Loading...
1/* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
2 *
3 * This program is free software; you can redistribute it and/or
4 * modify it under the terms of version 2 of the GNU General Public
5 * License as published by the Free Software Foundation.
6 */
7#include <linux/kernel.h>
8#include <linux/types.h>
9#include <linux/slab.h>
10#include <linux/bpf.h>
11#include <linux/filter.h>
12#include <linux/uaccess.h>
13#include <linux/ctype.h>
14#include "trace.h"
15
16/**
17 * trace_call_bpf - invoke BPF program
18 * @prog: BPF program
19 * @ctx: opaque context pointer
20 *
21 * kprobe handlers execute BPF programs via this helper.
22 * Can be used from static tracepoints in the future.
23 *
24 * Return: BPF programs always return an integer which is interpreted by
25 * kprobe handler as:
26 * 0 - return from kprobe (event is filtered out)
27 * 1 - store kprobe event into ring buffer
28 * Other values are reserved and currently alias to 1
29 */
30unsigned int trace_call_bpf(struct bpf_prog *prog, void *ctx)
31{
32 unsigned int ret;
33
34 if (in_nmi()) /* not supported yet */
35 return 1;
36
37 preempt_disable();
38
39 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
40 /*
41 * since some bpf program is already running on this cpu,
42 * don't call into another bpf program (same or different)
43 * and don't send kprobe event into ring-buffer,
44 * so return zero here
45 */
46 ret = 0;
47 goto out;
48 }
49
50 rcu_read_lock();
51 ret = BPF_PROG_RUN(prog, ctx);
52 rcu_read_unlock();
53
54 out:
55 __this_cpu_dec(bpf_prog_active);
56 preempt_enable();
57
58 return ret;
59}
60EXPORT_SYMBOL_GPL(trace_call_bpf);
61
62static u64 bpf_probe_read(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5)
63{
64 void *dst = (void *) (long) r1;
65 int size = (int) r2;
66 void *unsafe_ptr = (void *) (long) r3;
67
68 return probe_kernel_read(dst, unsafe_ptr, size);
69}
70
71static const struct bpf_func_proto bpf_probe_read_proto = {
72 .func = bpf_probe_read,
73 .gpl_only = true,
74 .ret_type = RET_INTEGER,
75 .arg1_type = ARG_PTR_TO_STACK,
76 .arg2_type = ARG_CONST_STACK_SIZE,
77 .arg3_type = ARG_ANYTHING,
78};
79
80/*
81 * limited trace_printk()
82 * only %d %u %x %ld %lu %lx %lld %llu %llx %p %s conversion specifiers allowed
83 */
84static u64 bpf_trace_printk(u64 r1, u64 fmt_size, u64 r3, u64 r4, u64 r5)
85{
86 char *fmt = (char *) (long) r1;
87 bool str_seen = false;
88 int mod[3] = {};
89 int fmt_cnt = 0;
90 u64 unsafe_addr;
91 char buf[64];
92 int i;
93
94 /*
95 * bpf_check()->check_func_arg()->check_stack_boundary()
96 * guarantees that fmt points to bpf program stack,
97 * fmt_size bytes of it were initialized and fmt_size > 0
98 */
99 if (fmt[--fmt_size] != 0)
100 return -EINVAL;
101
102 /* check format string for allowed specifiers */
103 for (i = 0; i < fmt_size; i++) {
104 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
105 return -EINVAL;
106
107 if (fmt[i] != '%')
108 continue;
109
110 if (fmt_cnt >= 3)
111 return -EINVAL;
112
113 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
114 i++;
115 if (fmt[i] == 'l') {
116 mod[fmt_cnt]++;
117 i++;
118 } else if (fmt[i] == 'p' || fmt[i] == 's') {
119 mod[fmt_cnt]++;
120 i++;
121 if (!isspace(fmt[i]) && !ispunct(fmt[i]) && fmt[i] != 0)
122 return -EINVAL;
123 fmt_cnt++;
124 if (fmt[i - 1] == 's') {
125 if (str_seen)
126 /* allow only one '%s' per fmt string */
127 return -EINVAL;
128 str_seen = true;
129
130 switch (fmt_cnt) {
131 case 1:
132 unsafe_addr = r3;
133 r3 = (long) buf;
134 break;
135 case 2:
136 unsafe_addr = r4;
137 r4 = (long) buf;
138 break;
139 case 3:
140 unsafe_addr = r5;
141 r5 = (long) buf;
142 break;
143 }
144 buf[0] = 0;
145 strncpy_from_unsafe(buf,
146 (void *) (long) unsafe_addr,
147 sizeof(buf));
148 }
149 continue;
150 }
151
152 if (fmt[i] == 'l') {
153 mod[fmt_cnt]++;
154 i++;
155 }
156
157 if (fmt[i] != 'd' && fmt[i] != 'u' && fmt[i] != 'x')
158 return -EINVAL;
159 fmt_cnt++;
160 }
161
162 return __trace_printk(1/* fake ip will not be printed */, fmt,
163 mod[0] == 2 ? r3 : mod[0] == 1 ? (long) r3 : (u32) r3,
164 mod[1] == 2 ? r4 : mod[1] == 1 ? (long) r4 : (u32) r4,
165 mod[2] == 2 ? r5 : mod[2] == 1 ? (long) r5 : (u32) r5);
166}
167
168static const struct bpf_func_proto bpf_trace_printk_proto = {
169 .func = bpf_trace_printk,
170 .gpl_only = true,
171 .ret_type = RET_INTEGER,
172 .arg1_type = ARG_PTR_TO_STACK,
173 .arg2_type = ARG_CONST_STACK_SIZE,
174};
175
176const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
177{
178 /*
179 * this program might be calling bpf_trace_printk,
180 * so allocate per-cpu printk buffers
181 */
182 trace_printk_init_buffers();
183
184 return &bpf_trace_printk_proto;
185}
186
187static u64 bpf_perf_event_read(u64 r1, u64 index, u64 r3, u64 r4, u64 r5)
188{
189 struct bpf_map *map = (struct bpf_map *) (unsigned long) r1;
190 struct bpf_array *array = container_of(map, struct bpf_array, map);
191 struct perf_event *event;
192 struct file *file;
193
194 if (unlikely(index >= array->map.max_entries))
195 return -E2BIG;
196
197 file = (struct file *)array->ptrs[index];
198 if (unlikely(!file))
199 return -ENOENT;
200
201 event = file->private_data;
202
203 /* make sure event is local and doesn't have pmu::count */
204 if (event->oncpu != smp_processor_id() ||
205 event->pmu->count)
206 return -EINVAL;
207
208 /*
209 * we don't know if the function is run successfully by the
210 * return value. It can be judged in other places, such as
211 * eBPF programs.
