Commit | Line | Data |
---|---|---|
179a0cc4 | 1 | // SPDX-License-Identifier: GPL-2.0 |
2541517c | 2 | /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com |
0515e599 | 3 | * Copyright (c) 2016 Facebook |
2541517c AS |
4 | */ |
5 | #include <linux/kernel.h> | |
6 | #include <linux/types.h> | |
7 | #include <linux/slab.h> | |
8 | #include <linux/bpf.h> | |
0515e599 | 9 | #include <linux/bpf_perf_event.h> |
c4d0bfb4 | 10 | #include <linux/btf.h> |
2541517c AS |
11 | #include <linux/filter.h> |
12 | #include <linux/uaccess.h> | |
9c959c86 | 13 | #include <linux/ctype.h> |
9802d865 | 14 | #include <linux/kprobes.h> |
ac5a72ea | 15 | #include <linux/spinlock.h> |
41bdc4b4 | 16 | #include <linux/syscalls.h> |
540adea3 | 17 | #include <linux/error-injection.h> |
c9a0f3b8 | 18 | #include <linux/btf_ids.h> |
6f100640 KS |
19 | #include <linux/bpf_lsm.h> |
20 | ||
8e4597c6 | 21 | #include <net/bpf_sk_storage.h> |
9802d865 | 22 | |
c4d0bfb4 AM |
23 | #include <uapi/linux/bpf.h> |
24 | #include <uapi/linux/btf.h> | |
25 | ||
c7b6f29b NA |
26 | #include <asm/tlb.h> |
27 | ||
9802d865 | 28 | #include "trace_probe.h" |
2541517c AS |
29 | #include "trace.h" |
30 | ||
ac5a72ea AM |
31 | #define CREATE_TRACE_POINTS |
32 | #include "bpf_trace.h" | |
33 | ||
e672db03 SF |
34 | #define bpf_event_rcu_dereference(p) \ |
35 | rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex)) | |
36 | ||
a38d1107 MM |
37 | #ifdef CONFIG_MODULES |
38 | struct bpf_trace_module { | |
39 | struct module *module; | |
40 | struct list_head list; | |
41 | }; | |
42 | ||
43 | static LIST_HEAD(bpf_trace_modules); | |
44 | static DEFINE_MUTEX(bpf_module_mutex); | |
45 | ||
46 | static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) | |
47 | { | |
48 | struct bpf_raw_event_map *btp, *ret = NULL; | |
49 | struct bpf_trace_module *btm; | |
50 | unsigned int i; | |
51 | ||
52 | mutex_lock(&bpf_module_mutex); | |
53 | list_for_each_entry(btm, &bpf_trace_modules, list) { | |
54 | for (i = 0; i < btm->module->num_bpf_raw_events; ++i) { | |
55 | btp = &btm->module->bpf_raw_events[i]; | |
56 | if (!strcmp(btp->tp->name, name)) { | |
57 | if (try_module_get(btm->module)) | |
58 | ret = btp; | |
59 | goto out; | |
60 | } | |
61 | } | |
62 | } | |
63 | out: | |
64 | mutex_unlock(&bpf_module_mutex); | |
65 | return ret; | |
66 | } | |
67 | #else | |
68 | static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name) | |
69 | { | |
70 | return NULL; | |
71 | } | |
72 | #endif /* CONFIG_MODULES */ | |
73 | ||
035226b9 | 74 | u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
c195651e | 75 | u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5); |
035226b9 | 76 | |
eb411377 AM |
77 | static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size, |
78 | u64 flags, const struct btf **btf, | |
79 | s32 *btf_id); | |
80 | ||
2541517c AS |
81 | /** |
82 | * trace_call_bpf - invoke BPF program | |
e87c6bc3 | 83 | * @call: tracepoint event |
2541517c AS |
84 | * @ctx: opaque context pointer |
85 | * | |
86 | * kprobe handlers execute BPF programs via this helper. | |
87 | * Can be used from static tracepoints in the future. | |
88 | * | |
89 | * Return: BPF programs always return an integer which is interpreted by | |
90 | * kprobe handler as: | |
91 | * 0 - return from kprobe (event is filtered out) | |
92 | * 1 - store kprobe event into ring buffer | |
93 | * Other values are reserved and currently alias to 1 | |
94 | */ | |
e87c6bc3 | 95 | unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx) |
2541517c AS |
96 | { |
97 | unsigned int ret; | |
98 | ||
b0a81b94 | 99 | cant_sleep(); |
2541517c AS |
100 | |
101 | if (unlikely(__this_cpu_inc_return(bpf_prog_active) != 1)) { | |
102 | /* | |
103 | * since some bpf program is already running on this cpu, | |
104 | * don't call into another bpf program (same or different) | |
105 | * and don't send kprobe event into ring-buffer, | |
106 | * so return zero here | |
107 | */ | |
108 | ret = 0; | |
109 | goto out; | |
110 | } | |
111 | ||
e87c6bc3 YS |
112 | /* |
113 | * Instead of moving rcu_read_lock/rcu_dereference/rcu_read_unlock | |
114 | * to all call sites, we did a bpf_prog_array_valid() there to check | |
115 | * whether call->prog_array is empty or not, which is | |
2b5894cc | 116 | * a heuristic to speed up execution. |
e87c6bc3 YS |
117 | * |
118 | * If bpf_prog_array_valid() fetched prog_array was | |
119 | * non-NULL, we go into trace_call_bpf() and do the actual | |
120 | * proper rcu_dereference() under RCU lock. | |
121 | * If it turns out that prog_array is NULL then, we bail out. | |
122 | * For the opposite, if the bpf_prog_array_valid() fetched pointer | |
123 | * was NULL, you'll skip the prog_array with the risk of missing | |
124 | * out of events when it was updated in between this and the | |
125 | * rcu_dereference() which is accepted risk. | |
126 | */ | |
7d08c2c9 | 127 | ret = BPF_PROG_RUN_ARRAY(call->prog_array, ctx, bpf_prog_run); |
2541517c AS |
128 | |
129 | out: | |
130 | __this_cpu_dec(bpf_prog_active); | |
2541517c AS |
131 | |
132 | return ret; | |
133 | } | |
2541517c | 134 | |
9802d865 JB |
135 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
136 | BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc) | |
137 | { | |
9802d865 | 138 | regs_set_return_value(regs, rc); |
540adea3 | 139 | override_function_with_return(regs); |
9802d865 JB |
140 | return 0; |
141 | } | |
142 | ||
143 | static const struct bpf_func_proto bpf_override_return_proto = { | |
144 | .func = bpf_override_return, | |
145 | .gpl_only = true, | |
146 | .ret_type = RET_INTEGER, | |
147 | .arg1_type = ARG_PTR_TO_CTX, | |
148 | .arg2_type = ARG_ANYTHING, | |
149 | }; | |
150 | #endif | |
151 | ||
8d92db5c CH |
152 | static __always_inline int |
153 | bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr) | |
2541517c | 154 | { |
8d92db5c | 155 | int ret; |
2541517c | 156 | |
c0ee37e8 | 157 | ret = copy_from_user_nofault(dst, unsafe_ptr, size); |
6ae08ae3 DB |
158 | if (unlikely(ret < 0)) |
159 | memset(dst, 0, size); | |
6ae08ae3 DB |
160 | return ret; |
161 | } | |
162 | ||
8d92db5c CH |
163 | BPF_CALL_3(bpf_probe_read_user, void *, dst, u32, size, |
164 | const void __user *, unsafe_ptr) | |
165 | { | |
166 | return bpf_probe_read_user_common(dst, size, unsafe_ptr); | |
167 | } | |
168 | ||
f470378c | 169 | const struct bpf_func_proto bpf_probe_read_user_proto = { |
6ae08ae3 DB |
170 | .func = bpf_probe_read_user, |
171 | .gpl_only = true, | |
172 | .ret_type = RET_INTEGER, | |
173 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
174 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
175 | .arg3_type = ARG_ANYTHING, | |
176 | }; | |
177 | ||
8d92db5c CH |
178 | static __always_inline int |
179 | bpf_probe_read_user_str_common(void *dst, u32 size, | |
180 | const void __user *unsafe_ptr) | |
6ae08ae3 | 181 | { |
8d92db5c | 182 | int ret; |
6ae08ae3 | 183 | |
6fa6d280 DX |
184 | /* |
185 | * NB: We rely on strncpy_from_user() not copying junk past the NUL | |
186 | * terminator into `dst`. | |
187 | * | |
188 | * strncpy_from_user() does long-sized strides in the fast path. If the | |
189 | * strncpy does not mask out the bytes after the NUL in `unsafe_ptr`, | |
190 | * then there could be junk after the NUL in `dst`. If user takes `dst` | |
191 | * and keys a hash map with it, then semantically identical strings can | |
192 | * occupy multiple entries in the map. | |
193 | */ | |
8d92db5c | 194 | ret = strncpy_from_user_nofault(dst, unsafe_ptr, size); |
6ae08ae3 DB |
195 | if (unlikely(ret < 0)) |
196 | memset(dst, 0, size); | |
6ae08ae3 DB |
197 | return ret; |
198 | } | |
199 | ||
8d92db5c CH |
200 | BPF_CALL_3(bpf_probe_read_user_str, void *, dst, u32, size, |
201 | const void __user *, unsafe_ptr) | |
202 | { | |
203 | return bpf_probe_read_user_str_common(dst, size, unsafe_ptr); | |
204 | } | |
205 | ||
f470378c | 206 | const struct bpf_func_proto bpf_probe_read_user_str_proto = { |
6ae08ae3 DB |
207 | .func = bpf_probe_read_user_str, |
208 | .gpl_only = true, | |
209 | .ret_type = RET_INTEGER, | |
210 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
211 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
212 | .arg3_type = ARG_ANYTHING, | |
213 | }; | |
214 | ||
215 | static __always_inline int | |
8d92db5c | 216 | bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr) |
6ae08ae3 | 217 | { |
ff40e510 | 218 | int ret; |
9d1f8be5 | 219 | |
fe557319 | 220 | ret = copy_from_kernel_nofault(dst, unsafe_ptr, size); |
074f528e | 221 | if (unlikely(ret < 0)) |
ff40e510 | 222 | memset(dst, 0, size); |
6ae08ae3 DB |
223 | return ret; |
224 | } | |
074f528e | 225 | |
6ae08ae3 DB |
226 | BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size, |
227 | const void *, unsafe_ptr) | |
228 | { | |
8d92db5c | 229 | return bpf_probe_read_kernel_common(dst, size, unsafe_ptr); |
6ae08ae3 DB |
230 | } |
231 | ||
f470378c | 232 | const struct bpf_func_proto bpf_probe_read_kernel_proto = { |
6ae08ae3 DB |
233 | .func = bpf_probe_read_kernel, |
234 | .gpl_only = true, | |
235 | .ret_type = RET_INTEGER, | |
236 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
237 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
238 | .arg3_type = ARG_ANYTHING, | |
239 | }; | |
240 | ||
6ae08ae3 | 241 | static __always_inline int |
8d92db5c | 242 | bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr) |
6ae08ae3 | 243 | { |
ff40e510 | 244 | int ret; |
8d92db5c | 245 | |
6ae08ae3 | 246 | /* |
8d92db5c CH |
247 | * The strncpy_from_kernel_nofault() call will likely not fill the |
248 | * entire buffer, but that's okay in this circumstance as we're probing | |
6ae08ae3 DB |
249 | * arbitrary memory anyway similar to bpf_probe_read_*() and might |
250 | * as well probe the stack. Thus, memory is explicitly cleared | |
251 | * only in error case, so that improper users ignoring return | |
252 | * code altogether don't copy garbage; otherwise length of string | |
253 | * is returned that can be used for bpf_perf_event_output() et al. | |
254 | */ | |
