Linux Audio

Check our new training course

Loading...
v5.4
   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 */
v5.14.15
   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/btf.h>
  11#include <linux/filter.h>
  12#include <linux/uaccess.h>
  13#include <linux/ctype.h>
  14#include <linux/kprobes.h>
  15#include <linux/spinlock.h>
  16#include <linux/syscalls.h>
  17#include <linux/error-injection.h>
  18#include <linux/btf_ids.h>
  19#include <linux/bpf_lsm.h>
  20
  21#include <net/bpf_sk_storage.h>
  22
  23#include <uapi/linux/bpf.h>
  24#include <uapi/linux/btf.h>
  25
  26#include <asm/tlb.h>
  27
  28#include "trace_probe.h"
  29#include "trace.h"
  30
  31#define CREATE_TRACE_POINTS
  32#include "bpf_trace.h"
  33
  34#define bpf_event_rcu_dereference(p)					\
  35	rcu_dereference_protected(p, lockdep_is_held(&bpf_event_mutex))
  36
  37#ifdef CONFIG_MODULES
  38struct bpf_trace_module {
  39	struct module *module;
  40	struct list_head list;
  41};
  42
  43static LIST_HEAD(bpf_trace_modules);
  44static DEFINE_MUTEX(bpf_module_mutex);
  45
  46static 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	}
  63out:
  64	mutex_unlock(&bpf_module_mutex);
  65	return ret;
  66}
  67#else
  68static struct bpf_raw_event_map *bpf_get_raw_tracepoint_module(const char *name)
  69{
  70	return NULL;
  71}
  72#endif /* CONFIG_MODULES */
  73
  74u64 bpf_get_stackid(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  75u64 bpf_get_stack(u64 r1, u64 r2, u64 r3, u64 r4, u64 r5);
  76
  77static 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
  81/**
  82 * trace_call_bpf - invoke BPF program
  83 * @call: tracepoint event
  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 */
  95unsigned int trace_call_bpf(struct trace_event_call *call, void *ctx)
  96{
  97	unsigned int ret;
  98
  99	cant_sleep();
 
 
 
 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
 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
 116	 * a heuristic to speed up execution.
 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	 */
 127	ret = BPF_PROG_RUN_ARRAY_CHECK(call->prog_array, ctx, BPF_PROG_RUN);
 128
 129 out:
 130	__this_cpu_dec(bpf_prog_active);
 
 131
 132	return ret;
 133}
 
 134
 135#ifdef CONFIG_BPF_KPROBE_OVERRIDE
 136BPF_CALL_2(bpf_override_return, struct pt_regs *, regs, unsigned long, rc)
 137{
 138	regs_set_return_value(regs, rc);
 139	override_function_with_return(regs);
 140	return 0;
 141}
 142
 143static 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
 152static __always_inline int
 153bpf_probe_read_user_common(void *dst, u32 size, const void __user *unsafe_ptr)
 154{
 155	int ret;
 156
 157	ret = copy_from_user_nofault(dst, unsafe_ptr, size);
 158	if (unlikely(ret < 0))
 159		memset(dst, 0, size);
 160	return ret;
 161}
 162
 163BPF_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
 169const struct bpf_func_proto bpf_probe_read_user_proto = {
 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
 178static __always_inline int
 179bpf_probe_read_user_str_common(void *dst, u32 size,
 180			       const void __user *unsafe_ptr)
 181{
 182	int ret;
 183
 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	 */
 194	ret = strncpy_from_user_nofault(dst, unsafe_ptr, size);
 195	if (unlikely(ret < 0))
 196		memset(dst, 0, size);
 197	return ret;
 198}
 199
 200BPF_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
 206const struct bpf_func_proto bpf_probe_read_user_str_proto = {
 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
 215static __always_inline int
 216bpf_probe_read_kernel_common(void *dst, u32 size, const void *unsafe_ptr)
 217{
 218	int ret;
 219
 220	ret = copy_from_kernel_nofault(dst, unsafe_ptr, size);
 221	if (unlikely(ret < 0))
 
 222		memset(dst, 0, size);
 223	return ret;
 224}
 225
 226BPF_CALL_3(bpf_probe_read_kernel, void *, dst, u32, size,
 227	   const void *, unsafe_ptr)
 228{
 229	return bpf_probe_read_kernel_common(dst, size, unsafe_ptr);
 230}
 231
 232const struct bpf_func_proto bpf_probe_read_kernel_proto = {
 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
 241static __always_inline int
 242bpf_probe_read_kernel_str_common(void *dst, u32 size, const void *unsafe_ptr)
 243{
 244	int ret;
 245
 246	/*
 247	 * The strncpy_from_kernel_nofault() call will likely not fill the
 248	 * entire buffer, but that's okay in this circumstance as we're probing
 249	 * arbitrary memory anyway similar to bpf_probe_read_*() and might
 250	 * as well probe the stack. Thus, memory is explicitly cleared
 251	 * only in error case, so that improper users ignoring return
 252	 * code altogether don't copy garbage; otherwise length of string
 253	 * is returned that can be used for bpf_perf_event_output() et al.
 254	 */
 255	ret = strncpy_from_kernel_nofault(dst, unsafe_ptr, size);
 256	if (unlikely(ret < 0))
 257		memset(dst, 0, size);
 258	return ret;
 259}
 260
 261BPF_CALL_3(bpf_probe_read_kernel_str, void *, dst, u32, size,
 262	   const void *, unsafe_ptr)
 263{
 264	return bpf_probe_read_kernel_str_common(dst, size, unsafe_ptr);
 265}
 266
 267const struct bpf_func_proto bpf_probe_read_kernel_str_proto = {
 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
 276#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
 277BPF_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
 287static 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
 296BPF_CALL_3(bpf_probe_read_compat_str, void *, dst, u32, size,
 297	   const void *, unsafe_ptr)
 298{
 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);
 304}
 305
 306static const struct bpf_func_proto bpf_probe_read_compat_str_proto = {
 307	.func		= bpf_probe_read_compat_str,
 308	.gpl_only	= true,
 309	.ret_type	= RET_INTEGER,
 310	.arg1_type	= ARG_PTR_TO_UNINIT_MEM,
 311	.arg2_type	= ARG_CONST_SIZE_OR_ZERO,
 312	.arg3_type	= ARG_ANYTHING,
 313};
 314#endif /* CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE */
 315
 316BPF_CALL_3(bpf_probe_write_user, void __user *, unsafe_ptr, const void *, src,
 317	   u32, size)
 318{
 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.
 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.
 330	 */
 331
 332	if (unlikely(in_interrupt() ||
 333		     current->flags & (PF_KTHREAD | PF_EXITING)))
 334		return -EPERM;
 335	if (unlikely(uaccess_kernel()))
 336		return -EPERM;
 337	if (unlikely(!nmi_uaccess_okay()))
 338		return -EPERM;
 
 
 339
 340	return copy_to_user_nofault(unsafe_ptr, src, size);
 341}
 342
 343static 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,
 348	.arg2_type	= ARG_PTR_TO_MEM,
 349	.arg3_type	= ARG_CONST_SIZE,
 350};
 351
 352static const struct bpf_func_proto *bpf_get_probe_write_proto(void)
 353{
 354	if (!capable(CAP_SYS_ADMIN))
 355		return NULL;
 356
 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
 363static DEFINE_RAW_SPINLOCK(trace_printk_lock);
 364
 365#define MAX_TRACE_PRINTK_VARARGS	3
 366#define BPF_TRACE_PRINTK_SIZE		1024
 367
 368BPF_CALL_5(bpf_trace_printk, char *, fmt, u32, fmt_size, u64, arg1,
 369	   u64, arg2, u64, arg3)
 370{
 371	u64 args[MAX_TRACE_PRINTK_VARARGS] = { arg1, arg2, arg3 };
 372	u32 *bin_args;
 373	static char buf[BPF_TRACE_PRINTK_SIZE];
 374	unsigned long flags;
 375	int ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 376
 377	ret = bpf_bprintf_prepare(fmt, fmt_size, args, &bin_args,
 378				  MAX_TRACE_PRINTK_VARARGS);
 379	if (ret < 0)
 380		return ret;
 381
 382	raw_spin_lock_irqsave(&trace_printk_lock, flags);
 383	ret = bstr_printf(buf, sizeof(buf), fmt, bin_args);
 384
 385	trace_bpf_trace_printk(buf);
 386	raw_spin_unlock_irqrestore(&trace_printk_lock, flags);
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 387
 388	bpf_bprintf_cleanup();
 
