bpf_helpers.h 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423
  1. /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
  2. #ifndef __BPF_HELPERS__
  3. #define __BPF_HELPERS__
  4. /*
  5. * Note that bpf programs need to include either
  6. * vmlinux.h (auto-generated from BTF) or linux/types.h
  7. * in advance since bpf_helper_defs.h uses such types
  8. * as __u64.
  9. */
  10. #include "bpf_helper_defs.h"
  11. #define __uint(name, val) int (*name)[val]
  12. #define __type(name, val) typeof(val) *name
  13. #define __array(name, val) typeof(val) *name[]
  14. #define __ulong(name, val) enum { ___bpf_concat(__unique_value, __COUNTER__) = val } name
  15. /*
  16. * Helper macro to place programs, maps, license in
  17. * different sections in elf_bpf file. Section names
  18. * are interpreted by libbpf depending on the context (BPF programs, BPF maps,
  19. * extern variables, etc).
  20. * To allow use of SEC() with externs (e.g., for extern .maps declarations),
  21. * make sure __attribute__((unused)) doesn't trigger compilation warning.
  22. */
  23. #if __GNUC__ && !__clang__
  24. /*
  25. * Pragma macros are broken on GCC
  26. * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55578
  27. * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90400
  28. */
  29. #define SEC(name) __attribute__((section(name), used))
  30. #else
  31. #define SEC(name) \
  32. _Pragma("GCC diagnostic push") \
  33. _Pragma("GCC diagnostic ignored \"-Wignored-attributes\"") \
  34. __attribute__((section(name), used)) \
  35. _Pragma("GCC diagnostic pop") \
  36. #endif
  37. /* Avoid 'linux/stddef.h' definition of '__always_inline'. */
  38. #undef __always_inline
  39. #define __always_inline inline __attribute__((always_inline))
  40. #ifndef __noinline
  41. #define __noinline __attribute__((noinline))
  42. #endif
  43. #ifndef __weak
  44. #define __weak __attribute__((weak))
  45. #endif
  46. /*
  47. * Use __hidden attribute to mark a non-static BPF subprogram effectively
  48. * static for BPF verifier's verification algorithm purposes, allowing more
  49. * extensive and permissive BPF verification process, taking into account
  50. * subprogram's caller context.
  51. */
  52. #define __hidden __attribute__((visibility("hidden")))
  53. /* When utilizing vmlinux.h with BPF CO-RE, user BPF programs can't include
  54. * any system-level headers (such as stddef.h, linux/version.h, etc), and
  55. * commonly-used macros like NULL and KERNEL_VERSION aren't available through
  56. * vmlinux.h. This just adds unnecessary hurdles and forces users to re-define
  57. * them on their own. So as a convenience, provide such definitions here.
  58. */
  59. #ifndef NULL
  60. #define NULL ((void *)0)
  61. #endif
  62. #ifndef KERNEL_VERSION
  63. #define KERNEL_VERSION(a, b, c) (((a) << 16) + ((b) << 8) + ((c) > 255 ? 255 : (c)))
  64. #endif
  65. /*
  66. * Helper macros to manipulate data structures
  67. */
  68. /* offsetof() definition that uses __builtin_offset() might not preserve field
  69. * offset CO-RE relocation properly, so force-redefine offsetof() using
  70. * old-school approach which works with CO-RE correctly
  71. */
  72. #undef offsetof
  73. #define offsetof(type, member) ((unsigned long)&((type *)0)->member)
  74. /* redefined container_of() to ensure we use the above offsetof() macro */
  75. #undef container_of
  76. #define container_of(ptr, type, member) \
  77. ({ \
  78. void *__mptr = (void *)(ptr); \
  79. ((type *)(__mptr - offsetof(type, member))); \
  80. })
  81. /*
  82. * Compiler (optimization) barrier.
  83. */
  84. #ifndef barrier
  85. #define barrier() asm volatile("" ::: "memory")
  86. #endif
  87. /* Variable-specific compiler (optimization) barrier. It's a no-op which makes
  88. * compiler believe that there is some black box modification of a given
  89. * variable and thus prevents compiler from making extra assumption about its
  90. * value and potential simplifications and optimizations on this variable.
