 #include <stdbool.h>
   8 #include <string.h>
   9 #include <stdint.h>
  10 #include <fcntl.h>
  11 #include <linux/bpf.h>
  12 #include <sys/ioctl.h>
  13 #include <sys/resource.h>
  14 #include <sys/types.h>
  15 #include <sys/stat.h>
  16 #include <linux/perf_event.h>
  17
  18 #include <bpf/libbpf.h>
  19 #include "bpf_load.h"
  20 #include "bpf_util.h"
  21 #include "perf-sys.h"
  22 #include "trace_helpers.h"
  23
  24 #define CHECK_PERROR_RET(condition) ({                  \
  25         int __ret = !!(condition);                      \
  26         if (__ret) {                                    \
  27                 printf("FAIL: %s:\n", __func__);        \
  28                 perror("    ");                 \
  29                 return -1;                              \
  30         }                                               \
  31 })
  32
  33 #define CHECK_AND_RET(condition) ({                     \
  34         int __ret = !!(condition);                      \
  35         if (__ret)                                      \
  36                 return -1;                              \
  37 })
  38
  39 static __u64 ptr_to_u64(void *ptr)
  40 {
  41         return (__u64) (unsigned long) ptr;
  42 }
  43
  44 #define PMU_TYPE_FILE "/sys/bus/event_source/devices/%s/type"
  45 static int bpf_find_probe_type(const char *event_type)
  46 {
  47         char buf[256];
  48         int fd, ret;
  49
  50         ret = snprintf(buf, sizeof(buf), PMU_TYPE_FILE, event_type);
  51         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  52
  53         fd = open(buf, O_RDONLY);
  54         CHECK_PERROR_RET(fd < 0);
  55
  56         ret = read(fd, buf, sizeof(buf));
  57         close(fd);
  58         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  59
  60         errno = 0;
  61         ret = (int)strtol(buf, NULL, 10);
  62         CHECK_PERROR_RET(errno);
  63         return ret;
  64 }
  65
  66 #define PMU_RETPROBE_FILE "/sys/bus/event_source/devices/%s/format/retprobe"
  67 static int bpf_get_retprobe_bit(const char *event_type)
  68 {
  69         char buf[256];
  70         int fd, ret;
  71
  72         ret = snprintf(buf, sizeof(buf), PMU_RETPROBE_FILE, event_type);
  73         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  74
  75         fd = open(buf, O_RDONLY);
  76         CHECK_PERROR_RET(fd < 0);
  77
  78         ret = read(fd, buf, sizeof(buf));
  79         close(fd);
  80         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  81         CHECK_PERROR_RET(strlen(buf) < strlen("config:"));
  82
  83         errno = 0;
  84         ret = (int)strtol(buf + strlen("config:"), NULL, 10);
  85         CHECK_PERROR_RET(errno);
  86         return ret;
  87 }
  88
  89 static int test_debug_fs_kprobe(int prog_fd_idx, const char *fn_name,
  90                                 __u32 expected_fd_type)
  91 {
  92         __u64 probe_offset, probe_addr;
  93         __u32 len, prog_id, fd_type;
  94         char buf[256];
  95         int err;
  96
  97         len = sizeof(buf);
  98         err = bpf_task_fd_query(getpid(), event_fd[prog_fd_idx], 0, buf, &len,
  99                                 &prog_id, &fd_type, &probe_offset,
 100                                 &probe_addr);
 101         if (err < 0) {
 102                 printf("FAIL: %s, for event_fd idx %d, fn_name %s\n",
 103                        __func__, prog_fd_idx, fn_name);
 104                 perror("    :");
 105                 return -1;
 106         }
 107         if (strcmp(buf, fn_name) != 0 ||
 108             fd_type != expected_fd_type ||
 109             probe_offset != 0x0 || probe_addr != 0x0) {
 110                 printf("FAIL: bpf_trace_event_query(event_fd[%d]):\n",
 111                        prog_fd_idx);
 112                 printf("buf: %s, fd_type: %u, probe_offset: 0x%llx,"
 113                        " probe_addr: 0x%llx\n",
 114                        buf, fd_type, probe_offset, probe_addr);
 115                 return -1;
 116         }
 117         return 0;
 118 }
 119
 120 static int test_nondebug_fs_kuprobe_common(const char *event_type,
