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_perf_event.h>
10 #include <linux/btf.h>
11 #include <linux/filter.h>
12 #include <linux/uaccess.h>
13 #include <linux/ctype.h>
14 #include <linux/kprobes.h>
15 #include <linux/spinlock.h>
16 #include <linux/syscalls.h>
17 #include <linux/error-injection.h>
18 #include <linux/btf_ids.h>
19 #include <linux/bpf_lsm.h>
21 #include <net/bpf_sk_storage.h>
23 #include <uapi/linux/bpf.h>
24 #include <uapi/linux/btf.h>
28 #include "trace_probe.h"
31 #define CREATE_TRACE_POINTS
32 #include "bpf_trace.h"
34 #define bpf_event_rcu_dereference(p) \
35 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
38 struct bpf_trace_module {
39 struct module *module;
40 struct list_head list;
43 static LIST_HEAD(bpf_trace_modules);
44 static DEFINE_MUTEX(bpf_module_mutex);
46 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
48 struct bpf_raw_event_map *btp, *ret = NULL;
49 struct bpf_trace_module *btm;
52 mutex_lock(&bpf_module_mutex);
53 list_for_each_entry(btm, &bpf_trace_modules, list) {
54 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
55 btp = &btm->module->bpf_raw_events[i];
56 if (!strcmp(btp->tp->name, name)) {
57 if (try_module_get(btm->module))
64 mutex_unlock(&bpf_module_mutex);
68 static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
72 #endif /* CONFIG_MODULES */
74 u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
75 u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
77 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
78 u64 flags, const struct btf **btf,
82 * trace_call_bpf - invoke BPF program
83 * @call: tracepoint event
84 * @ctx: opaque context pointer
86 * kprobe handlers execute BPF programs via this helper.
87 * Can be used from static tracepoints in the future.
89 * Return: BPF programs always return an integer which is interpreted by
91 * 0 - return from kprobe (event is filtered out)
92 * 1 - store kprobe event into ring buffer
93 * Other values are reserved and currently alias to 1
95 unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
101 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
103 * since some bpf program is already running on this cpu,
104 * don't call into another bpf program (same or different)
105 * and don't send kprobe event into ring-buffer,
106 * so return zero here
113 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
114 * to all call sites, we did a bpf_prog_array_valid() there to check
115 * whether call->prog_array is empty or not, which is
116 * a heuristic to speed up execution.
118 * If bpf_prog_array_valid() fetched prog_array was
119 * non-NULL, we go into trace_call_bpf() and do the actual
120 * proper rcu_dereference() under RCU lock.
121 * If it turns out that prog_array is NULL then, we bail out.
122 * For the opposite, if the bpf_prog_array_valid() fetched pointer
123 * was NULL, you'll skip the prog_array with the risk of missing
124 * out of events when it was updated in between this and the
125 * rcu_dereference() which is accepted risk.
127 ret = BPF_PROG_RUN_ARRAY(call->prog_array, ctx, bpf_prog_run);
130 __this_cpu_dec(bpf_prog_active);
135 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
136 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
138 regs_set_return_value(regs, rc);
139 override_function_with_return(regs);
143 static const struct bpf_func_proto bpf_override_return_proto = {
144 .func = bpf_override_return,
146 .ret_type = RET_INTEGER,
147 .arg1_type = ARG_PTR_TO_CTX,
148 .arg2_type = ARG_ANYTHING,
152 static __always_inline int
153 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
157 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
158 if (unlikely(ret < 0))
159 memset(dst, 0, size);
163 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
164 const void __user *, unsafe_ptr)
166 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
169 const struct bpf_func_proto bpf_probe_read_user_proto = {
170 .func = bpf_probe_read_user,
172 .ret_type = RET_INTEGER,
173 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
174 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
175 .arg3_type = ARG_ANYTHING,
178 static __always_inline int
179 bpf_probe_read_user_str_common(void *dst, u32 size,
180 const void __user *unsafe_ptr)
185 * NB: We rely on strncpy_from_user() not copying junk past the NUL
186 * terminator into `dst`.
188 * strncpy_from_user() does long-sized strides in the fast path. If the
189 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
190 * then there could be junk after the NUL in `dst`. If user takes `dst`
191 * and keys a hash map with it, then semantically identical strings can
192 * occupy multiple entries in the map.
194 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
195 if (unlikely(ret < 0))
196 memset(dst, 0, size);
200 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
201 const void __user *, unsafe_ptr)
203 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
206 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
207 .func = bpf_probe_read_user_str,
209 .ret_type = RET_INTEGER,
210 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
211 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
212 .arg3_type = ARG_ANYTHING,
215 static __always_inline int
216 bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
220 ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
221 if (unlikely(ret < 0))
222 memset(dst, 0, size);
226 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
227 const void *, unsafe_ptr)
229 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
232 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
233 .func = bpf_probe_read_kernel,
235 .ret_type = RET_INTEGER,
236 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
237 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
238 .arg3_type = ARG_ANYTHING,
241 static __always_inline int
242 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
247 * The strncpy_from_kernel_nofault() call will likely not fill the
248 * entire buffer, but that's okay in this circumstance as we're probing
249 * arbitrary memory anyway similar to bpf_probe_read_*() and might
250 * as well probe the stack. Thus, memory is explicitly cleared
251 * only in error case, so that improper users ignoring return
252 * code altogether don't copy garbage; otherwise length of string
253 * is returned that can be used for bpf_perf_event_output() et al.
255 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
256 if (unlikely(ret < 0))
257 memset(dst, 0, size);
261 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
262 const void *, unsafe_ptr)
264 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
267 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
268 .func = bpf_probe_read_kernel_str,
270 .ret_type = RET_INTEGER,
271 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
272 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
273 .arg3_type = ARG_ANYTHING,
276 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
277 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
278 const void *, unsafe_ptr)
280 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
281 return bpf_probe_read_user_common(dst, size,
282 (__force void __user *)unsafe_ptr);
284 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
287 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
288 .func = bpf_probe_read_compat,
290 .ret_type = RET_INTEGER,
291 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
292 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
293 .arg3_type = ARG_ANYTHING,
296 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
297 const void *, unsafe_ptr)
299 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
300 return bpf_probe_read_user_str_common(dst, size,
301 (__force void __user *)unsafe_ptr);
303 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
306 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
307 .func = bpf_probe_read_compat_str,
309 .ret_type = RET_INTEGER,
310 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
311 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
312 .arg3_type = ARG_ANYTHING,
314 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
316 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
320 * Ensure we're in user context which is safe for the helper to
321 * run. This helper has no business in a kthread.
