1 // SPDX-License-Identifier: GPL-2.0
2 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
5 #include <linux/kernel.h>
6 #include <linux/types.h>
7 #include <linux/slab.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 #include <linux/namei.h>
28 #include <net/bpf_sk_storage.h>
30 #include <uapi/linux/bpf.h>
31 #include <uapi/linux/btf.h>
35 #include "trace_probe.h"
38 #define CREATE_TRACE_POINTS
39 #include "bpf_trace.h"
41 #define bpf_event_rcu_dereference(p) \
42 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
44 #define MAX_UPROBE_MULTI_CNT (1U << 20)
45 #define MAX_KPROBE_MULTI_CNT (1U << 20)
48 struct bpf_trace_module {
49 struct module *module;
50 struct list_head list;
53 static LIST_HEAD(bpf_trace_modules);
54 static DEFINE_MUTEX(bpf_module_mutex);
56 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
58 struct bpf_raw_event_map *btp, *ret = NULL;
59 struct bpf_trace_module *btm;
62 mutex_lock(&bpf_module_mutex);
63 list_for_each_entry(btm, &bpf_trace_modules, list) {
64 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
65 btp = &btm->module->bpf_raw_events[i];
66 if (!strcmp(btp->tp->name, name)) {
67 if (try_module_get(btm->module))
74 mutex_unlock(&bpf_module_mutex);
78 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
82 #endif /* CONFIG_MODULES */
84 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
85 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
87 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
88 u64 flags, const struct btf **btf,
90 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx);
91 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
93 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx);
94 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx);
97 * trace_call_bpf - invoke BPF program
98 * @call: tracepoint event
99 * @ctx: opaque context pointer
101 * kprobe handlers execute BPF programs via this helper.
102 * Can be used from static tracepoints in the future.
104 * Return: BPF programs always return an integer which is interpreted by
106 * 0 - return from kprobe (event is filtered out)
107 * 1 - store kprobe event into ring buffer
108 * Other values are reserved and currently alias to 1
110 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
116 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
118 * since some bpf program is already running on this cpu,
119 * don't call into another bpf program (same or different)
120 * and don't send kprobe event into ring-buffer,
121 * so return zero here
128 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
129 * to all call sites, we did a bpf_prog_array_valid() there to check
130 * whether call->prog_array is empty or not, which is
131 * a heuristic to speed up execution.
133 * If bpf_prog_array_valid() fetched prog_array was
134 * non-NULL, we go into trace_call_bpf() and do the actual
135 * proper rcu_dereference() under RCU lock.
136 * If it turns out that prog_array is NULL then, we bail out.
137 * For the opposite, if the bpf_prog_array_valid() fetched pointer
138 * was NULL, you'll skip the prog_array with the risk of missing
139 * out of events when it was updated in between this and the
140 * rcu_dereference() which is accepted risk.
143 ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
148 __this_cpu_dec(bpf_prog_active);
153 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
154 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
156 regs_set_return_value(regs, rc);
157 override_function_with_return(regs);
161 static const struct bpf_func_proto bpf_override_return_proto = {
162 .func = bpf_override_return,
164 .ret_type = RET_INTEGER,
165 .arg1_type = ARG_PTR_TO_CTX,
166 .arg2_type = ARG_ANYTHING,
170 static __always_inline int
171 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
175 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
176 if (unlikely(ret < 0))
177 memset(dst, 0, size);
181 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
182 const void __user *, unsafe_ptr)
184 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
187 const struct bpf_func_proto bpf_probe_read_user_proto = {
188 .func = bpf_probe_read_user,
190 .ret_type = RET_INTEGER,
191 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
192 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
193 .arg3_type = ARG_ANYTHING,
196 static __always_inline int
197 bpf_probe_read_user_str_common(void *dst, u32 size,
198 const void __user *unsafe_ptr)
203 * NB: We rely on strncpy_from_user() not copying junk past the NUL
204 * terminator into `dst`.
206 * strncpy_from_user() does long-sized strides in the fast path. If the
207 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
208 * then there could be junk after the NUL in `dst`. If user takes `dst`
209 * and keys a hash map with it, then semantically identical strings can
210 * occupy multiple entries in the map.
212 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
213 if (unlikely(ret < 0))
214 memset(dst, 0, size);
218 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
219 const void __user *, unsafe_ptr)
221 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
224 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
225 .func = bpf_probe_read_user_str,
227 .ret_type = RET_INTEGER,
228 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
229 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
230 .arg3_type = ARG_ANYTHING,
233 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
234 const void *, unsafe_ptr)
236 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
239 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
240 .func = bpf_probe_read_kernel,
242 .ret_type = RET_INTEGER,
243 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
244 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
245 .arg3_type = ARG_ANYTHING,
248 static __always_inline int
249 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
254 * The strncpy_from_kernel_nofault() call will likely not fill the
255 * entire buffer, but that's okay in this circumstance as we're probing
256 * arbitrary memory anyway similar to bpf_probe_read_*() and might
257 * as well probe the stack. Thus, memory is explicitly cleared
258 * only in error case, so that improper users ignoring return
259 * code altogether don't copy garbage; otherwise length of string
260 * is returned that can be used for bpf_perf_event_output() et al.
262 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
263 if (unlikely(ret < 0))
264 memset(dst, 0, size);
268 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
269 const void *, unsafe_ptr)
271 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
274 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
275 .func = bpf_probe_read_kernel_str,
277 .ret_type = RET_INTEGER,
278 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
279 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
280 .arg3_type = ARG_ANYTHING,
283 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
284 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
285 const void *, unsafe_ptr)
287 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
288 return bpf_probe_read_user_common(dst, size,
289 (__force void __user *)unsafe_ptr);
291 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
294 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
295 .func = bpf_probe_read_compat,
297 .ret_type = RET_INTEGER,
298 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
299 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
300 .arg3_type = ARG_ANYTHING,
303 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
304 const void *, unsafe_ptr)
306 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
307 return bpf_probe_read_user_str_common(dst, size,
308 (__force void __user *)unsafe_ptr);
310 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
313 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
314 .func = bpf_probe_read_compat_str,
316 .ret_type = RET_INTEGER,
317 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
318 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
319 .arg3_type = ARG_ANYTHING,
321 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
323 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
327 * Ensure we're in user context which is safe for the helper to
328 * run. This helper has no business in a kthread.
330 * access_ok() should prevent writing to non-user memory, but in
331 * some situations (nommu, temporary switch, etc) access_ok() does
332 * not provide enough validation, hence the check on KERNEL_DS.
334 * nmi_uaccess_okay() ensures the probe is not run in an interim
335 * state, when the task or mm are switched. This is specifically
336 * required to prevent the use of temporary mm.
339 if (unlikely(in_interrupt() ||
340 current->flags & (PF_KTHREAD | PF_EXITING)))
342 if (unlikely(!nmi_uaccess_okay()))
345 return copy_to_user_nofault(unsafe_ptr, src, size);
348 static const struct bpf_func_proto bpf_probe_write_user_proto = {
349 .func = bpf_probe_write_user,
351 .ret_type = RET_INTEGER,
352 .arg1_type = ARG_ANYTHING,
353 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
354 .arg3_type = ARG_CONST_SIZE,
357 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
359 if (!capable(CAP_SYS_ADMIN))
362 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
363 current->comm, task_pid_nr(current));
365 return &bpf_probe_write_user_proto;
368 #define MAX_TRACE_PRINTK_VARARGS 3
369 #define BPF_TRACE_PRINTK_SIZE 1024
371 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
372 u64, arg2, u64, arg3)
374 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
375 struct bpf_bprintf_data data = {
376 .get_bin_args = true,
381 ret = bpf_bprintf_prepare(fmt, fmt_size, args,
382 MAX_TRACE_PRINTK_VARARGS, &data);
386 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
388 trace_bpf_trace_printk(data.buf);
390 bpf_bprintf_cleanup(&data);
395 static const struct bpf_func_proto bpf_trace_printk_proto = {
396 .func = bpf_trace_printk,
398 .ret_type = RET_INTEGER,
399 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
400 .arg2_type = ARG_CONST_SIZE,
403 static void __set_printk_clr_event(void)
406 * This program might be calling bpf_trace_printk,
407 * so enable the associated bpf_trace/bpf_trace_printk event.
408 * Repeat this each time as it is possible a user has
409 * disabled bpf_trace_printk events. By loading a program
410 * calling bpf_trace_printk() however the user has expressed
411 * the intent to see such events.
