1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
2 #ifndef __BPF_CORE_READ_H__
3 #define __BPF_CORE_READ_H__
4
5 #include "bpf_helpers.h"
6
7 /*
8 * enum bpf_field_info_kind is passed as a second argument into
9 * __builtin_preserve_field_info() built-in to get a specific aspect of
10 * a field, captured as a first argument. __builtin_preserve_field_info(field,
11 * info_kind) returns __u32 integer and produces BTF field relocation, which
12 * is understood and processed by libbpf during BPF object loading. See
13 * selftests/bpf for examples.
14 */
15 enum bpf_field_info_kind {
16 BPF_FIELD_BYTE_OFFSET = 0, /* field byte offset */
17 BPF_FIELD_BYTE_SIZE = 1,
18 BPF_FIELD_EXISTS = 2, /* field existence in target kernel */
19 BPF_FIELD_SIGNED = 3,
20 BPF_FIELD_LSHIFT_U64 = 4,
21 BPF_FIELD_RSHIFT_U64 = 5,
22 };
23
24 /* second argument to __builtin_btf_type_id() built-in */
25 enum bpf_type_id_kind {
26 BPF_TYPE_ID_LOCAL = 0, /* BTF type ID in local program */
27 BPF_TYPE_ID_TARGET = 1, /* BTF type ID in target kernel */
28 };
29
30 /* second argument to __builtin_preserve_type_info() built-in */
31 enum bpf_type_info_kind {
32 BPF_TYPE_EXISTS = 0, /* type existence in target kernel */
33 BPF_TYPE_SIZE = 1, /* type size in target kernel */
34 BPF_TYPE_MATCHES = 2, /* type match in target kernel */
35 };
36
37 /* second argument to __builtin_preserve_enum_value() built-in */
38 enum bpf_enum_value_kind {
39 BPF_ENUMVAL_EXISTS = 0, /* enum value existence in kernel */
40 BPF_ENUMVAL_VALUE = 1, /* enum value value relocation */
41 };
42
43 #define __CORE_RELO(src, field, info) \
44 __builtin_preserve_field_info((src)->field, BPF_FIELD_##info)
45
46 #if __BYTE_ORDER__ == __ORDER_LITTLE_ENDIAN__
47 #define __CORE_BITFIELD_PROBE_READ(dst, src, fld) \
48 bpf_probe_read_kernel( \
49 (void *)dst, \
50 __CORE_RELO(src, fld, BYTE_SIZE), \
51 (const void *)src + __CORE_RELO(src, fld, BYTE_OFFSET))
52 #else
53 /* semantics of LSHIFT_64 assumes loading values into low-ordered bytes, so
54 * for big-endian we need to adjust destination pointer accordingly, based on
55 * field byte size
56 */
57 #define __CORE_BITFIELD_PROBE_READ(dst, src, fld) \
58 bpf_probe_read_kernel( \
59 (void *)dst + (8 - __CORE_RELO(src, fld, BYTE_SIZE)), \
60 __CORE_RELO(src, fld, BYTE_SIZE), \
61 (const void *)src + __CORE_RELO(src, fld, BYTE_OFFSET))
62 #endif
63
64 /*
65 * Extract bitfield, identified by s->field, and return its value as u64.
66 * All this is done in relocatable manner, so bitfield changes such as
67 * signedness, bit size, offset changes, this will be handled automatically.
68 * This version of macro is using bpf_probe_read_kernel() to read underlying
69 * integer storage. Macro functions as an expression and its return type is
70 * bpf_probe_read_kernel()'s return value: 0, on success, <0 on error.
71 */
72 #define BPF_CORE_READ_BITFIELD_PROBED(s, field) ({ \
73 unsigned long long val = 0; \
74 \
75 __CORE_BITFIELD_PROBE_READ(&val, s, field); \
76 val <<= __CORE_RELO(s, field, LSHIFT_U64); \
77 if (__CORE_RELO(s, field, SIGNED)) \
78 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \
79 else \
80 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \
81 val; \
82 })
83
84 /*
85 * Extract bitfield, identified by s->field, and return its value as u64.
