tools/lib/bpf/bpf_helpers.h

   1 /* SPDX-License-Identifier: (LGPL-2.1 OR BSD-2-Clause) */
   2 #ifndef __BPF_HELPERS__
   3 #define __BPF_HELPERS__
   4
   5 /*
   6  * Note that bpf programs need to include either
   7  * vmlinux.h (auto-generated from BTF) or linux/types.h
   8  * in advance since bpf_helper_defs.h uses such types
   9  * as __u64.
  10  */
  11 #include "bpf_helper_defs.h"
  12
  13 #define __uint(name, val) int (*name)[val]
  14 #define __type(name, val) typeof(val) *name
  15 #define __array(name, val) typeof(val) *name[]
  16 #define __ulong(name, val) enum { ___bpf_concat(__unique_value, __COUNTER__) = val } name
  17
  18 /*
  19  * Helper macro to place programs, maps, license in
  20  * different sections in elf_bpf file. Section names
  21  * are interpreted by libbpf depending on the context (BPF programs, BPF maps,
  22  * extern variables, etc).
  23  * To allow use of SEC() with externs (e.g., for extern .maps declarations),
  24  * make sure __attribute__((unused)) doesn't trigger compilation warning.
  25  */
  26 #if __GNUC__ && !__clang__
  27
  28 /*
  29  * Pragma macros are broken on GCC
  30  * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=55578
  31  * https://gcc.gnu.org/bugzilla/show_bug.cgi?id=90400
  32  */
  33 #define SEC(name) __attribute__((section(name), used))
  34
  35 #else
  36
  37 #define SEC(name) \
  38         _Pragma("GCC diagnostic push")                                      \
  39         _Pragma("GCC diagnostic ignored \"-Wignored-attributes\"")          \
  40         __attribute__((section(name), used))                                \
  41         _Pragma("GCC diagnostic pop")                                       \
  42
  43 #endif
  44
  45 /* Avoid 'linux/stddef.h' definition of '__always_inline'. */
  46 #undef __always_inline
  47 #define __always_inline inline __attribute__((always_inline))
  48
  49 #ifndef __noinline
  50 #define __noinline __attribute__((noinline))
  51 #endif
  52 #ifndef __weak
  53 #define __weak __attribute__((weak))
  54 #endif
  55
  56 /*
  57  * Use __hidden attribute to mark a non-static BPF subprogram effectively
  58  * static for BPF verifier's verification algorithm purposes, allowing more
  59  * extensive and permissive BPF verification process, taking into account
  60  * subprogram's caller context.
  61  */
  62 #define __hidden __attribute__((visibility("hidden")))
  63
  64 /* When utilizing vmlinux.h with BPF CO-RE, user BPF programs can't include
  65  * any system-level headers (such as stddef.h, linux/version.h, etc), and
  66  * commonly-used macros like NULL and KERNEL_VERSION aren't available through
  67  * vmlinux.h. This just adds unnecessary hurdles and forces users to re-define
  68  * them on their own. So as a convenience, provide such definitions here.
  69  */
  70 #ifndef NULL
  71 #define NULL ((void *)0)
  72 #endif
  73
  74 #ifndef KERNEL_VERSION
  75 #define KERNEL_VERSION(a, b, c) (((a) << 16) + ((b) << 8) + ((c) > 255 ? 255 : (c)))
  76 #endif
  77
  78 /*
  79  * Helper macros to manipulate data structures
  80  */
  81
  82 /* offsetof() definition that uses __builtin_offset() might not preserve field
  83  * offset CO-RE relocation properly, so force-redefine offsetof() using
  84  * old-school approach which works with CO-RE correctly
  85  */
  86 #undef offsetof
  87 #define offsetof(type, member)  ((unsigned long)&((type *)0)->member)
  88
  89 /* redefined container_of() to ensure we use the above offsetof() macro */
  90 #undef container_of
  91 #define container_of(ptr, type, member)                         \
  92         ({                                                      \
  93                 void *__mptr = (void *)(ptr);                   \
  94                 ((type *)(__mptr - offsetof(type, member)));    \
  95         })
  96
  97 /*
  98  * Compiler (optimization) barrier.
  99  */
 100 #ifndef barrier
 101 #define barrier() asm volatile("" ::: "memory")
 102 #endif
 103
 104 /* Variable-specific compiler (optimization) barrier. It's a no-op which makes
 105  * compiler believe that there is some black box modification of a given
 106  * variable and thus prevents compiler from making extra assumption about its
 107  * value and potential simplifications and optimizations on this variable.