413 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
414 pr_warn_ratelimited("could not enable bpf_trace_printk events");
417 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
419 __set_printk_clr_event();
420 return &bpf_trace_printk_proto;
423 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, args,
426 struct bpf_bprintf_data data = {
427 .get_bin_args = true,
432 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
435 num_args = data_len / 8;
437 ret = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
441 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
443 trace_bpf_trace_printk(data.buf);
445 bpf_bprintf_cleanup(&data);
450 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
451 .func = bpf_trace_vprintk,
453 .ret_type = RET_INTEGER,
454 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
455 .arg2_type = ARG_CONST_SIZE,
456 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
457 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
460 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
462 __set_printk_clr_event();
463 return &bpf_trace_vprintk_proto;
466 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
467 const void *, args, u32, data_len)
469 struct bpf_bprintf_data data = {
470 .get_bin_args = true,
474 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
477 num_args = data_len / 8;
479 err = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
483 seq_bprintf(m, fmt, data.bin_args);
485 bpf_bprintf_cleanup(&data);
487 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
490 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
492 static const struct bpf_func_proto bpf_seq_printf_proto = {
493 .func = bpf_seq_printf,
495 .ret_type = RET_INTEGER,
496 .arg1_type = ARG_PTR_TO_BTF_ID,
497 .arg1_btf_id = &btf_seq_file_ids[0],
498 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
499 .arg3_type = ARG_CONST_SIZE,
500 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
501 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
504 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
506 return seq_write(m, data, len) ? -EOVERFLOW : 0;
509 static const struct bpf_func_proto bpf_seq_write_proto = {
510 .func = bpf_seq_write,
512 .ret_type = RET_INTEGER,
513 .arg1_type = ARG_PTR_TO_BTF_ID,
514 .arg1_btf_id = &btf_seq_file_ids[0],
515 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
516 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
519 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
520 u32, btf_ptr_size, u64, flags)
522 const struct btf *btf;
526 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
530 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
533 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
534 .func = bpf_seq_printf_btf,
536 .ret_type = RET_INTEGER,
537 .arg1_type = ARG_PTR_TO_BTF_ID,
538 .arg1_btf_id = &btf_seq_file_ids[0],
539 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
540 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
541 .arg4_type = ARG_ANYTHING,
544 static __always_inline int
545 get_map_perf_counter(struct bpf_map *map, u64 flags,
546 u64 *value, u64 *enabled, u64 *running)
548 struct bpf_array *array = container_of(map, struct bpf_array, map);
549 unsigned int cpu = smp_processor_id();
550 u64 index = flags & BPF_F_INDEX_MASK;
551 struct bpf_event_entry *ee;
553 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
555 if (index == BPF_F_CURRENT_CPU)
557 if (unlikely(index >= array->map.max_entries))
560 ee = READ_ONCE(array->ptrs[index]);
564 return perf_event_read_local(ee->event, value, enabled, running);
567 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
572 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
574 * this api is ugly since we miss [-22..-2] range of valid
575 * counter values, but that's uapi
582 static const struct bpf_func_proto bpf_perf_event_read_proto = {
583 .func = bpf_perf_event_read,
585 .ret_type = RET_INTEGER,
586 .arg1_type = ARG_CONST_MAP_PTR,
587 .arg2_type = ARG_ANYTHING,
590 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
591 struct bpf_perf_event_value *, buf, u32, size)
595 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
597 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
603 memset(buf, 0, size);
607 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
608 .func = bpf_perf_event_read_value,
610 .ret_type = RET_INTEGER,
611 .arg1_type = ARG_CONST_MAP_PTR,
612 .arg2_type = ARG_ANYTHING,
613 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
614 .arg4_type = ARG_CONST_SIZE,
617 static __always_inline u64
618 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
619 u64 flags, struct perf_sample_data *sd)
621 struct bpf_array *array = container_of(map, struct bpf_array, map);
622 unsigned int cpu = smp_processor_id();
623 u64 index = flags & BPF_F_INDEX_MASK;
624 struct bpf_event_entry *ee;
625 struct perf_event *event;
627 if (index == BPF_F_CURRENT_CPU)
629 if (unlikely(index >= array->map.max_entries))
632 ee = READ_ONCE(array->ptrs[index]);
637 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
638 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
641 if (unlikely(event->oncpu != cpu))
644 return perf_event_output(event, sd, regs);
648 * Support executing tracepoints in normal, irq, and nmi context that each call
649 * bpf_perf_event_output
651 struct bpf_trace_sample_data {
652 struct perf_sample_data sds[3];
655 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
656 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
657 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
658 u64, flags, void *, data, u64, size)
660 struct bpf_trace_sample_data *sds;
661 struct perf_raw_record raw = {
667 struct perf_sample_data *sd;
671 sds = this_cpu_ptr(&bpf_trace_sds);
672 nest_level = this_cpu_inc_return(bpf_trace_nest_level);
674 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
679 sd = &sds->sds[nest_level - 1];
681 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
686 perf_sample_data_init(sd, 0, 0);
687 perf_sample_save_raw_data(sd, &raw);
689 err = __bpf_perf_event_output(regs, map, flags, sd);
691 this_cpu_dec(bpf_trace_nest_level);
696 static const struct bpf_func_proto bpf_perf_event_output_proto = {
697 .func = bpf_perf_event_output,
699 .ret_type = RET_INTEGER,
700 .arg1_type = ARG_PTR_TO_CTX,
701 .arg2_type = ARG_CONST_MAP_PTR,
702 .arg3_type = ARG_ANYTHING,
703 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
704 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
707 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
708 struct bpf_nested_pt_regs {
709 struct pt_regs regs[3];
711 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
712 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
714 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
715 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
717 struct perf_raw_frag frag = {
722 struct perf_raw_record raw = {
725 .next = ctx_size ? &frag : NULL,
731 struct perf_sample_data *sd;
732 struct pt_regs *regs;
737 nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
739 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
743 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
744 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
746 perf_fetch_caller_regs(regs);
747 perf_sample_data_init(sd, 0, 0);
748 perf_sample_save_raw_data(sd, &raw);
750 ret = __bpf_perf_event_output(regs, map, flags, sd);
752 this_cpu_dec(bpf_event_output_nest_level);
757 BPF_CALL_0(bpf_get_current_task)
759 return (long) current;
762 const struct bpf_func_proto bpf_get_current_task_proto = {
763 .func = bpf_get_current_task,
765 .ret_type = RET_INTEGER,
768 BPF_CALL_0(bpf_get_current_task_btf)
770 return (unsigned long) current;
773 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
774 .func = bpf_get_current_task_btf,
776 .ret_type = RET_PTR_TO_BTF_ID_TRUSTED,
777 .ret_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
780 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
782 return (unsigned long) task_pt_regs(task);
785 BTF_ID_LIST(bpf_task_pt_regs_ids)
786 BTF_ID(struct, pt_regs)
788 const struct bpf_func_proto bpf_task_pt_regs_proto = {
789 .func = bpf_task_pt_regs,
791 .arg1_type = ARG_PTR_TO_BTF_ID,
792 .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
793 .ret_type = RET_PTR_TO_BTF_ID,
794 .ret_btf_id = &bpf_task_pt_regs_ids[0],
797 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
799 struct bpf_array *array = container_of(map, struct bpf_array, map);
802 if (unlikely(idx >= array->map.max_entries))
805 cgrp = READ_ONCE(array->ptrs[idx]);
809 return task_under_cgroup_hierarchy(current, cgrp);
812 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
813 .func = bpf_current_task_under_cgroup,
815 .ret_type = RET_INTEGER,
816 .arg1_type = ARG_CONST_MAP_PTR,
817 .arg2_type = ARG_ANYTHING,
820 struct send_signal_irq_work {
821 struct irq_work irq_work;
822 struct task_struct *task;
827 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
829 static void do_bpf_send_signal(struct irq_work *entry)
831 struct send_signal_irq_work *work;
833 work = container_of(entry, struct send_signal_irq_work, irq_work);
834 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
835 put_task_struct(work->task);
838 static int bpf_send_signal_common(u32 sig, enum pid_type type)
840 struct send_signal_irq_work *work = NULL;
842 /* Similar to bpf_probe_write_user, task needs to be
843 * in a sound condition and kernel memory access be
844 * permitted in order to send signal to the current
847 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
849 if (unlikely(!nmi_uaccess_okay()))
851 /* Task should not be pid=1 to avoid kernel panic. */
852 if (unlikely(is_global_init(current)))
855 if (irqs_disabled()) {
856 /* Do an early check on signal validity. Otherwise,
857 * the error is lost in deferred irq_work.
859 if (unlikely(!valid_signal(sig)))
862 work = this_cpu_ptr(&send_signal_work);
863 if (irq_work_is_busy(&work->irq_work))
866 /* Add the current task, which is the target of sending signal,
867 * to the irq_work. The current task may change when queued
868 * irq works get executed.
870 work->task = get_task_struct(current);
873 irq_work_queue(&work->irq_work);
877 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
880 BPF_CALL_1(bpf_send_signal, u32, sig)
882 return bpf_send_signal_common(sig, PIDTYPE_TGID);
885 static const struct bpf_func_proto bpf_send_signal_proto = {
886 .func = bpf_send_signal,
888 .ret_type = RET_INTEGER,
889 .arg1_type = ARG_ANYTHING,
892 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
894 return bpf_send_signal_common(sig, PIDTYPE_PID);
897 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
898 .func = bpf_send_signal_thread,
900 .ret_type = RET_INTEGER,
901 .arg1_type = ARG_ANYTHING,
904 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
914 * The path pointer is verified as trusted and safe to use,
915 * but let's double check it's valid anyway to workaround
916 * potentially broken verifier.