  91. *
  92. * E.g., compiler might often delay or even omit 32-bit to 64-bit casting of
  93. * a variable, making some code patterns unverifiable. Putting barrier_var()
  94. * in place will ensure that cast is performed before the barrier_var()
  95. * invocation, because compiler has to pessimistically assume that embedded
  96. * asm section might perform some extra operations on that variable.
  97. *
  98. * This is a variable-specific variant of more global barrier().
  99. */
  100. #ifndef barrier_var
  101. #define barrier_var(var) asm volatile("" : "+r"(var))
  102. #endif
  103. /*
  104. * Helper macro to throw a compilation error if __bpf_unreachable() gets
  105. * built into the resulting code. This works given BPF back end does not
  106. * implement __builtin_trap(). This is useful to assert that certain paths
  107. * of the program code are never used and hence eliminated by the compiler.
  108. *
  109. * For example, consider a switch statement that covers known cases used by
  110. * the program. __bpf_unreachable() can then reside in the default case. If
  111. * the program gets extended such that a case is not covered in the switch
  112. * statement, then it will throw a build error due to the default case not
  113. * being compiled out.
  114. */
  115. #ifndef __bpf_unreachable
  116. # define __bpf_unreachable() __builtin_trap()
  117. #endif
  118. /*
  119. * Helper function to perform a tail call with a constant/immediate map slot.
  120. */
  121. #if (defined(__clang__) && __clang_major__ >= 8) || (!defined(__clang__) && __GNUC__ > 12)
  122. #if defined(__bpf__)
  123. static __always_inline void
  124. bpf_tail_call_static(void *ctx, const void *map, const __u32 slot)
  125. {
  126. if (!__builtin_constant_p(slot))
  127. __bpf_unreachable();
  128. /*
  129. * Provide a hard guarantee that LLVM won't optimize setting r2 (map
  130. * pointer) and r3 (constant map index) from _different paths_ ending
  131. * up at the _same_ call insn as otherwise we won't be able to use the
  132. * jmpq/nopl retpoline-free patching by the x86-64 JIT in the kernel
  133. * given they mismatch. See also d2e4c1e6c294 ("bpf: Constant map key
  134. * tracking for prog array pokes") for details on verifier tracking.
  135. *
  136. * Note on clobber list: we need to stay in-line with BPF calling
  137. * convention, so even if we don't end up using r0, r4, r5, we need
  138. * to mark them as clobber so that LLVM doesn't end up using them
  139. * before / after the call.
  140. */
  141. asm volatile("r1 = %[ctx]\n\t"
  142. "r2 = %[map]\n\t"
  143. "r3 = %[slot]\n\t"
  144. "call 12"
  145. :: [ctx]"r"(ctx), [map]"r"(map), [slot]"i"(slot)
  146. : "r0", "r1", "r2", "r3", "r4", "r5");
  147. }
  148. #endif
  149. #endif
  150. enum libbpf_pin_type {
  151. LIBBPF_PIN_NONE,
  152. /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
  153. LIBBPF_PIN_BY_NAME,
  154. };
  155. enum libbpf_tristate {
  156. TRI_NO = 0,
  157. TRI_YES = 1,
  158. TRI_MODULE = 2,
  159. };
  160. #define __kconfig __attribute__((section(".kconfig")))
  161. #define __ksym __attribute__((section(".ksyms")))
  162. #define __kptr_untrusted __attribute__((btf_type_tag("kptr_untrusted")))
  163. #define __kptr __attribute__((btf_type_tag("kptr")))
