1 /* Copyright (c) 2011-2015 PLUMgrid, http://plumgrid.com
2 * Copyright (c) 2016 Facebook
4 * This program is free software; you can redistribute it and/or
5 * modify it under the terms of version 2 of the GNU General Public
6 * License as published by the Free Software Foundation.
8 #include <linux/kernel.h>
9 #include <linux/types.h>
10 #include <linux/slab.h>
11 #include <linux/bpf.h>
12 #include <linux/bpf_perf_event.h>
13 #include <linux/filter.h>
14 #include <linux/uaccess.h>
15 #include <linux/ctype.h>
19 * trace_call_bpf - invoke BPF program
21 * @ctx: opaque context pointer
23 * kprobe handlers execute BPF programs via this helper.
24 * Can be used from static tracepoints in the future.
26 * Return: BPF programs always return an integer which is interpreted by
28 * 0 - return from kprobe (event is filtered out)
29 * 1 - store kprobe event into ring buffer
30 * Other values are reserved and currently alias to 1
32 unsigned int trace_call_bpf(struct bpf_prog
*prog
, void *ctx
)
36 if (in_nmi()) /* not supported yet */
41 if (unlikely(__this_cpu_inc_return(bpf_prog_active
) != 1)) {
43 * since some bpf program is already running on this cpu,
44 * don't call into another bpf program (same or different)
45 * and don't send kprobe event into ring-buffer,
53 ret
= BPF_PROG_RUN(prog
, ctx
);
57 __this_cpu_dec(bpf_prog_active
);
62 EXPORT_SYMBOL_GPL(trace_call_bpf
);
64 BPF_CALL_3(bpf_probe_read
, void *, dst
, u32
, size
, const void *, unsafe_ptr
)
68 ret
= probe_kernel_read(dst
, unsafe_ptr
, size
);
69 if (unlikely(ret
< 0))
75 static const struct bpf_func_proto bpf_probe_read_proto
= {
76 .func
= bpf_probe_read
,
78 .ret_type
= RET_INTEGER
,
79 .arg1_type
= ARG_PTR_TO_UNINIT_MEM
,
80 .arg2_type
= ARG_CONST_SIZE
,
81 .arg3_type
= ARG_ANYTHING
,
84 BPF_CALL_3(bpf_probe_write_user
, void *, unsafe_ptr
, const void *, src
,
88 * Ensure we're in user context which is safe for the helper to
89 * run. This helper has no business in a kthread.
91 * access_ok() should prevent writing to non-user memory, but in
92 * some situations (nommu, temporary switch, etc) access_ok() does
93 * not provide enough validation, hence the check on KERNEL_DS.
96 if (unlikely(in_interrupt() ||
97 current
->flags
& (PF_KTHREAD
| PF_EXITING
)))
99 if (unlikely(uaccess_kernel()))
101 if (!access_ok(VERIFY_WRITE
, unsafe_ptr
, size
))
104 return probe_kernel_write(unsafe_ptr
, src
, size
);
107 static const struct bpf_func_proto bpf_probe_write_user_proto
= {
108 .func
= bpf_probe_write_user
,
110 .ret_type
= RET_INTEGER
,
111 .arg1_type
= ARG_ANYTHING
,
112 .arg2_type
= ARG_PTR_TO_MEM
,
113 .arg3_type
= ARG_CONST_SIZE
,
116 static const struct bpf_func_proto
*bpf_get_probe_write_proto(void)
118 pr_warn_ratelimited("%s[%d] is installing a program with bpf_probe_write_user helper that may corrupt user memory!",
119 current
->comm
, task_pid_nr(current
));