918 len = copy_from_kernel_nofault(©, path, sizeof(*path));
922 p = d_path(©, buf, sz);
927 memmove(buf, p, len);
933 BTF_SET_START(btf_allowlist_d_path)
934 #ifdef CONFIG_SECURITY
935 BTF_ID(func, security_file_permission)
936 BTF_ID(func, security_inode_getattr)
937 BTF_ID(func, security_file_open)
939 #ifdef CONFIG_SECURITY_PATH
940 BTF_ID(func, security_path_truncate)
942 BTF_ID(func, vfs_truncate)
943 BTF_ID(func, vfs_fallocate)
944 BTF_ID(func, dentry_open)
945 BTF_ID(func, vfs_getattr)
946 BTF_ID(func, filp_close)
947 BTF_SET_END(btf_allowlist_d_path)
949 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
951 if (prog->type == BPF_PROG_TYPE_TRACING &&
952 prog->expected_attach_type == BPF_TRACE_ITER)
955 if (prog->type == BPF_PROG_TYPE_LSM)
956 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
958 return btf_id_set_contains(&btf_allowlist_d_path,
959 prog->aux->attach_btf_id);
962 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
964 static const struct bpf_func_proto bpf_d_path_proto = {
967 .ret_type = RET_INTEGER,
968 .arg1_type = ARG_PTR_TO_BTF_ID,
969 .arg1_btf_id = &bpf_d_path_btf_ids[0],
970 .arg2_type = ARG_PTR_TO_MEM,
971 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
972 .allowed = bpf_d_path_allowed,
975 #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
976 BTF_F_PTR_RAW | BTF_F_ZERO)
978 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
979 u64 flags, const struct btf **btf,
982 const struct btf_type *t;
984 if (unlikely(flags & ~(BTF_F_ALL)))
987 if (btf_ptr_size != sizeof(struct btf_ptr))
990 *btf = bpf_get_btf_vmlinux();
992 if (IS_ERR_OR_NULL(*btf))
993 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
995 if (ptr->type_id > 0)
996 *btf_id = ptr->type_id;
1001 t = btf_type_by_id(*btf, *btf_id);
1002 if (*btf_id <= 0 || !t)
1008 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
1009 u32, btf_ptr_size, u64, flags)
1011 const struct btf *btf;
1015 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
1019 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1023 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1024 .func = bpf_snprintf_btf,
1026 .ret_type = RET_INTEGER,
1027 .arg1_type = ARG_PTR_TO_MEM,
1028 .arg2_type = ARG_CONST_SIZE,
1029 .arg3_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1030 .arg4_type = ARG_CONST_SIZE,
1031 .arg5_type = ARG_ANYTHING,
1034 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1036 /* This helper call is inlined by verifier. */
1037 return ((u64 *)ctx)[-2];
1040 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1041 .func = bpf_get_func_ip_tracing,
1043 .ret_type = RET_INTEGER,
1044 .arg1_type = ARG_PTR_TO_CTX,
1047 #ifdef CONFIG_X86_KERNEL_IBT
1048 static unsigned long get_entry_ip(unsigned long fentry_ip)
1052 /* Being extra safe in here in case entry ip is on the page-edge. */
1053 if (get_kernel_nofault(instr, (u32 *) fentry_ip - 1))
1055 if (is_endbr(instr))
1056 fentry_ip -= ENDBR_INSN_SIZE;
1060 #define get_entry_ip(fentry_ip) fentry_ip
1063 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1065 struct bpf_trace_run_ctx *run_ctx __maybe_unused;
1068 #ifdef CONFIG_UPROBES
1069 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1070 if (run_ctx->is_uprobe)
1071 return ((struct uprobe_dispatch_data *)current->utask->vaddr)->bp_addr;
1074 kp = kprobe_running();
1076 if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1079 return get_entry_ip((uintptr_t)kp->addr);
1082 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1083 .func = bpf_get_func_ip_kprobe,
1085 .ret_type = RET_INTEGER,
1086 .arg1_type = ARG_PTR_TO_CTX,
1089 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1091 return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1094 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1095 .func = bpf_get_func_ip_kprobe_multi,
1097 .ret_type = RET_INTEGER,
1098 .arg1_type = ARG_PTR_TO_CTX,
1101 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1103 return bpf_kprobe_multi_cookie(current->bpf_ctx);
1106 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1107 .func = bpf_get_attach_cookie_kprobe_multi,
1109 .ret_type = RET_INTEGER,
1110 .arg1_type = ARG_PTR_TO_CTX,
1113 BPF_CALL_1(bpf_get_func_ip_uprobe_multi, struct pt_regs *, regs)
1115 return bpf_uprobe_multi_entry_ip(current->bpf_ctx);
1118 static const struct bpf_func_proto bpf_get_func_ip_proto_uprobe_multi = {
1119 .func = bpf_get_func_ip_uprobe_multi,
1121 .ret_type = RET_INTEGER,
1122 .arg1_type = ARG_PTR_TO_CTX,
1125 BPF_CALL_1(bpf_get_attach_cookie_uprobe_multi, struct pt_regs *, regs)
1127 return bpf_uprobe_multi_cookie(current->bpf_ctx);
1130 static const struct bpf_func_proto bpf_get_attach_cookie_proto_umulti = {
1131 .func = bpf_get_attach_cookie_uprobe_multi,
1133 .ret_type = RET_INTEGER,
1134 .arg1_type = ARG_PTR_TO_CTX,
1137 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1139 struct bpf_trace_run_ctx *run_ctx;
1141 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1142 return run_ctx->bpf_cookie;
1145 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1146 .func = bpf_get_attach_cookie_trace,
1148 .ret_type = RET_INTEGER,
1149 .arg1_type = ARG_PTR_TO_CTX,
1152 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1154 return ctx->event->bpf_cookie;
1157 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1158 .func = bpf_get_attach_cookie_pe,
1160 .ret_type = RET_INTEGER,
1161 .arg1_type = ARG_PTR_TO_CTX,
1164 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1166 struct bpf_trace_run_ctx *run_ctx;
1168 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1169 return run_ctx->bpf_cookie;
1172 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1173 .func = bpf_get_attach_cookie_tracing,
1175 .ret_type = RET_INTEGER,
1176 .arg1_type = ARG_PTR_TO_CTX,
1179 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1184 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1185 u32 entry_cnt = size / br_entry_size;
1187 entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1189 if (unlikely(flags))
1195 return entry_cnt * br_entry_size;
1199 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1200 .func = bpf_get_branch_snapshot,
1202 .ret_type = RET_INTEGER,
1203 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1204 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1207 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1209 /* This helper call is inlined by verifier. */
1210 u64 nr_args = ((u64 *)ctx)[-1];
1212 if ((u64) n >= nr_args)
1214 *value = ((u64 *)ctx)[n];
1218 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1219 .func = get_func_arg,
1220 .ret_type = RET_INTEGER,
1221 .arg1_type = ARG_PTR_TO_CTX,
1222 .arg2_type = ARG_ANYTHING,
1223 .arg3_type = ARG_PTR_TO_LONG,
1226 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1228 /* This helper call is inlined by verifier. */
1229 u64 nr_args = ((u64 *)ctx)[-1];
1231 *value = ((u64 *)ctx)[nr_args];
1235 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1236 .func = get_func_ret,
1237 .ret_type = RET_INTEGER,
1238 .arg1_type = ARG_PTR_TO_CTX,
1239 .arg2_type = ARG_PTR_TO_LONG,
1242 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1244 /* This helper call is inlined by verifier. */
1245 return ((u64 *)ctx)[-1];
1248 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1249 .func = get_func_arg_cnt,
1250 .ret_type = RET_INTEGER,
1251 .arg1_type = ARG_PTR_TO_CTX,
1256 __diag_ignore_all("-Wmissing-prototypes",
1257 "kfuncs which will be used in BPF programs");
1260 * bpf_lookup_user_key - lookup a key by its serial
1261 * @serial: key handle serial number
1262 * @flags: lookup-specific flags
1264 * Search a key with a given *serial* and the provided *flags*.
1265 * If found, increment the reference count of the key by one, and
1266 * return it in the bpf_key structure.
1268 * The bpf_key structure must be passed to bpf_key_put() when done
1269 * with it, so that the key reference count is decremented and the
1270 * bpf_key structure is freed.
1272 * Permission checks are deferred to the time the key is used by
1273 * one of the available key-specific kfuncs.
1275 * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1276 * special keyring (e.g. session keyring), if it doesn't yet exist.
1277 * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1278 * for the key construction, and to retrieve uninstantiated keys (keys
1279 * without data attached to them).
1281 * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1282 * NULL pointer otherwise.
1284 __bpf_kfunc struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1287 struct bpf_key *bkey;
1289 if (flags & ~KEY_LOOKUP_ALL)
1293 * Permission check is deferred until the key is used, as the
1294 * intent of the caller is unknown here.
1296 key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1297 if (IS_ERR(key_ref))
1300 bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1302 key_put(key_ref_to_ptr(key_ref));
1306 bkey->key = key_ref_to_ptr(key_ref);
1307 bkey->has_ref = true;
1313 * bpf_lookup_system_key - lookup a key by a system-defined ID
1316 * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1317 * The key pointer is marked as invalid, to prevent bpf_key_put() from
1318 * attempting to decrement the key reference count on that pointer. The key
1319 * pointer set in such way is currently understood only by
1320 * verify_pkcs7_signature().
1322 * Set *id* to one of the values defined in include/linux/verification.h:
1323 * 0 for the primary keyring (immutable keyring of system keys);
1324 * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1325 * (where keys can be added only if they are vouched for by existing keys
1326 * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1327 * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1328 * kerned image and, possibly, the initramfs signature).
1330 * Return: a bpf_key pointer with an invalid key pointer set from the
1331 * pre-determined ID on success, a NULL pointer otherwise
1333 __bpf_kfunc struct bpf_key *bpf_lookup_system_key(u64 id)
1335 struct bpf_key *bkey;
1337 if (system_keyring_id_check(id) < 0)
1340 bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1344 bkey->key = (struct key *)(unsigned long)id;
1345 bkey->has_ref = false;
1351 * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1352 * @bkey: bpf_key structure
1354 * Decrement the reference count of the key inside *bkey*, if the pointer
1355 * is valid, and free *bkey*.