  164. #define __percpu_kptr __attribute__((btf_type_tag("percpu_kptr")))
  165. #if defined (__clang__)
  166. #define bpf_ksym_exists(sym) ({ \
  167. _Static_assert(!__builtin_constant_p(!!sym), \
  168. #sym " should be marked as __weak"); \
  169. !!sym; \
  170. })
  171. #elif __GNUC__ > 8
  172. #define bpf_ksym_exists(sym) ({ \
  173. _Static_assert(__builtin_has_attribute (*sym, __weak__), \
  174. #sym " should be marked as __weak"); \
  175. !!sym; \
  176. })
  177. #else
  178. #define bpf_ksym_exists(sym) !!sym
  179. #endif
  180. #define __arg_ctx __attribute__((btf_decl_tag("arg:ctx")))
  181. #define __arg_nonnull __attribute((btf_decl_tag("arg:nonnull")))
  182. #define __arg_nullable __attribute((btf_decl_tag("arg:nullable")))
  183. #define __arg_trusted __attribute((btf_decl_tag("arg:trusted")))
  184. #define __arg_arena __attribute((btf_decl_tag("arg:arena")))
  185. #ifndef ___bpf_concat
  186. #define ___bpf_concat(a, b) a ## b
  187. #endif
  188. #ifndef ___bpf_apply
  189. #define ___bpf_apply(fn, n) ___bpf_concat(fn, n)
  190. #endif
  191. #ifndef ___bpf_nth
  192. #define ___bpf_nth(_, _1, _2, _3, _4, _5, _6, _7, _8, _9, _a, _b, _c, N, ...) N
  193. #endif
  194. #ifndef ___bpf_narg
  195. #define ___bpf_narg(...) \
  196. ___bpf_nth(_, ##__VA_ARGS__, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)
  197. #endif
  198. #define ___bpf_fill0(arr, p, x) do {} while (0)
  199. #define ___bpf_fill1(arr, p, x) arr[p] = x
  200. #define ___bpf_fill2(arr, p, x, args...) arr[p] = x; ___bpf_fill1(arr, p + 1, args)
  201. #define ___bpf_fill3(arr, p, x, args...) arr[p] = x; ___bpf_fill2(arr, p + 1, args)
  202. #define ___bpf_fill4(arr, p, x, args...) arr[p] = x; ___bpf_fill3(arr, p + 1, args)
  203. #define ___bpf_fill5(arr, p, x, args...) arr[p] = x; ___bpf_fill4(arr, p + 1, args)
  204. #define ___bpf_fill6(arr, p, x, args...) arr[p] = x; ___bpf_fill5(arr, p + 1, args)
  205. #define ___bpf_fill7(arr, p, x, args...) arr[p] = x; ___bpf_fill6(arr, p + 1, args)
  206. #define ___bpf_fill8(arr, p, x, args...) arr[p] = x; ___bpf_fill7(arr, p + 1, args)
  207. #define ___bpf_fill9(arr, p, x, args...) arr[p] = x; ___bpf_fill8(arr, p + 1, args)
  208. #define ___bpf_fill10(arr, p, x, args...) arr[p] = x; ___bpf_fill9(arr, p + 1, args)
  209. #define ___bpf_fill11(arr, p, x, args...) arr[p] = x; ___bpf_fill10(arr, p + 1, args)
  210. #define ___bpf_fill12(arr, p, x, args...) arr[p] = x; ___bpf_fill11(arr, p + 1, args)
  211. #define ___bpf_fill(arr, args...) \
  212. ___bpf_apply(___bpf_fill, ___bpf_narg(args))(arr, 0, args)
  213. /*
  214. * BPF_SEQ_PRINTF to wrap bpf_seq_printf to-be-printed values
  215. * in a structure.
  216. */
  217. #define BPF_SEQ_PRINTF(seq, fmt, args...) \
  218. ({ \
  219. static const char ___fmt[] = fmt; \
  220. unsigned long long ___param[___bpf_narg(args)]; \
  221. \
  222. _Pragma("GCC diagnostic push") \
  223. _Pragma("GCC diagnostic ignored \"-Wint-conversion\"") \
  224. ___bpf_fill(___param, args); \
  225. _Pragma("GCC diagnostic pop") \
  226. \
  227. bpf_seq_printf(seq, ___fmt, sizeof(___fmt), \
  228. ___param, sizeof(___param)); \
  229. })
  230. /*
  231. * BPF_SNPRINTF wraps the bpf_snprintf helper with variadic arguments instead of
  232. * an array of u64.