323 * access_ok() should prevent writing to non-user memory, but in
324 * some situations (nommu, temporary switch, etc) access_ok() does
325 * not provide enough validation, hence the check on KERNEL_DS.
327 * nmi_uaccess_okay() ensures the probe is not run in an interim
328 * state, when the task or mm are switched. This is specifically
329 * required to prevent the use of temporary mm.
332 if (unlikely(in_interrupt() ||
333 current->flags & (PF_KTHREAD | PF_EXITING)))
335 if (unlikely(uaccess_kernel()))
337 if (unlikely(!nmi_uaccess_okay()))
340 return copy_to_user_nofault(unsafe_ptr, src, size);
343 static const struct bpf_func_proto bpf_probe_write_user_proto = {
344 .func = bpf_probe_write_user,
346 .ret_type = RET_INTEGER,
347 .arg1_type = ARG_ANYTHING,
348 .arg2_type = ARG_PTR_TO_MEM,
349 .arg3_type = ARG_CONST_SIZE,
352 static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
354 if (!capable(CAP_SYS_ADMIN))
357 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
358 current->comm, task_pid_nr(current));
360 return &bpf_probe_write_user_proto;
363 static DEFINE_RAW_SPINLOCK(trace_printk_lock);
365 #define MAX_TRACE_PRINTK_VARARGS 3
366 #define BPF_TRACE_PRINTK_SIZE 1024
368 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
369 u64, arg2, u64, arg3)
371 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
373 static char buf[BPF_TRACE_PRINTK_SIZE];
377 ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
378 MAX_TRACE_PRINTK_VARARGS);
382 raw_spin_lock_irqsave(&trace_printk_lock, flags);
383 ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
385 trace_bpf_trace_printk(buf);
386 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
388 bpf_bprintf_cleanup();
393 static const struct bpf_func_proto bpf_trace_printk_proto = {
394 .func = bpf_trace_printk,
396 .ret_type = RET_INTEGER,
397 .arg1_type = ARG_PTR_TO_MEM,
398 .arg2_type = ARG_CONST_SIZE,
401 static void __set_printk_clr_event(void)
404 * This program might be calling bpf_trace_printk,
405 * so enable the associated bpf_trace/bpf_trace_printk event.
406 * Repeat this each time as it is possible a user has
407 * disabled bpf_trace_printk events. By loading a program
408 * calling bpf_trace_printk() however the user has expressed
409 * the intent to see such events.
411 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
412 pr_warn_ratelimited("could not enable bpf_trace_printk events");
415 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
417 __set_printk_clr_event();
418 return &bpf_trace_printk_proto;
421 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, data,
424 static char buf[BPF_TRACE_PRINTK_SIZE];
429 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
432 num_args = data_len / 8;
434 ret = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
438 raw_spin_lock_irqsave(&trace_printk_lock, flags);
439 ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
441 trace_bpf_trace_printk(buf);
442 raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
444 bpf_bprintf_cleanup();
449 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
450 .func = bpf_trace_vprintk,
452 .ret_type = RET_INTEGER,
453 .arg1_type = ARG_PTR_TO_MEM,
454 .arg2_type = ARG_CONST_SIZE,
455 .arg3_type = ARG_PTR_TO_MEM_OR_NULL,
456 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
459 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
461 __set_printk_clr_event();
462 return &bpf_trace_vprintk_proto;
465 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
466 const void *, data, u32, data_len)
471 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
474 num_args = data_len / 8;
476 err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
480 seq_bprintf(m, fmt, bin_args);
482 bpf_bprintf_cleanup();
484 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
487 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
489 static const struct bpf_func_proto bpf_seq_printf_proto = {
490 .func = bpf_seq_printf,
492 .ret_type = RET_INTEGER,
493 .arg1_type = ARG_PTR_TO_BTF_ID,
494 .arg1_btf_id = &btf_seq_file_ids[0],
495 .arg2_type = ARG_PTR_TO_MEM,
496 .arg3_type = ARG_CONST_SIZE,
497 .arg4_type = ARG_PTR_TO_MEM_OR_NULL,
498 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
501 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
503 return seq_write(m, data, len) ? -EOVERFLOW : 0;
506 static const struct bpf_func_proto bpf_seq_write_proto = {
507 .func = bpf_seq_write,
509 .ret_type = RET_INTEGER,
510 .arg1_type = ARG_PTR_TO_BTF_ID,
511 .arg1_btf_id = &btf_seq_file_ids[0],
512 .arg2_type = ARG_PTR_TO_MEM,
513 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
516 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
517 u32, btf_ptr_size, u64, flags)
519 const struct btf *btf;
523 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
527 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
530 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
531 .func = bpf_seq_printf_btf,
533 .ret_type = RET_INTEGER,
534 .arg1_type = ARG_PTR_TO_BTF_ID,
535 .arg1_btf_id = &btf_seq_file_ids[0],
536 .arg2_type = ARG_PTR_TO_MEM,
537 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
538 .arg4_type = ARG_ANYTHING,
541 static __always_inline int
542 get_map_perf_counter(struct bpf_map *map, u64 flags,
543 u64 *value, u64 *enabled, u64 *running)
545 struct bpf_array *array = container_of(map, struct bpf_array, map);
546 unsigned int cpu = smp_processor_id();
547 u64 index = flags & BPF_F_INDEX_MASK;
548 struct bpf_event_entry *ee;
550 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
552 if (index == BPF_F_CURRENT_CPU)
554 if (unlikely(index >= array->map.max_entries))
557 ee = READ_ONCE(array->ptrs[index]);
561 return perf_event_read_local(ee->event, value, enabled, running);
564 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
569 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
571 * this api is ugly since we miss [-22..-2] range of valid
572 * counter values, but that's uapi
579 static const struct bpf_func_proto bpf_perf_event_read_proto = {
