1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486 487 488 489 490 491 492 493 494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587 588 589 590 591 592 593 594 595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417 1418 1419 1420 1421 1422 1423 1424 1425 1426 1427 1428 1429 1430 1431 1432 1433 1434 1435 1436 1437 1438 1439 1440 1441 1442 1443 1444 1445 1446 1447 1448 1449 1450 1451 1452 1453 1454 1455 1456 1457 1458 1459 1460 1461 1462 1463 1464 1465 1466 1467 1468 1469 1470 1471 1472 1473 1474 1475 1476 1477 1478 1479 1480 1481 1482 1483 1484 1485 1486 1487 1488 1489 1490 1491 1492 1493 1494 1495 1496 1497 1498 1499 1500 1501 1502 1503 1504 1505 1506 1507 1508 1509 1510 1511 1512 1513 1514 1515 1516 1517 1518 1519 1520 1521 1522 1523 1524 1525 1526 1527 1528 1529 1530 1531 1532 1533 1534 1535 1536 1537 1538 1539 1540 1541 1542 1543 1544 1545 1546 1547 1548 1549 1550 1551 1552 1553 1554 1555 1556 1557 1558 1559 1560 1561 1562 1563 1564 1565 1566 1567 1568 1569 1570 1571 1572 1573 1574 1575 1576 1577 1578 1579 1580 1581 1582 1583 1584 1585 1586 1587 1588 1589 1590 1591 1592 1593 1594 1595 1596 1597 1598 1599 1600 1601 1602 1603 1604 1605 1606 1607 1608 1609 1610 1611 1612 1613 1614 1615 1616 1617 1618 1619 1620 1621 1622 1623 1624 1625 1626 1627 1628 1629 1630 1631 1632 1633 1634 1635 1636 1637 1638 1639 1640 1641 1642 1643 1644 1645 1646 1647 1648 1649 1650 1651 1652 1653 1654 1655 1656 1657 1658 1659 1660 1661 1662 1663 1664 1665 1666 1667 1668 1669 1670 1671 1672 1673 1674 1675 1676 1677 1678 1679 1680 1681 1682 1683 1684 1685 1686 1687 1688 1689 1690 1691 1692 1693 1694 1695 1696 1697 1698 1699 1700 1701 1702 1703 1704 1705 1706 1707 1708 1709 1710 1711 1712 1713 1714 1715 1716 1717 1718 1719 1720 1721 1722 1723 1724 1725 1726 1727 1728 1729 1730 1731 1732 1733 1734 1735 1736 1737 1738 1739 1740 1741 1742 1743 1744 1745 1746 1747 1748 1749 1750 1751 1752 1753 1754 1755 1756 1757 1758 1759 1760 1761 1762 1763 1764 1765 1766 1767 1768 1769 1770 1771 1772 1773 1774 1775 1776 1777 1778 1779 1780 1781 1782 1783 1784 1785 1786 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 1801 1802 1803 1804 1805 1806 1807 1808 1809 1810 1811 1812 1813 1814 1815 1816 1817 1818 1819 1820 1821 1822 1823 1824 1825 1826 1827 1828 1829 1830 1831 1832 1833 1834 1835 1836 1837 1838 1839 1840 1841 1842 1843 1844 1845 1846 1847 1848 1849 1850 1851 1852 1853 1854 1855 1856 1857 1858 1859 1860 1861 1862 1863 1864 1865 1866 1867 1868 1869 1870 1871 1872 1873 1874 1875 1876 1877 1878 1879 1880 1881 1882 1883 1884 1885 1886 1887 1888 1889 1890 1891 1892 1893 1894 1895 1896 1897 1898 1899 1900 1901 1902 1903 1904 1905 1906 1907 1908 1909 1910 1911 1912 1913 1914 1915 1916 1917 1918 1919 1920 1921 1922 1923 1924 1925 1926 1927 1928 1929 1930 1931 1932 1933 1934 1935 1936 1937 1938 1939 1940 1941 1942 1943 1944 1945 1946 1947 1948 1949 1950 1951 1952 1953 1954 1955 1956 1957 1958 1959 1960 1961 1962 1963 1964 1965 1966 1967 1968 1969 1970 1971 1972 1973 1974 1975 1976 1977 1978 1979 1980 1981 1982 1983 1984 1985 1986 1987 1988 1989 1990 1991 1992 1993 1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030 2031 2032 2033 2034 2035 2036 2037 2038 2039 2040 2041 2042 2043 2044 2045 2046 2047 2048 2049 2050 2051 2052 2053 2054 2055 2056 2057 2058 2059 2060 2061 2062 2063 2064 2065 2066 2067 2068 2069 2070 2071 2072 2073 2074 2075 2076 2077 2078 2079 2080 2081 2082 2083 2084 2085 2086 2087 2088 2089 2090 2091 2092 2093 2094 2095 2096 2097 2098 2099 2100 2101 2102 2103 2104 2105 2106 2107 2108 2109 2110 2111 2112 2113 2114 2115 2116 2117 2118 2119 2120 2121 2122 2123 2124 2125 2126 2127 2128 2129 2130 2131 2132 2133 2134 2135 2136 2137 2138 2139 2140 2141 2142 2143 2144 2145 2146 2147 2148 2149 2150 2151 2152 2153 2154 2155
use {IntoBuf, BufMut}; use std::{cmp, fmt, mem, hash, ops, slice, ptr, usize}; use std::borrow::Borrow; use std::io::Cursor; use std::sync::atomic::{self, AtomicUsize, AtomicPtr}; use std::sync::atomic::Ordering::{Relaxed, Acquire, Release, AcqRel}; /// A reference counted contiguous slice of memory. /// /// `Bytes` is an efficient container for storing and operating on continguous /// slices of memory. It is intended for use primarily in networking code, but /// could have applications elsewhere as well. /// /// `Bytes` values facilitate zero-copy network programming by allowing multiple /// `Bytes` objects to point to the same underlying memory. This is managed by /// using a reference count to track when the memory is no longer needed and can /// be freed. /// /// ``` /// use bytes::Bytes; /// /// let mut mem = Bytes::from(&b"Hello world"[..]); /// let a = mem.slice(0, 5); /// /// assert_eq!(&a[..], b"Hello"); /// /// let b = mem.drain_to(6); /// /// assert_eq!(&mem[..], b"world"); /// assert_eq!(&b[..], b"Hello "); /// ``` /// /// # Memory layout /// /// The `Bytes` struct itself is fairly small, limited to a pointer to the /// memory and 4 `usize` fields used to track information about which segment of /// the underlying memory the `Bytes` handle has access to. /// /// The memory layout looks like this: /// /// ```text /// +-------+ /// | Bytes | /// +-------+ /// / \_____ /// | \ /// v v /// +-----+------------------------------------+ /// | Arc | | Data | | /// +-----+------------------------------------+ /// ``` /// /// `Bytes` keeps both a pointer to the shared `Arc` containing the full memory /// slice and a pointer to the start of the region visible by the handle. /// `Bytes` also tracks the length of its view into the memory. /// /// # Sharing /// /// The memory itself is reference counted, and multiple `Bytes` objects may /// point to the same region. Each `Bytes` handle point to different sections within /// the memory region, and `Bytes` handle may or may not have overlapping views /// into the memory. /// /// /// ```text /// /// Arc ptrs +---------+ /// ________________________ / | Bytes 2 | /// / +---------+ /// / +-----------+ | | /// |_________/ | Bytes 1 | | | /// | +-----------+ | | /// | | | ___/ data | tail /// | data | tail |/ | /// v v v v /// +-----+---------------------------------+-----+ /// | Arc | | | | | /// +-----+---------------------------------+-----+ /// ``` /// /// # Mutating /// /// While `Bytes` handles may potentially represent overlapping views of the /// underlying memory slice and may not be mutated, `BytesMut` handles are /// guaranteed to be the only handle able to view that slice of memory. As such, /// `BytesMut` handles are able to mutate the underlying memory. Note that /// holding a unique view to a region of memory does not mean that there are not /// other `Bytes` and `BytesMut` handles with disjoint views of the underlying /// memory. /// /// # Inline bytes. /// /// As an opitmization, when the slice referenced by a `Bytes` or `BytesMut` /// handle is small enough [1], `Bytes` will avoid the allocation by inlining /// the slice directly in the handle. In this case, a clone is no longer /// "shallow" and the data will be copied. /// /// [1] Small enough: 31 bytes on 64 bit systems, 15 on 32 bit systems. /// pub struct Bytes { inner: Inner2, } /// A unique reference to a continuous slice of memory. /// /// `BytesMut` represents a unique view into a potentially shared memory region. /// Given the uniqueness guarantee, owners of `BytesMut` handles are able to /// mutate the memory. /// /// For more detail, see [Bytes](struct.Bytes.html). /// /// ``` /// use bytes::{BytesMut, BufMut}; /// /// let mut buf = BytesMut::with_capacity(64); /// /// buf.put(b'h'); /// buf.put(b'e'); /// buf.put("llo"); /// /// assert_eq!