| 1 | |
| 2 | * parse_vdso.c: Linux reference vDSO parser
|
| 3 | * Written by Andrew Lutomirski, 2011-2014.
|
| 4 | *
|
| 5 | * This code is meant to be linked in to various programs that run on Linux.
|
| 6 | * As such, it is available with as few restrictions as possible. This file
|
| 7 | * is licensed under the Creative Commons Zero License, version 1.0,
|
| 8 | * available at http://creativecommons.org/publicdomain/zero/1.0/legalcode
|
| 9 | *
|
| 10 | * The vDSO is a regular ELF DSO that the kernel maps into user space when
|
| 11 | * it starts a program. It works equally well in statically and dynamically
|
| 12 | * linked binaries.
|
| 13 | *
|
| 14 | * This code is tested on x86. In principle it should work on any
|
| 15 | * architecture that has a vDSO.
|
| 16 |
|
| 17 |
|
| 18 | #include <stdbool.h>
|
| 19 | #include <stdint.h>
|
| 20 | #include <string.h>
|
| 21 | #include <limits.h>
|
| 22 | #include <elf.h>
|
| 23 |
|
| 24 | #include "parse_vdso.h"
|
| 25 |
|
| 26 |
|
| 27 | #ifndef ELF_BITS
|
| 28 | # if __SIZEOF_LONG__ >= 8
|
| 29 | # define ELF_BITS 64
|
| 30 | # else
|
| 31 | # define ELF_BITS 32
|
| 32 | # endif
|
| 33 | #endif
|
| 34 |
|
| 35 | #define ELF_BITS_XFORM2(bits, x) Elf##bits##_##x
|
| 36 | #define ELF_BITS_XFORM(bits, x) ELF_BITS_XFORM2(bits, x)
|
| 37 | #define ELF(x) ELF_BITS_XFORM(ELF_BITS, x)
|
| 38 |
|
| 39 | #ifdef __s390x__
|
| 40 | #define ELF_HASH_ENTRY ELF(Xword)
|
| 41 | #else
|
| 42 | #define ELF_HASH_ENTRY ELF(Word)
|
| 43 | #endif
|
| 44 |
|
| 45 | static struct vdso_info
|
| 46 | {
|
| 47 | bool valid;
|
| 48 |
|
| 49 |
|
| 50 | uintptr_t load_addr;
|
| 51 | uintptr_t load_offset;
|
| 52 |
|
| 53 |
|
| 54 | ELF(Sym) *symtab;
|
| 55 | const char *symstrings;
|
| 56 | ELF(Word) *gnu_hash, *gnu_bucket;
|
| 57 | ELF_HASH_ENTRY *bucket, *chain;
|
| 58 | ELF_HASH_ENTRY nbucket, nchain;
|
| 59 |
|
| 60 |
|
| 61 | ELF(Versym) *versym;
|
| 62 | ELF(Verdef) *verdef;
|
| 63 | } vdso_info;
|
| 64 |
|
| 65 | |
| 66 | * Straight from the ELF specification...and then tweaked slightly, in order to
|
| 67 | * avoid a few clang warnings.
|
| 68 |
|
| 69 | static unsigned long elf_hash(const char *name)
|
| 70 | {
|
| 71 | unsigned long h = 0, g;
|
| 72 | const unsigned char *uch_name = (const unsigned char *)name;
|
| 73 |
|
| 74 | while (*uch_name)
|
| 75 | {
|
| 76 | h = (h << 4) + *uch_name++;
|
| 77 | g = h & 0xf0000000;
|
| 78 | if (g)
|
| 79 | h ^= g >> 24;
|
| 80 | h &= ~g;
|
| 81 | }
|
| 82 | return h;
|
| 83 | }
|
| 84 |
|
| 85 | static uint32_t gnu_hash(const char *name)
|
| 86 | {
|
| 87 | const unsigned char *s = (void *)name;
|
| 88 | uint32_t h = 5381;
|
| 89 |
|
| 90 | for (; *s; s++)
|
| 91 | h += h * 32 + *s;
|
| 92 | return h;
|
| 93 | }
|
| 94 |
|
| 95 | void vdso_init_from_sysinfo_ehdr(uintptr_t base)
|
| 96 | {
|
| 97 | size_t i;
|
| 98 | bool found_vaddr = false;
|
| 99 |
|
| 100 | vdso_info.valid = false;
|
| 101 |
|
| 102 | vdso_info.load_addr = base;
|
| 103 |
|
| 104 | ELF(Ehdr) *hdr = (ELF(Ehdr)*)base;
|
| 105 | if (hdr->e_ident[EI_CLASS] !=
|
| 106 | (ELF_BITS == 32 ? ELFCLASS32 : ELFCLASS64)) {
|
| 107 | return;
|
| 108 | }
|
| 109 |
|
| 110 | ELF(Phdr) *pt = (ELF(Phdr)*)(vdso_info.load_addr + hdr->e_phoff);
|
| 111 | ELF(Dyn) *dyn = 0;
|
| 112 |
|
| 113 | |
| 114 | * We need two things from the segment table: the load offset
|
| 115 | * and the dynamic table.