86 * This version of macro is using direct memory reads and should be used from
87 * BPF program types that support such functionality (e.g., typed raw
88 * tracepoints).
89 */
90 #define BPF_CORE_READ_BITFIELD(s, field) ({ \
91 const void *p = (const void *)s + __CORE_RELO(s, field, BYTE_OFFSET); \
92 unsigned long long val; \
93 \
94 /* This is a so-called barrier_var() operation that makes specified \
95 * variable "a black box" for optimizing compiler. \
96 * It forces compiler to perform BYTE_OFFSET relocation on p and use \
97 * its calculated value in the switch below, instead of applying \
98 * the same relocation 4 times for each individual memory load. \
99 */ \
100 asm volatile("" : "=r"(p) : ""(p)); \
101 \
102 switch (__CORE_RELO(s, field, BYTE_SIZE)) { \
103 case 1: val = *(const unsigned char *)p; break; \
104 case 2: val = *(const unsigned short *)p; break; \
105 case 4: val = *(const unsigned int *)p; break; \
106 case 8: val = *(const unsigned long long *)p; break; \
107 default: val = 0; break; \
108 } \
109 val <<= __CORE_RELO(s, field, LSHIFT_U64); \
110 if (__CORE_RELO(s, field, SIGNED)) \
111 val = ((long long)val) >> __CORE_RELO(s, field, RSHIFT_U64); \
112 else \
113 val = val >> __CORE_RELO(s, field, RSHIFT_U64); \
114 val; \
115 })
116
117 /*
118 * Write to a bitfield, identified by s->field.
119 * This is the inverse of BPF_CORE_WRITE_BITFIELD().
120 */
121 #define BPF_CORE_WRITE_BITFIELD(s, field, new_val) ({ \
122 void *p = (void *)s + __CORE_RELO(s, field, BYTE_OFFSET); \
123 unsigned int byte_size = __CORE_RELO(s, field, BYTE_SIZE); \
124 unsigned int lshift = __CORE_RELO(s, field, LSHIFT_U64); \
125 unsigned int rshift = __CORE_RELO(s, field, RSHIFT_U64); \
126 unsigned long long mask, val, nval = new_val; \
127 unsigned int rpad = rshift - lshift; \
128 \
129 asm volatile("" : "+r"(p)); \
130 \
131 switch (byte_size) { \
132 case 1: val = *(unsigned char *)p; break; \
133 case 2: val = *(unsigned short *)p; break; \
134 case 4: val = *(unsigned int *)p; break; \
135 case 8: val = *(unsigned long long *)p; break; \
136 } \
137 \
138 mask = (~0ULL << rshift) >> lshift; \
139 val = (val & ~mask) | ((nval << rpad) & mask); \
140 \
141 switch (byte_size) { \
142 case 1: *(unsigned char *)p = val; break; \
143 case 2: *(unsigned short *)p = val; break; \
144 case 4: *(unsigned int *)p = val; break; \
145 case 8: *(unsigned long long *)p = val; break; \
146 } \
147 })
148
149 /* Differentiator between compilers builtin implementations. This is a
150 * requirement due to the compiler parsing differences where GCC optimizes
151 * early in parsing those constructs of type pointers to the builtin specific
152 * type, resulting in not being possible to collect the required type
153 * information in the builtin expansion.