(&buf[..], b"hello"); /// /// // Freeze the buffer so that it can be shared /// let a = buf.freeze(); /// /// // This does not allocate, instead `b` points to the same memory. /// let b = a.clone(); /// /// assert_eq!(&a[..], b"hello"); /// assert_eq!(&b[..], b"hello"); /// ``` pub struct BytesMut { inner: Inner2, } // Both `Bytes` and `BytesMut` are backed by `Inner` and functions are delegated // to `Inner` functions. The `Bytes` and `BytesMut` shims ensure that functions // that mutate the underlying buffer are only performed when the data range // being mutated is only available via a single `BytesMut` handle. // // # Data storage modes // // The goal of `bytes` is to be as efficient as possible across a wide range of // potential usage patterns. As such, `bytes` needs to be able to handle buffers // that are never shared, shared on a single thread, and shared across many // threads. `bytes` also needs to handle both tiny buffers as well as very large // buffers. For example, [Cassandra](http://cassandra.apache.org) values have // been known to be in the hundreds of megabyte, and HTTP header values can be a // few characters in size. // // To achieve high performance in these various situations, `Bytes` and // `BytesMut` use different strategies for storing the buffer depending on the // usage pattern. // // ## Delayed `Arc` allocation // // When a `Bytes` or `BytesMut` is first created, there is only one outstanding // handle referencing the buffer. Since sharing is not yet required, an `Arc`* is // not used and the buffer is backed by a `Vec<u8>` directly. Using an // `Arc<Vec<u8>>` requires two allocations, so if the buffer ends up never being // shared, that allocation is avoided. // // When sharing does become necessary (`clone`, `drain_to`, `split_off`), that // is when the buffer is promoted to being shareable. The `Vec<u8>` is moved // into an `Arc` and both the original handle and the new handle use the same // buffer via the `Arc`. // // * `Arc` is being used to signify an atomically reference counted cell. We // don't use the `Arc` implementation provided by `std` and instead use our own. // This ends up simplifying a number of the `unsafe` code snippets. // // ## Inlining small buffers // // The `Bytes` / `BytesMut` structs require 4 pointer sized fields. On 64 bit // systems, this ends up being 32 bytes, which is actually a lot of storage for // cases where `Bytes` is being used to represent small byte strings, such as // HTTP header names and values. // // To avoid any allocation at all in these cases, `Bytes` will use the struct // itself for storing the buffer, reserving 1 byte for meta data. This means // that, on 64 bit systems, 31 byte buffers require no allocation at all. // // The byte used for metadata stores a 1 bit flag used to indicate that the // buffer is stored inline as well as 7 bits for tracking the buffer length (the // return value of `Bytes::len`). // // ## Static buffers // // `Bytes` can also represent a static buffer, which is created with // `Bytes::from_static`. No copying or allocations are required for trackign // static buffers. The pointer to the `&'static [u8]`, the length, and a flag // tracking that the `Bytes` instance represents a static buffer is stored in // the `Bytes` struct. // // # Struct layout // // Both `Bytes` and `BytesMut` are wrappers around `Inner`, which provides the // data fields as well as all of the function implementations. // // The `Inner` struct is carefully laid out in order to support the // functionality described above as well as being as small as possible. Size is // important as growing the size of the `Bytes` struct from 32 bytes to 40 bytes // added as much as 15% overhead in benchmarks using `Bytes` in an HTTP header // map structure. // // The `Inner` struct contains the following fields: // // * `ptr: *mut u8` // * `len: usize` // * `cap: usize` // * `arc: AtomicPtr<Shared>` // // ## `ptr: *mut u8` // // A pointer to start of the handle's buffer view. When backed by a `Vec<u8>`, // this is always the `Vec`'s pointer. When backed by an `Arc<Vec<u8>>`, `ptr` // may have been shifted to point somewhere inside the buffer. // // When in "inlined" mode, `ptr` is used as part of the inlined buffer. // // ## `len: usize` // // The length of the handle's buffer view. When backed by a `Vec<u8>`, this is // always the `Vec`'s length. The slice represented by `ptr` and `len` should // (ideally) always be initialized memory. // // When in "inlined" mode, `len` is used as part of the inlined buffer. // // ## `cap: usize` // // The capacity of the handle's buffer view. When backed by a `Vec<u8>`, this is // always the `Vec`'s capacity. The slice represented by `ptr+len` and `cap-len` // may or may not be initialized memory. // // When in "inlined" mode, `cap` is used as part of the inlined buffer. // // ## `arc: AtomicPtr<Shared>` // // When `Inner` is in allocated mode (backed by Vec<u8> or Arc<Vec<u8>>), this // will be the pointer to the `Arc` structure tracking the ref count for the // underlying buffer. When the pointer is null, then the `Arc` has not been // allocated yet and `self` is the only outstanding handle for the underlying // buffer. // // The lower two bits of `arc` are used as flags to track the storage state of // `Inner`. `0b01` indicates inline storage and `0b10` indicates static storage. // Since pointers to allocated structures are aligned, the lower two bits of a // pointer will always be 0. This allows disambiguating between a pointer and // the two flags. // // When in "inlined" mode, the least significant byte of `arc` is also used to // store the length of the buffer view (vs. the capacity, which is a constant). // // The rest of `arc`'s bytes are used as part of the inline buffer, which means // that those bytes need to be located next to the `ptr`, `len`, and `cap` // fields, which make up the rest of the inline buffer. This requires special // casing the layout of `Inner` depending on if the target platform is bit or // little endian. // // On little endian platforms, the `arc` field must be the first field in the // struct. On big endian platforms, the `arc` field must be the last field in // the struct. Since a deterministic struct layout