 
 
 389
 390	return ret;
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 391}
 392
 393static const struct bpf_func_proto bpf_trace_printk_proto = {
 394	.func		= bpf_trace_printk,
 395	.gpl_only	= true,
 396	.ret_type	= RET_INTEGER,
 397	.arg1_type	= ARG_PTR_TO_MEM,
 398	.arg2_type	= ARG_CONST_SIZE,
 399};
 400
 401const struct bpf_func_proto *bpf_get_trace_printk_proto(void)
 402{
 403	/*
 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.
 410	 */
 411	if (trace_set_clr_event("bpf_trace", "bpf_trace_printk", 1))
 412		pr_warn_ratelimited("could not enable bpf_trace_printk events");
 413
 414	return &bpf_trace_printk_proto;
 415}
 416
 417#define MAX_SEQ_PRINTF_VARARGS		12
 418
 419BPF_CALL_5(bpf_seq_printf, struct seq_file *, m, char *, fmt, u32, fmt_size,
 420	   const void *, data, u32, data_len)
 421{
 422	int err, num_args;
 423	u32 *bin_args;
 424
 425	if (data_len & 7 || data_len > MAX_SEQ_PRINTF_VARARGS * 8 ||
 426	    (data_len && !data))
 427		return -EINVAL;
 428	num_args = data_len / 8;
 429
 430	err = bpf_bprintf_prepare(fmt, fmt_size, data, &bin_args, num_args);
 431	if (err < 0)
 432		return err;
 433
 434	seq_bprintf(m, fmt, bin_args);
 435
 436	bpf_bprintf_cleanup();
 437
 438	return seq_has_overflowed(m) ? -EOVERFLOW : 0;
 439}
 440
 441BTF_ID_LIST_SINGLE(btf_seq_file_ids, struct, seq_file)
 442
 443static 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,
 448	.arg1_btf_id	= &btf_seq_file_ids[0],
 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,
 453};
 454
 455BPF_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
 460static 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,
 465	.arg1_btf_id	= &btf_seq_file_ids[0],
 466	.arg2_type	= ARG_PTR_TO_MEM,
 467	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
 468};
 469
 470BPF_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
 484static 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],
 490	.arg2_type	= ARG_PTR_TO_MEM,
 491	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
 492	.arg4_type	= ARG_ANYTHING,
 493};
 494
 495static __always_inline int
 496get_map_perf_counter(struct bpf_map *map, u64 flags,
 497		     u64 *value, u64 *enabled, u64 *running)
 498{
 499	struct bpf_array *array = container_of(map, struct bpf_array, map);
 500	unsigned int cpu = smp_processor_id();
 501	u64 index = flags & BPF_F_INDEX_MASK;
 502	struct bpf_event_entry *ee;
 503
 504	if (unlikely(flags & ~(BPF_F_INDEX_MASK)))
 505		return -EINVAL;
 506	if (index == BPF_F_CURRENT_CPU)
 507		index = cpu;
 508	if (unlikely(index >= array->map.max_entries))
 509		return -E2BIG;
 510
 511	ee = READ_ONCE(array->ptrs[index]);
 512	if (!ee)
 513		return -ENOENT;
 514
 515	return perf_event_read_local(ee->event, value, enabled, running);
 516}
 517
 518BPF_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);
 524	/*
 525	 * this api is ugly since we miss [-22..-2] range of valid
 526	 * counter values, but that's uapi
 527	 */
 528	if (err)
 529		return err;
 530	return value;
 531}
 532
 533static const struct bpf_func_proto bpf_perf_event_read_proto = {
 534	.func		= bpf_perf_event_read,
 535	.gpl_only	= true,
 536	.ret_type	= RET_INTEGER,
 537	.arg1_type	= ARG_CONST_MAP_PTR,
 538	.arg2_type	= ARG_ANYTHING,
 539};
 540
 541BPF_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;
 553clear:
 554	memset(buf, 0, size);
 555	return err;
 556}
 557
 558static 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
 568static __always_inline u64
 569__bpf_perf_event_output(struct pt_regs *regs, struct bpf_map *map,
 570			u64 flags, struct perf_sample_data *sd)
 571{
 572	struct bpf_array *array = container_of(map, struct bpf_array, map);
 573	unsigned int cpu = smp_processor_id();
 574	u64 index = flags & BPF_F_INDEX_MASK;
 575	struct bpf_event_entry *ee;
 576	struct perf_event *event;
 577
 578	if (index == BPF_F_CURRENT_CPU)
 579		index = cpu;
 580	if (unlikely(index >= array->map.max_entries))
 581		return -E2BIG;
 582
 583	ee = READ_ONCE(array->ptrs[index]);
 584	if (!ee)
 585		return -ENOENT;
 586
 587	event = ee->event;
 588	if (unlikely(event->attr.type != PERF_TYPE_SOFTWARE ||
 589		     event->attr.config != PERF_COUNT_SW_BPF_OUTPUT))
 590		return -EINVAL;
 591
 592	if (unlikely(event->oncpu != cpu))
 593		return -EOPNOTSUPP;
 594
 595	return perf_event_output(event, sd, regs);
 596}
 597
 598/*
 599 * Support executing tracepoints in normal, irq, and nmi context that each call
 600 * bpf_perf_event_output
 601 */
 602struct bpf_trace_sample_data {
 603	struct perf_sample_data sds[3];
 604};
 605
 606static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_trace_sds);
 607static DEFINE_PER_CPU(int, bpf_trace_nest_level);
 608BPF_CALL_5(bpf_perf_event_output, struct pt_regs *, regs, struct bpf_map *, map,
 609	   u64, flags, void *, data, u64, size)
 610{
 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);
 613	struct perf_raw_record raw = {
 614		.frag = {
 615			.size = size,
 616			.data = data,
 617		},
 618	};
 619	struct perf_sample_data *sd;
 620	int err;
 621
 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	}
 633
 634	perf_sample_data_init(sd, 0, 0);
 635	sd->raw = &raw;
 636
 637	err = __bpf_perf_event_output(regs, map, flags, sd);
 638
 639out:
 640	this_cpu_dec(bpf_trace_nest_level);
 641	return err;
 642}
 643
 644static const struct bpf_func_proto bpf_perf_event_output_proto = {
 645	.func		= bpf_perf_event_output,
 646	.gpl_only	= true,
 647	.ret_type	= RET_INTEGER,
 648	.arg1_type	= ARG_PTR_TO_CTX,
 649	.arg2_type	= ARG_CONST_MAP_PTR,
 650	.arg3_type	= ARG_ANYTHING,
 651	.arg4_type	= ARG_PTR_TO_MEM,
 652	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
 653};
 654
 655static DEFINE_PER_CPU(int, bpf_event_output_nest_level);
 656struct bpf_nested_pt_regs {
 657	struct pt_regs regs[3];
 658};
 659static DEFINE_PER_CPU(struct bpf_nested_pt_regs, bpf_pt_regs);
 660static DEFINE_PER_CPU(struct bpf_trace_sample_data, bpf_misc_sds);
 661
 662u64 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)
 664{
 665	int nest_level = this_cpu_inc_return(bpf_event_output_nest_level);
 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 = {
 673			{
 674				.next	= ctx_size ? &frag : NULL,
 675			},
 676			.size	= meta_size,
 677			.data	= meta,
 678		},
 679	};
 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]);
 690
 691	perf_fetch_caller_regs(regs);
 692	perf_sample_data_init(sd, 0, 0);
 693	sd->raw = &raw;
 694
 695	ret = __bpf_perf_event_output(regs, map, flags, sd);
 696out:
 697	this_cpu_dec(bpf_event_output_nest_level);
 698	return ret;
 699}
 700
 701BPF_CALL_0(bpf_get_current_task)
 702{
 703	return (long) current;
 704}
 705
 706const struct bpf_func_proto bpf_get_current_task_proto = {
 707	.func		= bpf_get_current_task,
 708	.gpl_only	= true,
 709	.ret_type	= RET_INTEGER,
 710};
 711
 712BPF_CALL_0(bpf_get_current_task_btf)
 713{
 714	return (unsigned long) current;
 715}
 716
 717BTF_ID_LIST_SINGLE(bpf_get_current_btf_ids, struct, task_struct)
 718
 719static const struct bpf_func_proto bpf_get_current_task_btf_proto = {