1357 __bpf_kfunc void bpf_key_put(struct bpf_key *bkey)
1365 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1367 * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1368 * @data_ptr: data to verify
1369 * @sig_ptr: signature of the data
1370 * @trusted_keyring: keyring with keys trusted for signature verification
1372 * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1373 * with keys in a keyring referenced by *trusted_keyring*.
1375 * Return: 0 on success, a negative value on error.
1377 __bpf_kfunc int bpf_verify_pkcs7_signature(struct bpf_dynptr_kern *data_ptr,
1378 struct bpf_dynptr_kern *sig_ptr,
1379 struct bpf_key *trusted_keyring)
1383 if (trusted_keyring->has_ref) {
1385 * Do the permission check deferred in bpf_lookup_user_key().
1386 * See bpf_lookup_user_key() for more details.
1388 * A call to key_task_permission() here would be redundant, as
1389 * it is already done by keyring_search() called by
1390 * find_asymmetric_key().
1392 ret = key_validate(trusted_keyring->key);
1397 return verify_pkcs7_signature(data_ptr->data,
1398 __bpf_dynptr_size(data_ptr),
1400 __bpf_dynptr_size(sig_ptr),
1401 trusted_keyring->key,
1402 VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1405 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1409 BTF_SET8_START(key_sig_kfunc_set)
1410 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1411 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1412 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1413 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1414 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1416 BTF_SET8_END(key_sig_kfunc_set)
1418 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1419 .owner = THIS_MODULE,
1420 .set = &key_sig_kfunc_set,
1423 static int __init bpf_key_sig_kfuncs_init(void)
1425 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1426 &bpf_key_sig_kfunc_set);
1429 late_initcall(bpf_key_sig_kfuncs_init);
1430 #endif /* CONFIG_KEYS */
1432 static const struct bpf_func_proto *
1433 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1436 case BPF_FUNC_map_lookup_elem:
1437 return &bpf_map_lookup_elem_proto;
1438 case BPF_FUNC_map_update_elem:
1439 return &bpf_map_update_elem_proto;
1440 case BPF_FUNC_map_delete_elem:
1441 return &bpf_map_delete_elem_proto;
1442 case BPF_FUNC_map_push_elem:
1443 return &bpf_map_push_elem_proto;
1444 case BPF_FUNC_map_pop_elem:
1445 return &bpf_map_pop_elem_proto;
1446 case BPF_FUNC_map_peek_elem:
1447 return &bpf_map_peek_elem_proto;
1448 case BPF_FUNC_map_lookup_percpu_elem:
1449 return &bpf_map_lookup_percpu_elem_proto;
1450 case BPF_FUNC_ktime_get_ns:
1451 return &bpf_ktime_get_ns_proto;
1452 case BPF_FUNC_ktime_get_boot_ns:
1453 return &bpf_ktime_get_boot_ns_proto;
1454 case BPF_FUNC_tail_call:
1455 return &bpf_tail_call_proto;
1456 case BPF_FUNC_get_current_pid_tgid:
1457 return &bpf_get_current_pid_tgid_proto;
1458 case BPF_FUNC_get_current_task:
1459 return &bpf_get_current_task_proto;
1460 case BPF_FUNC_get_current_task_btf:
1461 return &bpf_get_current_task_btf_proto;
1462 case BPF_FUNC_task_pt_regs:
1463 return &bpf_task_pt_regs_proto;
1464 case BPF_FUNC_get_current_uid_gid:
1465 return &bpf_get_current_uid_gid_proto;
1466 case BPF_FUNC_get_current_comm:
1467 return &bpf_get_current_comm_proto;
1468 case BPF_FUNC_trace_printk:
1469 return bpf_get_trace_printk_proto();
1470 case BPF_FUNC_get_smp_processor_id:
1471 return &bpf_get_smp_processor_id_proto;
1472 case BPF_FUNC_get_numa_node_id:
1473 return &bpf_get_numa_node_id_proto;
1474 case BPF_FUNC_perf_event_read:
1475 return &bpf_perf_event_read_proto;
1476 case BPF_FUNC_current_task_under_cgroup:
1477 return &bpf_current_task_under_cgroup_proto;
1478 case BPF_FUNC_get_prandom_u32:
1479 return &bpf_get_prandom_u32_proto;
1480 case BPF_FUNC_probe_write_user:
1481 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1482 NULL : bpf_get_probe_write_proto();
1483 case BPF_FUNC_probe_read_user:
1484 return &bpf_probe_read_user_proto;
1485 case BPF_FUNC_probe_read_kernel:
1486 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1487 NULL : &bpf_probe_read_kernel_proto;
1488 case BPF_FUNC_probe_read_user_str:
1489 return &bpf_probe_read_user_str_proto;
1490 case BPF_FUNC_probe_read_kernel_str:
1491 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1492 NULL : &bpf_probe_read_kernel_str_proto;
1493 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1494 case BPF_FUNC_probe_read:
1495 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1496 NULL : &bpf_probe_read_compat_proto;
1497 case BPF_FUNC_probe_read_str:
1498 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1499 NULL : &bpf_probe_read_compat_str_proto;
1501 #ifdef CONFIG_CGROUPS
1502 case BPF_FUNC_cgrp_storage_get:
1503 return &bpf_cgrp_storage_get_proto;
1504 case BPF_FUNC_cgrp_storage_delete:
1505 return &bpf_cgrp_storage_delete_proto;
1507 case BPF_FUNC_send_signal:
1508 return &bpf_send_signal_proto;
1509 case BPF_FUNC_send_signal_thread:
1510 return &bpf_send_signal_thread_proto;
1511 case BPF_FUNC_perf_event_read_value:
1512 return &bpf_perf_event_read_value_proto;
1513 case BPF_FUNC_get_ns_current_pid_tgid:
1514 return &bpf_get_ns_current_pid_tgid_proto;
1515 case BPF_FUNC_ringbuf_output:
1516 return &bpf_ringbuf_output_proto;
1517 case BPF_FUNC_ringbuf_reserve:
1518 return &bpf_ringbuf_reserve_proto;
1519 case BPF_FUNC_ringbuf_submit:
1520 return &bpf_ringbuf_submit_proto;
1521 case BPF_FUNC_ringbuf_discard:
1522 return &bpf_ringbuf_discard_proto;
1523 case BPF_FUNC_ringbuf_query:
1524 return &bpf_ringbuf_query_proto;
1525 case BPF_FUNC_jiffies64:
1526 return &bpf_jiffies64_proto;
1527 case BPF_FUNC_get_task_stack:
1528 return &bpf_get_task_stack_proto;
1529 case BPF_FUNC_copy_from_user:
1530 return &bpf_copy_from_user_proto;
1531 case BPF_FUNC_copy_from_user_task:
1532 return &bpf_copy_from_user_task_proto;
1533 case BPF_FUNC_snprintf_btf:
1534 return &bpf_snprintf_btf_proto;
1535 case BPF_FUNC_per_cpu_ptr:
1536 return &bpf_per_cpu_ptr_proto;
1537 case BPF_FUNC_this_cpu_ptr:
1538 return &bpf_this_cpu_ptr_proto;
1539 case BPF_FUNC_task_storage_get:
1540 if (bpf_prog_check_recur(prog))
1541 return &bpf_task_storage_get_recur_proto;
1542 return &bpf_task_storage_get_proto;
1543 case BPF_FUNC_task_storage_delete:
1544 if (bpf_prog_check_recur(prog))
1545 return &bpf_task_storage_delete_recur_proto;
1546 return &bpf_task_storage_delete_proto;
1547 case BPF_FUNC_for_each_map_elem:
1548 return &bpf_for_each_map_elem_proto;
1549 case BPF_FUNC_snprintf:
1550 return &bpf_snprintf_proto;
1551 case BPF_FUNC_get_func_ip:
1552 return &bpf_get_func_ip_proto_tracing;
1553 case BPF_FUNC_get_branch_snapshot:
1554 return &bpf_get_branch_snapshot_proto;
1555 case BPF_FUNC_find_vma:
1556 return &bpf_find_vma_proto;
1557 case BPF_FUNC_trace_vprintk:
1558 return bpf_get_trace_vprintk_proto();
1560 return bpf_base_func_proto(func_id);
1564 static const struct bpf_func_proto *
1565 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1568 case BPF_FUNC_perf_event_output:
1569 return &bpf_perf_event_output_proto;
1570 case BPF_FUNC_get_stackid:
1571 return &bpf_get_stackid_proto;
1572 case BPF_FUNC_get_stack:
1573 return &bpf_get_stack_proto;
1574 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1575 case BPF_FUNC_override_return:
1576 return &bpf_override_return_proto;
1578 case BPF_FUNC_get_func_ip:
1579 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI)
1580 return &bpf_get_func_ip_proto_kprobe_multi;
1581 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI)
1582 return &bpf_get_func_ip_proto_uprobe_multi;
1583 return &bpf_get_func_ip_proto_kprobe;
1584 case BPF_FUNC_get_attach_cookie:
1585 if (prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI)
1586 return &bpf_get_attach_cookie_proto_kmulti;
1587 if (prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI)
1588 return &bpf_get_attach_cookie_proto_umulti;
1589 return &bpf_get_attach_cookie_proto_trace;
1591 return bpf_tracing_func_proto(func_id, prog);
1595 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1596 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1597 const struct bpf_prog *prog,
1598 struct bpf_insn_access_aux *info)
1600 if (off < 0 || off >= sizeof(struct pt_regs))
1602 if (type != BPF_READ)
1604 if (off % size != 0)
1607 * Assertion for 32 bit to make sure last 8 byte access
1608 * (BPF_DW) to the last 4 byte member is disallowed.