is required, `Inner` is // annotated with `#[repr(C)]`. // // # Thread safety // // `Bytes::clone()` returns a new `Bytes` handle with no copying. This is done // by bumping the buffer ref count and returning a new struct pointing to the // same buffer. However, the `Arc` structure is lazily allocated. This means // that if `Bytes` is stored itself in an `Arc` (`Arc<Bytes>`), the `clone` // function can be called concurrently from multiple threads. This is why an // `AtomicPtr` is used for the `arc` field vs. a `*const`. // // Care is taken to ensure that the need for synchronization is minimized. Most // operations do not require any synchronization. // #[cfg(target_endian = "little")] #[repr(C)] struct Inner { arc: AtomicPtr<Shared>, ptr: *mut u8, len: usize, cap: usize, } #[cfg(target_endian = "big")] #[repr(C)] struct Inner { ptr: *mut u8, len: usize, cap: usize, arc: AtomicPtr<Shared>, } // This struct is only here to make older versions of Rust happy. In older // versions of `Rust`, `repr(C)` structs could not have drop functions. While // this is no longer the case for newer rust versions, a number of major Rust // libraries still support older versions of Rust for which it is the case. To // get around this, `Inner` (the actual struct) is wrapped by `Inner2` which has // the drop fn implementation. struct Inner2 { inner: Inner, } // Thread-safe reference-counted container for the shared storage. This mostly // the same as `std::sync::Arc` but without the weak counter. The ref counting // fns are based on the ones found in `std`. // // The main reason to use `Shared` instead of `std::sync::Arc` is that it ends // up making the overall code simpler and easier to reason about. This is due to // some of the logic around setting `Inner::arc` and other ways the `arc` field // is used. Using `Arc` ended up requiring a number of funky transmutes and // other shenanigans to make it work. struct Shared { vec: Vec<u8>, ref_count: AtomicUsize, } // Buffer storage strategy flags. const KIND_INLINE: usize = 0b01; const KIND_STATIC: usize = 0b10; // Bit op constants for extracting the inline length value from the `arc` field. const INLINE_LEN_MASK: usize = 0b11111110; const INLINE_LEN_OFFSET: usize = 1; // Byte offset from the start of `Inner` to where the inline buffer data // starts. On little endian platforms, the first byte of the struct is the // storage flag, so the data is shifted by a byte. On big endian systems, the // data starts at the beginning of the struct. #[cfg(target_endian = "little")] const INLINE_DATA_OFFSET: isize = 1; #[cfg(target_endian = "big")] const INLINE_DATA_OFFSET: isize = 0; // Inline buffer capacity. This is the size of `Inner` minus 1 byte for the // metadata. #[cfg(target_pointer_width = "64")] const INLINE_CAP: usize = 4 * 8 - 1; #[cfg(target_pointer_width = "32")] const INLINE_CAP: usize = 4 * 4 - 1; /* * * ===== Bytes ===== * */ impl Bytes { /// Creates a new empty `Bytes` /// /// This will not allocate and the returned `Bytes` handle will be empty. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let b = Bytes::new(); /// assert_eq!(&b[..], b""); /// ``` #[inline] pub fn new() -> Bytes { Bytes { inner: Inner2 { inner: Inner { arc: AtomicPtr::new(ptr::null_mut()), ptr: ptr::null_mut(), len: 0, cap: 0, } } } } /// Creates a new `Bytes` from a static slice. /// /// The returned `Bytes` will point directly to the static slice. There is /// no allocating or copying. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let b = Bytes::from_static(b"hello"); /// assert_eq!(&b[..], b"hello"); /// ``` #[inline] pub fn from_static(bytes: &'static [u8]) -> Bytes { let ptr = bytes.as_ptr() as *mut u8; Bytes { inner: Inner2 { inner: Inner { // `arc` won't ever store a pointer. Instead, use it to // track the fact that the `Bytes` handle is backed by a // static buffer. arc: AtomicPtr::new(KIND_STATIC as *mut Shared), ptr: ptr, len: bytes.len(), cap: bytes.len(), } } } } /// Returns the number of bytes contained in this `Bytes`. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let b = Bytes::from(&b"hello"[..]); /// assert_eq!(b.len(), 5); /// ``` pub fn len(&self) -> usize { self.inner.len() } /// Returns true if the `Bytes` has a length of 0. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let b = Bytes::new(); /// assert!(b.is_empty()); /// ``` pub fn is_empty(&self) -> bool { self.inner.is_empty() } /// Returns a slice of self for the index range `[begin..end)`. /// /// This will increment the reference count for the underlying memory and /// return a new `Bytes` handle set to the slice. /// /// This operation is `O(1)`. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let a = Bytes::from(&b"hello world"[..]); /// let b = a.slice(2, 5); /// /// assert_eq!(&b[..], b"llo"); /// ``` /// /// # Panics /// /// Requires that `begin <= end` and `end <= self.len()`, otherwise slicing /// will panic. pub fn slice(&self, begin: usize, end: usize) -> Bytes { let mut ret = self.clone(); unsafe { ret.inner.set_end(end); ret.inner.set_start(begin); } ret } /// Returns a slice of self for the index range `[begin..self.len())`. /// /// This will increment the reference count for the underlying memory and /// return a new `Bytes` handle set to the slice. /// /// This operation is `O(1)` and is equivalent to `self.slice(begin, /// self.len())`. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let a = Bytes::from(&b"hello world"[..]); /// let b = a.slice_from(6); /// /// assert_eq!(&b[..], b"world"); /// ``` /// /// # Panics /// /// Requires that `begin <= self.len()`, otherwise slicing will panic. pub fn slice_from(&self, begin: usize) -> Bytes { self.slice(begin, self.len()) } /// Returns a slice of self for the index range `[0..end)`. /// /// This will increment the reference count for the underlying memory and /// return a new `Bytes` handle set to the slice. /// /// This operation is `O(1)` and is equivalent to `self.slice(0, end)`. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let a = Bytes::from(&b"hello world"[..]); /// let b = a.slice_to(5); /// /// assert_eq!(&b[..], b"hello"); /// ``` /// /// # Panics /// /// Requires that `end <= self.len()`, otherwise slicing will panic. pub fn slice_to(&self, end: usize) -> Bytes { self.slice(0, end) } /// Splits the bytes into two at the given index. /// /// Afterwards `self` contains elements `[0, at)`, and the returned `Bytes` /// contains elements `[at, len)`. /// /// This is an O(1) operation that just increases the reference count and /// sets a few indexes. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let mut a = Bytes::from(&b"hello world"[..]); /// let b = a.split_off(5); /// /// assert_eq!(&a[..], b"hello"); /// assert_eq!