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
124 bpf_prog_inc_misses_counters(rcu_dereference(call->prog_array));
131 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
132 * to all call sites, we did a bpf_prog_array_valid() there to check
133 * whether call->prog_array is empty or not, which is
134 * a heuristic to speed up execution.
136 * If bpf_prog_array_valid() fetched prog_array was
137 * non-NULL, we go into trace_call_bpf() and do the actual
138 * proper rcu_dereference() under RCU lock.
139 * If it turns out that prog_array is NULL then, we bail out.
140 * For the opposite, if the bpf_prog_array_valid() fetched pointer
141 * was NULL, you'll skip the prog_array with the risk of missing
142 * out of events when it was updated in between this and the
143 * rcu_dereference() which is accepted risk.
146 ret = bpf_prog_run_array(rcu_dereference(call->prog_array),
151 __this_cpu_dec(bpf_prog_active);
156 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
157 BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
159 regs_set_return_value(regs, rc);
160 override_function_with_return(regs);
164 static const struct bpf_func_proto bpf_override_return_proto = {
165 .func = bpf_override_return,
167 .ret_type = RET_INTEGER,
168 .arg1_type = ARG_PTR_TO_CTX,
169 .arg2_type = ARG_ANYTHING,
173 static __always_inline int
174 bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
178 ret = copy_from_user_nofault(dst, unsafe_ptr, size);
179 if (unlikely(ret < 0))
180 memset(dst, 0, size);
184 BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size,
185 const void __user *, unsafe_ptr)
187 return bpf_probe_read_user_common(dst, size, unsafe_ptr);
190 const struct bpf_func_proto bpf_probe_read_user_proto = {
191 .func = bpf_probe_read_user,
193 .ret_type = RET_INTEGER,
194 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
195 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
196 .arg3_type = ARG_ANYTHING,
199 static __always_inline int
200 bpf_probe_read_user_str_common(void *dst, u32 size,
201 const void __user *unsafe_ptr)
206 * NB: We rely on strncpy_from_user() not copying junk past the NUL
207 * terminator into `dst`.
209 * strncpy_from_user() does long-sized strides in the fast path. If the
210 * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`,
211 * then there could be junk after the NUL in `dst`. If user takes `dst`
212 * and keys a hash map with it, then semantically identical strings can
213 * occupy multiple entries in the map.
215 ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
216 if (unlikely(ret < 0))
217 memset(dst, 0, size);
221 BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size,
222 const void __user *, unsafe_ptr)
224 return bpf_probe_read_user_str_common(dst, size, unsafe_ptr);
227 const struct bpf_func_proto bpf_probe_read_user_str_proto = {
228 .func = bpf_probe_read_user_str,
230 .ret_type = RET_INTEGER,
231 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
232 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
233 .arg3_type = ARG_ANYTHING,
236 BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
237 const void *, unsafe_ptr)
239 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
242 const struct bpf_func_proto bpf_probe_read_kernel_proto = {
243 .func = bpf_probe_read_kernel,
245 .ret_type = RET_INTEGER,
246 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
247 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
248 .arg3_type = ARG_ANYTHING,
251 static __always_inline int
252 bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
257 * The strncpy_from_kernel_nofault() call will likely not fill the
258 * entire buffer, but that's okay in this circumstance as we're probing
259 * arbitrary memory anyway similar to bpf_probe_read_*() and might
260 * as well probe the stack. Thus, memory is explicitly cleared
261 * only in error case, so that improper users ignoring return
262 * code altogether don't copy garbage; otherwise length of string
263 * is returned that can be used for bpf_perf_event_output() et al.
265 ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
266 if (unlikely(ret < 0))
267 memset(dst, 0, size);
271 BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
272 const void *, unsafe_ptr)
274 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
277 const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
278 .func = bpf_probe_read_kernel_str,
280 .ret_type = RET_INTEGER,
281 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
282 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
283 .arg3_type = ARG_ANYTHING,
286 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
287 BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size,
288 const void *, unsafe_ptr)
290 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
291 return bpf_probe_read_user_common(dst, size,
292 (__force void __user *)unsafe_ptr);
294 return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
297 static const struct bpf_func_proto bpf_probe_read_compat_proto = {
298 .func = bpf_probe_read_compat,
300 .ret_type = RET_INTEGER,
301 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
302 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
303 .arg3_type = ARG_ANYTHING,
306 BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
307 const void *, unsafe_ptr)
309 if ((unsigned long)unsafe_ptr < TASK_SIZE) {
310 return bpf_probe_read_user_str_common(dst, size,
311 (__force void __user *)unsafe_ptr);
313 return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
316 static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
317 .func = bpf_probe_read_compat_str,
319 .ret_type = RET_INTEGER,
320 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
321 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
322 .arg3_type = ARG_ANYTHING,
324 #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
326 BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
330 * Ensure we're in user context which is safe for the helper to
331 * run. This helper has no business in a kthread.
333 * access_ok() should prevent writing to non-user memory, but in
334 * some situations (nommu, temporary switch, etc) access_ok() does
335 * not provide enough validation, hence the check on KERNEL_DS.
337 * nmi_uaccess_okay() ensures the probe is not run in an interim
338 * state, when the task or mm are switched. This is specifically
339 * required to prevent the use of temporary mm.
342 if (unlikely(in_interrupt() ||
343 current->flags & (PF_KTHREAD | PF_EXITING)))
345 if (unlikely(!nmi_uaccess_okay()))
348 return copy_to_user_nofault(unsafe_ptr, src, size);
351 static const struct bpf_func_proto bpf_probe_write_user_proto = {
352 .func = bpf_probe_write_user,
354 .ret_type = RET_INTEGER,
355 .arg1_type = ARG_ANYTHING,
356 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
357 .arg3_type = ARG_CONST_SIZE,
360 #define MAX_TRACE_PRINTK_VARARGS 3
361 #define BPF_TRACE_PRINTK_SIZE 1024
363 BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
364 u64, arg2, u64, arg3)
366 u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
367 struct bpf_bprintf_data data = {
368 .get_bin_args = true,
373 ret = bpf_bprintf_prepare(fmt, fmt_size, args,
374 MAX_TRACE_PRINTK_VARARGS, &data);
378 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
380 trace_bpf_trace_printk(data.buf);
382 bpf_bprintf_cleanup(&data);
387 static const struct bpf_func_proto bpf_trace_printk_proto = {
388 .func = bpf_trace_printk,
390 .ret_type = RET_INTEGER,
391 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
392 .arg2_type = ARG_CONST_SIZE,
395 static void __set_printk_clr_event(struct work_struct *work)
398 * This program might be calling bpf_trace_printk,
399 * so enable the associated bpf_trace/bpf_trace_printk event.
400 * Repeat this each time as it is possible a user has
401 * disabled bpf_trace_printk events. By loading a program
402 * calling bpf_trace_printk() however the user has expressed
403 * the intent to see such events.
405 if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
406 pr_warn_ratelimited("could not enable bpf_trace_printk events");
408 static DECLARE_WORK(set_printk_work, __set_printk_clr_event);
410 const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
412 schedule_work(&set_printk_work);
413 return &bpf_trace_printk_proto;
416 BPF_CALL_4(bpf_trace_vprintk, char *, fmt, u32, fmt_size, const void *, args,
419 struct bpf_bprintf_data data = {
420 .get_bin_args = true,
425 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
428 num_args = data_len / 8;
430 ret = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
434 ret = bstr_printf(data.buf, MAX_BPRINTF_BUF, fmt, data.bin_args);
436 trace_bpf_trace_printk(data.buf);
438 bpf_bprintf_cleanup(&data);
443 static const struct bpf_func_proto bpf_trace_vprintk_proto = {
444 .func = bpf_trace_vprintk,
446 .ret_type = RET_INTEGER,
447 .arg1_type = ARG_PTR_TO_MEM | MEM_RDONLY,
448 .arg2_type = ARG_CONST_SIZE,
449 .arg3_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
450 .arg4_type = ARG_CONST_SIZE_OR_ZERO,
453 const struct bpf_func_proto *bpf_get_trace_vprintk_proto(void)
455 schedule_work(&set_printk_work);
456 return &bpf_trace_vprintk_proto;
459 BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
460 const void *, args, u32, data_len)
462 struct bpf_bprintf_data data = {
463 .get_bin_args = true,
467 if (data_len & 7 || data_len > MAX_BPRINTF_VARARGS * 8 ||
470 num_args = data_len / 8;
472 err = bpf_bprintf_prepare(fmt, fmt_size, args, num_args, &data);
476 seq_bprintf(m, fmt, data.bin_args);
478 bpf_bprintf_cleanup(&data);
480 return seq_has_overflowed(m) ? -EOVERFLOW : 0;
483 BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
485 static const struct bpf_func_proto bpf_seq_printf_proto = {
486 .func = bpf_seq_printf,
488 .ret_type = RET_INTEGER,
489 .arg1_type = ARG_PTR_TO_BTF_ID,
490 .arg1_btf_id = &btf_seq_file_ids[0],
491 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
492 .arg3_type = ARG_CONST_SIZE,
493 .arg4_type = ARG_PTR_TO_MEM | PTR_MAYBE_NULL | MEM_RDONLY,
494 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
497 BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len)
499 return seq_write(m, data, len) ? -EOVERFLOW : 0;
502 static const struct bpf_func_proto bpf_seq_write_proto = {
503 .func = bpf_seq_write,
505 .ret_type = RET_INTEGER,
506 .arg1_type = ARG_PTR_TO_BTF_ID,
507 .arg1_btf_id = &btf_seq_file_ids[0],
508 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
509 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
512 BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr,
513 u32, btf_ptr_size, u64, flags)
515 const struct btf *btf;
519 ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
523 return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags);
526 static const struct bpf_func_proto bpf_seq_printf_btf_proto = {
527 .func = bpf_seq_printf_btf,
529 .ret_type = RET_INTEGER,
530 .arg1_type = ARG_PTR_TO_BTF_ID,
531 .arg1_btf_id = &btf_seq_file_ids[0],
532 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
533 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
534 .arg4_type = ARG_ANYTHING,
537 static __always_inline int
538 get_map_perf_counter(struct bpf_map *map, u64 flags,
539 u64 *value, u64 *enabled, u64 *running)
541 struct bpf_array *array = container_of(map, struct bpf_array, map);
542 unsigned int cpu = smp_processor_id();
