1 // SPDX-License-Identifier: GPL-2.0 2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com 3 * Copyright (c) 2016 Facebook 4 */ 5 #include <linux/kernel.h> 6 #include <linux/types.h> 7 #include <linux/slab.h> 8 #include <linux/bpf.h> 9 #include <linux/bpf_verifier.h> 10 #include <linux/bpf_perf_event.h> 11 #include <linux/btf.h> 12 #include <linux/filter.h> 13 #include <linux/uaccess.h> 14 #include <linux/ctype.h> 15 #include <linux/kprobes.h> 16 #include <linux/spinlock.h> 17 #include <linux/syscalls.h> 18 #include <linux/error-injection.h> 19 #include <linux/btf_ids.h> 20 #include <linux/bpf_lsm.h> 21 #include <linux/fprobe.h> 22 #include <linux/bsearch.h> 23 #include <linux/sort.h> 24 #include <linux/key.h> 25 #include <linux/verification.h> 26 27 #include <net/bpf_sk_storage.h> 28 29 #include <uapi/linux/bpf.h> 30 #include <uapi/linux/btf.h> 31 32 #include <asm/tlb.h> 33 34 #include "trace_probe.h" 35 #include "trace.h" 36 37 #define CREATE_TRACE_POINTS 38 #include "bpf_trace.h" 39 40 #define bpf_event_rcu_dereference(p) \ 41 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex)) 42 43 #ifdef CONFIG_MODULES 44 struct bpf_trace_module { 45 struct module *module; 46 struct list_head list; 47 }; 48 49 static LIST_HEAD(bpf_trace_modules); 50 static DEFINE_MUTEX(bpf_module_mutex); 51 52 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) 53 { 54 struct bpf_raw_event_map *btp, *ret = NULL; 55 struct bpf_trace_module *btm; 56 unsigned int i; 57 58 mutex_lock(&bpf_module_mutex); 59 list_for_each_entry(btm, &bpf_trace_modules, list) { 60 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) { 61 btp = &btm->module->bpf_raw_events[i]; 62 if (!strcmp(btp->tp->name, name)) { 63 if (try_module_get(btm->module)) 64 ret = btp; 65 goto out; 66 } 67 } 68 } 69 out: 70 mutex_unlock(&bpf_module_mutex); 71 return ret; 72 } 73 #else 74 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) 75 { 76 return NULL; 77 } 78 #endif /* CONFIG_MODULES */ 79 80 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 81 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); 82 83 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size, 84 u64 flags, const struct btf **btf, 85 s32 *btf_id); 86 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx); 87 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx); 88 89 /** 90 * trace_call_bpf - invoke BPF program 91 * @call: tracepoint event 92 * @ctx: opaque context pointer 93 * 94 * kprobe handlers execute BPF programs via this helper. 95 * Can be used from static tracepoints in the future. 96 * 97 * Return: BPF programs always return an integer which is interpreted by 98 * kprobe handler as: 99 * 0 - return from kprobe (event is filtered out) 100 * 1 - store kprobe event into ring buffer 101 * Other values are reserved and currently alias to 1 102 */ 103 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx) 104 { 105 unsigned int ret; 106 107 cant_sleep(); 108 109 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { 110 /* 111 * since some bpf program is already running on this cpu, 112 * don't call into another bpf program (same or different) 113 * and don't send kprobe event into ring-buffer, 114 * so return zero here 115 */ 116 ret = 0; 117 goto out; 118 } 119 120 /* 121 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock 122 * to all call sites, we did a bpf_prog_array_valid() there to check 123 * whether call->prog_array is empty or not, which is 124 * a heuristic to speed up execution. 125 * 126 * If bpf_prog_array_valid() fetched prog_array was 127 * non-NULL, we go into trace_call_bpf() and do the actual 128 * proper rcu_dereference() under RCU lock. 129 * If it turns out that prog_array is NULL then, we bail out. 130 * For the opposite, if the bpf_prog_array_valid() fetched pointer 131 * was NULL, you'll skip the prog_array with the risk of missing 132 * out of events when it was updated in between this and the 133 * rcu_dereference() which is accepted risk. 134 */ 135 rcu_read_lock(); 136 ret = bpf_prog_run_array(rcu_dereference(call->prog_array), 137 ctx, bpf_prog_run); 138 rcu_read_unlock(); 139 140 out: 141 __this_cpu_dec(bpf_prog_active); 142 143 return ret; 144 } 145 146 #ifdef CONFIG_BPF_KPROBE_OVERRIDE 147 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc) 148 { 149 regs_set_return_value(regs, rc); 150 override_function_with_return(regs); 151 return 0; 152 } 153 154 static const struct bpf_func_proto bpf_override_return_proto = { 155 .func = bpf_override_return, 156 .gpl_only = true, 157 .ret_type = RET_INTEGER, 158 .arg1_type = ARG_PTR_TO_CTX, 159 .arg2_type = ARG_ANYTHING, 160 }; 161 #endif 162 163 static __always_inline int 164 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr) 165 { 166 int ret; 167 168 ret = copy_from_user_nofault(dst, unsafe_ptr, size); 169 if (unlikely(ret < 0)) 170 memset(dst, 0, size); 171 return ret; 172 } 173 174 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size, 175 const void __user *, unsafe_ptr) 176 { 177 return bpf_probe_read_user_common(dst, size, unsafe_ptr); 178 } 179 180 const struct bpf_func_proto bpf_probe_read_user_proto = { 181 .func = bpf_probe_read_user, 182 .gpl_only = true, 183 .ret_type = RET_INTEGER, 184 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 185 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 186 .arg3_type = ARG_ANYTHING, 187 }; 188 189 static __always_inline int 190 bpf_probe_read_user_str_common(void *dst, u32 size, 191 const void __user *unsafe_ptr) 192 { 193 int ret; 194 195 /* 196 * NB: We rely on strncpy_from_user() not copying junk past the NUL 197 * terminator into `dst`. 198 * 199 * strncpy_from_user() does long-sized strides in the fast path. If the 200 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`, 201 * then there could be junk after the NUL in `dst`. If user takes `dst` 202 * and keys a hash map with it, then semantically identical strings can 203 * occupy multiple entries in the map. 204 */ 205 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size); 206 if (unlikely(ret < 0)) 207 memset(dst, 0, size); 208 return ret; 209 } 210 211 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size, 212 const void __user *, unsafe_ptr) 213 { 214 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr); 215 } 216 217 const struct bpf_func_proto bpf_probe_read_user_str_proto = { 218 .func = bpf_probe_read_user_str, 219 .gpl_only = true, 220 .ret_type = RET_INTEGER, 221 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 222 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 223 .arg3_type = ARG_ANYTHING, 224 }; 225 226 static __always_inline int 227 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr) 228 { 229 int ret; 230 231 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size); 232 if (unlikely(ret < 0)) 233 memset(dst, 0, size); 234 return ret; 235 } 236 237 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size, 238 const void *, unsafe_ptr) 239 { 240 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr); 241 } 242 243 const struct bpf_func_proto bpf_probe_read_kernel_proto = { 244 .func = bpf_probe_read_kernel, 245 .gpl_only = true, 246 .ret_type = RET_INTEGER, 247 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 248 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 249 .arg3_type = ARG_ANYTHING, 250 }; 251 252 static __always_inline int 253 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr) 254 { 255 int ret; 256 257 /* 258 * The strncpy_from_kernel_nofault() call will likely not fill the 259 * entire buffer, but that's okay in this circumstance as we're probing 260 * arbitrary memory anyway similar to bpf_probe_read_*() and might 261 * as well probe the stack. Thus, memory is explicitly cleared 262 * only in error case, so that improper users ignoring return 263 * code altogether don't copy garbage; otherwise length of string 264 * is returned that can be used for bpf_perf_event_output() et al. 265 */ 266 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size); 267 if (unlikely(ret < 0)) 268 memset(dst, 0, size); 269 return ret; 270 } 271 272 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size, 273 const void *, unsafe_ptr) 274 { 275 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr); 276 } 277 278 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = { 279 .func = bpf_probe_read_kernel_str, 280 .gpl_only = true, 281 .ret_type = RET_INTEGER, 282 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 283 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 284 .arg3_type = ARG_ANYTHING, 285 }; 286 287 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE 288 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size, 289 const void *, unsafe_ptr) 290 { 291 if ((unsigned long)unsafe_ptr < TASK_SIZE) { 292 return bpf_probe_read_user_common(dst, size, 293 (__force void __user *)unsafe_ptr); 294 } 295 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr); 296 } 297 298 static const struct bpf_func_proto bpf_probe_read_compat_proto = { 299 .func = bpf_probe_read_compat, 300 .gpl_only = true, 301 .ret_type = RET_INTEGER, 302 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 303 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 304 .arg3_type = ARG_ANYTHING, 305 }; 306 307 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size, 308 const void *, unsafe_ptr) 309 { 310 if ((unsigned long)unsafe_ptr < TASK_SIZE) { 311 return bpf_probe_read_user_str_common(dst, size, 312 (__force void __user *)unsafe_ptr); 313 } 314 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr); 315 } 316 317 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = { 318 .func = bpf_probe_read_compat_str, 319 .gpl_only = true, 320 .ret_type = RET_INTEGER, 321 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 322 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 323 .arg3_type = ARG_ANYTHING, 324 }; 325 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */ 326 327 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src, 328 u32, size) 329 { 330 /* 331 * Ensure we're in user context which is safe for the helper to 332 * run. This helper has no business in a kthread. 