212 */
213 return perf_event_read_local(event);
214}
215
216static const struct bpf_func_proto bpf_perf_event_read_proto = {
217 .func = bpf_perf_event_read,
218 .gpl_only = true,
219 .ret_type = RET_INTEGER,
220 .arg1_type = ARG_CONST_MAP_PTR,
221 .arg2_type = ARG_ANYTHING,
222};
223
224static u64 bpf_perf_event_output(u64 r1, u64 r2, u64 index, u64 r4, u64 size)
225{
226 struct pt_regs *regs = (struct pt_regs *) (long) r1;
227 struct bpf_map *map = (struct bpf_map *) (long) r2;
228 struct bpf_array *array = container_of(map, struct bpf_array, map);
229 void *data = (void *) (long) r4;
230 struct perf_sample_data sample_data;
231 struct perf_event *event;
232 struct file *file;
233 struct perf_raw_record raw = {
234 .size = size,
235 .data = data,
236 };
237
238 if (unlikely(index >= array->map.max_entries))
239 return -E2BIG;
240
241 file = (struct file *)array->ptrs[index];
242 if (unlikely(!file))
243 return -ENOENT;
244
245 event = file->private_data;
246
247 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
248 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
249 return -EINVAL;
250
251 if (unlikely(event->oncpu != smp_processor_id()))
252 return -EOPNOTSUPP;
253
254 perf_sample_data_init(&sample_data, 0, 0);
255 sample_data.raw = &raw;
256 perf_event_output(event, &sample_data, regs);
257 return 0;
258}
259
260static const struct bpf_func_proto bpf_perf_event_output_proto = {
261 .func = bpf_perf_event_output,
262 .gpl_only = true,
263 .ret_type = RET_INTEGER,
264 .arg1_type = ARG_PTR_TO_CTX,
265 .arg2_type = ARG_CONST_MAP_PTR,
266 .arg3_type = ARG_ANYTHING,
267 .arg4_type = ARG_PTR_TO_STACK,
268 .arg5_type = ARG_CONST_STACK_SIZE,
269};
270
271static const struct bpf_func_proto *kprobe_prog_func_proto(enum bpf_func_id func_id)
272{
273 switch (func_id) {
274 case BPF_FUNC_map_lookup_elem:
275 return &bpf_map_lookup_elem_proto;
276 case BPF_FUNC_map_update_elem:
277 return &bpf_map_update_elem_proto;
278 case BPF_FUNC_map_delete_elem:
279 return &bpf_map_delete_elem_proto;
280 case BPF_FUNC_probe_read:
281 return &bpf_probe_read_proto;
282 case BPF_FUNC_ktime_get_ns:
283 return &bpf_ktime_get_ns_proto;
284 case BPF_FUNC_tail_call:
285 return &bpf_tail_call_proto;
286 case BPF_FUNC_get_current_pid_tgid:
287 return &bpf_get_current_pid_tgid_proto;
288 case BPF_FUNC_get_current_uid_gid:
289 return &bpf_get_current_uid_gid_proto;
290 case BPF_FUNC_get_current_comm:
291 return &bpf_get_current_comm_proto;
292 case BPF_FUNC_trace_printk:
293 return bpf_get_trace_printk_proto();
294 case BPF_FUNC_get_smp_processor_id:
295 return &bpf_get_smp_processor_id_proto;
296 case BPF_FUNC_perf_event_read:
297 return &bpf_perf_event_read_proto;
298 case BPF_FUNC_perf_event_output:
299 return &bpf_perf_event_output_proto;
300 case BPF_FUNC_get_stackid:
301 return &bpf_get_stackid_proto;
302 default:
303 return NULL;
304 }
305}
306
307/* bpf+kprobe programs can access fields of 'struct pt_regs' */
308static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type)
309{
310 /* check bounds */
311 if (off < 0 || off >= sizeof(struct pt_regs))
312 return false;
313
314 /* only read is allowed */
315 if (type != BPF_READ)
316 return false;
317
318 /* disallow misaligned access */
319 if (off % size != 0)
320 return false;
321
322 return true;
323}
324
325static const struct bpf_verifier_ops kprobe_prog_ops = {
326 .get_func_proto = kprobe_prog_func_proto,
327 .is_valid_access = kprobe_prog_is_valid_access,
328};
329
330static struct bpf_prog_type_list kprobe_tl = {
331 .ops = &kprobe_prog_ops,
332 .type = BPF_PROG_TYPE_KPROBE,
333};
334
335static int __init register_kprobe_prog_ops(void)
336{
337 bpf_register_prog_type(&kprobe_tl);
338 return 0;
339}
340late_initcall(register_kprobe_prog_ops);
1// SPDX-License-Identifier: GPL-2.0
2/* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
3 * Copyright (c) 2016 Facebook
4 */
5#include <linux/kernel.h>
6#include <linux/types.h>
7#include <linux/slab.h>
8#include <linux/bpf.h>
9#include <linux/bpf_perf_event.h>
10#include <linux/filter.h>
11#include <linux/uaccess.h>
12#include <linux/ctype.h>
13#include <linux/kprobes.h>
14#include <linux/syscalls.h>
15#include <linux/error-injection.h>
16
17#include <asm/tlb.h>
18
19#include "trace_probe.h"
20#include "trace.h"
21
22#define bpf_event_rcu_dereference(p) \
23 rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
24
25#ifdef CONFIG_MODULES
26struct bpf_trace_module {
27 struct module *module;
28 struct list_head list;
29};
30
31static LIST_HEAD(bpf_trace_modules);
32static DEFINE_MUTEX(bpf_module_mutex);
33
34static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
35{
36 struct bpf_raw_event_map *btp, *ret = NULL;
37 struct bpf_trace_module *btm;
38 unsigned int i;
39
40 mutex_lock(&bpf_module_mutex);
41 list_for_each_entry(btm, &bpf_trace_modules, list) {
42 for (i = 0; i < btm->module->num_bpf_raw_events; ++i) {
43 btp = &btm->module->bpf_raw_events[i];
44 if (!strcmp(btp->tp->name, name)) {
45 if (try_module_get(btm->module))
46 ret = btp;
47 goto out;
48 }
49 }
50 }
51out:
52 mutex_unlock(&bpf_module_mutex);
53 return ret;
54}
55#else
56static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
57{
58 return NULL;
59}
60#endif /* CONFIG_MODULES */
61
62u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
63u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
64
65/**
66 * trace_call_bpf - invoke BPF program
67 * @call: tracepoint event
68 * @ctx: opaque context pointer
69 *
70 * kprobe handlers execute BPF programs via this helper.
71 * Can be used from static tracepoints in the future.
72 *
73 * Return: BPF programs always return an integer which is interpreted by
74 * kprobe handler as:
75 * 0 - return from kprobe (event is filtered out)
76 * 1 - store kprobe event into ring buffer
77 * Other values are reserved and currently alias to 1
78 */
79unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
80{
81 unsigned int ret;
82
83 if (in_nmi()) /* not supported yet */
84 return 1;
85
86 preempt_disable();
87
88 if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) {
89 /*
90 * since some bpf program is already running on this cpu,
91 * don't call into another bpf program (same or different)
92 * and don't send kprobe event into ring-buffer,
93 * so return zero here
94 */
95 ret = 0;
96 goto out;
97 }
98
99 /*
100 * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock
101 * to all call sites, we did a bpf_prog_array_valid() there to check
102 * whether call->prog_array is empty or not, which is
103 * a heurisitc to speed up execution.