|
| 116 |
|
| 117 | for (i = 0; i < hdr->e_phnum; i++)
|
| 118 | {
|
| 119 | if (pt[i].p_type == PT_LOAD && !found_vaddr) {
|
| 120 | found_vaddr = true;
|
| 121 | vdso_info.load_offset = base
|
| 122 | + (uintptr_t)pt[i].p_offset
|
| 123 | - (uintptr_t)pt[i].p_vaddr;
|
| 124 | } else if (pt[i].p_type == PT_DYNAMIC) {
|
| 125 | dyn = (ELF(Dyn)*)(base + pt[i].p_offset);
|
| 126 | }
|
| 127 | }
|
| 128 |
|
| 129 | if (!found_vaddr || !dyn)
|
| 130 | return;
|
| 131 |
|
| 132 | |
| 133 | * Fish out the useful bits of the dynamic table.
|
| 134 |
|
| 135 | ELF_HASH_ENTRY *hash = 0;
|
| 136 | vdso_info.symstrings = 0;
|
| 137 | vdso_info.gnu_hash = 0;
|
| 138 | vdso_info.symtab = 0;
|
| 139 | vdso_info.versym = 0;
|
| 140 | vdso_info.verdef = 0;
|
| 141 | for (i = 0; dyn[i].d_tag != DT_NULL; i++) {
|
| 142 | switch (dyn[i].d_tag) {
|
| 143 | case DT_STRTAB:
|
| 144 | vdso_info.symstrings = (const char *)
|
| 145 | ((uintptr_t)dyn[i].d_un.d_ptr
|
| 146 | + vdso_info.load_offset);
|
| 147 | break;
|
| 148 | case DT_SYMTAB:
|
| 149 | vdso_info.symtab = (ELF(Sym) *)
|
| 150 | ((uintptr_t)dyn[i].d_un.d_ptr
|
| 151 | + vdso_info.load_offset);
|
| 152 | break;
|
| 153 | case DT_HASH:
|
| 154 | hash = (ELF_HASH_ENTRY *)
|
| 155 | ((uintptr_t)dyn[i].d_un.d_ptr
|
| 156 | + vdso_info.load_offset);
|
| 157 | break;
|
| 158 | case DT_GNU_HASH:
|
| 159 | vdso_info.gnu_hash =
|
| 160 | (ELF(Word) *)((uintptr_t)dyn[i].d_un.d_ptr +
|
| 161 | vdso_info.load_offset);
|
| 162 | break;
|
| 163 | case DT_VERSYM:
|
| 164 | vdso_info.versym = (ELF(Versym) *)
|
| 165 | ((uintptr_t)dyn[i].d_un.d_ptr
|
| 166 | + vdso_info.load_offset);
|
| 167 | break;
|
| 168 | case DT_VERDEF:
|
| 169 | vdso_info.verdef = (ELF(Verdef) *)
|
| 170 | ((uintptr_t)dyn[i].d_un.d_ptr
|
| 171 | + vdso_info.load_offset);
|
| 172 | break;
|
| 173 | }
|
| 174 | }
|
| 175 | if (!vdso_info.symstrings || !vdso_info.symtab ||
|
| 176 | (!hash && !vdso_info.gnu_hash))
|
| 177 | return;
|
| 178 |
|
| 179 | if (!vdso_info.verdef)
|
| 180 | vdso_info.versym = 0;
|
| 181 |
|
| 182 |
|
| 183 | if (vdso_info.gnu_hash) {
|
| 184 | vdso_info.nbucket = vdso_info.gnu_hash[0];
|
| 185 | |
| 186 | * filter (size_t array of gnu_hash[2] elements).
|
| 187 |
|
| 188 | vdso_info.gnu_bucket = vdso_info.gnu_hash + 4 +
|
| 189 | sizeof(size_t) / 4 * vdso_info.gnu_hash[2];
|
| 190 | } else {
|
| 191 | vdso_info.nbucket = hash[0];
|
| 192 | vdso_info.nchain = hash[1];
|
| 193 | vdso_info.bucket = &hash[2];
|
| 194 | vdso_info.chain = &hash[vdso_info.nbucket + 2];
|
| 195 | }
|
| 196 |
|
| 197 |
|
| 198 | vdso_info.valid = true;
|
| 199 | }
|
| 200 |
|
| 201 | static bool vdso_match_version(ELF(Versym) ver,
|
| 202 | const char *name, ELF(Word) hash)
|
| 203 | {
|
| 204 | |
| 205 | * This is a helper function to check if the version indexed by
|
| 206 | * ver matches name (which hashes to hash).