333 * 334 * access_ok() should prevent writing to non-user memory, but in 335 * some situations (nommu, temporary switch, etc) access_ok() does 336 * not provide enough validation, hence the check on KERNEL_DS. 337 * 338 * nmi_uaccess_okay() ensures the probe is not run in an interim 339 * state, when the task or mm are switched. This is specifically 340 * required to prevent the use of temporary mm. 341 */ 342 343 if (unlikely(in_interrupt() || 344 current->flags & (PF_KTHREAD | PF_EXITING))) 345 return -EPERM; 346 if (unlikely(!nmi_uaccess_okay())) 347 return -EPERM; 348 349 return copy_to_user_nofault(unsafe_ptr, src, size); 350 } 351 352 static const struct bpf_func_proto bpf_probe_write_user_proto = { 353 .func = bpf_probe_write_user, 354 .gpl_only = true, 355 .ret_type = RET_INTEGER, 356 .arg1_type = ARG_ANYTHING, 357 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 358 .arg3_type = ARG_CONST_SIZE, 359 }; 360 361 static const struct bpf_func_proto *bpf_get_probe_write_proto(void) 362 { 363 if (!capable(CAP_SYS_ADMIN)) 364 return NULL; 365 366 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", 367 current->comm, task_pid_nr(current)); 368 369 return &bpf_probe_write_user_proto; 370 } 371 372 #define MAX_TRACE_PRINTK_VARARGS 3 373 #define BPF_TRACE_PRINTK_SIZE 1024 374 375 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, 376 u64, arg2, u64, arg3) 377 { 378 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 }; 379 struct bpf_bprintf_data data = { 380 .get_bin_args = true, 381 .get_buf = true, 382 }; 383 int ret; 384 385 ret = bpf_bprintf_prepare(fmt, fmt_size, args, 386 MAX_TRACE_PRINTK_VARARGS, &data); 387 if (ret < 0) 388 return ret; 389 390 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args); 391 392 trace_bpf_trace_printk(data.buf); 393 394 bpf_bprintf_cleanup(&data); 395 396 return ret; 397 } 398 399 static const struct bpf_func_proto bpf_trace_printk_proto = { 400 .func = bpf_trace_printk, 401 .gpl_only = true, 402 .ret_type = RET_INTEGER, 403 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, 404 .arg2_type = ARG_CONST_SIZE, 405 }; 406 407 static void __set_printk_clr_event(void) 408 { 409 /* 410 * This program might be calling bpf_trace_printk, 411 * so enable the associated bpf_trace/bpf_trace_printk event. 412 * Repeat this each time as it is possible a user has 413 * disabled bpf_trace_printk events. By loading a program 414 * calling bpf_trace_printk() however the user has expressed 415 * the intent to see such events. 416 */ 417 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1)) 418 pr_warn_ratelimited("could not enable bpf_trace_printk events"); 419 } 420 421 const struct bpf_func_proto *bpf_get_trace_printk_proto(void) 422 { 423 __set_printk_clr_event(); 424 return &bpf_trace_printk_proto; 425 } 426 427 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, args, 428 u32, data_len) 429 { 430 struct bpf_bprintf_data data = { 431 .get_bin_args = true, 432 .get_buf = true, 433 }; 434 int ret, num_args; 435 436 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 || 437 (data_len && !args)) 438 return -EINVAL; 439 num_args = data_len / 8; 440 441 ret = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data); 442 if (ret < 0) 443 return ret; 444 445 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args); 446 447 trace_bpf_trace_printk(data.buf); 448 449 bpf_bprintf_cleanup(&data); 450 451 return ret; 452 } 453 454 static const struct bpf_func_proto bpf_trace_vprintk_proto = { 455 .func = bpf_trace_vprintk, 456 .gpl_only = true, 457 .ret_type = RET_INTEGER, 458 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY, 459 .arg2_type = ARG_CONST_SIZE, 460 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 461 .arg4_type = ARG_CONST_SIZE_OR_ZERO, 462 }; 463 464 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void) 465 { 466 __set_printk_clr_event(); 467 return &bpf_trace_vprintk_proto; 468 } 469 470 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size, 471 const void *, args, u32, data_len) 472 { 473 struct bpf_bprintf_data data = { 474 .get_bin_args = true, 475 }; 476 int err, num_args; 477 478 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 || 479 (data_len && !args)) 480 return -EINVAL; 481 num_args = data_len / 8; 482 483 err = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data); 484 if (err < 0) 485 return err; 486 487 seq_bprintf(m, fmt, data.bin_args); 488 489 bpf_bprintf_cleanup(&data); 490 491 return seq_has_overflowed(m) ? -EOVERFLOW : 0; 492 } 493 494 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file) 495 496 static const struct bpf_func_proto bpf_seq_printf_proto = { 497 .func = bpf_seq_printf, 498 .gpl_only = true, 499 .ret_type = RET_INTEGER, 500 .arg1_type = ARG_PTR_TO_BTF_ID, 501 .arg1_btf_id = &btf_seq_file_ids[0], 502 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 503 .arg3_type = ARG_CONST_SIZE, 504 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY, 505 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 506 }; 507 508 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len) 509 { 510 return seq_write(m, data, len) ? -EOVERFLOW : 0; 511 } 512 513 static const struct bpf_func_proto bpf_seq_write_proto = { 514 .func = bpf_seq_write, 515 .gpl_only = true, 516 .ret_type = RET_INTEGER, 517 .arg1_type = ARG_PTR_TO_BTF_ID, 518 .arg1_btf_id = &btf_seq_file_ids[0], 519 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 520 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 521 }; 522 523 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr, 524 u32, btf_ptr_size, u64, flags) 525 { 526 const struct btf *btf; 527 s32 btf_id; 528 int ret; 529 530 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id); 531 if (ret) 532 return ret; 533 534 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags); 535 } 536 537 static const struct bpf_func_proto bpf_seq_printf_btf_proto = { 538 .func = bpf_seq_printf_btf, 539 .gpl_only = true, 540 .ret_type = RET_INTEGER, 541 .arg1_type = ARG_PTR_TO_BTF_ID, 542 .arg1_btf_id = &btf_seq_file_ids[0], 543 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 544 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 545 .arg4_type = ARG_ANYTHING, 546 }; 547 548 static __always_inline int 549 get_map_perf_counter(struct bpf_map *map, u64 flags, 550 u64 *value, u64 *enabled, u64 *running) 551 { 552 struct bpf_array *array = container_of(map, struct bpf_array, map); 553 unsigned int cpu = smp_processor_id(); 554 u64 index = flags & BPF_F_INDEX_MASK; 555 struct bpf_event_entry *ee; 556 557 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) 558 return -EINVAL; 559 if (index == BPF_F_CURRENT_CPU) 560 index = cpu; 561 if (unlikely(index >= array->map.max_entries)) 562 return -E2BIG; 563 564 ee = READ_ONCE(array->ptrs[index]); 565 if (!ee) 566 return -ENOENT; 567 568 return perf_event_read_local(ee->event, value, enabled, running); 569 } 570 571 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) 572 { 573 u64 value = 0; 574 int err; 575 576 err = get_map_perf_counter(map, flags, &value, NULL, NULL); 577 /* 578 * this api is ugly since we miss [-22..-2] range of valid 579 * counter values, but that's uapi 580 */ 581 if (err) 582 return err; 583 return value; 584 } 585 586 static const struct bpf_func_proto bpf_perf_event_read_proto = { 587 .func = bpf_perf_event_read, 588 .gpl_only = true, 589 .ret_type = RET_INTEGER, 590 .arg1_type = ARG_CONST_MAP_PTR, 591 .arg2_type = ARG_ANYTHING, 592 }; 593 594 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags, 595 struct bpf_perf_event_value *, buf, u32, size) 596 { 597 int err = -EINVAL; 598 599 if (unlikely(size != sizeof(struct bpf_perf_event_value))) 600 goto clear; 601 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled, 602 &buf->running); 603 if (unlikely(err)) 604 goto clear; 605 return 0; 606 clear: 607 memset(buf, 0, size); 608 return err; 609 } 610 611 static const struct bpf_func_proto bpf_perf_event_read_value_proto = { 612 .func = bpf_perf_event_read_value, 613 .gpl_only = true, 614 .ret_type = RET_INTEGER, 615 .arg1_type = ARG_CONST_MAP_PTR, 616 .arg2_type = ARG_ANYTHING, 617 .arg3_type = ARG_PTR_TO_UNINIT_MEM, 618 .arg4_type = ARG_CONST_SIZE, 619 }; 620 621 static __always_inline u64 622 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, 623 u64 flags, struct perf_sample_data *sd) 624 { 625 struct bpf_array *array = container_of(map, struct bpf_array, map); 626 unsigned int cpu = smp_processor_id(); 627 u64 index = flags & BPF_F_INDEX_MASK; 628 struct bpf_event_entry *ee; 629 struct perf_event *event; 630 631 if (index == BPF_F_CURRENT_CPU) 632 index = cpu; 633 if (unlikely(index >= array->map.max_entries)) 634 return -E2BIG; 635 636 ee = READ_ONCE(array->ptrs[index]); 637 if (!ee) 638 return -ENOENT; 639 640 event = ee->event; 641 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || 642 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) 643 return -EINVAL; 644 645 if (unlikely(event->oncpu != cpu)) 646 return -EOPNOTSUPP; 647 648 return perf_event_output(event, sd, regs); 649 } 650 651 /* 652 * Support executing tracepoints in normal, irq, and nmi context that each call 653 * bpf_perf_event_output 654 */ 655 struct bpf_trace_sample_data { 656 struct perf_sample_data sds[3]; 657 }; 658 659 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds); 660 static DEFINE_PER_CPU(int, bpf_trace_nest_level); 661 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, 662 u64, flags, void *, data, u64, size) 663 { 664 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds); 665 int nest_level = this_cpu_inc_return(bpf_trace_nest_level); 666 struct perf_raw_record raw = { 667 .frag = { 668 .size = size, 669 .data = data, 670 }, 671 }; 672 struct perf_sample_data *sd; 673 int err; 674 675 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) { 676 err = -EBUSY; 677 goto out; 678 } 679 680 sd = &sds->sds[nest_level - 1]; 681 682 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) { 683 err = -EINVAL; 684 goto out; 685 } 686 687 perf_sample_data_init(sd, 0, 0); 688 perf_sample_save_raw_data(sd, &raw); 689 690 err = __bpf_perf_event_output(regs, map, flags, sd); 691 692 out: 693 this_cpu_dec(bpf_trace_nest_level); 694 return err; 695 } 696 697 static const struct bpf_func_proto bpf_perf_event_output_proto = { 698 .func = bpf_perf_event_output, 699 .gpl_only = true, 700 .ret_type = RET_INTEGER, 701 .arg1_type = ARG_PTR_TO_CTX, 702 .arg2_type = ARG_CONST_MAP_PTR, 703 .arg3_type = ARG_ANYTHING, 704 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 705 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 706 }; 707 708 static DEFINE_PER_CPU(int, bpf_event_output_nest_level); 709 struct bpf_nested_pt_regs { 710 struct pt_regs regs[3]; 711 }; 712 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs); 713 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds); 714 715 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, 716 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) 717 { 718 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level); 719 struct perf_raw_frag frag = { 720 .copy = ctx_copy, 721 .size = ctx_size, 722 .data = ctx, 723 }; 724 struct perf_raw_record raw = { 725 .frag = { 726 { 727 .next = ctx_size ? &frag : NULL, 728 }, 729 .size = meta_size, 730 .data = meta, 731 }, 732 }; 733 struct perf_sample_data *sd; 734 struct pt_regs *regs; 735 u64 ret; 736 737 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) { 738 ret = -EBUSY; 739 goto out; 740 } 741 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]); 742 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]); 743 744 perf_fetch_caller_regs(regs); 745 perf_sample_data_init(sd, 0, 0); 746 perf_sample_save_raw_data(sd, &raw); 747 748 ret = __bpf_perf_event_output(regs, map, flags, sd); 749 out: 750 this_cpu_dec(bpf_event_output_nest_level); 751 return ret; 752 } 753 754 BPF_CALL_0(bpf_get_current_task) 755 { 756 return (long) current; 757 } 758 759 const struct bpf_func_proto bpf_get_current_task_proto = { 760 .func = bpf_get_current_task, 761 .gpl_only = true, 762 .ret_type = RET_INTEGER, 763 }; 764 765 BPF_CALL_0(bpf_get_current_task_btf) 766 { 767 return (unsigned long) current; 768 } 769 770 const struct bpf_func_proto bpf_get_current_task_btf_proto = { 771 .func = bpf_get_current_task_btf, 772 .gpl_only = true, 773 .ret_type = RET_PTR_TO_BTF_ID_TRUSTED, 774 .ret_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK], 775 }; 776 777 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task) 778 { 779 return (unsigned long) task_pt_regs(task); 780 } 781 782 BTF_ID_LIST(bpf_task_pt_regs_ids) 783 BTF_ID(struct, pt_regs) 784 785 const struct bpf_func_proto bpf_task_pt_regs_proto = { 786 .func = bpf_task_pt_regs, 787 .gpl_only = true, 788 .arg1_type = ARG_PTR_TO_BTF_ID, 789 .