 108  *
 109  * E.g., compiler might often delay or even omit 32-bit to 64-bit casting of
 110  * a variable, making some code patterns unverifiable. Putting barrier_var()
 111  * in place will ensure that cast is performed before the barrier_var()
 112  * invocation, because compiler has to pessimistically assume that embedded
 113  * asm section might perform some extra operations on that variable.
 114  *
 115  * This is a variable-specific variant of more global barrier().
 116  */
 117 #ifndef barrier_var
 118 #define barrier_var(var) asm volatile("" : "+r"(var))
 119 #endif
 120
 121 /*
 122  * Helper macro to throw a compilation error if __bpf_unreachable() gets
 123  * built into the resulting code. This works given BPF back end does not
 124  * implement __builtin_trap(). This is useful to assert that certain paths
 125  * of the program code are never used and hence eliminated by the compiler.
 126  *
 127  * For example, consider a switch statement that covers known cases used by
 128  * the program. __bpf_unreachable() can then reside in the default case. If
 129  * the program gets extended such that a case is not covered in the switch
 130  * statement, then it will throw a build error due to the default case not
 131  * being compiled out.
 132  */
 133 #ifndef __bpf_unreachable
 134 # define __bpf_unreachable()    __builtin_trap()
 135 #endif
 136
 137 /*
 138  * Helper function to perform a tail call with a constant/immediate map slot.
 139  */
 140 #if (defined(__clang__) && __clang_major__ >= 8) || (!defined(__clang__) && __GNUC__ > 12)
 141 #if defined(__bpf__)
 142 static __always_inline void
 143 bpf_tail_call_static(void *ctx, const void *map, const __u32 slot)
 144 {
 145         if (!__builtin_constant_p(slot))
 146                 __bpf_unreachable();
 147
 148         /*
 149          * Provide a hard guarantee that LLVM won't optimize setting r2 (map
 150          * pointer) and r3 (constant map index) from _different paths_ ending
 151          * up at the _same_ call insn as otherwise we won't be able to use the
 152          * jmpq/nopl retpoline-free patching by the x86-64 JIT in the kernel
 153          * given they mismatch. See also d2e4c1e6c294 ("bpf: Constant map key
 154          * tracking for prog array pokes") for details on verifier tracking.
 155          *
 156          * Note on clobber list: we need to stay in-line with BPF calling
 157          * convention, so even if we don't end up using r0, r4, r5, we need
 158          * to mark them as clobber so that LLVM doesn't end up using them
 159          * before / after the call.
 160          */
 161         asm volatile("r1 = %[ctx]\n\t"
 162                      "r2 = %[map]\n\t"
 163                      "r3 = %[slot]\n\t"
 164                      "call 12"
 165                      :: [ctx]"r"(ctx), [map]"r"(map), [slot]"i"(slot)
 166                      : "r0", "r1", "r2", "r3", "r4", "r5");
 167 }
 168 #endif
 169 #endif
 170
 171 enum libbpf_pin_type {
 172         LIBBPF_PIN_NONE,
 173         /* PIN_BY_NAME: pin maps by name (in /sys/fs/bpf by default) */
 174         LIBBPF_PIN_BY_NAME,
 175 };
 176
 177 enum libbpf_tristate {
 178         TRI_NO = 0,
 179         TRI_YES = 1,
 180         TRI_MODULE = 2,
 181 };
 182
 183 #define __kconfig __attribute__((section(".kconfig")))
 184 #define __ksym __attribute__((section(".ksyms")))
 185 #define __kptr_untrusted __attribute__((btf_type_tag("kptr_untrusted")))
 186 #define __kptr __attribute__((btf_type_tag("kptr")))
 187 #define __percpu_kptr __attribute__((btf_type_tag("percpu_kptr")))
 188 #define __uptr __attribute__((btf_type_tag("uptr")))
 189
 190 #if defined (__clang__)
 191 #define bpf_ksym_exists(sym) ({                                         \