580 .func = bpf_perf_event_read,
582 .ret_type = RET_INTEGER,
583 .arg1_type = ARG_CONST_MAP_PTR,
584 .arg2_type = ARG_ANYTHING,
587 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
588 struct bpf_perf_event_value *, buf, u32, size)
592 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
594 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
600 memset(buf, 0, size);
604 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
605 .func = bpf_perf_event_read_value,
607 .ret_type = RET_INTEGER,
608 .arg1_type = ARG_CONST_MAP_PTR,
609 .arg2_type = ARG_ANYTHING,
610 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
611 .arg4_type = ARG_CONST_SIZE,
614 static __always_inline u64
615 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
616 u64 flags, struct perf_sample_data *sd)
618 struct bpf_array *array = container_of(map, struct bpf_array, map);
619 unsigned int cpu = smp_processor_id();
620 u64 index = flags & BPF_F_INDEX_MASK;
621 struct bpf_event_entry *ee;
622 struct perf_event *event;
624 if (index == BPF_F_CURRENT_CPU)
626 if (unlikely(index >= array->map.max_entries))
629 ee = READ_ONCE(array->ptrs[index]);
634 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
635 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
638 if (unlikely(event->oncpu != cpu))
641 return perf_event_output(event, sd, regs);
645 * Support executing tracepoints in normal, irq, and nmi context that each call
646 * bpf_perf_event_output
648 struct bpf_trace_sample_data {
649 struct perf_sample_data sds[3];
652 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
653 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
654 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
655 u64, flags, void *, data, u64, size)
657 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
658 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
659 struct perf_raw_record raw = {
665 struct perf_sample_data *sd;
668 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
673 sd = &sds->sds[nest_level - 1];
675 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
680 perf_sample_data_init(sd, 0, 0);
683 err = __bpf_perf_event_output(regs, map, flags, sd);
686 this_cpu_dec(bpf_trace_nest_level);
690 static const struct bpf_func_proto bpf_perf_event_output_proto = {
691 .func = bpf_perf_event_output,
693 .ret_type = RET_INTEGER,
694 .arg1_type = ARG_PTR_TO_CTX,
695 .arg2_type = ARG_CONST_MAP_PTR,
696 .arg3_type = ARG_ANYTHING,
697 .arg4_type = ARG_PTR_TO_MEM,
698 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
701 static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
702 struct bpf_nested_pt_regs {
703 struct pt_regs regs[3];
705 static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
706 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
708 u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
709 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
711 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
712 struct perf_raw_frag frag = {
717 struct perf_raw_record raw = {
720 .next = ctx_size ? &frag : NULL,
726 struct perf_sample_data *sd;
727 struct pt_regs *regs;
730 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
734 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
735 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
737 perf_fetch_caller_regs(regs);
738 perf_sample_data_init(sd, 0, 0);
741 ret = __bpf_perf_event_output(regs, map, flags, sd);
743 this_cpu_dec(bpf_event_output_nest_level);
747 BPF_CALL_0(bpf_get_current_task)
749 return (long) current;
752 const struct bpf_func_proto bpf_get_current_task_proto = {
753 .func = bpf_get_current_task,
755 .ret_type = RET_INTEGER,
758 BPF_CALL_0(bpf_get_current_task_btf)
760 return (unsigned long) current;
763 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
764 .func = bpf_get_current_task_btf,
766 .ret_type = RET_PTR_TO_BTF_ID,
767 .ret_btf_id = &btf_task_struct_ids[0],
770 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
772 return (unsigned long) task_pt_regs(task);
775 BTF_ID_LIST(bpf_task_pt_regs_ids)
776 BTF_ID(struct, pt_regs)
778 const struct bpf_func_proto bpf_task_pt_regs_proto = {
779 .func = bpf_task_pt_regs,
781 .arg1_type = ARG_PTR_TO_BTF_ID,
782 .arg1_btf_id = &btf_task_struct_ids[0],
783 .ret_type = RET_PTR_TO_BTF_ID,
784 .ret_btf_id = &bpf_task_pt_regs_ids[0],
787 BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
789 struct bpf_array *array = container_of(map, struct bpf_array, map);
792 if (unlikely(idx >= array->map.max_entries))
795 cgrp = READ_ONCE(array->ptrs[idx]);
799 return task_under_cgroup_hierarchy(current, cgrp);
802 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
803 .func = bpf_current_task_under_cgroup,
805 .ret_type = RET_INTEGER,
806 .arg1_type = ARG_CONST_MAP_PTR,
807 .arg2_type = ARG_ANYTHING,
810 struct send_signal_irq_work {
811 struct irq_work irq_work;
812 struct task_struct *task;
817 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
819 static void do_bpf_send_signal(struct irq_work *entry)
821 struct send_signal_irq_work *work;
823 work = container_of(entry, struct send_signal_irq_work, irq_work);
824 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
827 static int bpf_send_signal_common(u32 sig, enum pid_type type)
829 struct send_signal_irq_work *work = NULL;
831 /* Similar to bpf_probe_write_user, task needs to be
832 * in a sound condition and kernel memory access be
833 * permitted in order to send signal to the current
836 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
838 if (unlikely(uaccess_kernel()))
840 if (unlikely(!nmi_uaccess_okay()))
843 if (irqs_disabled()) {
844 /* Do an early check on signal validity. Otherwise,
845 * the error is lost in deferred irq_work.
847 if (unlikely(!valid_signal(sig)))
850 work = this_cpu_ptr(&send_signal_work);
851 if (irq_work_is_busy(&work->irq_work))
854 /* Add the current task, which is the target of sending signal,
855 * to the irq_work. The current task may change when queued
856 * irq works get executed.