154 */
155 #ifdef __clang__
156 #define ___bpf_typeof(type) ((typeof(type) *) 0)
157 #else
158 #define ___bpf_typeof1(type, NR) ({ \
159 extern typeof(type) *___concat(bpf_type_tmp_, NR); \
160 ___concat(bpf_type_tmp_, NR); \
161 })
162 #define ___bpf_typeof(type) ___bpf_typeof1(type, __COUNTER__)
163 #endif
164
165 #ifdef __clang__
166 #define ___bpf_field_ref1(field) (field)
167 #define ___bpf_field_ref2(type, field) (___bpf_typeof(type)->field)
168 #else
169 #define ___bpf_field_ref1(field) (&(field))
170 #define ___bpf_field_ref2(type, field) (&(___bpf_typeof(type)->field))
171 #endif
172 #define ___bpf_field_ref(args...) \
173 ___bpf_apply(___bpf_field_ref, ___bpf_narg(args))(args)
174
175 /*
176 * Convenience macro to check that field actually exists in target kernel's.
177 * Returns:
178 * 1, if matching field is present in target kernel;
179 * 0, if no matching field found.
180 *
181 * Supports two forms:
182 * - field reference through variable access:
183 * bpf_core_field_exists(p->my_field);
184 * - field reference through type and field names:
185 * bpf_core_field_exists(struct my_type, my_field).
186 */
187 #define bpf_core_field_exists(field...) \
188 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_EXISTS)
189
190 /*
191 * Convenience macro to get the byte size of a field. Works for integers,
192 * struct/unions, pointers, arrays, and enums.
193 *
194 * Supports two forms:
195 * - field reference through variable access:
196 * bpf_core_field_size(p->my_field);
197 * - field reference through type and field names:
198 * bpf_core_field_size(struct my_type, my_field).
199 */
200 #define bpf_core_field_size(field...) \
201 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_BYTE_SIZE)
202
203 /*
204 * Convenience macro to get field's byte offset.
205 *
206 * Supports two forms:
207 * - field reference through variable access:
208 * bpf_core_field_offset(p->my_field);
209 * - field reference through type and field names:
210 * bpf_core_field_offset(struct my_type, my_field).
211 */
212 #define bpf_core_field_offset(field...) \
213 __builtin_preserve_field_info(___bpf_field_ref(field), BPF_FIELD_BYTE_OFFSET)
214
215 /*
216 * Convenience macro to get BTF type ID of a specified type, using a local BTF
217 * information. Return 32-bit unsigned integer with type ID from program's own
218 * BTF. Always succeeds.
219 */
220 #define bpf_core_type_id_local(type) \
221 __builtin_btf_type_id(*___bpf_typeof(type), BPF_TYPE_ID_LOCAL)
222
223 /*
224 * Convenience macro to get BTF type ID of a target kernel's type that matches
225 * specified local type.
226 * Returns:
227 * - valid 32-bit unsigned type ID in kernel BTF;
228 * - 0, if no matching type was found in a target kernel BTF.
229 */
230 #define bpf_core_type_id_kernel(type) \
231 __builtin_btf_type_id(*___bpf_typeof(type), BPF_TYPE_ID_TARGET)
232
233 /*
234 * Convenience macro to check that provided named type
235 * (struct/union/enum/typedef) exists in a target kernel.
236 * Returns:
237 * 1, if such type is present in target kernel's BTF;
238 * 0, if no matching type is found.
239 */
240 #define bpf_core_type_exists(type) \
241 __builtin_preserve_type_info(*___bpf_typeof(type), BPF_TYPE_EXISTS)
242
243 /*
244 * Convenience macro to check that provided named type
245 * (struct/union/enum/typedef) "matches" that in a target kernel.
246 * Returns:
247 * 1, if the type matches in the target kernel's BTF;
248 * 0, if the type does not match any in the target kernel
249 */
250 #define bpf_core_type_matches(type) \
251 __builtin_preserve_type_info(*___bpf_typeof(type), BPF_TYPE_MATCHES)
252
253 /*
254 * Convenience macro to get the byte size of a provided named type
255 * (struct/union/enum/typedef) in a target kernel.
256 * Returns:
257 * >= 0 size (in bytes), if type is present in target kernel's BTF;
258 * 0, if no matching type is found.
259 */
260 #define bpf_core_type_size(type) \
261 __builtin_preserve_type_info(*___bpf_typeof(type), BPF_TYPE_SIZE)
262
263 /*
264 * Convenience macro to check that provided enumerator value is defined in
265 * a target kernel.