543 u64 index = flags & BPF_F_INDEX_MASK;
544 struct bpf_event_entry *ee;
546 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
548 if (index == BPF_F_CURRENT_CPU)
550 if (unlikely(index >= array->map.max_entries))
553 ee = READ_ONCE(array->ptrs[index]);
557 return perf_event_read_local(ee->event, value, enabled, running);
560 BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
565 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
567 * this api is ugly since we miss [-22..-2] range of valid
568 * counter values, but that's uapi
575 const struct bpf_func_proto bpf_perf_event_read_proto = {
576 .func = bpf_perf_event_read,
578 .ret_type = RET_INTEGER,
579 .arg1_type = ARG_CONST_MAP_PTR,
580 .arg2_type = ARG_ANYTHING,
583 BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
584 struct bpf_perf_event_value *, buf, u32, size)
588 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
590 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
596 memset(buf, 0, size);
600 static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
601 .func = bpf_perf_event_read_value,
603 .ret_type = RET_INTEGER,
604 .arg1_type = ARG_CONST_MAP_PTR,
605 .arg2_type = ARG_ANYTHING,
606 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
607 .arg4_type = ARG_CONST_SIZE,
610 const struct bpf_func_proto *bpf_get_perf_event_read_value_proto(void)
612 return &bpf_perf_event_read_value_proto;
615 static __always_inline u64
616 __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
617 u64 flags, struct perf_raw_record *raw,
618 struct perf_sample_data *sd)
620 struct bpf_array *array = container_of(map, struct bpf_array, map);
621 unsigned int cpu = smp_processor_id();
622 u64 index = flags & BPF_F_INDEX_MASK;
623 struct bpf_event_entry *ee;
624 struct perf_event *event;
626 if (index == BPF_F_CURRENT_CPU)
628 if (unlikely(index >= array->map.max_entries))
631 ee = READ_ONCE(array->ptrs[index]);
636 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
637 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
640 if (unlikely(event->oncpu != cpu))
643 perf_sample_save_raw_data(sd, event, raw);
645 return perf_event_output(event, sd, regs);
649 * Support executing tracepoints in normal, irq, and nmi context that each call
650 * bpf_perf_event_output
652 struct bpf_trace_sample_data {
653 struct perf_sample_data sds[3];
656 static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
657 static DEFINE_PER_CPU(int, bpf_trace_nest_level);
658 BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
659 u64, flags, void *, data, u64, size)
661 struct bpf_trace_sample_data *sds;
662 struct perf_raw_record raw = {
668 struct perf_sample_data *sd;
672 sds = this_cpu_ptr(&bpf_trace_sds);
673 nest_level = this_cpu_inc_return(bpf_trace_nest_level);
675 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
680 sd = &sds->sds[nest_level - 1];
682 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
687 perf_sample_data_init(sd, 0, 0);
689 err = __bpf_perf_event_output(regs, map, flags, &raw, 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);
749 ret = __bpf_perf_event_output(regs, map, flags, &raw, sd);
751 this_cpu_dec(bpf_event_output_nest_level);
756 BPF_CALL_0(bpf_get_current_task)
758 return (long) current;
761 const struct bpf_func_proto bpf_get_current_task_proto = {
762 .func = bpf_get_current_task,
764 .ret_type = RET_INTEGER,
767 BPF_CALL_0(bpf_get_current_task_btf)
769 return (unsigned long) current;
772 const struct bpf_func_proto bpf_get_current_task_btf_proto = {
773 .func = bpf_get_current_task_btf,
775 .ret_type = RET_PTR_TO_BTF_ID_TRUSTED,
776 .ret_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
779 BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task)
781 return (unsigned long) task_pt_regs(task);
784 BTF_ID_LIST(bpf_task_pt_regs_ids)
785 BTF_ID(struct, pt_regs)
787 const struct bpf_func_proto bpf_task_pt_regs_proto = {
788 .func = bpf_task_pt_regs,
790 .arg1_type = ARG_PTR_TO_BTF_ID,
791 .arg1_btf_id = &btf_tracing_ids[BTF_TRACING_TYPE_TASK],
792 .ret_type = RET_PTR_TO_BTF_ID,
793 .ret_btf_id = &bpf_task_pt_regs_ids[0],
796 struct send_signal_irq_work {
797 struct irq_work irq_work;
798 struct task_struct *task;
802 struct kernel_siginfo info;
805 static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
807 static void do_bpf_send_signal(struct irq_work *entry)
809 struct send_signal_irq_work *work;
810 struct kernel_siginfo *siginfo;
812 work = container_of(entry, struct send_signal_irq_work, irq_work);
813 siginfo = work->has_siginfo ? &work->info : SEND_SIG_PRIV;
815 group_send_sig_info(work->sig, siginfo, work->task, work->type);
816 put_task_struct(work->task);
819 static int bpf_send_signal_common(u32 sig, enum pid_type type, struct task_struct *task, u64 value)
821 struct send_signal_irq_work *work = NULL;
822 struct kernel_siginfo info;
823 struct kernel_siginfo *siginfo;
827 siginfo = SEND_SIG_PRIV;
829 clear_siginfo(&info);
832 info.si_code = SI_KERNEL;
835 info.si_value.sival_ptr = (void *)(unsigned long)value;
839 /* Similar to bpf_probe_write_user, task needs to be
840 * in a sound condition and kernel memory access be
841 * permitted in order to send signal to the current
844 if (unlikely(task->flags & (PF_KTHREAD | PF_EXITING)))
846 if (unlikely(!nmi_uaccess_okay()))
848 /* Task should not be pid=1 to avoid kernel panic. */
849 if (unlikely(is_global_init(task)))
852 if (preempt_count() != 0 || irqs_disabled()) {
853 /* Do an early check on signal validity. Otherwise,
854 * the error is lost in deferred irq_work.
856 if (unlikely(!valid_signal(sig)))
859 work = this_cpu_ptr(&send_signal_work);
860 if (irq_work_is_busy(&work->irq_work))
863 /* Add the current task, which is the target of sending signal,
864 * to the irq_work. The current task may change when queued
865 * irq works get executed.
867 work->task = get_task_struct(task);
868 work->has_siginfo = siginfo == &info;
869 if (work->has_siginfo)
870 copy_siginfo(&work->info, &info);
873 irq_work_queue(&work->irq_work);
877 return group_send_sig_info(sig, siginfo, task, type);
880 BPF_CALL_1(bpf_send_signal, u32, sig)
882 return bpf_send_signal_common(sig, PIDTYPE_TGID, NULL, 0);
885 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, NULL, 0);
897 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 static inline unsigned long get_entry_ip(unsigned long fentry_ip)
1049 #ifdef CONFIG_X86_KERNEL_IBT
1050 if (is_endbr((void *)(fentry_ip - ENDBR_INSN_SIZE)))
1051 fentry_ip -= ENDBR_INSN_SIZE;
1056 BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs)
1058 struct bpf_trace_run_ctx *run_ctx __maybe_unused;
1061 #ifdef CONFIG_UPROBES
1062 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1063 if (run_ctx->is_uprobe)
1064 return ((struct uprobe_dispatch_data *)current->utask->vaddr)->bp_addr;
1067 kp = kprobe_running();
1069 if (!kp || !(kp->flags & KPROBE_FLAG_ON_FUNC_ENTRY))
1072 return get_entry_ip((uintptr_t)kp->addr);
1075 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = {
1076 .func = bpf_get_func_ip_kprobe,
1078 .ret_type = RET_INTEGER,
1079 .arg1_type = ARG_PTR_TO_CTX,
1082 BPF_CALL_1(bpf_get_func_ip_kprobe_multi, struct pt_regs *, regs)
1084 return bpf_kprobe_multi_entry_ip(current->bpf_ctx);
1087 static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe_multi = {
1088 .func = bpf_get_func_ip_kprobe_multi,
1090 .ret_type = RET_INTEGER,
1091 .arg1_type = ARG_PTR_TO_CTX,
1094 BPF_CALL_1(bpf_get_attach_cookie_kprobe_multi, struct pt_regs *, regs)
1096 return bpf_kprobe_multi_cookie(current->bpf_ctx);
1099 static const struct bpf_func_proto bpf_get_attach_cookie_proto_kmulti = {
1100 .func = bpf_get_attach_cookie_kprobe_multi,
1102 .ret_type = RET_INTEGER,
1103 .arg1_type = ARG_PTR_TO_CTX,
1106 BPF_CALL_1(bpf_get_func_ip_uprobe_multi, struct pt_regs *, regs)
1108 return bpf_uprobe_multi_entry_ip(current->bpf_ctx);
1111 static const struct bpf_func_proto bpf_get_func_ip_proto_uprobe_multi = {
1112 .func = bpf_get_func_ip_uprobe_multi,
1114 .ret_type = RET_INTEGER,
1115 .arg1_type = ARG_PTR_TO_CTX,
1118 BPF_CALL_1(bpf_get_attach_cookie_uprobe_multi, struct pt_regs *, regs)
1120 return bpf_uprobe_multi_cookie(current->bpf_ctx);
1123 static const struct bpf_func_proto bpf_get_attach_cookie_proto_umulti = {
1124 .func = bpf_get_attach_cookie_uprobe_multi,
1126 .ret_type = RET_INTEGER,
1127 .arg1_type = ARG_PTR_TO_CTX,
1130 BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx)
1132 struct bpf_trace_run_ctx *run_ctx;
1134 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1135 return run_ctx->bpf_cookie;
1138 static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = {
1139 .func = bpf_get_attach_cookie_trace,
1141 .ret_type = RET_INTEGER,
1142 .arg1_type = ARG_PTR_TO_CTX,
1145 BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx)
1147 return ctx->event->bpf_cookie;
1150 static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = {
1151 .func = bpf_get_attach_cookie_pe,
1153 .ret_type = RET_INTEGER,
1154 .arg1_type = ARG_PTR_TO_CTX,
1157 BPF_CALL_1(bpf_get_attach_cookie_tracing, void *, ctx)
1159 struct bpf_trace_run_ctx *run_ctx;
1161 run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx);
1162 return run_ctx->bpf_cookie;
1165 static const struct bpf_func_proto bpf_get_attach_cookie_proto_tracing = {
1166 .func = bpf_get_attach_cookie_tracing,
1168 .ret_type = RET_INTEGER,
1169 .arg1_type = ARG_PTR_TO_CTX,
1172 BPF_CALL_3(bpf_get_branch_snapshot, void *, buf, u32, size, u64, flags)
1174 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1175 u32 entry_cnt = size / br_entry_size;
1177 entry_cnt = static_call(perf_snapshot_branch_stack)(buf, entry_cnt);
1179 if (unlikely(flags))
1185 return entry_cnt * br_entry_size;
1188 const struct bpf_func_proto bpf_get_branch_snapshot_proto = {
1189 .func = bpf_get_branch_snapshot,
1191 .ret_type = RET_INTEGER,
1192 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
1193 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
1196 BPF_CALL_3(get_func_arg, void *, ctx, u32, n, u64 *, value)
1198 /* This helper call is inlined by verifier. */
1199 u64 nr_args = ((u64 *)ctx)[-1];
1201 if ((u64) n >= nr_args)
1203 *value = ((u64 *)ctx)[n];
1207 static const struct bpf_func_proto bpf_get_func_arg_proto = {
1208 .func = get_func_arg,
1209 .ret_type = RET_INTEGER,
1210 .arg1_type = ARG_PTR_TO_CTX,
1211 .arg2_type = ARG_ANYTHING,
1212 .arg3_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
1213 .arg3_size = sizeof(u64),
1216 BPF_CALL_2(get_func_ret, void *, ctx, u64 *, value)
1218 /* This helper call is inlined by verifier. */
1219 u64 nr_args = ((u64 *)ctx)[-1];
1221 *value = ((u64 *)ctx)[nr_args];
1225 static const struct bpf_func_proto bpf_get_func_ret_proto = {
1226 .func = get_func_ret,
1227 .ret_type = RET_INTEGER,
1228 .arg1_type = ARG_PTR_TO_CTX,
1229 .arg2_type = ARG_PTR_TO_FIXED_SIZE_MEM | MEM_UNINIT | MEM_WRITE | MEM_ALIGNED,
1230 .arg2_size = sizeof(u64),
1233 BPF_CALL_1(get_func_arg_cnt, void *, ctx)
1235 /* This helper call is inlined by verifier. */
1236 return ((u64 *)ctx)[-1];
1239 static const struct bpf_func_proto bpf_get_func_arg_cnt_proto = {
1240 .func = get_func_arg_cnt,
1241 .ret_type = RET_INTEGER,
1242 .arg1_type = ARG_PTR_TO_CTX,
1246 __bpf_kfunc_start_defs();
1249 * bpf_lookup_user_key - lookup a key by its serial
1250 * @serial: key handle serial number
1251 * @flags: lookup-specific flags
1253 * Search a key with a given *serial* and the provided *flags*.
