| 1 | /* Return the offset of one string within another. |
| 2 | Copyright (C) 1994-2021 Free Software Foundation, Inc. |
| 3 | This file is part of the GNU C Library. |
| 4 | |
| 5 | The GNU C Library is free software; you can redistribute it and/or |
| 6 | modify it under the terms of the GNU Lesser General Public |
| 7 | License as published by the Free Software Foundation; either |
| 8 | version 2.1 of the License, or (at your option) any later version. |
| 9 | |
| 10 | The GNU C Library is distributed in the hope that it will be useful, |
| 11 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 12 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU |
| 13 | Lesser General Public License for more details. |
| 14 | |
| 15 | You should have received a copy of the GNU Lesser General Public |
| 16 | License along with the GNU C Library; if not, see |
| 17 | <https://www.gnu.org/licenses/>. */ |
| 18 | |
| 19 | #ifndef _LIBC |
| 20 | # include <config.h> |
| 21 | #endif |
| 22 | |
| 23 | #include <string.h> |
| 24 | |
| 25 | #define RETURN_TYPE char * |
| 26 | #define AVAILABLE(h, h_l, j, n_l) \ |
| 27 | (((j) + (n_l) <= (h_l)) \ |
| 28 | || ((h_l) += __strnlen ((void*)((h) + (h_l)), (n_l) + 512), \ |
| 29 | (j) + (n_l) <= (h_l))) |
| 30 | #include "str-two-way.h" |
| 31 | |
| 32 | #undef strstr |
| 33 | |
| 34 | #ifndef STRSTR |
| 35 | #define STRSTR strstr |
| 36 | #endif |
| 37 | |
| 38 | static inline char * |
| 39 | strstr2 (const unsigned char *hs, const unsigned char *ne) |
| 40 | { |
| 41 | uint32_t h1 = (ne[0] << 16) | ne[1]; |
| 42 | uint32_t h2 = 0; |
| 43 | for (int c = hs[0]; h1 != h2 && c != 0; c = *++hs) |
| 44 | h2 = (h2 << 16) | c; |
| 45 | return h1 == h2 ? (char *)hs - 2 : NULL; |
| 46 | } |
| 47 | |
| 48 | static inline char * |
| 49 | strstr3 (const unsigned char *hs, const unsigned char *ne) |
| 50 | { |
| 51 | uint32_t h1 = ((uint32_t)ne[0] << 24) | (ne[1] << 16) | (ne[2] << 8); |
| 52 | uint32_t h2 = 0; |
| 53 | for (int c = hs[0]; h1 != h2 && c != 0; c = *++hs) |
| 54 | h2 = (h2 | c) << 8; |
| 55 | return h1 == h2 ? (char *)hs - 3 : NULL; |
| 56 | } |
| 57 | |
| 58 | /* Hash character pairs so a small shift table can be used. All bits of |
| 59 | p[0] are included, but not all bits from p[-1]. So if two equal hashes |
| 60 | match on p[-1], p[0] matches too. Hash collisions are harmless and result |
| 61 | in smaller shifts. */ |
| 62 | #define hash2(p) (((size_t)(p)[0] - ((size_t)(p)[-1] << 3)) % sizeof (shift)) |
| 63 | |
| 64 | /* Fast strstr algorithm with guaranteed linear-time performance. |
| 65 | Small needles up to size 3 use a dedicated linear search. Longer needles |
| 66 | up to size 256 use a novel modified Horspool algorithm. It hashes pairs |
| 67 | of characters to quickly skip past mismatches. The main search loop only |
| 68 | exits if the last 2 characters match, avoiding unnecessary calls to memcmp |
| 69 | and allowing for a larger skip if there is no match. A self-adapting |
| 70 | filtering check is used to quickly detect mismatches in long needles. |
| 71 | By limiting the needle length to 256, the shift table can be reduced to 8 |
| 72 | bits per entry, lowering preprocessing overhead and minimizing cache effects. |
| 73 | The limit also implies worst-case performance is linear. |