(&b[..], b" world"); /// ``` /// /// # Panics /// /// Panics if `at > len` pub fn split_off(&mut self, at: usize) -> Bytes { Bytes { inner: Inner2 { inner: self.inner.split_off(at), } } } /// Splits the bytes into two at the given index. /// /// Afterwards `self` contains elements `[at, len)`, and the returned /// `Bytes` contains elements `[0, at)`. /// /// This is an O(1) operation that just increases the reference count and /// sets a few indexes. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let mut a = Bytes::from(&b"hello world"[..]); /// let b = a.drain_to(5); /// /// assert_eq!(&a[..], b" world"); /// assert_eq!(&b[..], b"hello"); /// ``` /// /// # Panics /// /// Panics if `at > len` pub fn drain_to(&mut self, at: usize) -> Bytes { Bytes { inner: Inner2 { inner: self.inner.drain_to(at), } } } /// Attempt to convert into a `BytesMut` handle. /// /// This will only succeed if there are no other outstanding references to /// the underlying chunk of memory. `Bytes` handles that contain inlined /// bytes will always be convertable to `BytesMut`. /// /// # Examples /// /// ``` /// use bytes::Bytes; /// /// let a = Bytes::from(&b"Mary had a little lamb, little lamb, little lamb..."[..]); /// /// // Create a shallow clone /// let b = a.clone(); /// /// // This will fail because `b` shares a reference with `a` /// let a = a.try_mut().unwrap_err(); /// /// drop(b); /// /// // This will succeed /// let mut a = a.try_mut().unwrap(); /// /// a[0] = b'b'; /// /// assert_eq!(&a[..4], b"bary"); /// ``` pub fn try_mut(mut self) -> Result<BytesMut, Bytes> { if self.inner.is_mut_safe() { Ok(BytesMut { inner: self.inner }) } else { Err(self) } } } impl IntoBuf for Bytes { type Buf = Cursor<Self>; fn into_buf(self) -> Self::Buf { Cursor::new(self) } } impl<'a> IntoBuf for &'a Bytes { type Buf = Cursor<Self>; fn into_buf(self) -> Self::Buf { Cursor::new(self) } } impl Clone for Bytes { fn clone(&self) -> Bytes { Bytes { inner: Inner2 { inner: self.inner.shallow_clone(), } } } } impl AsRef<[u8]> for Bytes { fn as_ref(&self) -> &[u8] { self.inner.as_ref() } } impl ops::Deref for Bytes { type Target = [u8]; fn deref(&self) -> &[u8] { self.inner.as_ref() } } impl From<BytesMut> for Bytes { fn from(src: BytesMut) -> Bytes { src.freeze() } } impl From<Vec<u8>> for Bytes { fn from(src: Vec<u8>) -> Bytes { BytesMut::from(src).freeze() } } impl From<String> for Bytes { fn from(src: String) -> Bytes { BytesMut::from(src).freeze() } } impl<'a> From<&'a [u8]> for Bytes { fn from(src: &'a [u8]) -> Bytes { BytesMut::from(src).freeze() } } impl<'a> From<&'a str> for Bytes { fn from(src: &'a str) -> Bytes { BytesMut::from(src).freeze() } } impl PartialEq for Bytes { fn eq(&self, other: &Bytes) -> bool { self.inner.as_ref() == other.inner.as_ref() } } impl PartialOrd for Bytes { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { self.inner.as_ref().partial_cmp(other.inner.as_ref()) } } impl Ord for Bytes { fn cmp(&self, other: &Bytes) -> cmp::Ordering { self.inner.as_ref().cmp(other.inner.as_ref()) } } impl Eq for Bytes { } impl fmt::Debug for Bytes { fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result { fmt::Debug::fmt(&self.inner.as_ref(), fmt) } } impl hash::Hash for Bytes { fn hash<H>(&self, state: &mut H) where H: hash::Hasher { let s: &[u8] = self.as_ref(); s.hash(state); } } impl Borrow<[u8]> for Bytes { fn borrow(&self) -> &[u8] { self.as_ref() } } /* * * ===== BytesMut ===== * */ impl BytesMut { /// Create a new `BytesMut` with the specified capacity. /// /// The returned `BytesMut` will be able to hold at least `capacity` bytes /// without reallocating. If `capacity` is under `3 * size:of::<usize>()`, /// then `BytesMut` will not allocate. /// /// It is important to note that this function does not specify the length /// of the returned `BytesMut`, but only the capacity. /// /// # Examples /// /// ``` /// use bytes::{BytesMut, BufMut}; /// /// let mut bytes = BytesMut::with_capacity(64); /// /// // `bytes` contains no data, even though there is capacity /// assert_eq!(bytes.len(), 0); /// /// bytes.put(&b"hello world"[..]); /// /// assert_eq!(&bytes[..], b"hello world"); /// ``` #[inline] pub fn with_capacity(capacity: usize) -> BytesMut { if capacity <= INLINE_CAP { unsafe { // Using uninitialized memory is ~30% faster BytesMut { inner: Inner2 { inner: Inner { arc: AtomicPtr::new(KIND_INLINE as *mut Shared), .. mem::uninitialized() }, }, } } } else { BytesMut::from(Vec::with_capacity(capacity)) } } /// Returns the number of bytes contained in this `BytesMut`. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let b = BytesMut::from(&b"hello"[..]); /// assert_eq!(b.len(), 5); /// ``` #[inline] pub fn len(&self) -> usize { self.inner.len() } /// Returns true if the `BytesMut` has a length of 0. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let b = BytesMut::with_capacity(64); /// assert!(b.is_empty()); /// ``` #[inline] pub fn is_empty(&self) -> bool { self.len() == 0 } /// Returns the number of bytes the `BytesMut` can hold without reallocating. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let b = BytesMut::with_capacity(64); /// assert_eq!(b.capacity(), 64); /// ``` #[inline] pub fn capacity(&self) -> usize { self.inner.capacity() } /// Convert `self` into an immutable `Bytes` /// /// The conversion is zero cost and is used to indicate that the slice /// referenced by the handle will no longer be mutated. Once the conversion /// is done, the handle can be cloned and shared across threads. /// /// # Examples /// /// ``` /// use bytes::{BytesMut, BufMut}; /// use std::thread; /// /// let mut b = BytesMut::with_capacity(64); /// b.put("hello world"); /// let b1 = b.freeze(); /// let b2 = b1.clone(); /// /// let th = thread::spawn(move || { /// assert_eq!(&b1[..], b"hello world"); /// }); /// /// assert_eq!(&b2[..], b"hello world"); /// th.join().unwrap(); /// ``` #[inline] pub fn freeze(self) -> Bytes { Bytes { inner: self.inner } } /// Splits the bytes into two at the given index. /// /// Afterwards `self` contains elements `[0, at)`, and the returned /// `BytesMut` contains elements `[at, capacity)`. /// /// This is an O(1) operation that just increases the reference count /// and sets a few indexes. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let mut a = BytesMut::from(&b"hello world"[..]); /// let mut b = a.split_off(5); /// /// a[0] = b'j'; /// b[0] = b'!'; /// /// assert_eq!(&a[..], b"jello"); /// assert_eq!(&b[..], b"!world"); /// ``` /// /// # Panics /// /// Panics if `at > capacity` pub fn split_off(&mut self, at: usize) -> BytesMut { BytesMut { inner: Inner2 { inner: self.inner.split_off(at), } } } /// Remove the bytes from the current view, returning them in a new /// `BytesMut` handle. /// /// Afterwards, `self` will be empty, but will retain any additional /// capacity that it had before the operation. This is identical to /// `self.drain_to(self.len())`. /// /// This is an `O(1)` operation that just increases the reference count and /// sets a few indexes. /// /// # Examples /// /// ``` /// use bytes::{BytesMut, BufMut}; /// /// let mut buf = BytesMut::with_capacity(1024); /// buf.put(&b"hello world"[..]); /// /// let other = buf.drain(); /// /// assert!