| 1 | /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com |
| 2 | * |
| 3 | * This program is free software; you can redistribute it and/or |
| 4 | * modify it under the terms of version 2 of the GNU General Public |
| 5 | * License as published by the Free Software Foundation. |
| 6 | */ |
| 7 | #include <linux/kernel.h> |
| 8 | #include <linux/types.h> |
| 9 | #include <linux/slab.h> |
| 10 | #include <linux/bpf.h> |
| 11 | #include <linux/filter.h> |
| 12 | #include <linux/uaccess.h> |
| 13 | #include <linux/ctype.h> |
| 14 | #include "trace.h" |
| 15 | |
| 16 | /** |
| 17 | * trace_call_bpf - invoke BPF program |
| 18 | * @prog: BPF program |
| 19 | * @ctx: opaque context pointer |
| 20 | * |
| 21 | * kprobe handlers execute BPF programs via this helper. |
| 22 | * Can be used from static tracepoints in the future. |
| 23 | * |
| 24 | * Return: BPF programs always return an integer which is interpreted by |
| 25 | * kprobe handler as: |
| 26 | * 0 - return from kprobe (event is filtered out) |
| 27 | * 1 - store kprobe event into ring buffer |
| 28 | * Other values are reserved and currently alias to 1 |
| 29 | */ |
| 30 | unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx) |
| 31 | { |
| 32 | unsigned int ret; |
| 33 | |
| 34 | if (in_nmi()) /* not supported yet */ |
| 35 | return 1; |
| 36 | |
| 37 | preempt_disable(); |
| 38 | |
| 39 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { |
| 40 | /* |
| 41 | * since some bpf program is already running on this cpu, |
| 42 | * don't call into another bpf program (same or different) |
| 43 | * and don't send kprobe event into ring-buffer, |
| 44 | * so return zero here |
| 45 | */ |
| 46 | ret = 0; |
| 47 | goto out; |
| 48 | } |
| 49 | |
| 50 | rcu_read_lock(); |
| 51 | ret = BPF_PROG_RUN(prog, ctx); |
| 52 | rcu_read_unlock(); |
| 53 | |
| 54 | out: |
| 55 | __this_cpu_dec(bpf_prog_active); |
| 56 | preempt_enable(); |
| 57 | |
| 58 | return ret; |
| 59 | } |
| 60 | EXPORT_SYMBOL_GPL(trace_call_bpf); |
| 61 | |
| 62 | static u64 bpf_probe_read(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
| 63 | { |
| 64 | void *dst = (void *) (long) r1; |
| 65 | int ret, size = (int) r2; |
| 66 | void *unsafe_ptr = (void *) (long) r3; |
| 67 | |
| 68 | ret = probe_kernel_read(dst, unsafe_ptr, size); |
| 69 | if (unlikely(ret < 0)) |
| 70 | memset(dst, 0, size); |
| 71 | |
| 72 | return ret; |
| 73 | } |
| 74 | |
| 75 | static const struct bpf_func_proto bpf_probe_read_proto = { |
| 76 | .func = bpf_probe_read, |
| 77 | .gpl_only = true, |
| 78 | .ret_type = RET_INTEGER, |
| 79 | .arg1_type = ARG_PTR_TO_RAW_STACK, |
| 80 | .arg2_type = ARG_CONST_STACK_SIZE, |
| 81 | .arg3_type = ARG_ANYTHING, |
| 82 | }; |
| 83 | |
| 84 | /* |
| 85 | * limited trace_printk() |
| 86 | * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed |
| 87 | */ |
| 88 | static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5) |
| 89 | { |
| 90 | char *fmt = (char *) (long) r1; |
| 91 | bool str_seen = false; |
| 92 | int mod[3] = {}; |