8d92db5c | 255 | ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size); |
6ae08ae3 | 256 | if (unlikely(ret < 0)) |
ff40e510 | 257 | memset(dst, 0, size); |
074f528e | 258 | return ret; |
2541517c AS |
259 | } |
260 | ||
6ae08ae3 DB |
261 | BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size, |
262 | const void *, unsafe_ptr) | |
263 | { | |
8d92db5c | 264 | return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr); |
6ae08ae3 DB |
265 | } |
266 | ||
f470378c | 267 | const struct bpf_func_proto bpf_probe_read_kernel_str_proto = { |
6ae08ae3 DB |
268 | .func = bpf_probe_read_kernel_str, |
269 | .gpl_only = true, | |
270 | .ret_type = RET_INTEGER, | |
271 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
272 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
273 | .arg3_type = ARG_ANYTHING, | |
274 | }; | |
275 | ||
8d92db5c CH |
276 | #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE |
277 | BPF_CALL_3(bpf_probe_read_compat, void *, dst, u32, size, | |
278 | const void *, unsafe_ptr) | |
279 | { | |
280 | if ((unsigned long)unsafe_ptr < TASK_SIZE) { | |
281 | return bpf_probe_read_user_common(dst, size, | |
282 | (__force void __user *)unsafe_ptr); | |
283 | } | |
284 | return bpf_probe_read_kernel_common(dst, size, unsafe_ptr); | |
285 | } | |
286 | ||
287 | static const struct bpf_func_proto bpf_probe_read_compat_proto = { | |
288 | .func = bpf_probe_read_compat, | |
289 | .gpl_only = true, | |
290 | .ret_type = RET_INTEGER, | |
291 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, | |
292 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, | |
293 | .arg3_type = ARG_ANYTHING, | |
294 | }; | |
295 | ||
6ae08ae3 DB |
296 | BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size, |
297 | const void *, unsafe_ptr) | |
298 | { | |
8d92db5c CH |
299 | if ((unsigned long)unsafe_ptr < TASK_SIZE) { |
300 | return bpf_probe_read_user_str_common(dst, size, | |
301 | (__force void __user *)unsafe_ptr); | |
302 | } | |
303 | return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr); | |
6ae08ae3 DB |
304 | } |
305 | ||
306 | static const struct bpf_func_proto bpf_probe_read_compat_str_proto = { | |
307 | .func = bpf_probe_read_compat_str, | |
2541517c AS |
308 | .gpl_only = true, |
309 | .ret_type = RET_INTEGER, | |
39f19ebb | 310 | .arg1_type = ARG_PTR_TO_UNINIT_MEM, |
9c019e2b | 311 | .arg2_type = ARG_CONST_SIZE_OR_ZERO, |
2541517c AS |
312 | .arg3_type = ARG_ANYTHING, |
313 | }; | |
8d92db5c | 314 | #endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */ |
2541517c | 315 | |
eb1b6688 | 316 | BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src, |
f3694e00 | 317 | u32, size) |
96ae5227 | 318 | { |
96ae5227 SD |
319 | /* |
320 | * Ensure we're in user context which is safe for the helper to | |
321 | * run. This helper has no business in a kthread. | |
322 | * | |
323 | * access_ok() should prevent writing to non-user memory, but in | |
324 | * some situations (nommu, temporary switch, etc) access_ok() does | |
325 | * not provide enough validation, hence the check on KERNEL_DS. | |
c7b6f29b NA |
326 | * |
327 | * nmi_uaccess_okay() ensures the probe is not run in an interim | |
328 | * state, when the task or mm are switched. This is specifically | |
329 | * required to prevent the use of temporary mm. | |
96ae5227 SD |
330 | */ |
331 | ||
332 | if (unlikely(in_interrupt() || | |
333 | current->flags & (PF_KTHREAD | PF_EXITING))) | |
334 | return -EPERM; | |
db68ce10 | 335 | if (unlikely(uaccess_kernel())) |
96ae5227 | 336 | return -EPERM; |
c7b6f29b NA |
337 | if (unlikely(!nmi_uaccess_okay())) |
338 | return -EPERM; | |
96ae5227 | 339 | |
c0ee37e8 | 340 | return copy_to_user_nofault(unsafe_ptr, src, size); |
96ae5227 SD |
341 | } |
342 | ||
343 | static const struct bpf_func_proto bpf_probe_write_user_proto = { | |
344 | .func = bpf_probe_write_user, | |
345 | .gpl_only = true, | |
346 | .ret_type = RET_INTEGER, | |
347 | .arg1_type = ARG_ANYTHING, | |
39f19ebb AS |
348 | .arg2_type = ARG_PTR_TO_MEM, |
349 | .arg3_type = ARG_CONST_SIZE, | |
96ae5227 SD |
350 | }; |
351 | ||
352 | static const struct bpf_func_proto *bpf_get_probe_write_proto(void) | |
353 | { | |
2c78ee89 AS |
354 | if (!capable(CAP_SYS_ADMIN)) |
355 | return NULL; | |
356 | ||
96ae5227 SD |
357 | pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!", |
358 | current->comm, task_pid_nr(current)); | |
359 | ||
360 | return &bpf_probe_write_user_proto; | |
361 | } | |
362 | ||
ac5a72ea AM |
363 | static DEFINE_RAW_SPINLOCK(trace_printk_lock); |
364 | ||
d9c9e4db FR |
365 | #define MAX_TRACE_PRINTK_VARARGS 3 |
366 | #define BPF_TRACE_PRINTK_SIZE 1024 | |
ac5a72ea | 367 | |
d9c9e4db FR |
368 | BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1, |
369 | u64, arg2, u64, arg3) | |
ac5a72ea | 370 | { |
d9c9e4db | 371 | u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 }; |
48cac3f4 | 372 | u32 *bin_args; |
ac5a72ea AM |
373 | static char buf[BPF_TRACE_PRINTK_SIZE]; |
374 | unsigned long flags; | |
ac5a72ea AM |
375 | int ret; |
376 | ||
48cac3f4 FR |
377 | ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args, |
378 | MAX_TRACE_PRINTK_VARARGS); | |
d9c9e4db FR |
379 | if (ret < 0) |
380 | return ret; | |
381 | ||
38d26d89 | 382 | raw_spin_lock_irqsave(&trace_printk_lock, flags); |
48cac3f4 | 383 | ret = bstr_printf(buf, sizeof(buf), fmt, bin_args); |
d9c9e4db | 384 | |
ac5a72ea AM |
385 | trace_bpf_trace_printk(buf); |
386 | raw_spin_unlock_irqrestore(&trace_printk_lock, flags); | |
387 | ||
48cac3f4 | 388 | bpf_bprintf_cleanup(); |
9c959c86 | 389 | |
d9c9e4db | 390 | return ret; |
9c959c86 AS |
391 | } |
392 | ||
393 | static const struct bpf_func_proto bpf_trace_printk_proto = { | |
394 | .func = bpf_trace_printk, | |
395 | .gpl_only = true, | |
396 | .ret_type = RET_INTEGER, | |
39f19ebb AS |
397 | .arg1_type = ARG_PTR_TO_MEM, |
398 | .arg2_type = ARG_CONST_SIZE, | |
9c959c86 AS |
399 | }; |
400 | ||
0756ea3e AS |
401 | const struct bpf_func_proto *bpf_get_trace_printk_proto(void) |
402 | { | |
403 | /* | |
ac5a72ea AM |
404 | * This program might be calling bpf_trace_printk, |
405 | * so enable the associated bpf_trace/bpf_trace_printk event. | |
406 | * Repeat this each time as it is possible a user has | |
407 | * disabled bpf_trace_printk events. By loading a program | |
408 | * calling bpf_trace_printk() however the user has expressed | |
409 | * the intent to see such events. | |
0756ea3e | 410 | */ |
ac5a72ea AM |
411 | if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1)) |
412 | pr_warn_ratelimited("could not enable bpf_trace_printk events"); | |
0756ea3e AS |
413 | |
414 | return &bpf_trace_printk_proto; | |
415 | } | |
416 | ||
492e639f | 417 | #define MAX_SEQ_PRINTF_VARARGS 12 |
492e639f YS |
418 | |
419 | BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size, | |
420 | const void *, data, u32, data_len) | |
421 | { | |
d9c9e4db | 422 | int err, num_args; |
48cac3f4 | 423 | u32 *bin_args; |
492e639f | 424 | |
d9c9e4db FR |
425 | if (data_len & 7 || data_len > MAX_SEQ_PRINTF_VARARGS * 8 || |
426 | (data_len && !data)) | |
427 | return -EINVAL; | |
492e639f YS |
428 | num_args = data_len / 8; |
429 | ||
48cac3f4 | 430 | err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args); |
d9c9e4db FR |
431 | if (err < 0) |
432 | return err; | |
492e639f | 433 | |
48cac3f4 FR |
434 | seq_bprintf(m, fmt, bin_args); |
435 | ||
436 | bpf_bprintf_cleanup(); | |
d9c9e4db FR |
437 | |
438 | return seq_has_overflowed(m) ? -EOVERFLOW : 0; | |
492e639f YS |
439 | } |
440 | ||
9436ef6e | 441 | BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file) |
c9a0f3b8 | 442 | |
492e639f YS |
443 | static const struct bpf_func_proto bpf_seq_printf_proto = { |
444 | .func = bpf_seq_printf, | |
445 | .gpl_only = true, | |
446 | .ret_type = RET_INTEGER, | |
447 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
9436ef6e | 448 | .arg1_btf_id = &btf_seq_file_ids[0], |
492e639f YS |
449 | .arg2_type = ARG_PTR_TO_MEM, |
450 | .arg3_type = ARG_CONST_SIZE, | |
451 | .arg4_type = ARG_PTR_TO_MEM_OR_NULL, | |
452 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, | |
492e639f YS |
453 | }; |
454 | ||
455 | BPF_CALL_3(bpf_seq_write, struct seq_file *, m, const void *, data, u32, len) | |
456 | { | |
457 | return seq_write(m, data, len) ? -EOVERFLOW : 0; | |
458 | } | |
459 | ||
492e639f YS |
460 | static const struct bpf_func_proto bpf_seq_write_proto = { |
461 | .func = bpf_seq_write, | |
462 | .gpl_only = true, | |
463 | .ret_type = RET_INTEGER, | |
464 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
9436ef6e | 465 | .arg1_btf_id = &btf_seq_file_ids[0], |
492e639f YS |
466 | .arg2_type = ARG_PTR_TO_MEM, |
467 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
492e639f YS |
468 | }; |
469 | ||
eb411377 AM |
470 | BPF_CALL_4(bpf_seq_printf_btf, struct seq_file *, m, struct btf_ptr *, ptr, |
471 | u32, btf_ptr_size, u64, flags) | |
472 | { | |
473 | const struct btf *btf; | |
474 | s32 btf_id; | |
475 | int ret; | |
476 | ||
477 | ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id); | |
478 | if (ret) | |
479 | return ret; | |
480 | ||
481 | return btf_type_seq_show_flags(btf, btf_id, ptr->ptr, m, flags); | |
482 | } | |
483 | ||
484 | static const struct bpf_func_proto bpf_seq_printf_btf_proto = { | |
485 | .func = bpf_seq_printf_btf, | |
486 | .gpl_only = true, | |
487 | .ret_type = RET_INTEGER, | |
488 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
489 | .arg1_btf_id = &btf_seq_file_ids[0], | |
492e639f YS |
490 | .arg2_type = ARG_PTR_TO_MEM, |
491 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
eb411377 | 492 | .arg4_type = ARG_ANYTHING, |
492e639f YS |
493 | }; |
494 | ||
908432ca YS |
495 | static __always_inline int |
496 | get_map_perf_counter(struct bpf_map *map, u64 flags, | |
497 | u64 *value, u64 *enabled, u64 *running) | |
35578d79 | 498 | { |
35578d79 | 499 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
6816a7ff DB |
500 | unsigned int cpu = smp_processor_id(); |
501 | u64 index = flags & BPF_F_INDEX_MASK; | |
