| 1 | /* |
| 2 | Portions Copyright (c) 2016-Present, Facebook, Inc. |
| 3 | Portions Copyright (c) 2012,2013 Monty Program Ab |
| 4 | |
| 5 | This program is free software; you can redistribute it and/or modify |
| 6 | it under the terms of the GNU General Public License as published by |
| 7 | the Free Software Foundation; version 2 of the License. |
| 8 | |
| 9 | This program is distributed in the hope that it will be useful, |
| 10 | but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 11 | MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 12 | GNU General Public License for more details. |
| 13 | |
| 14 | You should have received a copy of the GNU General Public License |
| 15 | along with this program; if not, write to the Free Software |
| 16 | Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA */ |
| 17 | #pragma once |
| 18 | |
| 19 | #include <algorithm> |
| 20 | #include <string> |
| 21 | #include <vector> |
| 22 | |
| 23 | #ifdef _WIN32 |
| 24 | #include <stdlib.h> |
| 25 | #define htobe64 _byteswap_uint64 |
| 26 | #define be64toh _byteswap_uint64 |
| 27 | #define htobe32 _byteswap_ulong |
| 28 | #define be32toh _byteswap_ulong |
| 29 | #define htobe16 _byteswap_ushort |
| 30 | #define be16toh _byteswap_ushort |
| 31 | #endif |
| 32 | |
| 33 | #if defined(__APPLE__) |
| 34 | #include <libkern/OSByteOrder.h> |
| 35 | #define htobe64(x) OSSwapHostToBigInt64(x) |
| 36 | #define be64toh(x) OSSwapBigToHostInt64(x) |
| 37 | #define htobe32(x) OSSwapHostToBigInt32(x) |
| 38 | #define be32toh(x) OSSwapBigToHostInt32(x) |
| 39 | #define htobe16(x) OSSwapHostToBigInt16(x) |
| 40 | #define be16toh(x) OSSwapBigToHostInt16(x) |
| 41 | #endif |
| 42 | |
| 43 | namespace myrocks { |
| 44 | |
| 45 | /* |
| 46 | Basic composition functions for a network buffer presented as a MySQL String |
| 47 | ("netstr") which stores data in Network Byte Order (Big Endian). |
| 48 | */ |
| 49 | |
| 50 | inline void rdb_netstr_append_uint64(my_core::String *const out_netstr, |
| 51 | const uint64 &val) { |
| 52 | DBUG_ASSERT(out_netstr != nullptr); |
| 53 | |
| 54 | // Convert from host machine byte order (usually Little Endian) to network |
| 55 | // byte order (Big Endian). |
| 56 | uint64 net_val = htobe64(val); |
| 57 | out_netstr->append(reinterpret_cast<char *>(&net_val), sizeof(net_val)); |
| 58 | } |
| 59 | |
| 60 | inline void rdb_netstr_append_uint32(my_core::String *const out_netstr, |
| 61 | const uint32 &val) { |
| 62 | DBUG_ASSERT(out_netstr != nullptr); |
| 63 | |
| 64 | // Convert from host machine byte order (usually Little Endian) to network |
| 65 | // byte order (Big Endian). |
| 66 | uint32 net_val = htobe32(val); |
| 67 | out_netstr->append(reinterpret_cast<char *>(&net_val), sizeof(net_val)); |
| 68 | } |
| 69 | |
| 70 | inline void rdb_netstr_append_uint16(my_core::String *const out_netstr, |
| 71 | const uint16 &val) { |
| 72 | DBUG_ASSERT(out_netstr != nullptr); |
| 73 | |
| 74 | // Convert from host machine byte order (usually Little Endian) to network |
| 75 | // byte order (Big Endian). |
| 76 | uint16 net_val = htobe16(val); |
| 77 | out_netstr->append(reinterpret_cast<char *>(&net_val), sizeof(net_val)); |
| 78 | } |
| 79 | |
| 80 | /* |
| 81 | Basic network buffer ("netbuf") write helper functions. |
| 82 | */ |
| 83 | |
| 84 | inline void rdb_netbuf_store_uint64(uchar *const dst_netbuf, const uint64 &n) { |
