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/types.h>
  14 #include <sys/stat.h>
  15 #include <linux/perf_event.h>
  16
  17 #include <bpf/bpf.h>
  18 #include <bpf/libbpf.h>
  19 #include "bpf_util.h"
  20 #include "perf-sys.h"
  21 #include "trace_helpers.h"
  22
  23 static struct bpf_program *progs[2];
  24 static struct bpf_link *links[2];
  25
  26 #define CHECK_PERROR_RET(condition) ({                  \
  27         int __ret = !!(condition);                      \
  28         if (__ret) {                                    \
  29                 printf("FAIL: %s:\n", __func__);        \
  30                 perror("    ");                 \
  31                 return -1;                              \
  32         }                                               \
  33 })
  34
  35 #define CHECK_AND_RET(condition) ({                     \
  36         int __ret = !!(condition);                      \
  37         if (__ret)                                      \
  38                 return -1;                              \
  39 })
  40
  41 static __u64 ptr_to_u64(void *ptr)
  42 {
  43         return (__u64) (unsigned long) ptr;
  44 }
  45
  46 #define PMU_TYPE_FILE "/sys/bus/event_source/devices/%s/type"
  47 static int bpf_find_probe_type(const char *event_type)
  48 {
  49         char buf[256];
  50         int fd, ret;
  51
  52         ret = snprintf(buf, sizeof(buf), PMU_TYPE_FILE, event_type);
  53         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  54
  55         fd = open(buf, O_RDONLY);
  56         CHECK_PERROR_RET(fd < 0);
  57
  58         ret = read(fd, buf, sizeof(buf));
  59         close(fd);
  60         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  61
  62         errno = 0;
  63         ret = (int)strtol(buf, NULL, 10);
  64         CHECK_PERROR_RET(errno);
  65         return ret;
  66 }
  67
  68 #define PMU_RETPROBE_FILE "/sys/bus/event_source/devices/%s/format/retprobe"
  69 static int bpf_get_retprobe_bit(const char *event_type)
  70 {
  71         char buf[256];
  72         int fd, ret;
  73
  74         ret = snprintf(buf, sizeof(buf), PMU_RETPROBE_FILE, event_type);
  75         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  76
  77         fd = open(buf, O_RDONLY);
  78         CHECK_PERROR_RET(fd < 0);
  79
  80         ret = read(fd, buf, sizeof(buf));
  81         close(fd);
  82         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  83         CHECK_PERROR_RET(strlen(buf) < strlen("config:"));
  84
  85         errno = 0;
  86         ret = (int)strtol(buf + strlen("config:"), NULL, 10);
  87         CHECK_PERROR_RET(errno);
  88         return ret;
  89 }
  90
  91 static int test_debug_fs_kprobe(int link_idx, const char *fn_name,
  92                                 __u32 expected_fd_type)
  93 {
  94         __u64 probe_offset, probe_addr;
  95         __u32 len, prog_id, fd_type;
  96         int err, event_fd;
  97         char buf[256];
  98
  99         len = sizeof(buf);
 100         event_fd = bpf_link__fd(links[link_idx]);
 101         err = bpf_task_fd_query(getpid(), event_fd, 0, buf, &len,
 102                                 &prog_id, &fd_type, &probe_offset,
 103                                 &probe_addr);
 104         if (err < 0) {
 105                 printf("FAIL: %s, for event_fd idx %d, fn_name %s\n",
 106                        __func__, link_idx, fn_name);
 107                 perror("    :");
 108                 return -1;
 109         }
 110         if (strcmp(buf, fn_name) != 0 ||
 111             fd_type != expected_fd_type ||
 112             probe_offset != 0x0 || probe_addr != 0x0) {
 113                 printf("FAIL: bpf_trace_event_query(event_fd[%d]):\n",
 114                        link_idx);
 115                 printf("buf: %s, fd_type: %u, probe_offset: 0x%llx,"
 116                        " probe_addr: 0x%llx\n",
 117                        buf, fd_type, probe_offset, probe_addr);
 118                 return -1;
 119         }
 120         return 0;
 121 }
 122
 123 static int test_nondebug_fs_kuprobe_common(const char *event_type,
 124         const char *name, __u64 offset, __u64 addr, bool is_return,
 125         char *buf, __u32 *buf_len, __u32 *prog_id, __u32 *fd_type,
 126         __u64 *probe_offset, __u64 *probe_addr)
 127 {
 128         int is_return_bit = bpf_get_retprobe_bit(event_type);
 129         int type = bpf_find_probe_type(event_type);