3b1efb19 | 502 | struct bpf_event_entry *ee; |
35578d79 | 503 | |
6816a7ff DB |
504 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) |
505 | return -EINVAL; | |
506 | if (index == BPF_F_CURRENT_CPU) | |
507 | index = cpu; | |
35578d79 KX |
508 | if (unlikely(index >= array->map.max_entries)) |
509 | return -E2BIG; | |
510 | ||
3b1efb19 | 511 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 512 | if (!ee) |
35578d79 KX |
513 | return -ENOENT; |
514 | ||
908432ca YS |
515 | return perf_event_read_local(ee->event, value, enabled, running); |
516 | } | |
517 | ||
518 | BPF_CALL_2(bpf_perf_event_read, struct bpf_map *, map, u64, flags) | |
519 | { | |
520 | u64 value = 0; | |
521 | int err; | |
522 | ||
523 | err = get_map_perf_counter(map, flags, &value, NULL, NULL); | |
35578d79 | 524 | /* |
f91840a3 AS |
525 | * this api is ugly since we miss [-22..-2] range of valid |
526 | * counter values, but that's uapi | |
35578d79 | 527 | */ |
f91840a3 AS |
528 | if (err) |
529 | return err; | |
530 | return value; | |
35578d79 KX |
531 | } |
532 | ||
62544ce8 | 533 | static const struct bpf_func_proto bpf_perf_event_read_proto = { |
35578d79 | 534 | .func = bpf_perf_event_read, |
1075ef59 | 535 | .gpl_only = true, |
35578d79 KX |
536 | .ret_type = RET_INTEGER, |
537 | .arg1_type = ARG_CONST_MAP_PTR, | |
538 | .arg2_type = ARG_ANYTHING, | |
539 | }; | |
540 | ||
908432ca YS |
541 | BPF_CALL_4(bpf_perf_event_read_value, struct bpf_map *, map, u64, flags, |
542 | struct bpf_perf_event_value *, buf, u32, size) | |
543 | { | |
544 | int err = -EINVAL; | |
545 | ||
546 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
547 | goto clear; | |
548 | err = get_map_perf_counter(map, flags, &buf->counter, &buf->enabled, | |
549 | &buf->running); | |
550 | if (unlikely(err)) | |
551 | goto clear; | |
552 | return 0; | |
553 | clear: | |
554 | memset(buf, 0, size); | |
555 | return err; | |
556 | } | |
557 | ||
558 | static const struct bpf_func_proto bpf_perf_event_read_value_proto = { | |
559 | .func = bpf_perf_event_read_value, | |
560 | .gpl_only = true, | |
561 | .ret_type = RET_INTEGER, | |
562 | .arg1_type = ARG_CONST_MAP_PTR, | |
563 | .arg2_type = ARG_ANYTHING, | |
564 | .arg3_type = ARG_PTR_TO_UNINIT_MEM, | |
565 | .arg4_type = ARG_CONST_SIZE, | |
566 | }; | |
567 | ||
8e7a3920 DB |
568 | static __always_inline u64 |
569 | __bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map, | |
283ca526 | 570 | u64 flags, struct perf_sample_data *sd) |
a43eec30 | 571 | { |
a43eec30 | 572 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
d7931330 | 573 | unsigned int cpu = smp_processor_id(); |
1e33759c | 574 | u64 index = flags & BPF_F_INDEX_MASK; |
3b1efb19 | 575 | struct bpf_event_entry *ee; |
a43eec30 | 576 | struct perf_event *event; |
a43eec30 | 577 | |
1e33759c | 578 | if (index == BPF_F_CURRENT_CPU) |
d7931330 | 579 | index = cpu; |
a43eec30 AS |
580 | if (unlikely(index >= array->map.max_entries)) |
581 | return -E2BIG; | |
582 | ||
3b1efb19 | 583 | ee = READ_ONCE(array->ptrs[index]); |
1ca1cc98 | 584 | if (!ee) |
a43eec30 AS |
585 | return -ENOENT; |
586 | ||
3b1efb19 | 587 | event = ee->event; |
a43eec30 AS |
588 | if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE || |
589 | event->attr.config != PERF_COUNT_SW_BPF_OUTPUT)) | |
590 | return -EINVAL; | |
591 | ||
d7931330 | 592 | if (unlikely(event->oncpu != cpu)) |
a43eec30 AS |
593 | return -EOPNOTSUPP; |
594 | ||
56201969 | 595 | return perf_event_output(event, sd, regs); |
a43eec30 AS |
596 | } |
597 | ||
9594dc3c MM |
598 | /* |
599 | * Support executing tracepoints in normal, irq, and nmi context that each call | |
600 | * bpf_perf_event_output | |
601 | */ | |
602 | struct bpf_trace_sample_data { | |
603 | struct perf_sample_data sds[3]; | |
604 | }; | |
605 | ||
606 | static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds); | |
607 | static DEFINE_PER_CPU(int, bpf_trace_nest_level); | |
f3694e00 DB |
608 | BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map, |
609 | u64, flags, void *, data, u64, size) | |
8e7a3920 | 610 | { |
9594dc3c MM |
611 | struct bpf_trace_sample_data *sds = this_cpu_ptr(&bpf_trace_sds); |
612 | int nest_level = this_cpu_inc_return(bpf_trace_nest_level); | |
8e7a3920 DB |
613 | struct perf_raw_record raw = { |
614 | .frag = { | |
615 | .size = size, | |
616 | .data = data, | |
617 | }, | |
618 | }; | |
9594dc3c MM |
619 | struct perf_sample_data *sd; |
620 | int err; | |
8e7a3920 | 621 | |
9594dc3c MM |
622 | if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(sds->sds))) { |
623 | err = -EBUSY; | |
624 | goto out; | |
625 | } | |
626 | ||
627 | sd = &sds->sds[nest_level - 1]; | |
628 | ||
629 | if (unlikely(flags & ~(BPF_F_INDEX_MASK))) { | |
630 | err = -EINVAL; | |
631 | goto out; | |
632 | } | |
8e7a3920 | 633 | |
283ca526 DB |
634 | perf_sample_data_init(sd, 0, 0); |
635 | sd->raw = &raw; | |
636 | ||
9594dc3c MM |
637 | err = __bpf_perf_event_output(regs, map, flags, sd); |
638 | ||
639 | out: | |
640 | this_cpu_dec(bpf_trace_nest_level); | |
641 | return err; | |
8e7a3920 DB |
642 | } |
643 | ||
a43eec30 AS |
644 | static const struct bpf_func_proto bpf_perf_event_output_proto = { |
645 | .func = bpf_perf_event_output, | |
1075ef59 | 646 | .gpl_only = true, |
a43eec30 AS |
647 | .ret_type = RET_INTEGER, |
648 | .arg1_type = ARG_PTR_TO_CTX, | |
649 | .arg2_type = ARG_CONST_MAP_PTR, | |
650 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 651 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 652 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
a43eec30 AS |
653 | }; |
654 | ||
768fb61f AZ |
655 | static DEFINE_PER_CPU(int, bpf_event_output_nest_level); |
656 | struct bpf_nested_pt_regs { | |
657 | struct pt_regs regs[3]; | |
658 | }; | |
659 | static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs); | |
660 | static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds); | |
bd570ff9 | 661 | |
555c8a86 DB |
662 | u64 bpf_event_output(struct bpf_map *map, u64 flags, void *meta, u64 meta_size, |
663 | void *ctx, u64 ctx_size, bpf_ctx_copy_t ctx_copy) | |
bd570ff9 | 664 | { |
768fb61f | 665 | int nest_level = this_cpu_inc_return(bpf_event_output_nest_level); |
555c8a86 DB |
666 | struct perf_raw_frag frag = { |
667 | .copy = ctx_copy, | |
668 | .size = ctx_size, | |
669 | .data = ctx, | |
670 | }; | |
671 | struct perf_raw_record raw = { | |
672 | .frag = { | |
183fc153 AM |
673 | { |
674 | .next = ctx_size ? &frag : NULL, | |
675 | }, | |
555c8a86 DB |
676 | .size = meta_size, |
677 | .data = meta, | |
678 | }, | |
679 | }; | |
768fb61f AZ |
680 | struct perf_sample_data *sd; |
681 | struct pt_regs *regs; | |
682 | u64 ret; | |
683 | ||
684 | if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(bpf_misc_sds.sds))) { | |
685 | ret = -EBUSY; | |
686 | goto out; | |
687 | } | |
688 | sd = this_cpu_ptr(&bpf_misc_sds.sds[nest_level - 1]); | |
689 | regs = this_cpu_ptr(&bpf_pt_regs.regs[nest_level - 1]); | |
bd570ff9 DB |
690 | |
691 | perf_fetch_caller_regs(regs); | |
283ca526 DB |
692 | perf_sample_data_init(sd, 0, 0); |
693 | sd->raw = &raw; | |
bd570ff9 | 694 | |
768fb61f AZ |
695 | ret = __bpf_perf_event_output(regs, map, flags, sd); |
696 | out: | |
697 | this_cpu_dec(bpf_event_output_nest_level); | |
698 | return ret; | |
bd570ff9 DB |
699 | } |
700 | ||
f3694e00 | 701 | BPF_CALL_0(bpf_get_current_task) |
606274c5 AS |
702 | { |
703 | return (long) current; | |
704 | } | |
705 | ||
f470378c | 706 | const struct bpf_func_proto bpf_get_current_task_proto = { |
606274c5 AS |
707 | .func = bpf_get_current_task, |
708 | .gpl_only = true, | |
709 | .ret_type = RET_INTEGER, | |
710 | }; | |
711 | ||
3ca1032a KS |
712 | BPF_CALL_0(bpf_get_current_task_btf) |
713 | { | |
714 | return (unsigned long) current; | |
715 | } | |
716 | ||
a396eda5 | 717 | const struct bpf_func_proto bpf_get_current_task_btf_proto = { |
3ca1032a KS |
718 | .func = bpf_get_current_task_btf, |
719 | .gpl_only = true, | |
720 | .ret_type = RET_PTR_TO_BTF_ID, | |
33c5cb36 | 721 | .ret_btf_id = &btf_task_struct_ids[0], |
3ca1032a KS |
722 | }; |
723 | ||
dd6e10fb DX |
724 | BPF_CALL_1(bpf_task_pt_regs, struct task_struct *, task) |
725 | { | |
726 | return (unsigned long) task_pt_regs(task); | |
727 | } | |
728 | ||
729 | BTF_ID_LIST(bpf_task_pt_regs_ids) | |
730 | BTF_ID(struct, pt_regs) | |
731 | ||
732 | const struct bpf_func_proto bpf_task_pt_regs_proto = { | |
733 | .func = bpf_task_pt_regs, | |
734 | .gpl_only = true, | |
735 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
736 | .arg1_btf_id = &btf_task_struct_ids[0], | |
737 | .ret_type = RET_PTR_TO_BTF_ID, | |
738 | .ret_btf_id = &bpf_task_pt_regs_ids[0], | |
739 | }; | |
740 | ||
f3694e00 | 741 | BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx) |
60d20f91 | 742 | { |
60d20f91 SD |
743 | struct bpf_array *array = container_of(map, struct bpf_array, map); |
744 | struct cgroup *cgrp; | |
60d20f91 | 745 | |
60d20f91 SD |
746 | if (unlikely(idx >= array->map.max_entries)) |
747 | return -E2BIG; | |
748 | ||
749 | cgrp = READ_ONCE(array->ptrs[idx]); | |
750 | if (unlikely(!cgrp)) | |
751 | return -EAGAIN; | |
752 | ||
753 | return task_under_cgroup_hierarchy(current, cgrp); | |
754 | } | |
755 | ||
756 | static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = { | |
757 | .func = bpf_current_task_under_cgroup, | |
758 | .gpl_only = false, | |
759 | .ret_type = RET_INTEGER, | |
760 | .arg1_type = ARG_CONST_MAP_PTR, | |
761 | .arg2_type = ARG_ANYTHING, | |
762 | }; | |
763 | ||
8b401f9e YS |
764 | struct send_signal_irq_work { |
765 | struct irq_work irq_work; | |
766 | struct task_struct *task; | |
767 | u32 sig; | |
8482941f | 768 | enum pid_type type; |
8b401f9e YS |
769 | }; |
770 | ||
771 | static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work); | |
772 | ||
773 | static void do_bpf_send_signal(struct irq_work *entry) | |
774 | { | |
775 | struct send_signal_irq_work *work; | |
776 | ||
777 | work = container_of(entry, struct send_signal_irq_work, irq_work); | |
8482941f | 778 | group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type); |