1254 * If found, increment the reference count of the key by one, and
1255 * return it in the bpf_key structure.
1257 * The bpf_key structure must be passed to bpf_key_put() when done
1258 * with it, so that the key reference count is decremented and the
1259 * bpf_key structure is freed.
1261 * Permission checks are deferred to the time the key is used by
1262 * one of the available key-specific kfuncs.
1264 * Set *flags* with KEY_LOOKUP_CREATE, to attempt creating a requested
1265 * special keyring (e.g. session keyring), if it doesn't yet exist.
1266 * Set *flags* with KEY_LOOKUP_PARTIAL, to lookup a key without waiting
1267 * for the key construction, and to retrieve uninstantiated keys (keys
1268 * without data attached to them).
1270 * Return: a bpf_key pointer with a valid key pointer if the key is found, a
1271 * NULL pointer otherwise.
1273 __bpf_kfunc struct bpf_key *bpf_lookup_user_key(u32 serial, u64 flags)
1276 struct bpf_key *bkey;
1278 if (flags & ~KEY_LOOKUP_ALL)
1282 * Permission check is deferred until the key is used, as the
1283 * intent of the caller is unknown here.
1285 key_ref = lookup_user_key(serial, flags, KEY_DEFER_PERM_CHECK);
1286 if (IS_ERR(key_ref))
1289 bkey = kmalloc(sizeof(*bkey), GFP_KERNEL);
1291 key_put(key_ref_to_ptr(key_ref));
1295 bkey->key = key_ref_to_ptr(key_ref);
1296 bkey->has_ref = true;
1302 * bpf_lookup_system_key - lookup a key by a system-defined ID
1305 * Obtain a bpf_key structure with a key pointer set to the passed key ID.
1306 * The key pointer is marked as invalid, to prevent bpf_key_put() from
1307 * attempting to decrement the key reference count on that pointer. The key
1308 * pointer set in such way is currently understood only by
1309 * verify_pkcs7_signature().
1311 * Set *id* to one of the values defined in include/linux/verification.h:
1312 * 0 for the primary keyring (immutable keyring of system keys);
1313 * VERIFY_USE_SECONDARY_KEYRING for both the primary and secondary keyring
1314 * (where keys can be added only if they are vouched for by existing keys
1315 * in those keyrings); VERIFY_USE_PLATFORM_KEYRING for the platform
1316 * keyring (primarily used by the integrity subsystem to verify a kexec'ed
1317 * kerned image and, possibly, the initramfs signature).
1319 * Return: a bpf_key pointer with an invalid key pointer set from the
1320 * pre-determined ID on success, a NULL pointer otherwise
1322 __bpf_kfunc struct bpf_key *bpf_lookup_system_key(u64 id)
1324 struct bpf_key *bkey;
1326 if (system_keyring_id_check(id) < 0)
1329 bkey = kmalloc(sizeof(*bkey), GFP_ATOMIC);
1333 bkey->key = (struct key *)(unsigned long)id;
1334 bkey->has_ref = false;
1340 * bpf_key_put - decrement key reference count if key is valid and free bpf_key
1341 * @bkey: bpf_key structure
1343 * Decrement the reference count of the key inside *bkey*, if the pointer
1344 * is valid, and free *bkey*.
1346 __bpf_kfunc void bpf_key_put(struct bpf_key *bkey)
1354 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1356 * bpf_verify_pkcs7_signature - verify a PKCS#7 signature
1357 * @data_p: data to verify
1358 * @sig_p: signature of the data
1359 * @trusted_keyring: keyring with keys trusted for signature verification
1361 * Verify the PKCS#7 signature *sig_ptr* against the supplied *data_ptr*
1362 * with keys in a keyring referenced by *trusted_keyring*.
1364 * Return: 0 on success, a negative value on error.
1366 __bpf_kfunc int bpf_verify_pkcs7_signature(struct bpf_dynptr *data_p,
1367 struct bpf_dynptr *sig_p,
1368 struct bpf_key *trusted_keyring)
1370 struct bpf_dynptr_kern *data_ptr = (struct bpf_dynptr_kern *)data_p;
1371 struct bpf_dynptr_kern *sig_ptr = (struct bpf_dynptr_kern *)sig_p;
1372 const void *data, *sig;
1373 u32 data_len, sig_len;
1376 if (trusted_keyring->has_ref) {
1378 * Do the permission check deferred in bpf_lookup_user_key().
1379 * See bpf_lookup_user_key() for more details.
1381 * A call to key_task_permission() here would be redundant, as
1382 * it is already done by keyring_search() called by
1383 * find_asymmetric_key().
1385 ret = key_validate(trusted_keyring->key);
1390 data_len = __bpf_dynptr_size(data_ptr);
1391 data = __bpf_dynptr_data(data_ptr, data_len);
1392 sig_len = __bpf_dynptr_size(sig_ptr);
1393 sig = __bpf_dynptr_data(sig_ptr, sig_len);
1395 return verify_pkcs7_signature(data, data_len, sig, sig_len,
1396 trusted_keyring->key,
1397 VERIFYING_UNSPECIFIED_SIGNATURE, NULL,
1400 #endif /* CONFIG_SYSTEM_DATA_VERIFICATION */
1402 __bpf_kfunc_end_defs();
1404 BTF_KFUNCS_START(key_sig_kfunc_set)
1405 BTF_ID_FLAGS(func, bpf_lookup_user_key, KF_ACQUIRE | KF_RET_NULL | KF_SLEEPABLE)
1406 BTF_ID_FLAGS(func, bpf_lookup_system_key, KF_ACQUIRE | KF_RET_NULL)
1407 BTF_ID_FLAGS(func, bpf_key_put, KF_RELEASE)
1408 #ifdef CONFIG_SYSTEM_DATA_VERIFICATION
1409 BTF_ID_FLAGS(func, bpf_verify_pkcs7_signature, KF_SLEEPABLE)
1411 BTF_KFUNCS_END(key_sig_kfunc_set)
1413 static const struct btf_kfunc_id_set bpf_key_sig_kfunc_set = {
1414 .owner = THIS_MODULE,
1415 .set = &key_sig_kfunc_set,
1418 static int __init bpf_key_sig_kfuncs_init(void)
1420 return register_btf_kfunc_id_set(BPF_PROG_TYPE_TRACING,
1421 &bpf_key_sig_kfunc_set);
1424 late_initcall(bpf_key_sig_kfuncs_init);
1425 #endif /* CONFIG_KEYS */
1427 static const struct bpf_func_proto *
1428 bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1430 const struct bpf_func_proto *func_proto;
1433 case BPF_FUNC_get_smp_processor_id:
1434 return &bpf_get_smp_processor_id_proto;
1435 #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1436 case BPF_FUNC_probe_read:
1437 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1438 NULL : &bpf_probe_read_compat_proto;
1439 case BPF_FUNC_probe_read_str:
1440 return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1441 NULL : &bpf_probe_read_compat_str_proto;
1443 case BPF_FUNC_get_func_ip:
1444 return &bpf_get_func_ip_proto_tracing;
1449 func_proto = bpf_base_func_proto(func_id, prog);
1453 if (!bpf_token_capable(prog->aux->token, CAP_SYS_ADMIN))
1457 case BPF_FUNC_probe_write_user:
1458 return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
1459 NULL : &bpf_probe_write_user_proto;
1465 static bool is_kprobe_multi(const struct bpf_prog *prog)
1467 return prog->expected_attach_type == BPF_TRACE_KPROBE_MULTI ||
1468 prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION;
1471 static inline bool is_kprobe_session(const struct bpf_prog *prog)
1473 return prog->expected_attach_type == BPF_TRACE_KPROBE_SESSION;
1476 static inline bool is_uprobe_multi(const struct bpf_prog *prog)
1478 return prog->expected_attach_type == BPF_TRACE_UPROBE_MULTI ||
1479 prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION;
1482 static inline bool is_uprobe_session(const struct bpf_prog *prog)
1484 return prog->expected_attach_type == BPF_TRACE_UPROBE_SESSION;
1487 static const struct bpf_func_proto *
1488 kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1491 case BPF_FUNC_perf_event_output:
1492 return &bpf_perf_event_output_proto;
1493 case BPF_FUNC_get_stackid:
1494 return &bpf_get_stackid_proto;
1495 case BPF_FUNC_get_stack:
1496 return prog->sleepable ? &bpf_get_stack_sleepable_proto : &bpf_get_stack_proto;
1497 #ifdef CONFIG_BPF_KPROBE_OVERRIDE
1498 case BPF_FUNC_override_return:
1499 return &bpf_override_return_proto;
1501 case BPF_FUNC_get_func_ip:
1502 if (is_kprobe_multi(prog))
1503 return &bpf_get_func_ip_proto_kprobe_multi;
1504 if (is_uprobe_multi(prog))
1505 return &bpf_get_func_ip_proto_uprobe_multi;
1506 return &bpf_get_func_ip_proto_kprobe;
1507 case BPF_FUNC_get_attach_cookie:
1508 if (is_kprobe_multi(prog))
1509 return &bpf_get_attach_cookie_proto_kmulti;
1510 if (is_uprobe_multi(prog))
1511 return &bpf_get_attach_cookie_proto_umulti;
1512 return &bpf_get_attach_cookie_proto_trace;
1514 return bpf_tracing_func_proto(func_id, prog);
1518 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
1519 static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1520 const struct bpf_prog *prog,
1521 struct bpf_insn_access_aux *info)
1523 if (off < 0 || off >= sizeof(struct pt_regs))
1525 if (type != BPF_READ)
1527 if (off % size != 0)
1530 * Assertion for 32 bit to make sure last 8 byte access
1531 * (BPF_DW) to the last 4 byte member is disallowed.
1533 if (off + size > sizeof(struct pt_regs))
1539 const struct bpf_verifier_ops kprobe_verifier_ops = {
1540 .get_func_proto = kprobe_prog_func_proto,
1541 .is_valid_access = kprobe_prog_is_valid_access,
1544 const struct bpf_prog_ops kprobe_prog_ops = {
1547 BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1548 u64, flags, void *, data, u64, size)
1550 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1553 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1554 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1555 * from there and call the same bpf_perf_event_output() helper inline.
1557 return ____bpf_perf_event_output(regs, map, flags, data, size);
1560 static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1561 .func = bpf_perf_event_output_tp,
1563 .ret_type = RET_INTEGER,
1564 .arg1_type = ARG_PTR_TO_CTX,
1565 .arg2_type = ARG_CONST_MAP_PTR,
1566 .arg3_type = ARG_ANYTHING,
1567 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1568 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1571 BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1574 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1577 * Same comment as in bpf_perf_event_output_tp(), only that this time
1578 * the other helper's function body cannot be inlined due to being
1579 * external, thus we need to call raw helper function.