 720	.func		= bpf_get_current_task_btf,
 721	.gpl_only	= true,
 722	.ret_type	= RET_PTR_TO_BTF_ID,
 723	.ret_btf_id	= &bpf_get_current_btf_ids[0],
 724};
 725
 726BPF_CALL_2(bpf_current_task_under_cgroup, struct bpf_map *, map, u32, idx)
 727{
 728	struct bpf_array *array = container_of(map, struct bpf_array, map);
 729	struct cgroup *cgrp;
 730
 731	if (unlikely(idx >= array->map.max_entries))
 732		return -E2BIG;
 733
 734	cgrp = READ_ONCE(array->ptrs[idx]);
 735	if (unlikely(!cgrp))
 736		return -EAGAIN;
 737
 738	return task_under_cgroup_hierarchy(current, cgrp);
 739}
 740
 741static const struct bpf_func_proto bpf_current_task_under_cgroup_proto = {
 742	.func           = bpf_current_task_under_cgroup,
 743	.gpl_only       = false,
 744	.ret_type       = RET_INTEGER,
 745	.arg1_type      = ARG_CONST_MAP_PTR,
 746	.arg2_type      = ARG_ANYTHING,
 747};
 748
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 749struct send_signal_irq_work {
 750	struct irq_work irq_work;
 751	struct task_struct *task;
 752	u32 sig;
 753	enum pid_type type;
 754};
 755
 756static DEFINE_PER_CPU(struct send_signal_irq_work, send_signal_work);
 757
 758static void do_bpf_send_signal(struct irq_work *entry)
 759{
 760	struct send_signal_irq_work *work;
 761
 762	work = container_of(entry, struct send_signal_irq_work, irq_work);
 763	group_send_sig_info(work->sig, SEND_SIG_PRIV, work->task, work->type);
 764}
 765
 766static int bpf_send_signal_common(u32 sig, enum pid_type type)
 767{
 768	struct send_signal_irq_work *work = NULL;
 769
 770	/* Similar to bpf_probe_write_user, task needs to be
 771	 * in a sound condition and kernel memory access be
 772	 * permitted in order to send signal to the current
 773	 * task.
 774	 */
 775	if (unlikely(current->flags & (PF_KTHREAD | PF_EXITING)))
 776		return -EPERM;
 777	if (unlikely(uaccess_kernel()))
 778		return -EPERM;
 779	if (unlikely(!nmi_uaccess_okay()))
 780		return -EPERM;
 781
 782	if (irqs_disabled()) {
 783		/* Do an early check on signal validity. Otherwise,
 784		 * the error is lost in deferred irq_work.
 785		 */
 786		if (unlikely(!valid_signal(sig)))
 787			return -EINVAL;
 788
 789		work = this_cpu_ptr(&send_signal_work);
 790		if (irq_work_is_busy(&work->irq_work))
 791			return -EBUSY;
 792
 793		/* Add the current task, which is the target of sending signal,
 794		 * to the irq_work. The current task may change when queued
 795		 * irq works get executed.
 796		 */
 797		work->task = current;
 798		work->sig = sig;
 799		work->type = type;
 800		irq_work_queue(&work->irq_work);
 801		return 0;
 802	}
 803
 804	return group_send_sig_info(sig, SEND_SIG_PRIV, current, type);
 805}
 806
 807BPF_CALL_1(bpf_send_signal, u32, sig)
 808{
 809	return bpf_send_signal_common(sig, PIDTYPE_TGID);
 810}
 811
 812static const struct bpf_func_proto bpf_send_signal_proto = {
 813	.func		= bpf_send_signal,
 814	.gpl_only	= false,
 815	.ret_type	= RET_INTEGER,
 816	.arg1_type	= ARG_ANYTHING,
 817};
 818
 819BPF_CALL_1(bpf_send_signal_thread, u32, sig)
 820{
 821	return bpf_send_signal_common(sig, PIDTYPE_PID);
 822}
 823
 824static const struct bpf_func_proto bpf_send_signal_thread_proto = {
 825	.func		= bpf_send_signal_thread,
 826	.gpl_only	= false,
 827	.ret_type	= RET_INTEGER,
 828	.arg1_type	= ARG_ANYTHING,
 829};
 830
 831BPF_CALL_3(bpf_d_path, struct path *, path, char *, buf, u32, sz)
 832{
 833	long len;
 834	char *p;
 835
 836	if (!sz)
 837		return 0;
 838
 839	p = d_path(path, buf, sz);
 840	if (IS_ERR(p)) {
 841		len = PTR_ERR(p);
 842	} else {
 843		len = buf + sz - p;
 844		memmove(buf, p, len);
 845	}
 846
 847	return len;
 848}
 849
 850BTF_SET_START(btf_allowlist_d_path)
 851#ifdef CONFIG_SECURITY
 852BTF_ID(func, security_file_permission)
 853BTF_ID(func, security_inode_getattr)
 854BTF_ID(func, security_file_open)
 855#endif
 856#ifdef CONFIG_SECURITY_PATH
 857BTF_ID(func, security_path_truncate)
 858#endif
 859BTF_ID(func, vfs_truncate)
 860BTF_ID(func, vfs_fallocate)
 861BTF_ID(func, dentry_open)
 862BTF_ID(func, vfs_getattr)
 863BTF_ID(func, filp_close)
 864BTF_SET_END(btf_allowlist_d_path)
 865
 866static bool bpf_d_path_allowed(const struct bpf_prog *prog)
 867{
 868	if (prog->type == BPF_PROG_TYPE_TRACING &&
 869	    prog->expected_attach_type == BPF_TRACE_ITER)
 870		return true;
 871
 872	if (prog->type == BPF_PROG_TYPE_LSM)
 873		return bpf_lsm_is_sleepable_hook(prog->aux->attach_btf_id);
 874
 875	return btf_id_set_contains(&btf_allowlist_d_path,
 876				   prog->aux->attach_btf_id);
 877}
 878
 879BTF_ID_LIST_SINGLE(bpf_d_path_btf_ids, struct, path)
 880
 881static const struct bpf_func_proto bpf_d_path_proto = {
 882	.func		= bpf_d_path,
 883	.gpl_only	= false,
 884	.ret_type	= RET_INTEGER,
 885	.arg1_type	= ARG_PTR_TO_BTF_ID,
 886	.arg1_btf_id	= &bpf_d_path_btf_ids[0],
 887	.arg2_type	= ARG_PTR_TO_MEM,
 888	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
 889	.allowed	= bpf_d_path_allowed,
 890};
 891
 892#define BTF_F_ALL	(BTF_F_COMPACT  | BTF_F_NONAME | \
 893			 BTF_F_PTR_RAW | BTF_F_ZERO)
 894
 895static int bpf_btf_printf_prepare(struct btf_ptr *ptr, u32 btf_ptr_size,
 896				  u64 flags, const struct btf **btf,
 897				  s32 *btf_id)
 898{
 899	const struct btf_type *t;
 900
 901	if (unlikely(flags & ~(BTF_F_ALL)))
 902		return -EINVAL;
 903
 904	if (btf_ptr_size != sizeof(struct btf_ptr))
 905		return -EINVAL;
 906
 907	*btf = bpf_get_btf_vmlinux();
 908
 909	if (IS_ERR_OR_NULL(*btf))
 910		return IS_ERR(*btf) ? PTR_ERR(*btf) : -EINVAL;
 911
 912	if (ptr->type_id > 0)
 913		*btf_id = ptr->type_id;
 914	else
 915		return -EINVAL;
 916
 917	if (*btf_id > 0)
 918		t = btf_type_by_id(*btf, *btf_id);
 919	if (*btf_id <= 0 || !t)
 920		return -ENOENT;
 921
 922	return 0;
 923}
 924
 925BPF_CALL_5(bpf_snprintf_btf, char *, str, u32, str_size, struct btf_ptr *, ptr,
 926	   u32, btf_ptr_size, u64, flags)
 927{
 928	const struct btf *btf;
 929	s32 btf_id;
 930	int ret;
 931
 932	ret = bpf_btf_printf_prepare(ptr, btf_ptr_size, flags, &btf, &btf_id);
 933	if (ret)
 934		return ret;
 935
 936	return btf_type_snprintf_show(btf, btf_id, ptr->ptr, str, str_size,
 937				      flags);
 938}
 939
 940const struct bpf_func_proto bpf_snprintf_btf_proto = {
 941	.func		= bpf_snprintf_btf,
 942	.gpl_only	= false,
 943	.ret_type	= RET_INTEGER,
 944	.arg1_type	= ARG_PTR_TO_MEM,
 945	.arg2_type	= ARG_CONST_SIZE,
 946	.arg3_type	= ARG_PTR_TO_MEM,
 947	.arg4_type	= ARG_CONST_SIZE,
 948	.arg5_type	= ARG_ANYTHING,
 949};
 950
 951const struct bpf_func_proto *
 952bpf_tracing_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
 953{
 954	switch (func_id) {
 955	case BPF_FUNC_map_lookup_elem:
 956		return &bpf_map_lookup_elem_proto;
 957	case BPF_FUNC_map_update_elem:
 958		return &bpf_map_update_elem_proto;
 959	case BPF_FUNC_map_delete_elem:
 960		return &bpf_map_delete_elem_proto;
 961	case BPF_FUNC_map_push_elem:
 962		return &bpf_map_push_elem_proto;
 963	case BPF_FUNC_map_pop_elem:
 964		return &bpf_map_pop_elem_proto;
 965	case BPF_FUNC_map_peek_elem:
 966		return &bpf_map_peek_elem_proto;
 