  233. */
  234. #define BPF_SNPRINTF(out, out_size, fmt, args...) \
  235. ({ \
  236. static const char ___fmt[] = fmt; \
  237. unsigned long long ___param[___bpf_narg(args)]; \
  238. \
  239. _Pragma("GCC diagnostic push") \
  240. _Pragma("GCC diagnostic ignored \"-Wint-conversion\"") \
  241. ___bpf_fill(___param, args); \
  242. _Pragma("GCC diagnostic pop") \
  243. \
  244. bpf_snprintf(out, out_size, ___fmt, \
  245. ___param, sizeof(___param)); \
  246. })
  247. #ifdef BPF_NO_GLOBAL_DATA
  248. #define BPF_PRINTK_FMT_MOD
  249. #else
  250. #define BPF_PRINTK_FMT_MOD static const
  251. #endif
  252. #define __bpf_printk(fmt, ...) \
  253. ({ \
  254. BPF_PRINTK_FMT_MOD char ____fmt[] = fmt; \
  255. bpf_trace_printk(____fmt, sizeof(____fmt), \
  256. ##__VA_ARGS__); \
  257. })
  258. /*
  259. * __bpf_vprintk wraps the bpf_trace_vprintk helper with variadic arguments
  260. * instead of an array of u64.
  261. */
  262. #define __bpf_vprintk(fmt, args...) \
  263. ({ \
  264. static const char ___fmt[] = fmt; \
  265. unsigned long long ___param[___bpf_narg(args)]; \
  266. \
  267. _Pragma("GCC diagnostic push") \
  268. _Pragma("GCC diagnostic ignored \"-Wint-conversion\"") \
  269. ___bpf_fill(___param, args); \
  270. _Pragma("GCC diagnostic pop") \
  271. \
  272. bpf_trace_vprintk(___fmt, sizeof(___fmt), \
  273. ___param, sizeof(___param)); \
  274. })
  275. /* Use __bpf_printk when bpf_printk call has 3 or fewer fmt args
  276. * Otherwise use __bpf_vprintk
  277. */
  278. #define ___bpf_pick_printk(...) \
  279. ___bpf_nth(_, ##__VA_ARGS__, __bpf_vprintk, __bpf_vprintk, __bpf_vprintk, \
  280. __bpf_vprintk, __bpf_vprintk, __bpf_vprintk, __bpf_vprintk, \
  281. __bpf_vprintk, __bpf_vprintk, __bpf_printk /*3*/, __bpf_printk /*2*/,\
  282. __bpf_printk /*1*/, __bpf_printk /*0*/)
  283. /* Helper macro to print out debug messages */
  284. #define bpf_printk(fmt, args...) ___bpf_pick_printk(args)(fmt, ##args)
  285. struct bpf_iter_num;
  286. extern int bpf_iter_num_new(struct bpf_iter_num *it, int start, int end) __weak __ksym;
  287. extern int *bpf_iter_num_next(struct bpf_iter_num *it) __weak __ksym;
  288. extern void bpf_iter_num_destroy(struct bpf_iter_num *it) __weak __ksym;
  289. #ifndef bpf_for_each
  290. /* bpf_for_each(iter_type, cur_elem, args...) provides generic construct for
  291. * using BPF open-coded iterators without having to write mundane explicit
  292. * low-level loop logic. Instead, it provides for()-like generic construct
  293. * that can be used pretty naturally. E.g., for some hypothetical cgroup
  294. * iterator, you'd write:
  295. *
  296. * struct cgroup *cg, *parent_cg = <...>;
  297. *
  298. * bpf_for_each(cgroup, cg, parent_cg, CG_ITER_CHILDREN) {
  299. * bpf_printk("Child cgroup id = %d", cg->cgroup_id);
  300. * if (cg->cgroup_id == 123)
  301. * break;
  302. * }
  303. *
  304. * I.e., it looks almost like high-level for each loop in other languages,
  305. * supports continue/break, and is verifiable by BPF verifier.
  306. *
  307. * For iterating integers, the difference between bpf_for_each(num, i, N, M)
  308. * and bpf_for(i, N, M) is in that bpf_for() provides additional proof to
  309. * verifier that i is in [N, M) range, and in bpf_for_each() case i is `int
  310. * *`, not just `int`. So for integers bpf_for() is more convenient.