1581 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1585 static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1586 .func = bpf_get_stackid_tp,
1588 .ret_type = RET_INTEGER,
1589 .arg1_type = ARG_PTR_TO_CTX,
1590 .arg2_type = ARG_CONST_MAP_PTR,
1591 .arg3_type = ARG_ANYTHING,
1594 BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1597 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1599 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1600 (unsigned long) size, flags, 0);
1603 static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1604 .func = bpf_get_stack_tp,
1606 .ret_type = RET_INTEGER,
1607 .arg1_type = ARG_PTR_TO_CTX,
1608 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1609 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1610 .arg4_type = ARG_ANYTHING,
1613 static const struct bpf_func_proto *
1614 tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1617 case BPF_FUNC_perf_event_output:
1618 return &bpf_perf_event_output_proto_tp;
1619 case BPF_FUNC_get_stackid:
1620 return &bpf_get_stackid_proto_tp;
1621 case BPF_FUNC_get_stack:
1622 return &bpf_get_stack_proto_tp;
1623 case BPF_FUNC_get_attach_cookie:
1624 return &bpf_get_attach_cookie_proto_trace;
1626 return bpf_tracing_func_proto(func_id, prog);
1630 static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1631 const struct bpf_prog *prog,
1632 struct bpf_insn_access_aux *info)
1634 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1636 if (type != BPF_READ)
1638 if (off % size != 0)
1641 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1645 const struct bpf_verifier_ops tracepoint_verifier_ops = {
1646 .get_func_proto = tp_prog_func_proto,
1647 .is_valid_access = tp_prog_is_valid_access,
1650 const struct bpf_prog_ops tracepoint_prog_ops = {
1653 BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1654 struct bpf_perf_event_value *, buf, u32, size)
1658 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1660 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1666 memset(buf, 0, size);
1670 static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1671 .func = bpf_perf_prog_read_value,
1673 .ret_type = RET_INTEGER,
1674 .arg1_type = ARG_PTR_TO_CTX,
1675 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
1676 .arg3_type = ARG_CONST_SIZE,
1679 BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1680 void *, buf, u32, size, u64, flags)
1682 static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1683 struct perf_branch_stack *br_stack = ctx->data->br_stack;
1686 if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1689 if (unlikely(!(ctx->data->sample_flags & PERF_SAMPLE_BRANCH_STACK)))
1692 if (unlikely(!br_stack))
1695 if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1696 return br_stack->nr * br_entry_size;
1698 if (!buf || (size % br_entry_size != 0))
1701 to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1702 memcpy(buf, br_stack->entries, to_copy);
1707 static const struct bpf_func_proto bpf_read_branch_records_proto = {
1708 .func = bpf_read_branch_records,
1710 .ret_type = RET_INTEGER,
1711 .arg1_type = ARG_PTR_TO_CTX,
1712 .arg2_type = ARG_PTR_TO_MEM_OR_NULL,
1713 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1714 .arg4_type = ARG_ANYTHING,
1717 static const struct bpf_func_proto *
1718 pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1721 case BPF_FUNC_perf_event_output:
1722 return &bpf_perf_event_output_proto_tp;
1723 case BPF_FUNC_get_stackid:
1724 return &bpf_get_stackid_proto_pe;
1725 case BPF_FUNC_get_stack:
1726 return &bpf_get_stack_proto_pe;
1727 case BPF_FUNC_perf_prog_read_value:
1728 return &bpf_perf_prog_read_value_proto;
1729 case BPF_FUNC_read_branch_records:
1730 return &bpf_read_branch_records_proto;
1731 case BPF_FUNC_get_attach_cookie:
1732 return &bpf_get_attach_cookie_proto_pe;
1734 return bpf_tracing_func_proto(func_id, prog);
1739 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1740 * to avoid potential recursive reuse issue when/if tracepoints are added
1741 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1743 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1744 * in normal, irq, and nmi context.
1746 struct bpf_raw_tp_regs {
1747 struct pt_regs regs[3];
1749 static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1750 static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1751 static struct pt_regs *get_bpf_raw_tp_regs(void)
1753 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1754 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1756 if (nest_level > ARRAY_SIZE(tp_regs->regs)) {
1757 this_cpu_dec(bpf_raw_tp_nest_level);
1758 return ERR_PTR(-EBUSY);
1761 return &tp_regs->regs[nest_level - 1];
1764 static void put_bpf_raw_tp_regs(void)
1766 this_cpu_dec(bpf_raw_tp_nest_level);
1769 BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1770 struct bpf_map *, map, u64, flags, void *, data, u64, size)
1772 struct pt_regs *regs = get_bpf_raw_tp_regs();
1776 return PTR_ERR(regs);
1778 perf_fetch_caller_regs(regs);
1779 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1781 put_bpf_raw_tp_regs();
1785 static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1786 .func = bpf_perf_event_output_raw_tp,
1788 .ret_type = RET_INTEGER,
1789 .arg1_type = ARG_PTR_TO_CTX,
1790 .arg2_type = ARG_CONST_MAP_PTR,
1791 .arg3_type = ARG_ANYTHING,
1792 .arg4_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1793 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
1796 extern const struct bpf_func_proto bpf_skb_output_proto;
1797 extern const struct bpf_func_proto bpf_xdp_output_proto;
1798 extern const struct bpf_func_proto bpf_xdp_get_buff_len_trace_proto;
1800 BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1801 struct bpf_map *, map, u64, flags)
1803 struct pt_regs *regs = get_bpf_raw_tp_regs();
1807 return PTR_ERR(regs);
1809 perf_fetch_caller_regs(regs);
1810 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1811 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1813 put_bpf_raw_tp_regs();
1817 static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1818 .func = bpf_get_stackid_raw_tp,
1820 .ret_type = RET_INTEGER,
1821 .arg1_type = ARG_PTR_TO_CTX,
1822 .arg2_type = ARG_CONST_MAP_PTR,
1823 .arg3_type = ARG_ANYTHING,
1826 BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1827 void *, buf, u32, size, u64, flags)
1829 struct pt_regs *regs = get_bpf_raw_tp_regs();
1833 return PTR_ERR(regs);
1835 perf_fetch_caller_regs(regs);
1836 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1837 (unsigned long) size, flags, 0);
1838 put_bpf_raw_tp_regs();
1842 static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1843 .func = bpf_get_stack_raw_tp,
1845 .ret_type = RET_INTEGER,
1846 .arg1_type = ARG_PTR_TO_CTX,
1847 .arg2_type = ARG_PTR_TO_MEM | MEM_RDONLY,
1848 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1849 .arg4_type = ARG_ANYTHING,
1852 static const struct bpf_func_proto *
1853 raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1856 case BPF_FUNC_perf_event_output:
1857 return &bpf_perf_event_output_proto_raw_tp;
1858 case BPF_FUNC_get_stackid:
1859 return &bpf_get_stackid_proto_raw_tp;
1860 case BPF_FUNC_get_stack:
1861 return &bpf_get_stack_proto_raw_tp;
1862 case BPF_FUNC_get_attach_cookie:
1863 return &bpf_get_attach_cookie_proto_tracing;
1865 return bpf_tracing_func_proto(func_id, prog);
1869 const struct bpf_func_proto *
1870 tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1872 const struct bpf_func_proto *fn;
1876 case BPF_FUNC_skb_output:
1877 return &bpf_skb_output_proto;
1878 case BPF_FUNC_xdp_output:
1879 return &bpf_xdp_output_proto;
1880 case BPF_FUNC_skc_to_tcp6_sock:
1881 return &bpf_skc_to_tcp6_sock_proto;
1882 case BPF_FUNC_skc_to_tcp_sock:
1883 return &bpf_skc_to_tcp_sock_proto;
1884 case BPF_FUNC_skc_to_tcp_timewait_sock:
1885 return &bpf_skc_to_tcp_timewait_sock_proto;
1886 case BPF_FUNC_skc_to_tcp_request_sock:
1887 return &bpf_skc_to_tcp_request_sock_proto;
1888 case BPF_FUNC_skc_to_udp6_sock:
1889 return &bpf_skc_to_udp6_sock_proto;
1890 case BPF_FUNC_skc_to_unix_sock:
1891 return &bpf_skc_to_unix_sock_proto;
1892 case BPF_FUNC_skc_to_mptcp_sock:
1893 return &bpf_skc_to_mptcp_sock_proto;
1894 case BPF_FUNC_sk_storage_get:
1895 return &bpf_sk_storage_get_tracing_proto;
1896 case BPF_FUNC_sk_storage_delete:
1897 return &bpf_sk_storage_delete_tracing_proto;
1898 case BPF_FUNC_sock_from_file:
1899 return &bpf_sock_from_file_proto;
1900 case BPF_FUNC_get_socket_cookie:
1901 return &bpf_get_socket_ptr_cookie_proto;
1902 case BPF_FUNC_xdp_get_buff_len:
1903 return &bpf_xdp_get_buff_len_trace_proto;
1905 case BPF_FUNC_seq_printf:
1906 return prog->expected_attach_type == BPF_TRACE_ITER ?
1907 &bpf_seq_printf_proto :
1909 case BPF_FUNC_seq_write:
1910 return prog->expected_attach_type == BPF_TRACE_ITER ?
1911 &bpf_seq_write_proto :
1913 case BPF_FUNC_seq_printf_btf:
1914 return prog->expected_attach_type == BPF_TRACE_ITER ?