104 *
105 * If bpf_prog_array_valid() fetched prog_array was
106 * non-NULL, we go into trace_call_bpf() and do the actual
107 * proper rcu_dereference() under RCU lock.
108 * If it turns out that prog_array is NULL then, we bail out.
109 * For the opposite, if the bpf_prog_array_valid() fetched pointer
110 * was NULL, you'll skip the prog_array with the risk of missing
111 * out of events when it was updated in between this and the
112 * rcu_dereference() which is accepted risk.
113 */
114 ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
115
116 out:
117 __this_cpu_dec(bpf_prog_active);
118 preempt_enable();
119
120 return ret;
121}
122EXPORT_SYMBOL_GPL(trace_call_bpf);
123
124#ifdef CONFIG_BPF_KPROBE_OVERRIDE
125BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
126{
127 regs_set_return_value(regs, rc);
128 override_function_with_return(regs);
129 return 0;
130}
131
132static const struct bpf_func_proto bpf_override_return_proto = {
133 .func = bpf_override_return,
134 .gpl_only = true,
135 .ret_type = RET_INTEGER,
136 .arg1_type = ARG_PTR_TO_CTX,
137 .arg2_type = ARG_ANYTHING,
138};
139#endif
140
141BPF_CALL_3(bpf_probe_read, void *, dst, u32, size, const void *, unsafe_ptr)
142{
143 int ret;
144
145 ret = security_locked_down(LOCKDOWN_BPF_READ);
146 if (ret < 0)
147 goto out;
148
149 ret = probe_kernel_read(dst, unsafe_ptr, size);
150 if (unlikely(ret < 0))
151out:
152 memset(dst, 0, size);
153
154 return ret;
155}
156
157static const struct bpf_func_proto bpf_probe_read_proto = {
158 .func = bpf_probe_read,
159 .gpl_only = true,
160 .ret_type = RET_INTEGER,
161 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
162 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
163 .arg3_type = ARG_ANYTHING,
164};
165
166BPF_CALL_3(bpf_probe_write_user, void *, unsafe_ptr, const void *, src,
167 u32, size)
168{
169 /*
170 * Ensure we're in user context which is safe for the helper to
171 * run. This helper has no business in a kthread.
172 *
173 * access_ok() should prevent writing to non-user memory, but in
174 * some situations (nommu, temporary switch, etc) access_ok() does
175 * not provide enough validation, hence the check on KERNEL_DS.
176 *
177 * nmi_uaccess_okay() ensures the probe is not run in an interim
178 * state, when the task or mm are switched. This is specifically
179 * required to prevent the use of temporary mm.
180 */
181
182 if (unlikely(in_interrupt() ||
183 current->flags & (PF_KTHREAD | PF_EXITING)))
184 return -EPERM;
185 if (unlikely(uaccess_kernel()))
186 return -EPERM;
187 if (unlikely(!nmi_uaccess_okay()))
188 return -EPERM;
189 if (!access_ok(unsafe_ptr, size))
190 return -EPERM;
191
192 return probe_kernel_write(unsafe_ptr, src, size);
193}
194
195static const struct bpf_func_proto bpf_probe_write_user_proto = {
196 .func = bpf_probe_write_user,
197 .gpl_only = true,
198 .ret_type = RET_INTEGER,
199 .arg1_type = ARG_ANYTHING,
200 .arg2_type = ARG_PTR_TO_MEM,
201 .arg3_type = ARG_CONST_SIZE,
202};
203
204static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
205{
206 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
207 current->comm, task_pid_nr(current));
208
209 return &bpf_probe_write_user_proto;
210}
211
212/*
213 * Only limited trace_printk() conversion specifiers allowed:
214 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
215 */
216BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
217 u64, arg2, u64, arg3)
218{
219 bool str_seen = false;
220 int mod[3] = {};
221 int fmt_cnt = 0;
222 u64 unsafe_addr;
223 char buf[64];
224 int i;
225
226 /*
227 * bpf_check()->check_func_arg()->check_stack_boundary()
228 * guarantees that fmt points to bpf program stack,
229 * fmt_size bytes of it were initialized and fmt_size > 0
230 */
231 if (fmt[--fmt_size] != 0)
232 return -EINVAL;
233
234 /* check format string for allowed specifiers */
235 for (i = 0; i < fmt_size; i++) {
236 if ((!isprint(fmt[i]) && !isspace(fmt[i])) || !isascii(fmt[i]))
237 return -EINVAL;
238
239 if (fmt[i] != '%')
240 continue;
241
242 if (fmt_cnt >= 3)
243 return -EINVAL;
244
245 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
246 i++;
247 if (fmt[i] == 'l') {
248 mod[fmt_cnt]++;
249 i++;
250 } else if (fmt[i] == 'p' || fmt[i] == 's') {
251 mod[fmt_cnt]++;
252 /* disallow any further format extensions */
253 if (fmt[i + 1] != 0 &&
254 !isspace(fmt[i + 1]) &&
255 !ispunct(fmt[i + 1]))
256 return -EINVAL;
257 fmt_cnt++;
258 if (fmt[i] == 's') {
259 if (str_seen)
260 /* allow only one '%s' per fmt string */
261 return -EINVAL;
262 str_seen = true;
263
264 switch (fmt_cnt) {
265 case 1:
266 unsafe_addr = arg1;
267 arg1 = (long) buf;
268 break;
269 case 2:
270 unsafe_addr = arg2;
271 arg2 = (long) buf;
272 break;
273 case 3:
274 unsafe_addr = arg3;
275 arg3 = (long) buf;
276 break;
277 }
278 buf[0] = 0;
279 strncpy_from_unsafe(buf,
280 (void *) (long) unsafe_addr,
281 sizeof(buf));
282 }
283 continue;
284 }
285
286 if (fmt[i] == 'l') {
287 mod[fmt_cnt]++;
288 i++;
289 }
290
291 if (fmt[i] != 'i' && fmt[i] != 'd' &&
292 fmt[i] != 'u' && fmt[i] != 'x')
293 return -EINVAL;
294 fmt_cnt++;
295 }
296
297/* Horrid workaround for getting va_list handling working with different
298 * argument type combinations generically for 32 and 64 bit archs.