| 74 | Needles larger than 256 characters use the linear-time Two-Way algorithm. */ |
| 75 | char * |
| 76 | STRSTR (const char *haystack, const char *needle) |
| 77 | { |
| 78 | const unsigned char *hs = (const unsigned char *) haystack; |
| 79 | const unsigned char *ne = (const unsigned char *) needle; |
| 80 | |
| 81 | /* Handle short needle special cases first. */ |
| 82 | if (ne[0] == '\0') |
| 83 | return (char *)hs; |
| 84 | hs = (const unsigned char *)strchr ((const char*)hs, ne[0]); |
| 85 | if (hs == NULL || ne[1] == '\0') |
| 86 | return (char*)hs; |
| 87 | if (ne[2] == '\0') |
| 88 | return strstr2 (hs, ne); |
| 89 | if (ne[3] == '\0') |
| 90 | return strstr3 (hs, ne); |
| 91 | |
| 92 | /* Ensure haystack length is at least as long as needle length. |
| 93 | Since a match may occur early on in a huge haystack, use strnlen |
| 94 | and read ahead a few cachelines for improved performance. */ |
| 95 | size_t ne_len = strlen ((const char*)ne); |
| 96 | size_t hs_len = __strnlen ((const char*)hs, ne_len | 512); |
| 97 | if (hs_len < ne_len) |
| 98 | return NULL; |
| 99 | |
| 100 | /* Check whether we have a match. This improves performance since we |
| 101 | avoid initialization overheads. */ |
| 102 | if (memcmp (hs, ne, ne_len) == 0) |
| 103 | return (char *) hs; |
| 104 | |
| 105 | /* Use Two-Way algorithm for very long needles. */ |
| 106 | if (__glibc_unlikely (ne_len > 256)) |
| 107 | return two_way_long_needle (hs, hs_len, ne, ne_len); |
| 108 | |
| 109 | const unsigned char *end = hs + hs_len - ne_len; |
| 110 | uint8_t shift[256]; |
| 111 | size_t tmp, shift1; |
| 112 | size_t m1 = ne_len - 1; |
| 113 | size_t offset = 0; |
| 114 | |
| 115 | /* Initialize bad character shift hash table. */ |
| 116 | memset (shift, 0, sizeof (shift)); |
| 117 | for (int i = 1; i < m1; i++) |
| 118 | shift[hash2 (ne + i)] = i; |
| 119 | /* Shift1 is the amount we can skip after matching the hash of the |
| 120 | needle end but not the full needle. */ |
| 121 | shift1 = m1 - shift[hash2 (ne + m1)]; |
| 122 | shift[hash2 (ne + m1)] = m1; |
| 123 | |
| 124 | while (1) |
| 125 | { |
| 126 | if (__glibc_unlikely (hs > end)) |
| 127 | { |
| 128 | end += __strnlen ((const char*)end + m1 + 1, 2048); |
| 129 | if (hs > end) |
| 130 | return NULL; |
| 131 | } |
| 132 | |
| 133 | /* Skip past character pairs not in the needle. */ |
| 134 | do |
| 135 | { |
| 136 | hs += m1; |
| 137 | tmp = shift[hash2 (hs)]; |
| 138 | } |
| 139 | while (tmp == 0 && hs <= end); |
| 140 | |
| 141 | /* If the match is not at the end of the needle, shift to the end |
| 142 | and continue until we match the hash of the needle end. */ |
| 143 | hs -= tmp; |
| 144 | if (tmp < m1) |
| 145 | continue; |
| 146 | |
| 147 | /* Hash of the last 2 characters matches. If the needle is long, |
| 148 | try to quickly filter out mismatches. */ |
| 149 | if (m1 < 15 || memcmp (hs + offset, ne + offset, 8) == 0) |
| 150 | { |
| 151 | if (memcmp (hs, ne, m1) == 0) |
| 152 | return (void *) hs; |
| 153 | |
| 154 | /* Adjust filter offset when it doesn't find the mismatch. */ |
| 155 | offset = (offset >= 8 ? offset : m1) - 8; |
| 156 | } |
| 157 | |
| 158 | /* Skip based on matching the hash of the needle end. */ |
| 159 | hs += shift1; |
| 160 | } |
| 161 | } |
| 162 | libc_hidden_builtin_def (strstr) |
| 163 | |