1610 if (off + size > sizeof(struct pt_regs))
1616 const struct bpf_verifier_ops kprobe_verifier_ops = {
1617 .get_func_proto = kprobe_prog_func_proto,
1618 .is_valid_access = kprobe_prog_is_valid_access,
1621 const struct bpf_prog_ops kprobe_prog_ops = {
1624 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1625 u64, flags, void *, data, u64, size)
1627 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1630 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1631 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1632 * from there and call the same bpf_perf_event_output() helper inline.
1634 return ____bpf_perf_event_output(regs, map, flags, data, size);
1637 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1638 .func = bpf_perf_event_output_tp,
1640 .ret_type = RET_INTEGER,
1641 .arg1_type = ARG_PTR_TO_CTX,
1642 .arg2_type = ARG_CONST_MAP_PTR,
1643 .arg3_type = ARG_ANYTHING,
1644 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1645 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1648 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1651 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1654 * Same comment as in bpf_perf_event_output_tp(), only that this time
1655 * the other helper's function body cannot be inlined due to being
1656 * external, thus we need to call raw helper function.
1658 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1662 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1663 .func = bpf_get_stackid_tp,
1665 .ret_type = RET_INTEGER,
1666 .arg1_type = ARG_PTR_TO_CTX,
1667 .arg2_type = ARG_CONST_MAP_PTR,
1668 .arg3_type = ARG_ANYTHING,
1671 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1674 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1676 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1677 (unsigned long) size, flags, 0);
1680 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1681 .func = bpf_get_stack_tp,
1683 .ret_type = RET_INTEGER,
1684 .arg1_type = ARG_PTR_TO_CTX,
1685 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1686 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1687 .arg4_type = ARG_ANYTHING,
1690 static const struct bpf_func_proto *
1691 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1694 case BPF_FUNC_perf_event_output:
1695 return &bpf_perf_event_output_proto_tp;
1696 case BPF_FUNC_get_stackid:
1697 return &bpf_get_stackid_proto_tp;
1698 case BPF_FUNC_get_stack:
1699 return &bpf_get_stack_proto_tp;
1700 case BPF_FUNC_get_attach_cookie:
1701 return &bpf_get_attach_cookie_proto_trace;
1703 return bpf_tracing_func_proto(func_id, prog);
1707 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1708 const struct bpf_prog *prog,
1709 struct bpf_insn_access_aux *info)
1711 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1713 if (type != BPF_READ)
1715 if (off % size != 0)
1718 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1722 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1723 .get_func_proto = tp_prog_func_proto,
1724 .is_valid_access = tp_prog_is_valid_access,
1727 const struct bpf_prog_ops tracepoint_prog_ops = {
1730 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1731 struct bpf_perf_event_value *, buf, u32, size)
1735 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1737 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1743 memset(buf, 0, size);
1747 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1748 .func = bpf_perf_prog_read_value,
1750 .ret_type = RET_INTEGER,
1751 .arg1_type = ARG_PTR_TO_CTX,
1752 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1753 .arg3_type = ARG_CONST_SIZE,
1756 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1757 void *, buf, u32, size, u64, flags)
1759 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1760 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1763 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1766 if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1769 if (unlikely(!br_stack))
1772 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1773 return br_stack->nr * br_entry_size;
1775 if (!buf || (size % br_entry_size != 0))
1778 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1779 memcpy(buf, br_stack->entries, to_copy);
1784 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1785 .func = bpf_read_branch_records,
1787 .ret_type = RET_INTEGER,
1788 .arg1_type = ARG_PTR_TO_CTX,
1789 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1790 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1791 .arg4_type = ARG_ANYTHING,
1794 static const struct bpf_func_proto *
1795 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1798 case BPF_FUNC_perf_event_output:
1799 return &bpf_perf_event_output_proto_tp;
1800 case BPF_FUNC_get_stackid:
1801 return &bpf_get_stackid_proto_pe;
1802 case BPF_FUNC_get_stack:
1803 return &bpf_get_stack_proto_pe;
1804 case BPF_FUNC_perf_prog_read_value:
1805 return &bpf_perf_prog_read_value_proto;
1806 case BPF_FUNC_read_branch_records:
1807 return &bpf_read_branch_records_proto;
1808 case BPF_FUNC_get_attach_cookie:
1809 return &bpf_get_attach_cookie_proto_pe;
1811 return bpf_tracing_func_proto(func_id, prog);
1816 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1817 * to avoid potential recursive reuse issue when/if tracepoints are added
1818 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1820 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1821 * in normal, irq, and nmi context.
1823 struct bpf_raw_tp_regs {
1824 struct pt_regs regs[3];
1826 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1827 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1828 static struct pt_regs *get_bpf_raw_tp_regs(void)
1830 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1831 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1833 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1834 this_cpu_dec(bpf_raw_tp_nest_level);
1835 return ERR_PTR(-EBUSY);
1838 return &tp_regs->regs[nest_level - 1];
1841 static void put_bpf_raw_tp_regs(void)
1843 this_cpu_dec(bpf_raw_tp_nest_level);
1846 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1847 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1849 struct pt_regs *regs = get_bpf_raw_tp_regs();
1853 return PTR_ERR(regs);
1855 perf_fetch_caller_regs(regs);
1856 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1858 put_bpf_raw_tp_regs();
1862 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1863 .func = bpf_perf_event_output_raw_tp,
1865 .ret_type = RET_INTEGER,
1866 .arg1_type = ARG_PTR_TO_CTX,
1867 .arg2_type = ARG_CONST_MAP_PTR,
1868 .arg3_type = ARG_ANYTHING,
1869 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1870 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1873 extern const struct bpf_func_proto bpf_skb_output_proto;
1874 extern const struct bpf_func_proto bpf_xdp_output_proto;
1875 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1877 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1878 struct bpf_map *, map, u64, flags)
1880 struct pt_regs *regs = get_bpf_raw_tp_regs();
1884 return PTR_ERR(regs);
1886 perf_fetch_caller_regs(regs);
1887 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1888 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1890 put_bpf_raw_tp_regs();
1894 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1895 .func = bpf_get_stackid_raw_tp,
1897 .ret_type = RET_INTEGER,
1898 .arg1_type = ARG_PTR_TO_CTX,
1899 .arg2_type = ARG_CONST_MAP_PTR,
1900 .arg3_type = ARG_ANYTHING,
1903 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1904 void *, buf, u32, size, u64, flags)
1906 struct pt_regs *regs = get_bpf_raw_tp_regs();
1910 return PTR_ERR(regs);
1912 perf_fetch_caller_regs(regs);
1913 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1914 (unsigned long) size, flags, 0);
1915 put_bpf_raw_tp_regs();
1919 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1920 .func = bpf_get_stack_raw_tp,
1922 .ret_type = RET_INTEGER,
1923 .arg1_type = ARG_PTR_TO_CTX,
1924 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1925 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1926 .arg4_type = ARG_ANYTHING,
1929 static const struct bpf_func_proto *
1930 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1933 case BPF_FUNC_perf_event_output:
1934 return &bpf_perf_event_output_proto_raw_tp;
1935 case BPF_FUNC_get_stackid:
1936 return &bpf_get_stackid_proto_raw_tp;
1937 case BPF_FUNC_get_stack:
1938 return &bpf_get_stack_proto_raw_tp;
1940 return bpf_tracing_func_proto(func_id, prog);
1944 const struct bpf_func_proto *
1945 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1947 const struct bpf_func_proto *fn;
1951 case BPF_FUNC_skb_output:
1952 return &bpf_skb_output_proto;
1953 case BPF_FUNC_xdp_output:
1954 return &bpf_xdp_output_proto;
1955 case BPF_FUNC_skc_to_tcp6_sock:
1956 return &bpf_skc_to_tcp6_sock_proto;
1957 case BPF_FUNC_skc_to_tcp_sock:
1958 return &bpf_skc_to_tcp_sock_proto;
1959 case BPF_FUNC_skc_to_tcp_timewait_sock:
1960 return &bpf_skc_to_tcp_timewait_sock_proto;
1961 case BPF_FUNC_skc_to_tcp_request_sock:
1962 return &bpf_skc_to_tcp_request_sock_proto;
1963 case BPF_FUNC_skc_to_udp6_sock:
1964 return &bpf_skc_to_udp6_sock_proto;
1965 case BPF_FUNC_skc_to_unix_sock:
1966 return &bpf_skc_to_unix_sock_proto;
1967 case BPF_FUNC_skc_to_mptcp_sock:
1968 return &bpf_skc_to_mptcp_sock_proto;
1969 case BPF_FUNC_sk_storage_get:
1970 return &bpf_sk_storage_get_tracing_proto;
1971 case BPF_FUNC_sk_storage_delete:
1972 return &bpf_sk_storage_delete_tracing_proto;
1973 case BPF_FUNC_sock_from_file:
1974 return &bpf_sock_from_file_proto;
1975 case BPF_FUNC_get_socket_cookie:
1976 return &bpf_get_socket_ptr_cookie_proto;
1977 case BPF_FUNC_xdp_get_buff_len:
1978 return &bpf_xdp_get_buff_len_trace_proto;
1980 case BPF_FUNC_seq_printf:
1981 return prog->expected_attach_type == BPF_TRACE_ITER ?