 121         const char *name, __u64 offset, __u64 addr, bool is_return,
 122         char *buf, __u32 *buf_len, __u32 *prog_id, __u32 *fd_type,
 123         __u64 *probe_offset, __u64 *probe_addr)
 124 {
 125         int is_return_bit = bpf_get_retprobe_bit(event_type);
 126         int type = bpf_find_probe_type(event_type);
 127         struct perf_event_attr attr = {};
 128         int fd;
 129
 130         if (type < 0 || is_return_bit < 0) {
 131                 printf("FAIL: %s incorrect type (%d) or is_return_bit (%d)\n",
 132                         __func__, type, is_return_bit);
 133                 return -1;
 134         }
 135
 136         attr.sample_period = 1;
 137         attr.wakeup_events = 1;
 138         if (is_return)
 139                 attr.config |= 1 << is_return_bit;
 140
 141         if (name) {
 142                 attr.config1 = ptr_to_u64((void *)name);
 143                 attr.config2 = offset;
 144         } else {
 145                 attr.config1 = 0;
 146                 attr.config2 = addr;
 147         }
 148         attr.size = sizeof(attr);
 149         attr.type = type;
 150
 151         fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
 152         CHECK_PERROR_RET(fd < 0);
 153
 154         CHECK_PERROR_RET(ioctl(fd, PERF_EVENT_IOC_ENABLE, 0) < 0);
 155         CHECK_PERROR_RET(ioctl(fd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) < 0);
 156         CHECK_PERROR_RET(bpf_task_fd_query(getpid(), fd, 0, buf, buf_len,
 157                          prog_id, fd_type, probe_offset, probe_addr) < 0);
 158
 159         return 0;
 160 }
 161
 162 static int test_nondebug_fs_probe(const char *event_type, const char *name,
 163                                   __u64 offset, __u64 addr, bool is_return,
 164                                   __u32 expected_fd_type,
 165                                   __u32 expected_ret_fd_type,
 166                                   char *buf, __u32 buf_len)
 167 {
 168         __u64 probe_offset, probe_addr;
 169         __u32 prog_id, fd_type;
 170         int err;
 171
 172         err = test_nondebug_fs_kuprobe_common(event_type, name,
 173                                               offset, addr, is_return,
 174                                               buf, &buf_len, &prog_id,
 175                                               &fd_type, &probe_offset,
 176                                               &probe_addr);
 177         if (err < 0) {
 178                 printf("FAIL: %s, "
 179                        "for name %s, offset 0x%llx, addr 0x%llx, is_return %d\n",
 180                        __func__, name ? name : "", offset, addr, is_return);
 181                 perror("    :");
 182                 return -1;
 183         }
 184         if ((is_return && fd_type != expected_ret_fd_type) ||
 185             (!is_return && fd_type != expected_fd_type)) {
 186                 printf("FAIL: %s, incorrect fd_type %u\n",
 187                        __func__, fd_type);
 188                 return -1;
 189         }
 190         if (name) {
 191                 if (strcmp(name, buf) != 0) {
 192                         printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
 193                         return -1;
 194                 }
 195                 if (probe_offset != offset) {
 196                         printf("FAIL: %s, incorrect probe_offset 0x%llx\n",
 197                                __func__, probe_offset);
 198                         return -1;
 199                 }
 200         } else {
 201                 if (buf_len != 0) {
 202                         printf("FAIL: %s, incorrect buf %p\n",
 203                                __func__, buf);
 204                         return -1;
 205                 }
 206
 207                 if (probe_addr != addr) {
 208                         printf("FAIL: %s, incorrect probe_addr 0x%llx\n",
 209                                __func__, probe_addr);
 210                         return -1;
 211                 }
 212         }
 213         return 0;
 214 }
 215
 216 static int test_debug_fs_uprobe(char *binary_path, long offset, bool is_return)