 130         struct perf_event_attr attr = {};
 131         struct bpf_link *link;
 132         int fd, err = -1;
 133
 134         if (type < 0 || is_return_bit < 0) {
 135                 printf("FAIL: %s incorrect type (%d) or is_return_bit (%d)\n",
 136                         __func__, type, is_return_bit);
 137                 return err;
 138         }
 139
 140         attr.sample_period = 1;
 141         attr.wakeup_events = 1;
 142         if (is_return)
 143                 attr.config |= 1 << is_return_bit;
 144
 145         if (name) {
 146                 attr.config1 = ptr_to_u64((void *)name);
 147                 attr.config2 = offset;
 148         } else {
 149                 attr.config1 = 0;
 150                 attr.config2 = addr;
 151         }
 152         attr.size = sizeof(attr);
 153         attr.type = type;
 154
 155         fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
 156         link = bpf_program__attach_perf_event(progs[0], fd);
 157         if (libbpf_get_error(link)) {
 158                 printf("ERROR: bpf_program__attach_perf_event failed\n");
 159                 link = NULL;
 160                 close(fd);
 161                 goto cleanup;
 162         }
 163
 164         CHECK_PERROR_RET(bpf_task_fd_query(getpid(), fd, 0, buf, buf_len,
 165                          prog_id, fd_type, probe_offset, probe_addr) < 0);
 166         err = 0;
 167
 168 cleanup:
 169         bpf_link__destroy(link);
 170         return err;
 171 }
 172
 173 static int test_nondebug_fs_probe(const char *event_type, const char *name,
 174                                   __u64 offset, __u64 addr, bool is_return,
 175                                   __u32 expected_fd_type,
 176                                   __u32 expected_ret_fd_type,
 177                                   char *buf, __u32 buf_len)
 178 {
 179         __u64 probe_offset, probe_addr;
 180         __u32 prog_id, fd_type;
 181         int err;
 182
 183         err = test_nondebug_fs_kuprobe_common(event_type, name,
 184                                               offset, addr, is_return,
 185                                               buf, &buf_len, &prog_id,
 186                                               &fd_type, &probe_offset,
 187                                               &probe_addr);
 188         if (err < 0) {
 189                 printf("FAIL: %s, "
 190                        "for name %s, offset 0x%llx, addr 0x%llx, is_return %d\n",
 191                        __func__, name ? name : "", offset, addr, is_return);
 192                 perror("    :");
 193                 return -1;
 194         }
 195         if ((is_return && fd_type != expected_ret_fd_type) ||
 196             (!is_return && fd_type != expected_fd_type)) {
 197                 printf("FAIL: %s, incorrect fd_type %u\n",
 198                        __func__, fd_type);
 199                 return -1;
 200         }
 201         if (name) {
 202                 if (strcmp(name, buf) != 0) {
 203                         printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
 204                         return -1;
 205                 }
 206                 if (probe_offset != offset) {
 207                         printf("FAIL: %s, incorrect probe_offset 0x%llx\n",
 208                                __func__, probe_offset);
 209                         return -1;
 210                 }
 211         } else {
 212                 if (buf_len != 0) {
 213                         printf("FAIL: %s, incorrect buf %p\n",
 214                                __func__, buf);
 215                         return -1;
 216                 }
 217
 218                 if (probe_addr != addr) {
 219                         printf("FAIL: %s, incorrect probe_addr 0x%llx\n",
 220                                __func__, probe_addr);
 221                         return -1;
 222                 }
 223         }
 224         return 0;
 225 }
 226
 227 static int test_debug_fs_uprobe(char *binary_path, long offset, bool is_return)
 228 {
 229         char buf[256], event_alias[sizeof("test_1234567890")];
 230         const char *event_type = "uprobe";
 231         struct perf_event_attr attr = {};
 232         __u64 probe_offset, probe_addr;
 233         __u32 len, prog_id, fd_type;
 234         int err = -1, res, kfd, efd;
 235         struct bpf_link *link;
 236         ssize_t bytes;
 237