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK], 790 .ret_type = RET_PTR_TO_BTF_ID, 791 .ret_btf_id = &bpf_task_pt_regs_ids[0], 792 }; 793 794 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) 795 { 796 struct bpf_array *array = container_of(map, struct bpf_array, map); 797 struct cgroup *cgrp; 798 799 if (unlikely(idx >= array->map.max_entries)) 800 return -E2BIG; 801 802 cgrp = READ_ONCE(array->ptrs[idx]); 803 if (unlikely(!cgrp)) 804 return -EAGAIN; 805 806 return task_under_cgroup_hierarchy(current, cgrp); 807 } 808 809 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { 810 .func = bpf_current_task_under_cgroup, 811 .gpl_only = false, 812 .ret_type = RET_INTEGER, 813 .arg1_type = ARG_CONST_MAP_PTR, 814 .arg2_type = ARG_ANYTHING, 815 }; 816 817 struct send_signal_irq_work { 818 struct irq_work irq_work; 819 struct task_struct *task; 820 u32 sig; 821 enum pid_type type; 822 }; 823 824 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work); 825 826 static void do_bpf_send_signal(struct irq_work *entry) 827 { 828 struct send_signal_irq_work *work; 829 830 work = container_of(entry, struct send_signal_irq_work, irq_work); 831 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type); 832 put_task_struct(work->task); 833 } 834 835 static int bpf_send_signal_common(u32 sig, enum pid_type type) 836 { 837 struct send_signal_irq_work *work = NULL; 838 839 /* Similar to bpf_probe_write_user, task needs to be 840 * in a sound condition and kernel memory access be 841 * permitted in order to send signal to the current 842 * task. 843 */ 844 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING))) 845 return -EPERM; 846 if (unlikely(!nmi_uaccess_okay())) 847 return -EPERM; 848 /* Task should not be pid=1 to avoid kernel panic. */ 849 if (unlikely(is_global_init(current))) 850 return -EPERM; 851 852 if (irqs_disabled()) { 853 /* Do an early check on signal validity. Otherwise, 854 * the error is lost in deferred irq_work. 855 */ 856 if (unlikely(!valid_signal(sig))) 857 return -EINVAL; 858 859 work = this_cpu_ptr(&send_signal_work); 860 if (irq_work_is_busy(&work->irq_work)) 861 return -EBUSY; 862 863 /* Add the current task, which is the target of sending signal, 864 * to the irq_work. The current task may change when queued 865 * irq works get executed. 866 */ 867 work->task = get_task_struct(current); 868 work->sig = sig; 869 work->type = type; 870 irq_work_queue(&work->irq_work); 871 return 0; 872 } 873 874 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type); 875 } 876 877 BPF_CALL_1(bpf_send_signal, u32, sig) 878 { 879 return bpf_send_signal_common(sig, PIDTYPE_TGID); 880 } 881 882 static const struct bpf_func_proto bpf_send_signal_proto = { 883 .func = bpf_send_signal, 884 .gpl_only = false, 885 .ret_type = RET_INTEGER, 886 .arg1_type = ARG_ANYTHING, 887 }; 888 889 BPF_CALL_1(bpf_send_signal_thread, u32, sig) 890 { 891 return bpf_send_signal_common(sig, PIDTYPE_PID); 892 } 893 894 static const struct bpf_func_proto bpf_send_signal_thread_proto = { 895 .func = bpf_send_signal_thread, 896 .gpl_only = false, 897 .ret_type = RET_INTEGER, 898 .arg1_type = ARG_ANYTHING, 899 }; 900 901 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz) 902 { 903 struct path copy; 904 long len; 905 char *p; 906 907 if (!sz) 908 return 0; 909 910 /* 911 * The path pointer is verified as trusted and safe to use, 912 * but let's double check it's valid anyway to workaround 913 * potentially broken verifier. 914 */ 915 len = copy_from_kernel_nofault(©, path, sizeof(*path)); 916 if (len < 0) 917 return len; 918 919 p = d_path(©, buf, sz); 920 if (IS_ERR(p)) { 921 len = PTR_ERR(p); 922 } else { 923 len = buf + sz - p; 924 memmove(buf, p, len); 925 } 926 927 return len; 928 } 929 930 BTF_SET_START(btf_allowlist_d_path) 931 #ifdef CONFIG_SECURITY 932 BTF_ID(func, security_file_permission) 933 BTF_ID(func, security_inode_getattr) 934 BTF_ID(func, security_file_open) 935 #endif 936 #ifdef CONFIG_SECURITY_PATH 937 BTF_ID(func, security_path_truncate) 938 #endif 939 BTF_ID(func, vfs_truncate) 940 BTF_ID(func, vfs_fallocate) 941 BTF_ID(func, dentry_open) 942 BTF_ID(func, vfs_getattr) 943 BTF_ID(func, filp_close) 944 BTF_SET_END(btf_allowlist_d_path) 945 946 static bool bpf_d_path_allowed(const struct bpf_prog *prog) 947 { 948 if (prog->type == BPF_PROG_TYPE_TRACING && 949 prog->expected_attach_type == BPF_TRACE_ITER) 950 return true; 951 952 if (prog->type == BPF_PROG_TYPE_LSM) 953 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id); 954 955 return btf_id_set_contains(&btf_allowlist_d_path, 956 prog->aux->attach_btf_id); 957 } 958 959 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path) 960 961 static const struct bpf_func_proto bpf_d_path_proto = { 962 .func = bpf_d_path, 963 .gpl_only = false, 964 .ret_type = RET_INTEGER, 965 .arg1_type = ARG_PTR_TO_BTF_ID, 966 .arg1_btf_id = &bpf_d_path_btf_ids[0], 967 .arg2_type = ARG_PTR_TO_MEM, 968 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 969 .allowed = bpf_d_path_allowed, 970 }; 971 972 #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \ 973 BTF_F_PTR_RAW | BTF_F_ZERO) 974 975 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size, 976 u64 flags, const struct btf **btf, 977 s32 *btf_id) 978 { 979 const struct btf_type *t; 980 981 if (unlikely(flags & ~(BTF_F_ALL))) 982 return -EINVAL; 983 984 if (btf_ptr_size != sizeof(struct btf_ptr)) 985 return -EINVAL; 986 987 *btf = bpf_get_btf_vmlinux(); 988 989 if (IS_ERR_OR_NULL(*btf)) 990 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL; 991 992 if (ptr->type_id > 0) 993 *btf_id = ptr->type_id; 994 else 995 return -EINVAL; 996 997 if (*btf_id > 0) 998 t = btf_type_by_id(*btf, *btf_id); 999 if (*btf_id <= 0 || !t) 1000 return -ENOENT; 1001 1002 return 0; 1003 } 1004 1005 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr, 1006 u32, btf_ptr_size, u64, flags) 1007 { 1008 const struct btf *btf; 1009 s32 btf_id; 1010 int ret; 1011 1012 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id); 1013 if (ret) 1014 return ret; 1015 1016 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size, 1017 flags); 1018 } 1019 1020 const struct bpf_func_proto bpf_snprintf_btf_proto = { 1021 .func = bpf_snprintf_btf, 1022 .gpl_only = false, 1023 .ret_type = RET_INTEGER, 1024 .arg1_type = ARG_PTR_TO_MEM, 1025 .arg2_type = ARG_CONST_SIZE, 1026 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1027 .arg4_type = ARG_CONST_SIZE, 1028 .arg5_type = ARG_ANYTHING, 1029 }; 1030 1031 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx) 1032 { 1033 /* This helper call is inlined by verifier. */ 1034 return ((u64 *)ctx)[-2]; 1035 } 1036 1037 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = { 1038 .func = bpf_get_func_ip_tracing, 1039 .gpl_only = true, 1040 .ret_type = RET_INTEGER, 1041 .arg1_type = ARG_PTR_TO_CTX, 1042 }; 1043 1044 #ifdef CONFIG_X86_KERNEL_IBT 1045 static unsigned long get_entry_ip(unsigned long fentry_ip) 1046 { 1047 u32 instr; 1048 1049 /* Being extra safe in here in case entry ip is on the page-edge. */ 1050 if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1)) 1051 return fentry_ip; 1052 if (is_endbr(instr)) 1053 fentry_ip -= ENDBR_INSN_SIZE; 1054 return fentry_ip; 1055 } 1056 #else 1057 #define get_entry_ip(fentry_ip) fentry_ip 1058 #endif 1059 1060 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs) 1061 { 1062 struct kprobe *kp = kprobe_running(); 1063 1064 if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY)) 1065 return 0; 1066 1067 return get_entry_ip((uintptr_t)kp->addr); 1068 } 1069 1070 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = { 1071 .func = bpf_get_func_ip_kprobe, 1072 .gpl_only = true, 1073 .ret_type = RET_INTEGER, 1074 .arg1_type = ARG_PTR_TO_CTX, 1075 }; 1076 1077 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs) 1078 { 1079 return bpf_kprobe_multi_entry_ip(current->bpf_ctx); 1080 } 1081 1082 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = { 1083 .func = bpf_get_func_ip_kprobe_multi, 1084 .gpl_only = false, 1085 .ret_type = RET_INTEGER, 1086 .arg1_type = ARG_PTR_TO_CTX, 1087 }; 1088 1089 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs) 1090 { 1091 return bpf_kprobe_multi_cookie(current->bpf_ctx); 1092 } 1093 1094 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = { 1095 .func = bpf_get_attach_cookie_kprobe_multi, 1096 .gpl_only = false, 1097 .ret_type = RET_INTEGER, 1098 .arg1_type = ARG_PTR_TO_CTX, 1099 }; 1100 1101 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx) 1102 { 1103 struct bpf_trace_run_ctx *run_ctx; 1104 1105 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx); 1106 return run_ctx->bpf_cookie; 1107 } 1108 1109 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = { 1110 .func = bpf_get_attach_cookie_trace, 1111 .gpl_only = false, 1112 .ret_type = RET_INTEGER, 1113 .arg1_type = ARG_PTR_TO_CTX, 1114 }; 1115 1116 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx) 1117 { 1118 return ctx->event->bpf_cookie; 1119 } 1120 1121 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = { 1122 .func = bpf_get_attach_cookie_pe, 1123 .gpl_only = false, 1124 .ret_type = RET_INTEGER, 1125 .arg1_type = ARG_PTR_TO_CTX, 1126 }; 1127 1128 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx) 