| 85 | DBUG_ASSERT(dst_netbuf != nullptr); |
| 86 | |
| 87 | // Convert from host byte order (usually Little Endian) to network byte order |
| 88 | // (Big Endian). |
| 89 | uint64 net_val = htobe64(n); |
| 90 | memcpy(dst_netbuf, &net_val, sizeof(net_val)); |
| 91 | } |
| 92 | |
| 93 | inline void rdb_netbuf_store_uint32(uchar *const dst_netbuf, const uint32 &n) { |
| 94 | DBUG_ASSERT(dst_netbuf != nullptr); |
| 95 | |
| 96 | // Convert from host byte order (usually Little Endian) to network byte order |
| 97 | // (Big Endian). |
| 98 | uint32 net_val = htobe32(n); |
| 99 | memcpy(dst_netbuf, &net_val, sizeof(net_val)); |
| 100 | } |
| 101 | |
| 102 | inline void rdb_netbuf_store_uint16(uchar *const dst_netbuf, const uint16 &n) { |
| 103 | DBUG_ASSERT(dst_netbuf != nullptr); |
| 104 | |
| 105 | // Convert from host byte order (usually Little Endian) to network byte order |
| 106 | // (Big Endian). |
| 107 | uint16 net_val = htobe16(n); |
| 108 | memcpy(dst_netbuf, &net_val, sizeof(net_val)); |
| 109 | } |
| 110 | |
| 111 | inline void rdb_netbuf_store_byte(uchar *const dst_netbuf, const uchar &c) { |
| 112 | DBUG_ASSERT(dst_netbuf != nullptr); |
| 113 | |
| 114 | *dst_netbuf = c; |
| 115 | } |
| 116 | |
| 117 | inline void rdb_netbuf_store_index(uchar *const dst_netbuf, |
| 118 | const uint32 &number) { |
| 119 | DBUG_ASSERT(dst_netbuf != nullptr); |
| 120 | |
| 121 | rdb_netbuf_store_uint32(dst_netbuf, number); |
| 122 | } |
| 123 | |
| 124 | /* |
| 125 | Basic conversion helper functions from network byte order (Big Endian) to host |
| 126 | machine byte order (usually Little Endian). |
| 127 | */ |
| 128 | |
| 129 | inline uint64 rdb_netbuf_to_uint64(const uchar *const netbuf) { |
| 130 | DBUG_ASSERT(netbuf != nullptr); |
| 131 | |
| 132 | uint64 net_val; |
| 133 | memcpy(&net_val, netbuf, sizeof(net_val)); |
| 134 | |
| 135 | // Convert from network byte order (Big Endian) to host machine byte order |
| 136 | // (usually Little Endian). |
| 137 | return be64toh(net_val); |
| 138 | } |
| 139 | |
| 140 | inline uint32 rdb_netbuf_to_uint32(const uchar *const netbuf) { |
| 141 | DBUG_ASSERT(netbuf != nullptr); |
| 142 | |
| 143 | uint32 net_val; |
| 144 | memcpy(&net_val, netbuf, sizeof(net_val)); |
| 145 | |
| 146 | // Convert from network byte order (Big Endian) to host machine byte order |
| 147 | // (usually Little Endian). |
| 148 | return be32toh(net_val); |
| 149 | } |
| 150 | |
| 151 | inline uint16 rdb_netbuf_to_uint16(const uchar *const netbuf) { |
| 152 | DBUG_ASSERT(netbuf != nullptr); |
| 153 | |
| 154 | uint16 net_val; |
| 155 | memcpy(&net_val, netbuf, sizeof(net_val)); |
| 156 | |
| 157 | // Convert from network byte order (Big Endian) to host machine byte order |
| 158 | // (usually Little Endian). |
| 159 | return be16toh(net_val); |
| 160 | } |
| 161 | |
| 162 | inline uchar rdb_netbuf_to_byte(const uchar *const netbuf) { |
| 163 | DBUG_ASSERT(netbuf != nullptr); |
| 164 | |
| 165 | return (uchar)netbuf[0]; |
| 166 | } |
| 167 | |
| 168 | /* |
| 169 | Basic network buffer ("netbuf") read helper functions. |
| 170 | Network buffer stores data in Network Byte Order (Big Endian). |
| 171 | NB: The netbuf is passed as an input/output param, hence after reading, |
| 172 | the netbuf pointer gets advanced to the following byte. |
| 173 | */ |
| 174 | |
| 175 | inline uint64 rdb_netbuf_read_uint64(const uchar **netbuf_ptr) { |