121 return &bpf_probe_write_user_proto
;
125 * Only limited trace_printk() conversion specifiers allowed:
126 * %d %i %u %x %ld %li %lu %lx %lld %lli %llu %llx %p %s
128 BPF_CALL_5(bpf_trace_printk
, char *, fmt
, u32
, fmt_size
, u64
, arg1
,
129 u64
, arg2
, u64
, arg3
)
131 bool str_seen
= false;
139 * bpf_check()->check_func_arg()->check_stack_boundary()
140 * guarantees that fmt points to bpf program stack,
141 * fmt_size bytes of it were initialized and fmt_size > 0
143 if (fmt
[--fmt_size
] != 0)
146 /* check format string for allowed specifiers */
147 for (i
= 0; i
< fmt_size
; i
++) {
148 if ((!isprint(fmt
[i
]) && !isspace(fmt
[i
])) || !isascii(fmt
[i
]))
157 /* fmt[i] != 0 && fmt[last] == 0, so we can access fmt[i + 1] */
162 } else if (fmt
[i
] == 'p' || fmt
[i
] == 's') {
165 if (!isspace(fmt
[i
]) && !ispunct(fmt
[i
]) && fmt
[i
] != 0)
168 if (fmt
[i
- 1] == 's') {
170 /* allow only one '%s' per fmt string */
189 strncpy_from_unsafe(buf
,
190 (void *) (long) unsafe_addr
,
201 if (fmt
[i
] != 'i' && fmt
[i
] != 'd' &&
202 fmt
[i
] != 'u' && fmt
[i
] != 'x')
207 /* Horrid workaround for getting va_list handling working with different
208 * argument type combinations generically for 32 and 64 bit archs.
210 #define __BPF_TP_EMIT() __BPF_ARG3_TP()
211 #define __BPF_TP(...) \
212 __trace_printk(1 /* Fake ip will not be printed. */, \
215 #define __BPF_ARG1_TP(...) \
216 ((mod[0] == 2 || (mod[0] == 1 && __BITS_PER_LONG == 64)) \
217 ? __BPF_TP(arg1, ##__VA_ARGS__) \
218 : ((mod[0] == 1 || (mod[0] == 0 && __BITS_PER_LONG == 32)) \
219 ? __BPF_TP((long)arg1, ##__VA_ARGS__) \
220 : __BPF_TP((u32)arg1, ##__VA_ARGS__)))
222 #define __BPF_ARG2_TP(...) \
223 ((mod[1] == 2 || (mod[1] == 1 && __BITS_PER_LONG == 64)) \
224 ? __BPF_ARG1_TP(arg2, ##__VA_ARGS__) \
225 : ((mod[1] == 1 || (mod[1] == 0 && __BITS_PER_LONG == 32)) \
226 ? __BPF_ARG1_TP((long)arg2, ##__VA_ARGS__) \
227 : __BPF_ARG1_TP((u32)arg2, ##__VA_ARGS__)))
229 #define __BPF_ARG3_TP(...) \
230 ((mod[2] == 2 || (mod[2] == 1 && __BITS_PER_LONG == 64)) \
231 ? __BPF_ARG2_TP(arg3, ##__VA_ARGS__) \
232 : ((mod[2] == 1 || (mod[2] == 0 && __BITS_PER_LONG == 32)) \
233 ? __BPF_ARG2_TP((long)arg3, ##__VA_ARGS__) \
234 : __BPF_ARG2_TP((u32)arg3, ##__VA_ARGS__)))
236 return __BPF_TP_EMIT();
239 static const struct bpf_func_proto bpf_trace_printk_proto
= {
240 .func
= bpf_trace_printk
,
242 .ret_type
= RET_INTEGER
,
243 .arg1_type
= ARG_PTR_TO_MEM
,
244 .arg2_type
= ARG_CONST_SIZE
,
247 const struct bpf_func_proto
*bpf_get_trace_printk_proto(void)
250 * this program might be calling bpf_trace_printk,
251 * so allocate per-cpu printk buffers
253 trace_printk_init_buffers();
255 return &bpf_trace_printk_proto
;
258 static __always_inline
int
259 get_map_perf_counter(struct bpf_map
*map
, u64 flags
,
260 u64
*value
, u64
*enabled
, u64
*running
)
262 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