858 work->task = current;
861 irq_work_queue(&work->irq_work);
865 return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
868 BPF_CALL_1(bpf_send_signal, u32, sig)
870 return bpf_send_signal_common(sig, PIDTYPE_TGID);
873 static const struct bpf_func_proto bpf_send_signal_proto = {
874 .func = bpf_send_signal,
876 .ret_type = RET_INTEGER,
877 .arg1_type = ARG_ANYTHING,
880 BPF_CALL_1(bpf_send_signal_thread, u32, sig)
882 return bpf_send_signal_common(sig, PIDTYPE_PID);
885 static const struct bpf_func_proto bpf_send_signal_thread_proto = {
886 .func = bpf_send_signal_thread,
888 .ret_type = RET_INTEGER,
889 .arg1_type = ARG_ANYTHING,
892 BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
900 p = d_path(path, buf, sz);
905 memmove(buf, p, len);
911 BTF_SET_START(btf_allowlist_d_path)
912 #ifdef CONFIG_SECURITY
913 BTF_ID(func, security_file_permission)
914 BTF_ID(func, security_inode_getattr)
915 BTF_ID(func, security_file_open)
917 #ifdef CONFIG_SECURITY_PATH
918 BTF_ID(func, security_path_truncate)
920 BTF_ID(func, vfs_truncate)
921 BTF_ID(func, vfs_fallocate)
922 BTF_ID(func, dentry_open)
923 BTF_ID(func, vfs_getattr)
924 BTF_ID(func, filp_close)
925 BTF_SET_END(btf_allowlist_d_path)
927 static bool bpf_d_path_allowed(const struct bpf_prog *prog)
929 if (prog->type == BPF_PROG_TYPE_TRACING &&
930 prog->expected_attach_type == BPF_TRACE_ITER)
933 if (prog->type == BPF_PROG_TYPE_LSM)
934 return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
936 return btf_id_set_contains(&btf_allowlist_d_path,
937 prog->aux->attach_btf_id);
940 BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
942 static const struct bpf_func_proto bpf_d_path_proto = {
945 .ret_type = RET_INTEGER,
946 .arg1_type = ARG_PTR_TO_BTF_ID,
947 .arg1_btf_id = &bpf_d_path_btf_ids[0],
948 .arg2_type = ARG_PTR_TO_MEM,
949 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
950 .allowed = bpf_d_path_allowed,
953 #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \
954 BTF_F_PTR_RAW | BTF_F_ZERO)
956 static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
957 u64 flags, const struct btf **btf,
960 const struct btf_type *t;
962 if (unlikely(flags & ~(BTF_F_ALL)))
965 if (btf_ptr_size != sizeof(struct btf_ptr))
968 *btf = bpf_get_btf_vmlinux();
970 if (IS_ERR_OR_NULL(*btf))
971 return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
973 if (ptr->type_id > 0)
974 *btf_id = ptr->type_id;
979 t = btf_type_by_id(*btf, *btf_id);
980 if (*btf_id <= 0 || !t)
986 BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
987 u32, btf_ptr_size, u64, flags)
989 const struct btf *btf;
993 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
997 return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
1001 const struct bpf_func_proto bpf_snprintf_btf_proto = {
1002 .func = bpf_snprintf_btf,
1004 .ret_type = RET_INTEGER,
1005 .arg1_type = ARG_PTR_TO_MEM,
1006 .arg2_type = ARG_CONST_SIZE,
1007 .arg3_type = ARG_PTR_TO_MEM,
1008 .arg4_type = ARG_CONST_SIZE,
1009 .arg5_type = ARG_ANYTHING,
1012 BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx)
1014 /* This helper call is inlined by verifier. */
1015 return ((u64 *)ctx)[-1];
1018 static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = {
1019 .func = bpf_get_func_ip_tracing,
1021 .ret_type = RET_INTEGER,
1022 .arg1_type = ARG_PTR_TO_CTX,
1025 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1027 struct kprobe *kp = kprobe_running();
1029 return kp ? (uintptr_t)kp->addr : 0;
1032 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1033 .func = bpf_get_func_ip_kprobe,
1035 .ret_type = RET_INTEGER,
1036 .arg1_type = ARG_PTR_TO_CTX,
1039 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1041 struct bpf_trace_run_ctx *run_ctx;
1043 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1044 return run_ctx->bpf_cookie;
1047 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1048 .func = bpf_get_attach_cookie_trace,
1050 .ret_type = RET_INTEGER,
1051 .arg1_type = ARG_PTR_TO_CTX,
1054 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1056 return ctx->event->bpf_cookie;
1059 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1060 .func = bpf_get_attach_cookie_pe,
1062 .ret_type = RET_INTEGER,
1063 .arg1_type = ARG_PTR_TO_CTX,
1066 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1071 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1072 u32 entry_cnt = size / br_entry_size;
1074 entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1076 if (unlikely(flags))
1082 return entry_cnt * br_entry_size;
1086 static const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1087 .func = bpf_get_branch_snapshot,
1089 .ret_type = RET_INTEGER,
1090 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1091 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1094 static const struct bpf_func_proto *
1095 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1098 case BPF_FUNC_map_lookup_elem:
1099 return &bpf_map_lookup_elem_proto;
1100 case BPF_FUNC_map_update_elem:
1101 return &bpf_map_update_elem_proto;
1102 case BPF_FUNC_map_delete_elem:
1103 return &bpf_map_delete_elem_proto;
1104 case BPF_FUNC_map_push_elem:
1105 return &bpf_map_push_elem_proto;
1106 case BPF_FUNC_map_pop_elem:
1107 return &bpf_map_pop_elem_proto;
1108 case BPF_FUNC_map_peek_elem:
1109 return &bpf_map_peek_elem_proto;
1110 case BPF_FUNC_ktime_get_ns:
1111 return &bpf_ktime_get_ns_proto;
1112 case BPF_FUNC_ktime_get_boot_ns:
1113 return &bpf_ktime_get_boot_ns_proto;
1114 case BPF_FUNC_ktime_get_coarse_ns:
1115 return &bpf_ktime_get_coarse_ns_proto;
1116 case BPF_FUNC_tail_call:
1117 return &bpf_tail_call_proto;
1118 case BPF_FUNC_get_current_pid_tgid:
1119 return &bpf_get_current_pid_tgid_proto;
1120 case BPF_FUNC_get_current_task:
1121 return &bpf_get_current_task_proto;
1122 case BPF_FUNC_get_current_task_btf:
1123 return &bpf_get_current_task_btf_proto;
1124 case BPF_FUNC_task_pt_regs:
1125 return &bpf_task_pt_regs_proto;
1126 case BPF_FUNC_get_current_uid_gid:
1127 return &bpf_get_current_uid_gid_proto;
1128 case BPF_FUNC_get_current_comm:
1129 return &bpf_get_current_comm_proto;
1130 case BPF_FUNC_trace_printk:
1131 return bpf_get_trace_printk_proto();
1132 case BPF_FUNC_get_smp_processor_id:
1133 return &bpf_get_smp_processor_id_proto;
1134 case BPF_FUNC_get_numa_node_id:
1135 return &bpf_get_numa_node_id_proto;
1136 case BPF_FUNC_perf_event_read:
1137 return &bpf_perf_event_read_proto;
1138 case BPF_FUNC_current_task_under_cgroup:
1139 return &bpf_current_task_under_cgroup_proto;
1140 case BPF_FUNC_get_prandom_u32:
1141 return &bpf_get_prandom_u32_proto;
1142 case BPF_FUNC_probe_write_user:
1143 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1144 NULL : bpf_get_probe_write_proto();
1145 case BPF_FUNC_probe_read_user:
1146 return &bpf_probe_read_user_proto;
1147 case BPF_FUNC_probe_read_kernel:
1148 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1149 NULL : &bpf_probe_read_kernel_proto;
1150 case BPF_FUNC_probe_read_user_str:
1151 return &bpf_probe_read_user_str_proto;
1152 case BPF_FUNC_probe_read_kernel_str:
1153 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1154 NULL : &bpf_probe_read_kernel_str_proto;
1155 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1156 case BPF_FUNC_probe_read:
1157 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1158 NULL : &bpf_probe_read_compat_proto;
1159 case BPF_FUNC_probe_read_str:
1160 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1161 NULL : &bpf_probe_read_compat_str_proto;
1163 #ifdef CONFIG_CGROUPS
1164 case BPF_FUNC_get_current_cgroup_id:
1165 return &bpf_get_current_cgroup_id_proto;
1166 case BPF_FUNC_get_current_ancestor_cgroup_id:
1167 return &bpf_get_current_ancestor_cgroup_id_proto;
1169 case BPF_FUNC_send_signal:
1170 return &bpf_send_signal_proto;
1171 case BPF_FUNC_send_signal_thread:
1172 return &bpf_send_signal_thread_proto;
1173 case BPF_FUNC_perf_event_read_value:
1174 return &bpf_perf_event_read_value_proto;
1175 case BPF_FUNC_get_ns_current_pid_tgid:
1176 return &bpf_get_ns_current_pid_tgid_proto;
1177 case BPF_FUNC_ringbuf_output:
1178 return &bpf_ringbuf_output_proto;
1179 case BPF_FUNC_ringbuf_reserve:
1180 return &bpf_ringbuf_reserve_proto;
1181 case BPF_FUNC_ringbuf_submit:
1182 return &bpf_ringbuf_submit_proto;
1183 case BPF_FUNC_ringbuf_discard:
1184 return &bpf_ringbuf_discard_proto;
1185 case BPF_FUNC_ringbuf_query:
1186 return &bpf_ringbuf_query_proto;
1187 case BPF_FUNC_jiffies64:
1188 return &bpf_jiffies64_proto;
1189 case BPF_FUNC_get_task_stack:
1190 return &bpf_get_task_stack_proto;
1191 case BPF_FUNC_copy_from_user:
1192 return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1193 case BPF_FUNC_snprintf_btf:
1194 return &bpf_snprintf_btf_proto;
1195 case BPF_FUNC_per_cpu_ptr:
1196 return &bpf_per_cpu_ptr_proto;
1197 case BPF_FUNC_this_cpu_ptr:
1198 return &bpf_this_cpu_ptr_proto;
1199 case BPF_FUNC_task_storage_get:
1200 return &bpf_task_storage_get_proto;
1201 case BPF_FUNC_task_storage_delete:
1202 return &bpf_task_storage_delete_proto;
1203 case BPF_FUNC_for_each_map_elem:
1204 return &bpf_for_each_map_elem_proto;
1205 case BPF_FUNC_snprintf:
1206 return &bpf_snprintf_proto;
1207 case BPF_FUNC_get_func_ip:
1208 return &bpf_get_func_ip_proto_tracing;
1209 case BPF_FUNC_get_branch_snapshot:
1210 return &bpf_get_branch_snapshot_proto;
1211 case BPF_FUNC_trace_vprintk:
1212 return bpf_get_trace_vprintk_proto();
1214 return bpf_base_func_proto(func_id);
1218 static const struct bpf_func_proto *
1219 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1222 case BPF_FUNC_perf_event_output:
1223 return &bpf_perf_event_output_proto;
1224 case BPF_FUNC_get_stackid:
1225 return &bpf_get_stackid_proto;
1226 case BPF_FUNC_get_stack:
1227 return &bpf_get_stack_proto;
1228 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1229 case BPF_FUNC_override_return:
1230 return &bpf_override_return_proto;
1232 case BPF_FUNC_get_func_ip:
1233 return &bpf_get_func_ip_proto_kprobe;
1234 case BPF_FUNC_get_attach_cookie:
1235 return &bpf_get_attach_cookie_proto_trace;
1237 return bpf_tracing_func_proto(func_id, prog);
1241 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1242 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1243 const struct bpf_prog *prog,
1244 struct bpf_insn_access_aux *info)
1246 if (off < 0 || off >= sizeof(struct pt_regs))
1248 if (type != BPF_READ)
1250 if (off % size != 0)
1253 * Assertion for 32 bit to make sure last 8 byte access
1254 * (BPF_DW) to the last 4 byte member is disallowed.
1256 if (off + size > sizeof(struct pt_regs))
1262 const struct bpf_verifier_ops kprobe_verifier_ops = {
1263 .get_func_proto = kprobe_prog_func_proto,
1264 .is_valid_access = kprobe_prog_is_valid_access,
1267 const struct bpf_prog_ops kprobe_prog_ops = {
1270 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1271 u64, flags, void *, data, u64, size)
1273 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1276 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1277 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1278 * from there and call the same bpf_perf_event_output() helper inline.
1280 return ____bpf_perf_event_output(regs, map, flags, data, size);
1283 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1284 .func = bpf_perf_event_output_tp,
1286 .ret_type = RET_INTEGER,
1287 .arg1_type = ARG_PTR_TO_CTX,
1288 .arg2_type = ARG_CONST_MAP_PTR,
1289 .arg3_type = ARG_ANYTHING,
1290 .arg4_type = ARG_PTR_TO_MEM,
1291 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1294 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1297 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1300 * Same comment as in bpf_perf_event_output_tp(), only that this time
1301 * the other helper's function body cannot be inlined due to being
1302 * external, thus we need to call raw helper function.