 
 967	case BPF_FUNC_ktime_get_ns:
 968		return &bpf_ktime_get_ns_proto;
 969	case BPF_FUNC_ktime_get_boot_ns:
 970		return &bpf_ktime_get_boot_ns_proto;
 971	case BPF_FUNC_ktime_get_coarse_ns:
 972		return &bpf_ktime_get_coarse_ns_proto;
 973	case BPF_FUNC_tail_call:
 974		return &bpf_tail_call_proto;
 975	case BPF_FUNC_get_current_pid_tgid:
 976		return &bpf_get_current_pid_tgid_proto;
 977	case BPF_FUNC_get_current_task:
 978		return &bpf_get_current_task_proto;
 979	case BPF_FUNC_get_current_task_btf:
 980		return &bpf_get_current_task_btf_proto;
 981	case BPF_FUNC_get_current_uid_gid:
 982		return &bpf_get_current_uid_gid_proto;
 983	case BPF_FUNC_get_current_comm:
 984		return &bpf_get_current_comm_proto;
 985	case BPF_FUNC_trace_printk:
 986		return bpf_get_trace_printk_proto();
 987	case BPF_FUNC_get_smp_processor_id:
 988		return &bpf_get_smp_processor_id_proto;
 989	case BPF_FUNC_get_numa_node_id:
 990		return &bpf_get_numa_node_id_proto;
 991	case BPF_FUNC_perf_event_read:
 992		return &bpf_perf_event_read_proto;
 
 
 993	case BPF_FUNC_current_task_under_cgroup:
 994		return &bpf_current_task_under_cgroup_proto;
 995	case BPF_FUNC_get_prandom_u32:
 996		return &bpf_get_prandom_u32_proto;
 997	case BPF_FUNC_probe_write_user:
 998		return security_locked_down(LOCKDOWN_BPF_WRITE_USER) < 0 ?
 999		       NULL : bpf_get_probe_write_proto();
1000	case BPF_FUNC_probe_read_user:
1001		return &bpf_probe_read_user_proto;
1002	case BPF_FUNC_probe_read_kernel:
1003		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1004		       NULL : &bpf_probe_read_kernel_proto;
1005	case BPF_FUNC_probe_read_user_str:
1006		return &bpf_probe_read_user_str_proto;
1007	case BPF_FUNC_probe_read_kernel_str:
1008		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1009		       NULL : &bpf_probe_read_kernel_str_proto;
1010#ifdef CONFIG_ARCH_HAS_NON_OVERLAPPING_ADDRESS_SPACE
1011	case BPF_FUNC_probe_read:
1012		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1013		       NULL : &bpf_probe_read_compat_proto;
1014	case BPF_FUNC_probe_read_str:
1015		return security_locked_down(LOCKDOWN_BPF_READ_KERNEL) < 0 ?
1016		       NULL : &bpf_probe_read_compat_str_proto;
1017#endif
1018#ifdef CONFIG_CGROUPS
1019	case BPF_FUNC_get_current_cgroup_id:
1020		return &bpf_get_current_cgroup_id_proto;
1021	case BPF_FUNC_get_current_ancestor_cgroup_id:
1022		return &bpf_get_current_ancestor_cgroup_id_proto;
1023#endif
1024	case BPF_FUNC_send_signal:
1025		return &bpf_send_signal_proto;
1026	case BPF_FUNC_send_signal_thread:
1027		return &bpf_send_signal_thread_proto;
1028	case BPF_FUNC_perf_event_read_value:
1029		return &bpf_perf_event_read_value_proto;
1030	case BPF_FUNC_get_ns_current_pid_tgid:
1031		return &bpf_get_ns_current_pid_tgid_proto;
1032	case BPF_FUNC_ringbuf_output:
1033		return &bpf_ringbuf_output_proto;
1034	case BPF_FUNC_ringbuf_reserve:
1035		return &bpf_ringbuf_reserve_proto;
1036	case BPF_FUNC_ringbuf_submit:
1037		return &bpf_ringbuf_submit_proto;
1038	case BPF_FUNC_ringbuf_discard:
1039		return &bpf_ringbuf_discard_proto;
1040	case BPF_FUNC_ringbuf_query:
1041		return &bpf_ringbuf_query_proto;
1042	case BPF_FUNC_jiffies64:
1043		return &bpf_jiffies64_proto;
1044	case BPF_FUNC_get_task_stack:
1045		return &bpf_get_task_stack_proto;
1046	case BPF_FUNC_copy_from_user:
1047		return prog->aux->sleepable ? &bpf_copy_from_user_proto : NULL;
1048	case BPF_FUNC_snprintf_btf:
1049		return &bpf_snprintf_btf_proto;
1050	case BPF_FUNC_per_cpu_ptr:
1051		return &bpf_per_cpu_ptr_proto;
1052	case BPF_FUNC_this_cpu_ptr:
1053		return &bpf_this_cpu_ptr_proto;
1054	case BPF_FUNC_task_storage_get:
1055		return &bpf_task_storage_get_proto;
1056	case BPF_FUNC_task_storage_delete:
1057		return &bpf_task_storage_delete_proto;
1058	case BPF_FUNC_for_each_map_elem:
1059		return &bpf_for_each_map_elem_proto;
1060	case BPF_FUNC_snprintf:
1061		return &bpf_snprintf_proto;
1062	default:
1063		return NULL;
1064	}
1065}
1066
1067static const struct bpf_func_proto *
1068kprobe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1069{
1070	switch (func_id) {
1071	case BPF_FUNC_perf_event_output:
1072		return &bpf_perf_event_output_proto;
1073	case BPF_FUNC_get_stackid:
1074		return &bpf_get_stackid_proto;
1075	case BPF_FUNC_get_stack:
1076		return &bpf_get_stack_proto;
 