263 unsigned int cpu
= smp_processor_id();
264 u64 index
= flags
& BPF_F_INDEX_MASK
;
265 struct bpf_event_entry
*ee
;
267 if (unlikely(flags
& ~(BPF_F_INDEX_MASK
)))
269 if (index
== BPF_F_CURRENT_CPU
)
271 if (unlikely(index
>= array
->map
.max_entries
))
274 ee
= READ_ONCE(array
->ptrs
[index
]);
278 return perf_event_read_local(ee
->event
, value
, enabled
, running
);
281 BPF_CALL_2(bpf_perf_event_read
, struct bpf_map
*, map
, u64
, flags
)
286 err
= get_map_perf_counter(map
, flags
, &value
, NULL
, NULL
);
288 * this api is ugly since we miss [-22..-2] range of valid
289 * counter values, but that's uapi
296 static const struct bpf_func_proto bpf_perf_event_read_proto
= {
297 .func
= bpf_perf_event_read
,
299 .ret_type
= RET_INTEGER
,
300 .arg1_type
= ARG_CONST_MAP_PTR
,
301 .arg2_type
= ARG_ANYTHING
,
304 BPF_CALL_4(bpf_perf_event_read_value
, struct bpf_map
*, map
, u64
, flags
,
305 struct bpf_perf_event_value
*, buf
, u32
, size
)
309 if (unlikely(size
!= sizeof(struct bpf_perf_event_value
)))
311 err
= get_map_perf_counter(map
, flags
, &buf
->counter
, &buf
->enabled
,
317 memset(buf
, 0, size
);
321 static const struct bpf_func_proto bpf_perf_event_read_value_proto
= {
322 .func
= bpf_perf_event_read_value
,
324 .ret_type
= RET_INTEGER
,
325 .arg1_type
= ARG_CONST_MAP_PTR
,
326 .arg2_type
= ARG_ANYTHING
,
327 .arg3_type
= ARG_PTR_TO_UNINIT_MEM
,
328 .arg4_type
= ARG_CONST_SIZE
,
331 static DEFINE_PER_CPU(struct perf_sample_data
, bpf_sd
);
333 static __always_inline u64
334 __bpf_perf_event_output(struct pt_regs
*regs
, struct bpf_map
*map
,
335 u64 flags
, struct perf_raw_record
*raw
)
337 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
338 struct perf_sample_data
*sd
= this_cpu_ptr(&bpf_sd
);
339 unsigned int cpu
= smp_processor_id();
340 u64 index
= flags
& BPF_F_INDEX_MASK
;
341 struct bpf_event_entry
*ee
;
342 struct perf_event
*event
;
344 if (index
== BPF_F_CURRENT_CPU
)
346 if (unlikely(index
>= array
->map
.max_entries
))
349 ee
= READ_ONCE(array
->ptrs
[index
]);
354 if (unlikely(event
->attr
.type
!= PERF_TYPE_SOFTWARE
||
355 event
->attr
.config
!= PERF_COUNT_SW_BPF_OUTPUT
))
358 if (unlikely(event
->oncpu
!= cpu
))
361 perf_sample_data_init(sd
, 0, 0);
363 perf_event_output(event
, sd
, regs
);
367 BPF_CALL_5(bpf_perf_event_output
, struct pt_regs
*, regs
, struct bpf_map
*, map
,
368 u64
, flags
, void *, data
, u64
, size
)
370 struct perf_raw_record raw
= {
377 if (unlikely(flags
& ~(BPF_F_INDEX_MASK
)))
380 return __bpf_perf_event_output(regs
, map
, flags
, &raw
);
383 static const struct bpf_func_proto bpf_perf_event_output_proto
= {
384 .func
= bpf_perf_event_output
,
386 .ret_type
= RET_INTEGER
,
387 .arg1_type
= ARG_PTR_TO_CTX
,
388 .arg2_type
= ARG_CONST_MAP_PTR
,
389 .arg3_type
= ARG_ANYTHING
,
390 .arg4_type
= ARG_PTR_TO_MEM
,
391 .arg5_type
= ARG_CONST_SIZE
,