299 */
300#define __BPF_TP_EMIT() __BPF_ARG3_TP()
301#define __BPF_TP(...) \
302 __trace_printk(0 /* Fake ip */, \
303 fmt, ##__VA_ARGS__)
304
305#define __BPF_ARG1_TP(...) \
306 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
307 ? __BPF_TP(arg1, ##__VA_ARGS__) \
308 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
309 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
310 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
311
312#define __BPF_ARG2_TP(...) \
313 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
314 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
315 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
316 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
317 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
318
319#define __BPF_ARG3_TP(...) \
320 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
321 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
322 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
323 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
324 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
325
326 return __BPF_TP_EMIT();
327}
328
329static const struct bpf_func_proto bpf_trace_printk_proto = {
330 .func = bpf_trace_printk,
331 .gpl_only = true,
332 .ret_type = RET_INTEGER,
333 .arg1_type = ARG_PTR_TO_MEM,
334 .arg2_type = ARG_CONST_SIZE,
335};
336
337const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
338{
339 /*
340 * this program might be calling bpf_trace_printk,
341 * so allocate per-cpu printk buffers
342 */
343 trace_printk_init_buffers();
344
345 return &bpf_trace_printk_proto;
346}
347
348static __always_inline int
349get_map_perf_counter(struct bpf_map *map, u64 flags,
350 u64 *value, u64 *enabled, u64 *running)
351{
352 struct bpf_array *array = container_of(map, struct bpf_array, map);
353 unsigned int cpu = smp_processor_id();
354 u64 index = flags & BPF_F_INDEX_MASK;
355 struct bpf_event_entry *ee;
356
357 if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
358 return -EINVAL;
359 if (index == BPF_F_CURRENT_CPU)
360 index = cpu;
361 if (unlikely(index >= array->map.max_entries))
362 return -E2BIG;
363
364 ee = READ_ONCE(array->ptrs[index]);
365 if (!ee)
366 return -ENOENT;
367
368 return perf_event_read_local(ee->event, value, enabled, running);
369}
370
371BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags)
372{
373 u64 value = 0;
374 int err;
375
376 err = get_map_perf_counter(map, flags, &value, NULL, NULL);
377 /*
378 * this api is ugly since we miss [-22..-2] range of valid
379 * counter values, but that's uapi
380 */
381 if (err)
382 return err;
383 return value;
384}
385
386static const struct bpf_func_proto bpf_perf_event_read_proto = {
387 .func = bpf_perf_event_read,
388 .gpl_only = true,
389 .ret_type = RET_INTEGER,
390 .arg1_type = ARG_CONST_MAP_PTR,
391 .arg2_type = ARG_ANYTHING,
392};
393
394BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags,
395 struct bpf_perf_event_value *, buf, u32, size)
396{
397 int err = -EINVAL;
398
399 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
400 goto clear;
401 err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled,
402 &buf->running);
403 if (unlikely(err))
404 goto clear;
405 return 0;
406clear:
407 memset(buf, 0, size);
408 return err;
409}
410
411static const struct bpf_func_proto bpf_perf_event_read_value_proto = {
412 .func = bpf_perf_event_read_value,
413 .gpl_only = true,
414 .ret_type = RET_INTEGER,
415 .arg1_type = ARG_CONST_MAP_PTR,
416 .arg2_type = ARG_ANYTHING,
417 .arg3_type = ARG_PTR_TO_UNINIT_MEM,
418 .arg4_type = ARG_CONST_SIZE,
419};
420
421static __always_inline u64
422__bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
423 u64 flags, struct perf_sample_data *sd)
424{
425 struct bpf_array *array = container_of(map, struct bpf_array, map);
426 unsigned int cpu = smp_processor_id();
427 u64 index = flags & BPF_F_INDEX_MASK;
428 struct bpf_event_entry *ee;
429 struct perf_event *event;
430
431 if (index == BPF_F_CURRENT_CPU)
432 index = cpu;
433 if (unlikely(index >= array->map.max_entries))
434 return -E2BIG;
435
436 ee = READ_ONCE(array->ptrs[index]);
437 if (!ee)
438 return -ENOENT;
439
440 event = ee->event;
441 if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
442 event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
443 return -EINVAL;
444
445 if (unlikely(event->oncpu != cpu))
446 return -EOPNOTSUPP;
447
448 return perf_event_output(event, sd, regs);
449}
450
451/*
452 * Support executing tracepoints in normal, irq, and nmi context that each call
453 * bpf_perf_event_output
454 */
455struct bpf_trace_sample_data {
456 struct perf_sample_data sds[3];
457};
458
459static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
460static DEFINE_PER_CPU(int, bpf_trace_nest_level);
461BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
462 u64, flags, void *, data, u64, size)
463{
464 struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds);
465 int nest_level = this_cpu_inc_return(bpf_trace_nest_level);
466 struct perf_raw_record raw = {
467 .frag = {
468 .size = size,
469 .data = data,
470 },
471 };
472 struct perf_sample_data *sd;
473 int err;
474
475 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) {
476 err = -EBUSY;
477 goto out;
478 }
479
480 sd = &sds->sds[nest_level - 1];
481
482 if (unlikely(flags & ~(BPF_F_INDEX_MASK))) {
483 err = -EINVAL;
484 goto out;
485 }
486
487 perf_sample_data_init(sd, 0, 0);
488 sd->raw = &raw;
489
490 err = __bpf_perf_event_output(regs, map, flags, sd);
491
492out:
493 this_cpu_dec(bpf_trace_nest_level);
494 return err;
495}
496
497static const struct bpf_func_proto bpf_perf_event_output_proto = {
498 .func = bpf_perf_event_output,
499 .gpl_only = true,
500 .ret_type = RET_INTEGER,
501 .arg1_type = ARG_PTR_TO_CTX,
502 .arg2_type = ARG_CONST_MAP_PTR,
503 .arg3_type = ARG_ANYTHING,
504 .arg4_type = ARG_PTR_TO_MEM,
505 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
506};
507
508static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
509struct bpf_nested_pt_regs {
510 struct pt_regs regs[3];
511};
512static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
513static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
514
515u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size,
516 void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy)
517{
518 int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
519 struct perf_raw_frag frag = {
520 .copy = ctx_copy,
521 .size = ctx_size,
522 .data = ctx,
523 };
524 struct perf_raw_record raw = {
525 .frag = {
526 {
527 .next = ctx_size ? &frag : NULL,
528 },
529 .size = meta_size,
530 .data = meta,
531 },
532 };
533 struct perf_sample_data *sd;
534 struct pt_regs *regs;
535 u64 ret;
536
537 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) {
538 ret = -EBUSY;
539 goto out;
540 }
541 sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]);
542 regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]);
543
544 perf_fetch_caller_regs(regs);
545 perf_sample_data_init(sd, 0, 0);
546 sd->raw = &raw;
547
548 ret = __bpf_perf_event_output(regs, map, flags, sd);
549out:
550 this_cpu_dec(bpf_event_output_nest_level);
551 return ret;
552}
553
554BPF_CALL_0(bpf_get_current_task)
555{
556 return (long) current;
557}
558
559static const struct bpf_func_proto bpf_get_current_task_proto = {
560 .func = bpf_get_current_task,
561 .gpl_only = true,
562 .ret_type = RET_INTEGER,
563};
564
565BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
566{
567 struct bpf_array *array = container_of(map, struct bpf_array, map);
568 struct cgroup *cgrp;
569
570 if (unlikely(idx >= array->map.max_entries))
571 return -E2BIG;
572
573 cgrp = READ_ONCE(array->ptrs[idx]);
574 if (unlikely(!cgrp))
575 return -EAGAIN;
576
577 return task_under_cgroup_hierarchy(current, cgrp);
578}
579
580static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
581 .func = bpf_current_task_under_cgroup,
582 .gpl_only = false,
583 .ret_type = RET_INTEGER,
584 .arg1_type = ARG_CONST_MAP_PTR,
585 .arg2_type = ARG_ANYTHING,
586};
587
588BPF_CALL_3(bpf_probe_read_str, void *, dst, u32, size,
589 const void *, unsafe_ptr)
590{
591 int ret;
592
593 ret = security_locked_down(LOCKDOWN_BPF_READ);
594 if (ret < 0)
595 goto out;
596
597 /*
598 * The strncpy_from_unsafe() call will likely not fill the entire
599 * buffer, but that's okay in this circumstance as we're probing
600 * arbitrary memory anyway similar to bpf_probe_read() and might
601 * as well probe the stack. Thus, memory is explicitly cleared
602 * only in error case, so that improper users ignoring return
603 * code altogether don't copy garbage; otherwise length of string
604 * is returned that can be used for bpf_perf_event_output() et al.