266 * Returns:
267 * 1, if specified enum type and its enumerator value are present in target
268 * kernel's BTF;
269 * 0, if no matching enum and/or enum value within that enum is found.
270 */
271 #ifdef __clang__
272 #define bpf_core_enum_value_exists(enum_type, enum_value) \
273 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_EXISTS)
274 #else
275 #define bpf_core_enum_value_exists(enum_type, enum_value) \
276 __builtin_preserve_enum_value(___bpf_typeof(enum_type), enum_value, BPF_ENUMVAL_EXISTS)
277 #endif
278
279 /*
280 * Convenience macro to get the integer value of an enumerator value in
281 * a target kernel.
282 * Returns:
283 * 64-bit value, if specified enum type and its enumerator value are
284 * present in target kernel's BTF;
285 * 0, if no matching enum and/or enum value within that enum is found.
286 */
287 #ifdef __clang__
288 #define bpf_core_enum_value(enum_type, enum_value) \
289 __builtin_preserve_enum_value(*(typeof(enum_type) *)enum_value, BPF_ENUMVAL_VALUE)
290 #else
291 #define bpf_core_enum_value(enum_type, enum_value) \
292 __builtin_preserve_enum_value(___bpf_typeof(enum_type), enum_value, BPF_ENUMVAL_VALUE)
293 #endif
294
295 /*
296 * bpf_core_read() abstracts away bpf_probe_read_kernel() call and captures
297 * offset relocation for source address using __builtin_preserve_access_index()
298 * built-in, provided by Clang.
299 *
300 * __builtin_preserve_access_index() takes as an argument an expression of
301 * taking an address of a field within struct/union. It makes compiler emit
302 * a relocation, which records BTF type ID describing root struct/union and an
303 * accessor string which describes exact embedded field that was used to take
304 * an address. See detailed description of this relocation format and
305 * semantics in comments to struct bpf_core_relo in include/uapi/linux/bpf.h.
306 *
307 * This relocation allows libbpf to adjust BPF instruction to use correct
308 * actual field offset, based on target kernel BTF type that matches original
309 * (local) BTF, used to record relocation.
310 */
311 #define bpf_core_read(dst, sz, src) \
312 bpf_probe_read_kernel(dst, sz, (const void *)__builtin_preserve_access_index(src))
313
314 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */
315 #define bpf_core_read_user(dst, sz, src) \
316 bpf_probe_read_user(dst, sz, (const void *)__builtin_preserve_access_index(src))
317 /*
318 * bpf_core_read_str() is a thin wrapper around bpf_probe_read_str()
319 * additionally emitting BPF CO-RE field relocation for specified source
320 * argument.
321 */
322 #define bpf_core_read_str(dst, sz, src) \
323 bpf_probe_read_kernel_str(dst, sz, (const void *)__builtin_preserve_access_index(src))
324
325 /* NOTE: see comments for BPF_CORE_READ_USER() about the proper types use. */
326 #define bpf_core_read_user_str(dst, sz, src) \
327 bpf_probe_read_user_str(dst, sz, (const void *)__builtin_preserve_access_index(src))
328
329 extern void *bpf_rdonly_cast(const void *obj, __u32 btf_id) __ksym __weak;
330
331 /*
332 * Cast provided pointer *ptr* into a pointer to a specified *type* in such
333 * a way that BPF verifier will become aware of associated kernel-side BTF
334 * type. This allows to access members of kernel types directly without the
335 * need to use BPF_CORE_READ() macros.