1129 { 1130 struct bpf_trace_run_ctx *run_ctx; 1131 1132 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx); 1133 return run_ctx->bpf_cookie; 1134 } 1135 1136 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = { 1137 .func = bpf_get_attach_cookie_tracing, 1138 .gpl_only = false, 1139 .ret_type = RET_INTEGER, 1140 .arg1_type = ARG_PTR_TO_CTX, 1141 }; 1142 1143 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags) 1144 { 1145 #ifndef CONFIG_X86 1146 return -ENOENT; 1147 #else 1148 static const u32 br_entry_size = sizeof(struct perf_branch_entry); 1149 u32 entry_cnt = size / br_entry_size; 1150 1151 entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt); 1152 1153 if (unlikely(flags)) 1154 return -EINVAL; 1155 1156 if (!entry_cnt) 1157 return -ENOENT; 1158 1159 return entry_cnt * br_entry_size; 1160 #endif 1161 } 1162 1163 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = { 1164 .func = bpf_get_branch_snapshot, 1165 .gpl_only = true, 1166 .ret_type = RET_INTEGER, 1167 .arg1_type = ARG_PTR_TO_UNINIT_MEM, 1168 .arg2_type = ARG_CONST_SIZE_OR_ZERO, 1169 }; 1170 1171 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value) 1172 { 1173 /* This helper call is inlined by verifier. */ 1174 u64 nr_args = ((u64 *)ctx)[-1]; 1175 1176 if ((u64) n >= nr_args) 1177 return -EINVAL; 1178 *value = ((u64 *)ctx)[n]; 1179 return 0; 1180 } 1181 1182 static const struct bpf_func_proto bpf_get_func_arg_proto = { 1183 .func = get_func_arg, 1184 .ret_type = RET_INTEGER, 1185 .arg1_type = ARG_PTR_TO_CTX, 1186 .arg2_type = ARG_ANYTHING, 1187 .arg3_type = ARG_PTR_TO_LONG, 1188 }; 1189 1190 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value) 1191 { 1192 /* This helper call is inlined by verifier. */ 1193 u64 nr_args = ((u64 *)ctx)[-1]; 1194 1195 *value = ((u64 *)ctx)[nr_args]; 1196 return 0; 1197 } 1198 1199 static const struct bpf_func_proto bpf_get_func_ret_proto = { 1200 .func = get_func_ret, 1201 .ret_type = RET_INTEGER, 1202 .arg1_type = ARG_PTR_TO_CTX, 1203 .arg2_type = ARG_PTR_TO_LONG, 1204 }; 1205 1206 BPF_CALL_1(get_func_arg_cnt, void *, ctx) 1207 { 1208 /* This helper call is inlined by verifier. */ 1209 return ((u64 *)ctx)[-1]; 1210 } 1211 1212 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = { 1213 .func = get_func_arg_cnt, 1214 .ret_type = RET_INTEGER, 1215 .arg1_type = ARG_PTR_TO_CTX, 1216 }; 1217 1218 #ifdef CONFIG_KEYS 1219 __diag_push(); 1220 __diag_ignore_all("-Wmissing-prototypes", 1221 "kfuncs which will be used in BPF programs"); 1222 1223 /** 1224 * bpf_lookup_user_key - lookup a key by its serial 1225 * @serial: key handle serial number 1226 * @flags: lookup-specific flags 1227 * 1228 * Search a key with a given *serial* and the provided *flags*. 1229 * If found, increment the reference count of the key by one, and 1230 * return it in the bpf_key structure. 1231 * 1232 * The bpf_key structure must be passed to bpf_key_put() when done 1233 * with it, so that the key reference count is decremented and the 1234 * bpf_key structure is freed. 1235 * 1236 * Permission checks are deferred to the time the key is used by 1237 * one of the available key-specific kfuncs. 1238 * 1239 * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested 1240 * special keyring (e.g. session keyring), if it doesn't yet exist. 1241 * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting 1242 * for the key construction, and to retrieve uninstantiated keys (keys 1243 * without data attached to them). 1244 * 1245 * Return: a bpf_key pointer with a valid key pointer if the key is found, a 1246 * NULL pointer otherwise. 1247 */ 1248 __bpf_kfunc struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags) 1249 { 1250 key_ref_t key_ref; 1251 struct bpf_key *bkey; 1252 1253 if (flags & ~KEY_LOOKUP_ALL) 1254 return NULL; 1255 1256 /* 1257 * Permission check is deferred until the key is used, as the 1258 * intent of the caller is unknown here. 1259 */ 1260 key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK); 1261 if (IS_ERR(key_ref)) 1262 return NULL; 1263 1264 bkey = kmalloc(sizeof(*bkey), GFP_KERNEL); 1265 if (!bkey) { 1266 key_put(key_ref_to_ptr(key_ref)); 1267 return NULL; 1268 } 1269 1270 bkey->key = key_ref_to_ptr(key_ref); 1271 bkey->has_ref = true; 1272 1273 return bkey; 1274 } 1275 1276 /** 1277 * bpf_lookup_system_key - lookup a key by a system-defined ID 1278 * @id: key ID 1279 * 1280 * Obtain a bpf_key structure with a key pointer set to the passed key ID. 1281 * The key pointer is marked as invalid, to prevent bpf_key_put() from 1282 * attempting to decrement the key reference count on that pointer. The key 1283 * pointer set in such way is currently understood only by 1284 * verify_pkcs7_signature(). 1285 * 1286 * Set *id* to one of the values defined in include/linux/verification.h: 1287 * 0 for the primary keyring (immutable keyring of system keys); 1288 * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring 1289 * (where keys can be added only if they are vouched for by existing keys 1290 * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform 1291 * keyring (primarily used by the integrity subsystem to verify a kexec'ed 1292 * kerned image and, possibly, the initramfs signature). 1293 * 1294 * Return: a bpf_key pointer with an invalid key pointer set from the 1295 * pre-determined ID on success, a NULL pointer otherwise 1296 */ 1297 __bpf_kfunc struct bpf_key *bpf_lookup_system_key(u64 id) 1298 { 1299 struct bpf_key *bkey; 1300 1301 if (system_keyring_id_check(id) < 0) 1302 return NULL; 1303 1304 bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC); 1305 if (!bkey) 1306 return NULL; 1307 1308 bkey->key = (struct key *)(unsigned long)id; 1309 bkey->has_ref = false; 1310 1311 return bkey; 1312 } 1313 1314 /** 1315 * bpf_key_put - decrement key reference count if key is valid and free bpf_key 1316 * @bkey: bpf_key structure 1317 * 1318 * Decrement the reference count of the key inside *bkey*, if the pointer 1319 * is valid, and free *bkey*. 1320 */ 1321 __bpf_kfunc void bpf_key_put(struct bpf_key *bkey) 1322 { 1323 if (bkey->has_ref) 1324 key_put(bkey->key); 1325 1326 kfree(bkey); 1327 } 1328 1329 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION 1330 /** 1331 * bpf_verify_pkcs7_signature - verify a PKCS#7 signature 1332 * @data_ptr: data to verify 1333 * @sig_ptr: signature of the data 1334 * @trusted_keyring: keyring with keys trusted for signature verification 1335 * 1336 * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr* 1337 * with keys in a keyring referenced by *trusted_keyring*. 1338 * 1339 * Return: 0 on success, a negative value on error. 1340 */ 1341 __bpf_kfunc int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr, 1342 struct bpf_dynptr_kern *sig_ptr, 1343 struct bpf_key *trusted_keyring) 1344 { 1345 int ret; 1346 1347 if (trusted_keyring->has_ref) { 1348 /* 1349 * Do the permission check deferred in bpf_lookup_user_key(). 1350 * See bpf_lookup_user_key() for more details. 1351 * 1352 * A call to key_task_permission() here would be redundant, as 1353 * it is already done by keyring_search() called by 1354 * find_asymmetric_key(). 1355 */ 1356 ret = key_validate(trusted_keyring->key); 1357 if (ret < 0) 1358 return ret; 1359 } 1360 1361 return verify_pkcs7_signature(data_ptr->data, 1362 __bpf_dynptr_size(data_ptr), 1363 sig_ptr->data, 1364 __bpf_dynptr_size(sig_ptr), 1365 trusted_keyring->key, 1366 VERIFYING_UNSPECIFIED_SIGNATURE, NULL, 1367 NULL); 1368 } 1369 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */ 1370 1371 __diag_pop(); 1372 1373 BTF_SET8_START(key_sig_kfunc_set) 1374 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE) 1375 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL) 1376 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE) 1377 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION 1378 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE) 1379 #endif 1380 BTF_SET8_END(key_sig_kfunc_set) 1381 1382 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = { 1383 .owner = THIS_MODULE, 1384 .set = &key_sig_kfunc_set, 1385 }; 1386 1387 static int __init bpf_key_sig_kfuncs_init(void) 1388 { 1389 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING, 1390 &bpf_key_sig_kfunc_set); 1391 } 1392 1393 late_initcall(bpf_key_sig_kfuncs_init); 1394 #endif /* CONFIG_KEYS */ 1395 1396 static const struct bpf_func_proto * 1397 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1398 { 1399 switch (func_id) { 1400 case BPF_FUNC_map_lookup_elem: 1401 return &bpf_map_lookup_elem_proto; 1402 case BPF_FUNC_map_update_elem: 1403 return &bpf_map_update_elem_proto; 1404 case BPF_FUNC_map_delete_elem: 1405 return &bpf_map_delete_elem_proto; 1406 case BPF_FUNC_map_push_elem: 1407 return &bpf_map_push_elem_proto; 1408 case BPF_FUNC_map_pop_elem: 1409 return &bpf_map_pop_elem_proto; 1410 case BPF_FUNC_map_peek_elem: 1411 return &bpf_map_peek_elem_proto; 1412 case BPF_FUNC_map_lookup_percpu_elem: 1413 return &bpf_map_lookup_percpu_elem_proto; 1414 case BPF_FUNC_ktime_get_ns: 1415 return &bpf_ktime_get_ns_proto; 1416 case BPF_FUNC_ktime_get_boot_ns: 1417 return &bpf_ktime_get_boot_ns_proto; 1418 case BPF_FUNC_tail_call: 1419 return &bpf_tail_call_proto; 1420 case BPF_FUNC_get_current_pid_tgid: 1421 return &bpf_get_current_pid_tgid_proto; 1422 case BPF_FUNC_get_current_task: 1423 return &bpf_get_current_task_proto; 1424 case BPF_FUNC_get_current_task_btf: 1425 return &bpf_get_current_task_btf_proto; 1426 case BPF_FUNC_task_pt_regs: 1427 return &bpf_task_pt_regs_proto; 1428 case BPF_FUNC_get_current_uid_gid: 1429 return &bpf_get_current_uid_gid_proto; 1430 case BPF_FUNC_get_current_comm: 1431 return &bpf_get_current_comm_proto; 1432 case BPF_FUNC_trace_printk: 1433 return bpf_get_trace_printk_proto(); 1434 case BPF_FUNC_get_smp_processor_id: 1435 return &bpf_get_smp_processor_id_proto; 1436 case BPF_FUNC_get_numa_node_id: 1437 return &bpf_get_numa_node_id_proto; 1438 case BPF_FUNC_perf_event_read: 1439 return &bpf_perf_event_read_proto; 1440 case BPF_FUNC_current_task_under_cgroup: 1441 return &bpf_current_task_under_cgroup_proto; 1442 case BPF_FUNC_get_prandom_u32: 1443 return &bpf_get_prandom_u32_proto; 1444 case BPF_FUNC_probe_write_user: 1445 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ? 1446 NULL : bpf_get_probe_write_proto(); 1447 case BPF_FUNC_probe_read_user: 1448 return &bpf_probe_read_user_proto; 1449 case BPF_FUNC_probe_read_kernel: 1450 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? 1451 NULL : &bpf_probe_read_kernel_proto; 1452 case BPF_FUNC_probe_read_user_str: 1453 return &bpf_probe_read_user_str_proto; 1454 case BPF_FUNC_probe_read_kernel_str: 1455 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? 