1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
2 #ifndef __BPF_CORE_READ_H__
3 #define __BPF_CORE_READ_H__
6 * enum bpf_field_info_kind is passed as a second argument into
7 * __builtin_preserve_field_info() built-in to get a specific aspect of
8 * a field, captured as a first argument. __builtin_preserve_field_info(field,
9 * info_kind) returns __u32 integer and produces BTF field relocation, which
10 * is understood and processed by libbpf during BPF object loading. See
11 * selftests/bpf for examples.
13 enum bpf_field_info_kind {
14 BPF_FIELD_BYTE_OFFSET = 0, /* field byte offset */
15 BPF_FIELD_BYTE_SIZE = 1,
16 BPF_FIELD_EXISTS = 2, /* field existence in target kernel */
18 BPF_FIELD_LSHIFT_U64 = 4,
19 BPF_FIELD_RSHIFT_U64 = 5,
22 /* second argument to __builtin_btf_type_id() built-in */
23 enum bpf_type_id_kind {
24 BPF_TYPE_ID_LOCAL = 0, /* BTF type ID in local program */
25 BPF_TYPE_ID_TARGET = 1, /* BTF type ID in target kernel */
28 /* second argument to __builtin_preserve_type_info() built-in */
29 enum bpf_type_info_kind {
30 BPF_TYPE_EXISTS = 0, /* type existence in target kernel */
31 BPF_TYPE_SIZE = 1, /* type size in target kernel */
32 BPF_TYPE_MATCHES = 2, /* type match in target kernel */
35 /* second argument to __builtin_preserve_enum_value() built-in */
36 enum bpf_enum_value_kind {
37 BPF_ENUMVAL_EXISTS = 0, /* enum value existence in kernel */
38 BPF_ENUMVAL_VALUE = 1, /* enum value value relocation */
41 #define __CORE_RELO(src, field, info) \
42 __builtin_preserve_field_info((src)->field, BPF_FIELD_##info)
44 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
45 #define __CORE_BITFIELD_PROBE_READ(dst, src, fld) \
46 bpf_probe_read_kernel( \
48 __CORE_RELO(src, fld, BYTE_SIZE), \
49 (const void *)src + __CORE_RELO(src, fld, BYTE_OFFSET))
51 /* semantics of LSHIFT_64 assumes loading values into low-ordered bytes, so
52 * for big-endian we need to adjust destination pointer accordingly, based on
55 #define __CORE_BITFIELD_PROBE_READ(dst, src, fld) \
56 bpf_probe_read_kernel( \
57 (void *)dst + (8 - __CORE_RELO(src, fld, BYTE_SIZE)), \
58 __CORE_RELO(src, fld, BYTE_SIZE), \
59 (const void *)src + __CORE_RELO(src, fld, BYTE_OFFSET))
63 * Extract bitfield, identified by s->field, and return its value as u64.
64 * All this is done in relocatable manner, so bitfield changes such as
65 * signedness, bit size, offset changes, this will be handled automatically.
66 * This version of macro is using bpf_probe_read_kernel() to read underlying
67 * integer storage. Macro functions as an expression and its return type is
68 * bpf_probe_read_kernel()'s return value: 0, on success, <0 on error.
70 #define BPF_CORE_READ_BITFIELD_PROBED(s, field) ({ \
71 unsigned long long val = 0; \
73 __CORE_BITFIELD_PROBE_READ(&val, s, field); \
74 val <<= __CORE_RELO(s, field, LSHIFT_U64); \
75 if (__CORE_RELO(s, field, SIGNED)) \
76 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \
78 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \
83 * Extract bitfield, identified by s->field, and return its value as u64.
84 * This version of macro is using direct memory reads and should be used from
85 * BPF program types that support such functionality (e.g., typed raw
88 #define BPF_CORE_READ_BITFIELD(s, field) ({ \
89 const void *p = (const void *)s + __CORE_RELO(s, field, BYTE_OFFSET); \
90 unsigned long long val; \
92 /* This is a so-called barrier_var() operation that makes specified \
93 * variable "a black box" for optimizing compiler. \
94 * It forces compiler to perform BYTE_OFFSET relocation on p and use \
95 * its calculated value in the switch below, instead of applying \
96 * the same relocation 4 times for each individual memory load. \
98 asm volatile("" : "=r"(p) : "0"(p)); \
100 switch (__CORE_RELO(s, field, BYTE_SIZE)) { \
101 case 1: val = *(const unsigned char *)p; break; \
102 case 2: val = *(const unsigned short *)p; break; \
103 case 4: val = *(const unsigned int *)p; break; \
104 case 8: val = *(const unsigned long long *)p; break; \
106 val <<= __CORE_RELO(s, field, LSHIFT_U64); \
107 if (__CORE_RELO(s, field, SIGNED)) \
108 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \
110 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \
115 * Write to a bitfield, identified by s->field.