 
1077#ifdef CONFIG_BPF_KPROBE_OVERRIDE
1078	case BPF_FUNC_override_return:
1079		return &bpf_override_return_proto;
1080#endif
1081	default:
1082		return bpf_tracing_func_proto(func_id, prog);
1083	}
1084}
1085
1086/* bpf+kprobe programs can access fields of 'struct pt_regs' */
1087static bool kprobe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1088					const struct bpf_prog *prog,
1089					struct bpf_insn_access_aux *info)
1090{
1091	if (off < 0 || off >= sizeof(struct pt_regs))
1092		return false;
1093	if (type != BPF_READ)
1094		return false;
1095	if (off % size != 0)
1096		return false;
1097	/*
1098	 * Assertion for 32 bit to make sure last 8 byte access
1099	 * (BPF_DW) to the last 4 byte member is disallowed.
1100	 */
1101	if (off + size > sizeof(struct pt_regs))
1102		return false;
1103
1104	return true;
1105}
1106
1107const struct bpf_verifier_ops kprobe_verifier_ops = {
1108	.get_func_proto  = kprobe_prog_func_proto,
1109	.is_valid_access = kprobe_prog_is_valid_access,
1110};
1111
1112const struct bpf_prog_ops kprobe_prog_ops = {
1113};
1114
1115BPF_CALL_5(bpf_perf_event_output_tp, void *, tp_buff, struct bpf_map *, map,
1116	   u64, flags, void *, data, u64, size)
1117{
1118	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1119
1120	/*
1121	 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
1122	 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
1123	 * from there and call the same bpf_perf_event_output() helper inline.
1124	 */
1125	return ____bpf_perf_event_output(regs, map, flags, data, size);
1126}
1127
1128static const struct bpf_func_proto bpf_perf_event_output_proto_tp = {
1129	.func		= bpf_perf_event_output_tp,
1130	.gpl_only	= true,
1131	.ret_type	= RET_INTEGER,
1132	.arg1_type	= ARG_PTR_TO_CTX,
1133	.arg2_type	= ARG_CONST_MAP_PTR,
1134	.arg3_type	= ARG_ANYTHING,
1135	.arg4_type	= ARG_PTR_TO_MEM,
1136	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1137};
1138
1139BPF_CALL_3(bpf_get_stackid_tp, void *, tp_buff, struct bpf_map *, map,
1140	   u64, flags)
1141{
1142	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1143
1144	/*
1145	 * Same comment as in bpf_perf_event_output_tp(), only that this time
1146	 * the other helper's function body cannot be inlined due to being
1147	 * external, thus we need to call raw helper function.
1148	 */
1149	return bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1150			       flags, 0, 0);
1151}
1152
1153static const struct bpf_func_proto bpf_get_stackid_proto_tp = {
1154	.func		= bpf_get_stackid_tp,
1155	.gpl_only	= true,
1156	.ret_type	= RET_INTEGER,
1157	.arg1_type	= ARG_PTR_TO_CTX,
1158	.arg2_type	= ARG_CONST_MAP_PTR,
1159	.arg3_type	= ARG_ANYTHING,
1160};
1161
1162BPF_CALL_4(bpf_get_stack_tp, void *, tp_buff, void *, buf, u32, size,
1163	   u64, flags)
1164{
1165	struct pt_regs *regs = *(struct pt_regs **)tp_buff;
1166
1167	return bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1168			     (unsigned long) size, flags, 0);
1169}
1170
1171static const struct bpf_func_proto bpf_get_stack_proto_tp = {
1172	.func		= bpf_get_stack_tp,
1173	.gpl_only	= true,
1174	.ret_type	= RET_INTEGER,
1175	.arg1_type	= ARG_PTR_TO_CTX,
1176	.arg2_type	= ARG_PTR_TO_UNINIT_MEM,
1177	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1178	.arg4_type	= ARG_ANYTHING,
1179};
1180
1181static const struct bpf_func_proto *
1182tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1183{
1184	switch (func_id) {
1185	case BPF_FUNC_perf_event_output:
1186		return &bpf_perf_event_output_proto_tp;
1187	case BPF_FUNC_get_stackid:
1188		return &bpf_get_stackid_proto_tp;
1189	case BPF_FUNC_get_stack:
1190		return &bpf_get_stack_proto_tp;
1191	default:
1192		return bpf_tracing_func_proto(func_id, prog);
1193	}
1194}
1195
1196static bool tp_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1197				    const struct bpf_prog *prog,
1198				    struct bpf_insn_access_aux *info)
1199{
1200	if (off < sizeof(void *) || off >= PERF_MAX_TRACE_SIZE)
1201		return false;
1202	if (type != BPF_READ)
1203		return false;
1204	if (off % size != 0)
1205		return false;
1206
1207	BUILD_BUG_ON(PERF_MAX_TRACE_SIZE % sizeof(__u64));
1208	return true;
1209}
1210
1211const struct bpf_verifier_ops tracepoint_verifier_ops = {
1212	.get_func_proto  = tp_prog_func_proto,
1213	.is_valid_access = tp_prog_is_valid_access,
1214};
1215
1216const struct bpf_prog_ops tracepoint_prog_ops = {
1217};
1218
1219BPF_CALL_3(bpf_perf_prog_read_value, struct bpf_perf_event_data_kern *, ctx,
1220	   struct bpf_perf_event_value *, buf, u32, size)
1221{
1222	int err = -EINVAL;
1223
1224	if (unlikely(size != sizeof(struct bpf_perf_event_value)))
1225		goto clear;
1226	err = perf_event_read_local(ctx->event, &buf->counter, &buf->enabled,
1227				    &buf->running);
1228	if (unlikely(err))
1229		goto clear;
1230	return 0;
1231clear:
1232	memset(buf, 0, size);
1233	return err;
1234}
1235
1236static const struct bpf_func_proto bpf_perf_prog_read_value_proto = {
1237         .func           = bpf_perf_prog_read_value,
1238         .gpl_only       = true,
1239         .ret_type       = RET_INTEGER,
1240         .arg1_type      = ARG_PTR_TO_CTX,
1241         .arg2_type      = ARG_PTR_TO_UNINIT_MEM,
1242         .arg3_type      = ARG_CONST_SIZE,
1243};
1244
1245BPF_CALL_4(bpf_read_branch_records, struct bpf_perf_event_data_kern *, ctx,
1246	   void *, buf, u32, size, u64, flags)
1247{
1248#ifndef CONFIG_X86
1249	return -ENOENT;
1250#else
1251	static const u32 br_entry_size = sizeof(struct perf_branch_entry);
1252	struct perf_branch_stack *br_stack = ctx->data->br_stack;
1253	u32 to_copy;
1254
1255	if (unlikely(flags & ~BPF_F_GET_BRANCH_RECORDS_SIZE))
1256		return -EINVAL;
1257
1258	if (unlikely(!br_stack))
1259		return -EINVAL;
1260
1261	if (flags & BPF_F_GET_BRANCH_RECORDS_SIZE)
1262		return br_stack->nr * br_entry_size;
1263
1264	if (!buf || (size % br_entry_size != 0))
1265		return -EINVAL;
1266
1267	to_copy = min_t(u32, br_stack->nr * br_entry_size, size);
1268	memcpy(buf, br_stack->entries, to_copy);
1269
1270	return to_copy;
1271#endif
1272}
1273
1274static const struct bpf_func_proto bpf_read_branch_records_proto = {
1275	.func           = bpf_read_branch_records,
1276	.gpl_only       = true,
1277	.ret_type       = RET_INTEGER,
1278	.arg1_type      = ARG_PTR_TO_CTX,
1279	.arg2_type      = ARG_PTR_TO_MEM_OR_NULL,
1280	.arg3_type      = ARG_CONST_SIZE_OR_ZERO,
1281	.arg4_type      = ARG_ANYTHING,
1282};
1283
1284static const struct bpf_func_proto *