8b401f9e YS |
779 | } |
780 | ||
8482941f | 781 | static int bpf_send_signal_common(u32 sig, enum pid_type type) |
8b401f9e YS |
782 | { |
783 | struct send_signal_irq_work *work = NULL; | |
784 | ||
785 | /* Similar to bpf_probe_write_user, task needs to be | |
786 | * in a sound condition and kernel memory access be | |
787 | * permitted in order to send signal to the current | |
788 | * task. | |
789 | */ | |
790 | if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING))) | |
791 | return -EPERM; | |
792 | if (unlikely(uaccess_kernel())) | |
793 | return -EPERM; | |
794 | if (unlikely(!nmi_uaccess_okay())) | |
795 | return -EPERM; | |
796 | ||
1bc7896e | 797 | if (irqs_disabled()) { |
e1afb702 YS |
798 | /* Do an early check on signal validity. Otherwise, |
799 | * the error is lost in deferred irq_work. | |
800 | */ | |
801 | if (unlikely(!valid_signal(sig))) | |
802 | return -EINVAL; | |
803 | ||
8b401f9e | 804 | work = this_cpu_ptr(&send_signal_work); |
7a9f50a0 | 805 | if (irq_work_is_busy(&work->irq_work)) |
8b401f9e YS |
806 | return -EBUSY; |
807 | ||
808 | /* Add the current task, which is the target of sending signal, | |
809 | * to the irq_work. The current task may change when queued | |
810 | * irq works get executed. | |
811 | */ | |
812 | work->task = current; | |
813 | work->sig = sig; | |
8482941f | 814 | work->type = type; |
8b401f9e YS |
815 | irq_work_queue(&work->irq_work); |
816 | return 0; | |
817 | } | |
818 | ||
8482941f YS |
819 | return group_send_sig_info(sig, SEND_SIG_PRIV, current, type); |
820 | } | |
821 | ||
822 | BPF_CALL_1(bpf_send_signal, u32, sig) | |
823 | { | |
824 | return bpf_send_signal_common(sig, PIDTYPE_TGID); | |
8b401f9e YS |
825 | } |
826 | ||
827 | static const struct bpf_func_proto bpf_send_signal_proto = { | |
828 | .func = bpf_send_signal, | |
829 | .gpl_only = false, | |
830 | .ret_type = RET_INTEGER, | |
831 | .arg1_type = ARG_ANYTHING, | |
832 | }; | |
833 | ||
8482941f YS |
834 | BPF_CALL_1(bpf_send_signal_thread, u32, sig) |
835 | { | |
836 | return bpf_send_signal_common(sig, PIDTYPE_PID); | |
837 | } | |
838 | ||
839 | static const struct bpf_func_proto bpf_send_signal_thread_proto = { | |
840 | .func = bpf_send_signal_thread, | |
841 | .gpl_only = false, | |
842 | .ret_type = RET_INTEGER, | |
843 | .arg1_type = ARG_ANYTHING, | |
844 | }; | |
845 | ||
6e22ab9d JO |
846 | BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz) |
847 | { | |
848 | long len; | |
849 | char *p; | |
850 | ||
851 | if (!sz) | |
852 | return 0; | |
853 | ||
854 | p = d_path(path, buf, sz); | |
855 | if (IS_ERR(p)) { | |
856 | len = PTR_ERR(p); | |
857 | } else { | |
858 | len = buf + sz - p; | |
859 | memmove(buf, p, len); | |
860 | } | |
861 | ||
862 | return len; | |
863 | } | |
864 | ||
865 | BTF_SET_START(btf_allowlist_d_path) | |
a8a71796 JO |
866 | #ifdef CONFIG_SECURITY |
867 | BTF_ID(func, security_file_permission) | |
868 | BTF_ID(func, security_inode_getattr) | |
869 | BTF_ID(func, security_file_open) | |
870 | #endif | |
871 | #ifdef CONFIG_SECURITY_PATH | |
872 | BTF_ID(func, security_path_truncate) | |
873 | #endif | |
6e22ab9d JO |
874 | BTF_ID(func, vfs_truncate) |
875 | BTF_ID(func, vfs_fallocate) | |
876 | BTF_ID(func, dentry_open) | |
877 | BTF_ID(func, vfs_getattr) | |
878 | BTF_ID(func, filp_close) | |
879 | BTF_SET_END(btf_allowlist_d_path) | |
880 | ||
881 | static bool bpf_d_path_allowed(const struct bpf_prog *prog) | |
882 | { | |
3d06f34a SL |
883 | if (prog->type == BPF_PROG_TYPE_TRACING && |
884 | prog->expected_attach_type == BPF_TRACE_ITER) | |
885 | return true; | |
886 | ||
6f100640 KS |
887 | if (prog->type == BPF_PROG_TYPE_LSM) |
888 | return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id); | |
889 | ||
890 | return btf_id_set_contains(&btf_allowlist_d_path, | |
891 | prog->aux->attach_btf_id); | |
6e22ab9d JO |
892 | } |
893 | ||
9436ef6e | 894 | BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path) |
6e22ab9d JO |
895 | |
896 | static const struct bpf_func_proto bpf_d_path_proto = { | |
897 | .func = bpf_d_path, | |
898 | .gpl_only = false, | |
899 | .ret_type = RET_INTEGER, | |
900 | .arg1_type = ARG_PTR_TO_BTF_ID, | |
9436ef6e | 901 | .arg1_btf_id = &bpf_d_path_btf_ids[0], |
6e22ab9d JO |
902 | .arg2_type = ARG_PTR_TO_MEM, |
903 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
6e22ab9d JO |
904 | .allowed = bpf_d_path_allowed, |
905 | }; | |
906 | ||
c4d0bfb4 AM |
907 | #define BTF_F_ALL (BTF_F_COMPACT | BTF_F_NONAME | \ |
908 | BTF_F_PTR_RAW | BTF_F_ZERO) | |
909 | ||
910 | static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size, | |
911 | u64 flags, const struct btf **btf, | |
912 | s32 *btf_id) | |
913 | { | |
914 | const struct btf_type *t; | |
915 | ||
916 | if (unlikely(flags & ~(BTF_F_ALL))) | |
917 | return -EINVAL; | |
918 | ||
919 | if (btf_ptr_size != sizeof(struct btf_ptr)) | |
920 | return -EINVAL; | |
921 | ||
922 | *btf = bpf_get_btf_vmlinux(); | |
923 | ||
924 | if (IS_ERR_OR_NULL(*btf)) | |
abbaa433 | 925 | return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL; |
c4d0bfb4 AM |
926 | |
927 | if (ptr->type_id > 0) | |
928 | *btf_id = ptr->type_id; | |
929 | else | |
930 | return -EINVAL; | |
931 | ||
932 | if (*btf_id > 0) | |
933 | t = btf_type_by_id(*btf, *btf_id); | |
934 | if (*btf_id <= 0 || !t) | |
935 | return -ENOENT; | |
936 | ||
937 | return 0; | |
938 | } | |
939 | ||
940 | BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr, | |
941 | u32, btf_ptr_size, u64, flags) | |
942 | { | |
943 | const struct btf *btf; | |
944 | s32 btf_id; | |
945 | int ret; | |
946 | ||
947 | ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id); | |
948 | if (ret) | |
949 | return ret; | |
950 | ||
951 | return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size, | |
952 | flags); | |
953 | } | |
954 | ||
955 | const struct bpf_func_proto bpf_snprintf_btf_proto = { | |
956 | .func = bpf_snprintf_btf, | |
957 | .gpl_only = false, | |
958 | .ret_type = RET_INTEGER, | |
959 | .arg1_type = ARG_PTR_TO_MEM, | |
960 | .arg2_type = ARG_CONST_SIZE, | |
961 | .arg3_type = ARG_PTR_TO_MEM, | |
962 | .arg4_type = ARG_CONST_SIZE, | |
963 | .arg5_type = ARG_ANYTHING, | |
964 | }; | |
965 | ||
9b99edca JO |
966 | BPF_CALL_1(bpf_get_func_ip_tracing, void *, ctx) |
967 | { | |
968 | /* This helper call is inlined by verifier. */ | |
969 | return ((u64 *)ctx)[-1]; | |
970 | } | |
971 | ||
972 | static const struct bpf_func_proto bpf_get_func_ip_proto_tracing = { | |
973 | .func = bpf_get_func_ip_tracing, | |
974 | .gpl_only = true, | |
975 | .ret_type = RET_INTEGER, | |
976 | .arg1_type = ARG_PTR_TO_CTX, | |
977 | }; | |
978 | ||
9ffd9f3f JO |
979 | BPF_CALL_1(bpf_get_func_ip_kprobe, struct pt_regs *, regs) |
980 | { | |
981 | struct kprobe *kp = kprobe_running(); | |
982 | ||
16c5900b | 983 | return kp ? (uintptr_t)kp->addr : 0; |
9ffd9f3f JO |
984 | } |
985 | ||
986 | static const struct bpf_func_proto bpf_get_func_ip_proto_kprobe = { | |
987 | .func = bpf_get_func_ip_kprobe, | |
988 | .gpl_only = true, | |
989 | .ret_type = RET_INTEGER, | |
990 | .arg1_type = ARG_PTR_TO_CTX, | |
991 | }; | |
992 | ||
7adfc6c9 AN |
993 | BPF_CALL_1(bpf_get_attach_cookie_trace, void *, ctx) |
994 | { | |
995 | struct bpf_trace_run_ctx *run_ctx; | |
996 | ||
997 | run_ctx = container_of(current->bpf_ctx, struct bpf_trace_run_ctx, run_ctx); | |
998 | return run_ctx->bpf_cookie; | |
999 | } | |
1000 | ||
1001 | static const struct bpf_func_proto bpf_get_attach_cookie_proto_trace = { | |
1002 | .func = bpf_get_attach_cookie_trace, | |
1003 | .gpl_only = false, | |
1004 | .ret_type = RET_INTEGER, | |
1005 | .arg1_type = ARG_PTR_TO_CTX, | |
1006 | }; | |
1007 | ||
1008 | BPF_CALL_1(bpf_get_attach_cookie_pe, struct bpf_perf_event_data_kern *, ctx) | |
1009 | { | |
1010 | return ctx->event->bpf_cookie; | |
1011 | } | |
1012 | ||
1013 | static const struct bpf_func_proto bpf_get_attach_cookie_proto_pe = { | |
1014 | .func = bpf_get_attach_cookie_pe, | |
1015 | .gpl_only = false, | |
1016 | .ret_type = RET_INTEGER, | |
1017 | .arg1_type = ARG_PTR_TO_CTX, | |
1018 | }; | |
1019 | ||
1020 | static const struct bpf_func_proto * | |
fc611f47 | 1021 | bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) |
2541517c AS |
1022 | { |
1023 | switch (func_id) { | |
1024 | case BPF_FUNC_map_lookup_elem: | |
1025 | return &bpf_map_lookup_elem_proto; | |
1026 | case BPF_FUNC_map_update_elem: | |
1027 | return &bpf_map_update_elem_proto; | |
1028 | case BPF_FUNC_map_delete_elem: | |
1029 | return &bpf_map_delete_elem_proto; | |
02a8c817 AC |
1030 | case BPF_FUNC_map_push_elem: |
1031 | return &bpf_map_push_elem_proto; | |
1032 | case BPF_FUNC_map_pop_elem: | |
1033 | return &bpf_map_pop_elem_proto; | |
1034 | case BPF_FUNC_map_peek_elem: | |
1035 | return &bpf_map_peek_elem_proto; | |
d9847d31 AS |
1036 | case BPF_FUNC_ktime_get_ns: |
1037 | return &bpf_ktime_get_ns_proto; | |
71d19214 MÅ» |
1038 | case BPF_FUNC_ktime_get_boot_ns: |
1039 | return &bpf_ktime_get_boot_ns_proto; | |
d0551261 DB |
1040 | case BPF_FUNC_ktime_get_coarse_ns: |
1041 | return &bpf_ktime_get_coarse_ns_proto; | |
04fd61ab AS |
1042 | case BPF_FUNC_tail_call: |
1043 | return &bpf_tail_call_proto; | |
ffeedafb AS |
1044 | case BPF_FUNC_get_current_pid_tgid: |
1045 | return &bpf_get_current_pid_tgid_proto; | |
606274c5 AS |
1046 | case BPF_FUNC_get_current_task: |
1047 | return &bpf_get_current_task_proto; | |
3ca1032a KS |
1048 | case BPF_FUNC_get_current_task_btf: |
1049 | return &bpf_get_current_task_btf_proto; | |
dd6e10fb DX |
1050 | case BPF_FUNC_task_pt_regs: |
1051 | return &bpf_task_pt_regs_proto; | |
ffeedafb AS |
1052 | case BPF_FUNC_get_current_uid_gid: |
1053 | return &bpf_get_current_uid_gid_proto; | |
1054 | case BPF_FUNC_get_current_comm: | |
1055 | return &bpf_get_current_comm_proto; | |
9c959c86 | 1056 | case BPF_FUNC_trace_printk: |
0756ea3e | 1057 | return bpf_get_trace_printk_proto(); |
ab1973d3 AS |
1058 | case BPF_FUNC_get_smp_processor_id: |
1059 | return &bpf_get_smp_processor_id_proto; | |