 238         snprintf(buf, sizeof(buf), "/sys/kernel/tracing/%s_events",
 239                  event_type);
 240         kfd = open(buf, O_WRONLY | O_TRUNC, 0);
 241         CHECK_PERROR_RET(kfd < 0);
 242
 243         res = snprintf(event_alias, sizeof(event_alias), "test_%d", getpid());
 244         CHECK_PERROR_RET(res < 0 || res >= sizeof(event_alias));
 245
 246         res = snprintf(buf, sizeof(buf), "%c:%ss/%s %s:0x%lx",
 247                        is_return ? 'r' : 'p', event_type, event_alias,
 248                        binary_path, offset);
 249         CHECK_PERROR_RET(res < 0 || res >= sizeof(buf));
 250         CHECK_PERROR_RET(write(kfd, buf, strlen(buf)) < 0);
 251
 252         close(kfd);
 253         kfd = -1;
 254
 255         snprintf(buf, sizeof(buf), "/sys/kernel/tracing/events/%ss/%s/id",
 256                  event_type, event_alias);
 257         efd = open(buf, O_RDONLY, 0);
 258         CHECK_PERROR_RET(efd < 0);
 259
 260         bytes = read(efd, buf, sizeof(buf));
 261         CHECK_PERROR_RET(bytes <= 0 || bytes >= sizeof(buf));
 262         close(efd);
 263         buf[bytes] = '\0';
 264
 265         attr.config = strtol(buf, NULL, 0);
 266         attr.type = PERF_TYPE_TRACEPOINT;
 267         attr.sample_period = 1;
 268         attr.wakeup_events = 1;
 269
 270         kfd = sys_perf_event_open(&attr, -1, 0, -1, PERF_FLAG_FD_CLOEXEC);
 271         link = bpf_program__attach_perf_event(progs[0], kfd);
 272         if (libbpf_get_error(link)) {
 273                 printf("ERROR: bpf_program__attach_perf_event failed\n");
 274                 link = NULL;
 275                 close(kfd);
 276                 goto cleanup;
 277         }
 278
 279         len = sizeof(buf);
 280         err = bpf_task_fd_query(getpid(), kfd, 0, buf, &len,
 281                                 &prog_id, &fd_type, &probe_offset,
 282                                 &probe_addr);
 283         if (err < 0) {
 284                 printf("FAIL: %s, binary_path %s\n", __func__, binary_path);
 285                 perror("    :");
 286                 return -1;
 287         }
 288         if ((is_return && fd_type != BPF_FD_TYPE_URETPROBE) ||
 289             (!is_return && fd_type != BPF_FD_TYPE_UPROBE)) {
 290                 printf("FAIL: %s, incorrect fd_type %u\n", __func__,
 291                        fd_type);
 292                 return -1;
 293         }
 294         if (strcmp(binary_path, buf) != 0) {
 295                 printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
 296                 return -1;
 297         }
 298         if (probe_offset != offset) {
 299                 printf("FAIL: %s, incorrect probe_offset 0x%llx\n", __func__,
 300                        probe_offset);
 301                 return -1;
 302         }
 303         err = 0;
 304
 305 cleanup:
 306         bpf_link__destroy(link);
 307         return err;
 308 }
 309
 310 int main(int argc, char **argv)
 311 {
 312         extern char __executable_start;
 313         char filename[256], buf[256];
 314         __u64 uprobe_file_offset;
 315         struct bpf_program *prog;
 316         struct bpf_object *obj;
 317         int i = 0, err = -1;
 318
 319         if (load_kallsyms()) {
 320                 printf("failed to process /proc/kallsyms\n");
 321                 return err;
 322         }
 323
 324         snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 325         obj = bpf_object__open_file(filename, NULL);
 326         if (libbpf_get_error(obj)) {
 327                 fprintf(stderr, "ERROR: opening BPF object file failed\n");
 328                 return err;
 329         }
 330
 331         /* load BPF program */
 332         if (bpf_object__load(obj)) {
 333                 fprintf(stderr, "ERROR: loading BPF object file failed\n");
 334                 goto cleanup;
 335         }
 336
 337         bpf_object__for_each_program(prog, obj) {
 338                 progs[i] = prog;
 339                 links[i] = bpf_program__attach(progs[i]);
 340                 if (libbpf_get_error(links[i])) {
 341                         fprintf(stderr, "ERROR: bpf_program__attach failed\n");
 342                         links[i] = NULL;
 343                         goto cleanup;
 344                 }
 345                 i++;
 346         }
 347
 348         /* test two functions in the corresponding *_kern.c file */
 349         CHECK_AND_RET(test_debug_fs_kprobe(0, "blk_mq_start_request",