1982 &bpf_seq_printf_proto :
1984 case BPF_FUNC_seq_write:
1985 return prog->expected_attach_type == BPF_TRACE_ITER ?
1986 &bpf_seq_write_proto :
1988 case BPF_FUNC_seq_printf_btf:
1989 return prog->expected_attach_type == BPF_TRACE_ITER ?
1990 &bpf_seq_printf_btf_proto :
1992 case BPF_FUNC_d_path:
1993 return &bpf_d_path_proto;
1994 case BPF_FUNC_get_func_arg:
1995 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1996 case BPF_FUNC_get_func_ret:
1997 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1998 case BPF_FUNC_get_func_arg_cnt:
1999 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
2000 case BPF_FUNC_get_attach_cookie:
2001 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
2003 fn = raw_tp_prog_func_proto(func_id, prog);
2004 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
2005 fn = bpf_iter_get_func_proto(func_id, prog);
2010 static bool raw_tp_prog_is_valid_access(int off, int size,
2011 enum bpf_access_type type,
2012 const struct bpf_prog *prog,
2013 struct bpf_insn_access_aux *info)
2015 return bpf_tracing_ctx_access(off, size, type);
2018 static bool tracing_prog_is_valid_access(int off, int size,
2019 enum bpf_access_type type,
2020 const struct bpf_prog *prog,
2021 struct bpf_insn_access_aux *info)
2023 return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
2026 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
2027 const union bpf_attr *kattr,
2028 union bpf_attr __user *uattr)
2033 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
2034 .get_func_proto = raw_tp_prog_func_proto,
2035 .is_valid_access = raw_tp_prog_is_valid_access,
2038 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
2040 .test_run = bpf_prog_test_run_raw_tp,
2044 const struct bpf_verifier_ops tracing_verifier_ops = {
2045 .get_func_proto = tracing_prog_func_proto,
2046 .is_valid_access = tracing_prog_is_valid_access,
2049 const struct bpf_prog_ops tracing_prog_ops = {
2050 .test_run = bpf_prog_test_run_tracing,
2053 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
2054 enum bpf_access_type type,
2055 const struct bpf_prog *prog,
2056 struct bpf_insn_access_aux *info)
2059 if (size != sizeof(u64) || type != BPF_READ)
2061 info->reg_type = PTR_TO_TP_BUFFER;
2063 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
2066 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
2067 .get_func_proto = raw_tp_prog_func_proto,
2068 .is_valid_access = raw_tp_writable_prog_is_valid_access,
2071 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2074 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2075 const struct bpf_prog *prog,
2076 struct bpf_insn_access_aux *info)
2078 const int size_u64 = sizeof(u64);
2080 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2082 if (type != BPF_READ)
2084 if (off % size != 0) {
2085 if (sizeof(unsigned long) != 4)
2089 if (off % size != 4)
2094 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2095 bpf_ctx_record_field_size(info, size_u64);
2096 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2099 case bpf_ctx_range(struct bpf_perf_event_data, addr):
2100 bpf_ctx_record_field_size(info, size_u64);
2101 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2105 if (size != sizeof(long))
2112 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2113 const struct bpf_insn *si,
2114 struct bpf_insn *insn_buf,
2115 struct bpf_prog *prog, u32 *target_size)
2117 struct bpf_insn *insn = insn_buf;
2120 case offsetof(struct bpf_perf_event_data, sample_period):
2121 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2122 data), si->dst_reg, si->src_reg,
2123 offsetof(struct bpf_perf_event_data_kern, data));
2124 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2125 bpf_target_off(struct perf_sample_data, period, 8,
2128 case offsetof(struct bpf_perf_event_data, addr):
2129 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2130 data), si->dst_reg, si->src_reg,
2131 offsetof(struct bpf_perf_event_data_kern, data));
2132 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2133 bpf_target_off(struct perf_sample_data, addr, 8,
2137 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2138 regs), si->dst_reg, si->src_reg,
2139 offsetof(struct bpf_perf_event_data_kern, regs));
2140 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2145 return insn - insn_buf;
2148 const struct bpf_verifier_ops perf_event_verifier_ops = {
2149 .get_func_proto = pe_prog_func_proto,
2150 .is_valid_access = pe_prog_is_valid_access,
2151 .convert_ctx_access = pe_prog_convert_ctx_access,
2154 const struct bpf_prog_ops perf_event_prog_ops = {
2157 static DEFINE_MUTEX(bpf_event_mutex);
2159 #define BPF_TRACE_MAX_PROGS 64
2161 int perf_event_attach_bpf_prog(struct perf_event *event,
2162 struct bpf_prog *prog,
2165 struct bpf_prog_array *old_array;
2166 struct bpf_prog_array *new_array;
2170 * Kprobe override only works if they are on the function entry,
2171 * and only if they are on the opt-in list.
2173 if (prog->kprobe_override &&
2174 (!trace_kprobe_on_func_entry(event->tp_event) ||
2175 !trace_kprobe_error_injectable(event->tp_event)))
2178 mutex_lock(&bpf_event_mutex);
2183 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2185 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2190 ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2194 /* set the new array to event->tp_event and set event->prog */
2196 event->bpf_cookie = bpf_cookie;
2197 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2198 bpf_prog_array_free_sleepable(old_array);
2201 mutex_unlock(&bpf_event_mutex);
2205 void perf_event_detach_bpf_prog(struct perf_event *event)
2207 struct bpf_prog_array *old_array;
2208 struct bpf_prog_array *new_array;
2211 mutex_lock(&bpf_event_mutex);
2216 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2217 ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2221 bpf_prog_array_delete_safe(old_array, event->prog);
2223 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2224 bpf_prog_array_free_sleepable(old_array);
2227 bpf_prog_put(event->prog);
2231 mutex_unlock(&bpf_event_mutex);
2234 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2236 struct perf_event_query_bpf __user *uquery = info;
2237 struct perf_event_query_bpf query = {};
2238 struct bpf_prog_array *progs;
2239 u32 *ids, prog_cnt, ids_len;
2242 if (!perfmon_capable())
2244 if (event->attr.type != PERF_TYPE_TRACEPOINT)
2246 if (copy_from_user(&query, uquery, sizeof(query)))
2249 ids_len = query.ids_len;
2250 if (ids_len > BPF_TRACE_MAX_PROGS)
2252 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2256 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2257 * is required when user only wants to check for uquery->prog_cnt.
2258 * There is no need to check for it since the case is handled
2259 * gracefully in bpf_prog_array_copy_info.
2262 mutex_lock(&bpf_event_mutex);
2263 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2264 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2265 mutex_unlock(&bpf_event_mutex);
2267 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2268 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2275 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2276 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2278 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2280 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2282 for (; btp < __stop__bpf_raw_tp; btp++) {
2283 if (!strcmp(btp->tp->name, name))
2287 return bpf_get_raw_tracepoint_module(name);
2290 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2295 mod = __module_address((unsigned long)btp);
2300 static __always_inline
2301 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
2304 if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2305 bpf_prog_inc_misses_counter(prog);
2309 (void) bpf_prog_run(prog, args);
2312 this_cpu_dec(*(prog->active));
2315 #define UNPACK(...) __VA_ARGS__
2316 #define REPEAT_1(FN, DL, X, ...) FN(X)
2317 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2318 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2319 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2320 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2321 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2322 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2323 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2324 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2325 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2326 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2327 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2328 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
2330 #define SARG(X) u64 arg##X
2331 #define COPY(X) args[X] = arg##X
2333 #define __DL_COM (,)
2334 #define __DL_SEM (;)
2336 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2338 #define BPF_TRACE_DEFN_x(x) \
2339 void bpf_trace_run##x(struct bpf_prog *prog, \
2340 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
2343 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
2344 __bpf_trace_run(prog, args); \
2346 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2347 BPF_TRACE_DEFN_x(1);
2348 BPF_TRACE_DEFN_x(2);
2349 BPF_TRACE_DEFN_x(3);
2350 BPF_TRACE_DEFN_x(4);
2351 BPF_TRACE_DEFN_x(5);
2352 BPF_TRACE_DEFN_x(6);
2353 BPF_TRACE_DEFN_x(7);
2354 BPF_TRACE_DEFN_x(8);
2355 BPF_TRACE_DEFN_x(9);