2d0e30c3 DB |
1060 | case BPF_FUNC_get_numa_node_id: |
1061 | return &bpf_get_numa_node_id_proto; | |
35578d79 KX |
1062 | case BPF_FUNC_perf_event_read: |
1063 | return &bpf_perf_event_read_proto; | |
60d20f91 SD |
1064 | case BPF_FUNC_current_task_under_cgroup: |
1065 | return &bpf_current_task_under_cgroup_proto; | |
8937bd80 AS |
1066 | case BPF_FUNC_get_prandom_u32: |
1067 | return &bpf_get_prandom_u32_proto; | |
51e1bb9e DB |
1068 | case BPF_FUNC_probe_write_user: |
1069 | return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ? | |
1070 | NULL : bpf_get_probe_write_proto(); | |
6ae08ae3 DB |
1071 | case BPF_FUNC_probe_read_user: |
1072 | return &bpf_probe_read_user_proto; | |
1073 | case BPF_FUNC_probe_read_kernel: | |
71330842 | 1074 | return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? |
ff40e510 | 1075 | NULL : &bpf_probe_read_kernel_proto; |
6ae08ae3 DB |
1076 | case BPF_FUNC_probe_read_user_str: |
1077 | return &bpf_probe_read_user_str_proto; | |
1078 | case BPF_FUNC_probe_read_kernel_str: | |
71330842 | 1079 | return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? |
ff40e510 | 1080 | NULL : &bpf_probe_read_kernel_str_proto; |
0ebeea8c DB |
1081 | #ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE |
1082 | case BPF_FUNC_probe_read: | |
71330842 | 1083 | return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? |
ff40e510 | 1084 | NULL : &bpf_probe_read_compat_proto; |
a5e8c070 | 1085 | case BPF_FUNC_probe_read_str: |
71330842 | 1086 | return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ? |
ff40e510 | 1087 | NULL : &bpf_probe_read_compat_str_proto; |
0ebeea8c | 1088 | #endif |
34ea38ca | 1089 | #ifdef CONFIG_CGROUPS |
bf6fa2c8 YS |
1090 | case BPF_FUNC_get_current_cgroup_id: |
1091 | return &bpf_get_current_cgroup_id_proto; | |
95b861a7 NK |
1092 | case BPF_FUNC_get_current_ancestor_cgroup_id: |
1093 | return &bpf_get_current_ancestor_cgroup_id_proto; | |
34ea38ca | 1094 | #endif |
8b401f9e YS |
1095 | case BPF_FUNC_send_signal: |
1096 | return &bpf_send_signal_proto; | |
8482941f YS |
1097 | case BPF_FUNC_send_signal_thread: |
1098 | return &bpf_send_signal_thread_proto; | |
b80b033b SL |
1099 | case BPF_FUNC_perf_event_read_value: |
1100 | return &bpf_perf_event_read_value_proto; | |
b4490c5c CN |
1101 | case BPF_FUNC_get_ns_current_pid_tgid: |
1102 | return &bpf_get_ns_current_pid_tgid_proto; | |
457f4436 AN |
1103 | case BPF_FUNC_ringbuf_output: |
1104 | return &bpf_ringbuf_output_proto; | |
1105 | case BPF_FUNC_ringbuf_reserve: | |
1106 | return &bpf_ringbuf_reserve_proto; | |
1107 | case BPF_FUNC_ringbuf_submit: | |
1108 | return &bpf_ringbuf_submit_proto; | |
1109 | case BPF_FUNC_ringbuf_discard: | |
1110 | return &bpf_ringbuf_discard_proto; | |
1111 | case BPF_FUNC_ringbuf_query: | |
1112 | return &bpf_ringbuf_query_proto; | |
72e2b2b6 YS |
1113 | case BPF_FUNC_jiffies64: |
1114 | return &bpf_jiffies64_proto; | |
fa28dcb8 SL |
1115 | case BPF_FUNC_get_task_stack: |
1116 | return &bpf_get_task_stack_proto; | |
07be4c4a AS |
1117 | case BPF_FUNC_copy_from_user: |
1118 | return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL; | |
c4d0bfb4 AM |
1119 | case BPF_FUNC_snprintf_btf: |
1120 | return &bpf_snprintf_btf_proto; | |
b7906b70 | 1121 | case BPF_FUNC_per_cpu_ptr: |
eaa6bcb7 | 1122 | return &bpf_per_cpu_ptr_proto; |
b7906b70 | 1123 | case BPF_FUNC_this_cpu_ptr: |
63d9b80d | 1124 | return &bpf_this_cpu_ptr_proto; |
a10787e6 SL |
1125 | case BPF_FUNC_task_storage_get: |
1126 | return &bpf_task_storage_get_proto; | |
1127 | case BPF_FUNC_task_storage_delete: | |
1128 | return &bpf_task_storage_delete_proto; | |
69c087ba YS |
1129 | case BPF_FUNC_for_each_map_elem: |
1130 | return &bpf_for_each_map_elem_proto; | |
7b15523a FR |
1131 | case BPF_FUNC_snprintf: |
1132 | return &bpf_snprintf_proto; | |
9b99edca JO |
1133 | case BPF_FUNC_get_func_ip: |
1134 | return &bpf_get_func_ip_proto_tracing; | |
9fd82b61 | 1135 | default: |
b00628b1 | 1136 | return bpf_base_func_proto(func_id); |
9fd82b61 AS |
1137 | } |
1138 | } | |
1139 | ||
5e43f899 AI |
1140 | static const struct bpf_func_proto * |
1141 | kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
9fd82b61 AS |
1142 | { |
1143 | switch (func_id) { | |
a43eec30 AS |
1144 | case BPF_FUNC_perf_event_output: |
1145 | return &bpf_perf_event_output_proto; | |
d5a3b1f6 AS |
1146 | case BPF_FUNC_get_stackid: |
1147 | return &bpf_get_stackid_proto; | |
c195651e YS |
1148 | case BPF_FUNC_get_stack: |
1149 | return &bpf_get_stack_proto; | |
9802d865 JB |
1150 | #ifdef CONFIG_BPF_KPROBE_OVERRIDE |
1151 | case BPF_FUNC_override_return: | |
1152 | return &bpf_override_return_proto; | |
1153 | #endif | |
9ffd9f3f JO |
1154 | case BPF_FUNC_get_func_ip: |
1155 | return &bpf_get_func_ip_proto_kprobe; | |
7adfc6c9 AN |
1156 | case BPF_FUNC_get_attach_cookie: |
1157 | return &bpf_get_attach_cookie_proto_trace; | |
2541517c | 1158 | default: |
fc611f47 | 1159 | return bpf_tracing_func_proto(func_id, prog); |
2541517c AS |
1160 | } |
1161 | } | |
1162 | ||
1163 | /* bpf+kprobe programs can access fields of 'struct pt_regs' */ | |
19de99f7 | 1164 | static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
5e43f899 | 1165 | const struct bpf_prog *prog, |
23994631 | 1166 | struct bpf_insn_access_aux *info) |
2541517c | 1167 | { |
2541517c AS |
1168 | if (off < 0 || off >= sizeof(struct pt_regs)) |
1169 | return false; | |
2541517c AS |
1170 | if (type != BPF_READ) |
1171 | return false; | |
2541517c AS |
1172 | if (off % size != 0) |
1173 | return false; | |
2d071c64 DB |
1174 | /* |
1175 | * Assertion for 32 bit to make sure last 8 byte access | |
1176 | * (BPF_DW) to the last 4 byte member is disallowed. | |
1177 | */ | |
1178 | if (off + size > sizeof(struct pt_regs)) | |
1179 | return false; | |
1180 | ||
2541517c AS |
1181 | return true; |
1182 | } | |
1183 | ||
7de16e3a | 1184 | const struct bpf_verifier_ops kprobe_verifier_ops = { |
2541517c AS |
1185 | .get_func_proto = kprobe_prog_func_proto, |
1186 | .is_valid_access = kprobe_prog_is_valid_access, | |
1187 | }; | |
1188 | ||
7de16e3a JK |
1189 | const struct bpf_prog_ops kprobe_prog_ops = { |
1190 | }; | |
1191 | ||
f3694e00 DB |
1192 | BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map, |
1193 | u64, flags, void *, data, u64, size) | |
9940d67c | 1194 | { |
f3694e00 DB |
1195 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
1196 | ||
9940d67c AS |
1197 | /* |
1198 | * r1 points to perf tracepoint buffer where first 8 bytes are hidden | |
1199 | * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it | |
f3694e00 | 1200 | * from there and call the same bpf_perf_event_output() helper inline. |
9940d67c | 1201 | */ |
f3694e00 | 1202 | return ____bpf_perf_event_output(regs, map, flags, data, size); |
9940d67c AS |
1203 | } |
1204 | ||
1205 | static const struct bpf_func_proto bpf_perf_event_output_proto_tp = { | |
1206 | .func = bpf_perf_event_output_tp, | |
1207 | .gpl_only = true, | |
1208 | .ret_type = RET_INTEGER, | |
1209 | .arg1_type = ARG_PTR_TO_CTX, | |
1210 | .arg2_type = ARG_CONST_MAP_PTR, | |
1211 | .arg3_type = ARG_ANYTHING, | |
39f19ebb | 1212 | .arg4_type = ARG_PTR_TO_MEM, |
a60dd35d | 1213 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, |
9940d67c AS |
1214 | }; |
1215 | ||
f3694e00 DB |
1216 | BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map, |
1217 | u64, flags) | |
9940d67c | 1218 | { |
f3694e00 | 1219 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; |
9940d67c | 1220 | |
f3694e00 DB |
1221 | /* |
1222 | * Same comment as in bpf_perf_event_output_tp(), only that this time | |
1223 | * the other helper's function body cannot be inlined due to being | |
1224 | * external, thus we need to call raw helper function. | |
1225 | */ | |
1226 | return bpf_get_stackid((unsigned long) regs, (unsigned long) map, | |
1227 | flags, 0, 0); | |
9940d67c AS |
1228 | } |
1229 | ||
1230 | static const struct bpf_func_proto bpf_get_stackid_proto_tp = { | |
1231 | .func = bpf_get_stackid_tp, | |
1232 | .gpl_only = true, | |
1233 | .ret_type = RET_INTEGER, | |
1234 | .arg1_type = ARG_PTR_TO_CTX, | |
1235 | .arg2_type = ARG_CONST_MAP_PTR, | |
1236 | .arg3_type = ARG_ANYTHING, | |
1237 | }; | |
1238 | ||
c195651e YS |
1239 | BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size, |
1240 | u64, flags) | |
1241 | { | |
1242 | struct pt_regs *regs = *(struct pt_regs **)tp_buff; | |
1243 | ||
1244 | return bpf_get_stack((unsigned long) regs, (unsigned long) buf, | |
1245 | (unsigned long) size, flags, 0); | |
1246 | } | |
1247 | ||
1248 | static const struct bpf_func_proto bpf_get_stack_proto_tp = { | |
1249 | .func = bpf_get_stack_tp, | |
1250 | .gpl_only = true, | |
1251 | .ret_type = RET_INTEGER, | |
1252 | .arg1_type = ARG_PTR_TO_CTX, | |
1253 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
1254 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
1255 | .arg4_type = ARG_ANYTHING, | |
1256 | }; | |
1257 | ||
5e43f899 AI |
1258 | static const struct bpf_func_proto * |
1259 | tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
f005afed YS |
1260 | { |
1261 | switch (func_id) { | |
1262 | case BPF_FUNC_perf_event_output: | |
1263 | return &bpf_perf_event_output_proto_tp; | |
1264 | case BPF_FUNC_get_stackid: | |
1265 | return &bpf_get_stackid_proto_tp; | |
c195651e YS |
1266 | case BPF_FUNC_get_stack: |
1267 | return &bpf_get_stack_proto_tp; | |
7adfc6c9 AN |
1268 | case BPF_FUNC_get_attach_cookie: |
1269 | return &bpf_get_attach_cookie_proto_trace; | |
f005afed | 1270 | default: |
fc611f47 | 1271 | return bpf_tracing_func_proto(func_id, prog); |
f005afed YS |
1272 | } |
1273 | } | |
1274 | ||
1275 | static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type, | |
5e43f899 | 1276 | const struct bpf_prog *prog, |
f005afed YS |
1277 | struct bpf_insn_access_aux *info) |
1278 | { | |
1279 | if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE) | |
1280 | return false; | |
1281 | if (type != BPF_READ) | |
1282 | return false; | |
1283 | if (off % size != 0) | |
1284 | return false; | |