116 * This is the inverse of BPF_CORE_WRITE_BITFIELD().
118 #define BPF_CORE_WRITE_BITFIELD(s, field, new_val) ({ \
119 void *p = (void *)s + __CORE_RELO(s, field, BYTE_OFFSET); \
120 unsigned int byte_size = __CORE_RELO(s, field, BYTE_SIZE); \
121 unsigned int lshift = __CORE_RELO(s, field, LSHIFT_U64); \
122 unsigned int rshift = __CORE_RELO(s, field, RSHIFT_U64); \
123 unsigned long long mask, val, nval = new_val; \
124 unsigned int rpad = rshift - lshift; \
126 asm volatile("" : "+r"(p)); \
128 switch (byte_size) { \
129 case 1: val = *(unsigned char *)p; break; \
130 case 2: val = *(unsigned short *)p; break; \
131 case 4: val = *(unsigned int *)p; break; \
132 case 8: val = *(unsigned long long *)p; break; \
135 mask = (~0ULL << rshift) >> lshift; \
136 val = (val & ~mask) | ((nval << rpad) & mask); \
138 switch (byte_size) { \
139 case 1: *(unsigned char *)p = val; break; \
140 case 2: *(unsigned short *)p = val; break; \
141 case 4: *(unsigned int *)p = val; break; \
142 case 8: *(unsigned long long *)p = val; break; \
146 #define ___bpf_field_ref1(field) (field)
147 #define ___bpf_field_ref2(type, field) (((typeof(type) *)0)->field)
148 #define ___bpf_field_ref(args...) \
149 ___bpf_apply(___bpf_field_ref, ___bpf_narg(args))(args)
152 * Convenience macro to check that field actually exists in target kernel's.
154 * 1, if matching field is present in target kernel;
155 * 0, if no matching field found.
157 * Supports two forms:
158 * - field reference through variable access:
159 * bpf_core_field_exists(p->my_field);
160 * - field reference through type and field names:
161 * bpf_core_field_exists(struct my_type, my_field).
163 #define bpf_core_field_exists(field...) \
164 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_EXISTS)
167 * Convenience macro to get the byte size of a field. Works for integers,
168 * struct/unions, pointers, arrays, and enums.
170 * Supports two forms:
171 * - field reference through variable access:
172 * bpf_core_field_size(p->my_field);
173 * - field reference through type and field names:
174 * bpf_core_field_size(struct my_type, my_field).
176 #define bpf_core_field_size(field...) \
177 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_BYTE_SIZE)
180 * Convenience macro to get field's byte offset.
182 * Supports two forms:
183 * - field reference through variable access:
184 * bpf_core_field_offset(p->my_field);
185 * - field reference through type and field names:
186 * bpf_core_field_offset(struct my_type, my_field).
188 #define bpf_core_field_offset(field...) \
189 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_BYTE_OFFSET)
192 * Convenience macro to get BTF type ID of a specified type, using a local BTF
193 * information. Return 32-bit unsigned integer with type ID from program's own
194 * BTF. Always succeeds.
196 #define bpf_core_type_id_local(type) \
197 __builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_LOCAL)
200 * Convenience macro to get BTF type ID of a target kernel's type that matches
201 * specified local type.
203 * - valid 32-bit unsigned type ID in kernel BTF;
204 * - 0, if no matching type was found in a target kernel BTF.
206 #define bpf_core_type_id_kernel(type) \
207 __builtin_btf_type_id(*(typeof(type) *)0, BPF_TYPE_ID_TARGET)
210 * Convenience macro to check that provided named type
211 * (struct/union/enum/typedef) exists in a target kernel.
213 * 1, if such type is present in target kernel's BTF;
214 * 0, if no matching type is found.
216 #define bpf_core_type_exists(type) \
217 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_EXISTS)
220 * Convenience macro to check that provided named type
221 * (struct/union/enum/typedef) "matches" that in a target kernel.
223 * 1, if the type matches in the target kernel's BTF;
224 * 0, if the type does not match any in the target kernel
226 #define bpf_core_type_matches(type) \
227 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_MATCHES)
230 * Convenience macro to get the byte size of a provided named type
231 * (struct/union/enum/typedef) in a target kernel.