1304 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1308 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1309 .func = bpf_get_stackid_tp,
1311 .ret_type = RET_INTEGER,
1312 .arg1_type = ARG_PTR_TO_CTX,
1313 .arg2_type = ARG_CONST_MAP_PTR,
1314 .arg3_type = ARG_ANYTHING,
1317 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1320 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1322 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1323 (unsigned long) size, flags, 0);
1326 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1327 .func = bpf_get_stack_tp,
1329 .ret_type = RET_INTEGER,
1330 .arg1_type = ARG_PTR_TO_CTX,
1331 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1332 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1333 .arg4_type = ARG_ANYTHING,
1336 static const struct bpf_func_proto *
1337 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1340 case BPF_FUNC_perf_event_output:
1341 return &bpf_perf_event_output_proto_tp;
1342 case BPF_FUNC_get_stackid:
1343 return &bpf_get_stackid_proto_tp;
1344 case BPF_FUNC_get_stack:
1345 return &bpf_get_stack_proto_tp;
1346 case BPF_FUNC_get_attach_cookie:
1347 return &bpf_get_attach_cookie_proto_trace;
1349 return bpf_tracing_func_proto(func_id, prog);
1353 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1354 const struct bpf_prog *prog,
1355 struct bpf_insn_access_aux *info)
1357 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1359 if (type != BPF_READ)
1361 if (off % size != 0)
1364 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1368 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1369 .get_func_proto = tp_prog_func_proto,
1370 .is_valid_access = tp_prog_is_valid_access,
1373 const struct bpf_prog_ops tracepoint_prog_ops = {
1376 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1377 struct bpf_perf_event_value *, buf, u32, size)
1381 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1383 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1389 memset(buf, 0, size);
1393 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1394 .func = bpf_perf_prog_read_value,
1396 .ret_type = RET_INTEGER,
1397 .arg1_type = ARG_PTR_TO_CTX,
1398 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1399 .arg3_type = ARG_CONST_SIZE,
1402 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1403 void *, buf, u32, size, u64, flags)
1408 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1409 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1412 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1415 if (unlikely(!br_stack))
1418 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1419 return br_stack->nr * br_entry_size;
1421 if (!buf || (size % br_entry_size != 0))
1424 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1425 memcpy(buf, br_stack->entries, to_copy);
1431 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1432 .func = bpf_read_branch_records,
1434 .ret_type = RET_INTEGER,
1435 .arg1_type = ARG_PTR_TO_CTX,
1436 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1437 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1438 .arg4_type = ARG_ANYTHING,
1441 static const struct bpf_func_proto *
1442 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1445 case BPF_FUNC_perf_event_output:
1446 return &bpf_perf_event_output_proto_tp;
1447 case BPF_FUNC_get_stackid:
1448 return &bpf_get_stackid_proto_pe;
1449 case BPF_FUNC_get_stack:
1450 return &bpf_get_stack_proto_pe;
1451 case BPF_FUNC_perf_prog_read_value:
1452 return &bpf_perf_prog_read_value_proto;
1453 case BPF_FUNC_read_branch_records:
1454 return &bpf_read_branch_records_proto;
1455 case BPF_FUNC_get_attach_cookie:
1456 return &bpf_get_attach_cookie_proto_pe;
1458 return bpf_tracing_func_proto(func_id, prog);
1463 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1464 * to avoid potential recursive reuse issue when/if tracepoints are added
1465 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1467 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1468 * in normal, irq, and nmi context.
1470 struct bpf_raw_tp_regs {
1471 struct pt_regs regs[3];
1473 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1474 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1475 static struct pt_regs *get_bpf_raw_tp_regs(void)
1477 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1478 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1480 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1481 this_cpu_dec(bpf_raw_tp_nest_level);
1482 return ERR_PTR(-EBUSY);
1485 return &tp_regs->regs[nest_level - 1];
1488 static void put_bpf_raw_tp_regs(void)
1490 this_cpu_dec(bpf_raw_tp_nest_level);
1493 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1494 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1496 struct pt_regs *regs = get_bpf_raw_tp_regs();
1500 return PTR_ERR(regs);
1502 perf_fetch_caller_regs(regs);
1503 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1505 put_bpf_raw_tp_regs();
1509 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1510 .func = bpf_perf_event_output_raw_tp,
1512 .ret_type = RET_INTEGER,
1513 .arg1_type = ARG_PTR_TO_CTX,
1514 .arg2_type = ARG_CONST_MAP_PTR,
1515 .arg3_type = ARG_ANYTHING,
1516 .arg4_type = ARG_PTR_TO_MEM,
1517 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1520 extern const struct bpf_func_proto bpf_skb_output_proto;
1521 extern const struct bpf_func_proto bpf_xdp_output_proto;
1523 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1524 struct bpf_map *, map, u64, flags)
1526 struct pt_regs *regs = get_bpf_raw_tp_regs();
1530 return PTR_ERR(regs);
1532 perf_fetch_caller_regs(regs);
1533 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1534 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1536 put_bpf_raw_tp_regs();
1540 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1541 .func = bpf_get_stackid_raw_tp,
1543 .ret_type = RET_INTEGER,
1544 .arg1_type = ARG_PTR_TO_CTX,
1545 .arg2_type = ARG_CONST_MAP_PTR,
1546 .arg3_type = ARG_ANYTHING,
1549 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1550 void *, buf, u32, size, u64, flags)
1552 struct pt_regs *regs = get_bpf_raw_tp_regs();
1556 return PTR_ERR(regs);
1558 perf_fetch_caller_regs(regs);
1559 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1560 (unsigned long) size, flags, 0);
1561 put_bpf_raw_tp_regs();