394 static DEFINE_PER_CPU(struct pt_regs
, bpf_pt_regs
);
396 u64
bpf_event_output(struct bpf_map
*map
, u64 flags
, void *meta
, u64 meta_size
,
397 void *ctx
, u64 ctx_size
, bpf_ctx_copy_t ctx_copy
)
399 struct pt_regs
*regs
= this_cpu_ptr(&bpf_pt_regs
);
400 struct perf_raw_frag frag
= {
405 struct perf_raw_record raw
= {
408 .next
= ctx_size
? &frag
: NULL
,
415 perf_fetch_caller_regs(regs
);
417 return __bpf_perf_event_output(regs
, map
, flags
, &raw
);
420 BPF_CALL_0(bpf_get_current_task
)
422 return (long) current
;
425 static const struct bpf_func_proto bpf_get_current_task_proto
= {
426 .func
= bpf_get_current_task
,
428 .ret_type
= RET_INTEGER
,
431 BPF_CALL_2(bpf_current_task_under_cgroup
, struct bpf_map
*, map
, u32
, idx
)
433 struct bpf_array
*array
= container_of(map
, struct bpf_array
, map
);
436 if (unlikely(in_interrupt()))
438 if (unlikely(idx
>= array
->map
.max_entries
))
441 cgrp
= READ_ONCE(array
->ptrs
[idx
]);
445 return task_under_cgroup_hierarchy(current
, cgrp
);
448 static const struct bpf_func_proto bpf_current_task_under_cgroup_proto
= {
449 .func
= bpf_current_task_under_cgroup
,
451 .ret_type
= RET_INTEGER
,
452 .arg1_type
= ARG_CONST_MAP_PTR
,
453 .arg2_type
= ARG_ANYTHING
,
456 BPF_CALL_3(bpf_probe_read_str
, void *, dst
, u32
, size
,
457 const void *, unsafe_ptr
)
462 * The strncpy_from_unsafe() call will likely not fill the entire
463 * buffer, but that's okay in this circumstance as we're probing
464 * arbitrary memory anyway similar to bpf_probe_read() and might
465 * as well probe the stack. Thus, memory is explicitly cleared
466 * only in error case, so that improper users ignoring return
467 * code altogether don't copy garbage; otherwise length of string
468 * is returned that can be used for bpf_perf_event_output() et al.
470 ret
= strncpy_from_unsafe(dst
, unsafe_ptr
, size
);
471 if (unlikely(ret
< 0))
472 memset(dst
, 0, size
);
477 static const struct bpf_func_proto bpf_probe_read_str_proto
= {
478 .func
= bpf_probe_read_str
,
480 .ret_type
= RET_INTEGER
,
481 .arg1_type
= ARG_PTR_TO_UNINIT_MEM
,
482 .arg2_type
= ARG_CONST_SIZE
,
483 .arg3_type
= ARG_ANYTHING
,
486 static const struct bpf_func_proto
*tracing_func_proto(enum bpf_func_id func_id
)
489 case BPF_FUNC_map_lookup_elem
:
490 return &bpf_map_lookup_elem_proto
;
491 case BPF_FUNC_map_update_elem
:
492 return &bpf_map_update_elem_proto
;
493 case BPF_FUNC_map_delete_elem
:
494 return &bpf_map_delete_elem_proto
;
495 case BPF_FUNC_probe_read
:
496 return &bpf_probe_read_proto
;
497 case BPF_FUNC_ktime_get_ns
:
498 return &bpf_ktime_get_ns_proto
;
499 case BPF_FUNC_tail_call
:
500 return &bpf_tail_call_proto
;
501 case BPF_FUNC_get_current_pid_tgid
:
502 return &bpf_get_current_pid_tgid_proto
;
503 case BPF_FUNC_get_current_task
:
504 return &bpf_get_current_task_proto
;
505 case BPF_FUNC_get_current_uid_gid
:
506 return &bpf_get_current_uid_gid_proto