1456 NULL : &bpf_probe_read_kernel_str_proto; 1457 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE 1458 case BPF_FUNC_probe_read: 1459 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? 1460 NULL : &bpf_probe_read_compat_proto; 1461 case BPF_FUNC_probe_read_str: 1462 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? 1463 NULL : &bpf_probe_read_compat_str_proto; 1464 #endif 1465 #ifdef CONFIG_CGROUPS 1466 case BPF_FUNC_cgrp_storage_get: 1467 return &bpf_cgrp_storage_get_proto; 1468 case BPF_FUNC_cgrp_storage_delete: 1469 return &bpf_cgrp_storage_delete_proto; 1470 #endif 1471 case BPF_FUNC_send_signal: 1472 return &bpf_send_signal_proto; 1473 case BPF_FUNC_send_signal_thread: 1474 return &bpf_send_signal_thread_proto; 1475 case BPF_FUNC_perf_event_read_value: 1476 return &bpf_perf_event_read_value_proto; 1477 case BPF_FUNC_get_ns_current_pid_tgid: 1478 return &bpf_get_ns_current_pid_tgid_proto; 1479 case BPF_FUNC_ringbuf_output: 1480 return &bpf_ringbuf_output_proto; 1481 case BPF_FUNC_ringbuf_reserve: 1482 return &bpf_ringbuf_reserve_proto; 1483 case BPF_FUNC_ringbuf_submit: 1484 return &bpf_ringbuf_submit_proto; 1485 case BPF_FUNC_ringbuf_discard: 1486 return &bpf_ringbuf_discard_proto; 1487 case BPF_FUNC_ringbuf_query: 1488 return &bpf_ringbuf_query_proto; 1489 case BPF_FUNC_jiffies64: 1490 return &bpf_jiffies64_proto; 1491 case BPF_FUNC_get_task_stack: 1492 return &bpf_get_task_stack_proto; 1493 case BPF_FUNC_copy_from_user: 1494 return &bpf_copy_from_user_proto; 1495 case BPF_FUNC_copy_from_user_task: 1496 return &bpf_copy_from_user_task_proto; 1497 case BPF_FUNC_snprintf_btf: 1498 return &bpf_snprintf_btf_proto; 1499 case BPF_FUNC_per_cpu_ptr: 1500 return &bpf_per_cpu_ptr_proto; 1501 case BPF_FUNC_this_cpu_ptr: 1502 return &bpf_this_cpu_ptr_proto; 1503 case BPF_FUNC_task_storage_get: 1504 if (bpf_prog_check_recur(prog)) 1505 return &bpf_task_storage_get_recur_proto; 1506 return &bpf_task_storage_get_proto; 1507 case BPF_FUNC_task_storage_delete: 1508 if (bpf_prog_check_recur(prog)) 1509 return &bpf_task_storage_delete_recur_proto; 1510 return &bpf_task_storage_delete_proto; 1511 case BPF_FUNC_for_each_map_elem: 1512 return &bpf_for_each_map_elem_proto; 1513 case BPF_FUNC_snprintf: 1514 return &bpf_snprintf_proto; 1515 case BPF_FUNC_get_func_ip: 1516 return &bpf_get_func_ip_proto_tracing; 1517 case BPF_FUNC_get_branch_snapshot: 1518 return &bpf_get_branch_snapshot_proto; 1519 case BPF_FUNC_find_vma: 1520 return &bpf_find_vma_proto; 1521 case BPF_FUNC_trace_vprintk: 1522 return bpf_get_trace_vprintk_proto(); 1523 default: 1524 return bpf_base_func_proto(func_id); 1525 } 1526 } 1527 1528 static const struct bpf_func_proto * 1529 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1530 { 1531 switch (func_id) { 1532 case BPF_FUNC_perf_event_output: 1533 return &bpf_perf_event_output_proto; 1534 case BPF_FUNC_get_stackid: 1535 return &bpf_get_stackid_proto; 1536 case BPF_FUNC_get_stack: 1537 return &bpf_get_stack_proto; 1538 #ifdef CONFIG_BPF_KPROBE_OVERRIDE 1539 case BPF_FUNC_override_return: 1540 return &bpf_override_return_proto; 1541 #endif 1542 case BPF_FUNC_get_func_ip: 1543 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ? 1544 &bpf_get_func_ip_proto_kprobe_multi : 1545 &bpf_get_func_ip_proto_kprobe; 1546 case BPF_FUNC_get_attach_cookie: 1547 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ? 1548 &bpf_get_attach_cookie_proto_kmulti : 1549 &bpf_get_attach_cookie_proto_trace; 1550 default: 1551 return bpf_tracing_func_proto(func_id, prog); 1552 } 1553 } 1554 1555 /* bpf+kprobe programs can access fields of 'struct pt_regs' */ 1556 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 1557 const struct bpf_prog *prog, 1558 struct bpf_insn_access_aux *info) 1559 { 1560 if (off < 0 || off >= sizeof(struct pt_regs)) 1561 return false; 1562 if (type != BPF_READ) 1563 return false; 1564 if (off % size != 0) 1565 return false; 1566 /* 1567 * Assertion for 32 bit to make sure last 8 byte access 1568 * (BPF_DW) to the last 4 byte member is disallowed. 1569 */ 1570 if (off + size > sizeof(struct pt_regs)) 1571 return false; 1572 1573 return true; 1574 } 1575 1576 const struct bpf_verifier_ops kprobe_verifier_ops = { 1577 .get_func_proto = kprobe_prog_func_proto, 1578 .is_valid_access = kprobe_prog_is_valid_access, 1579 }; 1580 1581 const struct bpf_prog_ops kprobe_prog_ops = { 1582 }; 1583 1584 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, 1585 u64, flags, void *, data, u64, size) 1586 { 1587 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 1588 1589 /* 1590 * r1 points to perf tracepoint buffer where first 8 bytes are hidden 1591 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it 1592 * from there and call the same bpf_perf_event_output() helper inline. 1593 */ 1594 return ____bpf_perf_event_output(regs, map, flags, data, size); 1595 } 1596 1597 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { 1598 .func = bpf_perf_event_output_tp, 1599 .gpl_only = true, 1600 .ret_type = RET_INTEGER, 1601 .arg1_type = ARG_PTR_TO_CTX, 1602 .arg2_type = ARG_CONST_MAP_PTR, 1603 .arg3_type = ARG_ANYTHING, 1604 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1605 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 1606 }; 1607 1608 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, 1609 u64, flags) 1610 { 1611 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 1612 1613 /* 1614 * Same comment as in bpf_perf_event_output_tp(), only that this time 1615 * the other helper's function body cannot be inlined due to being 1616 * external, thus we need to call raw helper function. 1617 */ 1618 return bpf_get_stackid((unsigned long) regs, (unsigned long) map, 1619 flags, 0, 0); 1620 } 1621 1622 static const struct bpf_func_proto bpf_get_stackid_proto_tp = { 1623 .func = bpf_get_stackid_tp, 1624 .gpl_only = true, 1625 .ret_type = RET_INTEGER, 1626 .arg1_type = ARG_PTR_TO_CTX, 1627 .arg2_type = ARG_CONST_MAP_PTR, 1628 .arg3_type = ARG_ANYTHING, 1629 }; 1630 1631 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size, 1632 u64, flags) 1633 { 1634 struct pt_regs *regs = *(struct pt_regs **)tp_buff; 1635 1636 return bpf_get_stack((unsigned long) regs, (unsigned long) buf, 1637 (unsigned long) size, flags, 0); 1638 } 1639 1640 static const struct bpf_func_proto bpf_get_stack_proto_tp = { 1641 .func = bpf_get_stack_tp, 1642 .gpl_only = true, 1643 .ret_type = RET_INTEGER, 1644 .arg1_type = ARG_PTR_TO_CTX, 1645 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 1646 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 1647 .arg4_type = ARG_ANYTHING, 1648 }; 1649 1650 static const struct bpf_func_proto * 1651 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1652 { 1653 switch (func_id) { 1654 case BPF_FUNC_perf_event_output: 1655 return &bpf_perf_event_output_proto_tp; 1656 case BPF_FUNC_get_stackid: 1657 return &bpf_get_stackid_proto_tp; 1658 case BPF_FUNC_get_stack: 1659 return &bpf_get_stack_proto_tp; 1660 case BPF_FUNC_get_attach_cookie: 1661 return &bpf_get_attach_cookie_proto_trace; 1662 default: 1663 return bpf_tracing_func_proto(func_id, prog); 1664 } 1665 } 1666 1667 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, 1668 const struct bpf_prog *prog, 1669 struct bpf_insn_access_aux *info) 1670 { 1671 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) 1672 return false; 1673 if (type != BPF_READ) 1674 return false; 1675 if (off % size != 0) 1676 return false; 1677 1678 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); 1679 return true; 1680 } 1681 1682 const struct bpf_verifier_ops tracepoint_verifier_ops = { 1683 .get_func_proto = tp_prog_func_proto, 1684 .is_valid_access = tp_prog_is_valid_access, 1685 }; 1686 1687 const struct bpf_prog_ops tracepoint_prog_ops = { 1688 }; 1689 1690 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx, 1691 struct bpf_perf_event_value *, buf, u32, size) 1692 { 1693 int err = -EINVAL; 1694 1695 if (unlikely(size != sizeof(struct bpf_perf_event_value))) 1696 goto clear; 1697 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled, 1698 &buf->running); 1699 if (unlikely(err)) 1700 goto clear; 1701 return 0; 1702 clear: 1703 memset(buf, 0, size); 1704 return err; 1705 } 1706 1707 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = { 1708 .func = bpf_perf_prog_read_value, 1709 .gpl_only = true, 1710 .ret_type = RET_INTEGER, 1711 .arg1_type = ARG_PTR_TO_CTX, 1712 .arg2_type = ARG_PTR_TO_UNINIT_MEM, 1713 .arg3_type = ARG_CONST_SIZE, 1714 }; 1715 1716 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx, 1717 void *, buf, u32, size, u64, flags) 1718 { 1719 static const u32 br_entry_size = sizeof(struct perf_branch_entry); 1720 struct perf_branch_stack *br_stack = ctx->data->br_stack; 1721 u32 to_copy; 1722 1723 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE)) 1724 return -EINVAL; 1725 1726 if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK))) 1727 return -ENOENT; 1728 1729 if (unlikely(!br_stack)) 1730 return -ENOENT; 1731 1732 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE) 1733 return br_stack->nr * br_entry_size; 1734 1735 if (!buf || (size % br_entry_size != 0)) 1736 return -EINVAL; 1737 1738 to_copy = min_t(u32, br_stack->nr * br_entry_size, size); 1739 memcpy(buf, br_stack->entries, to_copy); 1740 1741 return to_copy; 1742 } 1743 1744 static const struct bpf_func_proto bpf_read_branch_records_proto = { 1745 .func = bpf_read_branch_records, 1746 .gpl_only = true, 1747 .ret_type = RET_INTEGER, 1748 .arg1_type = ARG_PTR_TO_CTX, 1749 .arg2_type = ARG_PTR_TO_MEM_OR_NULL, 1750 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 1751 .arg4_type = ARG_ANYTHING, 1752 }; 1753 1754 static const struct bpf_func_proto * 1755 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1756 { 1757 switch (func_id) { 1758 case BPF_FUNC_perf_event_output: 1759 return &bpf_perf_event_output_proto_tp; 1760 case BPF_FUNC_get_stackid: 1761 return &bpf_get_stackid_proto_pe; 1762 case BPF_FUNC_get_stack: 1763 return &bpf_get_stack_proto_pe; 1764 case BPF_FUNC_perf_prog_read_value: 1765 return &bpf_perf_prog_read_value_proto; 1766 case BPF_FUNC_read_branch_records: 1767 return &bpf_read_branch_records_proto; 1768 case BPF_FUNC_get_attach_cookie: 1769 return &bpf_get_attach_cookie_proto_pe; 1770 default: 1771 return bpf_tracing_func_proto(func_id, prog); 1772 } 1773 } 1774 1775 /* 1776 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp 1777 * to avoid potential recursive reuse issue when/if tracepoints are added 1778 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack. 1779 * 1780 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage 1781 * in normal, irq, and nmi context. 