| 176 | DBUG_ASSERT(netbuf_ptr != nullptr); |
| 177 | |
| 178 | // Convert from network byte order (Big Endian) to host machine byte order |
| 179 | // (usually Little Endian). |
| 180 | const uint64 host_val = rdb_netbuf_to_uint64(*netbuf_ptr); |
| 181 | |
| 182 | // Advance pointer. |
| 183 | *netbuf_ptr += sizeof(host_val); |
| 184 | |
| 185 | return host_val; |
| 186 | } |
| 187 | |
| 188 | inline uint32 rdb_netbuf_read_uint32(const uchar **netbuf_ptr) { |
| 189 | DBUG_ASSERT(netbuf_ptr != nullptr); |
| 190 | |
| 191 | // Convert from network byte order (Big Endian) to host machine byte order |
| 192 | // (usually Little Endian). |
| 193 | const uint32 host_val = rdb_netbuf_to_uint32(*netbuf_ptr); |
| 194 | |
| 195 | // Advance pointer. |
| 196 | *netbuf_ptr += sizeof(host_val); |
| 197 | |
| 198 | return host_val; |
| 199 | } |
| 200 | |
| 201 | inline uint16 rdb_netbuf_read_uint16(const uchar **netbuf_ptr) { |
| 202 | DBUG_ASSERT(netbuf_ptr != nullptr); |
| 203 | |
| 204 | // Convert from network byte order (Big Endian) to host machine byte order |
| 205 | // (usually Little Endian). |
| 206 | const uint16 host_val = rdb_netbuf_to_uint16(*netbuf_ptr); |
| 207 | |
| 208 | // Advance pointer. |
| 209 | *netbuf_ptr += sizeof(host_val); |
| 210 | |
| 211 | return host_val; |
| 212 | } |
| 213 | |
| 214 | inline void rdb_netbuf_read_gl_index(const uchar **netbuf_ptr, |
| 215 | GL_INDEX_ID *const gl_index_id) { |
| 216 | DBUG_ASSERT(gl_index_id != nullptr); |
| 217 | DBUG_ASSERT(netbuf_ptr != nullptr); |
| 218 | |
| 219 | gl_index_id->cf_id = rdb_netbuf_read_uint32(netbuf_ptr); |
| 220 | gl_index_id->index_id = rdb_netbuf_read_uint32(netbuf_ptr); |
| 221 | } |
| 222 | |
| 223 | /* |
| 224 | A simple string reader: |
| 225 | - it keeps position within the string that we read from |
| 226 | - it prevents one from reading beyond the end of the string. |
| 227 | */ |
| 228 | |
| 229 | class Rdb_string_reader { |
| 230 | const char *m_ptr; |
| 231 | uint m_len; |
| 232 | |
| 233 | private: |
| 234 | Rdb_string_reader &operator=(const Rdb_string_reader &) = default; |
| 235 | |
| 236 | public: |
| 237 | Rdb_string_reader(const Rdb_string_reader &) = default; |
| 238 | /* named constructor */ |
| 239 | static Rdb_string_reader read_or_empty(const rocksdb::Slice *const slice) { |
| 240 | if (!slice) { |
| 241 | return Rdb_string_reader("" ); |
| 242 | } else { |
| 243 | return Rdb_string_reader(slice); |
| 244 | } |
| 245 | } |
| 246 | |
| 247 | explicit Rdb_string_reader(const std::string &str) { |
| 248 | m_len = str.length(); |
| 249 | if (m_len) { |
| 250 | m_ptr = &str.at(0); |
| 251 | } else { |
| 252 | /* |
| 253 | One can a create a Rdb_string_reader for reading from an empty string |
| 254 | (although attempts to read anything will fail). |
| 255 | We must not access str.at(0), since len==0, we can set ptr to any |
| 256 | value. |
| 257 | */ |
| 258 | m_ptr = nullptr; |
| 259 | } |
| 260 | } |
| 261 | |
| 262 | explicit Rdb_string_reader(const rocksdb::Slice *const slice) { |
| 263 | m_ptr = slice->data(); |
| 264 | m_len = slice->size(); |
| 265 | } |
| 266 | |
| 267 | /* |
| 268 | Read the next @param size bytes. Returns pointer to the bytes read, or |
| 269 | nullptr if the remaining string doesn't have that many bytes. |
| 270 | */ |
| 271 | const char *read(const uint &size) { |
| 272 | const char *res; |
| 273 | if (m_len < size) { |