605 */
606 ret = strncpy_from_unsafe(dst, unsafe_ptr, size);
607 if (unlikely(ret < 0))
608out:
609 memset(dst, 0, size);
610
611 return ret;
612}
613
614static const struct bpf_func_proto bpf_probe_read_str_proto = {
615 .func = bpf_probe_read_str,
616 .gpl_only = true,
617 .ret_type = RET_INTEGER,
618 .arg1_type = ARG_PTR_TO_UNINIT_MEM,
619 .arg2_type = ARG_CONST_SIZE_OR_ZERO,
620 .arg3_type = ARG_ANYTHING,
621};
622
623struct send_signal_irq_work {
624 struct irq_work irq_work;
625 struct task_struct *task;
626 u32 sig;
627};
628
629static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
630
631static void do_bpf_send_signal(struct irq_work *entry)
632{
633 struct send_signal_irq_work *work;
634
635 work = container_of(entry, struct send_signal_irq_work, irq_work);
636 group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, PIDTYPE_TGID);
637}
638
639BPF_CALL_1(bpf_send_signal, u32, sig)
640{
641 struct send_signal_irq_work *work = NULL;
642
643 /* Similar to bpf_probe_write_user, task needs to be
644 * in a sound condition and kernel memory access be
645 * permitted in order to send signal to the current
646 * task.
647 */
648 if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
649 return -EPERM;
650 if (unlikely(uaccess_kernel()))
651 return -EPERM;
652 if (unlikely(!nmi_uaccess_okay()))
653 return -EPERM;
654
655 if (in_nmi()) {
656 /* Do an early check on signal validity. Otherwise,
657 * the error is lost in deferred irq_work.
658 */
659 if (unlikely(!valid_signal(sig)))
660 return -EINVAL;
661
662 work = this_cpu_ptr(&send_signal_work);
663 if (work->irq_work.flags & IRQ_WORK_BUSY)
664 return -EBUSY;
665
666 /* Add the current task, which is the target of sending signal,
667 * to the irq_work. The current task may change when queued
668 * irq works get executed.
669 */
670 work->task = current;
671 work->sig = sig;
672 irq_work_queue(&work->irq_work);
673 return 0;
674 }
675
676 return group_send_sig_info(sig, SEND_SIG_PRIV, current, PIDTYPE_TGID);
677}
678
679static const struct bpf_func_proto bpf_send_signal_proto = {
680 .func = bpf_send_signal,
681 .gpl_only = false,
682 .ret_type = RET_INTEGER,
683 .arg1_type = ARG_ANYTHING,
684};
685
686static const struct bpf_func_proto *
687tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
688{
689 switch (func_id) {
690 case BPF_FUNC_map_lookup_elem:
691 return &bpf_map_lookup_elem_proto;
692 case BPF_FUNC_map_update_elem:
693 return &bpf_map_update_elem_proto;
694 case BPF_FUNC_map_delete_elem:
695 return &bpf_map_delete_elem_proto;
696 case BPF_FUNC_map_push_elem:
697 return &bpf_map_push_elem_proto;
698 case BPF_FUNC_map_pop_elem:
699 return &bpf_map_pop_elem_proto;
700 case BPF_FUNC_map_peek_elem:
701 return &bpf_map_peek_elem_proto;
702 case BPF_FUNC_probe_read:
703 return &bpf_probe_read_proto;
704 case BPF_FUNC_ktime_get_ns:
705 return &bpf_ktime_get_ns_proto;
706 case BPF_FUNC_tail_call:
707 return &bpf_tail_call_proto;
708 case BPF_FUNC_get_current_pid_tgid:
709 return &bpf_get_current_pid_tgid_proto;
710 case BPF_FUNC_get_current_task:
711 return &bpf_get_current_task_proto;
712 case BPF_FUNC_get_current_uid_gid:
713 return &bpf_get_current_uid_gid_proto;
714 case BPF_FUNC_get_current_comm:
715 return &bpf_get_current_comm_proto;
716 case BPF_FUNC_trace_printk:
717 return bpf_get_trace_printk_proto();
718 case BPF_FUNC_get_smp_processor_id:
719 return &bpf_get_smp_processor_id_proto;
720 case BPF_FUNC_get_numa_node_id:
721 return &bpf_get_numa_node_id_proto;
722 case BPF_FUNC_perf_event_read:
723 return &bpf_perf_event_read_proto;
724 case BPF_FUNC_probe_write_user:
725 return bpf_get_probe_write_proto();
726 case BPF_FUNC_current_task_under_cgroup:
727 return &bpf_current_task_under_cgroup_proto;
728 case BPF_FUNC_get_prandom_u32:
729 return &bpf_get_prandom_u32_proto;
730 case BPF_FUNC_probe_read_str:
731 return &bpf_probe_read_str_proto;
732#ifdef CONFIG_CGROUPS
733 case BPF_FUNC_get_current_cgroup_id:
734 return &bpf_get_current_cgroup_id_proto;
735#endif
736 case BPF_FUNC_send_signal:
737 return &bpf_send_signal_proto;
738 default:
739 return NULL;
740 }
741}
742
743static const struct bpf_func_proto *
744kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
745{
746 switch (func_id) {
747 case BPF_FUNC_perf_event_output:
748 return &bpf_perf_event_output_proto;
749 case BPF_FUNC_get_stackid:
750 return &bpf_get_stackid_proto;
751 case BPF_FUNC_get_stack:
752 return &bpf_get_stack_proto;
753 case BPF_FUNC_perf_event_read_value:
754 return &bpf_perf_event_read_value_proto;
755#ifdef CONFIG_BPF_KPROBE_OVERRIDE
756 case BPF_FUNC_override_return:
757 return &bpf_override_return_proto;
758#endif
759 default:
760 return tracing_func_proto(func_id, prog);
761 }
762}
763
764/* bpf+kprobe programs can access fields of 'struct pt_regs' */
765static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
766 const struct bpf_prog *prog,
767 struct bpf_insn_access_aux *info)
768{
769 if (off < 0 || off >= sizeof(struct pt_regs))
770 return false;
771 if (type != BPF_READ)
772 return false;
773 if (off % size != 0)
774 return false;
775 /*
776 * Assertion for 32 bit to make sure last 8 byte access
777 * (BPF_DW) to the last 4 byte member is disallowed.
778 */
779 if (off + size > sizeof(struct pt_regs))
780 return false;
781
782 return true;
783}
784
785const struct bpf_verifier_ops kprobe_verifier_ops = {
786 .get_func_proto = kprobe_prog_func_proto,
787 .is_valid_access = kprobe_prog_is_valid_access,
788};
789
790const struct bpf_prog_ops kprobe_prog_ops = {
791};
792
793BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
794 u64, flags, void *, data, u64, size)
795{
796 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
797
798 /*
799 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
800 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
801 * from there and call the same bpf_perf_event_output() helper inline.
