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/bpf.h>
  19 #include <bpf/libbpf.h>
  20 #include "bpf_util.h"
  21 #include "perf-sys.h"
  22 #include "trace_helpers.h"
  23
  24 static struct bpf_program *progs[2];
  25 static struct bpf_link *links[2];
  26
  27 #define CHECK_PERROR_RET(condition) ({                  \
  28         int __ret = !!(condition);                      \
  29         if (__ret) {                                    \
  30                 printf("FAIL: %s:\n", __func__);        \
  31                 perror("    ");                 \
  32                 return -1;                              \
  33         }                                               \
  34 })
  35
  36 #define CHECK_AND_RET(condition) ({                     \
  37         int __ret = !!(condition);                      \
  38         if (__ret)                                      \
  39                 return -1;                              \
  40 })
  41
  42 static __u64 ptr_to_u64(void *ptr)
  43 {
  44         return (__u64) (unsigned long) ptr;
  45 }
  46
  47 #define PMU_TYPE_FILE "/sys/bus/event_source/devices/%s/type"
  48 static int bpf_find_probe_type(const char *event_type)
  49 {
  50         char buf[256];
  51         int fd, ret;
  52
  53         ret = snprintf(buf, sizeof(buf), PMU_TYPE_FILE, event_type);
  54         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  55
  56         fd = open(buf, O_RDONLY);
  57         CHECK_PERROR_RET(fd < 0);
  58
  59         ret = read(fd, buf, sizeof(buf));
  60         close(fd);
  61         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  62
  63         errno = 0;
  64         ret = (int)strtol(buf, NULL, 10);
  65         CHECK_PERROR_RET(errno);
  66         return ret;
  67 }
  68
  69 #define PMU_RETPROBE_FILE "/sys/bus/event_source/devices/%s/format/retprobe"
  70 static int bpf_get_retprobe_bit(const char *event_type)
  71 {
  72         char buf[256];
  73         int fd, ret;
  74
  75         ret = snprintf(buf, sizeof(buf), PMU_RETPROBE_FILE, event_type);
  76         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  77
  78         fd = open(buf, O_RDONLY);
  79         CHECK_PERROR_RET(fd < 0);
  80
  81         ret = read(fd, buf, sizeof(buf));
  82         close(fd);
  83         CHECK_PERROR_RET(ret < 0 || ret >= sizeof(buf));
  84         CHECK_PERROR_RET(strlen(buf) < strlen("config:"));
  85
  86         errno = 0;
  87         ret = (int)strtol(buf + strlen("config:"), NULL, 10);
  88         CHECK_PERROR_RET(errno);
  89         return ret;
  90 }
  91
  92 static int test_debug_fs_kprobe(int link_idx, const char *fn_name,
  93                                 __u32 expected_fd_type)
  94 {
  95         __u64 probe_offset, probe_addr;
  96         __u32 len, prog_id, fd_type;
  97         int err, event_fd;
  98         char buf[256];
  99
 100         len = sizeof(buf);
 101         event_fd = bpf_link__fd(links[link_idx]);
 102         err = bpf_task_fd_query(getpid(), event_fd, 0, buf, &len,
 103                                 &prog_id, &fd_type, &probe_offset,
 104                                 &probe_addr);
 105         if (err < 0) {
 106                 printf("FAIL: %s, for event_fd idx %d, fn_name %s\n",
 107                        __func__, link_idx, fn_name);
 108                 perror("    :");
 109                 return -1;
 110         }
 111         if (strcmp(buf, fn_name) != 0 ||
 112             fd_type != expected_fd_type ||
 113             probe_offset != 0x0 || probe_addr != 0x0) {
 114                 printf("FAIL: bpf_trace_event_query(event_fd[%d]):\n",
 115                        link_idx);
 116                 printf("buf: %s, fd_type: %u, probe_offset: 0x%llx,"
 117                        " probe_addr: 0x%llx\n",
 118                        buf, fd_type, probe_offset, probe_addr);
 119                 return -1;
 120         }
 121         return 0;
 122 }
 123
 124 static int test_nondebug_fs_kuprobe_common(const char *event_type,
 125         const char *name, __u64 offset, __u64 addr, bool is_return,
 126         char *buf, __u32 *buf_len, __u32 *prog_id, __u32 *fd_type,
 127         __u64 *probe_offset, __u64 *probe_addr)
 128 {
 129         int is_return_bit = bpf_get_retprobe_bit(event_type);
 130         int type = bpf_find_probe_type(event_type);
 131         struct perf_event_attr attr = {};
 132         struct bpf_link *link;