336 */
337 #define bpf_core_cast(ptr, type) \
338 ((typeof(type) *)bpf_rdonly_cast((ptr), bpf_core_type_id_kernel(type)))
339
340 #define ___concat(a, b) a ## b
341 #define ___apply(fn, n) ___concat(fn, n)
342 #define ___nth(_1, _2, _3, _4, _5, _6, _7, _8, _9, _10, __11, N, ...) N
343
344 /*
345 * return number of provided arguments; used for switch-based variadic macro
346 * definitions (see ___last, ___arrow, etc below)
347 */
348 #define ___narg(...) ___nth(_, ##__VA_ARGS__, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)
349 /*
350 * return 0 if no arguments are passed, N - otherwise; used for
351 * recursively-defined macros to specify termination (0) case, and generic
352 * (N) case (e.g., ___read_ptrs, ___core_read)
353 */
354 #define ___empty(...) ___nth(_, ##__VA_ARGS__, N, N, N, N, N, N, N, N, N, N, 0)
355
356 #define ___last1(x) x
357 #define ___last2(a, x) x
358 #define ___last3(a, b, x) x
359 #define ___last4(a, b, c, x) x
360 #define ___last5(a, b, c, d, x) x
361 #define ___last6(a, b, c, d, e, x) x
362 #define ___last7(a, b, c, d, e, f, x) x
363 #define ___last8(a, b, c, d, e, f, g, x) x
364 #define ___last9(a, b, c, d, e, f, g, h, x) x
365 #define ___last10(a, b, c, d, e, f, g, h, i, x) x
366 #define ___last(...) ___apply(___last, ___narg(__VA_ARGS__))(__VA_ARGS__)
367
368 #define ___nolast2(a, _) a
369 #define ___nolast3(a, b, _) a, b
370 #define ___nolast4(a, b, c, _) a, b, c
371 #define ___nolast5(a, b, c, d, _) a, b, c, d
372 #define ___nolast6(a, b, c, d, e, _) a, b, c, d, e
373 #define ___nolast7(a, b, c, d, e, f, _) a, b, c, d, e, f
374 #define ___nolast8(a, b, c, d, e, f, g, _) a, b, c, d, e, f, g
375 #define ___nolast9(a, b, c, d, e, f, g, h, _) a, b, c, d, e, f, g, h
376 #define ___nolast10(a, b, c, d, e, f, g, h, i, _) a, b, c, d, e, f, g, h, i
377 #define ___nolast(...) ___apply(___nolast, ___narg(__VA_ARGS__))(__VA_ARGS__)
378
379 #define ___arrow1(a) a
380 #define ___arrow2(a, b) a->b
381 #define ___arrow3(a, b, c) a->b->c
382 #define ___arrow4(a, b, c, d) a->b->c->d
383 #define ___arrow5(a, b, c, d, e) a->b->c->d->e
384 #define ___arrow6(a, b, c, d, e, f) a->b->c->d->e->f
385 #define ___arrow7(a, b, c, d, e, f, g) a->b->c->d->e->f->g
386 #define ___arrow8(a, b, c, d, e, f, g, h) a->b->c->d->e->f->g->h
387 #define ___arrow9(a, b, c, d, e, f, g, h, i) a->b->c->d->e->f->g->h->i
388 #define ___arrow10(a, b, c, d, e, f, g, h, i, j) a->b->c->d->e->f->g->h->i->j
389 #define ___arrow(...) ___apply(___arrow, ___narg(__VA_ARGS__))(__VA_ARGS__)
390
391 #define ___type(...) typeof(___arrow(__VA_ARGS__))
392
393 #define ___read(read_fn, dst, src_type, src, accessor) \
394 read_fn((void *)(dst), sizeof(*(dst)), &((src_type)(src))->accessor)
395
396 /* "recursively" read a sequence of inner pointers using local __t var */
397 #define ___rd_first(fn, src, a) ___read(fn, &__t, ___type(src), src, a);
398 #define ___rd_last(fn, ...) \
399 ___read(fn, &__t, ___type(___nolast(__VA_ARGS__)), __t, ___last(__VA_ARGS__));
400 #define ___rd_p1(fn, ...) const void *__t; ___rd_first(fn, __VA_ARGS__)