802 */
803 return ____bpf_perf_event_output(regs, map, flags, data, size);
804}
805
806static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
807 .func = bpf_perf_event_output_tp,
808 .gpl_only = true,
809 .ret_type = RET_INTEGER,
810 .arg1_type = ARG_PTR_TO_CTX,
811 .arg2_type = ARG_CONST_MAP_PTR,
812 .arg3_type = ARG_ANYTHING,
813 .arg4_type = ARG_PTR_TO_MEM,
814 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
815};
816
817BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
818 u64, flags)
819{
820 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
821
822 /*
823 * Same comment as in bpf_perf_event_output_tp(), only that this time
824 * the other helper's function body cannot be inlined due to being
825 * external, thus we need to call raw helper function.
826 */
827 return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
828 flags, 0, 0);
829}
830
831static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
832 .func = bpf_get_stackid_tp,
833 .gpl_only = true,
834 .ret_type = RET_INTEGER,
835 .arg1_type = ARG_PTR_TO_CTX,
836 .arg2_type = ARG_CONST_MAP_PTR,
837 .arg3_type = ARG_ANYTHING,
838};
839
840BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
841 u64, flags)
842{
843 struct pt_regs *regs = *(struct pt_regs **)tp_buff;
844
845 return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
846 (unsigned long) size, flags, 0);
847}
848
849static const struct bpf_func_proto bpf_get_stack_proto_tp = {
850 .func = bpf_get_stack_tp,
851 .gpl_only = true,
852 .ret_type = RET_INTEGER,
853 .arg1_type = ARG_PTR_TO_CTX,
854 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
855 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
856 .arg4_type = ARG_ANYTHING,
857};
858
859static const struct bpf_func_proto *
860tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
861{
862 switch (func_id) {
863 case BPF_FUNC_perf_event_output:
864 return &bpf_perf_event_output_proto_tp;
865 case BPF_FUNC_get_stackid:
866 return &bpf_get_stackid_proto_tp;
867 case BPF_FUNC_get_stack:
868 return &bpf_get_stack_proto_tp;
869 default:
870 return tracing_func_proto(func_id, prog);
871 }
872}
873
874static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
875 const struct bpf_prog *prog,
876 struct bpf_insn_access_aux *info)
877{
878 if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
879 return false;
880 if (type != BPF_READ)
881 return false;
882 if (off % size != 0)
883 return false;
884
885 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
886 return true;
887}
888
889const struct bpf_verifier_ops tracepoint_verifier_ops = {
890 .get_func_proto = tp_prog_func_proto,
891 .is_valid_access = tp_prog_is_valid_access,
892};
893
894const struct bpf_prog_ops tracepoint_prog_ops = {
895};
896
897BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
898 struct bpf_perf_event_value *, buf, u32, size)
899{
900 int err = -EINVAL;
901
902 if (unlikely(size != sizeof(struct bpf_perf_event_value)))
903 goto clear;
904 err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
905 &buf->running);
906 if (unlikely(err))
907 goto clear;
908 return 0;
909clear:
910 memset(buf, 0, size);
911 return err;
912}
913
914static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
915 .func = bpf_perf_prog_read_value,
916 .gpl_only = true,
917 .ret_type = RET_INTEGER,
918 .arg1_type = ARG_PTR_TO_CTX,
919 .arg2_type = ARG_PTR_TO_UNINIT_MEM,
920 .arg3_type = ARG_CONST_SIZE,
921};
922
923static const struct bpf_func_proto *
924pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
925{
926 switch (func_id) {
927 case BPF_FUNC_perf_event_output:
928 return &bpf_perf_event_output_proto_tp;
929 case BPF_FUNC_get_stackid:
930 return &bpf_get_stackid_proto_tp;
931 case BPF_FUNC_get_stack:
932 return &bpf_get_stack_proto_tp;
933 case BPF_FUNC_perf_prog_read_value:
934 return &bpf_perf_prog_read_value_proto;
935 default:
936 return tracing_func_proto(func_id, prog);
937 }
938}
939
940/*
941 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
942 * to avoid potential recursive reuse issue when/if tracepoints are added
943 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
944 *
945 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
946 * in normal, irq, and nmi context.
947 */
948struct bpf_raw_tp_regs {
949 struct pt_regs regs[3];
950};
951static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
952static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
953static struct pt_regs *get_bpf_raw_tp_regs(void)
954{
955 struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
956 int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
957
958 if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
959 this_cpu_dec(bpf_raw_tp_nest_level);
960 return ERR_PTR(-EBUSY);
961 }
962
963 return &tp_regs->regs[nest_level - 1];
964}
965
966static void put_bpf_raw_tp_regs(void)
967{
968 this_cpu_dec(bpf_raw_tp_nest_level);
969}
970
971BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
972 struct bpf_map *, map, u64, flags, void *, data, u64, size)
973{
974 struct pt_regs *regs = get_bpf_raw_tp_regs();
975 int ret;
976
977 if (IS_ERR(regs))
978 return PTR_ERR(regs);
979
980 perf_fetch_caller_regs(regs);
981 ret = ____bpf_perf_event_output(regs, map, flags, data, size);
982
983 put_bpf_raw_tp_regs();
984 return ret;
985}
986
987static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
988 .func = bpf_perf_event_output_raw_tp,
989 .gpl_only = true,
990 .ret_type = RET_INTEGER,
991 .arg1_type = ARG_PTR_TO_CTX,
992 .arg2_type = ARG_CONST_MAP_PTR,
993 .arg3_type = ARG_ANYTHING,
994 .arg4_type = ARG_PTR_TO_MEM,
995 .arg5_type = ARG_CONST_SIZE_OR_ZERO,
996};
997
998BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
999 struct bpf_map *, map, u64, flags)
1000{
1001 struct pt_regs *regs = get_bpf_raw_tp_regs();
1002 int ret;
1003
1004 if (IS_ERR(regs))
1005 return PTR_ERR(regs);
1006
1007 perf_fetch_caller_regs(regs);
1008 /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1009 ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1010 flags, 0, 0);
1011 put_bpf_raw_tp_regs();
1012 return ret;
1013}
1014
1015static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1016 .func = bpf_get_stackid_raw_tp,
1017 .gpl_only = true,
1018 .ret_type = RET_INTEGER,
1019 .arg1_type = ARG_PTR_TO_CTX,
1020 .arg2_type = ARG_CONST_MAP_PTR,
1021 .arg3_type = ARG_ANYTHING,
1022};
1023
1024BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1025 void *, buf, u32, size, u64, flags)
1026{
1027 struct pt_regs *regs = get_bpf_raw_tp_regs();
1028 int ret;
1029
1030 if (IS_ERR(regs))
1031 return PTR_ERR(regs);
1032
1033 perf_fetch_caller_regs(regs);
1034 ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1035 (unsigned long) size, flags, 0);
1036 put_bpf_raw_tp_regs();
1037 return ret;
1038}
1039
1040static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1041 .func = bpf_get_stack_raw_tp,
1042 .gpl_only = true,
1043 .ret_type = RET_INTEGER,
1044 .arg1_type = ARG_PTR_TO_CTX,
1045 .arg2_type = ARG_PTR_TO_MEM,
1046 .arg3_type = ARG_CONST_SIZE_OR_ZERO,
1047 .arg4_type = ARG_ANYTHING,