233 * >= 0 size (in bytes), if type is present in target kernel's BTF;
234 * 0, if no matching type is found.
236 #define bpf_core_type_size(type) \
237 __builtin_preserve_type_info(*(typeof(type) *)0, BPF_TYPE_SIZE)
240 * Convenience macro to check that provided enumerator value is defined in
243 * 1, if specified enum type and its enumerator value are present in target
245 * 0, if no matching enum and/or enum value within that enum is found.
247 #define bpf_core_enum_value_exists(enum_type, enum_value) \
248 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_EXISTS)
251 * Convenience macro to get the integer value of an enumerator value in
254 * 64-bit value, if specified enum type and its enumerator value are
255 * present in target kernel's BTF;
256 * 0, if no matching enum and/or enum value within that enum is found.
258 #define bpf_core_enum_value(enum_type, enum_value) \
259 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE)
262 * bpf_core_read() abstracts away bpf_probe_read_kernel() call and captures
263 * offset relocation for source address using __builtin_preserve_access_index()
264 * built-in, provided by Clang.
266 * __builtin_preserve_access_index() takes as an argument an expression of
267 * taking an address of a field within struct/union. It makes compiler emit
268 * a relocation, which records BTF type ID describing root struct/union and an
269 * accessor string which describes exact embedded field that was used to take
270 * an address. See detailed description of this relocation format and
271 * semantics in comments to struct bpf_field_reloc in libbpf_internal.h.
273 * This relocation allows libbpf to adjust BPF instruction to use correct
274 * actual field offset, based on target kernel BTF type that matches original
275 * (local) BTF, used to record relocation.
277 #define bpf_core_read(dst, sz, src) \
278 bpf_probe_read_kernel(dst, sz, (const void *)__builtin_preserve_access_index(src))
280 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */
281 #define bpf_core_read_user(dst, sz, src) \
282 bpf_probe_read_user(dst, sz, (const void *)__builtin_preserve_access_index(src))
284 * bpf_core_read_str() is a thin wrapper around bpf_probe_read_str()
285 * additionally emitting BPF CO-RE field relocation for specified source
288 #define bpf_core_read_str(dst, sz, src) \
289 bpf_probe_read_kernel_str(dst, sz, (const void *)__builtin_preserve_access_index(src))
291 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */
292 #define bpf_core_read_user_str(dst, sz, src) \
293 bpf_probe_read_user_str(dst, sz, (const void *)__builtin_preserve_access_index(src))
295 #define ___concat(a, b) a ## b
296 #define ___apply(fn, n) ___concat(fn, n)
297 #define ___nth(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, __11, N, ...) N
300 * return number of provided arguments; used for switch-based variadic macro
301 * definitions (see ___last, ___arrow, etc below)
303 #define ___narg(...) ___nth(_, ##__VA_ARGS__, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)
305 * return 0 if no arguments are passed, N - otherwise; used for
306 * recursively-defined macros to specify termination (0) case, and generic
307 * (N) case (e.g., ___read_ptrs, ___core_read)
309 #define ___empty(...) ___nth(_, ##__VA_ARGS__, N, N, N, N, N, N, N, N, N, N, 0)
311 #define ___last1(x) x
312 #define ___last2(a, x) x
313 #define ___last3(a, b, x) x
314 #define ___last4(a, b, c, x) x
315 #define ___last5(a, b, c, d, x) x
316 #define ___last6(a, b, c, d, e, x) x
317 #define ___last7(a, b, c, d, e, f, x) x
318 #define ___last8(a, b, c, d, e, f, g, x) x
319 #define ___last9(a, b, c, d, e, f, g, h, x) x
320 #define ___last10(a, b, c, d, e, f, g, h, i, x) x
321 #define ___last(...) ___apply(___last, ___narg(__VA_ARGS__))(__VA_ARGS__)
323 #define ___nolast2(a, _) a
324 #define ___nolast3(a, b, _) a, b