| 93 | int fmt_cnt = 0; |
| 94 | u64 unsafe_addr; |
| 95 | char buf[64]; |
| 96 | int i; |
| 97 | |
| 98 | /* |
| 99 | * bpf_check()->check_func_arg()->check_stack_boundary() |
| 100 | * guarantees that fmt points to bpf program stack, |
| 101 | * fmt_size bytes of it were initialized and fmt_size > 0 |
| 102 | */ |
| 103 | if (fmt[--fmt_size] != 0) |
| 104 | return -EINVAL; |
| 105 | |
| 106 | /* check format string for allowed specifiers */ |
| 107 | for (i = 0; i < fmt_size; i++) { |
| 108 | if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i])) |
| 109 | return -EINVAL; |
| 110 | |
| 111 | if (fmt[i] != '%') |
| 112 | continue; |
| 113 | |
| 114 | if (fmt_cnt >= 3) |
| 115 | return -EINVAL; |
| 116 | |
| 117 | /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */ |
| 118 | i++; |
| 119 | if (fmt[i] == 'l') { |
| 120 | mod[fmt_cnt]++; |
| 121 | i++; |
| 122 | } else if (fmt[i] == 'p' || fmt[i] == 's') { |
| 123 | mod[fmt_cnt]++; |
| 124 | i++; |
| 125 | if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0) |
| 126 | return -EINVAL; |
| 127 | fmt_cnt++; |
| 128 | if (fmt[i - 1] == 's') { |
| 129 | if (str_seen) |
| 130 | /* allow only one '%s' per fmt string */ |
| 131 | return -EINVAL; |
| 132 | str_seen = true; |
| 133 | |
| 134 | switch (fmt_cnt) { |
| 135 | case 1: |
| 136 | unsafe_addr = r3; |
| 137 | r3 = (long) buf; |
| 138 | break; |
| 139 | case 2: |
| 140 | unsafe_addr = r4; |
| 141 | r4 = (long) buf; |
| 142 | break; |
| 143 | case 3: |
| 144 | unsafe_addr = r5; |
| 145 | r5 = (long) buf; |
| 146 | break; |
| 147 | } |
| 148 | buf[0] = 0; |
| 149 | strncpy_from_unsafe(buf, |
| 150 | (void *) (long) unsafe_addr, |
| 151 | sizeof(buf)); |
| 152 | } |
| 153 | continue; |
| 154 | } |
| 155 | |
| 156 | if (fmt[i] == 'l') { |
| 157 | mod[fmt_cnt]++; |
| 158 | i++; |
| 159 | } |
| 160 | |
| 161 | if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x') |
| 162 | return -EINVAL; |
| 163 | fmt_cnt++; |
| 164 | } |
| 165 | |
| 166 | return __trace_printk(1/* fake ip will not be printed */, fmt, |
| 167 | mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3, |
| 168 | mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4, |
| 169 | mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5); |
| 170 | } |
| 171 | |
| 172 | static const struct bpf_func_proto bpf_trace_printk_proto = { |
| 173 | .func = bpf_trace_printk, |
| 174 | .gpl_only = true, |
| 175 | .ret_type = RET_INTEGER, |
| 176 | .arg1_type = ARG_PTR_TO_STACK, |
| 177 | .arg2_type = ARG_CONST_STACK_SIZE, |
| 178 | }; |
| 179 | |
| 180 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
| 181 | { |
| 182 | /* |
| 183 | * this program might be calling bpf_trace_printk, |
| 184 | * so allocate per-cpu printk buffers |
| 185 | */ |
| 186 | trace_printk_init_buffers(); |
| 187 | |
| 188 | return &bpf_trace_printk_proto; |
| 189 | } |
| 190 | |
| 191 | static u64 bpf_perf_event_read(u64 r1, u64 index, u64 r3, u64 r4, u64 r5) |
| 192 | { |
| 193 | struct bpf_map *map = (struct bpf_map *) (unsigned long) r1; |