1285pe_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1286{
1287	switch (func_id) {
1288	case BPF_FUNC_perf_event_output:
1289		return &bpf_perf_event_output_proto_tp;
1290	case BPF_FUNC_get_stackid:
1291		return &bpf_get_stackid_proto_pe;
1292	case BPF_FUNC_get_stack:
1293		return &bpf_get_stack_proto_pe;
1294	case BPF_FUNC_perf_prog_read_value:
1295		return &bpf_perf_prog_read_value_proto;
1296	case BPF_FUNC_read_branch_records:
1297		return &bpf_read_branch_records_proto;
1298	default:
1299		return bpf_tracing_func_proto(func_id, prog);
1300	}
1301}
1302
1303/*
1304 * bpf_raw_tp_regs are separate from bpf_pt_regs used from skb/xdp
1305 * to avoid potential recursive reuse issue when/if tracepoints are added
1306 * inside bpf_*_event_output, bpf_get_stackid and/or bpf_get_stack.
1307 *
1308 * Since raw tracepoints run despite bpf_prog_active, support concurrent usage
1309 * in normal, irq, and nmi context.
1310 */
1311struct bpf_raw_tp_regs {
1312	struct pt_regs regs[3];
1313};
1314static DEFINE_PER_CPU(struct bpf_raw_tp_regs, bpf_raw_tp_regs);
1315static DEFINE_PER_CPU(int, bpf_raw_tp_nest_level);
1316static struct pt_regs *get_bpf_raw_tp_regs(void)
1317{
1318	struct bpf_raw_tp_regs *tp_regs = this_cpu_ptr(&bpf_raw_tp_regs);
1319	int nest_level = this_cpu_inc_return(bpf_raw_tp_nest_level);
1320
1321	if (WARN_ON_ONCE(nest_level > ARRAY_SIZE(tp_regs->regs))) {
1322		this_cpu_dec(bpf_raw_tp_nest_level);
1323		return ERR_PTR(-EBUSY);
1324	}
1325
1326	return &tp_regs->regs[nest_level - 1];
1327}
1328
1329static void put_bpf_raw_tp_regs(void)
1330{
1331	this_cpu_dec(bpf_raw_tp_nest_level);
1332}
1333
1334BPF_CALL_5(bpf_perf_event_output_raw_tp, struct bpf_raw_tracepoint_args *, args,
1335	   struct bpf_map *, map, u64, flags, void *, data, u64, size)
1336{
1337	struct pt_regs *regs = get_bpf_raw_tp_regs();
1338	int ret;
1339
1340	if (IS_ERR(regs))
1341		return PTR_ERR(regs);
1342
1343	perf_fetch_caller_regs(regs);
1344	ret = ____bpf_perf_event_output(regs, map, flags, data, size);
1345
1346	put_bpf_raw_tp_regs();
1347	return ret;
1348}
1349
1350static const struct bpf_func_proto bpf_perf_event_output_proto_raw_tp = {
1351	.func		= bpf_perf_event_output_raw_tp,
1352	.gpl_only	= true,
1353	.ret_type	= RET_INTEGER,
1354	.arg1_type	= ARG_PTR_TO_CTX,
1355	.arg2_type	= ARG_CONST_MAP_PTR,
1356	.arg3_type	= ARG_ANYTHING,
1357	.arg4_type	= ARG_PTR_TO_MEM,
1358	.arg5_type	= ARG_CONST_SIZE_OR_ZERO,
1359};
1360
1361extern const struct bpf_func_proto bpf_skb_output_proto;
1362extern const struct bpf_func_proto bpf_xdp_output_proto;
1363
1364BPF_CALL_3(bpf_get_stackid_raw_tp, struct bpf_raw_tracepoint_args *, args,
1365	   struct bpf_map *, map, u64, flags)
1366{
1367	struct pt_regs *regs = get_bpf_raw_tp_regs();
1368	int ret;
1369
1370	if (IS_ERR(regs))
1371		return PTR_ERR(regs);
1372
1373	perf_fetch_caller_regs(regs);
1374	/* similar to bpf_perf_event_output_tp, but pt_regs fetched differently */
1375	ret = bpf_get_stackid((unsigned long) regs, (unsigned long) map,
1376			      flags, 0, 0);
1377	put_bpf_raw_tp_regs();
1378	return ret;
1379}
1380
1381static const struct bpf_func_proto bpf_get_stackid_proto_raw_tp = {
1382	.func		= bpf_get_stackid_raw_tp,
1383	.gpl_only	= true,
1384	.ret_type	= RET_INTEGER,
1385	.arg1_type	= ARG_PTR_TO_CTX,
1386	.arg2_type	= ARG_CONST_MAP_PTR,
1387	.arg3_type	= ARG_ANYTHING,
1388};
1389
1390BPF_CALL_4(bpf_get_stack_raw_tp, struct bpf_raw_tracepoint_args *, args,
1391	   void *, buf, u32, size, u64, flags)
1392{
1393	struct pt_regs *regs = get_bpf_raw_tp_regs();
1394	int ret;
1395
1396	if (IS_ERR(regs))
1397		return PTR_ERR(regs);
1398
1399	perf_fetch_caller_regs(regs);
1400	ret = bpf_get_stack((unsigned long) regs, (unsigned long) buf,
1401			    (unsigned long) size, flags, 0);
1402	put_bpf_raw_tp_regs();
1403	return ret;
1404}
1405
1406static const struct bpf_func_proto bpf_get_stack_proto_raw_tp = {
1407	.func		= bpf_get_stack_raw_tp,
1408	.gpl_only	= true,
1409	.ret_type	= RET_INTEGER,
1410	.arg1_type	= ARG_PTR_TO_CTX,
1411	.arg2_type	= ARG_PTR_TO_MEM,
1412	.arg3_type	= ARG_CONST_SIZE_OR_ZERO,
1413	.arg4_type	= ARG_ANYTHING,
1414};
1415
1416static const struct bpf_func_proto *
1417raw_tp_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1418{
1419	switch (func_id) {
1420	case BPF_FUNC_perf_event_output:
1421		return &bpf_perf_event_output_proto_raw_tp;
1422	case BPF_FUNC_get_stackid:
1423		return &bpf_get_stackid_proto_raw_tp;
1424	case BPF_FUNC_get_stack:
1425		return &bpf_get_stack_proto_raw_tp;
1426	default:
1427		return bpf_tracing_func_proto(func_id, prog);
1428	}
1429}
1430
1431const struct bpf_func_proto *
1432tracing_prog_func_proto(enum bpf_func_id func_id, const struct bpf_prog *prog)
1433{
1434	switch (func_id) {
1435#ifdef CONFIG_NET
1436	case BPF_FUNC_skb_output:
1437		return &bpf_skb_output_proto;
1438	case BPF_FUNC_xdp_output:
1439		return &bpf_xdp_output_proto;
1440	case BPF_FUNC_skc_to_tcp6_sock:
1441		return &bpf_skc_to_tcp6_sock_proto;
1442	case BPF_FUNC_skc_to_tcp_sock:
1443		return &bpf_skc_to_tcp_sock_proto;
1444	case BPF_FUNC_skc_to_tcp_timewait_sock:
1445		return &bpf_skc_to_tcp_timewait_sock_proto;
1446	case BPF_FUNC_skc_to_tcp_request_sock:
1447		return &bpf_skc_to_tcp_request_sock_proto;
1448	case BPF_FUNC_skc_to_udp6_sock:
1449		return &bpf_skc_to_udp6_sock_proto;
1450	case BPF_FUNC_sk_storage_get:
1451		return &bpf_sk_storage_get_tracing_proto;
1452	case BPF_FUNC_sk_storage_delete:
1453		return &bpf_sk_storage_delete_tracing_proto;
1454	case BPF_FUNC_sock_from_file:
1455		return &bpf_sock_from_file_proto;
1456	case BPF_FUNC_get_socket_cookie:
1457		return &bpf_get_socket_ptr_cookie_proto;
1458#endif
1459	case BPF_FUNC_seq_printf:
1460		return prog->expected_attach_type == BPF_TRACE_ITER ?
1461		       &bpf_seq_printf_proto :
1462		       NULL;
1463	case BPF_FUNC_seq_write:
1464		return prog->expected_attach_type == BPF_TRACE_ITER ?
1465		       &bpf_seq_write_proto :
1466		       NULL;
1467	case BPF_FUNC_seq_printf_btf:
1468		return prog->expected_attach_type == BPF_TRACE_ITER ?
1469		       &bpf_seq_printf_btf_proto :
1470		       NULL;
1471	case BPF_FUNC_d_path:
1472		return &bpf_d_path_proto;
1473	default:
1474		return raw_tp_prog_func_proto(func_id, prog);
1475	}
1476}
1477
1478static bool raw_tp_prog_is_valid_access(int off, int size,
1479					enum bpf_access_type type,
1480					const struct bpf_prog *prog,
1481					struct bpf_insn_access_aux *info)
1482{
1483	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
 