1565 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1566 .func = bpf_get_stack_raw_tp,
1568 .ret_type = RET_INTEGER,
1569 .arg1_type = ARG_PTR_TO_CTX,
1570 .arg2_type = ARG_PTR_TO_MEM,
1571 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1572 .arg4_type = ARG_ANYTHING,
1575 static const struct bpf_func_proto *
1576 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1579 case BPF_FUNC_perf_event_output:
1580 return &bpf_perf_event_output_proto_raw_tp;
1581 case BPF_FUNC_get_stackid:
1582 return &bpf_get_stackid_proto_raw_tp;
1583 case BPF_FUNC_get_stack:
1584 return &bpf_get_stack_proto_raw_tp;
1586 return bpf_tracing_func_proto(func_id, prog);
1590 const struct bpf_func_proto *
1591 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1593 const struct bpf_func_proto *fn;
1597 case BPF_FUNC_skb_output:
1598 return &bpf_skb_output_proto;
1599 case BPF_FUNC_xdp_output:
1600 return &bpf_xdp_output_proto;
1601 case BPF_FUNC_skc_to_tcp6_sock:
1602 return &bpf_skc_to_tcp6_sock_proto;
1603 case BPF_FUNC_skc_to_tcp_sock:
1604 return &bpf_skc_to_tcp_sock_proto;
1605 case BPF_FUNC_skc_to_tcp_timewait_sock:
1606 return &bpf_skc_to_tcp_timewait_sock_proto;
1607 case BPF_FUNC_skc_to_tcp_request_sock:
1608 return &bpf_skc_to_tcp_request_sock_proto;
1609 case BPF_FUNC_skc_to_udp6_sock:
1610 return &bpf_skc_to_udp6_sock_proto;
1611 case BPF_FUNC_skc_to_unix_sock:
1612 return &bpf_skc_to_unix_sock_proto;
1613 case BPF_FUNC_sk_storage_get:
1614 return &bpf_sk_storage_get_tracing_proto;
1615 case BPF_FUNC_sk_storage_delete:
1616 return &bpf_sk_storage_delete_tracing_proto;
1617 case BPF_FUNC_sock_from_file:
1618 return &bpf_sock_from_file_proto;
1619 case BPF_FUNC_get_socket_cookie:
1620 return &bpf_get_socket_ptr_cookie_proto;
1622 case BPF_FUNC_seq_printf:
1623 return prog->expected_attach_type == BPF_TRACE_ITER ?
1624 &bpf_seq_printf_proto :
1626 case BPF_FUNC_seq_write:
1627 return prog->expected_attach_type == BPF_TRACE_ITER ?
1628 &bpf_seq_write_proto :
1630 case BPF_FUNC_seq_printf_btf:
1631 return prog->expected_attach_type == BPF_TRACE_ITER ?
1632 &bpf_seq_printf_btf_proto :
1634 case BPF_FUNC_d_path:
1635 return &bpf_d_path_proto;
1637 fn = raw_tp_prog_func_proto(func_id, prog);
1638 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1639 fn = bpf_iter_get_func_proto(func_id, prog);
1644 static bool raw_tp_prog_is_valid_access(int off, int size,
1645 enum bpf_access_type type,
1646 const struct bpf_prog *prog,
1647 struct bpf_insn_access_aux *info)
1649 return bpf_tracing_ctx_access(off, size, type);
1652 static bool tracing_prog_is_valid_access(int off, int size,
1653 enum bpf_access_type type,
1654 const struct bpf_prog *prog,
1655 struct bpf_insn_access_aux *info)
1657 return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1660 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1661 const union bpf_attr *kattr,
1662 union bpf_attr __user *uattr)
1667 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1668 .get_func_proto = raw_tp_prog_func_proto,
1669 .is_valid_access = raw_tp_prog_is_valid_access,
1672 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1674 .test_run = bpf_prog_test_run_raw_tp,
1678 const struct bpf_verifier_ops tracing_verifier_ops = {
1679 .get_func_proto = tracing_prog_func_proto,
1680 .is_valid_access = tracing_prog_is_valid_access,
1683 const struct bpf_prog_ops tracing_prog_ops = {
1684 .test_run = bpf_prog_test_run_tracing,
1687 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1688 enum bpf_access_type type,
1689 const struct bpf_prog *prog,
1690 struct bpf_insn_access_aux *info)
1693 if (size != sizeof(u64) || type != BPF_READ)
1695 info->reg_type = PTR_TO_TP_BUFFER;
1697 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1700 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1701 .get_func_proto = raw_tp_prog_func_proto,
1702 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1705 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1708 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1709 const struct bpf_prog *prog,
1710 struct bpf_insn_access_aux *info)
1712 const int size_u64 = sizeof(u64);
1714 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1716 if (type != BPF_READ)
1718 if (off % size != 0) {
1719 if (sizeof(unsigned long) != 4)
1723 if (off % size != 4)
1728 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1729 bpf_ctx_record_field_size(info, size_u64);
1730 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1733 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1734 bpf_ctx_record_field_size(info, size_u64);
1735 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1739 if (size != sizeof(long))
1746 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1747 const struct bpf_insn *si,
1748 struct bpf_insn *insn_buf,
1749 struct bpf_prog *prog, u32 *target_size)
1751 struct bpf_insn *insn = insn_buf;
1754 case offsetof(struct bpf_perf_event_data, sample_period):
1755 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1756 data), si->dst_reg, si->src_reg,
1757 offsetof(struct bpf_perf_event_data_kern, data));
1758 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1759 bpf_target_off(struct perf_sample_data, period, 8,
1762 case offsetof(struct bpf_perf_event_data, addr):
1763 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1764 data), si->dst_reg, si->src_reg,
1765 offsetof(struct bpf_perf_event_data_kern, data));
1766 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1767 bpf_target_off(struct perf_sample_data, addr, 8,
1771 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1772 regs), si->dst_reg, si->src_reg,
1773 offsetof(struct bpf_perf_event_data_kern, regs));
1774 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1779 return insn - insn_buf;
1782 const struct bpf_verifier_ops perf_event_verifier_ops = {
1783 .get_func_proto = pe_prog_func_proto,
1784 .is_valid_access = pe_prog_is_valid_access,
1785 .convert_ctx_access = pe_prog_convert_ctx_access,
1788 const struct bpf_prog_ops perf_event_prog_ops = {
1791 static DEFINE_MUTEX(bpf_event_mutex);
1793 #define BPF_TRACE_MAX_PROGS 64
1795 int perf_event_attach_bpf_prog(struct perf_event *event,
1796 struct bpf_prog *prog,
1799 struct bpf_prog_array *old_array;
1800 struct bpf_prog_array *new_array;
1804 * Kprobe override only works if they are on the function entry,
1805 * and only if they are on the opt-in list.