1285 | ||
1286 | BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64)); | |
1287 | return true; | |
1288 | } | |
1289 | ||
1290 | const struct bpf_verifier_ops tracepoint_verifier_ops = { | |
1291 | .get_func_proto = tp_prog_func_proto, | |
1292 | .is_valid_access = tp_prog_is_valid_access, | |
1293 | }; | |
1294 | ||
1295 | const struct bpf_prog_ops tracepoint_prog_ops = { | |
1296 | }; | |
1297 | ||
1298 | BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx, | |
4bebdc7a YS |
1299 | struct bpf_perf_event_value *, buf, u32, size) |
1300 | { | |
1301 | int err = -EINVAL; | |
1302 | ||
1303 | if (unlikely(size != sizeof(struct bpf_perf_event_value))) | |
1304 | goto clear; | |
1305 | err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled, | |
1306 | &buf->running); | |
1307 | if (unlikely(err)) | |
1308 | goto clear; | |
1309 | return 0; | |
1310 | clear: | |
1311 | memset(buf, 0, size); | |
1312 | return err; | |
1313 | } | |
1314 | ||
f005afed YS |
1315 | static const struct bpf_func_proto bpf_perf_prog_read_value_proto = { |
1316 | .func = bpf_perf_prog_read_value, | |
4bebdc7a YS |
1317 | .gpl_only = true, |
1318 | .ret_type = RET_INTEGER, | |
1319 | .arg1_type = ARG_PTR_TO_CTX, | |
1320 | .arg2_type = ARG_PTR_TO_UNINIT_MEM, | |
1321 | .arg3_type = ARG_CONST_SIZE, | |
1322 | }; | |
1323 | ||
fff7b643 DX |
1324 | BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx, |
1325 | void *, buf, u32, size, u64, flags) | |
1326 | { | |
1327 | #ifndef CONFIG_X86 | |
1328 | return -ENOENT; | |
1329 | #else | |
1330 | static const u32 br_entry_size = sizeof(struct perf_branch_entry); | |
1331 | struct perf_branch_stack *br_stack = ctx->data->br_stack; | |
1332 | u32 to_copy; | |
1333 | ||
1334 | if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE)) | |
1335 | return -EINVAL; | |
1336 | ||
1337 | if (unlikely(!br_stack)) | |
1338 | return -EINVAL; | |
1339 | ||
1340 | if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE) | |
1341 | return br_stack->nr * br_entry_size; | |
1342 | ||
1343 | if (!buf || (size % br_entry_size != 0)) | |
1344 | return -EINVAL; | |
1345 | ||
1346 | to_copy = min_t(u32, br_stack->nr * br_entry_size, size); | |
1347 | memcpy(buf, br_stack->entries, to_copy); | |
1348 | ||
1349 | return to_copy; | |
1350 | #endif | |
1351 | } | |
1352 | ||
1353 | static const struct bpf_func_proto bpf_read_branch_records_proto = { | |
1354 | .func = bpf_read_branch_records, | |
1355 | .gpl_only = true, | |
1356 | .ret_type = RET_INTEGER, | |
1357 | .arg1_type = ARG_PTR_TO_CTX, | |
1358 | .arg2_type = ARG_PTR_TO_MEM_OR_NULL, | |
1359 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
1360 | .arg4_type = ARG_ANYTHING, | |
1361 | }; | |
1362 | ||
5e43f899 AI |
1363 | static const struct bpf_func_proto * |
1364 | pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
9fd82b61 AS |
1365 | { |
1366 | switch (func_id) { | |
1367 | case BPF_FUNC_perf_event_output: | |
9940d67c | 1368 | return &bpf_perf_event_output_proto_tp; |
9fd82b61 | 1369 | case BPF_FUNC_get_stackid: |
7b04d6d6 | 1370 | return &bpf_get_stackid_proto_pe; |
c195651e | 1371 | case BPF_FUNC_get_stack: |
7b04d6d6 | 1372 | return &bpf_get_stack_proto_pe; |
4bebdc7a | 1373 | case BPF_FUNC_perf_prog_read_value: |
f005afed | 1374 | return &bpf_perf_prog_read_value_proto; |
fff7b643 DX |
1375 | case BPF_FUNC_read_branch_records: |
1376 | return &bpf_read_branch_records_proto; | |
7adfc6c9 AN |
1377 | case BPF_FUNC_get_attach_cookie: |
1378 | return &bpf_get_attach_cookie_proto_pe; | |
9fd82b61 | 1379 | default: |
fc611f47 | 1380 | return bpf_tracing_func_proto(func_id, prog); |
9fd82b61 AS |
1381 | } |
1382 | } | |
1383 | ||
c4f6699d AS |
1384 | /* |
1385 | * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp | |
1386 | * to avoid potential recursive reuse issue when/if tracepoints are added | |
9594dc3c MM |
1387 | * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack. |
1388 | * | |
1389 | * Since raw tracepoints run despite bpf_prog_active, support concurrent usage | |
1390 | * in normal, irq, and nmi context. | |
c4f6699d | 1391 | */ |
9594dc3c MM |
1392 | struct bpf_raw_tp_regs { |
1393 | struct pt_regs regs[3]; | |
1394 | }; | |
1395 | static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs); | |
1396 | static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level); | |
1397 | static struct pt_regs *get_bpf_raw_tp_regs(void) | |
1398 | { | |
1399 | struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs); | |
1400 | int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level); | |
1401 | ||
1402 | if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) { | |
1403 | this_cpu_dec(bpf_raw_tp_nest_level); | |
1404 | return ERR_PTR(-EBUSY); | |
1405 | } | |
1406 | ||
1407 | return &tp_regs->regs[nest_level - 1]; | |
1408 | } | |
1409 | ||
1410 | static void put_bpf_raw_tp_regs(void) | |
1411 | { | |
1412 | this_cpu_dec(bpf_raw_tp_nest_level); | |
1413 | } | |
1414 | ||
c4f6699d AS |
1415 | BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args, |
1416 | struct bpf_map *, map, u64, flags, void *, data, u64, size) | |
1417 | { | |
9594dc3c MM |
1418 | struct pt_regs *regs = get_bpf_raw_tp_regs(); |
1419 | int ret; | |
1420 | ||
1421 | if (IS_ERR(regs)) | |
1422 | return PTR_ERR(regs); | |
c4f6699d AS |
1423 | |
1424 | perf_fetch_caller_regs(regs); | |
9594dc3c MM |
1425 | ret = ____bpf_perf_event_output(regs, map, flags, data, size); |
1426 | ||
1427 | put_bpf_raw_tp_regs(); | |
1428 | return ret; | |
c4f6699d AS |
1429 | } |
1430 | ||
1431 | static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = { | |
1432 | .func = bpf_perf_event_output_raw_tp, | |
1433 | .gpl_only = true, | |
1434 | .ret_type = RET_INTEGER, | |
1435 | .arg1_type = ARG_PTR_TO_CTX, | |
1436 | .arg2_type = ARG_CONST_MAP_PTR, | |
1437 | .arg3_type = ARG_ANYTHING, | |
1438 | .arg4_type = ARG_PTR_TO_MEM, | |
1439 | .arg5_type = ARG_CONST_SIZE_OR_ZERO, | |
1440 | }; | |
1441 | ||
a7658e1a | 1442 | extern const struct bpf_func_proto bpf_skb_output_proto; |
d831ee84 | 1443 | extern const struct bpf_func_proto bpf_xdp_output_proto; |
a7658e1a | 1444 | |
c4f6699d AS |
1445 | BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args, |
1446 | struct bpf_map *, map, u64, flags) | |
1447 | { | |
9594dc3c MM |
1448 | struct pt_regs *regs = get_bpf_raw_tp_regs(); |
1449 | int ret; | |
1450 | ||
1451 | if (IS_ERR(regs)) | |
1452 | return PTR_ERR(regs); | |
c4f6699d AS |
1453 | |
1454 | perf_fetch_caller_regs(regs); | |
1455 | /* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */ | |
9594dc3c MM |
1456 | ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map, |
1457 | flags, 0, 0); | |
1458 | put_bpf_raw_tp_regs(); | |
1459 | return ret; | |
c4f6699d AS |
1460 | } |
1461 | ||
1462 | static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = { | |
1463 | .func = bpf_get_stackid_raw_tp, | |
1464 | .gpl_only = true, | |
1465 | .ret_type = RET_INTEGER, | |
1466 | .arg1_type = ARG_PTR_TO_CTX, | |
1467 | .arg2_type = ARG_CONST_MAP_PTR, | |
1468 | .arg3_type = ARG_ANYTHING, | |
1469 | }; | |
1470 | ||
c195651e YS |
1471 | BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args, |
1472 | void *, buf, u32, size, u64, flags) | |
1473 | { | |
9594dc3c MM |
1474 | struct pt_regs *regs = get_bpf_raw_tp_regs(); |
1475 | int ret; | |
1476 | ||
1477 | if (IS_ERR(regs)) | |
1478 | return PTR_ERR(regs); | |
c195651e YS |
1479 | |
1480 | perf_fetch_caller_regs(regs); | |
9594dc3c MM |
1481 | ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf, |
1482 | (unsigned long) size, flags, 0); | |
1483 | put_bpf_raw_tp_regs(); | |
1484 | return ret; | |
c195651e YS |
1485 | } |
1486 | ||
1487 | static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = { | |
1488 | .func = bpf_get_stack_raw_tp, | |
1489 | .gpl_only = true, | |
1490 | .ret_type = RET_INTEGER, | |
1491 | .arg1_type = ARG_PTR_TO_CTX, | |
1492 | .arg2_type = ARG_PTR_TO_MEM, | |
1493 | .arg3_type = ARG_CONST_SIZE_OR_ZERO, | |
1494 | .arg4_type = ARG_ANYTHING, | |
1495 | }; | |
1496 | ||
5e43f899 AI |
1497 | static const struct bpf_func_proto * |
1498 | raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) | |
c4f6699d AS |
1499 | { |
1500 | switch (func_id) { | |
1501 | case BPF_FUNC_perf_event_output: | |
1502 | return &bpf_perf_event_output_proto_raw_tp; | |
1503 | case BPF_FUNC_get_stackid: | |
1504 | return &bpf_get_stackid_proto_raw_tp; | |
c195651e YS |
1505 | case BPF_FUNC_get_stack: |
1506 | return &bpf_get_stack_proto_raw_tp; | |
c4f6699d | 1507 | default: |
fc611f47 | 1508 | return bpf_tracing_func_proto(func_id, prog); |
c4f6699d AS |
1509 | } |
1510 | } | |
1511 | ||
958a3f2d | 1512 | const struct bpf_func_proto * |
f1b9509c AS |
1513 | tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog) |
1514 | { | |
3cee6fb8 MKL |
1515 | const struct bpf_func_proto *fn; |
1516 | ||
f1b9509c AS |
1517 | switch (func_id) { |
1518 | #ifdef CONFIG_NET | |
1519 | case BPF_FUNC_skb_output: | |
1520 | return &bpf_skb_output_proto; | |
d831ee84 EC |
1521 | case BPF_FUNC_xdp_output: |
1522 | return &bpf_xdp_output_proto; | |
af7ec138 YS |
1523 | case BPF_FUNC_skc_to_tcp6_sock: |
1524 | return &bpf_skc_to_tcp6_sock_proto; | |
478cfbdf YS |
1525 | case BPF_FUNC_skc_to_tcp_sock: |
1526 | return &bpf_skc_to_tcp_sock_proto; | |
1527 | case BPF_FUNC_skc_to_tcp_timewait_sock: | |
1528 | return &bpf_skc_to_tcp_timewait_sock_proto; | |
1529 | case BPF_FUNC_skc_to_tcp_request_sock: | |
1530 | return &bpf_skc_to_tcp_request_sock_proto; | |
0d4fad3e YS |
1531 | case BPF_FUNC_skc_to_udp6_sock: |
1532 | return &bpf_skc_to_udp6_sock_proto; | |
8e4597c6 MKL |
1533 | case BPF_FUNC_sk_storage_get: |
1534 | return &bpf_sk_storage_get_tracing_proto; | |
1535 | case BPF_FUNC_sk_storage_delete: | |
1536 | return &bpf_sk_storage_delete_tracing_proto; | |
b60da495 FR |
1537 | case BPF_FUNC_sock_from_file: |
1538 | return &bpf_sock_from_file_proto; | |
c5dbb89f FR |
1539 | case BPF_FUNC_get_socket_cookie: |
1540 | return &bpf_get_socket_ptr_cookie_proto; | |
f1b9509c | 1541 | #endif |