(buf.is_empty()); /// assert_eq!(1013, buf.capacity()); /// /// assert_eq!(other, b"hello world"[..]); /// ``` pub fn drain(&mut self) -> BytesMut { let len = self.len(); self.drain_to(len) } /// Splits the buffer into two at the given index. /// /// Afterwards `self` contains elements `[at, len)`, and the returned `BytesMut` /// contains elements `[0, at)`. /// /// This is an O(1) operation that just increases the reference count and /// sets a few indexes. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let mut a = BytesMut::from(&b"hello world"[..]); /// let mut b = a.drain_to(5); /// /// a[0] = b'!'; /// b[0] = b'j'; /// /// assert_eq!(&a[..], b"!world"); /// assert_eq!(&b[..], b"jello"); /// ``` /// /// # Panics /// /// Panics if `at > len` pub fn drain_to(&mut self, at: usize) -> BytesMut { BytesMut { inner: Inner2 { inner: self.inner.drain_to(at), } } } /// Shortens the buffer, keeping the first `len` bytes and dropping the /// rest. /// /// If `len` is greater than the buffer's current length, this has no /// effect. /// /// The [`split_off`] method can emulate `truncate`, but this causes the /// excess bytes to be returned instead of dropped. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let mut buf = BytesMut::from(&b"hello world"[..]); /// buf.truncate(5); /// assert_eq!(buf, b"hello"[..]); /// ``` /// /// [`split_off`]: #method.split_off pub fn truncate(&mut self, len: usize) { if len <= self.len() { unsafe { self.set_len(len); } } } /// Clears the buffer, removing all data. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let mut buf = BytesMut::from(&b"hello world"[..]); /// buf.clear(); /// assert!(buf.is_empty()); /// ``` pub fn clear(&mut self) { self.truncate(0); } /// Sets the length of the buffer /// /// This will explicitly set the size of the buffer without actually /// modifying the data, so it is up to the caller to ensure that the data /// has been initialized. /// /// # Examples /// /// ``` /// use bytes::BytesMut; /// /// let mut b = BytesMut::from(&b"hello world"[..]); /// /// unsafe { /// b.set_len(5); /// } /// /// assert_eq!(&b[..], b"hello"); /// /// unsafe { /// b.set_len(11); /// } /// /// assert_eq!(&b[..], b"hello world"); /// ``` /// /// # Panics /// /// This method will panic if `len` is out of bounds for the underlying /// slice or if it comes after the `end` of the configured window. pub unsafe fn set_len(&mut self, len: usize) { self.inner.set_len(len) } /// Reserves capacity for at least `additional` more bytes to be inserted /// into the given `BytesMut`. /// /// More than `additional` bytes may be reserved in order to avoid frequent /// reallocations. A call to `reserve` may result in an allocation. /// /// Before allocating new buffer space, the function will attempt to reclaim /// space in the existing buffer. If the current handle references a small /// view in the original buffer and all other handles have been dropped, /// and the requested capacity is less than or equal to the existing /// buffer's capacity, then the current view will be copied to the front of /// the buffer and the handle will take ownership of the full buffer. /// /// # Examples /// /// In the following example, a new buffer is allocated. /// /// ``` /// use bytes::BytesMut; /// /// let mut buf = BytesMut::from(&b"hello"[..]); /// buf.reserve(64); /// assert!(buf.capacity() >= 69); /// ``` /// /// In the following example, the existing buffer is reclaimed. /// /// ``` /// use bytes::{BytesMut, BufMut}; /// /// let mut buf = BytesMut::with_capacity(128); /// buf.put(&[0; 64][..]); /// /// let ptr = buf.as_ptr(); /// let other = buf.drain(); /// /// assert!(buf.is_empty()); /// assert_eq!(buf.capacity(), 64); /// /// drop(other); /// buf.reserve(128); /// /// assert_eq!(buf.capacity(), 128); /// assert_eq!(buf.as_ptr(), ptr); /// ``` /// /// # Panics /// /// Panics if the new capacity overflows usize. pub fn reserve(&mut self, additional: usize) { self.inner.reserve(additional) } } impl BufMut for BytesMut { #[inline] fn remaining_mut(&self) -> usize { self.capacity() - self.len() } #[inline] unsafe fn advance_mut(&mut self, cnt: usize) { let new_len = self.len() + cnt; self.inner.set_len(new_len); } #[inline] unsafe fn bytes_mut(&mut self) -> &mut [u8] { let len = self.len(); &mut self.inner.as_raw()[len..] } #[inline] fn put_slice(&mut self, src: &[u8]) { assert!(self.remaining_mut() >= src.len()); let len = src.len(); unsafe { self.bytes_mut()[..len].copy_from_slice(src); self.advance_mut(len); } } } impl IntoBuf for BytesMut { type Buf = Cursor<Self>; fn into_buf(self) -> Self::Buf { Cursor::new(self) } } impl<'a> IntoBuf for &'a BytesMut { type Buf = Cursor<&'a BytesMut>; fn into_buf(self) -> Self::Buf { Cursor::new(self) } } impl AsRef<[u8]> for BytesMut { fn as_ref(&self) -> &[u8] { self.inner.as_ref() } } impl ops::Deref for BytesMut { type Target = [u8]; fn deref(&self) -> &[u8] { self.as_ref() } } impl ops::DerefMut for BytesMut { fn deref_mut(&mut self) -> &mut [u8] { self.inner.as_mut() } } impl From<Vec<u8>> for BytesMut { fn from(mut src: Vec<u8>) -> BytesMut { let len = src.len(); let cap = src.capacity(); let ptr = src.as_mut_ptr(); mem::forget(src); BytesMut { inner: Inner2 { inner: Inner { arc: AtomicPtr::new(ptr::null_mut()), ptr: ptr, len: len, cap: cap, } }, } } } impl From<String> for BytesMut { fn from(src: String) -> BytesMut { BytesMut::from(src.into_bytes()) } } impl<'a> From<&'a [u8]> for BytesMut { fn from(src: &'a [u8]) -> BytesMut { let len = src.len(); if len <= INLINE_CAP { unsafe { let mut inner: Inner = mem::uninitialized(); // Set inline mask inner.arc = AtomicPtr::new(KIND_INLINE as *mut Shared); inner.set_inline_len(len); inner.as_raw()[0..len].copy_from_slice(src); BytesMut { inner: Inner2 { inner: inner, } } } } else { let mut buf = BytesMut::with_capacity(src.len()); buf.put(src.as_ref()); buf } } } impl<'a> From<&'a str> for BytesMut { fn from(src: &'a str) -> BytesMut { BytesMut::from(src.as_bytes()) } } impl From<Bytes> for BytesMut { fn from(src: Bytes) -> BytesMut { src.try_mut() .unwrap_or_else(|src| BytesMut::from(&src[..])) } } impl PartialEq for BytesMut { fn eq(&self, other: &BytesMut) -> bool { self.inner.as_ref() == other.inner.as_ref() } } impl PartialOrd for BytesMut { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { self.inner.as_ref().partial_cmp(other.inner.as_ref()) } } impl Ord for BytesMut { fn cmp(&self, other: &BytesMut) -> cmp::Ordering { self.inner.as_ref().cmp(other.inner.as_ref()) } } impl Eq for BytesMut { } impl fmt::Debug for BytesMut { fn fmt(&self, fmt: &mut fmt::Formatter) -> fmt::Result { fmt::Debug::fmt(self.inner.as_ref(), fmt) } } impl hash::Hash for BytesMut { fn hash<H>(&self, state: &mut H) where H: hash::Hasher { let s: &[u8] = self.as_ref(); s.hash(state); } } impl Borrow<[u8]> for BytesMut { fn borrow(&self) -> &[u8] { self.as_ref() } } impl fmt::Write for BytesMut { fn write_str(&mut self, s: &str) -> fmt::Result { BufMut::put(self, s); Ok(()) } fn write_fmt(&mut self, args: fmt::Arguments) -> fmt::Result { fmt::write(self, args) } } impl Clone for BytesMut { fn clone(&self) -> BytesMut { BytesMut::from(&self[..]) } } /* * * ===== Inner ===== * */ impl Inner { /// Return a slice for the handle's view into the shared buffer #[inline] fn as_ref(&self) -> &[u8] { unsafe { if self.is_inline() { slice::from_raw_parts(self.inline_ptr(), self.inline_len()) } else { slice::from_raw_parts(self.ptr, self.len) } } } /// Return a mutable slice for the handle's view into the shared buffer #[inline] fn as_mut(&mut self) -> &mut [u8] { debug_assert!