 217 {
 218         const char *event_type = "uprobe";
 219         struct perf_event_attr attr = {};
 220         char buf[256], event_alias[sizeof("test_1234567890")];
 221         __u64 probe_offset, probe_addr;
 222         __u32 len, prog_id, fd_type;
 223         int err, res, kfd, efd;
 224         ssize_t bytes;
 225
 226         snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/%s_events",
 227                  event_type);
 228         kfd = open(buf, O_WRONLY | O_APPEND, 0);
 229         CHECK_PERROR_RET(kfd < 0);
 230
 231         res = snprintf(event_alias, sizeof(event_alias), "test_%d", getpid());
 232         CHECK_PERROR_RET(res < 0 || res >= sizeof(event_alias));
 233
 234         res = snprintf(buf, sizeof(buf), "%c:%ss/%s %s:0x%lx",
 235                        is_return ? 'r' : 'p', event_type, event_alias,
 236                        binary_path, offset);
 237         CHECK_PERROR_RET(res < 0 || res >= sizeof(buf));
 238         CHECK_PERROR_RET(write(kfd, buf, strlen(buf)) < 0);
 239
 240         close(kfd);
 241         kfd = -1;
 242
 243         snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%ss/%s/id",
 244                  event_type, event_alias);
 245         efd = open(buf, O_RDONLY, 0);
 246         CHECK_PERROR_RET(efd < 0);
 247
 248         bytes = read(efd, buf, sizeof(buf));
 249         CHECK_PERROR_RET(bytes <= 0 || bytes >= sizeof(buf));
 250         close(efd);
 251         buf[bytes] = '\0';
 252
 253         attr.config = strtol(buf, NULL, 0);
 254         attr.type = PERF_TYPE_TRACEPOINT;
 255         attr.sample_period = 1;
 256         attr.wakeup_events = 1;
 257         kfd = sys_perf_event_open(&attr, -1, 0, -1, PERF_FLAG_FD_CLOEXEC);
 258         CHECK_PERROR_RET(kfd < 0);
 259         CHECK_PERROR_RET(ioctl(kfd, PERF_EVENT_IOC_SET_BPF, prog_fd[0]) < 0);
 260         CHECK_PERROR_RET(ioctl(kfd, PERF_EVENT_IOC_ENABLE, 0) < 0);
 261
 262         len = sizeof(buf);
 263         err = bpf_task_fd_query(getpid(), kfd, 0, buf, &len,
 264                                 &prog_id, &fd_type, &probe_offset,
 265                                 &probe_addr);
 266         if (err < 0) {
 267                 printf("FAIL: %s, binary_path %s\n", __func__, binary_path);
 268                 perror("    :");
 269                 return -1;
 270         }
 271         if ((is_return && fd_type != BPF_FD_TYPE_URETPROBE) ||
 272             (!is_return && fd_type != BPF_FD_TYPE_UPROBE)) {
 273                 printf("FAIL: %s, incorrect fd_type %u\n", __func__,
 274                        fd_type);
 275                 return -1;
 276         }
 277         if (strcmp(binary_path, buf) != 0) {
 278                 printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
 279                 return -1;
 280         }
 281         if (probe_offset != offset) {
 282                 printf("FAIL: %s, incorrect probe_offset 0x%llx\n", __func__,
 283                        probe_offset);
 284                 return -1;
 285         }
 286
 287         close(kfd);
 288         return 0;
 289 }
 290
 291 int main(int argc, char **argv)
 292 {
 293         struct rlimit r = {1024*1024, RLIM_INFINITY};
 294         extern char __executable_start;
 295         char filename[256], buf[256];
 296         __u64 uprobe_file_offset;
 297
 298         snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 299         if (setrlimit(RLIMIT_MEMLOCK, &r)) {
 300                 perror("setrlimit(RLIMIT_MEMLOCK)");
 301                 return 1;
 302         }
 303
 304         if (load_kallsyms()) {
 305                 printf("failed to process /proc/kallsyms\n");
 306                 return 1;
 307         }
 308
 309         if (load_bpf_file(filename)) {
 310                 printf("%s", bpf_log_buf);
 311                 return 1;
 312         }
 313
 314         /* test two functions in the corresponding *_kern.c file */
 315         CHECK_AND_RET(test_debug_fs_kprobe(0, "blk_mq_start_request",
 316                                            BPF_FD_TYPE_KPROBE));
 317         CHECK_AND_RET(test_debug_fs_kprobe(1, "blk_account_io_completion",