| 274 | res = nullptr; |
| 275 | } else { |
| 276 | res = m_ptr; |
| 277 | m_ptr += size; |
| 278 | m_len -= size; |
| 279 | } |
| 280 | return res; |
| 281 | } |
| 282 | |
| 283 | bool read_uint8(uint *const res) { |
| 284 | const uchar *p; |
| 285 | if (!(p = reinterpret_cast<const uchar *>(read(1)))) |
| 286 | return true; // error |
| 287 | else { |
| 288 | *res = *p; |
| 289 | return false; // Ok |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | bool read_uint16(uint *const res) { |
| 294 | const uchar *p; |
| 295 | if (!(p = reinterpret_cast<const uchar *>(read(2)))) |
| 296 | return true; // error |
| 297 | else { |
| 298 | *res = rdb_netbuf_to_uint16(p); |
| 299 | return false; // Ok |
| 300 | } |
| 301 | } |
| 302 | |
| 303 | bool read_uint64(uint64 *const res) { |
| 304 | const uchar *p; |
| 305 | if (!(p = reinterpret_cast<const uchar *>(read(sizeof(uint64))))) { |
| 306 | return true; // error |
| 307 | } else { |
| 308 | *res = rdb_netbuf_to_uint64(p); |
| 309 | return false; // Ok |
| 310 | } |
| 311 | } |
| 312 | |
| 313 | uint remaining_bytes() const { return m_len; } |
| 314 | |
| 315 | /* |
| 316 | Return pointer to data that will be read by next read() call (if there is |
| 317 | nothing left to read, returns pointer to beyond the end of previous read() |
| 318 | call) |
| 319 | */ |
| 320 | const char *get_current_ptr() const { return m_ptr; } |
| 321 | }; |
| 322 | |
| 323 | /* |
| 324 | @brief |
| 325 | A buffer one can write the data to. |
| 326 | |
| 327 | @detail |
| 328 | Suggested usage pattern: |
| 329 | |
| 330 | writer->clear(); |
| 331 | writer->write_XXX(...); |
| 332 | ... |
| 333 | // Ok, writer->ptr() points to the data written so far, |
| 334 | // and writer->get_current_pos() is the length of the data |
| 335 | |
| 336 | */ |
| 337 | |
| 338 | class Rdb_string_writer { |
| 339 | std::vector<uchar> m_data; |
| 340 | |
| 341 | public: |
| 342 | Rdb_string_writer(const Rdb_string_writer &) = delete; |
| 343 | Rdb_string_writer &operator=(const Rdb_string_writer &) = delete; |
| 344 | Rdb_string_writer() = default; |
| 345 | |
| 346 | void clear() { m_data.clear(); } |
| 347 | void write_uint8(const uint &val) { |
| 348 | m_data.push_back(static_cast<uchar>(val)); |
| 349 | } |
| 350 | |
| 351 | void write_uint16(const uint &val) { |
| 352 | const auto size = m_data.size(); |
| 353 | m_data.resize(size + 2); |
| 354 | rdb_netbuf_store_uint16(m_data.data() + size, val); |
| 355 | } |
| 356 | |
| 357 | void write_uint32(const uint &val) { |
| 358 | const auto size = m_data.size(); |
| 359 | m_data.resize(size + 4); |
| 360 | rdb_netbuf_store_uint32(m_data.data() + size, val); |
| 361 | } |
| 362 | |
| 363 | void write(const uchar *const new_data, const size_t &len) { |
| 364 | DBUG_ASSERT(new_data != nullptr); |
| 365 | m_data.insert(m_data.end(), new_data, new_data + len); |
| 366 | } |
| 367 | |
| 368 | uchar *ptr() { return m_data.data(); } |
| 369 | size_t get_current_pos() const { return m_data.size(); } |
| 370 | |
| 371 | void write_uint8_at(const size_t &pos, const uint &new_val) { |
| 372 | // This function will only overwrite what was written |
| 373 | DBUG_ASSERT(pos < get_current_pos()); |
| 374 | m_data.data()[pos] = new_val; |
| 375 | } |
| 376 | |
| 377 | void write_uint16_at(const size_t &pos, const uint &new_val) { |
| 378 | // This function will only overwrite what was written |