1048};
1049
1050static const struct bpf_func_proto *
1051raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1052{
1053 switch (func_id) {
1054 case BPF_FUNC_perf_event_output:
1055 return &bpf_perf_event_output_proto_raw_tp;
1056 case BPF_FUNC_get_stackid:
1057 return &bpf_get_stackid_proto_raw_tp;
1058 case BPF_FUNC_get_stack:
1059 return &bpf_get_stack_proto_raw_tp;
1060 default:
1061 return tracing_func_proto(func_id, prog);
1062 }
1063}
1064
1065static bool raw_tp_prog_is_valid_access(int off, int size,
1066 enum bpf_access_type type,
1067 const struct bpf_prog *prog,
1068 struct bpf_insn_access_aux *info)
1069{
1070 /* largest tracepoint in the kernel has 12 args */
1071 if (off < 0 || off >= sizeof(__u64) * 12)
1072 return false;
1073 if (type != BPF_READ)
1074 return false;
1075 if (off % size != 0)
1076 return false;
1077 return true;
1078}
1079
1080const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1081 .get_func_proto = raw_tp_prog_func_proto,
1082 .is_valid_access = raw_tp_prog_is_valid_access,
1083};
1084
1085const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1086};
1087
1088static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1089 enum bpf_access_type type,
1090 const struct bpf_prog *prog,
1091 struct bpf_insn_access_aux *info)
1092{
1093 if (off == 0) {
1094 if (size != sizeof(u64) || type != BPF_READ)
1095 return false;
1096 info->reg_type = PTR_TO_TP_BUFFER;
1097 }
1098 return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1099}
1100
1101const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1102 .get_func_proto = raw_tp_prog_func_proto,
1103 .is_valid_access = raw_tp_writable_prog_is_valid_access,
1104};
1105
1106const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1107};
1108
1109static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1110 const struct bpf_prog *prog,
1111 struct bpf_insn_access_aux *info)
1112{
1113 const int size_u64 = sizeof(u64);
1114
1115 if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1116 return false;
1117 if (type != BPF_READ)
1118 return false;
1119 if (off % size != 0) {
1120 if (sizeof(unsigned long) != 4)
1121 return false;
1122 if (size != 8)
1123 return false;
1124 if (off % size != 4)
1125 return false;
1126 }
1127
1128 switch (off) {
1129 case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1130 bpf_ctx_record_field_size(info, size_u64);
1131 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1132 return false;
1133 break;
1134 case bpf_ctx_range(struct bpf_perf_event_data, addr):
1135 bpf_ctx_record_field_size(info, size_u64);
1136 if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1137 return false;
1138 break;
1139 default:
1140 if (size != sizeof(long))
1141 return false;
1142 }
1143
1144 return true;
1145}
1146
1147static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1148 const struct bpf_insn *si,
1149 struct bpf_insn *insn_buf,
1150 struct bpf_prog *prog, u32 *target_size)
1151{
1152 struct bpf_insn *insn = insn_buf;
1153
1154 switch (si->off) {
1155 case offsetof(struct bpf_perf_event_data, sample_period):
1156 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1157 data), si->dst_reg, si->src_reg,
1158 offsetof(struct bpf_perf_event_data_kern, data));
1159 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1160 bpf_target_off(struct perf_sample_data, period, 8,
1161 target_size));
1162 break;
1163 case offsetof(struct bpf_perf_event_data, addr):
1164 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1165 data), si->dst_reg, si->src_reg,
1166 offsetof(struct bpf_perf_event_data_kern, data));
1167 *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1168 bpf_target_off(struct perf_sample_data, addr, 8,
1169 target_size));
1170 break;
1171 default:
1172 *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1173 regs), si->dst_reg, si->src_reg,
1174 offsetof(struct bpf_perf_event_data_kern, regs));
1175 *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1176 si->off);
1177 break;
1178 }
1179
1180 return insn - insn_buf;
1181}
1182
1183const struct bpf_verifier_ops perf_event_verifier_ops = {
1184 .get_func_proto = pe_prog_func_proto,
1185 .is_valid_access = pe_prog_is_valid_access,
1186 .convert_ctx_access = pe_prog_convert_ctx_access,
1187};
1188
1189const struct bpf_prog_ops perf_event_prog_ops = {
1190};
1191
1192static DEFINE_MUTEX(bpf_event_mutex);
1193
1194#define BPF_TRACE_MAX_PROGS 64
1195
1196int perf_event_attach_bpf_prog(struct perf_event *event,
1197 struct bpf_prog *prog)
1198{
1199 struct bpf_prog_array *old_array;
1200 struct bpf_prog_array *new_array;
1201 int ret = -EEXIST;
1202
1203 /*
1204 * Kprobe override only works if they are on the function entry,
1205 * and only if they are on the opt-in list.
1206 */
1207 if (prog->kprobe_override &&
1208 (!trace_kprobe_on_func_entry(event->tp_event) ||
1209 !trace_kprobe_error_injectable(event->tp_event)))
1210 return -EINVAL;
1211
1212 mutex_lock(&bpf_event_mutex);
1213
1214 if (event->prog)
1215 goto unlock;
1216
1217 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1218 if (old_array &&
1219 bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1220 ret = -E2BIG;
1221 goto unlock;
1222 }
1223
1224 ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1225 if (ret < 0)
1226 goto unlock;
1227
1228 /* set the new array to event->tp_event and set event->prog */
1229 event->prog = prog;
1230 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1231 bpf_prog_array_free(old_array);
1232
1233unlock:
1234 mutex_unlock(&bpf_event_mutex);
1235 return ret;
1236}
1237
1238void perf_event_detach_bpf_prog(struct perf_event *event)
1239{
1240 struct bpf_prog_array *old_array;
1241 struct bpf_prog_array *new_array;
1242 int ret;
1243
1244 mutex_lock(&bpf_event_mutex);
1245
1246 if (!event->prog)
1247 goto unlock;
1248
1249 old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1250 ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1251 if (ret == -ENOENT)
1252 goto unlock;
1253 if (ret < 0) {
1254 bpf_prog_array_delete_safe(old_array, event->prog);
1255 } else {
1256 rcu_assign_pointer(event->tp_event->prog_array, new_array);
1257 bpf_prog_array_free(old_array);
1258 }
1259
1260 bpf_prog_put(event->prog);
1261 event->prog = NULL;
1262
1263unlock:
1264 mutex_unlock(&bpf_event_mutex);
1265}
1266
1267int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1268{
1269 struct perf_event_query_bpf __user *uquery = info;
1270 struct perf_event_query_bpf query = {};
1271 struct bpf_prog_array *progs;
1272 u32 *ids, prog_cnt, ids_len;
1273 int ret;
1274
1275 if (!capable(CAP_SYS_ADMIN))
1276 return -EPERM;
1277 if (event->attr.type != PERF_TYPE_TRACEPOINT)
1278 return -EINVAL;
1279 if (copy_from_user(&query, uquery, sizeof(query)))
1280 return -EFAULT;
1281
1282 ids_len = query.ids_len;
1283 if (ids_len > BPF_TRACE_MAX_PROGS)
1284 return -E2BIG;
1285 ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1286 if (!ids)
1287 return -ENOMEM;
1288 /*
1289 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1290 * is required when user only wants to check for uquery->prog_cnt.