401 #define ___rd_p2(fn, ...) ___rd_p1(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
402 #define ___rd_p3(fn, ...) ___rd_p2(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
403 #define ___rd_p4(fn, ...) ___rd_p3(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
404 #define ___rd_p5(fn, ...) ___rd_p4(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
405 #define ___rd_p6(fn, ...) ___rd_p5(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
406 #define ___rd_p7(fn, ...) ___rd_p6(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
407 #define ___rd_p8(fn, ...) ___rd_p7(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
408 #define ___rd_p9(fn, ...) ___rd_p8(fn, ___nolast(__VA_ARGS__)) ___rd_last(fn, __VA_ARGS__)
409 #define ___read_ptrs(fn, src, ...) \
410 ___apply(___rd_p, ___narg(__VA_ARGS__))(fn, src, __VA_ARGS__)
411
412 #define ___core_read0(fn, fn_ptr, dst, src, a) \
413 ___read(fn, dst, ___type(src), src, a);
414 #define ___core_readN(fn, fn_ptr, dst, src, ...) \
415 ___read_ptrs(fn_ptr, src, ___nolast(__VA_ARGS__)) \
416 ___read(fn, dst, ___type(src, ___nolast(__VA_ARGS__)), __t, \
417 ___last(__VA_ARGS__));
418 #define ___core_read(fn, fn_ptr, dst, src, a, ...) \
419 ___apply(___core_read, ___empty(__VA_ARGS__))(fn, fn_ptr, dst, \
420 src, a, ##__VA_ARGS__)
421
422 /*
423 * BPF_CORE_READ_INTO() is a more performance-conscious variant of
424 * BPF_CORE_READ(), in which final field is read into user-provided storage.
425 * See BPF_CORE_READ() below for more details on general usage.
426 */
427 #define BPF_CORE_READ_INTO(dst, src, a, ...) ({ \
428 ___core_read(bpf_core_read, bpf_core_read, \
429 dst, (src), a, ##__VA_ARGS__) \
430 })
431
432 /*
433 * Variant of BPF_CORE_READ_INTO() for reading from user-space memory.
434 *
435 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use.
436 */
437 #define BPF_CORE_READ_USER_INTO(dst, src, a, ...) ({ \
438 ___core_read(bpf_core_read_user, bpf_core_read_user, \
439 dst, (src), a, ##__VA_ARGS__) \
440 })
441
442 /* Non-CO-RE variant of BPF_CORE_READ_INTO() */
443 #define BPF_PROBE_READ_INTO(dst, src, a, ...) ({ \
444 ___core_read(bpf_probe_read_kernel, bpf_probe_read_kernel, \
445 dst, (src), a, ##__VA_ARGS__) \
446 })
447
448 /* Non-CO-RE variant of BPF_CORE_READ_USER_INTO().
449 *
450 * As no CO-RE relocations are emitted, source types can be arbitrary and are
451 * not restricted to kernel types only.
452 */
453 #define BPF_PROBE_READ_USER_INTO(dst, src, a, ...) ({ \
454 ___core_read(bpf_probe_read_user, bpf_probe_read_user, \
455 dst, (src), a, ##__VA_ARGS__) \
456 })
457
458 /*
459 * BPF_CORE_READ_STR_INTO() does same "pointer chasing" as
460 * BPF_CORE_READ() for intermediate pointers, but then executes (and returns
461 * corresponding error code) bpf_core_read_str() for final string read.
462 */
463 #define BPF_CORE_READ_STR_INTO(dst, src, a, ...) ({ \
464 ___core_read(bpf_core_read_str, bpf_core_read, \
465 dst, (src), a, ##__VA_ARGS__) \
466 })
467
468 /*
469 * Variant of BPF_CORE_READ_STR_INTO() for reading from user-space memory.
470 *
471 * NOTE: see comments for BPF_CORE_READ_USER() about the proper types use.