325 #define ___nolast4(a, b, c, _) a, b, c
326 #define ___nolast5(a, b, c, d, _) a, b, c, d
327 #define ___nolast6(a, b, c, d, e, _) a, b, c, d, e
328 #define ___nolast7(a, b, c, d, e, f, _) a, b, c, d, e, f
329 #define ___nolast8(a, b, c, d, e, f, g, _) a, b, c, d, e, f, g
330 #define ___nolast9(a, b, c, d, e, f, g, h, _) a, b, c, d, e, f, g, h
331 #define ___nolast10(a, b, c, d, e, f, g, h, i, _) a, b, c, d, e, f, g, h, i
332 #define ___nolast(...) ___apply(___nolast, ___narg(__VA_ARGS__))(__VA_ARGS__)
334 #define ___arrow1(a) a
335 #define ___arrow2(a, b) a->b
336 #define ___arrow3(a, b, c) a->b->c
337 #define ___arrow4(a, b, c, d) a->b->c->d
338 #define ___arrow5(a, b, c, d, e) a->b->c->d->e
339 #define ___arrow6(a, b, c, d, e, f) a->b->c->d->e->f
340 #define ___arrow7(a, b, c, d, e, f, g) a->b->c->d->e->f->g
341 #define ___arrow8(a, b, c, d, e, f, g, h) a->b->c->d->e->f->g->h
342 #define ___arrow9(a, b, c, d, e, f, g, h, i) a->b->c->d->e->f->g->h->i
343 #define ___arrow10(a, b, c, d, e, f, g, h, i, j) a->b->c->d->e->f->g->h->i->j
344 #define ___arrow(...) ___apply(___arrow, ___narg(__VA_ARGS__))(__VA_ARGS__)
346 #define ___type(...) typeof(___arrow(__VA_ARGS__))
348 #define ___read(read_fn, dst, src_type, src, accessor) \
349 read_fn((void *)(dst), sizeof(*(dst)), &((src_type)(src))->accessor)
351 /* "recursively" read a sequence of inner pointers using local __t var */
352 #define ___rd_first(fn, src, a) ___read(fn, &__t, ___type(src), src, a);
353 #define ___rd_last(fn, ...) \
354 ___read(fn, &__t, ___type(___nolast(__VA_ARGS__)), __t, ___last(__VA_ARGS__));
355 #define ___rd_p1(fn, ...) const void *__t; ___rd_first(fn, __VA_ARGS__)
356 #define ___rd_p2(fn, ...) ___rd_p1(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
357 #define ___rd_p3(fn, ...) ___rd_p2(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
358 #define ___rd_p4(fn, ...) ___rd_p3(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
359 #define ___rd_p5(fn, ...) ___rd_p4(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
360 #define ___rd_p6(fn, ...) ___rd_p5(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
361 #define ___rd_p7(fn, ...) ___rd_p6(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
362 #define ___rd_p8(fn, ...) ___rd_p7(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
363 #define ___rd_p9(fn, ...) ___rd_p8(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
364 #define ___read_ptrs(fn, src, ...) \
365 ___apply(___rd_p, ___narg(__VA_ARGS__))(fn, src, __VA_ARGS__)
367 #define ___core_read0(fn, fn_ptr, dst, src, a) \
368 ___read(fn, dst, ___type(src), src, a);
369 #define ___core_readN(fn, fn_ptr, dst, src, ...) \
370 ___read_ptrs(fn_ptr, src, ___nolast(__VA_ARGS__)) \
371 ___read(fn, dst, ___type(src, ___nolast(__VA_ARGS__)), __t, \
372 ___last(__VA_ARGS__));
373 #define ___core_read(fn, fn_ptr, dst, src, a, ...) \
374 ___apply(___core_read, ___empty(__VA_ARGS__))(fn, fn_ptr, dst, \
375 src, a, ##__VA_ARGS__)
378 * BPF_CORE_READ_INTO() is a more performance-conscious variant of
379 * BPF_CORE_READ(), in which final field is read into user-provided storage.
380 * See BPF_CORE_READ() below for more details on general usage.
382 #define BPF_CORE_READ_INTO(dst, src, a, ...) ({ \
383 ___core_read(bpf_core_read, bpf_core_read, \
384 dst, (src), a, ##__VA_ARGS__) \
388 * Variant of BPF_CORE_READ_INTO() for reading from user-space memory.
390 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use.
392 #define BPF_CORE_READ_USER_INTO(dst, src, a, ...) ({ \
393 ___core_read(bpf_core_read_user, bpf_core_read_user, \
394 dst, (src), a, ##__VA_ARGS__) \
397 /* Non-CO-RE variant of BPF_CORE_READ_INTO() */
398 #define BPF_PROBE_READ_INTO(dst, src, a, ...) ({ \
399 ___core_read(bpf_probe_read_kernel, bpf_probe_read_kernel, \
400 dst, (src), a, ##__VA_ARGS__) \
403 /* Non-CO-RE variant of BPF_CORE_READ_USER_INTO().