1915 &bpf_seq_printf_btf_proto :
1917 case BPF_FUNC_d_path:
1918 return &bpf_d_path_proto;
1919 case BPF_FUNC_get_func_arg:
1920 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_proto : NULL;
1921 case BPF_FUNC_get_func_ret:
1922 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_ret_proto : NULL;
1923 case BPF_FUNC_get_func_arg_cnt:
1924 return bpf_prog_has_trampoline(prog) ? &bpf_get_func_arg_cnt_proto : NULL;
1925 case BPF_FUNC_get_attach_cookie:
1926 if (prog->type == BPF_PROG_TYPE_TRACING &&
1927 prog->expected_attach_type == BPF_TRACE_RAW_TP)
1928 return &bpf_get_attach_cookie_proto_tracing;
1929 return bpf_prog_has_trampoline(prog) ? &bpf_get_attach_cookie_proto_tracing : NULL;
1931 fn = raw_tp_prog_func_proto(func_id, prog);
1932 if (!fn && prog->expected_attach_type == BPF_TRACE_ITER)
1933 fn = bpf_iter_get_func_proto(func_id, prog);
1938 static bool raw_tp_prog_is_valid_access(int off, int size,
1939 enum bpf_access_type type,
1940 const struct bpf_prog *prog,
1941 struct bpf_insn_access_aux *info)
1943 return bpf_tracing_ctx_access(off, size, type);
1946 static bool tracing_prog_is_valid_access(int off, int size,
1947 enum bpf_access_type type,
1948 const struct bpf_prog *prog,
1949 struct bpf_insn_access_aux *info)
1951 return bpf_tracing_btf_ctx_access(off, size, type, prog, info);
1954 int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1955 const union bpf_attr *kattr,
1956 union bpf_attr __user *uattr)
1961 const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1962 .get_func_proto = raw_tp_prog_func_proto,
1963 .is_valid_access = raw_tp_prog_is_valid_access,
1966 const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1968 .test_run = bpf_prog_test_run_raw_tp,
1972 const struct bpf_verifier_ops tracing_verifier_ops = {
1973 .get_func_proto = tracing_prog_func_proto,
1974 .is_valid_access = tracing_prog_is_valid_access,
1977 const struct bpf_prog_ops tracing_prog_ops = {
1978 .test_run = bpf_prog_test_run_tracing,
1981 static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1982 enum bpf_access_type type,
1983 const struct bpf_prog *prog,
1984 struct bpf_insn_access_aux *info)
1987 if (size != sizeof(u64) || type != BPF_READ)
1989 info->reg_type = PTR_TO_TP_BUFFER;
1991 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1994 const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1995 .get_func_proto = raw_tp_prog_func_proto,
1996 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1999 const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
2002 static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
2003 const struct bpf_prog *prog,
2004 struct bpf_insn_access_aux *info)
2006 const int size_u64 = sizeof(u64);
2008 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
2010 if (type != BPF_READ)
2012 if (off % size != 0) {
2013 if (sizeof(unsigned long) != 4)
2017 if (off % size != 4)
2022 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
2023 bpf_ctx_record_field_size(info, size_u64);
2024 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2027 case bpf_ctx_range(struct bpf_perf_event_data, addr):
2028 bpf_ctx_record_field_size(info, size_u64);
2029 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
2033 if (size != sizeof(long))
2040 static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
2041 const struct bpf_insn *si,
2042 struct bpf_insn *insn_buf,
2043 struct bpf_prog *prog, u32 *target_size)
2045 struct bpf_insn *insn = insn_buf;
2048 case offsetof(struct bpf_perf_event_data, sample_period):
2049 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2050 data), si->dst_reg, si->src_reg,
2051 offsetof(struct bpf_perf_event_data_kern, data));
2052 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2053 bpf_target_off(struct perf_sample_data, period, 8,
2056 case offsetof(struct bpf_perf_event_data, addr):
2057 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2058 data), si->dst_reg, si->src_reg,
2059 offsetof(struct bpf_perf_event_data_kern, data));
2060 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
2061 bpf_target_off(struct perf_sample_data, addr, 8,
2065 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
2066 regs), si->dst_reg, si->src_reg,
2067 offsetof(struct bpf_perf_event_data_kern, regs));
2068 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
2073 return insn - insn_buf;
2076 const struct bpf_verifier_ops perf_event_verifier_ops = {
2077 .get_func_proto = pe_prog_func_proto,
2078 .is_valid_access = pe_prog_is_valid_access,
2079 .convert_ctx_access = pe_prog_convert_ctx_access,
2082 const struct bpf_prog_ops perf_event_prog_ops = {
2085 static DEFINE_MUTEX(bpf_event_mutex);
2087 #define BPF_TRACE_MAX_PROGS 64
2089 int perf_event_attach_bpf_prog(struct perf_event *event,
2090 struct bpf_prog *prog,
2093 struct bpf_prog_array *old_array;
2094 struct bpf_prog_array *new_array;
2098 * Kprobe override only works if they are on the function entry,
2099 * and only if they are on the opt-in list.
2101 if (prog->kprobe_override &&
2102 (!trace_kprobe_on_func_entry(event->tp_event) ||
2103 !trace_kprobe_error_injectable(event->tp_event)))
2106 mutex_lock(&bpf_event_mutex);
2111 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2113 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
2118 ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array);
2122 /* set the new array to event->tp_event and set event->prog */
2124 event->bpf_cookie = bpf_cookie;
2125 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2126 bpf_prog_array_free_sleepable(old_array);
2129 mutex_unlock(&bpf_event_mutex);
2133 void perf_event_detach_bpf_prog(struct perf_event *event)
2135 struct bpf_prog_array *old_array;
2136 struct bpf_prog_array *new_array;
2137 struct bpf_prog *prog = NULL;
2140 mutex_lock(&bpf_event_mutex);
2145 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
2149 ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array);
2151 bpf_prog_array_delete_safe(old_array, event->prog);
2153 rcu_assign_pointer(event->tp_event->prog_array, new_array);
2154 bpf_prog_array_free_sleepable(old_array);
2162 mutex_unlock(&bpf_event_mutex);
2166 * It could be that the bpf_prog is not sleepable (and will be freed
2167 * via normal RCU), but is called from a point that supports sleepable
2168 * programs and uses tasks-trace-RCU.
2170 synchronize_rcu_tasks_trace();
2176 int perf_event_query_prog_array(struct perf_event *event, void __user *info)
2178 struct perf_event_query_bpf __user *uquery = info;
2179 struct perf_event_query_bpf query = {};
2180 struct bpf_prog_array *progs;
2181 u32 *ids, prog_cnt, ids_len;
2184 if (!perfmon_capable())
2186 if (event->attr.type != PERF_TYPE_TRACEPOINT)
2188 if (copy_from_user(&query, uquery, sizeof(query)))
2191 ids_len = query.ids_len;
2192 if (ids_len > BPF_TRACE_MAX_PROGS)
2194 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
2198 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
2199 * is required when user only wants to check for uquery->prog_cnt.
2200 * There is no need to check for it since the case is handled
2201 * gracefully in bpf_prog_array_copy_info.
2204 mutex_lock(&bpf_event_mutex);
2205 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
2206 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
2207 mutex_unlock(&bpf_event_mutex);
2209 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
2210 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
2217 extern struct bpf_raw_event_map __start__bpf_raw_tp[];
2218 extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
2220 struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
2222 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
2224 for (; btp < __stop__bpf_raw_tp; btp++) {
2225 if (!strcmp(btp->tp->name, name))
2229 return bpf_get_raw_tracepoint_module(name);
2232 void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
2237 mod = __module_address((unsigned long)btp);
2241 static __always_inline
2242 void __bpf_trace_run(struct bpf_raw_tp_link *link, u64 *args)
2244 struct bpf_prog *prog = link->link.prog;
2245 struct bpf_run_ctx *old_run_ctx;
2246 struct bpf_trace_run_ctx run_ctx;
2249 if (unlikely(this_cpu_inc_return(*(prog->active)) != 1)) {
2250 bpf_prog_inc_misses_counter(prog);
2254 run_ctx.bpf_cookie = link->cookie;
2255 old_run_ctx = bpf_set_run_ctx(&run_ctx.run_ctx);
2258 (void) bpf_prog_run(prog, args);
2261 bpf_reset_run_ctx(old_run_ctx);
2263 this_cpu_dec(*(prog->active));
2266 #define UNPACK(...) __VA_ARGS__
2267 #define REPEAT_1(FN, DL, X, ...) FN(X)
2268 #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
2269 #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
2270 #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
2271 #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
2272 #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
2273 #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
2274 #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
2275 #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
2276 #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
2277 #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
2278 #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
2279 #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
2281 #define SARG(X) u64 arg##X
2282 #define COPY(X) args[X] = arg##X
2284 #define __DL_COM (,)
2285 #define __DL_SEM (;)
2287 #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
2289 #define BPF_TRACE_DEFN_x(x) \
2290 void bpf_trace_run##x(struct bpf_raw_tp_link *link, \
2291 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
2294 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
2295 __bpf_trace_run(link, args); \
2297 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
2298 BPF_TRACE_DEFN_x(1);
2299 BPF_TRACE_DEFN_x(2);
2300 BPF_TRACE_DEFN_x(3);
2301 BPF_TRACE_DEFN_x(4);
2302 BPF_TRACE_DEFN_x(5);
2303 BPF_TRACE_DEFN_x(6);
2304 BPF_TRACE_DEFN_x(7);