 133         int fd, err = -1;
 134
 135         if (type < 0 || is_return_bit < 0) {
 136                 printf("FAIL: %s incorrect type (%d) or is_return_bit (%d)\n",
 137                         __func__, type, is_return_bit);
 138                 return err;
 139         }
 140
 141         attr.sample_period = 1;
 142         attr.wakeup_events = 1;
 143         if (is_return)
 144                 attr.config |= 1 << is_return_bit;
 145
 146         if (name) {
 147                 attr.config1 = ptr_to_u64((void *)name);
 148                 attr.config2 = offset;
 149         } else {
 150                 attr.config1 = 0;
 151                 attr.config2 = addr;
 152         }
 153         attr.size = sizeof(attr);
 154         attr.type = type;
 155
 156         fd = sys_perf_event_open(&attr, -1, 0, -1, 0);
 157         link = bpf_program__attach_perf_event(progs[0], fd);
 158         if (libbpf_get_error(link)) {
 159                 printf("ERROR: bpf_program__attach_perf_event failed\n");
 160                 link = NULL;
 161                 close(fd);
 162                 goto cleanup;
 163         }
 164
 165         CHECK_PERROR_RET(bpf_task_fd_query(getpid(), fd, 0, buf, buf_len,
 166                          prog_id, fd_type, probe_offset, probe_addr) < 0);
 167         err = 0;
 168
 169 cleanup:
 170         bpf_link__destroy(link);
 171         return err;
 172 }
 173
 174 static int test_nondebug_fs_probe(const char *event_type, const char *name,
 175                                   __u64 offset, __u64 addr, bool is_return,
 176                                   __u32 expected_fd_type,
 177                                   __u32 expected_ret_fd_type,
 178                                   char *buf, __u32 buf_len)
 179 {
 180         __u64 probe_offset, probe_addr;
 181         __u32 prog_id, fd_type;
 182         int err;
 183
 184         err = test_nondebug_fs_kuprobe_common(event_type, name,
 185                                               offset, addr, is_return,
 186                                               buf, &buf_len, &prog_id,
 187                                               &fd_type, &probe_offset,
 188                                               &probe_addr);
 189         if (err < 0) {
 190                 printf("FAIL: %s, "
 191                        "for name %s, offset 0x%llx, addr 0x%llx, is_return %d\n",
 192                        __func__, name ? name : "", offset, addr, is_return);
 193                 perror("    :");
 194                 return -1;
 195         }
 196         if ((is_return && fd_type != expected_ret_fd_type) ||
 197             (!is_return && fd_type != expected_fd_type)) {
 198                 printf("FAIL: %s, incorrect fd_type %u\n",
 199                        __func__, fd_type);
 200                 return -1;
 201         }
 202         if (name) {
 203                 if (strcmp(name, buf) != 0) {
 204                         printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
 205                         return -1;
 206                 }
 207                 if (probe_offset != offset) {
 208                         printf("FAIL: %s, incorrect probe_offset 0x%llx\n",
 209                                __func__, probe_offset);
 210                         return -1;
 211                 }
 212         } else {
 213                 if (buf_len != 0) {
 214                         printf("FAIL: %s, incorrect buf %p\n",
 215                                __func__, buf);
 216                         return -1;
 217                 }
 218
 219                 if (probe_addr != addr) {
 220                         printf("FAIL: %s, incorrect probe_addr 0x%llx\n",
 221                                __func__, probe_addr);
 222                         return -1;
 223                 }
 224         }
 225         return 0;
 226 }
 227
 228 static int test_debug_fs_uprobe(char *binary_path, long offset, bool is_return)
 229 {
 230         char buf[256], event_alias[sizeof("test_1234567890")];
 231         const char *event_type = "uprobe";
 232         struct perf_event_attr attr = {};
 233         __u64 probe_offset, probe_addr;
 234         __u32 len, prog_id, fd_type;
 235         int err = -1, res, kfd, efd;
 236         struct bpf_link *link;
 237         ssize_t bytes;
 238
 239         snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/%s_events",