1291 * There is no need to check for it since the case is handled
1292 * gracefully in bpf_prog_array_copy_info.
1293 */
1294
1295 mutex_lock(&bpf_event_mutex);
1296 progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1297 ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1298 mutex_unlock(&bpf_event_mutex);
1299
1300 if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1301 copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1302 ret = -EFAULT;
1303
1304 kfree(ids);
1305 return ret;
1306}
1307
1308extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1309extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1310
1311struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1312{
1313 struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1314
1315 for (; btp < __stop__bpf_raw_tp; btp++) {
1316 if (!strcmp(btp->tp->name, name))
1317 return btp;
1318 }
1319
1320 return bpf_get_raw_tracepoint_module(name);
1321}
1322
1323void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1324{
1325 struct module *mod = __module_address((unsigned long)btp);
1326
1327 if (mod)
1328 module_put(mod);
1329}
1330
1331static __always_inline
1332void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1333{
1334 rcu_read_lock();
1335 preempt_disable();
1336 (void) BPF_PROG_RUN(prog, args);
1337 preempt_enable();
1338 rcu_read_unlock();
1339}
1340
1341#define UNPACK(...) __VA_ARGS__
1342#define REPEAT_1(FN, DL, X, ...) FN(X)
1343#define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1344#define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1345#define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1346#define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1347#define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1348#define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1349#define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1350#define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1351#define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1352#define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1353#define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1354#define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__)
1355
1356#define SARG(X) u64 arg##X
1357#define COPY(X) args[X] = arg##X
1358
1359#define __DL_COM (,)
1360#define __DL_SEM (;)
1361
1362#define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1363
1364#define BPF_TRACE_DEFN_x(x) \
1365 void bpf_trace_run##x(struct bpf_prog *prog, \
1366 REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \
1367 { \
1368 u64 args[x]; \
1369 REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \
1370 __bpf_trace_run(prog, args); \
1371 } \
1372 EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1373BPF_TRACE_DEFN_x(1);
1374BPF_TRACE_DEFN_x(2);
1375BPF_TRACE_DEFN_x(3);
1376BPF_TRACE_DEFN_x(4);
1377BPF_TRACE_DEFN_x(5);
1378BPF_TRACE_DEFN_x(6);
1379BPF_TRACE_DEFN_x(7);
1380BPF_TRACE_DEFN_x(8);
1381BPF_TRACE_DEFN_x(9);
1382BPF_TRACE_DEFN_x(10);
1383BPF_TRACE_DEFN_x(11);
1384BPF_TRACE_DEFN_x(12);
1385
1386static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1387{
1388 struct tracepoint *tp = btp->tp;
1389
1390 /*
1391 * check that program doesn't access arguments beyond what's
1392 * available in this tracepoint
1393 */
1394 if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1395 return -EINVAL;
1396
1397 if (prog->aux->max_tp_access > btp->writable_size)
1398 return -EINVAL;
1399
1400 return tracepoint_probe_register(tp, (void *)btp->bpf_func, prog);
1401}
1402
1403int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1404{
1405 return __bpf_probe_register(btp, prog);
1406}
1407
1408int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1409{
1410 return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1411}
1412
1413int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1414 u32 *fd_type, const char **buf,
1415 u64 *probe_offset, u64 *probe_addr)
1416{
1417 bool is_tracepoint, is_syscall_tp;
1418 struct bpf_prog *prog;
1419 int flags, err = 0;
1420
1421 prog = event->prog;
1422 if (!prog)
1423 return -ENOENT;
1424
1425 /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1426 if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1427 return -EOPNOTSUPP;
1428
1429 *prog_id = prog->aux->id;
1430 flags = event->tp_event->flags;
1431 is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1432 is_syscall_tp = is_syscall_trace_event(event->tp_event);
1433
1434 if (is_tracepoint || is_syscall_tp) {
1435 *buf = is_tracepoint ? event->tp_event->tp->name
1436 : event->tp_event->name;
1437 *fd_type = BPF_FD_TYPE_TRACEPOINT;
1438 *probe_offset = 0x0;
1439 *probe_addr = 0x0;
1440 } else {
1441 /* kprobe/uprobe */
1442 err = -EOPNOTSUPP;
1443#ifdef CONFIG_KPROBE_EVENTS
1444 if (flags & TRACE_EVENT_FL_KPROBE)
1445 err = bpf_get_kprobe_info(event, fd_type, buf,
1446 probe_offset, probe_addr,
1447 event->attr.type == PERF_TYPE_TRACEPOINT);
1448#endif
1449#ifdef CONFIG_UPROBE_EVENTS
1450 if (flags & TRACE_EVENT_FL_UPROBE)
1451 err = bpf_get_uprobe_info(event, fd_type, buf,
1452 probe_offset,
1453 event->attr.type == PERF_TYPE_TRACEPOINT);
1454#endif
1455 }
1456
1457 return err;
1458}
1459
1460static int __init send_signal_irq_work_init(void)
1461{
1462 int cpu;
1463 struct send_signal_irq_work *work;
1464
1465 for_each_possible_cpu(cpu) {
1466 work = per_cpu_ptr(&send_signal_work, cpu);
1467 init_irq_work(&work->irq_work, do_bpf_send_signal);
1468 }
1469 return 0;
1470}
1471
1472subsys_initcall(send_signal_irq_work_init);
1473
1474#ifdef CONFIG_MODULES
1475static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1476 void *module)
1477{
1478 struct bpf_trace_module *btm, *tmp;
1479 struct module *mod = module;
1480
1481 if (mod->num_bpf_raw_events == 0 ||
1482 (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1483 return 0;
1484
1485 mutex_lock(&bpf_module_mutex);
1486
1487 switch (op) {
1488 case MODULE_STATE_COMING:
1489 btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1490 if (btm) {
1491 btm->module = module;
1492 list_add(&btm->list, &bpf_trace_modules);
1493 }
1494 break;
1495 case MODULE_STATE_GOING:
1496 list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1497 if (btm->module == module) {
1498 list_del(&btm->list);
1499 kfree(btm);
1500 break;
1501 }
1502 }
1503 break;
1504 }
1505
1506 mutex_unlock(&bpf_module_mutex);
1507
1508 return 0;
1509}
1510
1511static struct notifier_block bpf_module_nb = {
1512 .notifier_call = bpf_event_notify,
1513};
1514
1515static int __init bpf_event_init(void)
1516{
1517 register_module_notifier(&bpf_module_nb);
1518 return 0;
1519}
1520
1521fs_initcall(bpf_event_init);
1522#endif /* CONFIG_MODULES */