(!self.is_static()); unsafe { if self.is_inline() { slice::from_raw_parts_mut(self.inline_ptr(), self.inline_len()) } else { slice::from_raw_parts_mut(self.ptr, self.len) } } } /// Return a mutable slice for the handle's view into the shared buffer /// including potentially uninitialized bytes. #[inline] unsafe fn as_raw(&mut self) -> &mut [u8] { debug_assert!(!self.is_static()); if self.is_inline() { slice::from_raw_parts_mut(self.inline_ptr(), INLINE_CAP) } else { slice::from_raw_parts_mut(self.ptr, self.cap) } } #[inline] fn len(&self) -> usize { if self.is_inline() { self.inline_len() } else { self.len } } /// Pointer to the start of the inline buffer #[inline] unsafe fn inline_ptr(&self) -> *mut u8 { (self as *const Inner as *mut Inner as *mut u8) .offset(INLINE_DATA_OFFSET) } #[inline] fn inline_len(&self) -> usize { let p: &usize = unsafe { mem::transmute(&self.arc) }; (p & INLINE_LEN_MASK) >> INLINE_LEN_OFFSET } /// Set the length of the inline buffer. This is done by writing to the /// least significant byte of the `arc` field. #[inline] fn set_inline_len(&mut self, len: usize) { debug_assert!(len <= INLINE_CAP); let p: &mut usize = unsafe { mem::transmute(&mut self.arc) }; *p = (*p & !INLINE_LEN_MASK) | (len << INLINE_LEN_OFFSET); } /// slice. #[inline] unsafe fn set_len(&mut self, len: usize) { if self.is_inline() { assert!(len <= INLINE_CAP); self.set_inline_len(len); } else { assert!(len <= self.cap); self.len = len; } } #[inline] fn is_empty(&self) -> bool { self.len() == 0 } #[inline] fn capacity(&self) -> usize { if self.is_inline() { INLINE_CAP } else { self.cap } } fn split_off(&mut self, at: usize) -> Inner { let mut other = self.shallow_clone(); unsafe { other.set_start(at); self.set_end(at); } return other } fn drain_to(&mut self, at: usize) -> Inner { let mut other = self.shallow_clone(); unsafe { other.set_end(at); self.set_start(at); } return other } unsafe fn set_start(&mut self, start: usize) { // This function should never be called when the buffer is still backed // by a `Vec<u8>` debug_assert!(self.is_shared()); // Setting the start to 0 is a no-op, so return early if this is the // case. if start == 0 { return; } // Always check `inline` first, because if the handle is using inline // data storage, all of the `Inner` struct fields will be gibberish. if self.is_inline() { assert!(start <= INLINE_CAP); let len = self.inline_len(); if len <= start { self.set_inline_len(0); } else { // `set_start` is essentially shifting data off the front of the // view. Inlined buffers only track the length of the slice. // So, to update the start, the data at the new starting point // is copied to the beginning of the buffer. let new_len = len - start; let dst = self.inline_ptr(); let src = (dst as *const u8).offset(start as isize); ptr::copy(src, dst, new_len); self.set_inline_len(new_len); } } else { assert!(start <= self.cap); // Updating the start of the view is setting `ptr` to point to the // new start and updating the `len` field to reflect the new length // of the view. self.ptr = self.ptr.offset(start as isize); if self.len >= start { self.len -= start; } else { self.len = 0; } self.cap -= start; } } unsafe fn set_end(&mut self, end: usize) { debug_assert!(self.is_shared()); // Always check `inline` first, because if the handle is using inline // data storage, all of the `Inner` struct fields will be gibberish. if self.is_inline() { assert!(end <= INLINE_CAP); let new_len = cmp::min(self.inline_len(), end); self.set_inline_len(new_len); } else { assert!(end <= self.cap); self.cap = end; self.len = cmp::min(self.len, end); } } /// Checks if it is safe to mutate the memory fn is_mut_safe(&mut self) -> bool { // Always check `inline` first, because if the handle is using inline // data storage, all of the `Inner` struct fields will be gibberish. if self.is_inline() { // Inlined buffers can always be mutated as the data is never shared // across handles. true } else { // The function requires `&mut self`, which guarantees a unique // reference to the current handle. This means that the `arc` field // *cannot* be concurrently mutated. As such, `Relaxed` ordering is // fine (since we aren't synchronizing with anything). // // TODO: No ordering? let arc = self.arc.load(Relaxed); // If the pointer is null, this is a non-shared handle and is mut // safe. if arc.is_null() { return true; } // Check if this is a static buffer if KIND_STATIC == arc as usize { return false; } // Otherwise, the underlying buffer is potentially shared with other // handles, so the ref_count needs to be checked. unsafe { return (*arc).is_unique(); } } } /// Increments the ref count. This should only be done if it is known that /// it can be done safely. As such, this fn is not public, instead other /// fns will use this one while maintaining the guarantees. fn shallow_clone(&self) -> Inner { // Always check `inline` first, because if the handle is using inline // data storage, all of the `Inner` struct fields will be gibberish. if self.is_inline() { // In this case, a shallow_clone still involves copying the data. unsafe { // TODO: Just copy the fields let mut inner: Inner = mem::uninitialized(); let len = self.inline_len(); inner.arc = AtomicPtr::new(KIND_INLINE as *mut Shared); inner.set_inline_len(len); inner.as_raw()[0..len].copy_from_slice(self.as_ref()); inner } } else { // The function requires `&self`, this means that `shallow_clone` // could be called concurrently. // // The first step is to load the value of `arc`. This will determine // how to proceed. The `Acquire` ordering synchronizes with the // `compare_and_swap` that comes later in this function. The goal is // to ensure that if `arc` is currently set to point to a `Shared`, // that the current thread acquires the associated memory. let mut arc = self.arc.load(Acquire); // If `arc` is null, then the buffer is still tracked in a // `Vec<u8>`. It is time to promote the vec to an `Arc`. This could // potentially be called concurrently, so some care must be taken. if arc.is_null() { unsafe { // First, allocate a new `Shared` instance containing the // `Vec` fields. It's important to note that `ptr`, `len`, // and `cap` cannot be mutated without having `&mut self`. // This means that these fields will not be concurrently // updated and since the buffer hasn't been promoted to an // `Arc`, those three fields still are the components of the // vector. let shared = Box::new(Shared { vec: Vec::from_raw_parts(self.ptr, self.len, self.cap), // Initialize refcount to 2. One for this reference, and one // for the new clone that will be returned from // `shallow_clone`. ref_count: AtomicUsize::new(2), }); let shared = Box::into_raw(shared); // The pointer should be aligned, so this assert should // always succeed. debug_assert!