2356 BPF_TRACE_DEFN_x(10);
2357 BPF_TRACE_DEFN_x(11);
2358 BPF_TRACE_DEFN_x(12);
2360 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2362 struct tracepoint *tp = btp->tp;
2365 * check that program doesn't access arguments beyond what's
2366 * available in this tracepoint
2368 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2371 if (prog->aux->max_tp_access > btp->writable_size)
2374 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2378 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2380 return __bpf_probe_register(btp, prog);
2383 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2385 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2388 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2389 u32 *fd_type, const char **buf,
2390 u64 *probe_offset, u64 *probe_addr)
2392 bool is_tracepoint, is_syscall_tp;
2393 struct bpf_prog *prog;
2400 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2401 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2404 *prog_id = prog->aux->id;
2405 flags = event->tp_event->flags;
2406 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2407 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2409 if (is_tracepoint || is_syscall_tp) {
2410 *buf = is_tracepoint ? event->tp_event->tp->name
2411 : event->tp_event->name;
2412 /* We allow NULL pointer for tracepoint */
2414 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2416 *probe_offset = 0x0;
2422 #ifdef CONFIG_KPROBE_EVENTS
2423 if (flags & TRACE_EVENT_FL_KPROBE)
2424 err = bpf_get_kprobe_info(event, fd_type, buf,
2425 probe_offset, probe_addr,
2426 event->attr.type == PERF_TYPE_TRACEPOINT);
2428 #ifdef CONFIG_UPROBE_EVENTS
2429 if (flags & TRACE_EVENT_FL_UPROBE)
2430 err = bpf_get_uprobe_info(event, fd_type, buf,
2431 probe_offset, probe_addr,
2432 event->attr.type == PERF_TYPE_TRACEPOINT);
2439 static int __init send_signal_irq_work_init(void)
2442 struct send_signal_irq_work *work;
2444 for_each_possible_cpu(cpu) {
2445 work = per_cpu_ptr(&send_signal_work, cpu);
2446 init_irq_work(&work->irq_work, do_bpf_send_signal);
2451 subsys_initcall(send_signal_irq_work_init);
2453 #ifdef CONFIG_MODULES
2454 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2457 struct bpf_trace_module *btm, *tmp;
2458 struct module *mod = module;
2461 if (mod->num_bpf_raw_events == 0 ||
2462 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2465 mutex_lock(&bpf_module_mutex);
2468 case MODULE_STATE_COMING:
2469 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2471 btm->module = module;
2472 list_add(&btm->list, &bpf_trace_modules);
2477 case MODULE_STATE_GOING:
2478 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2479 if (btm->module == module) {
2480 list_del(&btm->list);
2488 mutex_unlock(&bpf_module_mutex);
2491 return notifier_from_errno(ret);
2494 static struct notifier_block bpf_module_nb = {
2495 .notifier_call = bpf_event_notify,
2498 static int __init bpf_event_init(void)
2500 register_module_notifier(&bpf_module_nb);
2504 fs_initcall(bpf_event_init);
2505 #endif /* CONFIG_MODULES */
2507 #ifdef CONFIG_FPROBE
2508 struct bpf_kprobe_multi_link {
2509 struct bpf_link link;
2511 unsigned long *addrs;
2515 struct module **mods;
2519 struct bpf_kprobe_multi_run_ctx {
2520 struct bpf_run_ctx run_ctx;
2521 struct bpf_kprobe_multi_link *link;
2522 unsigned long entry_ip;
2530 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2532 unsigned long __user usymbol;
2533 const char **syms = NULL;
2534 char *buf = NULL, *p;
2538 syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2542 buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2546 for (p = buf, i = 0; i < cnt; i++) {
2547 if (__get_user(usymbol, usyms + i)) {
2551 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2552 if (err == KSYM_NAME_LEN)
2572 static void kprobe_multi_put_modules(struct module **mods, u32 cnt)
2576 for (i = 0; i < cnt; i++)
2577 module_put(mods[i]);
2580 static void free_user_syms(struct user_syms *us)
2586 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2588 struct bpf_kprobe_multi_link *kmulti_link;
2590 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2591 unregister_fprobe(&kmulti_link->fp);
2592 kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt);
2595 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2597 struct bpf_kprobe_multi_link *kmulti_link;
2599 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2600 kvfree(kmulti_link->addrs);
2601 kvfree(kmulti_link->cookies);
2602 kfree(kmulti_link->mods);
2606 static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,
2607 struct bpf_link_info *info)
2609 u64 __user *uaddrs = u64_to_user_ptr(info->kprobe_multi.addrs);
2610 struct bpf_kprobe_multi_link *kmulti_link;
2611 u32 ucount = info->kprobe_multi.count;
2614 if (!uaddrs ^ !ucount)
2617 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2618 info->kprobe_multi.count = kmulti_link->cnt;
2619 info->kprobe_multi.flags = kmulti_link->flags;
2623 if (ucount < kmulti_link->cnt)
2626 ucount = kmulti_link->cnt;
2628 if (kallsyms_show_value(current_cred())) {
2629 if (copy_to_user(uaddrs, kmulti_link->addrs, ucount * sizeof(u64)))
2632 for (i = 0; i < ucount; i++) {
2633 if (put_user(0, uaddrs + i))
2640 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2641 .release = bpf_kprobe_multi_link_release,
2642 .dealloc = bpf_kprobe_multi_link_dealloc,
2643 .fill_link_info = bpf_kprobe_multi_link_fill_link_info,
2646 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2648 const struct bpf_kprobe_multi_link *link = priv;
2649 unsigned long *addr_a = a, *addr_b = b;
2650 u64 *cookie_a, *cookie_b;
2652 cookie_a = link->cookies + (addr_a - link->addrs);
2653 cookie_b = link->cookies + (addr_b - link->addrs);
2655 /* swap addr_a/addr_b and cookie_a/cookie_b values */
2656 swap(*addr_a, *addr_b);
2657 swap(*cookie_a, *cookie_b);
2660 static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b)
2662 const unsigned long *addr_a = a, *addr_b = b;
2664 if (*addr_a == *addr_b)
2666 return *addr_a < *addr_b ? -1 : 1;
2669 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2671 return bpf_kprobe_multi_addrs_cmp(a, b);
2674 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2676 struct bpf_kprobe_multi_run_ctx *run_ctx;
2677 struct bpf_kprobe_multi_link *link;
2678 u64 *cookie, entry_ip;
2679 unsigned long *addr;
2681 if (WARN_ON_ONCE(!ctx))
2683 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2684 link = run_ctx->link;
2687 entry_ip = run_ctx->entry_ip;
2688 addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2689 bpf_kprobe_multi_addrs_cmp);
2692 cookie = link->cookies + (addr - link->addrs);
2696 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2698 struct bpf_kprobe_multi_run_ctx *run_ctx;
2700 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx, run_ctx);
2701 return run_ctx->entry_ip;
2705 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2706 unsigned long entry_ip, struct pt_regs *regs)
2708 struct bpf_kprobe_multi_run_ctx run_ctx = {
2710 .entry_ip = entry_ip,
2712 struct bpf_run_ctx *old_run_ctx;
2715 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2722 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2723 err = bpf_prog_run(link->link.prog, regs);
2724 bpf_reset_run_ctx(old_run_ctx);
2729 __this_cpu_dec(bpf_prog_active);
2734 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2735 unsigned long ret_ip, struct pt_regs *regs,
2738 struct bpf_kprobe_multi_link *link;
2740 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2741 kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2746 kprobe_multi_link_exit_handler(struct fprobe *fp, unsigned long fentry_ip,
2747 unsigned long ret_ip, struct pt_regs *regs,
2750 struct bpf_kprobe_multi_link *link;
2752 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2753 kprobe_multi_link_prog_run(link, get_entry_ip(fentry_ip), regs);
2756 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2758 const char **str_a = (const char **) a;
2759 const char **str_b = (const char **) b;
2761 return strcmp(*str_a, *str_b);
2764 struct multi_symbols_sort {
2769 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2771 const struct multi_symbols_sort *data = priv;
2772 const char **name_a = a, **name_b = b;
2774 swap(*name_a, *name_b);
2776 /* If defined, swap also related cookies. */
2777 if (data->cookies) {
2778 u64 *cookie_a, *cookie_b;
2780 cookie_a = data->cookies + (name_a - data->funcs);
2781 cookie_b = data->cookies + (name_b - data->funcs);
2782 swap(*cookie_a, *cookie_b);
2786 struct modules_array {
2787 struct module **mods;
2792 static int add_module(struct modules_array *arr, struct module *mod)
2794 struct module **mods;
2796 if (arr->mods_cnt == arr->mods_cap) {
2797 arr->mods_cap = max(16, arr->mods_cap * 3 / 2);
2798 mods = krealloc_array(arr->mods, arr->mods_cap, sizeof(*mods), GFP_KERNEL);
2804 arr->mods[arr->mods_cnt] = mod;
2809 static bool has_module(struct modules_array *arr, struct module *mod)
2813 for (i = arr->mods_cnt - 1; i >= 0; i--) {
2814 if (arr->mods[i] == mod)
2820 static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt)
2822 struct modules_array arr = {};
2825 for (i = 0; i < addrs_cnt; i++) {
2829 mod = __module_address(addrs[i]);
2830 /* Either no module or we it's already stored */
2831 if (!mod || has_module(&arr, mod)) {