2305 BPF_TRACE_DEFN_x(8);
2306 BPF_TRACE_DEFN_x(9);
2307 BPF_TRACE_DEFN_x(10);
2308 BPF_TRACE_DEFN_x(11);
2309 BPF_TRACE_DEFN_x(12);
2311 int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_raw_tp_link *link)
2313 struct tracepoint *tp = btp->tp;
2314 struct bpf_prog *prog = link->link.prog;
2317 * check that program doesn't access arguments beyond what's
2318 * available in this tracepoint
2320 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
2323 if (prog->aux->max_tp_access > btp->writable_size)
2326 return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func, link);
2329 int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_raw_tp_link *link)
2331 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, link);
2334 int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
2335 u32 *fd_type, const char **buf,
2336 u64 *probe_offset, u64 *probe_addr,
2337 unsigned long *missed)
2339 bool is_tracepoint, is_syscall_tp;
2340 struct bpf_prog *prog;
2347 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
2348 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
2351 *prog_id = prog->aux->id;
2352 flags = event->tp_event->flags;
2353 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
2354 is_syscall_tp = is_syscall_trace_event(event->tp_event);
2356 if (is_tracepoint || is_syscall_tp) {
2357 *buf = is_tracepoint ? event->tp_event->tp->name
2358 : event->tp_event->name;
2359 /* We allow NULL pointer for tracepoint */
2361 *fd_type = BPF_FD_TYPE_TRACEPOINT;
2363 *probe_offset = 0x0;
2369 #ifdef CONFIG_KPROBE_EVENTS
2370 if (flags & TRACE_EVENT_FL_KPROBE)
2371 err = bpf_get_kprobe_info(event, fd_type, buf,
2372 probe_offset, probe_addr, missed,
2373 event->attr.type == PERF_TYPE_TRACEPOINT);
2375 #ifdef CONFIG_UPROBE_EVENTS
2376 if (flags & TRACE_EVENT_FL_UPROBE)
2377 err = bpf_get_uprobe_info(event, fd_type, buf,
2378 probe_offset, probe_addr,
2379 event->attr.type == PERF_TYPE_TRACEPOINT);
2386 static int __init send_signal_irq_work_init(void)
2389 struct send_signal_irq_work *work;
2391 for_each_possible_cpu(cpu) {
2392 work = per_cpu_ptr(&send_signal_work, cpu);
2393 init_irq_work(&work->irq_work, do_bpf_send_signal);
2398 subsys_initcall(send_signal_irq_work_init);
2400 #ifdef CONFIG_MODULES
2401 static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
2404 struct bpf_trace_module *btm, *tmp;
2405 struct module *mod = module;
2408 if (mod->num_bpf_raw_events == 0 ||
2409 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
2412 mutex_lock(&bpf_module_mutex);
2415 case MODULE_STATE_COMING:
2416 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
2418 btm->module = module;
2419 list_add(&btm->list, &bpf_trace_modules);
2424 case MODULE_STATE_GOING:
2425 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
2426 if (btm->module == module) {
2427 list_del(&btm->list);
2435 mutex_unlock(&bpf_module_mutex);
2438 return notifier_from_errno(ret);
2441 static struct notifier_block bpf_module_nb = {
2442 .notifier_call = bpf_event_notify,
2445 static int __init bpf_event_init(void)
2447 register_module_notifier(&bpf_module_nb);
2451 fs_initcall(bpf_event_init);
2452 #endif /* CONFIG_MODULES */
2454 struct bpf_session_run_ctx {
2455 struct bpf_run_ctx run_ctx;
2460 #ifdef CONFIG_FPROBE
2461 struct bpf_kprobe_multi_link {
2462 struct bpf_link link;
2464 unsigned long *addrs;
2468 struct module **mods;
2472 struct bpf_kprobe_multi_run_ctx {
2473 struct bpf_session_run_ctx session_ctx;
2474 struct bpf_kprobe_multi_link *link;
2475 unsigned long entry_ip;
2483 #ifndef CONFIG_HAVE_FTRACE_REGS_HAVING_PT_REGS
2484 static DEFINE_PER_CPU(struct pt_regs, bpf_kprobe_multi_pt_regs);
2485 #define bpf_kprobe_multi_pt_regs_ptr() this_cpu_ptr(&bpf_kprobe_multi_pt_regs)
2487 #define bpf_kprobe_multi_pt_regs_ptr() (NULL)
2490 static unsigned long ftrace_get_entry_ip(unsigned long fentry_ip)
2492 unsigned long ip = ftrace_get_symaddr(fentry_ip);
2494 return ip ? : fentry_ip;
2497 static int copy_user_syms(struct user_syms *us, unsigned long __user *usyms, u32 cnt)
2499 unsigned long __user usymbol;
2500 const char **syms = NULL;
2501 char *buf = NULL, *p;
2505 syms = kvmalloc_array(cnt, sizeof(*syms), GFP_KERNEL);
2509 buf = kvmalloc_array(cnt, KSYM_NAME_LEN, GFP_KERNEL);
2513 for (p = buf, i = 0; i < cnt; i++) {
2514 if (__get_user(usymbol, usyms + i)) {
2518 err = strncpy_from_user(p, (const char __user *) usymbol, KSYM_NAME_LEN);
2519 if (err == KSYM_NAME_LEN)
2539 static void kprobe_multi_put_modules(struct module **mods, u32 cnt)
2543 for (i = 0; i < cnt; i++)
2544 module_put(mods[i]);
2547 static void free_user_syms(struct user_syms *us)
2553 static void bpf_kprobe_multi_link_release(struct bpf_link *link)
2555 struct bpf_kprobe_multi_link *kmulti_link;
2557 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2558 unregister_fprobe(&kmulti_link->fp);
2559 kprobe_multi_put_modules(kmulti_link->mods, kmulti_link->mods_cnt);
2562 static void bpf_kprobe_multi_link_dealloc(struct bpf_link *link)
2564 struct bpf_kprobe_multi_link *kmulti_link;
2566 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2567 kvfree(kmulti_link->addrs);
2568 kvfree(kmulti_link->cookies);
2569 kfree(kmulti_link->mods);
2573 static int bpf_kprobe_multi_link_fill_link_info(const struct bpf_link *link,
2574 struct bpf_link_info *info)
2576 u64 __user *ucookies = u64_to_user_ptr(info->kprobe_multi.cookies);
2577 u64 __user *uaddrs = u64_to_user_ptr(info->kprobe_multi.addrs);
2578 struct bpf_kprobe_multi_link *kmulti_link;
2579 u32 ucount = info->kprobe_multi.count;
2582 if (!uaddrs ^ !ucount)
2584 if (ucookies && !ucount)
2587 kmulti_link = container_of(link, struct bpf_kprobe_multi_link, link);
2588 info->kprobe_multi.count = kmulti_link->cnt;
2589 info->kprobe_multi.flags = kmulti_link->flags;
2590 info->kprobe_multi.missed = kmulti_link->fp.nmissed;
2594 if (ucount < kmulti_link->cnt)
2597 ucount = kmulti_link->cnt;
2600 if (kmulti_link->cookies) {
2601 if (copy_to_user(ucookies, kmulti_link->cookies, ucount * sizeof(u64)))
2604 for (i = 0; i < ucount; i++) {
2605 if (put_user(0, ucookies + i))
2611 if (kallsyms_show_value(current_cred())) {
2612 if (copy_to_user(uaddrs, kmulti_link->addrs, ucount * sizeof(u64)))
2615 for (i = 0; i < ucount; i++) {
2616 if (put_user(0, uaddrs + i))
2623 static const struct bpf_link_ops bpf_kprobe_multi_link_lops = {
2624 .release = bpf_kprobe_multi_link_release,
2625 .dealloc_deferred = bpf_kprobe_multi_link_dealloc,
2626 .fill_link_info = bpf_kprobe_multi_link_fill_link_info,
2629 static void bpf_kprobe_multi_cookie_swap(void *a, void *b, int size, const void *priv)
2631 const struct bpf_kprobe_multi_link *link = priv;
2632 unsigned long *addr_a = a, *addr_b = b;
2633 u64 *cookie_a, *cookie_b;
2635 cookie_a = link->cookies + (addr_a - link->addrs);
2636 cookie_b = link->cookies + (addr_b - link->addrs);
2638 /* swap addr_a/addr_b and cookie_a/cookie_b values */
2639 swap(*addr_a, *addr_b);
2640 swap(*cookie_a, *cookie_b);
2643 static int bpf_kprobe_multi_addrs_cmp(const void *a, const void *b)
2645 const unsigned long *addr_a = a, *addr_b = b;
2647 if (*addr_a == *addr_b)
2649 return *addr_a < *addr_b ? -1 : 1;
2652 static int bpf_kprobe_multi_cookie_cmp(const void *a, const void *b, const void *priv)
2654 return bpf_kprobe_multi_addrs_cmp(a, b);
2657 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
2659 struct bpf_kprobe_multi_run_ctx *run_ctx;
2660 struct bpf_kprobe_multi_link *link;
2661 u64 *cookie, entry_ip;
2662 unsigned long *addr;
2664 if (WARN_ON_ONCE(!ctx))
2666 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx,
2667 session_ctx.run_ctx);
2668 link = run_ctx->link;
2671 entry_ip = run_ctx->entry_ip;
2672 addr = bsearch(&entry_ip, link->addrs, link->cnt, sizeof(entry_ip),
2673 bpf_kprobe_multi_addrs_cmp);
2676 cookie = link->cookies + (addr - link->addrs);
2680 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
2682 struct bpf_kprobe_multi_run_ctx *run_ctx;
2684 run_ctx = container_of(current->bpf_ctx, struct bpf_kprobe_multi_run_ctx,
2685 session_ctx.run_ctx);
2686 return run_ctx->entry_ip;
2690 kprobe_multi_link_prog_run(struct bpf_kprobe_multi_link *link,
2691 unsigned long entry_ip, struct ftrace_regs *fregs,
2692 bool is_return, void *data)
2694 struct bpf_kprobe_multi_run_ctx run_ctx = {
2696 .is_return = is_return,
2700 .entry_ip = entry_ip,
2702 struct bpf_run_ctx *old_run_ctx;
2703 struct pt_regs *regs;
2706 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
2707 bpf_prog_inc_misses_counter(link->link.prog);
2714 regs = ftrace_partial_regs(fregs, bpf_kprobe_multi_pt_regs_ptr());
2715 old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx);
2716 err = bpf_prog_run(link->link.prog, regs);
2717 bpf_reset_run_ctx(old_run_ctx);
2722 __this_cpu_dec(bpf_prog_active);
2727 kprobe_multi_link_handler(struct fprobe *fp, unsigned long fentry_ip,
2728 unsigned long ret_ip, struct ftrace_regs *fregs,
2731 struct bpf_kprobe_multi_link *link;
2734 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2735 err = kprobe_multi_link_prog_run(link, ftrace_get_entry_ip(fentry_ip),
2736 fregs, false, data);
2737 return is_kprobe_session(link->link.prog) ? err : 0;
2741 kprobe_multi_link_exit_handler(struct fprobe *fp, unsigned long fentry_ip,
2742 unsigned long ret_ip, struct ftrace_regs *fregs,
2745 struct bpf_kprobe_multi_link *link;
2747 link = container_of(fp, struct bpf_kprobe_multi_link, fp);
2748 kprobe_multi_link_prog_run(link, ftrace_get_entry_ip(fentry_ip),
2752 static int symbols_cmp_r(const void *a, const void *b, const void *priv)
2754 const char **str_a = (const char **) a;
2755 const char **str_b = (const char **) b;
2757 return strcmp(*str_a, *str_b);
2760 struct multi_symbols_sort {
2765 static void symbols_swap_r(void *a, void *b, int size, const void *priv)
2767 const struct multi_symbols_sort *data = priv;
2768 const char **name_a = a, **name_b = b;
2770 swap(*name_a, *name_b);