1484		return false;
1485	if (type != BPF_READ)
1486		return false;
1487	if (off % size != 0)
1488		return false;
1489	return true;
1490}
1491
1492static bool tracing_prog_is_valid_access(int off, int size,
1493					 enum bpf_access_type type,
1494					 const struct bpf_prog *prog,
1495					 struct bpf_insn_access_aux *info)
1496{
1497	if (off < 0 || off >= sizeof(__u64) * MAX_BPF_FUNC_ARGS)
1498		return false;
1499	if (type != BPF_READ)
1500		return false;
1501	if (off % size != 0)
1502		return false;
1503	return btf_ctx_access(off, size, type, prog, info);
1504}
1505
1506int __weak bpf_prog_test_run_tracing(struct bpf_prog *prog,
1507				     const union bpf_attr *kattr,
1508				     union bpf_attr __user *uattr)
1509{
1510	return -ENOTSUPP;
1511}
1512
1513const struct bpf_verifier_ops raw_tracepoint_verifier_ops = {
1514	.get_func_proto  = raw_tp_prog_func_proto,
1515	.is_valid_access = raw_tp_prog_is_valid_access,
1516};
1517
1518const struct bpf_prog_ops raw_tracepoint_prog_ops = {
1519#ifdef CONFIG_NET
1520	.test_run = bpf_prog_test_run_raw_tp,
1521#endif
1522};
1523
1524const struct bpf_verifier_ops tracing_verifier_ops = {
1525	.get_func_proto  = tracing_prog_func_proto,
1526	.is_valid_access = tracing_prog_is_valid_access,
1527};
1528
1529const struct bpf_prog_ops tracing_prog_ops = {
1530	.test_run = bpf_prog_test_run_tracing,
1531};
1532
1533static bool raw_tp_writable_prog_is_valid_access(int off, int size,
1534						 enum bpf_access_type type,
1535						 const struct bpf_prog *prog,
1536						 struct bpf_insn_access_aux *info)
1537{
1538	if (off == 0) {
1539		if (size != sizeof(u64) || type != BPF_READ)
1540			return false;
1541		info->reg_type = PTR_TO_TP_BUFFER;
1542	}
1543	return raw_tp_prog_is_valid_access(off, size, type, prog, info);
1544}
1545
1546const struct bpf_verifier_ops raw_tracepoint_writable_verifier_ops = {
1547	.get_func_proto  = raw_tp_prog_func_proto,
1548	.is_valid_access = raw_tp_writable_prog_is_valid_access,
1549};
1550
1551const struct bpf_prog_ops raw_tracepoint_writable_prog_ops = {
1552};
1553
1554static bool pe_prog_is_valid_access(int off, int size, enum bpf_access_type type,
1555				    const struct bpf_prog *prog,
1556				    struct bpf_insn_access_aux *info)
1557{
1558	const int size_u64 = sizeof(u64);
1559
1560	if (off < 0 || off >= sizeof(struct bpf_perf_event_data))
1561		return false;
1562	if (type != BPF_READ)
1563		return false;
1564	if (off % size != 0) {
1565		if (sizeof(unsigned long) != 4)
1566			return false;
1567		if (size != 8)
1568			return false;
1569		if (off % size != 4)
1570			return false;
1571	}
1572
1573	switch (off) {
1574	case bpf_ctx_range(struct bpf_perf_event_data, sample_period):
1575		bpf_ctx_record_field_size(info, size_u64);
1576		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1577			return false;
1578		break;
1579	case bpf_ctx_range(struct bpf_perf_event_data, addr):
1580		bpf_ctx_record_field_size(info, size_u64);
1581		if (!bpf_ctx_narrow_access_ok(off, size, size_u64))
1582			return false;
1583		break;
1584	default:
1585		if (size != sizeof(long))
1586			return false;
1587	}
1588
1589	return true;
1590}
1591
1592static u32 pe_prog_convert_ctx_access(enum bpf_access_type type,
1593				      const struct bpf_insn *si,
1594				      struct bpf_insn *insn_buf,
1595				      struct bpf_prog *prog, u32 *target_size)
1596{
1597	struct bpf_insn *insn = insn_buf;
1598
1599	switch (si->off) {
1600	case offsetof(struct bpf_perf_event_data, sample_period):
1601		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1602						       data), si->dst_reg, si->src_reg,
1603				      offsetof(struct bpf_perf_event_data_kern, data));
1604		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1605				      bpf_target_off(struct perf_sample_data, period, 8,
1606						     target_size));
1607		break;
1608	case offsetof(struct bpf_perf_event_data, addr):
1609		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1610						       data), si->dst_reg, si->src_reg,
1611				      offsetof(struct bpf_perf_event_data_kern, data));
1612		*insn++ = BPF_LDX_MEM(BPF_DW, si->dst_reg, si->dst_reg,
1613				      bpf_target_off(struct perf_sample_data, addr, 8,
1614						     target_size));
1615		break;
1616	default:
1617		*insn++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern,
1618						       regs), si->dst_reg, si->src_reg,
1619				      offsetof(struct bpf_perf_event_data_kern, regs));
1620		*insn++ = BPF_LDX_MEM(BPF_SIZEOF(long), si->dst_reg, si->dst_reg,
1621				      si->off);
1622		break;
1623	}
1624
1625	return insn - insn_buf;
1626}
1627
1628const struct bpf_verifier_ops perf_event_verifier_ops = {
1629	.get_func_proto		= pe_prog_func_proto,
1630	.is_valid_access	= pe_prog_is_valid_access,
1631	.convert_ctx_access	= pe_prog_convert_ctx_access,
1632};
1633
1634const struct bpf_prog_ops perf_event_prog_ops = {
1635};
1636
1637static DEFINE_MUTEX(bpf_event_mutex);
1638
1639#define BPF_TRACE_MAX_PROGS 64
1640
1641int perf_event_attach_bpf_prog(struct perf_event *event,
1642			       struct bpf_prog *prog)
1643{
1644	struct bpf_prog_array *old_array;
1645	struct bpf_prog_array *new_array;
1646	int ret = -EEXIST;
1647
1648	/*
1649	 * Kprobe override only works if they are on the function entry,
1650	 * and only if they are on the opt-in list.
1651	 */
1652	if (prog->kprobe_override &&
1653	    (!trace_kprobe_on_func_entry(event->tp_event) ||
1654	     !trace_kprobe_error_injectable(event->tp_event)))
1655		return -EINVAL;
1656
1657	mutex_lock(&bpf_event_mutex);
1658
1659	if (event->prog)
1660		goto unlock;
1661
1662	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1663	if (old_array &&
1664	    bpf_prog_array_length(old_array) >= BPF_TRACE_MAX_PROGS) {
1665		ret = -E2BIG;
1666		goto unlock;
1667	}
1668
1669	ret = bpf_prog_array_copy(old_array, NULL, prog, &new_array);
1670	if (ret < 0)
1671		goto unlock;
1672
1673	/* set the new array to event->tp_event and set event->prog */
1674	event->prog = prog;
1675	rcu_assign_pointer(event->tp_event->prog_array, new_array);
1676	bpf_prog_array_free(old_array);
1677
1678unlock:
1679	mutex_unlock(&bpf_event_mutex);
1680	return ret;
1681}
1682
1683void perf_event_detach_bpf_prog(struct perf_event *event)
1684{
1685	struct bpf_prog_array *old_array;
1686	struct bpf_prog_array *new_array;
1687	int ret;
1688
1689	mutex_lock(&bpf_event_mutex);
1690
1691	if (!event->prog)
1692		goto unlock;
1693
1694	old_array = bpf_event_rcu_dereference(event->tp_event->prog_array);
1695	ret = bpf_prog_array_copy(old_array, event->prog, NULL, &new_array);
1696	if (ret == -ENOENT)
1697		goto unlock;
1698	if (ret < 0) {
1699		bpf_prog_array_delete_safe(old_array, event->prog);
1700	} else {
1701		rcu_assign_pointer(event->tp_event->prog_array, new_array);
1702		bpf_prog_array_free(old_array);
1703	}
1704
1705	bpf_prog_put(event->prog);
1706	event->prog = NULL;
1707
1708unlock:
1709	mutex_unlock(&bpf_event_mutex);
1710}
1711
1712int perf_event_query_prog_array(struct perf_event *event, void __user *info)
1713{
1714	struct perf_event_query_bpf __user *uquery = info;
1715	struct perf_event_query_bpf query = {};
1716	struct bpf_prog_array *progs;
1717	u32 *ids, prog_cnt, ids_len;
1718	int ret;
1719
1720	if (!perfmon_capable())
1721		return -EPERM;
1722	if (event->attr.type != PERF_TYPE_TRACEPOINT)
1723		return -EINVAL;
1724	if (copy_from_user(&query, uquery, sizeof(query)))
1725		return -EFAULT;
1726
1727	ids_len = query.ids_len;
1728	if (ids_len > BPF_TRACE_MAX_PROGS)
1729		return -E2BIG;
1730	ids = kcalloc(ids_len, sizeof(u32), GFP_USER | __GFP_NOWARN);
1731	if (!ids)
1732		return -ENOMEM;
1733	/*
1734	 * The above kcalloc returns ZERO_SIZE_PTR when ids_len = 0, which
1735	 * is required when user only wants to check for uquery->prog_cnt.
1736	 * There is no need to check for it since the case is handled
1737	 * gracefully in bpf_prog_array_copy_info.
1738	 */
1739
1740	mutex_lock(&bpf_event_mutex);
1741	progs = bpf_event_rcu_dereference(event->tp_event->prog_array);
1742	ret = bpf_prog_array_copy_info(progs, ids, ids_len, &prog_cnt);
1743	mutex_unlock(&bpf_event_mutex);
1744
1745	if (copy_to_user(&uquery->prog_cnt, &prog_cnt, sizeof(prog_cnt)) ||
1746	    copy_to_user(uquery->ids, ids, ids_len * sizeof(u32)))
1747		ret = -EFAULT;
1748
1749	kfree(ids);
1750	return ret;
1751}
1752
1753extern struct bpf_raw_event_map __start__bpf_raw_tp[];
1754extern struct bpf_raw_event_map __stop__bpf_raw_tp[];
1755
1756struct bpf_raw_event_map *bpf_get_raw_tracepoint(const char *name)
1757{
1758	struct bpf_raw_event_map *btp = __start__bpf_raw_tp;
1759
1760	for (; btp < __stop__bpf_raw_tp; btp++) {
1761		if (!strcmp(btp->tp->name, name))
1762			return btp;
1763	}
1764
1765	return bpf_get_raw_tracepoint_module(name);
1766}
1767
1768void bpf_put_raw_tracepoint(struct bpf_raw_event_map *btp)
1769{
1770	struct module *mod;
1771
1772	preempt_disable();
1773	mod = __module_address((unsigned long)btp);
1774	module_put(mod);
1775	preempt_enable();
1776}
1777
1778static __always_inline
1779void __bpf_trace_run(struct bpf_prog *prog, u64 *args)
1780{
1781	cant_sleep();
1782	rcu_read_lock();
 