2835 if (!try_module_get(mod))
2840 err = add_module(&arr, mod);
2847 /* We return either err < 0 in case of error, ... */
2849 kprobe_multi_put_modules(arr.mods, arr.mods_cnt);
2854 /* or number of modules found if everything is ok. */
2856 return arr.mods_cnt;
2859 static int addrs_check_error_injection_list(unsigned long *addrs, u32 cnt)
2863 for (i = 0; i < cnt; i++) {
2864 if (!within_error_injection_list(addrs[i]))
2870 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2872 struct bpf_kprobe_multi_link *link = NULL;
2873 struct bpf_link_primer link_primer;
2874 void __user *ucookies;
2875 unsigned long *addrs;
2876 u32 flags, cnt, size;
2877 void __user *uaddrs;
2878 u64 *cookies = NULL;
2882 /* no support for 32bit archs yet */
2883 if (sizeof(u64) != sizeof(void *))
2886 if (prog->expected_attach_type != BPF_TRACE_KPROBE_MULTI)
2889 flags = attr->link_create.kprobe_multi.flags;
2890 if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2893 uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2894 usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2895 if (!!uaddrs == !!usyms)
2898 cnt = attr->link_create.kprobe_multi.cnt;
2901 if (cnt > MAX_KPROBE_MULTI_CNT)
2904 size = cnt * sizeof(*addrs);
2905 addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2909 ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2911 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2916 if (copy_from_user(cookies, ucookies, size)) {
2923 if (copy_from_user(addrs, uaddrs, size)) {
2928 struct multi_symbols_sort data = {
2931 struct user_syms us;
2933 err = copy_user_syms(&us, usyms, cnt);
2938 data.funcs = us.syms;
2940 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2941 symbols_swap_r, &data);
2943 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2944 free_user_syms(&us);
2949 if (prog->kprobe_override && addrs_check_error_injection_list(addrs, cnt)) {
2954 link = kzalloc(sizeof(*link), GFP_KERNEL);
2960 bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2961 &bpf_kprobe_multi_link_lops, prog);
2963 err = bpf_link_prime(&link->link, &link_primer);
2967 if (flags & BPF_F_KPROBE_MULTI_RETURN)
2968 link->fp.exit_handler = kprobe_multi_link_exit_handler;
2970 link->fp.entry_handler = kprobe_multi_link_handler;
2972 link->addrs = addrs;
2973 link->cookies = cookies;
2975 link->flags = flags;
2979 * Sorting addresses will trigger sorting cookies as well
2980 * (check bpf_kprobe_multi_cookie_swap). This way we can
2981 * find cookie based on the address in bpf_get_attach_cookie
2984 sort_r(addrs, cnt, sizeof(*addrs),
2985 bpf_kprobe_multi_cookie_cmp,
2986 bpf_kprobe_multi_cookie_swap,
2990 err = get_modules_for_addrs(&link->mods, addrs, cnt);
2992 bpf_link_cleanup(&link_primer);
2995 link->mods_cnt = err;
2997 err = register_fprobe_ips(&link->fp, addrs, cnt);
2999 kprobe_multi_put_modules(link->mods, link->mods_cnt);
3000 bpf_link_cleanup(&link_primer);
3004 return bpf_link_settle(&link_primer);
3012 #else /* !CONFIG_FPROBE */
3013 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3017 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
3021 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3027 #ifdef CONFIG_UPROBES
3028 struct bpf_uprobe_multi_link;
3031 struct bpf_uprobe_multi_link *link;
3034 struct uprobe_consumer consumer;
3037 struct bpf_uprobe_multi_link {
3039 struct bpf_link link;
3041 struct bpf_uprobe *uprobes;
3042 struct task_struct *task;
3045 struct bpf_uprobe_multi_run_ctx {
3046 struct bpf_run_ctx run_ctx;
3047 unsigned long entry_ip;
3048 struct bpf_uprobe *uprobe;
3051 static void bpf_uprobe_unregister(struct path *path, struct bpf_uprobe *uprobes,
3056 for (i = 0; i < cnt; i++) {
3057 uprobe_unregister(d_real_inode(path->dentry), uprobes[i].offset,
3058 &uprobes[i].consumer);
3062 static void bpf_uprobe_multi_link_release(struct bpf_link *link)
3064 struct bpf_uprobe_multi_link *umulti_link;
3066 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3067 bpf_uprobe_unregister(&umulti_link->path, umulti_link->uprobes, umulti_link->cnt);
3070 static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)
3072 struct bpf_uprobe_multi_link *umulti_link;
3074 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3075 if (umulti_link->task)
3076 put_task_struct(umulti_link->task);
3077 path_put(&umulti_link->path);
3078 kvfree(umulti_link->uprobes);
3082 static const struct bpf_link_ops bpf_uprobe_multi_link_lops = {
3083 .release = bpf_uprobe_multi_link_release,
3084 .dealloc = bpf_uprobe_multi_link_dealloc,
3087 static int uprobe_prog_run(struct bpf_uprobe *uprobe,
3088 unsigned long entry_ip,
3089 struct pt_regs *regs)
3091 struct bpf_uprobe_multi_link *link = uprobe->link;
3092 struct bpf_uprobe_multi_run_ctx run_ctx = {
3093 .entry_ip = entry_ip,
3096 struct bpf_prog *prog = link->link.prog;
3097 bool sleepable = prog->aux->sleepable;
3098 struct bpf_run_ctx *old_run_ctx;
3101 if (link->task && current != link->task)
3105 rcu_read_lock_trace();
3111 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
3112 err = bpf_prog_run(link->link.prog, regs);
3113 bpf_reset_run_ctx(old_run_ctx);
3118 rcu_read_unlock_trace();
3125 uprobe_multi_link_filter(struct uprobe_consumer *con, enum uprobe_filter_ctx ctx,
3126 struct mm_struct *mm)
3128 struct bpf_uprobe *uprobe;
3130 uprobe = container_of(con, struct bpf_uprobe, consumer);
3131 return uprobe->link->task->mm == mm;
3135 uprobe_multi_link_handler(struct uprobe_consumer *con, struct pt_regs *regs)
3137 struct bpf_uprobe *uprobe;
3139 uprobe = container_of(con, struct bpf_uprobe, consumer);
3140 return uprobe_prog_run(uprobe, instruction_pointer(regs), regs);
3144 uprobe_multi_link_ret_handler(struct uprobe_consumer *con, unsigned long func, struct pt_regs *regs)
3146 struct bpf_uprobe *uprobe;
3148 uprobe = container_of(con, struct bpf_uprobe, consumer);
3149 return uprobe_prog_run(uprobe, func, regs);
3152 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3154 struct bpf_uprobe_multi_run_ctx *run_ctx;
3156 run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx, run_ctx);
3157 return run_ctx->entry_ip;
3160 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3162 struct bpf_uprobe_multi_run_ctx *run_ctx;
3164 run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx, run_ctx);
3165 return run_ctx->uprobe->cookie;
3168 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3170 struct bpf_uprobe_multi_link *link = NULL;
3171 unsigned long __user *uref_ctr_offsets;
3172 unsigned long *ref_ctr_offsets = NULL;
3173 struct bpf_link_primer link_primer;
3174 struct bpf_uprobe *uprobes = NULL;
3175 struct task_struct *task = NULL;
3176 unsigned long __user *uoffsets;
3177 u64 __user *ucookies;
3185 /* no support for 32bit archs yet */
3186 if (sizeof(u64) != sizeof(void *))
3189 if (prog->expected_attach_type != BPF_TRACE_UPROBE_MULTI)
3192 flags = attr->link_create.uprobe_multi.flags;
3193 if (flags & ~BPF_F_UPROBE_MULTI_RETURN)
3197 * path, offsets and cnt are mandatory,
3198 * ref_ctr_offsets and cookies are optional
3200 upath = u64_to_user_ptr(attr->link_create.uprobe_multi.path);
3201 uoffsets = u64_to_user_ptr(attr->link_create.uprobe_multi.offsets);
3202 cnt = attr->link_create.uprobe_multi.cnt;
3204 if (!upath || !uoffsets || !cnt)
3206 if (cnt > MAX_UPROBE_MULTI_CNT)
3209 uref_ctr_offsets = u64_to_user_ptr(attr->link_create.uprobe_multi.ref_ctr_offsets);
3210 ucookies = u64_to_user_ptr(attr->link_create.uprobe_multi.cookies);
3212 name = strndup_user(upath, PATH_MAX);
3214 err = PTR_ERR(name);
3218 err = kern_path(name, LOOKUP_FOLLOW, &path);
3223 if (!d_is_reg(path.dentry)) {
3225 goto error_path_put;
3228 pid = attr->link_create.uprobe_multi.pid;
3231 task = get_pid_task(find_vpid(pid), PIDTYPE_PID);
3235 goto error_path_put;
3241 link = kzalloc(sizeof(*link), GFP_KERNEL);
3242 uprobes = kvcalloc(cnt, sizeof(*uprobes), GFP_KERNEL);
3244 if (!uprobes || !link)
3247 if (uref_ctr_offsets) {
3248 ref_ctr_offsets = kvcalloc(cnt, sizeof(*ref_ctr_offsets), GFP_KERNEL);
3249 if (!ref_ctr_offsets)
3253 for (i = 0; i < cnt; i++) {
3254 if (ucookies && __get_user(uprobes[i].cookie, ucookies + i)) {
3258 if (uref_ctr_offsets && __get_user(ref_ctr_offsets[i], uref_ctr_offsets + i)) {
3262 if (__get_user(uprobes[i].offset, uoffsets + i)) {
3267 uprobes[i].link = link;
3269 if (flags & BPF_F_UPROBE_MULTI_RETURN)
3270 uprobes[i].consumer.ret_handler = uprobe_multi_link_ret_handler;
3272 uprobes[i].consumer.handler = uprobe_multi_link_handler;
3275 uprobes[i].consumer.filter = uprobe_multi_link_filter;
3279 link->uprobes = uprobes;
3283 bpf_link_init(&link->link, BPF_LINK_TYPE_UPROBE_MULTI,
3284 &bpf_uprobe_multi_link_lops, prog);
3286 for (i = 0; i < cnt; i++) {
3287 err = uprobe_register_refctr(d_real_inode(link->path.dentry),
3289 ref_ctr_offsets ? ref_ctr_offsets[i] : 0,
3290 &uprobes[i].consumer);
3292 bpf_uprobe_unregister(&path, uprobes, i);
3297 err = bpf_link_prime(&link->link, &link_primer);
3301 kvfree(ref_ctr_offsets);
3302 return bpf_link_settle(&link_primer);
3305 kvfree(ref_ctr_offsets);
3309 put_task_struct(task);
3314 #else /* !CONFIG_UPROBES */
3315 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3319 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3323 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3327 #endif /* CONFIG_UPROBES */