(0 == (shared as usize & 0b11)); // Try compare & swapping the pointer into the `arc` field. // `Release` is used synchronize with other threads that // will load the `arc` field. // // If the `compare_and_swap` fails, then the thread lost the // race to promote the buffer to shared. The `Acquire` // ordering will synchronize with the `compare_and_swap` // that happened in the other thread and the `Shared` // pointed to by `actual` will be visible. let actual = self.arc.compare_and_swap(arc, shared, AcqRel); if actual.is_null() { // The upgrade was successful, the new handle can be // returned. return Inner { arc: AtomicPtr::new(shared), .. *self }; } // The upgrade failed, a concurrent clone happened. Release // the allocation that was made in this thread, it will not // be needed. let shared: Box<Shared> = mem::transmute(shared); mem::forget(*shared); // Update the `arc` local variable and fall through to a ref // count update arc = actual; } } else if KIND_STATIC == arc as usize { // Static buffer return Inner { arc: AtomicPtr::new(arc), .. *self }; } // Buffer already promoted to shared storage, so increment ref // count. unsafe { // Relaxed ordering is acceptable as the memory has already been // acquired via the `Acquire` load above. let old_size = (*arc).ref_count.fetch_add(1, Relaxed); if old_size == usize::MAX { panic!(); // TODO: abort } } Inner { arc: AtomicPtr::new(arc), .. *self } } } #[inline] fn reserve(&mut self, additional: usize) { let len = self.len(); let rem = self.capacity() - len; if additional <= rem { // The handle can already store at least `additional` more bytes, so // there is no further work needed to be done. return; } // Always check `inline` first, because if the handle is using inline // data storage, all of the `Inner` struct fields will be gibberish. if self.is_inline() { let new_cap = len + additional; // Promote to a vector let mut v = Vec::with_capacity(new_cap); v.extend_from_slice(self.as_ref()); self.ptr = v.as_mut_ptr(); self.len = v.len(); self.cap = v.capacity(); self.arc = AtomicPtr::new(ptr::null_mut()); mem::forget(v); return; } // `Relaxed` is Ok here (and really, no synchronization is necessary) // due to having a `&mut self` pointer. The `&mut self` pointer ensures // that there is no concurrent access on `self`. let arc = self.arc.load(Relaxed); if arc.is_null() { // Currently backed by a vector, so just use `Vector::reserve`. unsafe { let mut v = Vec::from_raw_parts(self.ptr, self.len, self.cap); v.reserve(additional); // Update the info self.ptr = v.as_mut_ptr(); self.len = v.len(); self.cap = v.capacity(); // Drop the vec reference mem::forget(v); return; } } debug_assert!(!self.is_static()); // Reserving involves abandoning the currently shared buffer and // allocating a new vector with the requested capacity. // // Compute the new capacity let mut new_cap = len + additional; unsafe { // First, try to reclaim the buffer. This is possible if the current // handle is the only outstanding handle pointing to the buffer. if (*arc).is_unique() { // This is the only handle to the buffer. It can be reclaimed. // However, before doing the work of copying data, check to make // sure that the vector has enough capacity. let v = &mut (*arc).vec; if v.capacity() >= new_cap { // The capacity is sufficient, reclaim the buffer let ptr = v.as_mut_ptr(); ptr::copy(self.ptr, ptr, len); self.ptr = ptr; self.cap = v.capacity(); return; } // The vector capacity is not sufficient. The reserve request is // asking for more than the initial buffer capacity. Allocate more // than requested if `new_cap` is not much bigger than the current // capacity. new_cap = cmp::max(v.capacity() << 1, new_cap); } } // Create a new vector to store the data let mut v = Vec::with_capacity(new_cap.next_power_of_two()); // Copy the bytes v.extend_from_slice(self.as_ref()); // Release the shared handle. This must be done *after* the bytes are // copied. release_shared(arc); // Update self self.ptr = v.as_mut_ptr(); self.len = v.len(); self.cap = v.capacity(); self.arc = AtomicPtr::new(ptr::null_mut()); // Forget the vector handle mem::forget(v); } /// Returns true if the buffer is stored inline #[inline] fn is_inline(&self) -> bool { // This function is going to probably raise some eyebrows. The function // returns true if the buffer is stored inline. This is done by checking // the least significant bit in the `arc` field. // // Now, you may notice that `arc` is an `AtomicPtr` and this is // accessing it as a normal field without performing an atomic load... // // Again, the function only cares about the least significant bit, and // this bit is set when `Inner` is created and never changed after that. // All platforms have atomic "word" operations and won't randomly flip // bits, so even without any explicit atomic operations, reading the // flag will be correct. // // This function is very critical performance wise as it is called for // every operation. Performing an atomic load would mess with the // compiler's ability to optimize. Simple benchmarks show up to a 10% // slowdown using a `Relaxed` atomic load on x86. #[cfg(target_endian = "little")] #[inline] fn imp(arc: &AtomicPtr<Shared>) -> bool { unsafe { let p: &u8 = mem::transmute(arc); *p & (KIND_INLINE as u8) == (KIND_INLINE as u8) } } #[cfg(target_endian = "big")] #[inline] fn imp(arc: &AtomicPtr<Shared>) -> bool { unsafe { let p: &usize = mem::transmute(arc); *p & KIND_INLINE == KIND_INLINE } } imp(&self.arc) } /// Used for `debug_assert` statements #[inline] fn is_shared(&self) -> bool { self.is_inline() || !self.arc.load(Relaxed).is_null() } /// Used for `debug_assert` statements #[inline] fn is_static(&self) -> bool { !self.is_inline() && self.arc.load(Relaxed) as usize == KIND_STATIC } } impl Drop for Inner2 { fn drop(&mut self) { // Always check `inline` first, because if the handle is using inline // data storage, all of the `Inner` struct fields will be gibberish. if self.is_inline() { return; } // Acquire is needed here to ensure that the `Shared` memory is // visible. let arc = self.arc.load(Acquire); if arc as usize == KIND_STATIC { // Static buffer, no work to do return; } if arc.is_null() { // Vector storage, free the vector unsafe { let _ = Vec::from_raw_parts(self.ptr, self.len, self.cap); } return; } release_shared(arc); } } fn