1783	(void) BPF_PROG_RUN(prog, args);
 
1784	rcu_read_unlock();
1785}
1786
1787#define UNPACK(...)			__VA_ARGS__
1788#define REPEAT_1(FN, DL, X, ...)	FN(X)
1789#define REPEAT_2(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_1(FN, DL, __VA_ARGS__)
1790#define REPEAT_3(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_2(FN, DL, __VA_ARGS__)
1791#define REPEAT_4(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_3(FN, DL, __VA_ARGS__)
1792#define REPEAT_5(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_4(FN, DL, __VA_ARGS__)
1793#define REPEAT_6(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_5(FN, DL, __VA_ARGS__)
1794#define REPEAT_7(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_6(FN, DL, __VA_ARGS__)
1795#define REPEAT_8(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_7(FN, DL, __VA_ARGS__)
1796#define REPEAT_9(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_8(FN, DL, __VA_ARGS__)
1797#define REPEAT_10(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_9(FN, DL, __VA_ARGS__)
1798#define REPEAT_11(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_10(FN, DL, __VA_ARGS__)
1799#define REPEAT_12(FN, DL, X, ...)	FN(X) UNPACK DL REPEAT_11(FN, DL, __VA_ARGS__)
1800#define REPEAT(X, FN, DL, ...)		REPEAT_##X(FN, DL, __VA_ARGS__)
1801
1802#define SARG(X)		u64 arg##X
1803#define COPY(X)		args[X] = arg##X
1804
1805#define __DL_COM	(,)
1806#define __DL_SEM	(;)
1807
1808#define __SEQ_0_11	0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11
1809
1810#define BPF_TRACE_DEFN_x(x)						\
1811	void bpf_trace_run##x(struct bpf_prog *prog,			\
1812			      REPEAT(x, SARG, __DL_COM, __SEQ_0_11))	\
1813	{								\
1814		u64 args[x];						\
1815		REPEAT(x, COPY, __DL_SEM, __SEQ_0_11);			\
1816		__bpf_trace_run(prog, args);				\
1817	}								\
1818	EXPORT_SYMBOL_GPL(bpf_trace_run##x)
1819BPF_TRACE_DEFN_x(1);
1820BPF_TRACE_DEFN_x(2);
1821BPF_TRACE_DEFN_x(3);
1822BPF_TRACE_DEFN_x(4);
1823BPF_TRACE_DEFN_x(5);
1824BPF_TRACE_DEFN_x(6);
1825BPF_TRACE_DEFN_x(7);
1826BPF_TRACE_DEFN_x(8);
1827BPF_TRACE_DEFN_x(9);
1828BPF_TRACE_DEFN_x(10);
1829BPF_TRACE_DEFN_x(11);
1830BPF_TRACE_DEFN_x(12);
1831
1832static int __bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1833{
1834	struct tracepoint *tp = btp->tp;
1835
1836	/*
1837	 * check that program doesn't access arguments beyond what's
1838	 * available in this tracepoint
1839	 */
1840	if (prog->aux->max_ctx_offset > btp->num_args * sizeof(u64))
1841		return -EINVAL;
1842
1843	if (prog->aux->max_tp_access > btp->writable_size)
1844		return -EINVAL;
1845
1846	return tracepoint_probe_register_may_exist(tp, (void *)btp->bpf_func,
1847						   prog);
1848}
1849
1850int bpf_probe_register(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1851{
1852	return __bpf_probe_register(btp, prog);
1853}
1854
1855int bpf_probe_unregister(struct bpf_raw_event_map *btp, struct bpf_prog *prog)
1856{
1857	return tracepoint_probe_unregister(btp->tp, (void *)btp->bpf_func, prog);
1858}
1859
1860int bpf_get_perf_event_info(const struct perf_event *event, u32 *prog_id,
1861			    u32 *fd_type, const char **buf,
1862			    u64 *probe_offset, u64 *probe_addr)
1863{
1864	bool is_tracepoint, is_syscall_tp;
1865	struct bpf_prog *prog;
1866	int flags, err = 0;
1867
1868	prog = event->prog;
1869	if (!prog)
1870		return -ENOENT;
1871
1872	/* not supporting BPF_PROG_TYPE_PERF_EVENT yet */
1873	if (prog->type == BPF_PROG_TYPE_PERF_EVENT)
1874		return -EOPNOTSUPP;
1875
1876	*prog_id = prog->aux->id;
1877	flags = event->tp_event->flags;
1878	is_tracepoint = flags & TRACE_EVENT_FL_TRACEPOINT;
1879	is_syscall_tp = is_syscall_trace_event(event->tp_event);
1880
1881	if (is_tracepoint || is_syscall_tp) {
1882		*buf = is_tracepoint ? event->tp_event->tp->name
1883				     : event->tp_event->name;
1884		*fd_type = BPF_FD_TYPE_TRACEPOINT;
1885		*probe_offset = 0x0;
1886		*probe_addr = 0x0;
1887	} else {
1888		/* kprobe/uprobe */
1889		err = -EOPNOTSUPP;
1890#ifdef CONFIG_KPROBE_EVENTS
1891		if (flags & TRACE_EVENT_FL_KPROBE)
1892			err = bpf_get_kprobe_info(event, fd_type, buf,
1893						  probe_offset, probe_addr,
1894						  event->attr.type == PERF_TYPE_TRACEPOINT);
1895#endif
1896#ifdef CONFIG_UPROBE_EVENTS
1897		if (flags & TRACE_EVENT_FL_UPROBE)
1898			err = bpf_get_uprobe_info(event, fd_type, buf,
1899						  probe_offset,
1900						  event->attr.type == PERF_TYPE_TRACEPOINT);
1901#endif
1902	}
1903
1904	return err;
1905}
1906
1907static int __init send_signal_irq_work_init(void)
1908{
1909	int cpu;
1910	struct send_signal_irq_work *work;
1911
1912	for_each_possible_cpu(cpu) {
1913		work = per_cpu_ptr(&send_signal_work, cpu);
1914		init_irq_work(&work->irq_work, do_bpf_send_signal);
1915	}
1916	return 0;
1917}
1918
1919subsys_initcall(send_signal_irq_work_init);
1920
1921#ifdef CONFIG_MODULES
1922static int bpf_event_notify(struct notifier_block *nb, unsigned long op,
1923			    void *module)
1924{
1925	struct bpf_trace_module *btm, *tmp;
1926	struct module *mod = module;
1927	int ret = 0;
1928
1929	if (mod->num_bpf_raw_events == 0 ||
1930	    (op != MODULE_STATE_COMING && op != MODULE_STATE_GOING))
1931		goto out;
1932
1933	mutex_lock(&bpf_module_mutex);
1934
1935	switch (op) {
1936	case MODULE_STATE_COMING:
1937		btm = kzalloc(sizeof(*btm), GFP_KERNEL);
1938		if (btm) {
1939			btm->module = module;
1940			list_add(&btm->list, &bpf_trace_modules);
1941		} else {
1942			ret = -ENOMEM;
1943		}
1944		break;
1945	case MODULE_STATE_GOING:
1946		list_for_each_entry_safe(btm, tmp, &bpf_trace_modules, list) {
1947			if (btm->module == module) {
1948				list_del(&btm->list);
1949				kfree(btm);
1950				break;
1951			}
1952		}
1953		break;
1954	}
1955
1956	mutex_unlock(&bpf_module_mutex);
1957
1958out:
1959	return notifier_from_errno(ret);
1960}
1961
1962static struct notifier_block bpf_module_nb = {
1963	.notifier_call = bpf_event_notify,
1964};
1965
1966static int __init bpf_event_init(void)
1967{
1968	register_module_notifier(&bpf_module_nb);
1969	return 0;
1970}
1971
1972fs_initcall(bpf_event_init);
1973#endif /* CONFIG_MODULES */