 192         _Static_assert(!__builtin_constant_p(!!sym),                    \
 193                        #sym " should be marked as __weak");             \
 194         !!sym;                                                          \
 195 })
 196 #elif __GNUC__ > 8
 197 #define bpf_ksym_exists(sym) ({                                         \
 198         _Static_assert(__builtin_has_attribute (*sym, __weak__),        \
 199                        #sym " should be marked as __weak");             \
 200         !!sym;                                                          \
 201 })
 202 #else
 203 #define bpf_ksym_exists(sym) !!sym
 204 #endif
 205
 206 #define __arg_ctx __attribute__((btf_decl_tag("arg:ctx")))
 207 #define __arg_nonnull __attribute((btf_decl_tag("arg:nonnull")))
 208 #define __arg_nullable __attribute((btf_decl_tag("arg:nullable")))
 209 #define __arg_trusted __attribute((btf_decl_tag("arg:trusted")))
 210 #define __arg_arena __attribute((btf_decl_tag("arg:arena")))
 211
 212 #ifndef ___bpf_concat
 213 #define ___bpf_concat(a, b) a ## b
 214 #endif
 215 #ifndef ___bpf_apply
 216 #define ___bpf_apply(fn, n) ___bpf_concat(fn, n)
 217 #endif
 218 #ifndef ___bpf_nth
 219 #define ___bpf_nth(_, _1, _2, _3, _4, _5, _6, _7, _8, _9, _a, _b, _c, N, ...) N
 220 #endif
 221 #ifndef ___bpf_narg
 222 #define ___bpf_narg(...) \
 223         ___bpf_nth(_, ##__VA_ARGS__, 12, 11, 10, 9, 8, 7, 6, 5, 4, 3, 2, 1, 0)
 224 #endif
 225
 226 #define ___bpf_fill0(arr, p, x) do {} while (0)
 227 #define ___bpf_fill1(arr, p, x) arr[p] = x
 228 #define ___bpf_fill2(arr, p, x, args...) arr[p] = x; ___bpf_fill1(arr, p + 1, args)
 229 #define ___bpf_fill3(arr, p, x, args...) arr[p] = x; ___bpf_fill2(arr, p + 1, args)
 230 #define ___bpf_fill4(arr, p, x, args...) arr[p] = x; ___bpf_fill3(arr, p + 1, args)
 231 #define ___bpf_fill5(arr, p, x, args...) arr[p] = x; ___bpf_fill4(arr, p + 1, args)
 232 #define ___bpf_fill6(arr, p, x, args...) arr[p] = x; ___bpf_fill5(arr, p + 1, args)
 233 #define ___bpf_fill7(arr, p, x, args...) arr[p] = x; ___bpf_fill6(arr, p + 1, args)
 234 #define ___bpf_fill8(arr, p, x, args...) arr[p] = x; ___bpf_fill7(arr, p + 1, args)
 235 #define ___bpf_fill9(arr, p, x, args...) arr[p] = x; ___bpf_fill8(arr, p + 1, args)
 236 #define ___bpf_fill10(arr, p, x, args...) arr[p] = x; ___bpf_fill9(arr, p + 1, args)
 237 #define ___bpf_fill11(arr, p, x, args...) arr[p] = x; ___bpf_fill10(arr, p + 1, args)
 238 #define ___bpf_fill12(arr, p, x, args...) arr[p] = x; ___bpf_fill11(arr, p + 1, args)
 239 #define ___bpf_fill(arr, args...) \
 240         ___bpf_apply(___bpf_fill, ___bpf_narg(args))(arr, 0, args)
 241
 242 /*
 243  * BPF_SEQ_PRINTF to wrap bpf_seq_printf to-be-printed values
 244  * in a structure.
 245  */
 246 #define BPF_SEQ_PRINTF(seq, fmt, args...)                       \
 247 ({                                                              \
 248         static const char ___fmt[] = fmt;                       \
 249         unsigned long long ___param[___bpf_narg(args)];         \
 250                                                                 \
 251         _Pragma("GCC diagnostic push")                          \
 252         _Pragma("GCC diagnostic ignored \"-Wint-conversion\"")  \
 253         ___bpf_fill(___param, args);                            \
 254         _Pragma("GCC diagnostic pop")                           \
 255                                                                 \
 256         bpf_seq_printf(seq, ___fmt, sizeof(___fmt),             \
 257                        ___param, sizeof(___param));             \
 258 })
 259
 260 /*
 261  * BPF_SNPRINTF wraps the bpf_snprintf helper with variadic arguments instead of
 262  * an array of u64.