release_shared(ptr: *mut Shared) { // `Shared` storage... follow the drop steps from Arc. unsafe { if (*ptr).ref_count.fetch_sub(1, Release) != 1 { return; } // This fence is needed to prevent reordering of use of the data and // deletion of the data. Because it is marked `Release`, the decreasing // of the reference count synchronizes with this `Acquire` fence. This // means that use of the data happens before decreasing the reference // count, which happens before this fence, which happens before the // deletion of the data. // // As explained in the [Boost documentation][1], // // > It is important to enforce any possible access to the object in one // > thread (through an existing reference) to *happen before* deleting // > the object in a different thread. This is achieved by a "release" // > operation after dropping a reference (any access to the object // > through this reference must obviously happened before), and an // > "acquire" operation before deleting the object. // // [1]: (www.boost.org/doc/libs/1_55_0/doc/html/atomic/usage_examples.html) atomic::fence(Acquire); // Drop the data let _: Box<Shared> = mem::transmute(ptr); } } impl Shared { fn is_unique(&self) -> bool { // The goal is to check if the current handle is the only handle // that currently has access to the buffer. This is done by // checking if the `ref_count` is currently 1. // // The `Acquire` ordering synchronizes with the `Release` as // part of the `fetch_sub` in `release_shared`. The `fetch_sub` // operation guarantees that any mutations done in other threads // are ordered before the `ref_count` is decremented. As such, // this `Acquire` will guarantee that those mutations are // visible to the current thread. self.ref_count.load(Acquire) == 1 } } unsafe impl Send for Inner {} unsafe impl Sync for Inner {} /* * * ===== impl Inner2 ===== * */ impl ops::Deref for Inner2 { type Target = Inner; fn deref(&self) -> &Inner { &self.inner } } impl ops::DerefMut for Inner2 { fn deref_mut(&mut self) -> &mut Inner { &mut self.inner } } /* * * ===== PartialEq / PartialOrd ===== * */ impl PartialEq<[u8]> for BytesMut { fn eq(&self, other: &[u8]) -> bool { &**self == other } } impl PartialOrd<[u8]> for BytesMut { fn partial_cmp(&self, other: &[u8]) -> Option<cmp::Ordering> { (**self).partial_cmp(other) } } impl PartialEq<BytesMut> for [u8] { fn eq(&self, other: &BytesMut) -> bool { *other == *self } } impl PartialOrd<BytesMut> for [u8] { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<str> for BytesMut { fn eq(&self, other: &str) -> bool { &**self == other.as_bytes() } } impl PartialOrd<str> for BytesMut { fn partial_cmp(&self, other: &str) -> Option<cmp::Ordering> { (**self).partial_cmp(other.as_bytes()) } } impl PartialEq<BytesMut> for str { fn eq(&self, other: &BytesMut) -> bool { *other == *self } } impl PartialOrd<BytesMut> for str { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<Vec<u8>> for BytesMut { fn eq(&self, other: &Vec<u8>) -> bool { *self == &other[..] } } impl PartialOrd<Vec<u8>> for BytesMut { fn partial_cmp(&self, other: &Vec<u8>) -> Option<cmp::Ordering> { (**self).partial_cmp(&other[..]) } } impl PartialEq<BytesMut> for Vec<u8> { fn eq(&self, other: &BytesMut) -> bool { *other == *self } } impl PartialOrd<BytesMut> for Vec<u8> { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<String> for BytesMut { fn eq(&self, other: &String) -> bool { *self == &other[..] } } impl PartialOrd<String> for BytesMut { fn partial_cmp(&self, other: &String) -> Option<cmp::Ordering> { (**self).partial_cmp(other.as_bytes()) } } impl PartialEq<BytesMut> for String { fn eq(&self, other: &BytesMut) -> bool { *other == *self } } impl PartialOrd<BytesMut> for String { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl<'a, T: ?Sized> PartialEq<&'a T> for BytesMut where BytesMut: PartialEq<T> { fn eq(&self, other: &&'a T) -> bool { *self == **other } } impl<'a, T: ?Sized> PartialOrd<&'a T> for BytesMut where BytesMut: PartialOrd<T> { fn partial_cmp(&self, other: &&'a T) -> Option<cmp::Ordering> { self.partial_cmp(*other) } } impl<'a> PartialEq<BytesMut> for &'a [u8] { fn eq(&self, other: &BytesMut) -> bool { *other == *self } } impl<'a> PartialOrd<BytesMut> for &'a [u8] { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl<'a> PartialEq<BytesMut> for &'a str { fn eq(&self, other: &BytesMut) -> bool { *other == *self } } impl<'a> PartialOrd<BytesMut> for &'a str { fn partial_cmp(&self, other: &BytesMut) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<[u8]> for Bytes { fn eq(&self, other: &[u8]) -> bool { self.inner.as_ref() == other } } impl PartialOrd<[u8]> for Bytes { fn partial_cmp(&self, other: &[u8]) -> Option<cmp::Ordering> { self.inner.as_ref().partial_cmp(other) } } impl PartialEq<Bytes> for [u8] { fn eq(&self, other: &Bytes) -> bool { *other == *self } } impl PartialOrd<Bytes> for [u8] { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<str> for Bytes { fn eq(&self, other: &str) -> bool { self.inner.as_ref() == other.as_bytes() } } impl PartialOrd<str> for Bytes { fn partial_cmp(&self, other: &str) -> Option<cmp::Ordering> { self.inner.as_ref().partial_cmp(other.as_bytes()) } } impl PartialEq<Bytes> for str { fn eq(&self, other: &Bytes) -> bool { *other == *self } } impl PartialOrd<Bytes> for str { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<Vec<u8>> for Bytes { fn eq(&self, other: &Vec<u8>) -> bool { *self == &other[..] } } impl PartialOrd<Vec<u8>> for Bytes { fn partial_cmp(&self, other: &Vec<u8>) -> Option<cmp::Ordering> { self.inner.as_ref().partial_cmp(&other[..]) } } impl PartialEq<Bytes> for Vec<u8> { fn eq(&self, other: &Bytes) -> bool { *other == *self } } impl PartialOrd<Bytes> for Vec<u8> { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl PartialEq<String> for Bytes { fn eq(&self, other: &String) -> bool { *self == &other[..] } } impl PartialOrd<String> for Bytes { fn partial_cmp(&self, other: &String) -> Option<cmp::Ordering> { self.inner.as_ref().partial_cmp(other.as_bytes()) } } impl PartialEq<Bytes> for String { fn eq(&self, other: &Bytes) -> bool { *other == *self } } impl PartialOrd<Bytes> for String { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl<'a> PartialEq<Bytes> for &'a [u8] { fn eq(&self, other: &Bytes) -> bool { *other == *self } } impl<'a> PartialOrd<Bytes> for &'a [u8] { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl<'a> PartialEq<Bytes> for &'a str { fn eq(&self, other: &Bytes) -> bool { *other == *self } } impl<'a> PartialOrd<Bytes> for &'a str { fn partial_cmp(&self, other: &Bytes) -> Option<cmp::Ordering> { other.partial_cmp(self) } } impl<'a, T: ?Sized> PartialEq<&'a T> for Bytes where Bytes: PartialEq<T> { fn eq(&self, other: &&'a T) -> bool { *self == **other } } impl<'a, T: ?Sized> PartialOrd<&'a T> for Bytes where Bytes: PartialOrd<T> { fn partial_cmp(&self, other: &&'a T) -> Option<cmp::Ordering> { self.partial_cmp(&**other) } }