| 194 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
| 195 | struct perf_event *event; |
| 196 | struct file *file; |
| 197 | |
| 198 | if (unlikely(index >= array->map.max_entries)) |
| 199 | return -E2BIG; |
| 200 | |
| 201 | file = (struct file *)array->ptrs[index]; |
| 202 | if (unlikely(!file)) |
| 203 | return -ENOENT; |
| 204 | |
| 205 | event = file->private_data; |
| 206 | |
| 207 | /* make sure event is local and doesn't have pmu::count */ |
| 208 | if (event->oncpu != smp_processor_id() || |
| 209 | event->pmu->count) |
| 210 | return -EINVAL; |
| 211 | |
| 212 | /* |
| 213 | * we don't know if the function is run successfully by the |
| 214 | * return value. It can be judged in other places, such as |
| 215 | * eBPF programs. |
| 216 | */ |
| 217 | return perf_event_read_local(event); |
| 218 | } |
| 219 | |
| 220 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
| 221 | .func = bpf_perf_event_read, |
| 222 | .gpl_only = true, |
| 223 | .ret_type = RET_INTEGER, |
| 224 | .arg1_type = ARG_CONST_MAP_PTR, |
| 225 | .arg2_type = ARG_ANYTHING, |
| 226 | }; |
| 227 | |
| 228 | static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size) |
| 229 | { |
| 230 | struct pt_regs *regs = (struct pt_regs *) (long) r1; |
| 231 | struct bpf_map *map = (struct bpf_map *) (long) r2; |
| 232 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
| 233 | u64 index = flags & BPF_F_INDEX_MASK; |
| 234 | void *data = (void *) (long) r4; |
| 235 | struct perf_sample_data sample_data; |
| 236 | struct perf_event *event; |
| 237 | struct file *file; |
| 238 | struct perf_raw_record raw = { |
| 239 | .size = size, |
| 240 | .data = data, |
| 241 | }; |
| 242 | |
| 243 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
| 244 | return -EINVAL; |
| 245 | if (index == BPF_F_CURRENT_CPU) |
| 246 | index = raw_smp_processor_id(); |
| 247 | if (unlikely(index >= array->map.max_entries)) |
| 248 | return -E2BIG; |
| 249 | |
| 250 | file = (struct file *)array->ptrs[index]; |
| 251 | if (unlikely(!file)) |
| 252 | return -ENOENT; |
| 253 | |
| 254 | event = file->private_data; |
| 255 | |
| 256 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
| 257 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) |
| 258 | return -EINVAL; |
| 259 | |
| 260 | if (unlikely(event->oncpu != smp_processor_id())) |
| 261 | return -EOPNOTSUPP; |
| 262 | |
| 263 | perf_sample_data_init(&sample_data, 0, 0); |
| 264 | sample_data.raw = &raw; |
| 265 | perf_event_output(event, &sample_data, regs); |
| 266 | return 0; |
| 267 | } |
| 268 | |
| 269 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
| 270 | .func = bpf_perf_event_output, |
| 271 | .gpl_only = true, |
| 272 | .ret_type = RET_INTEGER, |
| 273 | .arg1_type = ARG_PTR_TO_CTX, |
| 274 | .arg2_type = ARG_CONST_MAP_PTR, |
| 275 | .arg3_type = ARG_ANYTHING, |
| 276 | .arg4_type = ARG_PTR_TO_STACK, |
| 277 | .arg5_type = ARG_CONST_STACK_SIZE, |
| 278 | }; |
| 279 | |
| 280 | static DEFINE_PER_CPU(struct pt_regs, bpf_pt_regs); |
| 281 | |