1782 */ 1783 struct bpf_raw_tp_regs { 1784 struct pt_regs regs[3]; 1785 }; 1786 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs); 1787 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level); 1788 static struct pt_regs *get_bpf_raw_tp_regs(void) 1789 { 1790 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs); 1791 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level); 1792 1793 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) { 1794 this_cpu_dec(bpf_raw_tp_nest_level); 1795 return ERR_PTR(-EBUSY); 1796 } 1797 1798 return &tp_regs->regs[nest_level - 1]; 1799 } 1800 1801 static void put_bpf_raw_tp_regs(void) 1802 { 1803 this_cpu_dec(bpf_raw_tp_nest_level); 1804 } 1805 1806 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args, 1807 struct bpf_map *, map, u64, flags, void *, data, u64, size) 1808 { 1809 struct pt_regs *regs = get_bpf_raw_tp_regs(); 1810 int ret; 1811 1812 if (IS_ERR(regs)) 1813 return PTR_ERR(regs); 1814 1815 perf_fetch_caller_regs(regs); 1816 ret = ____bpf_perf_event_output(regs, map, flags, data, size); 1817 1818 put_bpf_raw_tp_regs(); 1819 return ret; 1820 } 1821 1822 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = { 1823 .func = bpf_perf_event_output_raw_tp, 1824 .gpl_only = true, 1825 .ret_type = RET_INTEGER, 1826 .arg1_type = ARG_PTR_TO_CTX, 1827 .arg2_type = ARG_CONST_MAP_PTR, 1828 .arg3_type = ARG_ANYTHING, 1829 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1830 .arg5_type = ARG_CONST_SIZE_OR_ZERO, 1831 }; 1832 1833 extern const struct bpf_func_proto bpf_skb_output_proto; 1834 extern const struct bpf_func_proto bpf_xdp_output_proto; 1835 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto; 1836 1837 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args, 1838 struct bpf_map *, map, u64, flags) 1839 { 1840 struct pt_regs *regs = get_bpf_raw_tp_regs(); 1841 int ret; 1842 1843 if (IS_ERR(regs)) 1844 return PTR_ERR(regs); 1845 1846 perf_fetch_caller_regs(regs); 1847 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */ 1848 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map, 1849 flags, 0, 0); 1850 put_bpf_raw_tp_regs(); 1851 return ret; 1852 } 1853 1854 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = { 1855 .func = bpf_get_stackid_raw_tp, 1856 .gpl_only = true, 1857 .ret_type = RET_INTEGER, 1858 .arg1_type = ARG_PTR_TO_CTX, 1859 .arg2_type = ARG_CONST_MAP_PTR, 1860 .arg3_type = ARG_ANYTHING, 1861 }; 1862 1863 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args, 1864 void *, buf, u32, size, u64, flags) 1865 { 1866 struct pt_regs *regs = get_bpf_raw_tp_regs(); 1867 int ret; 1868 1869 if (IS_ERR(regs)) 1870 return PTR_ERR(regs); 1871 1872 perf_fetch_caller_regs(regs); 1873 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf, 1874 (unsigned long) size, flags, 0); 1875 put_bpf_raw_tp_regs(); 1876 return ret; 1877 } 1878 1879 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = { 1880 .func = bpf_get_stack_raw_tp, 1881 .gpl_only = true, 1882 .ret_type = RET_INTEGER, 1883 .arg1_type = ARG_PTR_TO_CTX, 1884 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY, 1885 .arg3_type = ARG_CONST_SIZE_OR_ZERO, 1886 .arg4_type = ARG_ANYTHING, 1887 }; 1888 1889 static const struct bpf_func_proto * 1890 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1891 { 1892 switch (func_id) { 1893 case BPF_FUNC_perf_event_output: 1894 return &bpf_perf_event_output_proto_raw_tp; 1895 case BPF_FUNC_get_stackid: 1896 return &bpf_get_stackid_proto_raw_tp; 1897 case BPF_FUNC_get_stack: 1898 return &bpf_get_stack_proto_raw_tp; 1899 default: 1900 return bpf_tracing_func_proto(func_id, prog); 1901 } 1902 } 1903 1904 const struct bpf_func_proto * 1905 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) 1906 { 1907 const struct bpf_func_proto *fn; 1908 1909 switch (func_id) { 1910 #ifdef CONFIG_NET 1911 case BPF_FUNC_skb_output: 1912 return &bpf_skb_output_proto; 1913 case BPF_FUNC_xdp_output: 1914 return &bpf_xdp_output_proto; 1915 case BPF_FUNC_skc_to_tcp6_sock: 1916 return &bpf_skc_to_tcp6_sock_proto; 1917 case BPF_FUNC_skc_to_tcp_sock: 1918 return &bpf_skc_to_tcp_sock_proto; 1919 case BPF_FUNC_skc_to_tcp_timewait_sock: 1920 return &bpf_skc_to_tcp_timewait_sock_proto; 1921 case BPF_FUNC_skc_to_tcp_request_sock: 1922 return &bpf_skc_to_tcp_request_sock_proto; 1923 case BPF_FUNC_skc_to_udp6_sock: 1924 return &bpf_skc_to_udp6_sock_proto; 1925 case BPF_FUNC_skc_to_unix_sock: 1926 return &bpf_skc_to_unix_sock_proto; 1927 case BPF_FUNC_skc_to_mptcp_sock: 1928 return &bpf_skc_to_mptcp_sock_proto; 1929 case BPF_FUNC_sk_storage_get: 1930 return &bpf_sk_storage_get_tracing_proto; 1931 case BPF_FUNC_sk_storage_delete: 1932 return &bpf_sk_storage_delete_tracing_proto; 1933 case BPF_FUNC_sock_from_file: 1934 return &bpf_sock_from_file_proto; 1935 case BPF_FUNC_get_socket_cookie: 1936 return &bpf_get_socket_ptr_cookie_proto; 1937 case BPF_FUNC_xdp_get_buff_len: 1938 return &bpf_xdp_get_buff_len_trace_proto; 1939 #endif 1940 case BPF_FUNC_seq_printf: 1941 return prog->expected_attach_type == BPF_TRACE_ITER ? 1942 &bpf_seq_printf_proto : 1943 NULL; 1944 case BPF_FUNC_seq_write: 1945 return prog->expected_attach_type == BPF_TRACE_ITER ? 1946 &bpf_seq_write_proto : 1947 NULL; 1948 case BPF_FUNC_seq_printf_btf: 1949 return prog->expected_attach_type == BPF_TRACE_ITER ? 1950 &bpf_seq_printf_btf_proto : 1951 NULL; 1952 case BPF_FUNC_d_path: 1953 return &bpf_d_path_proto; 1954 case BPF_FUNC_get_func_arg: 1955 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL; 1956 case BPF_FUNC_get_func_ret: 1957 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL; 1958 case BPF_FUNC_get_func_arg_cnt: 1959 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL; 1960 case BPF_FUNC_get_attach_cookie: 1961 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL; 1962 default: 1963 fn = raw_tp_prog_func_proto(func_id, prog); 1964 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER) 1965 fn = bpf_iter_get_func_proto(func_id, prog); 1966 return fn; 1967 } 1968 } 1969 1970 static bool raw_tp_prog_is_valid_access(int off, int size, 1971 enum bpf_access_type type, 1972 const struct bpf_prog *prog, 1973 struct bpf_insn_access_aux *info) 1974 { 1975 return bpf_tracing_ctx_access(off, size, type); 1976 } 1977 1978 static bool tracing_prog_is_valid_access(int off, int size, 1979 enum bpf_access_type type, 1980 const struct bpf_prog *prog, 1981 struct bpf_insn_access_aux *info) 1982 { 1983 return bpf_tracing_btf_ctx_access(off, size, type, prog, info); 1984 } 1985 1986 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog, 1987 const union bpf_attr *kattr, 1988 union bpf_attr __user *uattr) 1989 { 1990 return -ENOTSUPP; 1991 } 1992 1993 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = { 1994 .get_func_proto = raw_tp_prog_func_proto, 1995 .is_valid_access = raw_tp_prog_is_valid_access, 1996 }; 1997 1998 const struct bpf_prog_ops raw_tracepoint_prog_ops = { 1999 #ifdef CONFIG_NET 2000 .test_run = bpf_prog_test_run_raw_tp, 2001 #endif 2002 }; 2003 2004 const struct bpf_verifier_ops tracing_verifier_ops = { 2005 .get_func_proto = tracing_prog_func_proto, 2006 .is_valid_access = tracing_prog_is_valid_access, 2007 }; 2008 2009 const struct bpf_prog_ops tracing_prog_ops = { 2010 .test_run = bpf_prog_test_run_tracing, 2011 }; 2012 2013 static bool raw_tp_writable_prog_is_valid_access(int off, int size, 2014 enum bpf_access_type type, 2015 const struct bpf_prog *prog, 2016 struct bpf_insn_access_aux *info) 2017 { 2018 if (off == 0) { 2019 if (size != sizeof(u64) || type != BPF_READ) 2020 return false; 2021 info->reg_type = PTR_TO_TP_BUFFER; 2022 } 2023 return raw_tp_prog_is_valid_access(off, size, type, prog, info); 2024 } 2025 2026 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = { 2027 .get_func_proto = raw_tp_prog_func_proto, 2028 .is_valid_access = raw_tp_writable_prog_is_valid_access, 2029 }; 2030 2031 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = { 2032 }; 2033 2034 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, 2035 const struct bpf_prog *prog, 2036 struct bpf_insn_access_aux *info) 2037 { 2038 const int size_u64 = sizeof(u64); 2039 2040 if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) 2041 return false; 2042 if (type != BPF_READ) 2043 return false; 2044 if (off % size != 0) { 2045 if (sizeof(unsigned long) != 4) 2046 return false; 2047 if (size != 8) 2048 return false; 2049 if (off % size != 4) 2050 return false; 2051 } 2052 2053 switch (off) { 2054 case bpf_ctx_range(struct bpf_perf_event_data, sample_period): 2055 bpf_ctx_record_field_size(info, size_u64); 2056 if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) 2057 return false; 2058 break; 2059 case bpf_ctx_range(struct bpf_perf_event_data, addr): 2060 bpf_ctx_record_field_size(info, size_u64); 2061 if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) 2062 return false; 2063 break; 2064 default: 2065 if (size != sizeof(long)) 2066 return false; 2067 } 2068 2069 return true; 2070 } 2071 2072 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, 2073 const struct bpf_insn *si, 2074 struct bpf_insn *insn_buf, 2075 struct bpf_prog *prog, u32 *target_size) 2076 { 2077 struct bpf_insn *insn = insn_buf; 2078 2079 switch (si->off) { 2080 case offsetof(struct bpf_perf_event_data, sample_period): 2081 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 2082 data), si->dst_reg, si->src_reg, 2083 offsetof(struct bpf_perf_event_data_kern, data)); 2084 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 2085 bpf_target_off(struct perf_sample_data, period, 8, 2086 target_size)); 2087 break; 2088 case offsetof(struct bpf_perf_event_data, addr): 2089 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 2090 data), si->dst_reg, si->src_reg, 2091 offsetof(struct bpf_perf_event_data_kern, data)); 2092 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, 2093 bpf_target_off(struct perf_sample_data, addr, 8, 2094 target_size)); 2095 break; 2096 default: 2097 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, 2098 regs), si->dst_reg, si->src_reg, 2099 offsetof(struct bpf_perf_event_data_kern, regs)); 2100 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, 2101 si->off); 2102 break; 2103 } 2104 2105 return insn - insn_buf; 2106 } 2107 2108 const struct bpf_verifier_ops perf_event_verifier_ops = { 2109 .get_func_proto = pe_prog_func_proto, 2110 .is_valid_access = pe_prog_is_valid_access, 2111 .convert_ctx_access = pe_prog_convert_ctx_access, 2112 }; 2113 2114 const struct bpf_prog_ops perf_event_prog_ops = { 2115 }; 2116 2117 static DEFINE_MUTEX(bpf_event_mutex); 2118 2119 #define BPF_TRACE_MAX_PROGS 64 2120 2121 int perf_event_attach_bpf_prog(struct perf_event *event, 2122 struct bpf_prog *prog, 2123 u64 bpf_cookie) 2124 { 2125 struct bpf_prog_array *old_array; 2126 struct