492e639f YS |
1542 | case BPF_FUNC_seq_printf: |
1543 | return prog->expected_attach_type == BPF_TRACE_ITER ? | |
1544 | &bpf_seq_printf_proto : | |
1545 | NULL; | |
1546 | case BPF_FUNC_seq_write: | |
1547 | return prog->expected_attach_type == BPF_TRACE_ITER ? | |
1548 | &bpf_seq_write_proto : | |
1549 | NULL; | |
eb411377 AM |
1550 | case BPF_FUNC_seq_printf_btf: |
1551 | return prog->expected_attach_type == BPF_TRACE_ITER ? | |
1552 | &bpf_seq_printf_btf_proto : | |
1553 | NULL; | |
6e22ab9d JO |
1554 | case BPF_FUNC_d_path: |
1555 | return &bpf_d_path_proto; | |
f1b9509c | 1556 | default: |
3cee6fb8 MKL |
1557 | fn = raw_tp_prog_func_proto(func_id, prog); |
1558 | if (!fn && prog->expected_attach_type == BPF_TRACE_ITER) | |
1559 | fn = bpf_iter_get_func_proto(func_id, prog); | |
1560 | return fn; | |
f1b9509c AS |
1561 | } |
1562 | } | |
1563 | ||
c4f6699d AS |
1564 | static bool raw_tp_prog_is_valid_access(int off, int size, |
1565 | enum bpf_access_type type, | |
5e43f899 | 1566 | const struct bpf_prog *prog, |
c4f6699d AS |
1567 | struct bpf_insn_access_aux *info) |
1568 | { | |
f1b9509c AS |
1569 | if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS) |
1570 | return false; | |
1571 | if (type != BPF_READ) | |
1572 | return false; | |
1573 | if (off % size != 0) | |
1574 | return false; | |
1575 | return true; | |
1576 | } | |
1577 | ||
1578 | static bool tracing_prog_is_valid_access(int off, int size, | |
1579 | enum bpf_access_type type, | |
1580 | const struct bpf_prog *prog, | |
1581 | struct bpf_insn_access_aux *info) | |
1582 | { | |
1583 | if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS) | |
c4f6699d AS |
1584 | return false; |
1585 | if (type != BPF_READ) | |
1586 | return false; | |
1587 | if (off % size != 0) | |
1588 | return false; | |
9e15db66 | 1589 | return btf_ctx_access(off, size, type, prog, info); |
c4f6699d AS |
1590 | } |
1591 | ||
3e7c67d9 KS |
1592 | int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog, |
1593 | const union bpf_attr *kattr, | |
1594 | union bpf_attr __user *uattr) | |
1595 | { | |
1596 | return -ENOTSUPP; | |
1597 | } | |
1598 | ||
c4f6699d AS |
1599 | const struct bpf_verifier_ops raw_tracepoint_verifier_ops = { |
1600 | .get_func_proto = raw_tp_prog_func_proto, | |
1601 | .is_valid_access = raw_tp_prog_is_valid_access, | |
1602 | }; | |
1603 | ||
1604 | const struct bpf_prog_ops raw_tracepoint_prog_ops = { | |
ebfb4d40 | 1605 | #ifdef CONFIG_NET |
1b4d60ec | 1606 | .test_run = bpf_prog_test_run_raw_tp, |
ebfb4d40 | 1607 | #endif |
c4f6699d AS |
1608 | }; |
1609 | ||
f1b9509c AS |
1610 | const struct bpf_verifier_ops tracing_verifier_ops = { |
1611 | .get_func_proto = tracing_prog_func_proto, | |
1612 | .is_valid_access = tracing_prog_is_valid_access, | |
1613 | }; | |
1614 | ||
1615 | const struct bpf_prog_ops tracing_prog_ops = { | |
da00d2f1 | 1616 | .test_run = bpf_prog_test_run_tracing, |
f1b9509c AS |
1617 | }; |
1618 | ||
9df1c28b MM |
1619 | static bool raw_tp_writable_prog_is_valid_access(int off, int size, |
1620 | enum bpf_access_type type, | |
1621 | const struct bpf_prog *prog, | |
1622 | struct bpf_insn_access_aux *info) | |
1623 | { | |
1624 | if (off == 0) { | |
1625 | if (size != sizeof(u64) || type != BPF_READ) | |
1626 | return false; | |
1627 | info->reg_type = PTR_TO_TP_BUFFER; | |
1628 | } | |
1629 | return raw_tp_prog_is_valid_access(off, size, type, prog, info); | |
1630 | } | |
1631 | ||
1632 | const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = { | |
1633 | .get_func_proto = raw_tp_prog_func_proto, | |
1634 | .is_valid_access = raw_tp_writable_prog_is_valid_access, | |
1635 | }; | |
1636 | ||
1637 | const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = { | |
1638 | }; | |
1639 | ||
0515e599 | 1640 | static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type, |
5e43f899 | 1641 | const struct bpf_prog *prog, |
23994631 | 1642 | struct bpf_insn_access_aux *info) |
0515e599 | 1643 | { |
95da0cdb | 1644 | const int size_u64 = sizeof(u64); |
31fd8581 | 1645 | |
0515e599 AS |
1646 | if (off < 0 || off >= sizeof(struct bpf_perf_event_data)) |
1647 | return false; | |
1648 | if (type != BPF_READ) | |
1649 | return false; | |
bc23105c DB |
1650 | if (off % size != 0) { |
1651 | if (sizeof(unsigned long) != 4) | |
1652 | return false; | |
1653 | if (size != 8) | |
1654 | return false; | |
1655 | if (off % size != 4) | |
1656 | return false; | |
1657 | } | |
31fd8581 | 1658 | |
f96da094 DB |
1659 | switch (off) { |
1660 | case bpf_ctx_range(struct bpf_perf_event_data, sample_period): | |
95da0cdb TQ |
1661 | bpf_ctx_record_field_size(info, size_u64); |
1662 | if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) | |
1663 | return false; | |
1664 | break; | |
1665 | case bpf_ctx_range(struct bpf_perf_event_data, addr): | |
1666 | bpf_ctx_record_field_size(info, size_u64); | |
1667 | if (!bpf_ctx_narrow_access_ok(off, size, size_u64)) | |
23994631 | 1668 | return false; |
f96da094 DB |
1669 | break; |
1670 | default: | |
0515e599 AS |
1671 | if (size != sizeof(long)) |
1672 | return false; | |
1673 | } | |
f96da094 | 1674 | |
0515e599 AS |
1675 | return true; |
1676 | } | |
1677 | ||
6b8cc1d1 DB |
1678 | static u32 pe_prog_convert_ctx_access(enum bpf_access_type type, |
1679 | const struct bpf_insn *si, | |
0515e599 | 1680 | struct bpf_insn *insn_buf, |
f96da094 | 1681 | struct bpf_prog *prog, u32 *target_size) |
0515e599 AS |
1682 | { |
1683 | struct bpf_insn *insn = insn_buf; | |
1684 | ||
6b8cc1d1 | 1685 | switch (si->off) { |
0515e599 | 1686 | case offsetof(struct bpf_perf_event_data, sample_period): |
f035a515 | 1687 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 1688 | data), si->dst_reg, si->src_reg, |
0515e599 | 1689 | offsetof(struct bpf_perf_event_data_kern, data)); |
6b8cc1d1 | 1690 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, |
f96da094 DB |
1691 | bpf_target_off(struct perf_sample_data, period, 8, |
1692 | target_size)); | |
0515e599 | 1693 | break; |
95da0cdb TQ |
1694 | case offsetof(struct bpf_perf_event_data, addr): |
1695 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, | |
1696 | data), si->dst_reg, si->src_reg, | |
1697 | offsetof(struct bpf_perf_event_data_kern, data)); | |
1698 | *insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg, | |
1699 | bpf_target_off(struct perf_sample_data, addr, 8, | |
1700 | target_size)); | |
1701 | break; | |
0515e599 | 1702 | default: |
f035a515 | 1703 | *insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern, |
6b8cc1d1 | 1704 | regs), si->dst_reg, si->src_reg, |
0515e599 | 1705 | offsetof(struct bpf_perf_event_data_kern, regs)); |
6b8cc1d1 DB |
1706 | *insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg, |
1707 | si->off); | |
0515e599 AS |
1708 | break; |
1709 | } | |
1710 | ||
1711 | return insn - insn_buf; | |
1712 | } | |
1713 | ||
7de16e3a | 1714 | const struct bpf_verifier_ops perf_event_verifier_ops = { |
f005afed | 1715 | .get_func_proto = pe_prog_func_proto, |
0515e599 AS |
1716 | .is_valid_access = pe_prog_is_valid_access, |
1717 | .convert_ctx_access = pe_prog_convert_ctx_access, | |
1718 | }; | |
7de16e3a JK |
1719 | |
1720 | const struct bpf_prog_ops perf_event_prog_ops = { | |
1721 | }; | |
e87c6bc3 YS |
1722 | |
1723 | static DEFINE_MUTEX(bpf_event_mutex); | |
1724 | ||
c8c088ba YS |
1725 | #define BPF_TRACE_MAX_PROGS 64 |
1726 | ||
e87c6bc3 | 1727 | int perf_event_attach_bpf_prog(struct perf_event *event, |
82e6b1ee AN |
1728 | struct bpf_prog *prog, |
1729 | u64 bpf_cookie) | |
e87c6bc3 | 1730 | { |
e672db03 | 1731 | struct bpf_prog_array *old_array; |
e87c6bc3 YS |
1732 | struct bpf_prog_array *new_array; |
1733 | int ret = -EEXIST; | |
1734 | ||
9802d865 | 1735 | /* |
b4da3340 MH |
1736 | * Kprobe override only works if they are on the function entry, |
1737 | * and only if they are on the opt-in list. | |
9802d865 JB |
1738 | */ |
1739 | if (prog->kprobe_override && | |
b4da3340 | 1740 | (!trace_kprobe_on_func_entry(event->tp_event) || |
9802d865 JB |
1741 | !trace_kprobe_error_injectable(event->tp_event))) |
1742 | return -EINVAL; | |
1743 | ||
e87c6bc3 YS |
1744 | mutex_lock(&bpf_event_mutex); |
1745 | ||
1746 | if (event->prog) | |
07c41a29 | 1747 | goto unlock; |
e87c6bc3 | 1748 | |
e672db03 | 1749 | old_array = bpf_event_rcu_dereference(event->tp_event->prog_array); |
c8c088ba YS |
1750 | if (old_array && |
1751 | bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) { | |
1752 | ret = -E2BIG; | |
1753 | goto unlock; | |
1754 | } | |
1755 | ||
82e6b1ee | 1756 | ret = bpf_prog_array_copy(old_array, NULL, prog, bpf_cookie, &new_array); |
e87c6bc3 | 1757 | if (ret < 0) |
07c41a29 | 1758 | goto unlock; |
e87c6bc3 YS |
1759 | |
1760 | /* set the new array to event->tp_event and set event->prog */ | |
1761 | event->prog = prog; | |
82e6b1ee | 1762 | event->bpf_cookie = bpf_cookie; |
e87c6bc3 YS |
1763 | rcu_assign_pointer(event->tp_event->prog_array, new_array); |
1764 | bpf_prog_array_free(old_array); | |
1765 | ||
07c41a29 | 1766 | unlock: |
e87c6bc3 YS |
1767 | mutex_unlock(&bpf_event_mutex); |
1768 | return ret; | |
1769 | } | |
1770 | ||
1771 | void perf_event_detach_bpf_prog(struct perf_event *event) | |
1772 | { | |
e672db03 | 1773 | struct bpf_prog_array *old_array; |
e87c6bc3 YS |
1774 | struct bpf_prog_array *new_array; |
1775 | int ret; | |
1776 | ||
1777 | mutex_lock(&bpf_event_mutex); | |
1778 | ||
1779 | if (!event->prog) | |
07c41a29 | 1780 | goto unlock; |
e87c6bc3 | 1781 | |
e672db03 | 1782 | old_array = bpf_event_rcu_dereference(event->tp_event->prog_array); |
82e6b1ee | 1783 | ret = bpf_prog_array_copy(old_array, event->prog, NULL, 0, &new_array); |
170a7e3e SY |
1784 | if (ret == -ENOENT) |
1785 | goto unlock; | |
e87c6bc3 YS |
1786 | if (ret < 0) { |
1787 | bpf_prog_array_delete_safe(old_array, event->prog); | |
1788 | } else { | |
1789 | rcu_assign_pointer(event->tp_event->prog_array, new_array); | |
1790 | bpf_prog_array_free(old_array); | |
1791 | } | |
1792 | ||
1793 | bpf_prog_put(event->prog); | |
1794 | event->prog = NULL; | |