| 282 | static u64 bpf_event_output(u64 r1, u64 r2, u64 flags, u64 r4, u64 size) |
| 283 | { |
| 284 | struct pt_regs *regs = this_cpu_ptr(&bpf_pt_regs); |
| 285 | |
| 286 | perf_fetch_caller_regs(regs); |
| 287 | |
| 288 | return bpf_perf_event_output((long)regs, r2, flags, r4, size); |
| 289 | } |
| 290 | |
| 291 | static const struct bpf_func_proto bpf_event_output_proto = { |
| 292 | .func = bpf_event_output, |
| 293 | .gpl_only = true, |
| 294 | .ret_type = RET_INTEGER, |
| 295 | .arg1_type = ARG_PTR_TO_CTX, |
| 296 | .arg2_type = ARG_CONST_MAP_PTR, |
| 297 | .arg3_type = ARG_ANYTHING, |
| 298 | .arg4_type = ARG_PTR_TO_STACK, |
| 299 | .arg5_type = ARG_CONST_STACK_SIZE, |
| 300 | }; |
| 301 | |
| 302 | const struct bpf_func_proto *bpf_get_event_output_proto(void) |
| 303 | { |
| 304 | return &bpf_event_output_proto; |
| 305 | } |
| 306 | |
| 307 | static const struct bpf_func_proto *tracing_func_proto(enum bpf_func_id func_id) |
| 308 | { |
| 309 | switch (func_id) { |
| 310 | case BPF_FUNC_map_lookup_elem: |
| 311 | return &bpf_map_lookup_elem_proto; |
| 312 | case BPF_FUNC_map_update_elem: |
| 313 | return &bpf_map_update_elem_proto; |
| 314 | case BPF_FUNC_map_delete_elem: |
| 315 | return &bpf_map_delete_elem_proto; |
| 316 | case BPF_FUNC_probe_read: |
| 317 | return &bpf_probe_read_proto; |
| 318 | case BPF_FUNC_ktime_get_ns: |
| 319 | return &bpf_ktime_get_ns_proto; |
| 320 | case BPF_FUNC_tail_call: |
| 321 | return &bpf_tail_call_proto; |
| 322 | case BPF_FUNC_get_current_pid_tgid: |
| 323 | return &bpf_get_current_pid_tgid_proto; |
| 324 | case BPF_FUNC_get_current_uid_gid: |
| 325 | return &bpf_get_current_uid_gid_proto; |
| 326 | case BPF_FUNC_get_current_comm: |
| 327 | return &bpf_get_current_comm_proto; |
| 328 | case BPF_FUNC_trace_printk: |
| 329 | return bpf_get_trace_printk_proto(); |
| 330 | case BPF_FUNC_get_smp_processor_id: |
| 331 | return &bpf_get_smp_processor_id_proto; |
| 332 | case BPF_FUNC_perf_event_read: |
| 333 | return &bpf_perf_event_read_proto; |
| 334 | default: |
| 335 | return NULL; |
| 336 | } |
| 337 | } |
| 338 | |
| 339 | static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id) |
| 340 | { |
| 341 | switch (func_id) { |
| 342 | case BPF_FUNC_perf_event_output: |
| 343 | return &bpf_perf_event_output_proto; |
| 344 | case BPF_FUNC_get_stackid: |
| 345 | return &bpf_get_stackid_proto; |
| 346 | default: |
| 347 | return tracing_func_proto(func_id); |
| 348 | } |
| 349 | } |
| 350 | |
| 351 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ |
| 352 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type) |
| 353 | { |
| 354 | /* check bounds */ |
| 355 | if (off < 0 || off >= sizeof(struct pt_regs)) |
| 356 | return false; |
| 357 | |
| 358 | /* only read is allowed */ |
| 359 | if (type != BPF_READ) |
| 360 | return false; |
| 361 | |
| 362 | /* disallow misaligned access */ |
| 363 | if (off % size != 0) |
| 364 | return false; |
| 365 | |
| 366 | return true; |
| 367 | } |
| 368 | |
| 369 | static const struct bpf_verifier_ops kprobe_prog_ops = { |