|
| 207 | *
|
| 208 | * The version definition table is a mess, and I don't know how
|
| 209 | * to do this in better than linear time without allocating memory
|
| 210 | * to build an index. I also don't know why the table has
|
| 211 | * variable size entries in the first place.
|
| 212 | *
|
| 213 | * For added fun, I can't find a comprehensible specification of how
|
| 214 | * to parse all the weird flags in the table.
|
| 215 | *
|
| 216 | * So I just parse the whole table every time.
|
| 217 |
|
| 218 |
|
| 219 |
|
| 220 | ver &= 0x7fff;
|
| 221 | ELF(Verdef) *def = vdso_info.verdef;
|
| 222 | while(true) {
|
| 223 | if ((def->vd_flags & VER_FLG_BASE) == 0
|
| 224 | && (def->vd_ndx & 0x7fff) == ver)
|
| 225 | break;
|
| 226 |
|
| 227 | if (def->vd_next == 0)
|
| 228 | return false;
|
| 229 |
|
| 230 | def = (ELF(Verdef) *)((char *)def + def->vd_next);
|
| 231 | }
|
| 232 |
|
| 233 |
|
| 234 | ELF(Verdaux) *aux = (ELF(Verdaux)*)((char *)def + def->vd_aux);
|
| 235 | return def->vd_hash == hash
|
| 236 | && !strcmp(name, vdso_info.symstrings + aux->vda_name);
|
| 237 | }
|
| 238 |
|
| 239 | static bool check_sym(ELF(Sym) *sym, ELF(Word) i, const char *name,
|
| 240 | const char *version, unsigned long ver_hash)
|
| 241 | {
|
| 242 |
|
| 243 | if (ELF64_ST_TYPE(sym->st_info) != STT_FUNC)
|
| 244 | return false;
|
| 245 | if (ELF64_ST_BIND(sym->st_info) != STB_GLOBAL &&
|
| 246 | ELF64_ST_BIND(sym->st_info) != STB_WEAK)
|
| 247 | return false;
|
| 248 | if (strcmp(name, vdso_info.symstrings + sym->st_name))
|
| 249 | return false;
|
| 250 |
|
| 251 |
|
| 252 | if (vdso_info.versym &&
|
| 253 | !vdso_match_version(vdso_info.versym[i], version, ver_hash))
|
| 254 | return false;
|
| 255 |
|
| 256 | return true;
|
| 257 | }
|
| 258 |
|
| 259 | void *vdso_sym(const char *version, const char *name)
|
| 260 | {
|
| 261 | unsigned long ver_hash;
|
| 262 | if (!vdso_info.valid)
|
| 263 | return 0;
|
| 264 |
|
| 265 | ver_hash = elf_hash(version);
|
| 266 | ELF(Word) i;
|
| 267 |
|
| 268 | if (vdso_info.gnu_hash) {
|
| 269 | uint32_t h1 = gnu_hash(name), h2, *hashval;
|
| 270 |
|
| 271 | i = vdso_info.gnu_bucket[h1 % vdso_info.nbucket];
|
| 272 | if (i == 0)
|
| 273 | return 0;
|
| 274 | h1 |= 1;
|
| 275 | hashval = vdso_info.gnu_bucket + vdso_info.nbucket +
|
| 276 | (i - vdso_info.gnu_hash[1]);
|
| 277 | for (;; i++) {
|
| 278 | ELF(Sym) *sym = &vdso_info.symtab[i];
|
| 279 | h2 = *hashval++;
|
| 280 | if (h1 == (h2 | 1) &&
|
| 281 | check_sym(sym, i, name, version, ver_hash))
|
| 282 | return (void *)(vdso_info.load_offset +
|
| 283 | sym->st_value);
|
| 284 | if (h2 & 1)
|
| 285 | break;
|
| 286 | }
|
| 287 | } else {
|
| 288 | i = vdso_info.bucket[elf_hash(name) % vdso_info.nbucket];
|
| 289 | for (; i; i = vdso_info.chain[i]) {
|
| 290 | ELF(Sym) *sym = &vdso_info.symtab[i];
|
| 291 | if (sym->st_shndx != SHN_UNDEF &&
|
| 292 | check_sym(sym, i, name, version, ver_hash))
|
| 293 | return (void *)(vdso_info.load_offset +
|
| 294 | sym->st_value);
|
| 295 | }
|
| 296 | }
|
| 297 |
|
| 298 | return 0;
|
| 299 | }
|