  311. *
  312. * Note: this macro relies on C99 feature of allowing to declare variables
  313. * inside for() loop, bound to for() loop lifetime. It also utilizes GCC
  314. * extension: __attribute__((cleanup(<func>))), supported by both GCC and
  315. * Clang.
  316. */
  317. #define bpf_for_each(type, cur, args...) for ( \
  318. /* initialize and define destructor */ \
  319. struct bpf_iter_##type ___it __attribute__((aligned(8), /* enforce, just in case */, \
  320. cleanup(bpf_iter_##type##_destroy))), \
  321. /* ___p pointer is just to call bpf_iter_##type##_new() *once* to init ___it */ \
  322. *___p __attribute__((unused)) = ( \
  323. bpf_iter_##type##_new(&___it, ##args), \
  324. /* this is a workaround for Clang bug: it currently doesn't emit BTF */ \
  325. /* for bpf_iter_##type##_destroy() when used from cleanup() attribute */ \
  326. (void)bpf_iter_##type##_destroy, (void *)0); \
  327. /* iteration and termination check */ \
  328. (((cur) = bpf_iter_##type##_next(&___it))); \
  329. )
  330. #endif /* bpf_for_each */
  331. #ifndef bpf_for
  332. /* bpf_for(i, start, end) implements a for()-like looping construct that sets
  333. * provided integer variable *i* to values starting from *start* through,
  334. * but not including, *end*. It also proves to BPF verifier that *i* belongs
  335. * to range [start, end), so this can be used for accessing arrays without
  336. * extra checks.
  337. *
  338. * Note: *start* and *end* are assumed to be expressions with no side effects
  339. * and whose values do not change throughout bpf_for() loop execution. They do
  340. * not have to be statically known or constant, though.
  341. *
  342. * Note: similarly to bpf_for_each(), it relies on C99 feature of declaring for()
  343. * loop bound variables and cleanup attribute, supported by GCC and Clang.
  344. */
  345. #define bpf_for(i, start, end) for ( \
  346. /* initialize and define destructor */ \
  347. struct bpf_iter_num ___it __attribute__((aligned(8), /* enforce, just in case */ \
  348. cleanup(bpf_iter_num_destroy))), \
  349. /* ___p pointer is necessary to call bpf_iter_num_new() *once* to init ___it */ \
  350. *___p __attribute__((unused)) = ( \
  351. bpf_iter_num_new(&___it, (start), (end)), \
  352. /* this is a workaround for Clang bug: it currently doesn't emit BTF */ \
  353. /* for bpf_iter_num_destroy() when used from cleanup() attribute */ \
  354. (void)bpf_iter_num_destroy, (void *)0); \
  355. ({ \
  356. /* iteration step */ \
  357. int *___t = bpf_iter_num_next(&___it); \
  358. /* termination and bounds check */ \
  359. (___t && ((i) = *___t, (i) >= (start) && (i) < (end))); \
  360. }); \
  361. )
  362. #endif /* bpf_for */
  363. #ifndef bpf_repeat
  364. /* bpf_repeat(N) performs N iterations without exposing iteration number
  365. *
  366. * Note: similarly to bpf_for_each(), it relies on C99 feature of declaring for()
  367. * loop bound variables and cleanup attribute, supported by GCC and Clang.
  368. */
  369. #define bpf_repeat(N) for ( \
  370. /* initialize and define destructor */ \
  371. struct bpf_iter_num ___it __attribute__((aligned(8), /* enforce, just in case */ \
  372. cleanup(bpf_iter_num_destroy))), \
  373. /* ___p pointer is necessary to call bpf_iter_num_new() *once* to init ___it */ \
  374. *___p __attribute__((unused)) = ( \
  375. bpf_iter_num_new(&___it, 0, (N)), \
  376. /* this is a workaround for Clang bug: it currently doesn't emit BTF */ \
  377. /* for bpf_iter_num_destroy() when used from cleanup() attribute */ \
  378. (void)bpf_iter_num_destroy, (void *)0); \
  379. bpf_iter_num_next(&___it); \
  380. /* nothing here */ \
  381. )
  382. #endif /* bpf_repeat */
  383. #endif