 240                  event_type);
 241         kfd = open(buf, O_WRONLY | O_TRUNC, 0);
 242         CHECK_PERROR_RET(kfd < 0);
 243
 244         res = snprintf(event_alias, sizeof(event_alias), "test_%d", getpid());
 245         CHECK_PERROR_RET(res < 0 || res >= sizeof(event_alias));
 246
 247         res = snprintf(buf, sizeof(buf), "%c:%ss/%s %s:0x%lx",
 248                        is_return ? 'r' : 'p', event_type, event_alias,
 249                        binary_path, offset);
 250         CHECK_PERROR_RET(res < 0 || res >= sizeof(buf));
 251         CHECK_PERROR_RET(write(kfd, buf, strlen(buf)) < 0);
 252
 253         close(kfd);
 254         kfd = -1;
 255
 256         snprintf(buf, sizeof(buf), "/sys/kernel/debug/tracing/events/%ss/%s/id",
 257                  event_type, event_alias);
 258         efd = open(buf, O_RDONLY, 0);
 259         CHECK_PERROR_RET(efd < 0);
 260
 261         bytes = read(efd, buf, sizeof(buf));
 262         CHECK_PERROR_RET(bytes <= 0 || bytes >= sizeof(buf));
 263         close(efd);
 264         buf[bytes] = '\0';
 265
 266         attr.config = strtol(buf, NULL, 0);
 267         attr.type = PERF_TYPE_TRACEPOINT;
 268         attr.sample_period = 1;
 269         attr.wakeup_events = 1;
 270
 271         kfd = sys_perf_event_open(&attr, -1, 0, -1, PERF_FLAG_FD_CLOEXEC);
 272         link = bpf_program__attach_perf_event(progs[0], kfd);
 273         if (libbpf_get_error(link)) {
 274                 printf("ERROR: bpf_program__attach_perf_event failed\n");
 275                 link = NULL;
 276                 close(kfd);
 277                 goto cleanup;
 278         }
 279
 280         len = sizeof(buf);
 281         err = bpf_task_fd_query(getpid(), kfd, 0, buf, &len,
 282                                 &prog_id, &fd_type, &probe_offset,
 283                                 &probe_addr);
 284         if (err < 0) {
 285                 printf("FAIL: %s, binary_path %s\n", __func__, binary_path);
 286                 perror("    :");
 287                 return -1;
 288         }
 289         if ((is_return && fd_type != BPF_FD_TYPE_URETPROBE) ||
 290             (!is_return && fd_type != BPF_FD_TYPE_UPROBE)) {
 291                 printf("FAIL: %s, incorrect fd_type %u\n", __func__,
 292                        fd_type);
 293                 return -1;
 294         }
 295         if (strcmp(binary_path, buf) != 0) {
 296                 printf("FAIL: %s, incorrect buf %s\n", __func__, buf);
 297                 return -1;
 298         }
 299         if (probe_offset != offset) {
 300                 printf("FAIL: %s, incorrect probe_offset 0x%llx\n", __func__,
 301                        probe_offset);
 302                 return -1;
 303         }
 304         err = 0;
 305
 306 cleanup:
 307         bpf_link__destroy(link);
 308         return err;
 309 }
 310
 311 int main(int argc, char **argv)
 312 {
 313         extern char __executable_start;
 314         char filename[256], buf[256];
 315         __u64 uprobe_file_offset;
 316         struct bpf_program *prog;
 317         struct bpf_object *obj;
 318         int i = 0, err = -1;
 319
 320         if (load_kallsyms()) {
 321                 printf("failed to process /proc/kallsyms\n");
 322                 return err;
 323         }
 324
 325         snprintf(filename, sizeof(filename), "%s_kern.o", argv[0]);
 326         obj = bpf_object__open_file(filename, NULL);
 327         if (libbpf_get_error(obj)) {
 328                 fprintf(stderr, "ERROR: opening BPF object file failed\n");
 329                 return err;
 330         }
 331
 332         /* load BPF program */
 333         if (bpf_object__load(obj)) {
 334                 fprintf(stderr, "ERROR: loading BPF object file failed\n");
 335                 goto cleanup;
 336         }
 337
 338         bpf_object__for_each_program(prog, obj) {
 339                 progs[i] = prog;
 340                 links[i] = bpf_program__attach(progs[i]);
 341                 if (libbpf_get_error(links[i])) {
 342                         fprintf(stderr, "ERROR: bpf_program__attach failed\n");
 343                         links[i] = NULL;
 344                         goto cleanup;
 345                 }
 346                 i++;
 347         }
 348
 349         /* test two functions in the corresponding *_kern.c file */
 350         CHECK_AND_RET(test_debug_fs_kprobe(0, "blk_mq_start_request",
 351                                            BPF_FD_TYPE_KPROBE));