| 370 | .get_func_proto = kprobe_prog_func_proto, |
| 371 | .is_valid_access = kprobe_prog_is_valid_access, |
| 372 | }; |
| 373 | |
| 374 | static struct bpf_prog_type_list kprobe_tl = { |
| 375 | .ops = &kprobe_prog_ops, |
| 376 | .type = BPF_PROG_TYPE_KPROBE, |
| 377 | }; |
| 378 | |
| 379 | static u64 bpf_perf_event_output_tp(u64 r1, u64 r2, u64 index, u64 r4, u64 size) |
| 380 | { |
| 381 | /* |
| 382 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden |
| 383 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it |
| 384 | * from there and call the same bpf_perf_event_output() helper |
| 385 | */ |
| 386 | u64 ctx = *(long *)(uintptr_t)r1; |
| 387 | |
| 388 | return bpf_perf_event_output(ctx, r2, index, r4, size); |
| 389 | } |
| 390 | |
| 391 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { |
| 392 | .func = bpf_perf_event_output_tp, |
| 393 | .gpl_only = true, |
| 394 | .ret_type = RET_INTEGER, |
| 395 | .arg1_type = ARG_PTR_TO_CTX, |
| 396 | .arg2_type = ARG_CONST_MAP_PTR, |
| 397 | .arg3_type = ARG_ANYTHING, |
| 398 | .arg4_type = ARG_PTR_TO_STACK, |
| 399 | .arg5_type = ARG_CONST_STACK_SIZE, |
| 400 | }; |
| 401 | |
| 402 | static u64 bpf_get_stackid_tp(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5) |
| 403 | { |
| 404 | u64 ctx = *(long *)(uintptr_t)r1; |
| 405 | |
| 406 | return bpf_get_stackid(ctx, r2, r3, r4, r5); |
| 407 | } |
| 408 | |
| 409 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { |
| 410 | .func = bpf_get_stackid_tp, |
| 411 | .gpl_only = true, |
| 412 | .ret_type = RET_INTEGER, |
| 413 | .arg1_type = ARG_PTR_TO_CTX, |
| 414 | .arg2_type = ARG_CONST_MAP_PTR, |
| 415 | .arg3_type = ARG_ANYTHING, |
| 416 | }; |
| 417 | |
| 418 | static const struct bpf_func_proto *tp_prog_func_proto(enum bpf_func_id func_id) |
| 419 | { |
| 420 | switch (func_id) { |
| 421 | case BPF_FUNC_perf_event_output: |
| 422 | return &bpf_perf_event_output_proto_tp; |
| 423 | case BPF_FUNC_get_stackid: |
| 424 | return &bpf_get_stackid_proto_tp; |
| 425 | default: |
| 426 | return tracing_func_proto(func_id); |
| 427 | } |
| 428 | } |
| 429 | |
| 430 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type) |
| 431 | { |
| 432 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) |
| 433 | return false; |
| 434 | if (type != BPF_READ) |
| 435 | return false; |
| 436 | if (off % size != 0) |
| 437 | return false; |
| 438 | return true; |
| 439 | } |
| 440 | |
| 441 | static const struct bpf_verifier_ops tracepoint_prog_ops = { |
| 442 | .get_func_proto = tp_prog_func_proto, |
| 443 | .is_valid_access = tp_prog_is_valid_access, |
| 444 | }; |
| 445 | |
| 446 | static struct bpf_prog_type_list tracepoint_tl = { |
| 447 | .ops = &tracepoint_prog_ops, |
| 448 | .type = BPF_PROG_TYPE_TRACEPOINT, |
| 449 | }; |
| 450 | |
| 451 | static int __init register_kprobe_prog_ops(void) |
| 452 | { |
| 453 | bpf_register_prog_type(&kprobe_tl); |
| 454 | bpf_register_prog_type(&tracepoint_tl); |
| 455 | return 0; |
| 456 | } |
| 457 | late_initcall(register_kprobe_prog_ops); |