 318                                            BPF_FD_TYPE_KRETPROBE));
 319
 320         /* test nondebug fs kprobe */
 321         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
 322                                              false, BPF_FD_TYPE_KPROBE,
 323                                              BPF_FD_TYPE_KRETPROBE,
 324                                              buf, sizeof(buf)));
 325 #ifdef __x86_64__
 326         /* set a kprobe on "bpf_check + 0x5", which is x64 specific */
 327         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x5, 0x0,
 328                                              false, BPF_FD_TYPE_KPROBE,
 329                                              BPF_FD_TYPE_KRETPROBE,
 330                                              buf, sizeof(buf)));
 331 #endif
 332         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
 333                                              true, BPF_FD_TYPE_KPROBE,
 334                                              BPF_FD_TYPE_KRETPROBE,
 335                                              buf, sizeof(buf)));
 336         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 337                                              ksym_get_addr("bpf_check"), false,
 338                                              BPF_FD_TYPE_KPROBE,
 339                                              BPF_FD_TYPE_KRETPROBE,
 340                                              buf, sizeof(buf)));
 341         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 342                                              ksym_get_addr("bpf_check"), false,
 343                                              BPF_FD_TYPE_KPROBE,
 344                                              BPF_FD_TYPE_KRETPROBE,
 345                                              NULL, 0));
 346         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 347                                              ksym_get_addr("bpf_check"), true,
 348                                              BPF_FD_TYPE_KPROBE,
 349                                              BPF_FD_TYPE_KRETPROBE,
 350                                              buf, sizeof(buf)));
 351         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 352                                              ksym_get_addr("bpf_check"), true,
 353                                              BPF_FD_TYPE_KPROBE,
 354                                              BPF_FD_TYPE_KRETPROBE,
 355                                              0, 0));
 356
 357         /* test nondebug fs uprobe */
 358         /* the calculation of uprobe file offset is based on gcc 7.3.1 on x64
 359          * and the default linker script, which defines __executable_start as
 360          * the start of the .text section. The calculation could be different
 361          * on different systems with different compilers. The right way is
 362          * to parse the ELF file. We took a shortcut here.
 363          */
 364         uprobe_file_offset = (__u64)main - (__u64)&__executable_start;
 365         CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
 366                                              uprobe_file_offset, 0x0, false,
 367                                              BPF_FD_TYPE_UPROBE,
 368                                              BPF_FD_TYPE_URETPROBE,
 369                                              buf, sizeof(buf)));
 370         CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
 371                                              uprobe_file_offset, 0x0, true,
 372                                              BPF_FD_TYPE_UPROBE,
 373                                              BPF_FD_TYPE_URETPROBE,
 374                                              buf, sizeof(buf)));
 375
 376         /* test debug fs uprobe */
 377         CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
 378                                            false));
 379         CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
 380                                            true));
 381
 382         return 0;
 383 }