1795 | ||
07c41a29 | 1796 | unlock: |
e87c6bc3 YS |
1797 | mutex_unlock(&bpf_event_mutex); |
1798 | } | |
f371b304 | 1799 | |
f4e2298e | 1800 | int perf_event_query_prog_array(struct perf_event *event, void __user *info) |
f371b304 YS |
1801 | { |
1802 | struct perf_event_query_bpf __user *uquery = info; | |
1803 | struct perf_event_query_bpf query = {}; | |
e672db03 | 1804 | struct bpf_prog_array *progs; |
3a38bb98 | 1805 | u32 *ids, prog_cnt, ids_len; |
f371b304 YS |
1806 | int ret; |
1807 | ||
031258da | 1808 | if (!perfmon_capable()) |
f371b304 YS |
1809 | return -EPERM; |
1810 | if (event->attr.type != PERF_TYPE_TRACEPOINT) | |
1811 | return -EINVAL; | |
1812 | if (copy_from_user(&query, uquery, sizeof(query))) | |
1813 | return -EFAULT; | |
3a38bb98 YS |
1814 | |
1815 | ids_len = query.ids_len; | |
1816 | if (ids_len > BPF_TRACE_MAX_PROGS) | |
9c481b90 | 1817 | return -E2BIG; |
3a38bb98 YS |
1818 | ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN); |
1819 | if (!ids) | |
1820 | return -ENOMEM; | |
1821 | /* | |
1822 | * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which | |
1823 | * is required when user only wants to check for uquery->prog_cnt. | |
1824 | * There is no need to check for it since the case is handled | |
1825 | * gracefully in bpf_prog_array_copy_info. | |
1826 | */ | |
f371b304 YS |
1827 | |
1828 | mutex_lock(&bpf_event_mutex); | |
e672db03 SF |
1829 | progs = bpf_event_rcu_dereference(event->tp_event->prog_array); |
1830 | ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt); | |
f371b304 YS |
1831 | mutex_unlock(&bpf_event_mutex); |
1832 | ||
3a38bb98 YS |
1833 | if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) || |
1834 | copy_to_user(uquery->ids, ids, ids_len * sizeof(u32))) | |
1835 | ret = -EFAULT; | |
1836 | ||
1837 | kfree(ids); | |
f371b304 YS |
1838 | return ret; |
1839 | } | |
c4f6699d AS |
1840 | |
1841 | extern struct bpf_raw_event_map __start__bpf_raw_tp[]; | |
1842 | extern struct bpf_raw_event_map __stop__bpf_raw_tp[]; | |
1843 | ||
a38d1107 | 1844 | struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name) |
c4f6699d AS |
1845 | { |
1846 | struct bpf_raw_event_map *btp = __start__bpf_raw_tp; | |
1847 | ||
1848 | for (; btp < __stop__bpf_raw_tp; btp++) { | |
1849 | if (!strcmp(btp->tp->name, name)) | |
1850 | return btp; | |
1851 | } | |
a38d1107 MM |
1852 | |
1853 | return bpf_get_raw_tracepoint_module(name); | |
1854 | } | |
1855 | ||
1856 | void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp) | |
1857 | { | |
12cc126d | 1858 | struct module *mod; |
a38d1107 | 1859 | |
12cc126d AN |
1860 | preempt_disable(); |
1861 | mod = __module_address((unsigned long)btp); | |
1862 | module_put(mod); | |
1863 | preempt_enable(); | |
c4f6699d AS |
1864 | } |
1865 | ||
1866 | static __always_inline | |
1867 | void __bpf_trace_run(struct bpf_prog *prog, u64 *args) | |
1868 | { | |
f03efe49 | 1869 | cant_sleep(); |
c4f6699d | 1870 | rcu_read_lock(); |
fb7dd8bc | 1871 | (void) bpf_prog_run(prog, args); |
c4f6699d AS |
1872 | rcu_read_unlock(); |
1873 | } | |
1874 | ||
1875 | #define UNPACK(...) __VA_ARGS__ | |
1876 | #define REPEAT_1(FN, DL, X, ...) FN(X) | |
1877 | #define REPEAT_2(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__) | |
1878 | #define REPEAT_3(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__) | |
1879 | #define REPEAT_4(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__) | |
1880 | #define REPEAT_5(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__) | |
1881 | #define REPEAT_6(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__) | |
1882 | #define REPEAT_7(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__) | |
1883 | #define REPEAT_8(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__) | |
1884 | #define REPEAT_9(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__) | |
1885 | #define REPEAT_10(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__) | |
1886 | #define REPEAT_11(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__) | |
1887 | #define REPEAT_12(FN, DL, X, ...) FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__) | |
1888 | #define REPEAT(X, FN, DL, ...) REPEAT_##X(FN, DL, __VA_ARGS__) | |
1889 | ||
1890 | #define SARG(X) u64 arg##X | |
1891 | #define COPY(X) args[X] = arg##X | |
1892 | ||
1893 | #define __DL_COM (,) | |
1894 | #define __DL_SEM (;) | |
1895 | ||
1896 | #define __SEQ_0_11 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11 | |
1897 | ||
1898 | #define BPF_TRACE_DEFN_x(x) \ | |
1899 | void bpf_trace_run##x(struct bpf_prog *prog, \ | |
1900 | REPEAT(x, SARG, __DL_COM, __SEQ_0_11)) \ | |
1901 | { \ | |
1902 | u64 args[x]; \ | |
1903 | REPEAT(x, COPY, __DL_SEM, __SEQ_0_11); \ | |
1904 | __bpf_trace_run(prog, args); \ | |
1905 | } \ | |
1906 | EXPORT_SYMBOL_GPL(bpf_trace_run##x) | |
1907 | BPF_TRACE_DEFN_x(1); | |
1908 | BPF_TRACE_DEFN_x(2); | |
1909 | BPF_TRACE_DEFN_x(3); | |
1910 | BPF_TRACE_DEFN_x(4); | |
1911 | BPF_TRACE_DEFN_x(5); | |
1912 | BPF_TRACE_DEFN_x(6); | |
1913 | BPF_TRACE_DEFN_x(7); | |
1914 | BPF_TRACE_DEFN_x(8); | |
1915 | BPF_TRACE_DEFN_x(9); | |
1916 | BPF_TRACE_DEFN_x(10); | |
1917 | BPF_TRACE_DEFN_x(11); | |
1918 | BPF_TRACE_DEFN_x(12); | |
1919 | ||
1920 | static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1921 | { | |
1922 | struct tracepoint *tp = btp->tp; | |
1923 | ||
1924 | /* | |
1925 | * check that program doesn't access arguments beyond what's | |
1926 | * available in this tracepoint | |
1927 | */ | |
1928 | if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64)) | |
1929 | return -EINVAL; | |
1930 | ||
9df1c28b MM |
1931 | if (prog->aux->max_tp_access > btp->writable_size) |
1932 | return -EINVAL; | |
1933 | ||
9913d574 SRV |
1934 | return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func, |
1935 | prog); | |
c4f6699d AS |
1936 | } |
1937 | ||
1938 | int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1939 | { | |
e16ec340 | 1940 | return __bpf_probe_register(btp, prog); |
c4f6699d AS |
1941 | } |
1942 | ||
1943 | int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog) | |
1944 | { | |
e16ec340 | 1945 | return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog); |
c4f6699d | 1946 | } |
41bdc4b4 YS |
1947 | |
1948 | int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id, | |
1949 | u32 *fd_type, const char **buf, | |
1950 | u64 *probe_offset, u64 *probe_addr) | |
1951 | { | |
1952 | bool is_tracepoint, is_syscall_tp; | |
1953 | struct bpf_prog *prog; | |
1954 | int flags, err = 0; | |
1955 | ||
1956 | prog = event->prog; | |
1957 | if (!prog) | |
1958 | return -ENOENT; | |
1959 | ||
1960 | /* not supporting BPF_PROG_TYPE_PERF_EVENT yet */ | |
1961 | if (prog->type == BPF_PROG_TYPE_PERF_EVENT) | |
1962 | return -EOPNOTSUPP; | |
1963 | ||
1964 | *prog_id = prog->aux->id; | |
1965 | flags = event->tp_event->flags; | |
1966 | is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT; | |
1967 | is_syscall_tp = is_syscall_trace_event(event->tp_event); | |
1968 | ||
1969 | if (is_tracepoint || is_syscall_tp) { | |
1970 | *buf = is_tracepoint ? event->tp_event->tp->name | |
1971 | : event->tp_event->name; | |
1972 | *fd_type = BPF_FD_TYPE_TRACEPOINT; | |
1973 | *probe_offset = 0x0; | |
1974 | *probe_addr = 0x0; | |
1975 | } else { | |
1976 | /* kprobe/uprobe */ | |
1977 | err = -EOPNOTSUPP; | |
1978 | #ifdef CONFIG_KPROBE_EVENTS | |
1979 | if (flags & TRACE_EVENT_FL_KPROBE) | |
1980 | err = bpf_get_kprobe_info(event, fd_type, buf, | |
1981 | probe_offset, probe_addr, | |
1982 | event->attr.type == PERF_TYPE_TRACEPOINT); | |
1983 | #endif | |
1984 | #ifdef CONFIG_UPROBE_EVENTS | |
1985 | if (flags & TRACE_EVENT_FL_UPROBE) | |
1986 | err = bpf_get_uprobe_info(event, fd_type, buf, | |
1987 | probe_offset, | |
1988 | event->attr.type == PERF_TYPE_TRACEPOINT); | |
1989 | #endif | |
1990 | } | |
1991 | ||
1992 | return err; | |
1993 | } | |
a38d1107 | 1994 | |
9db1ff0a YS |
1995 | static int __init send_signal_irq_work_init(void) |
1996 | { | |
1997 | int cpu; | |
1998 | struct send_signal_irq_work *work; | |
1999 | ||
2000 | for_each_possible_cpu(cpu) { | |
2001 | work = per_cpu_ptr(&send_signal_work, cpu); | |
2002 | init_irq_work(&work->irq_work, do_bpf_send_signal); | |
2003 | } | |
2004 | return 0; | |
2005 | } | |
2006 | ||
2007 | subsys_initcall(send_signal_irq_work_init); | |
2008 | ||
a38d1107 | 2009 | #ifdef CONFIG_MODULES |
390e99cf SF |
2010 | static int bpf_event_notify(struct notifier_block *nb, unsigned long op, |
2011 | void *module) | |
a38d1107 MM |
2012 | { |
2013 | struct bpf_trace_module *btm, *tmp; | |
2014 | struct module *mod = module; | |
0340a6b7 | 2015 | int ret = 0; |
a38d1107 MM |
2016 | |
2017 | if (mod->num_bpf_raw_events == 0 || | |
2018 | (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING)) | |
0340a6b7 | 2019 | goto out; |
a38d1107 MM |
2020 | |
2021 | mutex_lock(&bpf_module_mutex); | |
2022 | ||
2023 | switch (op) { | |
2024 | case MODULE_STATE_COMING: | |
2025 | btm = kzalloc(sizeof(*btm), GFP_KERNEL); | |
2026 | if (btm) { | |
2027 | btm->module = module; | |
2028 | list_add(&btm->list, &bpf_trace_modules); | |
0340a6b7 PZ |
2029 | } else { |
2030 | ret = -ENOMEM; | |
a38d1107 MM |
2031 | } |
2032 | break; | |
2033 | case MODULE_STATE_GOING: | |
2034 | list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) { | |
2035 | if (btm->module == module) { | |
2036 | list_del(&btm->list); | |
2037 | kfree(btm); | |
2038 | break; | |
2039 | } | |
2040 | } | |
2041 | break; | |
2042 | } | |
2043 | ||
2044 | mutex_unlock(&bpf_module_mutex); | |
2045 | ||
0340a6b7 PZ |
2046 | out: |
2047 | return notifier_from_errno(ret); | |
a38d1107 MM |
2048 | } |
2049 | ||
2050 | static struct notifier_block bpf_module_nb = { | |
2051 | .notifier_call = bpf_event_notify, | |
2052 | }; | |
2053 | ||
390e99cf | 2054 | static int __init bpf_event_init(void) |
a38d1107 MM |
2055 | { |
2056 | register_module_notifier(&bpf_module_nb); | |
2057 | return 0; | |
2058 | } | |
2059 | ||
2060 | fs_initcall(bpf_event_init); | |
2061 | #endif /* CONFIG_MODULES */ |