;
507 case BPF_FUNC_get_current_comm
:
508 return &bpf_get_current_comm_proto
;
509 case BPF_FUNC_trace_printk
:
510 return bpf_get_trace_printk_proto();
511 case BPF_FUNC_get_smp_processor_id
:
512 return &bpf_get_smp_processor_id_proto
;
513 case BPF_FUNC_get_numa_node_id
:
514 return &bpf_get_numa_node_id_proto
;
515 case BPF_FUNC_perf_event_read
:
516 return &bpf_perf_event_read_proto
;
517 case BPF_FUNC_probe_write_user
:
518 return bpf_get_probe_write_proto();
519 case BPF_FUNC_current_task_under_cgroup
:
520 return &bpf_current_task_under_cgroup_proto
;
521 case BPF_FUNC_get_prandom_u32
:
522 return &bpf_get_prandom_u32_proto
;
523 case BPF_FUNC_probe_read_str
:
524 return &bpf_probe_read_str_proto
;
530 static const struct bpf_func_proto
*kprobe_prog_func_proto(enum bpf_func_id func_id
)
533 case BPF_FUNC_perf_event_output
:
534 return &bpf_perf_event_output_proto
;
535 case BPF_FUNC_get_stackid
:
536 return &bpf_get_stackid_proto
;
537 case BPF_FUNC_perf_event_read_value
:
538 return &bpf_perf_event_read_value_proto
;
540 return tracing_func_proto(func_id
);
544 /* bpf+kprobe programs can access fields of 'struct pt_regs' */
545 static bool kprobe_prog_is_valid_access(int off
, int size
, enum bpf_access_type type
,
546 struct bpf_insn_access_aux
*info
)
548 if (off
< 0 || off
>= sizeof(struct pt_regs
))
550 if (type
!= BPF_READ
)
555 * Assertion for 32 bit to make sure last 8 byte access
556 * (BPF_DW) to the last 4 byte member is disallowed.
558 if (off
+ size
> sizeof(struct pt_regs
))
564 const struct bpf_verifier_ops kprobe_verifier_ops
= {
565 .get_func_proto
= kprobe_prog_func_proto
,
566 .is_valid_access
= kprobe_prog_is_valid_access
,
569 const struct bpf_prog_ops kprobe_prog_ops
= {
572 BPF_CALL_5(bpf_perf_event_output_tp
, void *, tp_buff
, struct bpf_map
*, map
,
573 u64
, flags
, void *, data
, u64
, size
)
575 struct pt_regs
*regs
= *(struct pt_regs
**)tp_buff
;
578 * r1 points to perf tracepoint buffer where first 8 bytes are hidden
579 * from bpf program and contain a pointer to 'struct pt_regs'. Fetch it
580 * from there and call the same bpf_perf_event_output() helper inline.
582 return ____bpf_perf_event_output(regs
, map
, flags
, data
, size
);
585 static const struct bpf_func_proto bpf_perf_event_output_proto_tp
= {
586 .func
= bpf_perf_event_output_tp
,
588 .ret_type
= RET_INTEGER
,
589 .arg1_type
= ARG_PTR_TO_CTX
,
590 .arg2_type
= ARG_CONST_MAP_PTR
,
591 .arg3_type
= ARG_ANYTHING
,
592 .arg4_type
= ARG_PTR_TO_MEM
,
593 .arg5_type
= ARG_CONST_SIZE
,
596 BPF_CALL_3(bpf_get_stackid_tp
, void *, tp_buff
, struct bpf_map
*, map
,
599 struct pt_regs
*regs
= *(struct pt_regs
**)tp_buff
;
602 * Same comment as in bpf_perf_event_output_tp(), only that this time
603 * the other helper's function body cannot be inlined due to being
604 * external, thus we need to call raw helper function.