2772 /* If defined, swap also related cookies. */
2773 if (data->cookies) {
2774 u64 *cookie_a, *cookie_b;
2776 cookie_a = data->cookies + (name_a - data->funcs);
2777 cookie_b = data->cookies + (name_b - data->funcs);
2778 swap(*cookie_a, *cookie_b);
2782 struct modules_array {
2783 struct module **mods;
2788 static int add_module(struct modules_array *arr, struct module *mod)
2790 struct module **mods;
2792 if (arr->mods_cnt == arr->mods_cap) {
2793 arr->mods_cap = max(16, arr->mods_cap * 3 / 2);
2794 mods = krealloc_array(arr->mods, arr->mods_cap, sizeof(*mods), GFP_KERNEL);
2800 arr->mods[arr->mods_cnt] = mod;
2805 static bool has_module(struct modules_array *arr, struct module *mod)
2809 for (i = arr->mods_cnt - 1; i >= 0; i--) {
2810 if (arr->mods[i] == mod)
2816 static int get_modules_for_addrs(struct module ***mods, unsigned long *addrs, u32 addrs_cnt)
2818 struct modules_array arr = {};
2821 for (i = 0; i < addrs_cnt; i++) {
2822 bool skip_add = false;
2826 mod = __module_address(addrs[i]);
2827 /* Either no module or it's already stored */
2828 if (!mod || has_module(&arr, mod)) {
2830 break; /* scoped_guard */
2832 if (!try_module_get(mod))
2839 err = add_module(&arr, mod);
2846 /* We return either err < 0 in case of error, ... */
2848 kprobe_multi_put_modules(arr.mods, arr.mods_cnt);
2853 /* or number of modules found if everything is ok. */
2855 return arr.mods_cnt;
2858 static int addrs_check_error_injection_list(unsigned long *addrs, u32 cnt)
2862 for (i = 0; i < cnt; i++) {
2863 if (!within_error_injection_list(addrs[i]))
2869 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
2871 struct bpf_kprobe_multi_link *link = NULL;
2872 struct bpf_link_primer link_primer;
2873 void __user *ucookies;
2874 unsigned long *addrs;
2875 u32 flags, cnt, size;
2876 void __user *uaddrs;
2877 u64 *cookies = NULL;
2881 /* no support for 32bit archs yet */
2882 if (sizeof(u64) != sizeof(void *))
2885 if (attr->link_create.flags)
2888 if (!is_kprobe_multi(prog))
2891 flags = attr->link_create.kprobe_multi.flags;
2892 if (flags & ~BPF_F_KPROBE_MULTI_RETURN)
2895 uaddrs = u64_to_user_ptr(attr->link_create.kprobe_multi.addrs);
2896 usyms = u64_to_user_ptr(attr->link_create.kprobe_multi.syms);
2897 if (!!uaddrs == !!usyms)
2900 cnt = attr->link_create.kprobe_multi.cnt;
2903 if (cnt > MAX_KPROBE_MULTI_CNT)
2906 size = cnt * sizeof(*addrs);
2907 addrs = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2911 ucookies = u64_to_user_ptr(attr->link_create.kprobe_multi.cookies);
2913 cookies = kvmalloc_array(cnt, sizeof(*addrs), GFP_KERNEL);
2918 if (copy_from_user(cookies, ucookies, size)) {
2925 if (copy_from_user(addrs, uaddrs, size)) {
2930 struct multi_symbols_sort data = {
2933 struct user_syms us;
2935 err = copy_user_syms(&us, usyms, cnt);
2940 data.funcs = us.syms;
2942 sort_r(us.syms, cnt, sizeof(*us.syms), symbols_cmp_r,
2943 symbols_swap_r, &data);
2945 err = ftrace_lookup_symbols(us.syms, cnt, addrs);
2946 free_user_syms(&us);
2951 if (prog->kprobe_override && addrs_check_error_injection_list(addrs, cnt)) {
2956 link = kzalloc(sizeof(*link), GFP_KERNEL);
2962 bpf_link_init(&link->link, BPF_LINK_TYPE_KPROBE_MULTI,
2963 &bpf_kprobe_multi_link_lops, prog);
2965 err = bpf_link_prime(&link->link, &link_primer);
2969 if (!(flags & BPF_F_KPROBE_MULTI_RETURN))
2970 link->fp.entry_handler = kprobe_multi_link_handler;
2971 if ((flags & BPF_F_KPROBE_MULTI_RETURN) || is_kprobe_session(prog))
2972 link->fp.exit_handler = kprobe_multi_link_exit_handler;
2973 if (is_kprobe_session(prog))
2974 link->fp.entry_data_size = sizeof(u64);
2976 link->addrs = addrs;
2977 link->cookies = cookies;
2979 link->flags = flags;
2983 * Sorting addresses will trigger sorting cookies as well
2984 * (check bpf_kprobe_multi_cookie_swap). This way we can
2985 * find cookie based on the address in bpf_get_attach_cookie
2988 sort_r(addrs, cnt, sizeof(*addrs),
2989 bpf_kprobe_multi_cookie_cmp,
2990 bpf_kprobe_multi_cookie_swap,
2994 err = get_modules_for_addrs(&link->mods, addrs, cnt);
2996 bpf_link_cleanup(&link_primer);
2999 link->mods_cnt = err;
3001 err = register_fprobe_ips(&link->fp, addrs, cnt);
3003 kprobe_multi_put_modules(link->mods, link->mods_cnt);
3004 bpf_link_cleanup(&link_primer);
3008 return bpf_link_settle(&link_primer);
3016 #else /* !CONFIG_FPROBE */
3017 int bpf_kprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3021 static u64 bpf_kprobe_multi_cookie(struct bpf_run_ctx *ctx)
3025 static u64 bpf_kprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3031 #ifdef CONFIG_UPROBES
3032 struct bpf_uprobe_multi_link;
3035 struct bpf_uprobe_multi_link *link;
3037 unsigned long ref_ctr_offset;
3039 struct uprobe *uprobe;
3040 struct uprobe_consumer consumer;
3044 struct bpf_uprobe_multi_link {
3046 struct bpf_link link;
3049 struct bpf_uprobe *uprobes;
3050 struct task_struct *task;
3053 struct bpf_uprobe_multi_run_ctx {
3054 struct bpf_session_run_ctx session_ctx;
3055 unsigned long entry_ip;
3056 struct bpf_uprobe *uprobe;
3059 static void bpf_uprobe_unregister(struct bpf_uprobe *uprobes, u32 cnt)
3063 for (i = 0; i < cnt; i++)
3064 uprobe_unregister_nosync(uprobes[i].uprobe, &uprobes[i].consumer);
3067 uprobe_unregister_sync();
3070 static void bpf_uprobe_multi_link_release(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 bpf_uprobe_unregister(umulti_link->uprobes, umulti_link->cnt);
3076 if (umulti_link->task)
3077 put_task_struct(umulti_link->task);
3078 path_put(&umulti_link->path);
3081 static void bpf_uprobe_multi_link_dealloc(struct bpf_link *link)
3083 struct bpf_uprobe_multi_link *umulti_link;
3085 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3086 kvfree(umulti_link->uprobes);
3090 static int bpf_uprobe_multi_link_fill_link_info(const struct bpf_link *link,
3091 struct bpf_link_info *info)
3093 u64 __user *uref_ctr_offsets = u64_to_user_ptr(info->uprobe_multi.ref_ctr_offsets);
3094 u64 __user *ucookies = u64_to_user_ptr(info->uprobe_multi.cookies);
3095 u64 __user *uoffsets = u64_to_user_ptr(info->uprobe_multi.offsets);
3096 u64 __user *upath = u64_to_user_ptr(info->uprobe_multi.path);
3097 u32 upath_size = info->uprobe_multi.path_size;
3098 struct bpf_uprobe_multi_link *umulti_link;
3099 u32 ucount = info->uprobe_multi.count;
3104 if (!upath ^ !upath_size)
3107 if ((uoffsets || uref_ctr_offsets || ucookies) && !ucount)
3110 umulti_link = container_of(link, struct bpf_uprobe_multi_link, link);
3111 info->uprobe_multi.count = umulti_link->cnt;
3112 info->uprobe_multi.flags = umulti_link->flags;
3113 info->uprobe_multi.pid = umulti_link->task ?
3114 task_pid_nr_ns(umulti_link->task, task_active_pid_ns(current)) : 0;
3116 upath_size = upath_size ? min_t(u32, upath_size, PATH_MAX) : PATH_MAX;
3117 buf = kmalloc(upath_size, GFP_KERNEL);
3120 p = d_path(&umulti_link->path, buf, upath_size);
3125 upath_size = buf + upath_size - p;
3128 left = copy_to_user(upath, p, upath_size);
3132 info->uprobe_multi.path_size = upath_size;
3134 if (!uoffsets && !ucookies && !uref_ctr_offsets)
3137 if (ucount < umulti_link->cnt)
3140 ucount = umulti_link->cnt;
3142 for (i = 0; i < ucount; i++) {
3144 put_user(umulti_link->uprobes[i].offset, uoffsets + i))
3146 if (uref_ctr_offsets &&
3147 put_user(umulti_link->uprobes[i].ref_ctr_offset, uref_ctr_offsets + i))
3150 put_user(umulti_link->uprobes[i].cookie, ucookies + i))
3157 static const struct bpf_link_ops bpf_uprobe_multi_link_lops = {
3158 .release = bpf_uprobe_multi_link_release,
3159 .dealloc_deferred = bpf_uprobe_multi_link_dealloc,
3160 .fill_link_info = bpf_uprobe_multi_link_fill_link_info,
3163 static int uprobe_prog_run(struct bpf_uprobe *uprobe,
3164 unsigned long entry_ip,
3165 struct pt_regs *regs,
3166 bool is_return, void *data)
3168 struct bpf_uprobe_multi_link *link = uprobe->link;
3169 struct bpf_uprobe_multi_run_ctx run_ctx = {
3171 .is_return = is_return,
3174 .entry_ip = entry_ip,
3177 struct bpf_prog *prog = link->link.prog;
3178 bool sleepable = prog->sleepable;
3179 struct bpf_run_ctx *old_run_ctx;
3182 if (link->task && !same_thread_group(current, link->task))
3186 rcu_read_lock_trace();
3192 old_run_ctx = bpf_set_run_ctx(&run_ctx.session_ctx.run_ctx);
3193 err = bpf_prog_run(link->link.prog, regs);
3194 bpf_reset_run_ctx(old_run_ctx);
3199 rcu_read_unlock_trace();
3206 uprobe_multi_link_filter(struct uprobe_consumer *con, struct mm_struct *mm)
3208 struct bpf_uprobe *uprobe;
3210 uprobe = container_of(con, struct bpf_uprobe, consumer);
3211 return uprobe->link->task->mm == mm;
3215 uprobe_multi_link_handler(struct uprobe_consumer *con, struct pt_regs *regs,
3218 struct bpf_uprobe *uprobe;
3221 uprobe = container_of(con, struct bpf_uprobe, consumer);
3222 ret = uprobe_prog_run(uprobe, instruction_pointer(regs), regs, false, data);
3223 if (uprobe->session)
3224 return ret ? UPROBE_HANDLER_IGNORE : 0;
3229 uprobe_multi_link_ret_handler(struct uprobe_consumer *con, unsigned long func, struct pt_regs *regs,
3232 struct bpf_uprobe *uprobe;
3234 uprobe = container_of(con, struct bpf_uprobe, consumer);
3235 uprobe_prog_run(uprobe, func, regs, true, data);
3239 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3241 struct bpf_uprobe_multi_run_ctx *run_ctx;
3243 run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx,
3244 session_ctx.run_ctx);
3245 return run_ctx->entry_ip;
3248 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3250 struct bpf_uprobe_multi_run_ctx *run_ctx;
3252 run_ctx = container_of(current->bpf_ctx, struct bpf_uprobe_multi_run_ctx,
3253 session_ctx.run_ctx);
3254 return run_ctx->uprobe->cookie;
3257 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3259 struct bpf_uprobe_multi_link *link = NULL;
3260 unsigned long __user *uref_ctr_offsets;
3261 struct bpf_link_primer link_primer;
3262 struct bpf_uprobe *uprobes = NULL;
3263 struct task_struct *task = NULL;
3264 unsigned long __user *uoffsets;
3265 u64 __user *ucookies;
3273 /* no support for 32bit archs yet */