| 379 | DBUG_ASSERT(pos < get_current_pos() && (pos + 1) < get_current_pos()); |
| 380 | rdb_netbuf_store_uint16(m_data.data() + pos, new_val); |
| 381 | } |
| 382 | |
| 383 | void truncate(const size_t &pos) { |
| 384 | DBUG_ASSERT(pos < m_data.size()); |
| 385 | m_data.resize(pos); |
| 386 | } |
| 387 | |
| 388 | void allocate(const size_t &len, const uchar &val = 0) { |
| 389 | DBUG_ASSERT(len > 0); |
| 390 | m_data.resize(m_data.size() + len, val); |
| 391 | } |
| 392 | |
| 393 | /* |
| 394 | An awful hack to deallocate the buffer without relying on the deconstructor. |
| 395 | This is needed to suppress valgrind errors in rocksdb.partition |
| 396 | */ |
| 397 | void free() { std::vector<uchar>().swap(m_data); } |
| 398 | }; |
| 399 | |
| 400 | /* |
| 401 | A helper class for writing bits into Rdb_string_writer. |
| 402 | |
| 403 | The class assumes (but doesn't check) that nobody tries to write |
| 404 | anything to the Rdb_string_writer that it is writing to. |
| 405 | */ |
| 406 | class Rdb_bit_writer { |
| 407 | Rdb_string_writer *m_writer; |
| 408 | uchar m_offset; |
| 409 | |
| 410 | public: |
| 411 | Rdb_bit_writer(const Rdb_bit_writer &) = delete; |
| 412 | Rdb_bit_writer &operator=(const Rdb_bit_writer &) = delete; |
| 413 | |
| 414 | explicit Rdb_bit_writer(Rdb_string_writer *writer_arg) |
| 415 | : m_writer(writer_arg), m_offset(0) {} |
| 416 | |
| 417 | void write(uint size, const uint &value) { |
| 418 | DBUG_ASSERT((value & ((1 << size) - 1)) == value); |
| 419 | |
| 420 | while (size > 0) { |
| 421 | if (m_offset == 0) { |
| 422 | m_writer->write_uint8(0); |
| 423 | } |
| 424 | // number of bits to put in this byte |
| 425 | const uint bits = std::min(size, (uint)(8 - m_offset)); |
| 426 | uchar *const last_byte = |
| 427 | m_writer->ptr() + m_writer->get_current_pos() - 1; |
| 428 | *last_byte |= (uchar)((value >> (size - bits)) & ((1 << bits) - 1)) |
| 429 | << m_offset; |
| 430 | size -= bits; |
| 431 | m_offset = (m_offset + bits) & 0x7; |
| 432 | } |
| 433 | } |
| 434 | }; |
| 435 | |
| 436 | class Rdb_bit_reader { |
| 437 | const uchar *m_cur; |
| 438 | uchar m_offset; |
| 439 | uint m_ret; |
| 440 | Rdb_string_reader *const m_reader; |
| 441 | |
| 442 | public: |
| 443 | Rdb_bit_reader(const Rdb_bit_reader &) = delete; |
| 444 | Rdb_bit_reader &operator=(const Rdb_bit_reader &) = delete; |
| 445 | |
| 446 | explicit Rdb_bit_reader(Rdb_string_reader *const reader) |
| 447 | : m_cur(nullptr), m_offset(0), m_reader(reader) {} |
| 448 | |
| 449 | // Returns a pointer to an uint containing the bits read. On subsequent |
| 450 | // reads, the value being pointed to will be overwritten. Returns nullptr |
| 451 | // on failure. |
| 452 | uint *read(uint size) { |
| 453 | m_ret = 0; |
| 454 | DBUG_ASSERT(size <= 32); |
| 455 | |
| 456 | while (size > 0) { |
| 457 | if (m_offset == 0) { |
| 458 | m_cur = (const uchar *)m_reader->read(1); |
| 459 | if (m_cur == nullptr) { |
| 460 | return nullptr; |
| 461 | } |
| 462 | } |
| 463 | // how many bits from the current byte? |
| 464 | const uint bits = std::min((uint)(8 - m_offset), size); |
| 465 | m_ret <<= bits; |
| 466 | m_ret |= (*m_cur >> m_offset) & ((1 << bits) - 1); |
| 467 | size -= bits; |
| 468 | m_offset = (m_offset + bits) & 0x7; |
| 469 | } |
| 470 | |
| 471 | return &m_ret; |
| 472 | } |
| 473 | }; |
| 474 | |
| 475 | } // namespace myrocks |
| 476 | |