 263  */
 264 #define BPF_SNPRINTF(out, out_size, fmt, args...)               \
 265 ({                                                              \
 266         static const char ___fmt[] = fmt;                       \
 267         unsigned long long ___param[___bpf_narg(args)];         \
 268                                                                 \
 269         _Pragma("GCC diagnostic push")                          \
 270         _Pragma("GCC diagnostic ignored \"-Wint-conversion\"")  \
 271         ___bpf_fill(___param, args);                            \
 272         _Pragma("GCC diagnostic pop")                           \
 273                                                                 \
 274         bpf_snprintf(out, out_size, ___fmt,                     \
 275                      ___param, sizeof(___param));               \
 276 })
 277
 278 #ifdef BPF_NO_GLOBAL_DATA
 279 #define BPF_PRINTK_FMT_MOD
 280 #else
 281 #define BPF_PRINTK_FMT_MOD static const
 282 #endif
 283
 284 #define __bpf_printk(fmt, ...)                          \
 285 ({                                                      \
 286         BPF_PRINTK_FMT_MOD char ____fmt[] = fmt;        \
 287         bpf_trace_printk(____fmt, sizeof(____fmt),      \
 288                          ##__VA_ARGS__);                \
 289 })
 290
 291 /*
 292  * __bpf_vprintk wraps the bpf_trace_vprintk helper with variadic arguments
 293  * instead of an array of u64.
 294  */
 295 #define __bpf_vprintk(fmt, args...)                             \
 296 ({                                                              \
 297         static const char ___fmt[] = fmt;                       \
 298         unsigned long long ___param[___bpf_narg(args)];         \
 299                                                                 \
 300         _Pragma("GCC diagnostic push")                          \
 301         _Pragma("GCC diagnostic ignored \"-Wint-conversion\"")  \
 302         ___bpf_fill(___param, args);                            \
 303         _Pragma("GCC diagnostic pop")                           \
 304                                                                 \
 305         bpf_trace_vprintk(___fmt, sizeof(___fmt),               \
 306                           ___param, sizeof(___param));          \
 307 })
 308
 309 /* Use __bpf_printk when bpf_printk call has 3 or fewer fmt args
 310  * Otherwise use __bpf_vprintk
 311  */
 312 #define ___bpf_pick_printk(...) \
 313         ___bpf_nth(_, ##__VA_ARGS__, __bpf_vprintk, __bpf_vprintk, __bpf_vprintk,       \
 314                    __bpf_vprintk, __bpf_vprintk, __bpf_vprintk, __bpf_vprintk,          \
 315                    __bpf_vprintk, __bpf_vprintk, __bpf_printk /*3*/, __bpf_printk /*2*/,\
 316                    __bpf_printk /*1*/, __bpf_printk /*0*/)
 317
 318 /* Helper macro to print out debug messages */
 319 #define bpf_printk(fmt, args...) ___bpf_pick_printk(args)(fmt, ##args)
 320
 321 struct bpf_iter_num;
 322
 323 extern int bpf_iter_num_new(struct bpf_iter_num *it, int start, int end) __weak __ksym;
 324 extern int *bpf_iter_num_next(struct bpf_iter_num *it) __weak __ksym;
 325 extern void bpf_iter_num_destroy(struct bpf_iter_num *it) __weak __ksym;
 326
 327 #ifndef bpf_for_each
 328 /* bpf_for_each(iter_type, cur_elem, args...) provides generic construct for
 329  * using BPF open-coded iterators without having to write mundane explicit
 330  * low-level loop logic. Instead, it provides for()-like generic construct
 331  * that can be used pretty naturally. E.g., for some hypothetical cgroup
 332  * iterator, you'd write:
 333  *
 334  * struct cgroup *cg, *parent_cg = <...>;
 335  *
 336  * bpf_for_each(cgroup, cg, parent_cg, CG_ITER_CHILDREN) {
 337  *     bpf_printk("Child cgroup id = %d", cg->cgroup_id);
 338  *     if (cg->cgroup_id == 123)
 339  *         break;
 340  * }
 341  *
 342  * I.e., it looks almost like high-level for each loop in other languages,
 343  * supports continue/break, and is verifiable by BPF verifier.
 344  *
 345  * For iterating integers, the difference between bpf_for_each(num, i, N, M)
 346  * and bpf_for(i, N, M) is in that bpf_for() provides additional proof to
 347  * verifier that i is in [N, M) range, and in bpf_for_each() case i is `int
 348  * *`, not just `int`. So for integers bpf_for() is more convenient.