3274 if (sizeof(u64) != sizeof(void *))
3277 if (attr->link_create.flags)
3280 if (!is_uprobe_multi(prog))
3283 flags = attr->link_create.uprobe_multi.flags;
3284 if (flags & ~BPF_F_UPROBE_MULTI_RETURN)
3288 * path, offsets and cnt are mandatory,
3289 * ref_ctr_offsets and cookies are optional
3291 upath = u64_to_user_ptr(attr->link_create.uprobe_multi.path);
3292 uoffsets = u64_to_user_ptr(attr->link_create.uprobe_multi.offsets);
3293 cnt = attr->link_create.uprobe_multi.cnt;
3294 pid = attr->link_create.uprobe_multi.pid;
3296 if (!upath || !uoffsets || !cnt || pid < 0)
3298 if (cnt > MAX_UPROBE_MULTI_CNT)
3301 uref_ctr_offsets = u64_to_user_ptr(attr->link_create.uprobe_multi.ref_ctr_offsets);
3302 ucookies = u64_to_user_ptr(attr->link_create.uprobe_multi.cookies);
3304 name = strndup_user(upath, PATH_MAX);
3306 err = PTR_ERR(name);
3310 err = kern_path(name, LOOKUP_FOLLOW, &path);
3315 if (!d_is_reg(path.dentry)) {
3317 goto error_path_put;
3322 task = get_pid_task(find_vpid(pid), PIDTYPE_TGID);
3326 goto error_path_put;
3332 link = kzalloc(sizeof(*link), GFP_KERNEL);
3333 uprobes = kvcalloc(cnt, sizeof(*uprobes), GFP_KERNEL);
3335 if (!uprobes || !link)
3338 for (i = 0; i < cnt; i++) {
3339 if (__get_user(uprobes[i].offset, uoffsets + i)) {
3343 if (uprobes[i].offset < 0) {
3347 if (uref_ctr_offsets && __get_user(uprobes[i].ref_ctr_offset, uref_ctr_offsets + i)) {
3351 if (ucookies && __get_user(uprobes[i].cookie, ucookies + i)) {
3356 uprobes[i].link = link;
3358 if (!(flags & BPF_F_UPROBE_MULTI_RETURN))
3359 uprobes[i].consumer.handler = uprobe_multi_link_handler;
3360 if (flags & BPF_F_UPROBE_MULTI_RETURN || is_uprobe_session(prog))
3361 uprobes[i].consumer.ret_handler = uprobe_multi_link_ret_handler;
3362 if (is_uprobe_session(prog))
3363 uprobes[i].session = true;
3365 uprobes[i].consumer.filter = uprobe_multi_link_filter;
3369 link->uprobes = uprobes;
3372 link->flags = flags;
3374 bpf_link_init(&link->link, BPF_LINK_TYPE_UPROBE_MULTI,
3375 &bpf_uprobe_multi_link_lops, prog);
3377 for (i = 0; i < cnt; i++) {
3378 uprobes[i].uprobe = uprobe_register(d_real_inode(link->path.dentry),
3380 uprobes[i].ref_ctr_offset,
3381 &uprobes[i].consumer);
3382 if (IS_ERR(uprobes[i].uprobe)) {
3383 err = PTR_ERR(uprobes[i].uprobe);
3385 goto error_unregister;
3389 err = bpf_link_prime(&link->link, &link_primer);
3391 goto error_unregister;
3393 return bpf_link_settle(&link_primer);
3396 bpf_uprobe_unregister(uprobes, link->cnt);
3402 put_task_struct(task);
3407 #else /* !CONFIG_UPROBES */
3408 int bpf_uprobe_multi_link_attach(const union bpf_attr *attr, struct bpf_prog *prog)
3412 static u64 bpf_uprobe_multi_cookie(struct bpf_run_ctx *ctx)
3416 static u64 bpf_uprobe_multi_entry_ip(struct bpf_run_ctx *ctx)
3420 #endif /* CONFIG_UPROBES */
3422 __bpf_kfunc_start_defs();
3424 __bpf_kfunc bool bpf_session_is_return(void)
3426 struct bpf_session_run_ctx *session_ctx;
3428 session_ctx = container_of(current->bpf_ctx, struct bpf_session_run_ctx, run_ctx);
3429 return session_ctx->is_return;
3432 __bpf_kfunc __u64 *bpf_session_cookie(void)
3434 struct bpf_session_run_ctx *session_ctx;
3436 session_ctx = container_of(current->bpf_ctx, struct bpf_session_run_ctx, run_ctx);
3437 return session_ctx->data;
3440 __bpf_kfunc_end_defs();
3442 BTF_KFUNCS_START(kprobe_multi_kfunc_set_ids)
3443 BTF_ID_FLAGS(func, bpf_session_is_return)
3444 BTF_ID_FLAGS(func, bpf_session_cookie)
3445 BTF_KFUNCS_END(kprobe_multi_kfunc_set_ids)
3447 static int bpf_kprobe_multi_filter(const struct bpf_prog *prog, u32 kfunc_id)
3449 if (!btf_id_set8_contains(&kprobe_multi_kfunc_set_ids, kfunc_id))
3452 if (!is_kprobe_session(prog) && !is_uprobe_session(prog))
3458 static const struct btf_kfunc_id_set bpf_kprobe_multi_kfunc_set = {
3459 .owner = THIS_MODULE,
3460 .set = &kprobe_multi_kfunc_set_ids,
3461 .filter = bpf_kprobe_multi_filter,
3464 static int __init bpf_kprobe_multi_kfuncs_init(void)
3466 return register_btf_kfunc_id_set(BPF_PROG_TYPE_KPROBE, &bpf_kprobe_multi_kfunc_set);
3469 late_initcall(bpf_kprobe_multi_kfuncs_init);
3471 typedef int (*copy_fn_t)(void *dst, const void *src, u32 size, struct task_struct *tsk);
3474 * The __always_inline is to make sure the compiler doesn't
3475 * generate indirect calls into callbacks, which is expensive,
3476 * on some kernel configurations. This allows compiler to put
3477 * direct calls into all the specific callback implementations
3478 * (copy_user_data_sleepable, copy_user_data_nofault, and so on)
3480 static __always_inline int __bpf_dynptr_copy_str(struct bpf_dynptr *dptr, u32 doff, u32 size,
3481 const void *unsafe_src,
3482 copy_fn_t str_copy_fn,
3483 struct task_struct *tsk)
3485 struct bpf_dynptr_kern *dst;
3491 dst_slice = bpf_dynptr_slice_rdwr(dptr, doff, NULL, size);
3492 if (likely(dst_slice))
3493 return str_copy_fn(dst_slice, unsafe_src, size, tsk);
3495 dst = (struct bpf_dynptr_kern *)dptr;
3496 if (bpf_dynptr_check_off_len(dst, doff, size))
3499 for (off = 0; off < size; off += chunk_sz - 1) {
3500 chunk_sz = min_t(u32, sizeof(buf), size - off);
3501 /* Expect str_copy_fn to return count of copied bytes, including
3502 * zero terminator. Next iteration increment off by chunk_sz - 1 to
3505 cnt = str_copy_fn(buf, unsafe_src + off, chunk_sz, tsk);
3508 err = __bpf_dynptr_write(dst, doff + off, buf, cnt, 0);
3511 if (cnt < chunk_sz || chunk_sz == 1) /* we are done */
3517 static __always_inline int __bpf_dynptr_copy(const struct bpf_dynptr *dptr, u32 doff,
3518 u32 size, const void *unsafe_src,
3519 copy_fn_t copy_fn, struct task_struct *tsk)
3521 struct bpf_dynptr_kern *dst;
3527 dst_slice = bpf_dynptr_slice_rdwr(dptr, doff, NULL, size);
3528 if (likely(dst_slice))
3529 return copy_fn(dst_slice, unsafe_src, size, tsk);
3531 dst = (struct bpf_dynptr_kern *)dptr;
3532 if (bpf_dynptr_check_off_len(dst, doff, size))
3535 for (off = 0; off < size; off += chunk_sz) {
3536 chunk_sz = min_t(u32, sizeof(buf), size - off);
3537 err = copy_fn(buf, unsafe_src + off, chunk_sz, tsk);
3540 err = __bpf_dynptr_write(dst, doff + off, buf, chunk_sz, 0);
3547 static __always_inline int copy_user_data_nofault(void *dst, const void *unsafe_src,
3548 u32 size, struct task_struct *tsk)
3550 return copy_from_user_nofault(dst, (const void __user *)unsafe_src, size);
3553 static __always_inline int copy_user_data_sleepable(void *dst, const void *unsafe_src,
3554 u32 size, struct task_struct *tsk)
3558 if (!tsk) { /* Read from the current task */
3559 ret = copy_from_user(dst, (const void __user *)unsafe_src, size);
3565 ret = access_process_vm(tsk, (unsigned long)unsafe_src, dst, size, 0);
3571 static __always_inline int copy_kernel_data_nofault(void *dst, const void *unsafe_src,
3572 u32 size, struct task_struct *tsk)
3574 return copy_from_kernel_nofault(dst, unsafe_src, size);
3577 static __always_inline int copy_user_str_nofault(void *dst, const void *unsafe_src,
3578 u32 size, struct task_struct *tsk)
3580 return strncpy_from_user_nofault(dst, (const void __user *)unsafe_src, size);
3583 static __always_inline int copy_user_str_sleepable(void *dst, const void *unsafe_src,
3584 u32 size, struct task_struct *tsk)
3588 if (unlikely(size == 0))
3592 ret = copy_remote_vm_str(tsk, (unsigned long)unsafe_src, dst, size, 0);
3594 ret = strncpy_from_user(dst, (const void __user *)unsafe_src, size - 1);
3595 /* strncpy_from_user does not guarantee NUL termination */
3597 ((char *)dst)[ret] = '\0';
3605 static __always_inline int copy_kernel_str_nofault(void *dst, const void *unsafe_src,
3606 u32 size, struct task_struct *tsk)
3608 return strncpy_from_kernel_nofault(dst, unsafe_src, size);
3611 __bpf_kfunc_start_defs();
3613 __bpf_kfunc int bpf_send_signal_task(struct task_struct *task, int sig, enum pid_type type,
3616 if (type != PIDTYPE_PID && type != PIDTYPE_TGID)
3619 return bpf_send_signal_common(sig, type, task, value);
3622 __bpf_kfunc int bpf_probe_read_user_dynptr(struct bpf_dynptr *dptr, u32 off,
3623 u32 size, const void __user *unsafe_ptr__ign)
3625 return __bpf_dynptr_copy(dptr, off, size, (const void *)unsafe_ptr__ign,
3626 copy_user_data_nofault, NULL);
3629 __bpf_kfunc int bpf_probe_read_kernel_dynptr(struct bpf_dynptr *dptr, u32 off,
3630 u32 size, const void *unsafe_ptr__ign)
3632 return __bpf_dynptr_copy(dptr, off, size, unsafe_ptr__ign,
3633 copy_kernel_data_nofault, NULL);
3636 __bpf_kfunc int bpf_probe_read_user_str_dynptr(struct bpf_dynptr *dptr, u32 off,
3637 u32 size, const void __user *unsafe_ptr__ign)
3639 return __bpf_dynptr_copy_str(dptr, off, size, (const void *)unsafe_ptr__ign,
3640 copy_user_str_nofault, NULL);
3643 __bpf_kfunc int bpf_probe_read_kernel_str_dynptr(struct bpf_dynptr *dptr, u32 off,
3644 u32 size, const void *unsafe_ptr__ign)
3646 return __bpf_dynptr_copy_str(dptr, off, size, unsafe_ptr__ign,
3647 copy_kernel_str_nofault, NULL);
3650 __bpf_kfunc int bpf_copy_from_user_dynptr(struct bpf_dynptr *dptr, u32 off,
3651 u32 size, const void __user *unsafe_ptr__ign)
3653 return __bpf_dynptr_copy(dptr, off, size, (const void *)unsafe_ptr__ign,
3654 copy_user_data_sleepable, NULL);
3657 __bpf_kfunc int bpf_copy_from_user_str_dynptr(struct bpf_dynptr *dptr, u32 off,
3658 u32 size, const void __user *unsafe_ptr__ign)
3660 return __bpf_dynptr_copy_str(dptr, off, size, (const void *)unsafe_ptr__ign,
3661 copy_user_str_sleepable, NULL);
3664 __bpf_kfunc int bpf_copy_from_user_task_dynptr(struct bpf_dynptr *dptr, u32 off,
3665 u32 size, const void __user *unsafe_ptr__ign,
3666 struct task_struct *tsk)
3668 return __bpf_dynptr_copy(dptr, off, size, (const void *)unsafe_ptr__ign,
3669 copy_user_data_sleepable, tsk);
3672 __bpf_kfunc int bpf_copy_from_user_task_str_dynptr(struct bpf_dynptr *dptr, u32 off,
3673 u32 size, const void __user *unsafe_ptr__ign,
3674 struct task_struct *tsk)
3676 return __bpf_dynptr_copy_str(dptr, off, size, (const void *)unsafe_ptr__ign,
3677 copy_user_str_sleepable, tsk);
3680 __bpf_kfunc_end_defs();