bpf_prog_array *new_array; 2127 int ret = -EEXIST; 2128 2129 /* 2130 * Kprobe override only works if they are on the function entry, 2131 * and only if they are on the opt-in list. 2132 */ 2133 if (prog->kprobe_override && 2134 (!trace_kprobe_on_func_entry(event->tp_event) || 2135 !trace_kprobe_error_injectable(event->tp_event))) 2136 return -EINVAL; 2137 2138 mutex_lock(&bpf_event_mutex); 2139 2140 if (event->prog) 2141 goto unlock; 2142 2143 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array); 2144 if (old_array && 2145 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) { 2146 ret = -E2BIG; 2147 goto unlock; 2148 } 2149 2150 ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array); 2151 if (ret < 0) 2152 goto unlock; 2153 2154 /* set the new array to event->tp_event and set event->prog */ 2155 event->prog = prog; 2156 event->bpf_cookie = bpf_cookie; 2157 rcu_assign_pointer(event->tp_event->prog_array, new_array); 2158 bpf_prog_array_free_sleepable(old_array); 2159 2160 unlock: 2161 mutex_unlock(&bpf_event_mutex); 2162 return ret; 2163 } 2164 2165 void perf_event_detach_bpf_prog(struct perf_event *event) 2166 { 2167 struct bpf_prog_array *old_array; 2168 struct bpf_prog_array *new_array; 2169 int ret; 2170 2171 mutex_lock(&bpf_event_mutex); 2172 2173 if (!event->prog) 2174 goto unlock; 2175 2176 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array); 2177 ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array); 2178 if (ret == -ENOENT) 2179 goto unlock; 2180 if (ret < 0) { 2181 bpf_prog_array_delete_safe(old_array, event->prog); 2182 } else { 2183 rcu_assign_pointer(event->tp_event->prog_array, new_array); 2184 bpf_prog_array_free_sleepable(old_array); 2185 } 2186 2187 bpf_prog_put(event->prog); 2188 event->prog = NULL; 2189 2190 unlock: 2191 mutex_unlock(&bpf_event_mutex); 2192 } 2193 2194 int perf_event_query_prog_array(struct perf_event *event, void __user *info) 2195 { 2196 struct perf_event_query_bpf __user *uquery = info; 2197 struct perf_event_query_bpf query = {}; 2198 struct bpf_prog_array *progs; 2199 u32 *ids, prog_cnt, ids_len; 2200 int ret; 2201 2202 if (!perfmon_capable()) 2203 return -EPERM; 2204 if (event->attr.type != PERF_TYPE_TRACEPOINT) 2205 return -EINVAL; 2206 if (copy_from_user(&query, uquery, sizeof(query))) 2207 return -EFAULT; 2208 2209 ids_len = query.ids_len; 2210 if (ids_len > BPF_TRACE_MAX_PROGS) 2211 return -E2BIG; 2212 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN); 2213 if (!ids) 2214 return -ENOMEM; 2215 /* 2216 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which 2217 * is required when user only wants to check for uquery->prog_cnt. 2218 * There is no need to check for it since the case is handled 2219 * gracefully in bpf_prog_array_copy_info. 2220 */ 2221 2222 mutex_lock(&bpf_event_mutex); 2223 progs = bpf_event_rcu_dereference(event->tp_event->prog_array); 2224 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt); 2225 mutex_unlock(&bpf_event_mutex); 2226 2227 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) || 2228 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32))) 2229 ret = -EFAULT; 2230 2231 kfree(ids); 2232 return ret; 2233 } 2234 2235 extern struct bpf_raw_event_map __start__bpf_raw_tp[]; 2236 extern struct bpf_raw_event_map __stop__bpf_raw_tp[]; 2237 2238 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name) 2239 { 2240 struct bpf_raw_event_map *btp = __start__bpf_raw_tp; 2241 2242 for (; btp < __stop__bpf_raw_tp; btp++) { 2243 if (!strcmp(btp->tp->name, name)) 2244 return btp; 2245 } 2246 2247 return bpf_get_raw_tracepoint_module(name); 2248 } 2249 2250 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp) 2251 { 2252 struct module *mod; 2253 2254 preempt_disable(); 2255 mod = __module_address((unsigned long)btp); 2256 module_put(mod); 2257 preempt_enable(); 2258 } 2259 2260 static __always_inline 2261 void __bpf_trace_run(struct bpf_prog *prog, u64 *args) 2262 { 2263 cant_sleep(); 2264 if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) { 2265 bpf_prog_inc_misses_counter(prog); 2266 goto out; 2267 } 2268 rcu_read_lock(); 2269 (void) bpf_prog_run(prog, args); 2270 rcu_read_unlock(); 2271 out: 2272 this_cpu_dec(*(prog->active)); 2273 } 2274 2275 #define UNPACK(...) __VA_ARGS__ 2276 #define REPEAT_1(FN, DL, X, ...) FN(X) 2277 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__) 2278 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__) 2279 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__) 2280 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__) 2281 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__) 2282 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__) 2283 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__) 2284 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__) 2285 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__) 2286 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__) 2287 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__) 2288 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__) 2289 2290 #define SARG(X) u64 arg##X 2291 #define COPY(X) args[X] = arg##X 2292 2293 #define __DL_COM (,) 2294 #define __DL_SEM (;) 2295 2296 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 2297 2298 #define BPF_TRACE_DEFN_x(x) \ 2299 void bpf_trace_run##x(struct bpf_prog *prog, \ 2300 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \ 2301 { \ 2302 u64 args[x]; \ 2303 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \ 2304 __bpf_trace_run(prog, args); \ 2305 } \ 2306 EXPORT_SYMBOL_GPL(bpf_trace_run##x) 2307 BPF_TRACE_DEFN_x(1); 2308 BPF_TRACE_DEFN_x(2); 2309 BPF_TRACE_DEFN_x(3); 2310 BPF_TRACE_DEFN_x(4); 2311 BPF_TRACE_DEFN_x(5); 2312 BPF_TRACE_DEFN_x(6); 2313 BPF_TRACE_DEFN_x(7); 2314 BPF_TRACE_DEFN_x(8); 2315 BPF_TRACE_DEFN_x(9); 2316 BPF_TRACE_DEFN_x(10); 2317 BPF_TRACE_DEFN_x(11); 2318 BPF_TRACE_DEFN_x(12); 2319 2320 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) 2321 { 2322 struct tracepoint *tp = btp->tp; 2323 2324 /* 2325 * check that program doesn't access arguments beyond what's 2326 * available in this tracepoint 2327 */ 2328 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64)) 2329 return -EINVAL; 2330 2331 if (prog->aux->max_tp_access > btp->writable_size) 2332 return -EINVAL; 2333 2334 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func, 2335 prog); 2336 } 2337 2338 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) 2339 { 2340 return __bpf_probe_register(btp, prog); 2341 } 2342 2343 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog) 2344 { 2345 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog); 2346 } 2347 2348 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id, 2349 u32 *fd_type, const char **buf, 2350 u64 *probe_offset, u64 *probe_addr) 2351 { 2352 bool is_tracepoint, is_syscall_tp; 2353 struct bpf_prog *prog; 2354 int flags, err = 0; 2355 2356 prog = event->prog; 2357 if (!prog) 2358 return -ENOENT; 2359 2360 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */ 2361 if (prog->type == BPF_PROG_TYPE_PERF_EVENT) 2362 return -EOPNOTSUPP; 2363 2364 *prog_id = prog->aux->id; 2365 flags = event->tp_event->flags; 2366 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT; 2367 is_syscall_tp = is_syscall_trace_event(event->tp_event); 2368 2369 if (is_tracepoint || is_syscall_tp) { 2370 *buf = is_tracepoint ? event->tp_event->tp->name 2371 : event->tp_event->name; 2372 *fd_type = BPF_FD_TYPE_TRACEPOINT; 2373 *probe_offset = 0x0; 2374 *probe_addr = 0x0; 2375 } else { 2376 /* kprobe/uprobe */ 2377 err = -EOPNOTSUPP; 2378 #ifdef CONFIG_KPROBE_EVENTS 2379 if (flags & TRACE_EVENT_FL_KPROBE) 2380 err = bpf_get_kprobe_info(event, fd_type, buf, 2381 probe_offset, probe_addr, 2382 event->attr.type == PERF_TYPE_TRACEPOINT); 2383 #endif 2384 #ifdef CONFIG_UPROBE_EVENTS 2385 if (flags & TRACE_EVENT_FL_UPROBE) 2386 err = bpf_get_uprobe_info(event, fd_type, buf, 2387 probe_offset, 2388 event->attr.type == PERF_TYPE_TRACEPOINT); 2389 #endif 2390 } 2391 2392 return err; 2393 } 2394 2395 static int __init send_signal_irq_work_init(void) 2396 { 2397 int cpu; 2398 struct send_signal_irq_work *work; 2399 2400 for_each_possible_cpu(cpu) { 2401 work = per_cpu_ptr(&send_signal_work, cpu); 2402 init_irq_work(&work->irq_work, do_bpf_send_signal); 2403 } 2404 return 0; 2405 } 2406 2407 subsys_initcall(send_signal_irq_work_init); 2408 2409 #ifdef CONFIG_MODULES 2410 static int bpf_event_notify(struct notifier_block *nb, unsigned long op, 2411 void *module) 2412 { 2413 struct bpf_trace_module *btm, *tmp; 2414 struct module *mod = module; 2415 int ret = 0; 2416 2417 if (mod->num_bpf_raw_events == 0 || 2418 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING)) 2419 goto out; 2420 2421 mutex_lock(&bpf_module_mutex); 2422 2423 switch (op) { 2424 case MODULE_STATE_COMING: 2425 btm = kzalloc(sizeof(*btm), GFP_KERNEL); 2426 if (btm) { 2427 btm->module = module; 2428 list_add(&btm->list, &bpf_trace_modules); 2429 } else { 2430 ret = -ENOMEM; 2431 } 2432 break; 2433 case MODULE_STATE_GOING: 2434 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) { 2435 if (btm->module == module) { 2436 list_del(&btm->list); 2437 kfree(btm); 2438 break; 2439 } 2440 } 2441 break; 2442 } 2443 2444 mutex_unlock(&bpf_module_mutex); 2445 2446 out: 2447 return notifier_from_errno(ret); 2448 } 2449 2450 static struct notifier_block bpf_module_nb = { 2451 .notifier_call = bpf_event_notify, 2452 }; 2453 2454 static int __init bpf_event_init(void) 2455 { 2456 register_module_notifier(&bpf_module_nb); 2457 return 0; 2458 } 2459 2460 fs_initcall(bpf_event_init); 2461 #endif /* CONFIG_MODULES */ 2462 2463 #ifdef CONFIG_FPROBE 2464 struct bpf_kprobe_multi_link { 2465 struct bpf_link link; 2466 struct fprobe fp; 2467 unsigned long *addrs; 2468 u64 *cookies; 2469 u32 cnt; 2470 u32 mods_cnt; 2471 struct module **mods; 2472 }; 2473 2474 struct bpf_kprobe_multi_run_ctx { 2475 struct bpf_run_ctx run_ctx; 2476 struct bpf_kprobe_multi_link *link; 2477 unsigned long entry_ip; 2478 }; 2479 2480 struct user_syms { 2481 const char **syms; 2482 char *buf; 2483 }; 2484 2485 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt) 2486 { 2487 unsigned long __user usymbol; 2488 const char **syms = NULL; 2489 char *buf = NULL, *p; 2490 int err = -ENOMEM; 2491 unsigned int i; 2492 2493 syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL); 2494 if (!syms) 2495 goto error; 2496 2497 buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL); 2498 if (!buf) 2499 goto error; 2500 2501 for (p = buf, i = 0; i < cnt; i++) { 2502 if (__get_user(usymbol, usyms + i)) { 2503 err = -EFAULT; 2504 goto error; 2505 } 2506 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN); 2507 if (err == KSYM_NAME_LEN) 2508 err = -E2BIG; 2509 if (err < 0) 2510 goto error; 2511 syms[i] = p; 2512 p += err + 1; 2513 } 2514 2515 us->syms = syms; 2516 us->buf = buf; 2517 return 0; 2518 2519 error: 2520 if (err) { 2521 kvfree(syms); 2522 kvfree(buf); 2523 } 2524 return err; 2525 } 2526 2527 static void kprobe_multi_put_modules(struct module **mods, u32 cnt) 2528 { 2529 u32 i; 2530 2531 for (i = 0; i < cnt; i++) 2532 module_put(mods[i]); 2533 } 2534 2535 static void free_user_syms(struct user_syms *us) 2536 { 2537 kvfree(us->syms); 2538 kvfree(us->buf); 2539 } 2540 2541 static void bpf_kprobe_multi_link_release(struct bpf_link *link) 2542 { 2543 struct bpf_kprobe_multi_link *kmulti_link; 2544 2545 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link); 2546 unregister_fprobe(&kmulti_link->fp); 2547 kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt); 2548 } 2549 2550 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link) 2551 { 2552 struct bpf_kprobe_multi_link *kmulti_link; 2553 2554 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link); 2555 kvfree(kmulti_link->addrs); 2556 kvfree(kmulti_link->cookies); 2557 kfree(kmulti_link->mods); 2558 kfree(kmulti_link); 2559 } 2560 2561 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = { 2562 .release = bpf_kprobe_multi_link_release, 2563 .dealloc = bpf_kprobe_multi_link_dealloc, 2564 }; 2565 2566 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv) 2567 { 2568 const struct bpf_kprobe_multi_link *link = priv; 2569 unsigned long *addr_a = a, *addr_b = b; 2570 u64 *cookie_a, *cookie_b; 2571 2572 cookie_a = link->cookies + (addr_a - link->addrs); 2573 cookie_b = link->cookies + (addr_b - link->addrs); 2574 2575 /* swap addr_a/addr_b and cookie_a/cookie_b values */ 2576 swap(*addr_a, *addr_b); 2577 swap(*cookie_a, *cookie_b); 2578 } 2579 2580 static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b) 2581 { 2582 const unsigned long *addr_a = a, *addr_b = b; 2583 2584 if (*addr_a == *addr_b) 2585 return 0; 2586 return *addr_a < *addr_b ? -1 : 1; 2587 } 2588 2589 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv) 2590 { 2591 return bpf_kprobe_multi_addrs_cmp(a, b); 2592 } 2593 2594 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx) 2595 { 2596 struct bpf_kprobe_multi_run_ctx *run_ctx; 2597 struct bpf_kprobe_multi_link *link; 2598 u64 *cookie, entry_ip; 2599 unsigned long *addr; 2600 2601 if (WARN_ON_ONCE(!ctx)) 2602 return 0; 2603 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx); 2604 link = run_ctx->link; 2605 if (!link->cookies) 2606 return 0; 2607 entry_ip = run_ctx->entry_ip; 2608 addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip), 2609 bpf_kprobe_multi_addrs_cmp); 2610 if (!addr) 2611 return 0; 2612 cookie = link->cookies + (addr - link->addrs); 2613 return *cookie; 2614 } 2615 2616 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx) 2617 { 2618 struct bpf_kprobe_multi_run_ctx *run_ctx; 2619 2620 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx); 2621 return run_ctx->entry_ip; 2622 } 2623 2624 static int 2625 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link, 2626 unsigned long entry_ip, struct pt_regs *regs) 2627 { 2628 struct bpf_kprobe_multi_run_ctx run_ctx = { 2629 .link = link, 2630 .entry_ip = entry_ip, 2631 }; 2632 struct bpf_run_ctx *old_run_ctx; 2633 int err; 2634 2635 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { 2636 err = 0; 2637 goto out; 2638 } 2639 2640 migrate_disable(); 2641 rcu_read_lock(); 2642 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx); 2643 err = bpf_prog_run(link->link.prog, regs); 2644 bpf_reset_run_ctx(old_run_ctx); 2645 rcu_read_unlock(); 2646 migrate_enable(); 2647 2648 out: 2649 __this_cpu_dec(bpf_prog_active); 2650 return err; 2651 } 2652 2653 static int 2654 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip, 2655 unsigned long ret_ip, struct pt_regs *regs, 2656 void *data) 2657 { 2658 struct bpf_kprobe_multi_link *link; 2659 2660 link = container_of(fp, struct bpf_kprobe_multi_link, fp); 2661 kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs); 2662 return 0; 2663 } 2664 2665 static void 2666 kprobe_multi_link_exit_handler(struct fprobe *fp, unsigned long fentry_ip, 2667 unsigned long ret_ip, struct pt_regs *regs, 2668 void *data) 2669 { 2670 struct bpf_kprobe_multi_link *link; 2671 2672 link = container_of(fp, struct bpf_kprobe_multi_link, fp); 2673 kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs); 2674 } 2675 2676 static int symbols_cmp_r(const void *a, const void *b, const void *priv) 2677 { 2678 const char **str_a = (const char **) a; 2679 const char **str_b = (const char **) b; 2680 2681 return strcmp(*str_a, *str_b); 2682 } 2683 2684 struct multi_symbols_sort { 2685 const char **funcs; 2686 u64 *cookies; 2687 }; 2688 2689 static void symbols_swap_r(void *a, void *b, int size, const void *priv) 2690 { 2691 const struct multi_symbols_sort *data = priv; 2692 const char **name_a = a, **name_b = b; 2693 2694 swap(*name_a, *name_b); 2695 2696 /* If defined, swap also related cookies. */ 2697 if (data->cookies) { 2698 u64 *cookie_a, *cookie_b; 2699 2700 cookie_a = data->cookies + (name_a - data->funcs); 2701 cookie_b = data->cookies + (name_b - data->funcs); 2702 swap(*cookie_a, *cookie_b); 2703 } 2704 } 2705 2706 struct modules_array { 2707 struct module **mods; 2708 int mods_cnt; 2709 int mods_cap; 2710 }; 2711 2712 static int add_module(struct modules_array *arr, struct module *mod) 2713 { 2714 struct module **mods; 2715 2716 if (arr->mods_cnt == arr->mods_cap) { 2717 arr->mods_cap = max(16, arr->mods_cap * 3 / 2); 2718 mods = krealloc_array(arr->mods, arr->mods_cap, sizeof(*mods), GFP_KERNEL); 2719 if (!mods) 2720 return -ENOMEM; 2721 arr->mods = mods; 2722 } 2723 2724 arr->mods[arr->mods_cnt] = mod; 2725 arr->mods_cnt++; 2726 return 0; 2727 } 2728 2729 static bool has_module(struct modules_array *arr, struct module *mod) 2730 { 2731 int i; 2732 2733 for (i = arr->mods_cnt - 1; i >= 0; i--) { 2734 if (arr->mods[i] == mod) 2735 return true; 2736 } 2737 return false; 2738 } 2739 2740 static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt) 2741 { 2742 struct modules_array arr = {}; 2743 u32 i, err = 0; 2744 2745 for (i = 0; i < addrs_cnt; i++) { 2746 struct module *mod; 2747 2748 preempt_disable(); 2749 mod = __module_address(addrs[i]); 2750 /* Either no module or we it's already stored */ 2751 if (!mod || has_module(&arr, mod)) { 2752 preempt_enable(); 2753 continue; 2754 } 2755 if (!try_module_get(mod)) 2756 err = -EINVAL; 2757 preempt_enable(); 2758 if (err) 2759 break; 2760 err = add_module(&arr, mod); 2761 if (err) { 2762 module_put(mod); 2763 break; 2764 } 2765 } 2766 2767 /* We return either err < 0 in case of error, ... */ 2768 if (err) { 2769 kprobe_multi_put_modules(arr.mods, arr.mods_cnt); 2770 kfree(arr.mods); 2771 return err; 2772 } 2773 2774 /* or number of modules found if everything is ok. */ 2775 *mods = arr.mods; 2776 return arr.mods_cnt; 2777 } 2778 2779 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 2780 { 2781 struct bpf_kprobe_multi_link *link = NULL; 2782 struct bpf_link_primer link_primer; 2783 void __user *ucookies; 2784 unsigned long *addrs; 2785 u32 flags, cnt, size; 2786 void __user *uaddrs; 2787 u64 *cookies = NULL; 2788 void __user *usyms; 2789 int err; 2790 2791 /* no support for 32bit archs yet */ 2792 if (sizeof(u64) != sizeof(void *)) 2793 return -EOPNOTSUPP; 2794 2795 if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI) 2796 return -EINVAL; 2797 2798 flags = attr->link_create.kprobe_multi.flags; 2799 if (flags & ~BPF_F_KPROBE_MULTI_RETURN) 2800 return -EINVAL; 2801 2802 uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs); 2803 usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms); 2804 if (!!uaddrs == !!usyms) 2805 return -EINVAL; 2806 2807 cnt = attr->link_create.kprobe_multi.cnt; 2808 if (!cnt) 2809 return -EINVAL; 2810 2811 size = cnt * sizeof(*addrs); 2812 addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL); 2813 if (!addrs) 2814 return -ENOMEM; 2815 2816 ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies); 2817 if (ucookies) { 2818 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL); 2819 if (!cookies) { 2820 err = -ENOMEM; 2821 goto error; 2822 } 2823 if (copy_from_user(cookies, ucookies, size)) { 2824 err = -EFAULT; 2825 goto error; 2826 } 2827 } 2828 2829 if (uaddrs) { 2830 if (copy_from_user(addrs, uaddrs, size)) { 2831 err = -EFAULT; 2832 goto error; 2833 } 2834 } else { 2835 struct multi_symbols_sort data = { 2836 .cookies = cookies, 2837 }; 2838 struct user_syms us; 2839 2840 err = copy_user_syms(&us, usyms, cnt); 2841 if (err) 2842 goto error; 2843 2844 if (cookies) 2845 data.funcs = us.syms; 2846 2847 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r, 2848 symbols_swap_r, &data); 2849 2850 err = ftrace_lookup_symbols(us.syms, cnt, addrs); 2851 free_user_syms(&us); 2852 if (err) 2853 goto error; 2854 } 2855 2856 link = kzalloc(sizeof(*link), GFP_KERNEL); 2857 if (!link) { 2858 err = -ENOMEM; 2859 goto error; 2860 } 2861 2862 bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI, 2863 &bpf_kprobe_multi_link_lops, prog); 2864 2865 err = bpf_link_prime(&link->link, &link_primer); 2866 if (err) 2867 goto error; 2868 2869 if (flags & BPF_F_KPROBE_MULTI_RETURN) 2870 link->fp.exit_handler = kprobe_multi_link_exit_handler; 2871 else 2872 link->fp.entry_handler = kprobe_multi_link_handler; 2873 2874 link->addrs = addrs; 2875 link->cookies = cookies; 2876 link->cnt = cnt; 2877 2878 if (cookies) { 2879 /* 2880 * Sorting addresses will trigger sorting cookies as well 2881 * (check bpf_kprobe_multi_cookie_swap). This way we can 2882 * find cookie based on the address in bpf_get_attach_cookie 2883 * helper. 2884 */ 2885 sort_r(addrs, cnt, sizeof(*addrs), 2886 bpf_kprobe_multi_cookie_cmp, 2887 bpf_kprobe_multi_cookie_swap, 2888 link); 2889 } 2890 2891 err = get_modules_for_addrs(&link->mods, addrs, cnt); 2892 if (err < 0) { 2893 bpf_link_cleanup(&link_primer); 2894 return err; 2895 } 2896 link->mods_cnt = err; 2897 2898 err = register_fprobe_ips(&link->fp, addrs, cnt); 2899 if (err) { 2900 kprobe_multi_put_modules(link->mods, link->mods_cnt); 2901 bpf_link_cleanup(&link_primer); 2902 return err; 2903 } 2904 2905 return bpf_link_settle(&link_primer); 2906 2907 error: 2908 kfree(link); 2909 kvfree(addrs); 2910 kvfree(cookies); 2911 return err; 2912 } 2913 #else /* !CONFIG_FPROBE */ 2914 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog) 2915 { 2916 return -EOPNOTSUPP; 2917 } 2918 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx) 2919 { 2920 return 0; 2921 } 2922 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx) 2923 { 2924 return 0; 2925 } 2926 #endif 2927