 352         CHECK_AND_RET(test_debug_fs_kprobe(1, "blk_account_io_done",
 353                                            BPF_FD_TYPE_KRETPROBE));
 354
 355         /* test nondebug fs kprobe */
 356         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
 357                                              false, BPF_FD_TYPE_KPROBE,
 358                                              BPF_FD_TYPE_KRETPROBE,
 359                                              buf, sizeof(buf)));
 360 #ifdef __x86_64__
 361         /* set a kprobe on "bpf_check + 0x5", which is x64 specific */
 362         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x5, 0x0,
 363                                              false, BPF_FD_TYPE_KPROBE,
 364                                              BPF_FD_TYPE_KRETPROBE,
 365                                              buf, sizeof(buf)));
 366 #endif
 367         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", "bpf_check", 0x0, 0x0,
 368                                              true, BPF_FD_TYPE_KPROBE,
 369                                              BPF_FD_TYPE_KRETPROBE,
 370                                              buf, sizeof(buf)));
 371         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 372                                              ksym_get_addr("bpf_check"), false,
 373                                              BPF_FD_TYPE_KPROBE,
 374                                              BPF_FD_TYPE_KRETPROBE,
 375                                              buf, sizeof(buf)));
 376         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 377                                              ksym_get_addr("bpf_check"), false,
 378                                              BPF_FD_TYPE_KPROBE,
 379                                              BPF_FD_TYPE_KRETPROBE,
 380                                              NULL, 0));
 381         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 382                                              ksym_get_addr("bpf_check"), true,
 383                                              BPF_FD_TYPE_KPROBE,
 384                                              BPF_FD_TYPE_KRETPROBE,
 385                                              buf, sizeof(buf)));
 386         CHECK_AND_RET(test_nondebug_fs_probe("kprobe", NULL, 0x0,
 387                                              ksym_get_addr("bpf_check"), true,
 388                                              BPF_FD_TYPE_KPROBE,
 389                                              BPF_FD_TYPE_KRETPROBE,
 390                                              0, 0));
 391
 392         /* test nondebug fs uprobe */
 393         /* the calculation of uprobe file offset is based on gcc 7.3.1 on x64
 394          * and the default linker script, which defines __executable_start as
 395          * the start of the .text section. The calculation could be different
 396          * on different systems with different compilers. The right way is
 397          * to parse the ELF file. We took a shortcut here.
 398          */
 399         uprobe_file_offset = (__u64)main - (__u64)&__executable_start;
 400         CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
 401                                              uprobe_file_offset, 0x0, false,
 402                                              BPF_FD_TYPE_UPROBE,
 403                                              BPF_FD_TYPE_URETPROBE,
 404                                              buf, sizeof(buf)));
 405         CHECK_AND_RET(test_nondebug_fs_probe("uprobe", (char *)argv[0],
 406                                              uprobe_file_offset, 0x0, true,
 407                                              BPF_FD_TYPE_UPROBE,
 408                                              BPF_FD_TYPE_URETPROBE,
 409                                              buf, sizeof(buf)));
 410
 411         /* test debug fs uprobe */
 412         CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
 413                                            false));
 414         CHECK_AND_RET(test_debug_fs_uprobe((char *)argv[0], uprobe_file_offset,
 415                                            true));
 416         err = 0;
 417
 418 cleanup:
 419         for (i--; i >= 0; i--)
 420                 bpf_link__destroy(links[i]);
 421
 422         bpf_object__close(obj);
 423         return err;
 424 }