472 */
473 #define BPF_CORE_READ_USER_STR_INTO(dst, src, a, ...) ({ \
474 ___core_read(bpf_core_read_user_str, bpf_core_read_user, \
475 dst, (src), a, ##__VA_ARGS__) \
476 })
477
478 /* Non-CO-RE variant of BPF_CORE_READ_STR_INTO() */
479 #define BPF_PROBE_READ_STR_INTO(dst, src, a, ...) ({ \
480 ___core_read(bpf_probe_read_kernel_str, bpf_probe_read_kernel, \
481 dst, (src), a, ##__VA_ARGS__) \
482 })
483
484 /*
485 * Non-CO-RE variant of BPF_CORE_READ_USER_STR_INTO().
486 *
487 * As no CO-RE relocations are emitted, source types can be arbitrary and are
488 * not restricted to kernel types only.
489 */
490 #define BPF_PROBE_READ_USER_STR_INTO(dst, src, a, ...) ({ \
491 ___core_read(bpf_probe_read_user_str, bpf_probe_read_user, \
492 dst, (src), a, ##__VA_ARGS__) \
493 })
494
495 /*
496 * BPF_CORE_READ() is used to simplify BPF CO-RE relocatable read, especially
497 * when there are few pointer chasing steps.
498 * E.g., what in non-BPF world (or in BPF w/ BCC) would be something like:
499 * int x = s->a.b.c->d.e->f->g;
500 * can be succinctly achieved using BPF_CORE_READ as:
501 * int x = BPF_CORE_READ(s, a.b.c, d.e, f, g);
502 *
503 * BPF_CORE_READ will decompose above statement into 4 bpf_core_read (BPF
504 * CO-RE relocatable bpf_probe_read_kernel() wrapper) calls, logically
505 * equivalent to:
506 * 1. const void *__t = s->a.b.c;
507 * 2. __t = __t->d.e;
508 * 3. __t = __t->f;
509 * 4. return __t->g;
510 *
511 * Equivalence is logical, because there is a heavy type casting/preservation
512 * involved, as well as all the reads are happening through
513 * bpf_probe_read_kernel() calls using __builtin_preserve_access_index() to
514 * emit CO-RE relocations.
515 *
516 * N.B. Only up to 9 "field accessors" are supported, which should be more
517 * than enough for any practical purpose.
518 */
519 #define BPF_CORE_READ(src, a, ...) ({ \
520 ___type((src), a, ##__VA_ARGS__) __r; \
521 BPF_CORE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \
522 __r; \
523 })
524
525 /*
526 * Variant of BPF_CORE_READ() for reading from user-space memory.
527 *
528 * NOTE: all the source types involved are still *kernel types* and need to
529 * exist in kernel (or kernel module) BTF, otherwise CO-RE relocation will
530 * fail. Custom user types are not relocatable with CO-RE.
531 * The typical situation in which BPF_CORE_READ_USER() might be used is to
532 * read kernel UAPI types from the user-space memory passed in as a syscall
533 * input argument.
534 */
535 #define BPF_CORE_READ_USER(src, a, ...) ({ \
536 ___type((src), a, ##__VA_ARGS__) __r; \
537 BPF_CORE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \
538 __r; \
539 })
540
541 /* Non-CO-RE variant of BPF_CORE_READ() */
542 #define BPF_PROBE_READ(src, a, ...) ({ \
543 ___type((src), a, ##__VA_ARGS__) __r; \
544 BPF_PROBE_READ_INTO(&__r, (src), a, ##__VA_ARGS__); \
545 __r; \
546 })
547
548 /*
549 * Non-CO-RE variant of BPF_CORE_READ_USER().
550 *
551 * As no CO-RE relocations are emitted, source types can be arbitrary and are
552 * not restricted to kernel types only.
553 */
554 #define BPF_PROBE_READ_USER(src, a, ...) ({ \
555 ___type((src), a, ##__VA_ARGS__) __r; \
556 BPF_PROBE_READ_USER_INTO(&__r, (src), a, ##__VA_ARGS__); \
557 __r; \
558 })
559
560 #endif
561
562
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.