606 return bpf_get_stackid((unsigned long) regs
, (unsigned long) map
,
610 static const struct bpf_func_proto bpf_get_stackid_proto_tp
= {
611 .func
= bpf_get_stackid_tp
,
613 .ret_type
= RET_INTEGER
,
614 .arg1_type
= ARG_PTR_TO_CTX
,
615 .arg2_type
= ARG_CONST_MAP_PTR
,
616 .arg3_type
= ARG_ANYTHING
,
619 BPF_CALL_3(bpf_perf_prog_read_value_tp
, struct bpf_perf_event_data_kern
*, ctx
,
620 struct bpf_perf_event_value
*, buf
, u32
, size
)
624 if (unlikely(size
!= sizeof(struct bpf_perf_event_value
)))
626 err
= perf_event_read_local(ctx
->event
, &buf
->counter
, &buf
->enabled
,
632 memset(buf
, 0, size
);
636 static const struct bpf_func_proto bpf_perf_prog_read_value_proto_tp
= {
637 .func
= bpf_perf_prog_read_value_tp
,
639 .ret_type
= RET_INTEGER
,
640 .arg1_type
= ARG_PTR_TO_CTX
,
641 .arg2_type
= ARG_PTR_TO_UNINIT_MEM
,
642 .arg3_type
= ARG_CONST_SIZE
,
645 static const struct bpf_func_proto
*tp_prog_func_proto(enum bpf_func_id func_id
)
648 case BPF_FUNC_perf_event_output
:
649 return &bpf_perf_event_output_proto_tp
;
650 case BPF_FUNC_get_stackid
:
651 return &bpf_get_stackid_proto_tp
;
652 case BPF_FUNC_perf_prog_read_value
:
653 return &bpf_perf_prog_read_value_proto_tp
;
655 return tracing_func_proto(func_id
);
659 static bool tp_prog_is_valid_access(int off
, int size
, enum bpf_access_type type
,
660 struct bpf_insn_access_aux
*info
)
662 if (off
< sizeof(void *) || off
>= PERF_MAX_TRACE_SIZE
)
664 if (type
!= BPF_READ
)
669 BUILD_BUG_ON(PERF_MAX_TRACE_SIZE
% sizeof(__u64
));
673 const struct bpf_verifier_ops tracepoint_verifier_ops
= {
674 .get_func_proto
= tp_prog_func_proto
,
675 .is_valid_access
= tp_prog_is_valid_access
,
678 const struct bpf_prog_ops tracepoint_prog_ops
= {
681 static bool pe_prog_is_valid_access(int off
, int size
, enum bpf_access_type type
,
682 struct bpf_insn_access_aux
*info
)
684 const int size_sp
= FIELD_SIZEOF(struct bpf_perf_event_data
,
687 if (off
< 0 || off
>= sizeof(struct bpf_perf_event_data
))
689 if (type
!= BPF_READ
)
695 case bpf_ctx_range(struct bpf_perf_event_data
, sample_period
):
696 bpf_ctx_record_field_size(info
, size_sp
);
697 if (!bpf_ctx_narrow_access_ok(off
, size
, size_sp
))
701 if (size
!= sizeof(long))
708 static u32
pe_prog_convert_ctx_access(enum bpf_access_type type
,
709 const struct bpf_insn
*si
,
710 struct bpf_insn
*insn_buf
,
711 struct bpf_prog
*prog
, u32
*target_size
)
713 struct bpf_insn
*insn
= insn_buf
;
716 case offsetof(struct bpf_perf_event_data
, sample_period
):
717 *insn
++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern
,
718 data
), si
->dst_reg
, si
->src_reg
,
719 offsetof(struct bpf_perf_event_data_kern
, data
));
720 *insn
++ = BPF_LDX_MEM(BPF_DW
, si
->dst_reg
, si
->dst_reg
,
721 bpf_target_off(struct perf_sample_data
, period
, 8,
725 *insn
++ = BPF_LDX_MEM(BPF_FIELD_SIZEOF(struct bpf_perf_event_data_kern
,
726 regs
), si
->dst_reg
, si
->src_reg
,
727 offsetof(struct bpf_perf_event_data_kern
, regs
));
728 *insn
++ = BPF_LDX_MEM(BPF_SIZEOF(long), si
->dst_reg
, si
->dst_reg
,
733 return insn
- insn_buf
;
736 const struct bpf_verifier_ops perf_event_verifier_ops
= {
737 .get_func_proto
= tp_prog_func_proto
,
738 .is_valid_access
= pe_prog_is_valid_access
,
739 .convert_ctx_access
= pe_prog_convert_ctx_access
,
742 const struct bpf_prog_ops perf_event_prog_ops
= {