 350                                            BPF_FD_TYPE_KPROBE));
 351         CHECK_AND_RET(test_debug_fs_kprobe(1, "__blk_account_io_done",
 352                                            BPF_FD_TYPE_KRETPROBE));
 353
 354         /* test nondebug fs kprobe */
 355         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
 356                                              false, BPF_FD_TYPE_KPROBE,
 357                                              BPF_FD_TYPE_KRETPROBE,
 358                                              buf, sizeof(buf)));
 359 #ifdef __x86_64__
 360         /* set a kprobe on "bpf_check + 0x5", which is x64 specific */
 361         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x5, 0x0,
 362                                              false, BPF_FD_TYPE_KPROBE,
 363                                              BPF_FD_TYPE_KRETPROBE,
 364                                              buf, sizeof(buf)));
 365 #endif
 366         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
 367                                              true, BPF_FD_TYPE_KPROBE,
 368                                              BPF_FD_TYPE_KRETPROBE,
 369                                              buf, sizeof(buf)));
 370         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 371                                              ksym_get_addr("bpf_check"), false,
 372                                              BPF_FD_TYPE_KPROBE,
 373                                              BPF_FD_TYPE_KRETPROBE,
 374                                              buf, sizeof(buf)));
 375         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 376                                              ksym_get_addr("bpf_check"), false,
 377                                              BPF_FD_TYPE_KPROBE,
 378                                              BPF_FD_TYPE_KRETPROBE,
 379                                              NULL, 0));
 380         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 381                                              ksym_get_addr("bpf_check"), true,
 382                                              BPF_FD_TYPE_KPROBE,
 383                                              BPF_FD_TYPE_KRETPROBE,
 384                                              buf, sizeof(buf)));
 385         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 386                                              ksym_get_addr("bpf_check"), true,
 387                                              BPF_FD_TYPE_KPROBE,
 388                                              BPF_FD_TYPE_KRETPROBE,
 389                                              0, 0));
 390
 391         /* test nondebug fs uprobe */
 392         /* the calculation of uprobe file offset is based on gcc 7.3.1 on x64
 393          * and the default linker script, which defines __executable_start as
 394          * the start of the .text section. The calculation could be different
 395          * on different systems with different compilers. The right way is
 396          * to parse the ELF file. We took a shortcut here.
 397          */
 398         uprobe_file_offset = (unsigned long)main - (unsigned long)&__executable_start;
 399         CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
 400                                              uprobe_file_offset, 0x0, false,
 401                                              BPF_FD_TYPE_UPROBE,
 402                                              BPF_FD_TYPE_URETPROBE,
 403                                              buf, sizeof(buf)));
 404         CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
 405                                              uprobe_file_offset, 0x0, true,
 406                                              BPF_FD_TYPE_UPROBE,
 407                                              BPF_FD_TYPE_URETPROBE,
 408                                              buf, sizeof(buf)));
 409
 410         /* test debug fs uprobe */
 411         CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
 412                                            false));
 413         CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
 414                                            true));
 415         err = 0;
 416
 417 cleanup:
 418         for (i--; i >= 0; i--)
 419                 bpf_link__destroy(links[i]);
 420
 421         bpf_object__close(obj);
 422         return err;
 423 }