 349  *
 350  * Note: this macro relies on C99 feature of allowing to declare variables
 351  * inside for() loop, bound to for() loop lifetime. It also utilizes GCC
 352  * extension: __attribute__((cleanup(<func>))), supported by both GCC and
 353  * Clang.
 354  */
 355 #define bpf_for_each(type, cur, args...) for (                                                  \
 356         /* initialize and define destructor */                                                  \
 357         struct bpf_iter_##type ___it __attribute__((aligned(8), /* enforce, just in case */,    \
 358                                                     cleanup(bpf_iter_##type##_destroy))),       \
 359         /* ___p pointer is just to call bpf_iter_##type##_new() *once* to init ___it */         \
 360                                *___p __attribute__((unused)) = (                                \
 361                                         bpf_iter_##type##_new(&___it, ##args),                  \
 362         /* this is a workaround for Clang bug: it currently doesn't emit BTF */                 \
 363         /* for bpf_iter_##type##_destroy() when used from cleanup() attribute */                \
 364                                         (void)bpf_iter_##type##_destroy, (void *)0);            \
 365         /* iteration and termination check */                                                   \
 366         (((cur) = bpf_iter_##type##_next(&___it)));                                             \
 367 )
 368 #endif /* bpf_for_each */
 369
 370 #ifndef bpf_for
 371 /* bpf_for(i, start, end) implements a for()-like looping construct that sets
 372  * provided integer variable *i* to values starting from *start* through,
 373  * but not including, *end*. It also proves to BPF verifier that *i* belongs
 374  * to range [start, end), so this can be used for accessing arrays without
 375  * extra checks.
 376  *
 377  * Note: *start* and *end* are assumed to be expressions with no side effects
 378  * and whose values do not change throughout bpf_for() loop execution. They do
 379  * not have to be statically known or constant, though.
 380  *
 381  * Note: similarly to bpf_for_each(), it relies on C99 feature of declaring for()
 382  * loop bound variables and cleanup attribute, supported by GCC and Clang.
 383  */
 384 #define bpf_for(i, start, end) for (                                                            \
 385         /* initialize and define destructor */                                                  \
 386         struct bpf_iter_num ___it __attribute__((aligned(8), /* enforce, just in case */        \
 387                                                  cleanup(bpf_iter_num_destroy))),               \
 388         /* ___p pointer is necessary to call bpf_iter_num_new() *once* to init ___it */         \
 389                             *___p __attribute__((unused)) = (                                   \
 390                                 bpf_iter_num_new(&___it, (start), (end)),                       \
 391         /* this is a workaround for Clang bug: it currently doesn't emit BTF */                 \
 392         /* for bpf_iter_num_destroy() when used from cleanup() attribute */                     \
 393                                 (void)bpf_iter_num_destroy, (void *)0);                         \
 394         ({                                                                                      \
 395                 /* iteration step */                                                            \
 396                 int *___t = bpf_iter_num_next(&___it);                                          \
 397                 /* termination and bounds check */                                              \
 398                 (___t && ((i) = *___t, (i) >= (start) && (i) < (end)));                         \
 399         });                                                                                     \
 400 )
 401 #endif /* bpf_for */
 402
 403 #ifndef bpf_repeat
 404 /* bpf_repeat(N) performs N iterations without exposing iteration number
 405  *
 406  * Note: similarly to bpf_for_each(), it relies on C99 feature of declaring for()
 407  * loop bound variables and cleanup attribute, supported by GCC and Clang.
 408  */
 409 #define bpf_repeat(N) for (                                                                     \
 410         /* initialize and define destructor */                                                  \
 411         struct bpf_iter_num ___it __attribute__((aligned(8), /* enforce, just in case */        \
 412                                                  cleanup(bpf_iter_num_destroy))),               \
 413         /* ___p pointer is necessary to call bpf_iter_num_new() *once* to init ___it */         \
 414                             *___p __attribute__((unused)) = (                                   \
 415                                 bpf_iter_num_new(&___it, 0, (N)),                               \
 416         /* this is a workaround for Clang bug: it currently doesn't emit BTF */                 \
 417         /* for bpf_iter_num_destroy() when used from cleanup() attribute */                     \
 418                                 (void)bpf_iter_num_destroy, (void *)0);                         \
 419         bpf_iter_num_next(&___it);                                                              \
 420         /* nothing here  */                                                                     \
 421 )
 422 #endif /* bpf_repeat */
 423
 424 #endif