1807 if (prog->kprobe_override &&
1808 (!trace_kprobe_on_func_entry(event->tp_event) ||
1809 !trace_kprobe_error_injectable(event->tp_event)))
1812 mutex_lock(&bpf_event_mutex);
1817 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1819 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1824 ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
1828 /* set the new array to event->tp_event and set event->prog */
1830 event->bpf_cookie = bpf_cookie;
1831 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1832 bpf_prog_array_free(old_array);
1835 mutex_unlock(&bpf_event_mutex);
1839 void perf_event_detach_bpf_prog(struct perf_event *event)
1841 struct bpf_prog_array *old_array;
1842 struct bpf_prog_array *new_array;
1845 mutex_lock(&bpf_event_mutex);
1850 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1851 ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
1855 bpf_prog_array_delete_safe(old_array, event->prog);
1857 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1858 bpf_prog_array_free(old_array);
1861 bpf_prog_put(event->prog);
1865 mutex_unlock(&bpf_event_mutex);
1868 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1870 struct perf_event_query_bpf __user *uquery = info;
1871 struct perf_event_query_bpf query = {};
1872 struct bpf_prog_array *progs;
1873 u32 *ids, prog_cnt, ids_len;
1876 if (!perfmon_capable())
1878 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1880 if (copy_from_user(&query, uquery, sizeof(query)))
1883 ids_len = query.ids_len;
1884 if (ids_len > BPF_TRACE_MAX_PROGS)
1886 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1890 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1891 * is required when user only wants to check for uquery->prog_cnt.
1892 * There is no need to check for it since the case is handled
1893 * gracefully in bpf_prog_array_copy_info.
1896 mutex_lock(&bpf_event_mutex);
1897 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1898 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1899 mutex_unlock(&bpf_event_mutex);
1901 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1902 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1909 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1910 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1912 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1914 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1916 for (; btp < __stop__bpf_raw_tp; btp++) {
1917 if (!strcmp(btp->tp->name, name))
1921 return bpf_get_raw_tracepoint_module(name);
1924 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1929 mod = __module_address((unsigned long)btp);
1934 static __always_inline
1935 void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1939 (void) bpf_prog_run(prog, args);
1943 #define UNPACK(...) __VA_ARGS__
1944 #define REPEAT_1(FN, DL, X, ...) FN(X)
1945 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1946 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1947 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1948 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1949 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1950 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1951 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1952 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1953 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1954 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1955 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1956 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1958 #define SARG(X) u64 arg##X
1959 #define COPY(X) args[X] = arg##X
1961 #define __DL_COM (,)
1962 #define __DL_SEM (;)
1964 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1966 #define BPF_TRACE_DEFN_x(x) \
1967 void bpf_trace_run##x(struct bpf_prog *prog, \
1968 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1971 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1972 __bpf_trace_run(prog, args); \
1974 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1975 BPF_TRACE_DEFN_x(1);
1976 BPF_TRACE_DEFN_x(2);
1977 BPF_TRACE_DEFN_x(3);
1978 BPF_TRACE_DEFN_x(4);
1979 BPF_TRACE_DEFN_x(5);
1980 BPF_TRACE_DEFN_x(6);
1981 BPF_TRACE_DEFN_x(7);
1982 BPF_TRACE_DEFN_x(8);
1983 BPF_TRACE_DEFN_x(9);
1984 BPF_TRACE_DEFN_x(10);
1985 BPF_TRACE_DEFN_x(11);
1986 BPF_TRACE_DEFN_x(12);
1988 static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1990 struct tracepoint *tp = btp->tp;
1993 * check that program doesn't access arguments beyond what's
1994 * available in this tracepoint
1996 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1999 if (prog->aux->max_tp_access > btp->writable_size)
2002 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
2006 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2008 return __bpf_probe_register(btp, prog);
2011 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
2013 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
2016 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2017 u32 *fd_type, const char **buf,
2018 u64 *probe_offset, u64 *probe_addr)
2020 bool is_tracepoint, is_syscall_tp;
2021 struct bpf_prog *prog;
2028 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2029 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2032 *prog_id = prog->aux->id;
2033 flags = event->tp_event->flags;
2034 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2035 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2037 if (is_tracepoint || is_syscall_tp) {
2038 *buf = is_tracepoint ? event->tp_event->tp->name
2039 : event->tp_event->name;
2040 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2041 *probe_offset = 0x0;
2046 #ifdef CONFIG_KPROBE_EVENTS
2047 if (flags & TRACE_EVENT_FL_KPROBE)
2048 err = bpf_get_kprobe_info(event, fd_type, buf,
2049 probe_offset, probe_addr,
2050 event->attr.type == PERF_TYPE_TRACEPOINT);
2052 #ifdef CONFIG_UPROBE_EVENTS
2053 if (flags & TRACE_EVENT_FL_UPROBE)
2054 err = bpf_get_uprobe_info(event, fd_type, buf,
2056 event->attr.type == PERF_TYPE_TRACEPOINT);
2063 static int __init send_signal_irq_work_init(void)
2066 struct send_signal_irq_work *work;
2068 for_each_possible_cpu(cpu) {
2069 work = per_cpu_ptr(&send_signal_work, cpu);
2070 init_irq_work(&work->irq_work, do_bpf_send_signal);
2075 subsys_initcall(send_signal_irq_work_init);
2077 #ifdef CONFIG_MODULES
2078 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2081 struct bpf_trace_module *btm, *tmp;
2082 struct module *mod = module;
2085 if (mod->num_bpf_raw_events == 0 ||
2086 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2089 mutex_lock(&bpf_module_mutex);
2092 case MODULE_STATE_COMING:
2093 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2095 btm->module = module;
2096 list_add(&btm->list, &bpf_trace_modules);
2101 case MODULE_STATE_GOING:
2102 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2103 if (btm->module == module) {
2104 list_del(&btm->list);
2112 mutex_unlock(&bpf_module_mutex);
2115 return notifier_from_errno(ret);
2118 static struct notifier_block bpf_module_nb = {
2119 .notifier_call = bpf_event_notify,
2122 static int __init bpf_event_init(void)
2124 register_module_notifier(&bpf_module_nb);
2128 fs_initcall(bpf_event_init);
2129 #endif /* CONFIG_MODULES */