405 * As no CO-RE relocations are emitted, source types can be arbitrary and are
406 * not restricted to kernel types only.
408 #define BPF_PROBE_READ_USER_INTO(dst, src, a, ...) ({ \
409 ___core_read(bpf_probe_read_user, bpf_probe_read_user, \
410 dst, (src), a, ##__VA_ARGS__) \
414 * BPF_CORE_READ_STR_INTO() does same "pointer chasing" as
415 * BPF_CORE_READ() for intermediate pointers, but then executes (and returns
416 * corresponding error code) bpf_core_read_str() for final string read.
418 #define BPF_CORE_READ_STR_INTO(dst, src, a, ...) ({ \
419 ___core_read(bpf_core_read_str, bpf_core_read, \
420 dst, (src), a, ##__VA_ARGS__) \
424 * Variant of BPF_CORE_READ_STR_INTO() for reading from user-space memory.
426 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use.
428 #define BPF_CORE_READ_USER_STR_INTO(dst, src, a, ...) ({ \
429 ___core_read(bpf_core_read_user_str, bpf_core_read_user, \
430 dst, (src), a, ##__VA_ARGS__) \
433 /* Non-CO-RE variant of BPF_CORE_READ_STR_INTO() */
434 #define BPF_PROBE_READ_STR_INTO(dst, src, a, ...) ({ \
435 ___core_read(bpf_probe_read_kernel_str, bpf_probe_read_kernel, \
436 dst, (src), a, ##__VA_ARGS__) \
440 * Non-CO-RE variant of BPF_CORE_READ_USER_STR_INTO().
442 * As no CO-RE relocations are emitted, source types can be arbitrary and are
443 * not restricted to kernel types only.
445 #define BPF_PROBE_READ_USER_STR_INTO(dst, src, a, ...) ({ \
446 ___core_read(bpf_probe_read_user_str, bpf_probe_read_user, \
447 dst, (src), a, ##__VA_ARGS__) \
451 * BPF_CORE_READ() is used to simplify BPF CO-RE relocatable read, especially
452 * when there are few pointer chasing steps.
453 * E.g., what in non-BPF world (or in BPF w/ BCC) would be something like:
454 * int x = s->a.b.c->d.e->f->g;
455 * can be succinctly achieved using BPF_CORE_READ as:
456 * int x = BPF_CORE_READ(s, a.b.c, d.e, f, g);
458 * BPF_CORE_READ will decompose above statement into 4 bpf_core_read (BPF
459 * CO-RE relocatable bpf_probe_read_kernel() wrapper) calls, logically
461 * 1. const void *__t = s->a.b.c;
466 * Equivalence is logical, because there is a heavy type casting/preservation
467 * involved, as well as all the reads are happening through
468 * bpf_probe_read_kernel() calls using __builtin_preserve_access_index() to
469 * emit CO-RE relocations.
471 * N.B. Only up to 9 "field accessors" are supported, which should be more
472 * than enough for any practical purpose.
474 #define BPF_CORE_READ(src, a, ...) ({ \
475 ___type((src), a, ##__VA_ARGS__) __r; \
476 BPF_CORE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \
481 * Variant of BPF_CORE_READ() for reading from user-space memory.
483 * NOTE: all the source types involved are still *kernel types* and need to
484 * exist in kernel (or kernel module) BTF, otherwise CO-RE relocation will
485 * fail. Custom user types are not relocatable with CO-RE.
486 * The typical situation in which BPF_CORE_READ_USER() might be used is to
487 * read kernel UAPI types from the user-space memory passed in as a syscall
490 #define BPF_CORE_READ_USER(src, a, ...) ({ \
491 ___type((src), a, ##__VA_ARGS__) __r; \
492 BPF_CORE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \
496 /* Non-CO-RE variant of BPF_CORE_READ() */
497 #define BPF_PROBE_READ(src, a, ...) ({ \
498 ___type((src), a, ##__VA_ARGS__) __r; \
499 BPF_PROBE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \
504 * Non-CO-RE variant of BPF_CORE_READ_USER().
506 * As no CO-RE relocations are emitted, source types can be arbitrary and are
507 * not restricted to kernel types only.
509 #define BPF_PROBE_READ_USER(src, a, ...) ({ \
510 ___type((src), a, ##__VA_ARGS__) __r; \
511 BPF_PROBE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \