sys/vfs/ufs/ffs_softdep.c
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 2156 2157 2158 2159 2160 2161 2162 2163 2164 2165 2166 2167 2168 2169 2170 2171 2172 2173 2174 2175 2176 2177 2178 2179 2180 2181 2182 2183 2184 2185 2186 2187 2188 2189 2190 2191 2192 2193 2194 2195 2196 2197 2198 2199 2200 2201 2202 2203 2204 2205 2206 2207 2208 2209 2210 2211 2212 2213 2214 2215 2216 2217 2218 2219 2220 2221 2222 2223 2224 2225 2226 2227 2228 2229 2230 2231 2232 2233 2234 2235 2236 2237 2238 2239 2240 2241 2242 2243 2244 2245 2246 2247 2248 2249 2250 2251 2252 2253 2254 2255 2256 2257 2258 2259 2260 2261 2262 2263 2264 2265 2266 2267 2268 2269 2270 2271 2272 2273 2274 2275 2276 2277 2278 2279 2280 2281 2282 2283 2284 2285 2286 2287 2288 2289 2290 2291 2292 2293 2294 2295 2296 2297 2298 2299 2300 2301 2302 2303 2304 2305 2306 2307 2308 2309 2310 2311 2312 2313 2314 2315 2316 2317 2318 2319 2320 2321 2322 2323 2324 2325 2326 2327 2328 2329 2330 2331 2332 2333 2334 2335 2336 2337 2338 2339 2340 2341 2342 2343 2344 2345 2346 2347 2348 2349 2350 2351 2352 2353 2354 2355 2356 2357 2358 2359 2360 2361 2362 2363 2364 2365 2366 2367 2368 2369 2370 2371 2372 2373 2374 2375 2376 2377 2378 2379 2380 2381 2382 2383 2384 2385 2386 2387 2388 2389 2390 2391 2392 2393 2394 2395 2396 2397 2398 2399 2400 2401 2402 2403 2404 2405 2406 2407 2408 2409 2410 2411 2412 2413 2414 2415 2416 2417 2418 2419 2420 2421 2422 2423 2424 2425 2426 2427 2428 2429 2430 2431 2432 2433 2434 2435 2436 2437 2438 2439 2440 2441 2442 2443 2444 2445 2446 2447 2448 2449 2450 2451 2452 2453 2454 2455 2456 2457 2458 2459 2460 2461 2462 2463 2464 2465 2466 2467 2468 2469 2470 2471 2472 2473 2474 2475 2476 2477 2478 2479 2480 2481 2482 2483 2484 2485 2486 2487 2488 2489 2490 2491 2492 2493 2494 2495 2496 2497 2498 2499 2500 2501 2502 2503 2504 2505 2506 2507 2508 2509 2510 2511 2512 2513 2514 2515 2516 2517 2518 2519 2520 2521 2522 2523 2524 2525 2526 2527 2528 2529 2530 2531 2532 2533 2534 2535 2536 2537 2538 2539 2540 2541 2542 2543 2544 2545 2546 2547 2548 2549 2550 2551 2552 2553 2554 2555 2556 2557 2558 2559 2560 2561 2562 2563 2564 2565 2566 2567 2568 2569 2570 2571 2572 2573 2574 2575 2576 2577 2578 2579 2580 2581 2582 2583 2584 2585 2586 2587 2588 2589 2590 2591 2592 2593 2594 2595 2596 2597 2598 2599 2600 2601 2602 2603 2604 2605 2606 2607 2608 2609 2610 2611 2612 2613 2614 2615 2616 2617 2618 2619 2620 2621 2622 2623 2624 2625 2626 2627 2628 2629 2630 2631 2632 2633 2634 2635 2636 2637 2638 2639 2640 2641 2642 2643 2644 2645 2646 2647 2648 2649 2650 2651 2652 2653 2654 2655 2656 2657 2658 2659 2660 2661 2662 2663 2664 2665 2666 2667 2668 2669 2670 2671 2672 2673 2674 2675 2676 2677 2678 2679 2680 2681 2682 2683 2684 2685 2686 2687 2688 2689 2690 2691 2692 2693 2694 2695 2696 2697 2698 2699 2700 2701 2702 2703 2704 2705 2706 2707 2708 2709 2710 2711 2712 2713 2714 2715 2716 2717 2718 2719 2720 2721 2722 2723 2724 2725 2726 2727 2728 2729 2730 2731 2732 2733 2734 2735 2736 2737 2738 2739 2740 2741 2742 2743 2744 2745 2746 2747 2748 2749 2750 2751 2752 2753 2754 2755 2756 2757 2758 2759 2760 2761 2762 2763 2764 2765 2766 2767 2768 2769 2770 2771 2772 2773 2774 2775 2776 2777 2778 2779 2780 2781 2782 2783 2784 2785 2786 2787 2788 2789 2790 2791 2792 2793 2794 2795 2796 2797 2798 2799 2800 2801 2802 2803 2804 2805 2806 2807 2808 2809 2810 2811 2812 2813 2814 2815 2816 2817 2818 2819 2820 2821 2822 2823 2824 2825 2826 2827 2828 2829 2830 2831 2832 2833 2834 2835 2836 2837 2838 2839 2840 2841 2842 2843 2844 2845 2846 2847 2848 2849 2850 2851 2852 2853 2854 2855 2856 2857 2858 2859 2860 2861 2862 2863 2864 2865 2866 2867 2868 2869 2870 2871 2872 2873 2874 2875 2876 2877 2878 2879 2880 2881 2882 2883 2884 2885 2886 2887 2888 2889 2890 2891 2892 2893 2894 2895 2896 2897 2898 2899 2900 2901 2902 2903 2904 2905 2906 2907 2908 2909 2910 2911 2912 2913 2914 2915 2916 2917 2918 2919 2920 2921 2922 2923 2924 2925 2926 2927 2928 2929 2930 2931 2932 2933 2934 2935 2936 2937 2938 2939 2940 2941 2942 2943 2944 2945 2946 2947 2948 2949 2950 2951 2952 2953 2954 2955 2956 2957 2958 2959 2960 2961 2962 2963 2964 2965 2966 2967 2968 2969 2970 2971 2972 2973 2974 2975 2976 2977 2978 2979 2980 2981 2982 2983 2984 2985 2986 2987 2988 2989 2990 2991 2992 2993 2994 2995 2996 2997 2998 2999 3000 3001 3002 3003 3004 3005 3006 3007 3008 3009 3010 3011 3012 3013 3014 3015 3016 3017 3018 3019 3020 3021 3022 3023 3024 3025 3026 3027 3028 3029 3030 3031 3032 3033 3034 3035 3036 3037 3038 3039 3040 3041 3042 3043 3044 3045 3046 3047 3048 3049 3050 3051 3052 3053 3054 3055 3056 3057 3058 3059 3060 3061 3062 3063 3064 3065 3066 3067 3068 3069 3070 3071 3072 3073 3074 3075 3076 3077 3078 3079 3080 3081 3082 3083 3084 3085 3086 3087 3088 3089 3090 3091 3092 3093 3094 3095 3096 3097 3098 3099 3100 3101 3102 3103 3104 3105 3106 3107 3108 3109 3110 3111 3112 3113 3114 3115 3116 3117 3118 3119 3120 3121 3122 3123 3124 3125 3126 3127 3128 3129 3130 3131 3132 3133 3134 3135 3136 3137 3138 3139 3140 3141 3142 3143 3144 3145 3146 3147 3148 3149 3150 3151 3152 3153 3154 3155 3156 3157 3158 3159 3160 3161 3162 3163 3164 3165 3166 3167 3168 3169 3170 3171 3172 3173 3174 3175 3176 3177 3178 3179 3180 3181 3182 3183 3184 3185 3186 3187 3188 3189 3190 3191 3192 3193 3194 3195 3196 3197 3198 3199 3200 3201 3202 3203 3204 3205 3206 3207 3208 3209 3210 3211 3212 3213 3214 3215 3216 3217 3218 3219 3220 3221 3222 3223 3224 3225 3226 3227 3228 3229 3230 3231 3232 3233 3234 3235 3236 3237 3238 3239 3240 3241 3242 3243 3244 3245 3246 3247 3248 3249 3250 3251 3252 3253 3254 3255 3256 3257 3258 3259 3260 3261 3262 3263 3264 3265 3266 3267 3268 3269 3270 3271 3272 3273 3274 3275 3276 3277 3278 3279 3280 3281 3282 3283 3284 3285 3286 3287 3288 3289 3290 3291 3292 3293 3294 3295 3296 3297 3298 3299 3300 3301 3302 3303 3304 3305 3306 3307 3308 3309 3310 3311 3312 3313 3314 3315 3316 3317 3318 3319 3320 3321 3322 3323 3324 3325 3326 3327 3328 3329 3330 3331 3332 3333 3334 3335 3336 3337 3338 3339 3340 3341 3342 3343 3344 3345 3346 3347 3348 3349 3350 3351 3352 3353 3354 3355 3356 3357 3358 3359 3360 3361 3362 3363 3364 3365 3366 3367 3368 3369 3370 3371 3372 3373 3374 3375 3376 3377 3378 3379 3380 3381 3382 3383 3384 3385 3386 3387 3388 3389 3390 3391 3392 3393 3394 3395 3396 3397 3398 3399 3400 3401 3402 3403 3404 3405 3406 3407 3408 3409 3410 3411 3412 3413 3414 3415 3416 3417 3418 3419 3420 3421 3422 3423 3424 3425 3426 3427 3428 3429 3430 3431 3432 3433 3434 3435 3436 3437 3438 3439 3440 3441 3442 3443 3444 3445 3446 3447 3448 3449 3450 3451 3452 3453 3454 3455 3456 3457 3458 3459 3460 3461 3462 3463 3464 3465 3466 3467 3468 3469 3470 3471 3472 3473 3474 3475 3476 3477 3478 3479 3480 3481 3482 3483 3484 3485 3486 3487 3488 3489 3490 3491 3492 3493 3494 3495 3496 3497 3498 3499 3500 3501 3502 3503 3504 3505 3506 3507 3508 3509 3510 3511 3512 3513 3514 3515 3516 3517 3518 3519 3520 3521 3522 3523 3524 3525 3526 3527 3528 3529 3530 3531 3532 3533 3534 3535 3536 3537 3538 3539 3540 3541 3542 3543 3544 3545 3546 3547 3548 3549 3550 3551 3552 3553 3554 3555 3556 3557 3558 3559 3560 3561 3562 3563 3564 3565 3566 3567 3568 3569 3570 3571 3572 3573 3574 3575 3576 3577 3578 3579 3580 3581 3582 3583 3584 3585 3586 3587 3588 3589 3590 3591 3592 3593 3594 3595 3596 3597 3598 3599 3600 3601 3602 3603 3604 3605 3606 3607 3608 3609 3610 3611 3612 3613 3614 3615 3616 3617 3618 3619 3620 3621 3622 3623 3624 3625 3626 3627 3628 3629 3630 3631 3632 3633 3634 3635 3636 3637 3638 3639 3640 3641 3642 3643 3644 3645 3646 3647 3648 3649 3650 3651 3652 3653 3654 3655 3656 3657 3658 3659 3660 3661 3662 3663 3664 3665 3666 3667 3668 3669 3670 3671 3672 3673 3674 3675 3676 3677 3678 3679 3680 3681 3682 3683 3684 3685 3686 3687 3688 3689 3690 3691 3692 3693 3694 3695 3696 3697 3698 3699 3700 3701 3702 3703 3704 3705 3706 3707 3708 3709 3710 3711 3712 3713 3714 3715 3716 3717 3718 3719 3720 3721 3722 3723 3724 3725 3726 3727 3728 3729 3730 3731 3732 3733 3734 3735 3736 3737 3738 3739 3740 3741 3742 3743 3744 3745 3746 3747 3748 3749 3750 3751 3752 3753 3754 3755 3756 3757 3758 3759 3760 3761 3762 3763 3764 3765 3766 3767 3768 3769 3770 3771 3772 3773 3774 3775 3776 3777 3778 3779 3780 3781 3782 3783 3784 3785 3786 3787 3788 3789 3790 3791 3792 3793 3794 3795 3796 3797 3798 3799 3800 3801 3802 3803 3804 3805 3806 3807 3808 3809 3810 3811 3812 3813 3814 3815 3816 3817 3818 3819 3820 3821 3822 3823 3824 3825 3826 3827 3828 3829 3830 3831 3832 3833 3834 3835 3836 3837 3838 3839 3840 3841 3842 3843 3844 3845 3846 3847 3848 3849 3850 3851 3852 3853 3854 3855 3856 3857 3858 3859 3860 3861 3862 3863 3864 3865 3866 3867 3868 3869 3870 3871 3872 3873 3874 3875 3876 3877 3878 3879 3880 3881 3882 3883 3884 3885 3886 3887 3888 3889 3890 3891 3892 3893 3894 3895 3896 3897 3898 3899 3900 3901 3902 3903 3904 3905 3906 3907 3908 3909 3910 3911 3912 3913 3914 3915 3916 3917 3918 3919 3920 3921 3922 3923 3924 3925 3926 3927 3928 3929 3930 3931 3932 3933 3934 3935 3936 3937 3938 3939 3940 3941 3942 3943 3944 3945 3946 3947 3948 3949 3950 3951 3952 3953 3954 3955 3956 3957 3958 3959 3960 3961 3962 3963 3964 3965 3966 3967 3968 3969 3970 3971 3972 3973 3974 3975 3976 3977 3978 3979 3980 3981 3982 3983 3984 3985 3986 3987 3988 3989 3990 3991 3992 3993 3994 3995 3996 3997 3998 3999 4000 4001 4002 4003 4004 4005 4006 4007 4008 4009 4010 4011 4012 4013 4014 4015 4016 4017 4018 4019 4020 4021 4022 4023 4024 4025 4026 4027 4028 4029 4030 4031 4032 4033 4034 4035 4036 4037 4038 4039 4040 4041 4042 4043 4044 4045 4046 4047 4048 4049 4050 4051 4052 4053 4054 4055 4056 4057 4058 4059 4060 4061 4062 4063 4064 4065 4066 4067 4068 4069 4070 4071 4072 4073 4074 4075 4076 4077 4078 4079 4080 4081 4082 4083 4084 4085 4086 4087 4088 4089 4090 4091 4092 4093 4094 4095 4096 4097 4098 4099 4100 4101 4102 4103 4104 4105 4106 4107 4108 4109 4110 4111 4112 4113 4114 4115 4116 4117 4118 4119 4120 4121 4122 4123 4124 4125 4126 4127 4128 4129 4130 4131 4132 4133 4134 4135 4136 4137 4138 4139 4140 4141 4142 4143 4144 4145 4146 4147 4148 4149 4150 4151 4152 4153 4154 4155 4156 4157 4158 4159 4160 4161 4162 4163 4164 4165 4166 4167 4168 4169 4170 4171 4172 4173 4174 4175 4176 4177 4178 4179 4180 4181 4182 4183 4184 4185 4186 4187 4188 4189 4190 4191 4192 4193 4194 4195 4196 4197 4198 4199 4200 4201 4202 4203 4204 4205 4206 4207 4208 4209 4210 4211 4212 4213 4214 4215 4216 4217 4218 4219 4220 4221 4222 4223 4224 4225 4226 4227 4228 4229 4230 4231 4232 4233 4234 4235 4236 4237 4238 4239 4240 4241 4242 4243 4244 4245 4246 4247 4248 4249 4250 4251 4252 4253 4254 4255 4256 4257 4258 4259 4260 4261 4262 4263 4264 4265 4266 4267 4268 4269 4270 4271 4272 4273 4274 4275 4276 4277 4278 4279 4280 4281 4282 4283 4284 4285 4286 4287 4288 4289 4290 4291 4292 4293 4294 4295 4296 4297 4298 4299 4300 4301 4302 4303 4304 4305 4306 4307 4308 4309 4310 4311 4312 4313 4314 4315 4316 4317 4318 4319 4320 4321 4322 4323 4324 4325 4326 4327 4328 4329 4330 4331 4332 4333 4334 4335 4336 4337 4338 4339 4340 4341 4342 4343 4344 4345 4346 4347 4348 4349 4350 4351 4352 4353 4354 4355 4356 4357 4358 4359 4360 4361 4362 4363 4364 4365 4366 4367 4368 4369 4370 4371 4372 4373 4374 4375 4376 4377 4378 4379 4380 4381 4382 4383 4384 4385 4386 4387 4388 4389 4390 4391 4392 4393 4394 4395 4396 4397 4398 4399 4400 4401 4402 4403 4404 4405 4406 4407 4408 4409 4410 4411 4412 4413 4414 4415 4416 4417 4418 4419 4420 4421 4422 4423 4424 4425 4426 4427 4428 4429 4430 4431 4432 4433 4434 4435 4436 4437 4438 4439 4440 4441 4442 4443 4444 4445 4446 4447 4448 4449 4450 4451 4452 4453 4454 4455 4456 4457 4458 4459 4460 4461 4462 4463 4464 4465 4466 4467 4468 4469 4470 4471 4472 4473 4474 4475 4476 4477 4478 4479 4480 4481 4482 4483 4484 4485 4486 4487 4488 4489 4490 4491 4492 4493 4494 4495 4496 4497 4498 4499 4500 4501 4502 4503 4504 4505 4506 4507 4508 4509 4510 4511 4512 4513 4514 4515 4516 4517 4518 4519 4520 4521 4522 4523 4524 4525 4526 4527 4528 4529 4530 4531 4532 4533 4534 4535 4536 4537 4538 4539 4540 4541 4542 4543 4544 4545 4546 4547 4548 4549 4550 4551 4552 4553 4554 4555 4556 4557 4558 4559 4560 4561 4562 4563 4564 4565 4566 4567 4568 4569 4570 4571 4572 4573 4574 4575 4576 4577 4578 4579 4580 4581 4582 4583 4584 4585 4586 4587 4588 4589 4590 4591 4592 4593 4594 4595 4596 4597 4598 4599 4600 4601 4602 4603 4604 4605 4606 4607 4608 4609 4610 4611 4612 4613 4614 4615 4616 4617 4618 4619 4620 4621 4622 4623 4624 4625 4626 4627 4628 4629 4630 4631 4632 4633 4634 4635 4636 4637 4638 4639 4640 4641 4642 4643 4644 4645 4646 4647 4648 4649 4650 4651 4652 4653 4654 4655 4656 4657 4658 4659 4660 4661 4662 4663 4664 4665 4666 4667 4668 4669 4670 4671 4672 4673 4674 4675 4676 4677 4678 4679 4680 4681 4682 4683 4684 4685 4686 4687 4688 4689 4690 4691 4692 4693 4694 4695 4696 4697 4698 4699 4700 4701 4702 4703 4704 4705 4706 4707 4708 4709 4710 4711 4712 4713 4714 4715 4716 4717 4718 4719 4720 4721 4722 4723 4724 4725 4726 4727 4728 4729 4730 4731 4732 4733 4734 4735 4736 4737 4738 4739 4740 4741 4742 4743 4744 4745 4746 4747 4748 4749 4750 4751 4752 4753 4754 4755 4756 4757 4758 4759 4760 4761 4762 4763 4764 4765 4766 4767 4768 4769 4770 4771 4772 4773 4774 4775 4776 4777 4778 4779 4780 4781 4782 4783 4784 4785 4786 4787 4788 4789 4790 4791 4792 4793 4794 4795 4796 4797 4798 4799 4800 4801 4802 4803 4804 4805 4806 4807 4808 4809 4810 4811 4812 4813 4814 4815 4816 4817 4818 4819 4820 4821 4822 4823 4824 4825 4826 4827 4828 4829 4830 4831 4832 4833 4834 4835 4836 4837 4838 4839 4840 4841 4842 4843 4844 4845 4846 4847 4848 4849 4850 4851 4852 4853 4854 4855 4856 4857 4858 4859 4860 4861 4862 4863 4864 4865 4866 4867 4868 4869 4870 4871 4872 4873 4874 4875 4876 4877 4878 4879 4880 4881 4882 4883 4884 4885 4886 4887 4888 4889 4890 4891 4892 4893 4894 4895 4896 4897 4898 4899 4900 4901 4902 4903 4904 4905 4906 4907 4908 4909 4910 4911 4912 4913 4914 4915 4916 4917 4918 4919 4920 4921 4922 4923 4924 4925 4926 4927 4928 4929 4930 4931 4932 4933 4934 4935 4936 4937 4938 4939 4940 4941 4942 4943 4944 4945 4946 4947 4948 4949 4950 4951 4952 4953 4954 4955 4956 4957 4958 4959 4960 4961 4962 4963 4964 4965 4966 4967 4968 4969 4970 4971 4972 4973 4974 4975 4976 4977 4978 4979 4980 4981 4982 4983 4984 4985 4986 4987 4988 4989 4990 4991 4992 4993 4994 4995 4996 4997 4998 4999 5000 5001 5002 5003 5004 5005 5006 5007 5008 5009 5010 5011 5012 5013 5014 5015 5016 5017 5018 5019 5020 5021 5022 5023 5024 5025 5026 5027 5028 5029 5030 5031 5032 5033 5034 5035 5036 5037 5038 5039 5040 5041 5042 5043 5044 5045 5046 5047 5048 5049 5050 5051 5052 5053 5054 5055 5056 5057 5058 5059 5060 5061 5062 5063 5064 5065 5066 5067 5068 5069 5070 5071 5072 5073 5074 5075 5076 5077 5078 5079 5080 5081 5082 5083 5084 5085 5086 5087 5088 5089 5090 5091 5092 5093 5094 5095 5096 5097 5098 5099 5100 5101 5102 5103 5104 5105 5106 5107 5108 5109 5110 5111 5112 | /* * Copyright 1998, 2000 Marshall Kirk McKusick. All Rights Reserved. * * The soft updates code is derived from the appendix of a University * of Michigan technical report (Gregory R. Ganger and Yale N. Patt, * "Soft Updates: A Solution to the Metadata Update Problem in File * Systems", CSE-TR-254-95, August 1995). * * Further information about soft updates can be obtained from: * * Marshall Kirk McKusick http://www.mckusick.com/softdep/ * 1614 Oxford Street mckusick@mckusick.com * Berkeley, CA 94709-1608 +1-510-843-9542 * USA * * Redistribution and use in source and binary forms, with or without * modification, are permitted provided that the following conditions * are met: * * 1. Redistributions of source code must retain the above copyright * notice, this list of conditions and the following disclaimer. * 2. Redistributions in binary form must reproduce the above copyright * notice, this list of conditions and the following disclaimer in the * documentation and/or other materials provided with the distribution. * * THIS SOFTWARE IS PROVIDED BY MARSHALL KIRK MCKUSICK ``AS IS'' AND ANY * EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED * WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE * DISCLAIMED. IN NO EVENT SHALL MARSHALL KIRK MCKUSICK BE LIABLE FOR * ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF * SUCH DAMAGE. * * from: @(#)ffs_softdep.c 9.59 (McKusick) 6/21/00 * $FreeBSD: src/sys/ufs/ffs/ffs_softdep.c,v 1.57.2.11 2002/02/05 18:46:53 dillon Exp $ */ /* * For now we want the safety net that the DIAGNOSTIC and DEBUG flags provide. */ #ifndef DIAGNOSTIC #define DIAGNOSTIC #endif #ifndef DEBUG #define DEBUG #endif #include <sys/param.h> #include <sys/kernel.h> #include <sys/systm.h> #include <sys/buf.h> #include <sys/malloc.h> #include <sys/mount.h> #include <sys/proc.h> #include <sys/spinlock2.h> #include <sys/syslog.h> #include <sys/vnode.h> #include <sys/conf.h> #include <machine/inttypes.h> #include "dir.h" #include "quota.h" #include "inode.h" #include "ufsmount.h" #include "fs.h" #include "softdep.h" #include "ffs_extern.h" #include "ufs_extern.h" #include <sys/buf2.h> #include <sys/lock.h> /* * These definitions need to be adapted to the system to which * this file is being ported. */ /* * malloc types defined for the softdep system. */ MALLOC_DEFINE(M_PAGEDEP, "pagedep","File page dependencies"); MALLOC_DEFINE(M_INODEDEP, "inodedep","Inode dependencies"); MALLOC_DEFINE(M_NEWBLK, "newblk","New block allocation"); MALLOC_DEFINE(M_BMSAFEMAP, "bmsafemap","Block or frag allocated from cyl group map"); MALLOC_DEFINE(M_ALLOCDIRECT, "allocdirect","Block or frag dependency for an inode"); MALLOC_DEFINE(M_INDIRDEP, "indirdep","Indirect block dependencies"); MALLOC_DEFINE(M_ALLOCINDIR, "allocindir","Block dependency for an indirect block"); MALLOC_DEFINE(M_FREEFRAG, "freefrag","Previously used frag for an inode"); MALLOC_DEFINE(M_FREEBLKS, "freeblks","Blocks freed from an inode"); MALLOC_DEFINE(M_FREEFILE, "freefile","Inode deallocated"); MALLOC_DEFINE(M_DIRADD, "diradd","New directory entry"); MALLOC_DEFINE(M_MKDIR, "mkdir","New directory"); MALLOC_DEFINE(M_DIRREM, "dirrem","Directory entry deleted"); #define M_SOFTDEP_FLAGS (M_WAITOK | M_USE_RESERVE) #define D_PAGEDEP 0 #define D_INODEDEP 1 #define D_NEWBLK 2 #define D_BMSAFEMAP 3 #define D_ALLOCDIRECT 4 #define D_INDIRDEP 5 #define D_ALLOCINDIR 6 #define D_FREEFRAG 7 #define D_FREEBLKS 8 #define D_FREEFILE 9 #define D_DIRADD 10 #define D_MKDIR 11 #define D_DIRREM 12 #define D_LAST D_DIRREM /* * translate from workitem type to memory type * MUST match the defines above, such that memtype[D_XXX] == M_XXX */ static struct malloc_type *memtype[] = { M_PAGEDEP, M_INODEDEP, M_NEWBLK, M_BMSAFEMAP, M_ALLOCDIRECT, M_INDIRDEP, M_ALLOCINDIR, M_FREEFRAG, M_FREEBLKS, M_FREEFILE, M_DIRADD, M_MKDIR, M_DIRREM }; #define DtoM(type) (memtype[type]) /* * Names of malloc types. */ #define TYPENAME(type) \ ((unsigned)(type) < D_LAST ? memtype[type]->ks_shortdesc : "???") /* * End system adaptaion definitions. */ /* * Internal function prototypes. */ static void softdep_error(char *, int); static void drain_output(struct vnode *, int); static int getdirtybuf(struct buf **, int); static void clear_remove(struct thread *); static void clear_inodedeps(struct thread *); static int flush_pagedep_deps(struct vnode *, struct mount *, struct diraddhd *); static int flush_inodedep_deps(struct fs *, ino_t); static int handle_written_filepage(struct pagedep *, struct buf *); static void diradd_inode_written(struct diradd *, struct inodedep *); static int handle_written_inodeblock(struct inodedep *, struct buf *); static void handle_allocdirect_partdone(struct allocdirect *); static void handle_allocindir_partdone(struct allocindir *); static void initiate_write_filepage(struct pagedep *, struct buf *); static void handle_written_mkdir(struct mkdir *, int); static void initiate_write_inodeblock(struct inodedep *, struct buf *); static void handle_workitem_freefile(struct freefile *); static void handle_workitem_remove(struct dirrem *); static struct dirrem *newdirrem(struct buf *, struct inode *, struct inode *, int, struct dirrem **); static void free_diradd(struct diradd *); static void free_allocindir(struct allocindir *, struct inodedep *); static int indir_trunc (struct inode *, off_t, int, ufs_lbn_t, long *); static void deallocate_dependencies(struct buf *, struct inodedep *); static void free_allocdirect(struct allocdirectlst *, struct allocdirect *, int); static int check_inode_unwritten(struct inodedep *); static int free_inodedep(struct inodedep *); static void handle_workitem_freeblocks(struct freeblks *); static void merge_inode_lists(struct inodedep *); static void setup_allocindir_phase2(struct buf *, struct inode *, struct allocindir *); static struct allocindir *newallocindir(struct inode *, int, ufs_daddr_t, ufs_daddr_t); static void handle_workitem_freefrag(struct freefrag *); static struct freefrag *newfreefrag(struct inode *, ufs_daddr_t, long); static void allocdirect_merge(struct allocdirectlst *, struct allocdirect *, struct allocdirect *); static struct bmsafemap *bmsafemap_lookup(struct buf *); static int newblk_lookup(struct fs *, ufs_daddr_t, int, struct newblk **); static int inodedep_lookup(struct fs *, ino_t, int, struct inodedep **); static int pagedep_lookup(struct inode *, ufs_lbn_t, int, struct pagedep **); static int request_cleanup(int); static int process_worklist_item(struct mount *, int); static void add_to_worklist(struct worklist *); /* * Exported softdep operations. */ static void softdep_disk_io_initiation(struct buf *); static void softdep_disk_write_complete(struct buf *); static void softdep_deallocate_dependencies(struct buf *); static int softdep_fsync(struct vnode *); static int softdep_process_worklist(struct mount *); static void softdep_move_dependencies(struct buf *, struct buf *); static int softdep_count_dependencies(struct buf *bp, int); static int softdep_checkread(struct buf *bp); static int softdep_checkwrite(struct buf *bp); static struct bio_ops softdep_bioops = { .io_start = softdep_disk_io_initiation, .io_complete = softdep_disk_write_complete, .io_deallocate = softdep_deallocate_dependencies, .io_fsync = softdep_fsync, .io_sync = softdep_process_worklist, .io_movedeps = softdep_move_dependencies, .io_countdeps = softdep_count_dependencies, .io_checkread = softdep_checkread, .io_checkwrite = softdep_checkwrite }; /* * Locking primitives. */ static void acquire_lock(struct lock *); static void free_lock(struct lock *); #ifdef INVARIANTS static int lock_held(struct lock *); #endif static struct lock lk; #define ACQUIRE_LOCK(lkp) acquire_lock(lkp) #define FREE_LOCK(lkp) free_lock(lkp) static void acquire_lock(struct lock *lkp) { lockmgr(lkp, LK_EXCLUSIVE); } static void free_lock(struct lock *lkp) { lockmgr(lkp, LK_RELEASE); } #ifdef INVARIANTS static int lock_held(struct lock *lkp) { return lockinuse(lkp); } #endif /* * Place holder for real semaphores. */ struct sema { int value; thread_t holder; char *name; int timo; struct spinlock spin; }; static void sema_init(struct sema *, char *, int); static int sema_get(struct sema *, struct lock *); static void sema_release(struct sema *, struct lock *); #define NOHOLDER ((struct thread *) -1) static void sema_init(struct sema *semap, char *name, int timo) { semap->holder = NOHOLDER; semap->value = 0; semap->name = name; semap->timo = timo; spin_init(&semap->spin, "ufssema"); } /* * Obtain exclusive access, semaphore is protected by the interlock. * If interlock is NULL we must protect the semaphore ourselves. */ static int sema_get(struct sema *semap, struct lock *interlock) { int rv; if (interlock) { if (semap->value > 0) { ++semap->value; /* serves as wakeup flag */ lksleep(semap, interlock, 0, semap->name, semap->timo); rv = 0; } else { semap->value = 1; /* serves as owned flag */ semap->holder = curthread; rv = 1; } } else { spin_lock(&semap->spin); if (semap->value > 0) { ++semap->value; /* serves as wakeup flag */ ssleep(semap, &semap->spin, 0, semap->name, semap->timo); spin_unlock(&semap->spin); rv = 0; } else { semap->value = 1; /* serves as owned flag */ semap->holder = curthread; spin_unlock(&semap->spin); rv = 1; } } return (rv); } static void sema_release(struct sema *semap, struct lock *lk) { if (semap->value <= 0 || semap->holder != curthread) panic("sema_release: not held"); if (lk) { semap->holder = NOHOLDER; if (--semap->value > 0) { semap->value = 0; wakeup(semap); } } else { spin_lock(&semap->spin); semap->holder = NOHOLDER; if (--semap->value > 0) { semap->value = 0; spin_unlock(&semap->spin); wakeup(semap); } else { spin_unlock(&semap->spin); } } } /* * Worklist queue management. * These routines require that the lock be held. */ static void worklist_insert(struct workhead *, struct worklist *); static void worklist_remove(struct worklist *); static void workitem_free(struct worklist *, int); #define WORKLIST_INSERT_BP(bp, item) do { \ (bp)->b_ops = &softdep_bioops; \ worklist_insert(&(bp)->b_dep, item); \ } while (0) #define WORKLIST_INSERT(head, item) worklist_insert(head, item) #define WORKLIST_REMOVE(item) worklist_remove(item) #define WORKITEM_FREE(item, type) workitem_free((struct worklist *)item, type) static void worklist_insert(struct workhead *head, struct worklist *item) { KKASSERT(lock_held(&lk)); if (item->wk_state & ONWORKLIST) { panic("worklist_insert: already on list"); } item->wk_state |= ONWORKLIST; LIST_INSERT_HEAD(head, item, wk_list); } static void worklist_remove(struct worklist *item) { KKASSERT(lock_held(&lk)); if ((item->wk_state & ONWORKLIST) == 0) panic("worklist_remove: not on list"); item->wk_state &= ~ONWORKLIST; LIST_REMOVE(item, wk_list); } static void workitem_free(struct worklist *item, int type) { if (item->wk_state & ONWORKLIST) panic("workitem_free: still on list"); if (item->wk_type != type) panic("workitem_free: type mismatch"); kfree(item, DtoM(type)); } /* * Workitem queue management */ static struct workhead softdep_workitem_pending; static int num_on_worklist; /* number of worklist items to be processed */ static int softdep_worklist_busy; /* 1 => trying to do unmount */ static int softdep_worklist_req; /* serialized waiters */ static int max_softdeps; /* maximum number of structs before slowdown */ static int tickdelay = 2; /* number of ticks to pause during slowdown */ static int *stat_countp; /* statistic to count in proc_waiting timeout */ static int proc_waiting; /* tracks whether we have a timeout posted */ static struct thread *filesys_syncer; /* proc of filesystem syncer process */ static int req_clear_inodedeps; /* syncer process flush some inodedeps */ #define FLUSH_INODES 1 static int req_clear_remove; /* syncer process flush some freeblks */ #define FLUSH_REMOVE 2 /* * runtime statistics */ static int stat_worklist_push; /* number of worklist cleanups */ static int stat_blk_limit_push; /* number of times block limit neared */ static int stat_ino_limit_push; /* number of times inode limit neared */ static int stat_blk_limit_hit; /* number of times block slowdown imposed */ static int stat_ino_limit_hit; /* number of times inode slowdown imposed */ static int stat_sync_limit_hit; /* number of synchronous slowdowns imposed */ static int stat_indir_blk_ptrs; /* bufs redirtied as indir ptrs not written */ static int stat_inode_bitmap; /* bufs redirtied as inode bitmap not written */ static int stat_direct_blk_ptrs;/* bufs redirtied as direct ptrs not written */ static int stat_dir_entry; /* bufs redirtied as dir entry cannot write */ #ifdef DEBUG #include <vm/vm.h> #include <sys/sysctl.h> SYSCTL_INT(_debug, OID_AUTO, max_softdeps, CTLFLAG_RW, &max_softdeps, 0, "Maximum soft dependencies before slowdown occurs"); SYSCTL_INT(_debug, OID_AUTO, tickdelay, CTLFLAG_RW, &tickdelay, 0, "Ticks to delay before allocating during slowdown"); SYSCTL_INT(_debug, OID_AUTO, worklist_push, CTLFLAG_RW, &stat_worklist_push, 0, "Number of worklist cleanups"); SYSCTL_INT(_debug, OID_AUTO, blk_limit_push, CTLFLAG_RW, &stat_blk_limit_push, 0, "Number of times block limit neared"); SYSCTL_INT(_debug, OID_AUTO, ino_limit_push, CTLFLAG_RW, &stat_ino_limit_push, 0, "Number of times inode limit neared"); SYSCTL_INT(_debug, OID_AUTO, blk_limit_hit, CTLFLAG_RW, &stat_blk_limit_hit, 0, "Number of times block slowdown imposed"); SYSCTL_INT(_debug, OID_AUTO, ino_limit_hit, CTLFLAG_RW, &stat_ino_limit_hit, 0, "Number of times inode slowdown imposed "); SYSCTL_INT(_debug, OID_AUTO, sync_limit_hit, CTLFLAG_RW, &stat_sync_limit_hit, 0, "Number of synchronous slowdowns imposed"); SYSCTL_INT(_debug, OID_AUTO, indir_blk_ptrs, CTLFLAG_RW, &stat_indir_blk_ptrs, 0, "Bufs redirtied as indir ptrs not written"); SYSCTL_INT(_debug, OID_AUTO, inode_bitmap, CTLFLAG_RW, &stat_inode_bitmap, 0, "Bufs redirtied as inode bitmap not written"); SYSCTL_INT(_debug, OID_AUTO, direct_blk_ptrs, CTLFLAG_RW, &stat_direct_blk_ptrs, 0, "Bufs redirtied as direct ptrs not written"); SYSCTL_INT(_debug, OID_AUTO, dir_entry, CTLFLAG_RW, &stat_dir_entry, 0, "Bufs redirtied as dir entry cannot write"); #endif /* DEBUG */ /* * Add an item to the end of the work queue. * This routine requires that the lock be held. * This is the only routine that adds items to the list. * The following routine is the only one that removes items * and does so in order from first to last. */ static void add_to_worklist(struct worklist *wk) { static struct worklist *worklist_tail; if (wk->wk_state & ONWORKLIST) { panic("add_to_worklist: already on list"); } wk->wk_state |= ONWORKLIST; if (LIST_FIRST(&softdep_workitem_pending) == NULL) LIST_INSERT_HEAD(&softdep_workitem_pending, wk, wk_list); else LIST_INSERT_AFTER(worklist_tail, wk, wk_list); worklist_tail = wk; num_on_worklist += 1; } /* * Process that runs once per second to handle items in the background queue. * * Note that we ensure that everything is done in the order in which they * appear in the queue. The code below depends on this property to ensure * that blocks of a file are freed before the inode itself is freed. This * ordering ensures that no new <vfsid, inum, lbn> triples will be generated * until all the old ones have been purged from the dependency lists. * * bioops callback - hold io_token */ static int softdep_process_worklist(struct mount *matchmnt) { thread_t td = curthread; int matchcnt, loopcount; int starttime; ACQUIRE_LOCK(&lk); /* * Record the process identifier of our caller so that we can give * this process preferential treatment in request_cleanup below. */ filesys_syncer = td; matchcnt = 0; /* * There is no danger of having multiple processes run this * code, but we have to single-thread it when softdep_flushfiles() * is in operation to get an accurate count of the number of items * related to its mount point that are in the list. */ if (matchmnt == NULL) { if (softdep_worklist_busy < 0) { matchcnt = -1; goto done; } softdep_worklist_busy += 1; } /* * If requested, try removing inode or removal dependencies. */ if (req_clear_inodedeps) { clear_inodedeps(td); req_clear_inodedeps -= 1; wakeup_one(&proc_waiting); } if (req_clear_remove) { clear_remove(td); req_clear_remove -= 1; wakeup_one(&proc_waiting); } loopcount = 1; starttime = ticks; while (num_on_worklist > 0) { matchcnt += process_worklist_item(matchmnt, 0); /* * If a umount operation wants to run the worklist * accurately, abort. */ if (softdep_worklist_req && matchmnt == NULL) { matchcnt = -1; break; } /* * If requested, try removing inode or removal dependencies. */ if (req_clear_inodedeps) { clear_inodedeps(td); req_clear_inodedeps -= 1; wakeup_one(&proc_waiting); } if (req_clear_remove) { clear_remove(td); req_clear_remove -= 1; wakeup_one(&proc_waiting); } /* * We do not generally want to stop for buffer space, but if * we are really being a buffer hog, we will stop and wait. */ if (loopcount++ % 128 == 0) { FREE_LOCK(&lk); bwillinode(1); ACQUIRE_LOCK(&lk); } /* * Never allow processing to run for more than one * second. Otherwise the other syncer tasks may get * excessively backlogged. * * Use ticks to avoid boundary condition w/time_second or * time_uptime. */ if ((ticks - starttime) > hz && matchmnt == NULL) { matchcnt = -1; break; } } if (matchmnt == NULL) { --softdep_worklist_busy; if (softdep_worklist_req && softdep_worklist_busy == 0) wakeup(&softdep_worklist_req); } done: FREE_LOCK(&lk); return (matchcnt); } /* * Process one item on the worklist. */ static int process_worklist_item(struct mount *matchmnt, int flags) { struct ufsmount *ump; struct worklist *wk; struct dirrem *dirrem; struct fs *matchfs; struct vnode *vp; int matchcnt = 0; KKASSERT(lock_held(&lk)); matchfs = NULL; if (matchmnt != NULL) matchfs = VFSTOUFS(matchmnt)->um_fs; /* * Normally we just process each item on the worklist in order. * However, if we are in a situation where we cannot lock any * inodes, we have to skip over any dirrem requests whose * vnodes are resident and locked. */ LIST_FOREACH(wk, &softdep_workitem_pending, wk_list) { if ((flags & LK_NOWAIT) == 0 || wk->wk_type != D_DIRREM) break; dirrem = WK_DIRREM(wk); ump = VFSTOUFS(dirrem->dm_mnt); lwkt_gettoken(&ump->um_mountp->mnt_token); vp = ufs_ihashlookup(ump, ump->um_dev, dirrem->dm_oldinum); lwkt_reltoken(&ump->um_mountp->mnt_token); if (vp == NULL || !vn_islocked(vp)) break; } if (wk == NULL) { return (0); } WORKLIST_REMOVE(wk); num_on_worklist -= 1; FREE_LOCK(&lk); switch (wk->wk_type) { case D_DIRREM: /* removal of a directory entry */ if (WK_DIRREM(wk)->dm_mnt == matchmnt) matchcnt += 1; handle_workitem_remove(WK_DIRREM(wk)); break; case D_FREEBLKS: /* releasing blocks and/or fragments from a file */ if (WK_FREEBLKS(wk)->fb_fs == matchfs) matchcnt += 1; handle_workitem_freeblocks(WK_FREEBLKS(wk)); break; case D_FREEFRAG: /* releasing a fragment when replaced as a file grows */ if (WK_FREEFRAG(wk)->ff_fs == matchfs) matchcnt += 1; handle_workitem_freefrag(WK_FREEFRAG(wk)); break; case D_FREEFILE: /* releasing an inode when its link count drops to 0 */ if (WK_FREEFILE(wk)->fx_fs == matchfs) matchcnt += 1; handle_workitem_freefile(WK_FREEFILE(wk)); break; default: panic("%s_process_worklist: Unknown type %s", "softdep", TYPENAME(wk->wk_type)); /* NOTREACHED */ } ACQUIRE_LOCK(&lk); return (matchcnt); } /* * Move dependencies from one buffer to another. * * bioops callback - hold io_token */ static void softdep_move_dependencies(struct buf *oldbp, struct buf *newbp) { struct worklist *wk, *wktail; if (LIST_FIRST(&newbp->b_dep) != NULL) panic("softdep_move_dependencies: need merge code"); wktail = NULL; ACQUIRE_LOCK(&lk); while ((wk = LIST_FIRST(&oldbp->b_dep)) != NULL) { LIST_REMOVE(wk, wk_list); if (wktail == NULL) LIST_INSERT_HEAD(&newbp->b_dep, wk, wk_list); else LIST_INSERT_AFTER(wktail, wk, wk_list); wktail = wk; newbp->b_ops = &softdep_bioops; } FREE_LOCK(&lk); } /* * Purge the work list of all items associated with a particular mount point. */ int softdep_flushfiles(struct mount *oldmnt, int flags) { struct vnode *devvp; int error, loopcnt; /* * Await our turn to clear out the queue, then serialize access. */ ACQUIRE_LOCK(&lk); while (softdep_worklist_busy != 0) { softdep_worklist_req += 1; lksleep(&softdep_worklist_req, &lk, 0, "softflush", 0); softdep_worklist_req -= 1; } softdep_worklist_busy = -1; FREE_LOCK(&lk); if ((error = ffs_flushfiles(oldmnt, flags)) != 0) { softdep_worklist_busy = 0; if (softdep_worklist_req) wakeup(&softdep_worklist_req); return (error); } /* * Alternately flush the block device associated with the mount * point and process any dependencies that the flushing * creates. In theory, this loop can happen at most twice, * but we give it a few extra just to be sure. */ devvp = VFSTOUFS(oldmnt)->um_devvp; for (loopcnt = 10; loopcnt > 0; ) { if (softdep_process_worklist(oldmnt) == 0) { loopcnt--; /* * Do another flush in case any vnodes were brought in * as part of the cleanup operations. */ if ((error = ffs_flushfiles(oldmnt, flags)) != 0) break; /* * If we still found nothing to do, we are really done. */ if (softdep_process_worklist(oldmnt) == 0) break; } vn_lock(devvp, LK_EXCLUSIVE | LK_RETRY); error = VOP_FSYNC(devvp, MNT_WAIT, 0); vn_unlock(devvp); if (error) break; } ACQUIRE_LOCK(&lk); softdep_worklist_busy = 0; if (softdep_worklist_req) wakeup(&softdep_worklist_req); FREE_LOCK(&lk); /* * If we are unmounting then it is an error to fail. If we * are simply trying to downgrade to read-only, then filesystem * activity can keep us busy forever, so we just fail with EBUSY. */ if (loopcnt == 0) { if (oldmnt->mnt_kern_flag & MNTK_UNMOUNT) panic("softdep_flushfiles: looping"); error = EBUSY; } return (error); } /* * Structure hashing. * * There are three types of structures that can be looked up: * 1) pagedep structures identified by mount point, inode number, * and logical block. * 2) inodedep structures identified by mount point and inode number. * 3) newblk structures identified by mount point and * physical block number. * * The "pagedep" and "inodedep" dependency structures are hashed * separately from the file blocks and inodes to which they correspond. * This separation helps when the in-memory copy of an inode or * file block must be replaced. It also obviates the need to access * an inode or file page when simply updating (or de-allocating) * dependency structures. Lookup of newblk structures is needed to * find newly allocated blocks when trying to associate them with * their allocdirect or allocindir structure. * * The lookup routines optionally create and hash a new instance when * an existing entry is not found. */ #define DEPALLOC 0x0001 /* allocate structure if lookup fails */ #define NODELAY 0x0002 /* cannot do background work */ /* * Structures and routines associated with pagedep caching. */ LIST_HEAD(pagedep_hashhead, pagedep) *pagedep_hashtbl; u_long pagedep_hash; /* size of hash table - 1 */ #define PAGEDEP_HASH(mp, inum, lbn) \ (&pagedep_hashtbl[((((register_t)(mp)) >> 13) + (inum) + (lbn)) & \ pagedep_hash]) static struct sema pagedep_in_progress; /* * Helper routine for pagedep_lookup() */ static __inline struct pagedep * pagedep_find(struct pagedep_hashhead *pagedephd, ino_t ino, ufs_lbn_t lbn, struct mount *mp) { struct pagedep *pagedep; LIST_FOREACH(pagedep, pagedephd, pd_hash) { if (ino == pagedep->pd_ino && lbn == pagedep->pd_lbn && mp == pagedep->pd_mnt) { return (pagedep); } } return(NULL); } /* * Look up a pagedep. Return 1 if found, 0 if not found. * If not found, allocate if DEPALLOC flag is passed. * Found or allocated entry is returned in pagedeppp. * This routine must be called with splbio interrupts blocked. */ static int pagedep_lookup(struct inode *ip, ufs_lbn_t lbn, int flags, struct pagedep **pagedeppp) { struct pagedep *pagedep; struct pagedep_hashhead *pagedephd; struct mount *mp; int i; KKASSERT(lock_held(&lk)); mp = ITOV(ip)->v_mount; pagedephd = PAGEDEP_HASH(mp, ip->i_number, lbn); top: *pagedeppp = pagedep_find(pagedephd, ip->i_number, lbn, mp); if (*pagedeppp) return(1); if ((flags & DEPALLOC) == 0) return (0); if (sema_get(&pagedep_in_progress, &lk) == 0) goto top; FREE_LOCK(&lk); pagedep = kmalloc(sizeof(struct pagedep), M_PAGEDEP, M_SOFTDEP_FLAGS | M_ZERO); ACQUIRE_LOCK(&lk); if (pagedep_find(pagedephd, ip->i_number, lbn, mp)) { kprintf("pagedep_lookup: blocking race avoided\n"); sema_release(&pagedep_in_progress, &lk); kfree(pagedep, M_PAGEDEP); goto top; } pagedep->pd_list.wk_type = D_PAGEDEP; pagedep->pd_mnt = mp; pagedep->pd_ino = ip->i_number; pagedep->pd_lbn = lbn; LIST_INIT(&pagedep->pd_dirremhd); LIST_INIT(&pagedep->pd_pendinghd); for (i = 0; i < DAHASHSZ; i++) LIST_INIT(&pagedep->pd_diraddhd[i]); LIST_INSERT_HEAD(pagedephd, pagedep, pd_hash); sema_release(&pagedep_in_progress, &lk); *pagedeppp = pagedep; return (0); } /* * Structures and routines associated with inodedep caching. */ LIST_HEAD(inodedep_hashhead, inodedep) *inodedep_hashtbl; static u_long inodedep_hash; /* size of hash table - 1 */ static long num_inodedep; /* number of inodedep allocated */ #define INODEDEP_HASH(fs, inum) \ (&inodedep_hashtbl[((((register_t)(fs)) >> 13) + (inum)) & inodedep_hash]) static struct sema inodedep_in_progress; /* * Helper routine for inodedep_lookup() */ static __inline struct inodedep * inodedep_find(struct inodedep_hashhead *inodedephd, struct fs *fs, ino_t inum) { struct inodedep *inodedep; LIST_FOREACH(inodedep, inodedephd, id_hash) { if (inum == inodedep->id_ino && fs == inodedep->id_fs) return(inodedep); } return (NULL); } /* * Look up a inodedep. Return 1 if found, 0 if not found. * If not found, allocate if DEPALLOC flag is passed. * Found or allocated entry is returned in inodedeppp. * This routine must be called with splbio interrupts blocked. */ static int inodedep_lookup(struct fs *fs, ino_t inum, int flags, struct inodedep **inodedeppp) { struct inodedep *inodedep; struct inodedep_hashhead *inodedephd; KKASSERT(lock_held(&lk)); inodedephd = INODEDEP_HASH(fs, inum); top: *inodedeppp = inodedep_find(inodedephd, fs, inum); if (*inodedeppp) return (1); if ((flags & DEPALLOC) == 0) return (0); /* * If we are over our limit, try to improve the situation. */ if (num_inodedep > max_softdeps / 2) speedup_syncer(NULL); if (num_inodedep > max_softdeps && (flags & NODELAY) == 0 && request_cleanup(FLUSH_INODES)) { goto top; } if (sema_get(&inodedep_in_progress, &lk) == 0) goto top; FREE_LOCK(&lk); inodedep = kmalloc(sizeof(struct inodedep), M_INODEDEP, M_SOFTDEP_FLAGS | M_ZERO); ACQUIRE_LOCK(&lk); if (inodedep_find(inodedephd, fs, inum)) { kprintf("inodedep_lookup: blocking race avoided\n"); sema_release(&inodedep_in_progress, &lk); kfree(inodedep, M_INODEDEP); goto top; } inodedep->id_list.wk_type = D_INODEDEP; inodedep->id_fs = fs; inodedep->id_ino = inum; inodedep->id_state = ALLCOMPLETE; inodedep->id_nlinkdelta = 0; inodedep->id_savedino = NULL; inodedep->id_savedsize = -1; inodedep->id_buf = NULL; LIST_INIT(&inodedep->id_pendinghd); LIST_INIT(&inodedep->id_inowait); LIST_INIT(&inodedep->id_bufwait); TAILQ_INIT(&inodedep->id_inoupdt); TAILQ_INIT(&inodedep->id_newinoupdt); num_inodedep += 1; LIST_INSERT_HEAD(inodedephd, inodedep, id_hash); sema_release(&inodedep_in_progress, &lk); *inodedeppp = inodedep; return (0); } /* * Structures and routines associated with newblk caching. */ LIST_HEAD(newblk_hashhead, newblk) *newblk_hashtbl; u_long newblk_hash; /* size of hash table - 1 */ #define NEWBLK_HASH(fs, inum) \ (&newblk_hashtbl[((((register_t)(fs)) >> 13) + (inum)) & newblk_hash]) static struct sema newblk_in_progress; /* * Helper routine for newblk_lookup() */ static __inline struct newblk * newblk_find(struct newblk_hashhead *newblkhd, struct fs *fs, ufs_daddr_t newblkno) { struct newblk *newblk; LIST_FOREACH(newblk, newblkhd, nb_hash) { if (newblkno == newblk->nb_newblkno && fs == newblk->nb_fs) return (newblk); } return(NULL); } /* * Look up a newblk. Return 1 if found, 0 if not found. * If not found, allocate if DEPALLOC flag is passed. * Found or allocated entry is returned in newblkpp. */ static int newblk_lookup(struct fs *fs, ufs_daddr_t newblkno, int flags, struct newblk **newblkpp) { struct newblk *newblk; struct newblk_hashhead *newblkhd; newblkhd = NEWBLK_HASH(fs, newblkno); top: *newblkpp = newblk_find(newblkhd, fs, newblkno); if (*newblkpp) return(1); if ((flags & DEPALLOC) == 0) return (0); if (sema_get(&newblk_in_progress, NULL) == 0) goto top; newblk = kmalloc(sizeof(struct newblk), M_NEWBLK, M_SOFTDEP_FLAGS | M_ZERO); if (newblk_find(newblkhd, fs, newblkno)) { kprintf("newblk_lookup: blocking race avoided\n"); sema_release(&pagedep_in_progress, NULL); kfree(newblk, M_NEWBLK); goto top; } newblk->nb_state = 0; newblk->nb_fs = fs; newblk->nb_newblkno = newblkno; LIST_INSERT_HEAD(newblkhd, newblk, nb_hash); sema_release(&newblk_in_progress, NULL); *newblkpp = newblk; return (0); } /* * Executed during filesystem system initialization before * mounting any filesystems. */ void softdep_initialize(void) { size_t idsize = sizeof(struct inodedep); int hsize = vfs_inodehashsize(); LIST_INIT(&mkdirlisthd); LIST_INIT(&softdep_workitem_pending); max_softdeps = min(maxvnodes * 8, M_INODEDEP->ks_limit / (2 * idsize)); /* * Cap it at 100,000, having more just gets kinda silly. */ max_softdeps = min(max_softdeps, 100000); pagedep_hashtbl = hashinit(hsize / 4, M_PAGEDEP, &pagedep_hash); lockinit(&lk, "ffs_softdep", 0, LK_CANRECURSE); sema_init(&pagedep_in_progress, "pagedep", 0); inodedep_hashtbl = hashinit(hsize, M_INODEDEP, &inodedep_hash); sema_init(&inodedep_in_progress, "inodedep", 0); newblk_hashtbl = hashinit(64, M_NEWBLK, &newblk_hash); sema_init(&newblk_in_progress, "newblk", 0); add_bio_ops(&softdep_bioops); } /* * Called at mount time to notify the dependency code that a * filesystem wishes to use it. */ int softdep_mount(struct vnode *devvp, struct mount *mp, struct fs *fs) { struct csum cstotal; struct cg *cgp; struct buf *bp; int error, cyl; mp->mnt_flag &= ~MNT_ASYNC; mp->mnt_flag |= MNT_SOFTDEP; mp->mnt_bioops = &softdep_bioops; /* * When doing soft updates, the counters in the * superblock may have gotten out of sync, so we have * to scan the cylinder groups and recalculate them. */ if (fs->fs_clean != 0) return (0); bzero(&cstotal, sizeof cstotal); for (cyl = 0; cyl < fs->fs_ncg; cyl++) { if ((error = bread(devvp, fsbtodoff(fs, cgtod(fs, cyl)), fs->fs_cgsize, &bp)) != 0) { brelse(bp); return (error); } cgp = (struct cg *)bp->b_data; cstotal.cs_nffree += cgp->cg_cs.cs_nffree; cstotal.cs_nbfree += cgp->cg_cs.cs_nbfree; cstotal.cs_nifree += cgp->cg_cs.cs_nifree; cstotal.cs_ndir += cgp->cg_cs.cs_ndir; fs->fs_cs(fs, cyl) = cgp->cg_cs; brelse(bp); } #ifdef DEBUG if (bcmp(&cstotal, &fs->fs_cstotal, sizeof cstotal)) kprintf("ffs_mountfs: superblock updated for soft updates\n"); #endif bcopy(&cstotal, &fs->fs_cstotal, sizeof cstotal); return (0); } /* * Protecting the freemaps (or bitmaps). * * To eliminate the need to execute fsck before mounting a filesystem * after a power failure, one must (conservatively) guarantee that the * on-disk copy of the bitmaps never indicate that a live inode or block is * free. So, when a block or inode is allocated, the bitmap should be * updated (on disk) before any new pointers. When a block or inode is * freed, the bitmap should not be updated until all pointers have been * reset. The latter dependency is handled by the delayed de-allocation * approach described below for block and inode de-allocation. The former * dependency is handled by calling the following procedure when a block or * inode is allocated. When an inode is allocated an "inodedep" is created * with its DEPCOMPLETE flag cleared until its bitmap is written to disk. * Each "inodedep" is also inserted into the hash indexing structure so * that any additional link additions can be made dependent on the inode * allocation. * * The ufs filesystem maintains a number of free block counts (e.g., per * cylinder group, per cylinder and per <cylinder, rotational position> pair) * in addition to the bitmaps. These counts are used to improve efficiency * during allocation and therefore must be consistent with the bitmaps. * There is no convenient way to guarantee post-crash consistency of these * counts with simple update ordering, for two main reasons: (1) The counts * and bitmaps for a single cylinder group block are not in the same disk * sector. If a disk write is interrupted (e.g., by power failure), one may * be written and the other not. (2) Some of the counts are located in the * superblock rather than the cylinder group block. So, we focus our soft * updates implementation on protecting the bitmaps. When mounting a * filesystem, we recompute the auxiliary counts from the bitmaps. */ /* * Called just after updating the cylinder group block to allocate an inode. * * Parameters: * bp: buffer for cylgroup block with inode map * ip: inode related to allocation * newinum: new inode number being allocated */ void softdep_setup_inomapdep(struct buf *bp, struct inode *ip, ino_t newinum) { struct inodedep *inodedep; struct bmsafemap *bmsafemap; /* * Create a dependency for the newly allocated inode. * Panic if it already exists as something is seriously wrong. * Otherwise add it to the dependency list for the buffer holding * the cylinder group map from which it was allocated. */ ACQUIRE_LOCK(&lk); if ((inodedep_lookup(ip->i_fs, newinum, DEPALLOC|NODELAY, &inodedep))) { panic("softdep_setup_inomapdep: found inode"); } inodedep->id_buf = bp; inodedep->id_state &= ~DEPCOMPLETE; bmsafemap = bmsafemap_lookup(bp); LIST_INSERT_HEAD(&bmsafemap->sm_inodedephd, inodedep, id_deps); FREE_LOCK(&lk); } /* * Called just after updating the cylinder group block to * allocate block or fragment. * * Parameters: * bp: buffer for cylgroup block with block map * fs: filesystem doing allocation * newblkno: number of newly allocated block */ void softdep_setup_blkmapdep(struct buf *bp, struct fs *fs, ufs_daddr_t newblkno) { struct newblk *newblk; struct bmsafemap *bmsafemap; /* * Create a dependency for the newly allocated block. * Add it to the dependency list for the buffer holding * the cylinder group map from which it was allocated. */ if (newblk_lookup(fs, newblkno, DEPALLOC, &newblk) != 0) panic("softdep_setup_blkmapdep: found block"); ACQUIRE_LOCK(&lk); newblk->nb_bmsafemap = bmsafemap = bmsafemap_lookup(bp); LIST_INSERT_HEAD(&bmsafemap->sm_newblkhd, newblk, nb_deps); FREE_LOCK(&lk); } /* * Find the bmsafemap associated with a cylinder group buffer. * If none exists, create one. The buffer must be locked when * this routine is called and this routine must be called with * splbio interrupts blocked. */ static struct bmsafemap * bmsafemap_lookup(struct buf *bp) { struct bmsafemap *bmsafemap; struct worklist *wk; KKASSERT(lock_held(&lk)); LIST_FOREACH(wk, &bp->b_dep, wk_list) { if (wk->wk_type == D_BMSAFEMAP) return (WK_BMSAFEMAP(wk)); } FREE_LOCK(&lk); bmsafemap = kmalloc(sizeof(struct bmsafemap), M_BMSAFEMAP, M_SOFTDEP_FLAGS); bmsafemap->sm_list.wk_type = D_BMSAFEMAP; bmsafemap->sm_list.wk_state = 0; bmsafemap->sm_buf = bp; LIST_INIT(&bmsafemap->sm_allocdirecthd); LIST_INIT(&bmsafemap->sm_allocindirhd); LIST_INIT(&bmsafemap->sm_inodedephd); LIST_INIT(&bmsafemap->sm_newblkhd); ACQUIRE_LOCK(&lk); WORKLIST_INSERT_BP(bp, &bmsafemap->sm_list); return (bmsafemap); } /* * Direct block allocation dependencies. * * When a new block is allocated, the corresponding disk locations must be * initialized (with zeros or new data) before the on-disk inode points to * them. Also, the freemap from which the block was allocated must be * updated (on disk) before the inode's pointer. These two dependencies are * independent of each other and are needed for all file blocks and indirect * blocks that are pointed to directly by the inode. Just before the * "in-core" version of the inode is updated with a newly allocated block * number, a procedure (below) is called to setup allocation dependency * structures. These structures are removed when the corresponding * dependencies are satisfied or when the block allocation becomes obsolete * (i.e., the file is deleted, the block is de-allocated, or the block is a * fragment that gets upgraded). All of these cases are handled in * procedures described later. * * When a file extension causes a fragment to be upgraded, either to a larger * fragment or to a full block, the on-disk location may change (if the * previous fragment could not simply be extended). In this case, the old * fragment must be de-allocated, but not until after the inode's pointer has * been updated. In most cases, this is handled by later procedures, which * will construct a "freefrag" structure to be added to the workitem queue * when the inode update is complete (or obsolete). The main exception to * this is when an allocation occurs while a pending allocation dependency * (for the same block pointer) remains. This case is handled in the main * allocation dependency setup procedure by immediately freeing the * unreferenced fragments. * * Parameters: * ip: inode to which block is being added * lbn: block pointer within inode * newblkno: disk block number being added * oldblkno: previous block number, 0 unless frag * newsize: size of new block * oldsize: size of new block * bp: bp for allocated block */ void softdep_setup_allocdirect(struct inode *ip, ufs_lbn_t lbn, ufs_daddr_t newblkno, ufs_daddr_t oldblkno, long newsize, long oldsize, struct buf *bp) { struct allocdirect *adp, *oldadp; struct allocdirectlst *adphead; struct bmsafemap *bmsafemap; struct inodedep *inodedep; struct pagedep *pagedep; struct newblk *newblk; adp = kmalloc(sizeof(struct allocdirect), M_ALLOCDIRECT, M_SOFTDEP_FLAGS | M_ZERO); adp->ad_list.wk_type = D_ALLOCDIRECT; adp->ad_lbn = lbn; adp->ad_newblkno = newblkno; adp->ad_oldblkno = oldblkno; adp->ad_newsize = newsize; adp->ad_oldsize = oldsize; adp->ad_state = ATTACHED; if (newblkno == oldblkno) adp->ad_freefrag = NULL; else adp->ad_freefrag = newfreefrag(ip, oldblkno, oldsize); if (newblk_lookup(ip->i_fs, newblkno, 0, &newblk) == 0) panic("softdep_setup_allocdirect: lost block"); ACQUIRE_LOCK(&lk); inodedep_lookup(ip->i_fs, ip->i_number, DEPALLOC | NODELAY, &inodedep); adp->ad_inodedep = inodedep; if (newblk->nb_state == DEPCOMPLETE) { adp->ad_state |= DEPCOMPLETE; adp->ad_buf = NULL; } else { bmsafemap = newblk->nb_bmsafemap; adp->ad_buf = bmsafemap->sm_buf; LIST_REMOVE(newblk, nb_deps); LIST_INSERT_HEAD(&bmsafemap->sm_allocdirecthd, adp, ad_deps); } LIST_REMOVE(newblk, nb_hash); kfree(newblk, M_NEWBLK); WORKLIST_INSERT_BP(bp, &adp->ad_list); if (lbn >= UFS_NDADDR) { /* allocating an indirect block */ if (oldblkno != 0) { panic("softdep_setup_allocdirect: non-zero indir"); } } else { /* * Allocating a direct block. * * If we are allocating a directory block, then we must * allocate an associated pagedep to track additions and * deletions. */ if ((ip->i_mode & IFMT) == IFDIR && pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0) { WORKLIST_INSERT_BP(bp, &pagedep->pd_list); } } /* * The list of allocdirects must be kept in sorted and ascending * order so that the rollback routines can quickly determine the * first uncommitted block (the size of the file stored on disk * ends at the end of the lowest committed fragment, or if there * are no fragments, at the end of the highest committed block). * Since files generally grow, the typical case is that the new * block is to be added at the end of the list. We speed this * special case by checking against the last allocdirect in the * list before laboriously traversing the list looking for the * insertion point. */ adphead = &inodedep->id_newinoupdt; oldadp = TAILQ_LAST(adphead, allocdirectlst); if (oldadp == NULL || oldadp->ad_lbn <= lbn) { /* insert at end of list */ TAILQ_INSERT_TAIL(adphead, adp, ad_next); if (oldadp != NULL && oldadp->ad_lbn == lbn) allocdirect_merge(adphead, adp, oldadp); FREE_LOCK(&lk); return; } TAILQ_FOREACH(oldadp, adphead, ad_next) { if (oldadp->ad_lbn >= lbn) break; } if (oldadp == NULL) { panic("softdep_setup_allocdirect: lost entry"); } /* insert in middle of list */ TAILQ_INSERT_BEFORE(oldadp, adp, ad_next); if (oldadp->ad_lbn == lbn) allocdirect_merge(adphead, adp, oldadp); FREE_LOCK(&lk); } /* * Replace an old allocdirect dependency with a newer one. * This routine must be called with splbio interrupts blocked. * * Parameters: * adphead: head of list holding allocdirects * newadp: allocdirect being added * oldadp: existing allocdirect being checked */ static void allocdirect_merge(struct allocdirectlst *adphead, struct allocdirect *newadp, struct allocdirect *oldadp) { struct freefrag *freefrag; KKASSERT(lock_held(&lk)); if (newadp->ad_oldblkno != oldadp->ad_newblkno || newadp->ad_oldsize != oldadp->ad_newsize || newadp->ad_lbn >= UFS_NDADDR) { panic("allocdirect_check: old %d != new %d || lbn %ld >= %d", newadp->ad_oldblkno, oldadp->ad_newblkno, newadp->ad_lbn, UFS_NDADDR); } newadp->ad_oldblkno = oldadp->ad_oldblkno; newadp->ad_oldsize = oldadp->ad_oldsize; /* * If the old dependency had a fragment to free or had never * previously had a block allocated, then the new dependency * can immediately post its freefrag and adopt the old freefrag. * This action is done by swapping the freefrag dependencies. * The new dependency gains the old one's freefrag, and the * old one gets the new one and then immediately puts it on * the worklist when it is freed by free_allocdirect. It is * not possible to do this swap when the old dependency had a * non-zero size but no previous fragment to free. This condition * arises when the new block is an extension of the old block. * Here, the first part of the fragment allocated to the new * dependency is part of the block currently claimed on disk by * the old dependency, so cannot legitimately be freed until the * conditions for the new dependency are fulfilled. */ if (oldadp->ad_freefrag != NULL || oldadp->ad_oldblkno == 0) { freefrag = newadp->ad_freefrag; newadp->ad_freefrag = oldadp->ad_freefrag; oldadp->ad_freefrag = freefrag; } free_allocdirect(adphead, oldadp, 0); } /* * Allocate a new freefrag structure if needed. */ static struct freefrag * newfreefrag(struct inode *ip, ufs_daddr_t blkno, long size) { struct freefrag *freefrag; struct fs *fs; if (blkno == 0) return (NULL); fs = ip->i_fs; if (fragnum(fs, blkno) + numfrags(fs, size) > fs->fs_frag) panic("newfreefrag: frag size"); freefrag = kmalloc(sizeof(struct freefrag), M_FREEFRAG, M_SOFTDEP_FLAGS); freefrag->ff_list.wk_type = D_FREEFRAG; freefrag->ff_state = ip->i_uid & ~ONWORKLIST; /* XXX - used below */ freefrag->ff_inum = ip->i_number; freefrag->ff_fs = fs; freefrag->ff_devvp = ip->i_devvp; freefrag->ff_blkno = blkno; freefrag->ff_fragsize = size; return (freefrag); } /* * This workitem de-allocates fragments that were replaced during * file block allocation. */ static void handle_workitem_freefrag(struct freefrag *freefrag) { struct inode tip; tip.i_fs = freefrag->ff_fs; tip.i_devvp = freefrag->ff_devvp; tip.i_dev = freefrag->ff_devvp->v_rdev; tip.i_number = freefrag->ff_inum; tip.i_uid = freefrag->ff_state & ~ONWORKLIST; /* XXX - set above */ ffs_blkfree(&tip, freefrag->ff_blkno, freefrag->ff_fragsize); kfree(freefrag, M_FREEFRAG); } /* * Indirect block allocation dependencies. * * The same dependencies that exist for a direct block also exist when * a new block is allocated and pointed to by an entry in a block of * indirect pointers. The undo/redo states described above are also * used here. Because an indirect block contains many pointers that * may have dependencies, a second copy of the entire in-memory indirect * block is kept. The buffer cache copy is always completely up-to-date. * The second copy, which is used only as a source for disk writes, * contains only the safe pointers (i.e., those that have no remaining * update dependencies). The second copy is freed when all pointers * are safe. The cache is not allowed to replace indirect blocks with * pending update dependencies. If a buffer containing an indirect * block with dependencies is written, these routines will mark it * dirty again. It can only be successfully written once all the * dependencies are removed. The ffs_fsync routine in conjunction with * softdep_sync_metadata work together to get all the dependencies * removed so that a file can be successfully written to disk. Three * procedures are used when setting up indirect block pointer * dependencies. The division is necessary because of the organization * of the "balloc" routine and because of the distinction between file * pages and file metadata blocks. */ /* * Allocate a new allocindir structure. * * Parameters: * ip: inode for file being extended * ptrno: offset of pointer in indirect block * newblkno: disk block number being added * oldblkno: previous block number, 0 if none */ static struct allocindir * newallocindir(struct inode *ip, int ptrno, ufs_daddr_t newblkno, ufs_daddr_t oldblkno) { struct allocindir *aip; aip = kmalloc(sizeof(struct allocindir), M_ALLOCINDIR, M_SOFTDEP_FLAGS | M_ZERO); aip->ai_list.wk_type = D_ALLOCINDIR; aip->ai_state = ATTACHED; aip->ai_offset = ptrno; aip->ai_newblkno = newblkno; aip->ai_oldblkno = oldblkno; aip->ai_freefrag = newfreefrag(ip, oldblkno, ip->i_fs->fs_bsize); return (aip); } /* * Called just before setting an indirect block pointer * to a newly allocated file page. * * Parameters: * ip: inode for file being extended * lbn: allocated block number within file * bp: buffer with indirect blk referencing page * ptrno: offset of pointer in indirect block * newblkno: disk block number being added * oldblkno: previous block number, 0 if none * nbp: buffer holding allocated page */ void softdep_setup_allocindir_page(struct inode *ip, ufs_lbn_t lbn, struct buf *bp, int ptrno, ufs_daddr_t newblkno, ufs_daddr_t oldblkno, struct buf *nbp) { struct allocindir *aip; struct pagedep *pagedep; aip = newallocindir(ip, ptrno, newblkno, oldblkno); ACQUIRE_LOCK(&lk); /* * If we are allocating a directory page, then we must * allocate an associated pagedep to track additions and * deletions. */ if ((ip->i_mode & IFMT) == IFDIR && pagedep_lookup(ip, lbn, DEPALLOC, &pagedep) == 0) WORKLIST_INSERT_BP(nbp, &pagedep->pd_list); WORKLIST_INSERT_BP(nbp, &aip->ai_list); FREE_LOCK(&lk); setup_allocindir_phase2(bp, ip, aip); } /* * Called just before setting an indirect block pointer to a * newly allocated indirect block. * Parameters: * nbp: newly allocated indirect block * ip: inode for file being extended * bp: indirect block referencing allocated block * ptrno: offset of pointer in indirect block * newblkno: disk block number being added */ void softdep_setup_allocindir_meta(struct buf *nbp, struct inode *ip, struct buf *bp, int ptrno, ufs_daddr_t newblkno) { struct allocindir *aip; aip = newallocindir(ip, ptrno, newblkno, 0); ACQUIRE_LOCK(&lk); WORKLIST_INSERT_BP(nbp, &aip->ai_list); FREE_LOCK(&lk); setup_allocindir_phase2(bp, ip, aip); } /* * Called to finish the allocation of the "aip" allocated * by one of the two routines above. * * Parameters: * bp: in-memory copy of the indirect block * ip: inode for file being extended * aip: allocindir allocated by the above routines */ static void setup_allocindir_phase2(struct buf *bp, struct inode *ip, struct allocindir *aip) { struct worklist *wk; struct indirdep *indirdep, *newindirdep; struct bmsafemap *bmsafemap; struct allocindir *oldaip; struct freefrag *freefrag; struct newblk *newblk; if (bp->b_loffset >= 0) panic("setup_allocindir_phase2: not indir blk"); for (indirdep = NULL, newindirdep = NULL; ; ) { ACQUIRE_LOCK(&lk); LIST_FOREACH(wk, &bp->b_dep, wk_list) { if (wk->wk_type != D_INDIRDEP) continue; indirdep = WK_INDIRDEP(wk); break; } if (indirdep == NULL && newindirdep) { indirdep = newindirdep; WORKLIST_INSERT_BP(bp, &indirdep->ir_list); newindirdep = NULL; } FREE_LOCK(&lk); if (indirdep) { if (newblk_lookup(ip->i_fs, aip->ai_newblkno, 0, &newblk) == 0) panic("setup_allocindir: lost block"); ACQUIRE_LOCK(&lk); if (newblk->nb_state == DEPCOMPLETE) { aip->ai_state |= DEPCOMPLETE; aip->ai_buf = NULL; } else { bmsafemap = newblk->nb_bmsafemap; aip->ai_buf = bmsafemap->sm_buf; LIST_REMOVE(newblk, nb_deps); LIST_INSERT_HEAD(&bmsafemap->sm_allocindirhd, aip, ai_deps); } LIST_REMOVE(newblk, nb_hash); kfree(newblk, M_NEWBLK); aip->ai_indirdep = indirdep; /* * Check to see if there is an existing dependency * for this block. If there is, merge the old * dependency into the new one. */ if (aip->ai_oldblkno == 0) oldaip = NULL; else LIST_FOREACH(oldaip, &indirdep->ir_deplisthd, ai_next) if (oldaip->ai_offset == aip->ai_offset) break; if (oldaip != NULL) { if (oldaip->ai_newblkno != aip->ai_oldblkno) { panic("setup_allocindir_phase2: blkno"); } aip->ai_oldblkno = oldaip->ai_oldblkno; freefrag = oldaip->ai_freefrag; oldaip->ai_freefrag = aip->ai_freefrag; aip->ai_freefrag = freefrag; free_allocindir(oldaip, NULL); } LIST_INSERT_HEAD(&indirdep->ir_deplisthd, aip, ai_next); ((ufs_daddr_t *)indirdep->ir_savebp->b_data) [aip->ai_offset] = aip->ai_oldblkno; FREE_LOCK(&lk); } if (newindirdep) { /* * Avoid any possibility of data corruption by * ensuring that our old version is thrown away. */ newindirdep->ir_savebp->b_flags |= B_INVAL | B_NOCACHE; brelse(newindirdep->ir_savebp); WORKITEM_FREE((caddr_t)newindirdep, D_INDIRDEP); } if (indirdep) break; newindirdep = kmalloc(sizeof(struct indirdep), M_INDIRDEP, M_SOFTDEP_FLAGS); newindirdep->ir_list.wk_type = D_INDIRDEP; newindirdep->ir_state = ATTACHED; LIST_INIT(&newindirdep->ir_deplisthd); LIST_INIT(&newindirdep->ir_donehd); if (bp->b_bio2.bio_offset == NOOFFSET) { VOP_BMAP(bp->b_vp, bp->b_bio1.bio_offset, &bp->b_bio2.bio_offset, NULL, NULL, BUF_CMD_WRITE); } KKASSERT(bp->b_bio2.bio_offset != NOOFFSET); newindirdep->ir_savebp = getblk(ip->i_devvp, bp->b_bio2.bio_offset, bp->b_bcount, 0, 0); BUF_KERNPROC(newindirdep->ir_savebp); bcopy(bp->b_data, newindirdep->ir_savebp->b_data, bp->b_bcount); } } /* * Block de-allocation dependencies. * * When blocks are de-allocated, the on-disk pointers must be nullified before * the blocks are made available for use by other files. (The true * requirement is that old pointers must be nullified before new on-disk * pointers are set. We chose this slightly more stringent requirement to * reduce complexity.) Our implementation handles this dependency by updating * the inode (or indirect block) appropriately but delaying the actual block * de-allocation (i.e., freemap and free space count manipulation) until * after the updated versions reach stable storage. After the disk is * updated, the blocks can be safely de-allocated whenever it is convenient. * This implementation handles only the common case of reducing a file's * length to zero. Other cases are handled by the conventional synchronous * write approach. * * The ffs implementation with which we worked double-checks * the state of the block pointers and file size as it reduces * a file's length. Some of this code is replicated here in our * soft updates implementation. The freeblks->fb_chkcnt field is * used to transfer a part of this information to the procedure * that eventually de-allocates the blocks. * * This routine should be called from the routine that shortens * a file's length, before the inode's size or block pointers * are modified. It will save the block pointer information for * later release and zero the inode so that the calling routine * can release it. */ struct softdep_setup_freeblocks_info { struct fs *fs; struct inode *ip; }; static int softdep_setup_freeblocks_bp(struct buf *bp, void *data); /* * Parameters: * ip: The inode whose length is to be reduced * length: The new length for the file */ void softdep_setup_freeblocks(struct inode *ip, off_t length) { struct softdep_setup_freeblocks_info info; struct freeblks *freeblks; struct inodedep *inodedep; struct allocdirect *adp; struct vnode *vp; struct buf *bp; struct fs *fs; int i, error, delay; int count; fs = ip->i_fs; if (length != 0) panic("softde_setup_freeblocks: non-zero length"); freeblks = kmalloc(sizeof(struct freeblks), M_FREEBLKS, M_SOFTDEP_FLAGS | M_ZERO); freeblks->fb_list.wk_type = D_FREEBLKS; freeblks->fb_state = ATTACHED; freeblks->fb_uid = ip->i_uid; freeblks->fb_previousinum = ip->i_number; freeblks->fb_devvp = ip->i_devvp; freeblks->fb_fs = fs; freeblks->fb_oldsize = ip->i_size; freeblks->fb_newsize = length; freeblks->fb_chkcnt = ip->i_blocks; for (i = 0; i < UFS_NDADDR; i++) { freeblks->fb_dblks[i] = ip->i_db[i]; ip->i_db[i] = 0; } for (i = 0; i < UFS_NIADDR; i++) { freeblks->fb_iblks[i] = ip->i_ib[i]; ip->i_ib[i] = 0; } ip->i_blocks = 0; ip->i_size = 0; /* * Push the zero'ed inode to to its disk buffer so that we are free * to delete its dependencies below. Once the dependencies are gone * the buffer can be safely released. */ if ((error = bread(ip->i_devvp, fsbtodoff(fs, ino_to_fsba(fs, ip->i_number)), (int)fs->fs_bsize, &bp)) != 0) softdep_error("softdep_setup_freeblocks", error); *((struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, ip->i_number)) = ip->i_din; /* * Find and eliminate any inode dependencies. */ ACQUIRE_LOCK(&lk); (void) inodedep_lookup(fs, ip->i_number, DEPALLOC, &inodedep); if ((inodedep->id_state & IOSTARTED) != 0) { panic("softdep_setup_freeblocks: inode busy"); } /* * Add the freeblks structure to the list of operations that * must await the zero'ed inode being written to disk. If we * still have a bitmap dependency (delay == 0), then the inode * has never been written to disk, so we can process the * freeblks below once we have deleted the dependencies. */ delay = (inodedep->id_state & DEPCOMPLETE); if (delay) WORKLIST_INSERT(&inodedep->id_bufwait, &freeblks->fb_list); /* * Because the file length has been truncated to zero, any * pending block allocation dependency structures associated * with this inode are obsolete and can simply be de-allocated. * We must first merge the two dependency lists to get rid of * any duplicate freefrag structures, then purge the merged list. */ merge_inode_lists(inodedep); while ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != NULL) free_allocdirect(&inodedep->id_inoupdt, adp, 1); FREE_LOCK(&lk); bdwrite(bp); /* * We must wait for any I/O in progress to finish so that * all potential buffers on the dirty list will be visible. * Once they are all there, walk the list and get rid of * any dependencies. */ vp = ITOV(ip); ACQUIRE_LOCK(&lk); drain_output(vp, 1); info.fs = fs; info.ip = ip; lwkt_gettoken(&vp->v_token); do { count = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, softdep_setup_freeblocks_bp, &info); } while (count != 0); lwkt_reltoken(&vp->v_token); if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) != 0) (void)free_inodedep(inodedep); if (delay) { freeblks->fb_state |= DEPCOMPLETE; /* * If the inode with zeroed block pointers is now on disk * we can start freeing blocks. Add freeblks to the worklist * instead of calling handle_workitem_freeblocks directly as * it is more likely that additional IO is needed to complete * the request here than in the !delay case. */ if ((freeblks->fb_state & ALLCOMPLETE) == ALLCOMPLETE) add_to_worklist(&freeblks->fb_list); } FREE_LOCK(&lk); /* * If the inode has never been written to disk (delay == 0), * then we can process the freeblks now that we have deleted * the dependencies. */ if (!delay) handle_workitem_freeblocks(freeblks); } static int softdep_setup_freeblocks_bp(struct buf *bp, void *data) { struct softdep_setup_freeblocks_info *info = data; struct inodedep *inodedep; if (getdirtybuf(&bp, MNT_WAIT) == 0) { kprintf("softdep_setup_freeblocks_bp(1): caught bp %p going away\n", bp); return(-1); } if (bp->b_vp != ITOV(info->ip) || (bp->b_flags & B_DELWRI) == 0) { kprintf("softdep_setup_freeblocks_bp(2): caught bp %p going away\n", bp); BUF_UNLOCK(bp); return(-1); } (void) inodedep_lookup(info->fs, info->ip->i_number, 0, &inodedep); deallocate_dependencies(bp, inodedep); bp->b_flags |= B_INVAL | B_NOCACHE; FREE_LOCK(&lk); brelse(bp); ACQUIRE_LOCK(&lk); return(1); } /* * Reclaim any dependency structures from a buffer that is about to * be reallocated to a new vnode. The buffer must be locked, thus, * no I/O completion operations can occur while we are manipulating * its associated dependencies. The mutex is held so that other I/O's * associated with related dependencies do not occur. */ static void deallocate_dependencies(struct buf *bp, struct inodedep *inodedep) { struct worklist *wk; struct indirdep *indirdep; struct allocindir *aip; struct pagedep *pagedep; struct dirrem *dirrem; struct diradd *dap; int i; while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) { switch (wk->wk_type) { case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); /* * None of the indirect pointers will ever be visible, * so they can simply be tossed. GOINGAWAY ensures * that allocated pointers will be saved in the buffer * cache until they are freed. Note that they will * only be able to be found by their physical address * since the inode mapping the logical address will * be gone. The save buffer used for the safe copy * was allocated in setup_allocindir_phase2 using * the physical address so it could be used for this * purpose. Hence we swap the safe copy with the real * copy, allowing the safe copy to be freed and holding * on to the real copy for later use in indir_trunc. * * NOTE: ir_savebp is relative to the block device * so b_bio1 contains the device block number. */ if (indirdep->ir_state & GOINGAWAY) { panic("deallocate_dependencies: already gone"); } indirdep->ir_state |= GOINGAWAY; while ((aip = LIST_FIRST(&indirdep->ir_deplisthd)) != NULL) free_allocindir(aip, inodedep); if (bp->b_bio1.bio_offset >= 0 || bp->b_bio2.bio_offset != indirdep->ir_savebp->b_bio1.bio_offset) { panic("deallocate_dependencies: not indir"); } bcopy(bp->b_data, indirdep->ir_savebp->b_data, bp->b_bcount); WORKLIST_REMOVE(wk); WORKLIST_INSERT_BP(indirdep->ir_savebp, wk); continue; case D_PAGEDEP: pagedep = WK_PAGEDEP(wk); /* * None of the directory additions will ever be * visible, so they can simply be tossed. */ for (i = 0; i < DAHASHSZ; i++) while ((dap = LIST_FIRST(&pagedep->pd_diraddhd[i]))) free_diradd(dap); while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL) free_diradd(dap); /* * Copy any directory remove dependencies to the list * to be processed after the zero'ed inode is written. * If the inode has already been written, then they * can be dumped directly onto the work list. */ LIST_FOREACH(dirrem, &pagedep->pd_dirremhd, dm_next) { LIST_REMOVE(dirrem, dm_next); dirrem->dm_dirinum = pagedep->pd_ino; if (inodedep == NULL || (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) add_to_worklist(&dirrem->dm_list); else WORKLIST_INSERT(&inodedep->id_bufwait, &dirrem->dm_list); } WORKLIST_REMOVE(&pagedep->pd_list); LIST_REMOVE(pagedep, pd_hash); WORKITEM_FREE(pagedep, D_PAGEDEP); continue; case D_ALLOCINDIR: free_allocindir(WK_ALLOCINDIR(wk), inodedep); continue; case D_ALLOCDIRECT: case D_INODEDEP: panic("deallocate_dependencies: Unexpected type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ default: panic("deallocate_dependencies: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } } /* * Free an allocdirect. Generate a new freefrag work request if appropriate. * This routine must be called with splbio interrupts blocked. */ static void free_allocdirect(struct allocdirectlst *adphead, struct allocdirect *adp, int delay) { KKASSERT(lock_held(&lk)); if ((adp->ad_state & DEPCOMPLETE) == 0) LIST_REMOVE(adp, ad_deps); TAILQ_REMOVE(adphead, adp, ad_next); if ((adp->ad_state & COMPLETE) == 0) WORKLIST_REMOVE(&adp->ad_list); if (adp->ad_freefrag != NULL) { if (delay) WORKLIST_INSERT(&adp->ad_inodedep->id_bufwait, &adp->ad_freefrag->ff_list); else add_to_worklist(&adp->ad_freefrag->ff_list); } WORKITEM_FREE(adp, D_ALLOCDIRECT); } /* * Prepare an inode to be freed. The actual free operation is not * done until the zero'ed inode has been written to disk. */ void softdep_freefile(struct vnode *pvp, ino_t ino, int mode) { struct inode *ip = VTOI(pvp); struct inodedep *inodedep; struct freefile *freefile; /* * This sets up the inode de-allocation dependency. */ freefile = kmalloc(sizeof(struct freefile), M_FREEFILE, M_SOFTDEP_FLAGS); freefile->fx_list.wk_type = D_FREEFILE; freefile->fx_list.wk_state = 0; freefile->fx_mode = mode; freefile->fx_oldinum = ino; freefile->fx_devvp = ip->i_devvp; freefile->fx_fs = ip->i_fs; /* * If the inodedep does not exist, then the zero'ed inode has * been written to disk. If the allocated inode has never been * written to disk, then the on-disk inode is zero'ed. In either * case we can free the file immediately. */ ACQUIRE_LOCK(&lk); if (inodedep_lookup(ip->i_fs, ino, 0, &inodedep) == 0 || check_inode_unwritten(inodedep)) { FREE_LOCK(&lk); handle_workitem_freefile(freefile); return; } WORKLIST_INSERT(&inodedep->id_inowait, &freefile->fx_list); FREE_LOCK(&lk); } /* * Check to see if an inode has never been written to disk. If * so free the inodedep and return success, otherwise return failure. * This routine must be called with splbio interrupts blocked. * * If we still have a bitmap dependency, then the inode has never * been written to disk. Drop the dependency as it is no longer * necessary since the inode is being deallocated. We set the * ALLCOMPLETE flags since the bitmap now properly shows that the * inode is not allocated. Even if the inode is actively being * written, it has been rolled back to its zero'ed state, so we * are ensured that a zero inode is what is on the disk. For short * lived files, this change will usually result in removing all the * dependencies from the inode so that it can be freed immediately. */ static int check_inode_unwritten(struct inodedep *inodedep) { if ((inodedep->id_state & DEPCOMPLETE) != 0 || LIST_FIRST(&inodedep->id_pendinghd) != NULL || LIST_FIRST(&inodedep->id_bufwait) != NULL || LIST_FIRST(&inodedep->id_inowait) != NULL || TAILQ_FIRST(&inodedep->id_inoupdt) != NULL || TAILQ_FIRST(&inodedep->id_newinoupdt) != NULL || inodedep->id_nlinkdelta != 0) return (0); /* * Another process might be in initiate_write_inodeblock * trying to allocate memory without holding "Softdep Lock". */ if ((inodedep->id_state & IOSTARTED) != 0 && inodedep->id_savedino == NULL) return(0); inodedep->id_state |= ALLCOMPLETE; LIST_REMOVE(inodedep, id_deps); inodedep->id_buf = NULL; if (inodedep->id_state & ONWORKLIST) WORKLIST_REMOVE(&inodedep->id_list); if (inodedep->id_savedino != NULL) { kfree(inodedep->id_savedino, M_INODEDEP); inodedep->id_savedino = NULL; } if (free_inodedep(inodedep) == 0) { panic("check_inode_unwritten: busy inode"); } return (1); } /* * Try to free an inodedep structure. Return 1 if it could be freed. */ static int free_inodedep(struct inodedep *inodedep) { if ((inodedep->id_state & ONWORKLIST) != 0 || (inodedep->id_state & ALLCOMPLETE) != ALLCOMPLETE || LIST_FIRST(&inodedep->id_pendinghd) != NULL || LIST_FIRST(&inodedep->id_bufwait) != NULL || LIST_FIRST(&inodedep->id_inowait) != NULL || TAILQ_FIRST(&inodedep->id_inoupdt) != NULL || TAILQ_FIRST(&inodedep->id_newinoupdt) != NULL || inodedep->id_nlinkdelta != 0 || inodedep->id_savedino != NULL) return (0); LIST_REMOVE(inodedep, id_hash); WORKITEM_FREE(inodedep, D_INODEDEP); num_inodedep -= 1; return (1); } /* * This workitem routine performs the block de-allocation. * The workitem is added to the pending list after the updated * inode block has been written to disk. As mentioned above, * checks regarding the number of blocks de-allocated (compared * to the number of blocks allocated for the file) are also * performed in this function. */ static void handle_workitem_freeblocks(struct freeblks *freeblks) { struct inode tip; ufs_daddr_t bn; struct fs *fs; int i, level, bsize; long nblocks, blocksreleased = 0; int error, allerror = 0; ufs_lbn_t baselbns[UFS_NIADDR], tmpval; tip.i_number = freeblks->fb_previousinum; tip.i_devvp = freeblks->fb_devvp; tip.i_dev = freeblks->fb_devvp->v_rdev; tip.i_fs = freeblks->fb_fs; tip.i_size = freeblks->fb_oldsize; tip.i_uid = freeblks->fb_uid; fs = freeblks->fb_fs; tmpval = 1; baselbns[0] = UFS_NDADDR; for (i = 1; i < UFS_NIADDR; i++) { tmpval *= NINDIR(fs); baselbns[i] = baselbns[i - 1] + tmpval; } nblocks = btodb(fs->fs_bsize); blocksreleased = 0; /* * Indirect blocks first. */ for (level = (UFS_NIADDR - 1); level >= 0; level--) { if ((bn = freeblks->fb_iblks[level]) == 0) continue; if ((error = indir_trunc(&tip, fsbtodoff(fs, bn), level, baselbns[level], &blocksreleased)) == 0) allerror = error; ffs_blkfree(&tip, bn, fs->fs_bsize); blocksreleased += nblocks; } /* * All direct blocks or frags. */ for (i = (UFS_NDADDR - 1); i >= 0; i--) { if ((bn = freeblks->fb_dblks[i]) == 0) continue; bsize = blksize(fs, &tip, i); ffs_blkfree(&tip, bn, bsize); blocksreleased += btodb(bsize); } #ifdef DIAGNOSTIC if (freeblks->fb_chkcnt != blocksreleased) kprintf("handle_workitem_freeblocks: block count\n"); if (allerror) softdep_error("handle_workitem_freeblks", allerror); #endif /* DIAGNOSTIC */ WORKITEM_FREE(freeblks, D_FREEBLKS); } /* * Release blocks associated with the inode ip and stored in the indirect * block at doffset. If level is greater than SINGLE, the block is an * indirect block and recursive calls to indirtrunc must be used to * cleanse other indirect blocks. */ static int indir_trunc(struct inode *ip, off_t doffset, int level, ufs_lbn_t lbn, long *countp) { struct buf *bp; ufs_daddr_t *bap; ufs_daddr_t nb; struct fs *fs; struct worklist *wk; struct indirdep *indirdep; int i, lbnadd, nblocks; int error, allerror = 0; fs = ip->i_fs; lbnadd = 1; for (i = level; i > 0; i--) lbnadd *= NINDIR(fs); /* * Get buffer of block pointers to be freed. This routine is not * called until the zero'ed inode has been written, so it is safe * to free blocks as they are encountered. Because the inode has * been zero'ed, calls to bmap on these blocks will fail. So, we * have to use the on-disk address and the block device for the * filesystem to look them up. If the file was deleted before its * indirect blocks were all written to disk, the routine that set * us up (deallocate_dependencies) will have arranged to leave * a complete copy of the indirect block in memory for our use. * Otherwise we have to read the blocks in from the disk. */ ACQUIRE_LOCK(&lk); if ((bp = findblk(ip->i_devvp, doffset, FINDBLK_TEST)) != NULL && (wk = LIST_FIRST(&bp->b_dep)) != NULL) { /* * bp must be ir_savebp, which is held locked for our use. */ if (wk->wk_type != D_INDIRDEP || (indirdep = WK_INDIRDEP(wk))->ir_savebp != bp || (indirdep->ir_state & GOINGAWAY) == 0) { panic("indir_trunc: lost indirdep"); } WORKLIST_REMOVE(wk); WORKITEM_FREE(indirdep, D_INDIRDEP); if (LIST_FIRST(&bp->b_dep) != NULL) { panic("indir_trunc: dangling dep"); } FREE_LOCK(&lk); } else { FREE_LOCK(&lk); error = bread(ip->i_devvp, doffset, (int)fs->fs_bsize, &bp); if (error) return (error); } /* * Recursively free indirect blocks. */ bap = (ufs_daddr_t *)bp->b_data; nblocks = btodb(fs->fs_bsize); for (i = NINDIR(fs) - 1; i >= 0; i--) { if ((nb = bap[i]) == 0) continue; if (level != 0) { if ((error = indir_trunc(ip, fsbtodoff(fs, nb), level - 1, lbn + (i * lbnadd), countp)) != 0) allerror = error; } ffs_blkfree(ip, nb, fs->fs_bsize); *countp += nblocks; } bp->b_flags |= B_INVAL | B_NOCACHE; brelse(bp); return (allerror); } /* * Free an allocindir. * This routine must be called with splbio interrupts blocked. */ static void free_allocindir(struct allocindir *aip, struct inodedep *inodedep) { struct freefrag *freefrag; KKASSERT(lock_held(&lk)); if ((aip->ai_state & DEPCOMPLETE) == 0) LIST_REMOVE(aip, ai_deps); if (aip->ai_state & ONWORKLIST) WORKLIST_REMOVE(&aip->ai_list); LIST_REMOVE(aip, ai_next); if ((freefrag = aip->ai_freefrag) != NULL) { if (inodedep == NULL) add_to_worklist(&freefrag->ff_list); else WORKLIST_INSERT(&inodedep->id_bufwait, &freefrag->ff_list); } WORKITEM_FREE(aip, D_ALLOCINDIR); } /* * Directory entry addition dependencies. * * When adding a new directory entry, the inode (with its incremented link * count) must be written to disk before the directory entry's pointer to it. * Also, if the inode is newly allocated, the corresponding freemap must be * updated (on disk) before the directory entry's pointer. These requirements * are met via undo/redo on the directory entry's pointer, which consists * simply of the inode number. * * As directory entries are added and deleted, the free space within a * directory block can become fragmented. The ufs filesystem will compact * a fragmented directory block to make space for a new entry. When this * occurs, the offsets of previously added entries change. Any "diradd" * dependency structures corresponding to these entries must be updated with * the new offsets. */ /* * This routine is called after the in-memory inode's link * count has been incremented, but before the directory entry's * pointer to the inode has been set. * * Parameters: * bp: buffer containing directory block * dp: inode for directory * diroffset: offset of new entry in directory * newinum: inode referenced by new directory entry * newdirbp: non-NULL => contents of new mkdir */ void softdep_setup_directory_add(struct buf *bp, struct inode *dp, off_t diroffset, ino_t newinum, struct buf *newdirbp) { int offset; /* offset of new entry within directory block */ ufs_lbn_t lbn; /* block in directory containing new entry */ struct fs *fs; struct diradd *dap; struct pagedep *pagedep; struct inodedep *inodedep; struct mkdir *mkdir1, *mkdir2; /* * Whiteouts have no dependencies. */ if (newinum == UFS_WINO) { if (newdirbp != NULL) bdwrite(newdirbp); return; } fs = dp->i_fs; lbn = lblkno(fs, diroffset); offset = blkoff(fs, diroffset); dap = kmalloc(sizeof(struct diradd), M_DIRADD, M_SOFTDEP_FLAGS | M_ZERO); dap->da_list.wk_type = D_DIRADD; dap->da_offset = offset; dap->da_newinum = newinum; dap->da_state = ATTACHED; if (newdirbp == NULL) { dap->da_state |= DEPCOMPLETE; ACQUIRE_LOCK(&lk); } else { dap->da_state |= MKDIR_BODY | MKDIR_PARENT; mkdir1 = kmalloc(sizeof(struct mkdir), M_MKDIR, M_SOFTDEP_FLAGS); mkdir1->md_list.wk_type = D_MKDIR; mkdir1->md_state = MKDIR_BODY; mkdir1->md_diradd = dap; mkdir2 = kmalloc(sizeof(struct mkdir), M_MKDIR, M_SOFTDEP_FLAGS); mkdir2->md_list.wk_type = D_MKDIR; mkdir2->md_state = MKDIR_PARENT; mkdir2->md_diradd = dap; /* * Dependency on "." and ".." being written to disk. */ mkdir1->md_buf = newdirbp; ACQUIRE_LOCK(&lk); LIST_INSERT_HEAD(&mkdirlisthd, mkdir1, md_mkdirs); WORKLIST_INSERT_BP(newdirbp, &mkdir1->md_list); FREE_LOCK(&lk); bdwrite(newdirbp); /* * Dependency on link count increase for parent directory */ ACQUIRE_LOCK(&lk); if (inodedep_lookup(dp->i_fs, dp->i_number, 0, &inodedep) == 0 || (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) { dap->da_state &= ~MKDIR_PARENT; WORKITEM_FREE(mkdir2, D_MKDIR); } else { LIST_INSERT_HEAD(&mkdirlisthd, mkdir2, md_mkdirs); WORKLIST_INSERT(&inodedep->id_bufwait,&mkdir2->md_list); } } /* * Link into parent directory pagedep to await its being written. */ if (pagedep_lookup(dp, lbn, DEPALLOC, &pagedep) == 0) WORKLIST_INSERT_BP(bp, &pagedep->pd_list); dap->da_pagedep = pagedep; LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap, da_pdlist); /* * Link into its inodedep. Put it on the id_bufwait list if the inode * is not yet written. If it is written, do the post-inode write * processing to put it on the id_pendinghd list. */ (void) inodedep_lookup(fs, newinum, DEPALLOC, &inodedep); if ((inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) diradd_inode_written(dap, inodedep); else WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list); FREE_LOCK(&lk); } /* * This procedure is called to change the offset of a directory * entry when compacting a directory block which must be owned * exclusively by the caller. Note that the actual entry movement * must be done in this procedure to ensure that no I/O completions * occur while the move is in progress. * * Parameters: * dp: inode for directory * base: address of dp->i_offset * oldloc: address of old directory location * newloc: address of new directory location * entrysize: size of directory entry */ void softdep_change_directoryentry_offset(struct inode *dp, caddr_t base, caddr_t oldloc, caddr_t newloc, int entrysize) { int offset, oldoffset, newoffset; struct pagedep *pagedep; struct diradd *dap; ufs_lbn_t lbn; ACQUIRE_LOCK(&lk); lbn = lblkno(dp->i_fs, dp->i_offset); offset = blkoff(dp->i_fs, dp->i_offset); if (pagedep_lookup(dp, lbn, 0, &pagedep) == 0) goto done; oldoffset = offset + (oldloc - base); newoffset = offset + (newloc - base); LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(oldoffset)], da_pdlist) { if (dap->da_offset != oldoffset) continue; dap->da_offset = newoffset; if (DIRADDHASH(newoffset) == DIRADDHASH(oldoffset)) break; LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(newoffset)], dap, da_pdlist); break; } if (dap == NULL) { LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist) { if (dap->da_offset == oldoffset) { dap->da_offset = newoffset; break; } } } done: bcopy(oldloc, newloc, entrysize); FREE_LOCK(&lk); } /* * Free a diradd dependency structure. This routine must be called * with splbio interrupts blocked. */ static void free_diradd(struct diradd *dap) { struct dirrem *dirrem; struct pagedep *pagedep; struct inodedep *inodedep; struct mkdir *mkdir, *nextmd; KKASSERT(lock_held(&lk)); WORKLIST_REMOVE(&dap->da_list); LIST_REMOVE(dap, da_pdlist); if ((dap->da_state & DIRCHG) == 0) { pagedep = dap->da_pagedep; } else { dirrem = dap->da_previous; pagedep = dirrem->dm_pagedep; dirrem->dm_dirinum = pagedep->pd_ino; add_to_worklist(&dirrem->dm_list); } if (inodedep_lookup(VFSTOUFS(pagedep->pd_mnt)->um_fs, dap->da_newinum, 0, &inodedep) != 0) (void) free_inodedep(inodedep); if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) { for (mkdir = LIST_FIRST(&mkdirlisthd); mkdir; mkdir = nextmd) { nextmd = LIST_NEXT(mkdir, md_mkdirs); if (mkdir->md_diradd != dap) continue; dap->da_state &= ~mkdir->md_state; WORKLIST_REMOVE(&mkdir->md_list); LIST_REMOVE(mkdir, md_mkdirs); WORKITEM_FREE(mkdir, D_MKDIR); } if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) != 0) { panic("free_diradd: unfound ref"); } } WORKITEM_FREE(dap, D_DIRADD); } /* * Directory entry removal dependencies. * * When removing a directory entry, the entry's inode pointer must be * zero'ed on disk before the corresponding inode's link count is decremented * (possibly freeing the inode for re-use). This dependency is handled by * updating the directory entry but delaying the inode count reduction until * after the directory block has been written to disk. After this point, the * inode count can be decremented whenever it is convenient. */ /* * This routine should be called immediately after removing * a directory entry. The inode's link count should not be * decremented by the calling procedure -- the soft updates * code will do this task when it is safe. * * Parameters: * bp: buffer containing directory block * dp: inode for the directory being modified * ip: inode for directory entry being removed * isrmdir: indicates if doing RMDIR */ void softdep_setup_remove(struct buf *bp, struct inode *dp, struct inode *ip, int isrmdir) { struct dirrem *dirrem, *prevdirrem; /* * Allocate a new dirrem if appropriate and ACQUIRE_LOCK. */ dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem); /* * If the COMPLETE flag is clear, then there were no active * entries and we want to roll back to a zeroed entry until * the new inode is committed to disk. If the COMPLETE flag is * set then we have deleted an entry that never made it to * disk. If the entry we deleted resulted from a name change, * then the old name still resides on disk. We cannot delete * its inode (returned to us in prevdirrem) until the zeroed * directory entry gets to disk. The new inode has never been * referenced on the disk, so can be deleted immediately. */ if ((dirrem->dm_state & COMPLETE) == 0) { LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd, dirrem, dm_next); FREE_LOCK(&lk); } else { if (prevdirrem != NULL) LIST_INSERT_HEAD(&dirrem->dm_pagedep->pd_dirremhd, prevdirrem, dm_next); dirrem->dm_dirinum = dirrem->dm_pagedep->pd_ino; FREE_LOCK(&lk); handle_workitem_remove(dirrem); } } /* * Allocate a new dirrem if appropriate and return it along with * its associated pagedep. Called without a lock, returns with lock. */ static long num_dirrem; /* number of dirrem allocated */ /* * Parameters: * bp: buffer containing directory block * dp: inode for the directory being modified * ip: inode for directory entry being removed * isrmdir: indicates if doing RMDIR * prevdirremp: previously referenced inode, if any */ static struct dirrem * newdirrem(struct buf *bp, struct inode *dp, struct inode *ip, int isrmdir, struct dirrem **prevdirremp) { int offset; ufs_lbn_t lbn; struct diradd *dap; struct dirrem *dirrem; struct pagedep *pagedep; /* * Whiteouts have no deletion dependencies. */ if (ip == NULL) panic("newdirrem: whiteout"); /* * If we are over our limit, try to improve the situation. * Limiting the number of dirrem structures will also limit * the number of freefile and freeblks structures. */ if (num_dirrem > max_softdeps / 4) speedup_syncer(NULL); if (num_dirrem > max_softdeps / 2) { ACQUIRE_LOCK(&lk); request_cleanup(FLUSH_REMOVE); FREE_LOCK(&lk); } num_dirrem += 1; dirrem = kmalloc(sizeof(struct dirrem), M_DIRREM, M_SOFTDEP_FLAGS | M_ZERO); dirrem->dm_list.wk_type = D_DIRREM; dirrem->dm_state = isrmdir ? RMDIR : 0; dirrem->dm_mnt = ITOV(ip)->v_mount; dirrem->dm_oldinum = ip->i_number; *prevdirremp = NULL; ACQUIRE_LOCK(&lk); lbn = lblkno(dp->i_fs, dp->i_offset); offset = blkoff(dp->i_fs, dp->i_offset); if (pagedep_lookup(dp, lbn, DEPALLOC, &pagedep) == 0) WORKLIST_INSERT_BP(bp, &pagedep->pd_list); dirrem->dm_pagedep = pagedep; /* * Check for a diradd dependency for the same directory entry. * If present, then both dependencies become obsolete and can * be de-allocated. Check for an entry on both the pd_dirraddhd * list and the pd_pendinghd list. */ LIST_FOREACH(dap, &pagedep->pd_diraddhd[DIRADDHASH(offset)], da_pdlist) if (dap->da_offset == offset) break; if (dap == NULL) { LIST_FOREACH(dap, &pagedep->pd_pendinghd, da_pdlist) if (dap->da_offset == offset) break; if (dap == NULL) return (dirrem); } /* * Must be ATTACHED at this point. */ if ((dap->da_state & ATTACHED) == 0) { panic("newdirrem: not ATTACHED"); } if (dap->da_newinum != ip->i_number) { panic("newdirrem: inum %"PRId64" should be %"PRId64, ip->i_number, dap->da_newinum); } /* * If we are deleting a changed name that never made it to disk, * then return the dirrem describing the previous inode (which * represents the inode currently referenced from this entry on disk). */ if ((dap->da_state & DIRCHG) != 0) { *prevdirremp = dap->da_previous; dap->da_state &= ~DIRCHG; dap->da_pagedep = pagedep; } /* * We are deleting an entry that never made it to disk. * Mark it COMPLETE so we can delete its inode immediately. */ dirrem->dm_state |= COMPLETE; free_diradd(dap); return (dirrem); } /* * Directory entry change dependencies. * * Changing an existing directory entry requires that an add operation * be completed first followed by a deletion. The semantics for the addition * are identical to the description of adding a new entry above except * that the rollback is to the old inode number rather than zero. Once * the addition dependency is completed, the removal is done as described * in the removal routine above. */ /* * This routine should be called immediately after changing * a directory entry. The inode's link count should not be * decremented by the calling procedure -- the soft updates * code will perform this task when it is safe. * * Parameters: * bp: buffer containing directory block * dp: inode for the directory being modified * ip: inode for directory entry being removed * newinum: new inode number for changed entry * isrmdir: indicates if doing RMDIR */ void softdep_setup_directory_change(struct buf *bp, struct inode *dp, struct inode *ip, ino_t newinum, int isrmdir) { int offset; struct diradd *dap = NULL; struct dirrem *dirrem, *prevdirrem; struct pagedep *pagedep; struct inodedep *inodedep; offset = blkoff(dp->i_fs, dp->i_offset); /* * Whiteouts do not need diradd dependencies. */ if (newinum != UFS_WINO) { dap = kmalloc(sizeof(struct diradd), M_DIRADD, M_SOFTDEP_FLAGS | M_ZERO); dap->da_list.wk_type = D_DIRADD; dap->da_state = DIRCHG | ATTACHED | DEPCOMPLETE; dap->da_offset = offset; dap->da_newinum = newinum; } /* * Allocate a new dirrem and ACQUIRE_LOCK. */ dirrem = newdirrem(bp, dp, ip, isrmdir, &prevdirrem); pagedep = dirrem->dm_pagedep; /* * The possible values for isrmdir: * 0 - non-directory file rename * 1 - directory rename within same directory * inum - directory rename to new directory of given inode number * When renaming to a new directory, we are both deleting and * creating a new directory entry, so the link count on the new * directory should not change. Thus we do not need the followup * dirrem which is usually done in handle_workitem_remove. We set * the DIRCHG flag to tell handle_workitem_remove to skip the * followup dirrem. */ if (isrmdir > 1) dirrem->dm_state |= DIRCHG; /* * Whiteouts have no additional dependencies, * so just put the dirrem on the correct list. */ if (newinum == UFS_WINO) { if ((dirrem->dm_state & COMPLETE) == 0) { LIST_INSERT_HEAD(&pagedep->pd_dirremhd, dirrem, dm_next); } else { dirrem->dm_dirinum = pagedep->pd_ino; add_to_worklist(&dirrem->dm_list); } FREE_LOCK(&lk); return; } /* * If the COMPLETE flag is clear, then there were no active * entries and we want to roll back to the previous inode until * the new inode is committed to disk. If the COMPLETE flag is * set, then we have deleted an entry that never made it to disk. * If the entry we deleted resulted from a name change, then the old * inode reference still resides on disk. Any rollback that we do * needs to be to that old inode (returned to us in prevdirrem). If * the entry we deleted resulted from a create, then there is * no entry on the disk, so we want to roll back to zero rather * than the uncommitted inode. In either of the COMPLETE cases we * want to immediately free the unwritten and unreferenced inode. */ if ((dirrem->dm_state & COMPLETE) == 0) { dap->da_previous = dirrem; } else { if (prevdirrem != NULL) { dap->da_previous = prevdirrem; } else { dap->da_state &= ~DIRCHG; dap->da_pagedep = pagedep; } dirrem->dm_dirinum = pagedep->pd_ino; add_to_worklist(&dirrem->dm_list); } /* * Link into its inodedep. Put it on the id_bufwait list if the inode * is not yet written. If it is written, do the post-inode write * processing to put it on the id_pendinghd list. */ if (inodedep_lookup(dp->i_fs, newinum, DEPALLOC, &inodedep) == 0 || (inodedep->id_state & ALLCOMPLETE) == ALLCOMPLETE) { dap->da_state |= COMPLETE; LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list); } else { LIST_INSERT_HEAD(&pagedep->pd_diraddhd[DIRADDHASH(offset)], dap, da_pdlist); WORKLIST_INSERT(&inodedep->id_bufwait, &dap->da_list); } FREE_LOCK(&lk); } /* * Called whenever the link count on an inode is changed. * It creates an inode dependency so that the new reference(s) * to the inode cannot be committed to disk until the updated * inode has been written. * * Parameters: * ip: the inode with the increased link count */ void softdep_change_linkcnt(struct inode *ip) { struct inodedep *inodedep; ACQUIRE_LOCK(&lk); (void) inodedep_lookup(ip->i_fs, ip->i_number, DEPALLOC, &inodedep); if (ip->i_nlink < ip->i_effnlink) { panic("softdep_change_linkcnt: bad delta"); } inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; FREE_LOCK(&lk); } /* * This workitem decrements the inode's link count. * If the link count reaches zero, the file is removed. */ static void handle_workitem_remove(struct dirrem *dirrem) { struct inodedep *inodedep; struct vnode *vp; struct inode *ip; ino_t oldinum; int error; error = VFS_VGET(dirrem->dm_mnt, NULL, dirrem->dm_oldinum, &vp); if (error) { softdep_error("handle_workitem_remove: vget", error); return; } ip = VTOI(vp); ACQUIRE_LOCK(&lk); if ((inodedep_lookup(ip->i_fs, dirrem->dm_oldinum, 0, &inodedep)) == 0){ panic("handle_workitem_remove: lost inodedep"); } /* * Normal file deletion. */ if ((dirrem->dm_state & RMDIR) == 0) { ip->i_nlink--; ip->i_flag |= IN_CHANGE; if (ip->i_nlink < ip->i_effnlink) { panic("handle_workitem_remove: bad file delta"); } inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; FREE_LOCK(&lk); vput(vp); num_dirrem -= 1; WORKITEM_FREE(dirrem, D_DIRREM); return; } /* * Directory deletion. Decrement reference count for both the * just deleted parent directory entry and the reference for ".". * Next truncate the directory to length zero. When the * truncation completes, arrange to have the reference count on * the parent decremented to account for the loss of "..". */ ip->i_nlink -= 2; ip->i_flag |= IN_CHANGE; if (ip->i_nlink < ip->i_effnlink) { panic("handle_workitem_remove: bad dir delta"); } inodedep->id_nlinkdelta = ip->i_nlink - ip->i_effnlink; FREE_LOCK(&lk); if ((error = ffs_truncate(vp, (off_t)0, 0, proc0.p_ucred)) != 0) softdep_error("handle_workitem_remove: truncate", error); /* * Rename a directory to a new parent. Since, we are both deleting * and creating a new directory entry, the link count on the new * directory should not change. Thus we skip the followup dirrem. */ if (dirrem->dm_state & DIRCHG) { vput(vp); num_dirrem -= 1; WORKITEM_FREE(dirrem, D_DIRREM); return; } /* * If the inodedep does not exist, then the zero'ed inode has * been written to disk. If the allocated inode has never been * written to disk, then the on-disk inode is zero'ed. In either * case we can remove the file immediately. */ ACQUIRE_LOCK(&lk); dirrem->dm_state = 0; oldinum = dirrem->dm_oldinum; dirrem->dm_oldinum = dirrem->dm_dirinum; if (inodedep_lookup(ip->i_fs, oldinum, 0, &inodedep) == 0 || check_inode_unwritten(inodedep)) { FREE_LOCK(&lk); vput(vp); handle_workitem_remove(dirrem); return; } WORKLIST_INSERT(&inodedep->id_inowait, &dirrem->dm_list); FREE_LOCK(&lk); ip->i_flag |= IN_CHANGE; ffs_update(vp, 0); vput(vp); } /* * Inode de-allocation dependencies. * * When an inode's link count is reduced to zero, it can be de-allocated. We * found it convenient to postpone de-allocation until after the inode is * written to disk with its new link count (zero). At this point, all of the * on-disk inode's block pointers are nullified and, with careful dependency * list ordering, all dependencies related to the inode will be satisfied and * the corresponding dependency structures de-allocated. So, if/when the * inode is reused, there will be no mixing of old dependencies with new * ones. This artificial dependency is set up by the block de-allocation * procedure above (softdep_setup_freeblocks) and completed by the * following procedure. */ static void handle_workitem_freefile(struct freefile *freefile) { struct vnode vp; struct inode tip; struct inodedep *idp; int error; #ifdef DEBUG ACQUIRE_LOCK(&lk); error = inodedep_lookup(freefile->fx_fs, freefile->fx_oldinum, 0, &idp); FREE_LOCK(&lk); if (error) panic("handle_workitem_freefile: inodedep survived"); #endif tip.i_devvp = freefile->fx_devvp; tip.i_dev = freefile->fx_devvp->v_rdev; tip.i_fs = freefile->fx_fs; vp.v_data = &tip; if ((error = ffs_freefile(&vp, freefile->fx_oldinum, freefile->fx_mode)) != 0) softdep_error("handle_workitem_freefile", error); WORKITEM_FREE(freefile, D_FREEFILE); } /* * Helper function which unlinks marker element from work list and returns * the next element on the list. */ static __inline struct worklist * markernext(struct worklist *marker) { struct worklist *next; next = LIST_NEXT(marker, wk_list); LIST_REMOVE(marker, wk_list); return next; } /* * checkread, checkwrite * * bioops callback - hold io_token */ static int softdep_checkread(struct buf *bp) { /* nothing to do, mp lock not needed */ return(0); } /* * bioops callback - hold io_token */ static int softdep_checkwrite(struct buf *bp) { /* nothing to do, mp lock not needed */ return(0); } /* * Disk writes. * * The dependency structures constructed above are most actively used when file * system blocks are written to disk. No constraints are placed on when a * block can be written, but unsatisfied update dependencies are made safe by * modifying (or replacing) the source memory for the duration of the disk * write. When the disk write completes, the memory block is again brought * up-to-date. * * In-core inode structure reclamation. * * Because there are a finite number of "in-core" inode structures, they are * reused regularly. By transferring all inode-related dependencies to the * in-memory inode block and indexing them separately (via "inodedep"s), we * can allow "in-core" inode structures to be reused at any time and avoid * any increase in contention. * * Called just before entering the device driver to initiate a new disk I/O. * The buffer must be locked, thus, no I/O completion operations can occur * while we are manipulating its associated dependencies. * * bioops callback - hold io_token * * Parameters: * bp: structure describing disk write to occur */ static void softdep_disk_io_initiation(struct buf *bp) { struct worklist *wk; struct worklist marker; struct indirdep *indirdep; /* * We only care about write operations. There should never * be dependencies for reads. */ if (bp->b_cmd == BUF_CMD_READ) panic("softdep_disk_io_initiation: read"); ACQUIRE_LOCK(&lk); marker.wk_type = D_LAST + 1; /* Not a normal workitem */ /* * Do any necessary pre-I/O processing. */ for (wk = LIST_FIRST(&bp->b_dep); wk; wk = markernext(&marker)) { LIST_INSERT_AFTER(wk, &marker, wk_list); switch (wk->wk_type) { case D_PAGEDEP: initiate_write_filepage(WK_PAGEDEP(wk), bp); continue; case D_INODEDEP: initiate_write_inodeblock(WK_INODEDEP(wk), bp); continue; case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); if (indirdep->ir_state & GOINGAWAY) panic("disk_io_initiation: indirdep gone"); /* * If there are no remaining dependencies, this * will be writing the real pointers, so the * dependency can be freed. */ if (LIST_FIRST(&indirdep->ir_deplisthd) == NULL) { indirdep->ir_savebp->b_flags |= B_INVAL | B_NOCACHE; brelse(indirdep->ir_savebp); /* inline expand WORKLIST_REMOVE(wk); */ wk->wk_state &= ~ONWORKLIST; LIST_REMOVE(wk, wk_list); WORKITEM_FREE(indirdep, D_INDIRDEP); continue; } /* * Replace up-to-date version with safe version. */ indirdep->ir_saveddata = kmalloc(bp->b_bcount, M_INDIRDEP, M_SOFTDEP_FLAGS); ACQUIRE_LOCK(&lk); indirdep->ir_state &= ~ATTACHED; indirdep->ir_state |= UNDONE; bcopy(bp->b_data, indirdep->ir_saveddata, bp->b_bcount); bcopy(indirdep->ir_savebp->b_data, bp->b_data, bp->b_bcount); FREE_LOCK(&lk); continue; case D_MKDIR: case D_BMSAFEMAP: case D_ALLOCDIRECT: case D_ALLOCINDIR: continue; default: panic("handle_disk_io_initiation: Unexpected type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } FREE_LOCK(&lk); } /* * Called from within the procedure above to deal with unsatisfied * allocation dependencies in a directory. The buffer must be locked, * thus, no I/O completion operations can occur while we are * manipulating its associated dependencies. */ static void initiate_write_filepage(struct pagedep *pagedep, struct buf *bp) { struct diradd *dap; struct direct *ep; int i; if (pagedep->pd_state & IOSTARTED) { /* * This can only happen if there is a driver that does not * understand chaining. Here biodone will reissue the call * to strategy for the incomplete buffers. */ kprintf("initiate_write_filepage: already started\n"); return; } pagedep->pd_state |= IOSTARTED; ACQUIRE_LOCK(&lk); for (i = 0; i < DAHASHSZ; i++) { LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) { ep = (struct direct *) ((char *)bp->b_data + dap->da_offset); if (ep->d_ino != dap->da_newinum) { panic("%s: dir inum %d != new %"PRId64, "initiate_write_filepage", ep->d_ino, dap->da_newinum); } if (dap->da_state & DIRCHG) ep->d_ino = dap->da_previous->dm_oldinum; else ep->d_ino = 0; dap->da_state &= ~ATTACHED; dap->da_state |= UNDONE; } } FREE_LOCK(&lk); } /* * Called from within the procedure above to deal with unsatisfied * allocation dependencies in an inodeblock. The buffer must be * locked, thus, no I/O completion operations can occur while we * are manipulating its associated dependencies. * * Parameters: * bp: The inode block */ static void initiate_write_inodeblock(struct inodedep *inodedep, struct buf *bp) { struct allocdirect *adp, *lastadp; struct ufs1_dinode *dp; struct ufs1_dinode *sip; struct fs *fs; ufs_lbn_t prevlbn = 0; int i, deplist; if (inodedep->id_state & IOSTARTED) panic("initiate_write_inodeblock: already started"); inodedep->id_state |= IOSTARTED; fs = inodedep->id_fs; dp = (struct ufs1_dinode *)bp->b_data + ino_to_fsbo(fs, inodedep->id_ino); /* * If the bitmap is not yet written, then the allocated * inode cannot be written to disk. */ if ((inodedep->id_state & DEPCOMPLETE) == 0) { if (inodedep->id_savedino != NULL) panic("initiate_write_inodeblock: already doing I/O"); sip = kmalloc(sizeof(struct ufs1_dinode), M_INODEDEP, M_SOFTDEP_FLAGS); inodedep->id_savedino = sip; *inodedep->id_savedino = *dp; bzero((caddr_t)dp, sizeof(struct ufs1_dinode)); dp->di_gen = inodedep->id_savedino->di_gen; return; } /* * If no dependencies, then there is nothing to roll back. */ inodedep->id_savedsize = dp->di_size; if (TAILQ_FIRST(&inodedep->id_inoupdt) == NULL) return; /* * Set the dependencies to busy. */ ACQUIRE_LOCK(&lk); for (deplist = 0, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = TAILQ_NEXT(adp, ad_next)) { #ifdef DIAGNOSTIC if (deplist != 0 && prevlbn >= adp->ad_lbn) { panic("softdep_write_inodeblock: lbn order"); } prevlbn = adp->ad_lbn; if (adp->ad_lbn < UFS_NDADDR && dp->di_db[adp->ad_lbn] != adp->ad_newblkno) { panic("%s: direct pointer #%ld mismatch %d != %d", "softdep_write_inodeblock", adp->ad_lbn, dp->di_db[adp->ad_lbn], adp->ad_newblkno); } if (adp->ad_lbn >= UFS_NDADDR && dp->di_ib[adp->ad_lbn - UFS_NDADDR] != adp->ad_newblkno) { panic("%s: indirect pointer #%ld mismatch %d != %d", "softdep_write_inodeblock", adp->ad_lbn - UFS_NDADDR, dp->di_ib[adp->ad_lbn - UFS_NDADDR], adp->ad_newblkno); } deplist |= 1 << adp->ad_lbn; if ((adp->ad_state & ATTACHED) == 0) { panic("softdep_write_inodeblock: Unknown state 0x%x", adp->ad_state); } #endif /* DIAGNOSTIC */ adp->ad_state &= ~ATTACHED; adp->ad_state |= UNDONE; } /* * The on-disk inode cannot claim to be any larger than the last * fragment that has been written. Otherwise, the on-disk inode * might have fragments that were not the last block in the file * which would corrupt the filesystem. */ for (lastadp = NULL, adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; lastadp = adp, adp = TAILQ_NEXT(adp, ad_next)) { if (adp->ad_lbn >= UFS_NDADDR) break; dp->di_db[adp->ad_lbn] = adp->ad_oldblkno; /* keep going until hitting a rollback to a frag */ if (adp->ad_oldsize == 0 || adp->ad_oldsize == fs->fs_bsize) continue; dp->di_size = fs->fs_bsize * adp->ad_lbn + adp->ad_oldsize; for (i = adp->ad_lbn + 1; i < UFS_NDADDR; i++) { #ifdef DIAGNOSTIC if (dp->di_db[i] != 0 && (deplist & (1 << i)) == 0) { panic("softdep_write_inodeblock: lost dep1"); } #endif /* DIAGNOSTIC */ dp->di_db[i] = 0; } for (i = 0; i < UFS_NIADDR; i++) { #ifdef DIAGNOSTIC if (dp->di_ib[i] != 0 && (deplist & ((1 << UFS_NDADDR) << i)) == 0) { panic("softdep_write_inodeblock: lost dep2"); } #endif /* DIAGNOSTIC */ dp->di_ib[i] = 0; } FREE_LOCK(&lk); return; } /* * If we have zero'ed out the last allocated block of the file, * roll back the size to the last currently allocated block. * We know that this last allocated block is a full-sized as * we already checked for fragments in the loop above. */ if (lastadp != NULL && dp->di_size <= (lastadp->ad_lbn + 1) * fs->fs_bsize) { for (i = lastadp->ad_lbn; i >= 0; i--) if (dp->di_db[i] != 0) break; dp->di_size = (i + 1) * fs->fs_bsize; } /* * The only dependencies are for indirect blocks. * * The file size for indirect block additions is not guaranteed. * Such a guarantee would be non-trivial to achieve. The conventional * synchronous write implementation also does not make this guarantee. * Fsck should catch and fix discrepancies. Arguably, the file size * can be over-estimated without destroying integrity when the file * moves into the indirect blocks (i.e., is large). If we want to * postpone fsck, we are stuck with this argument. */ for (; adp; adp = TAILQ_NEXT(adp, ad_next)) dp->di_ib[adp->ad_lbn - UFS_NDADDR] = 0; FREE_LOCK(&lk); } /* * This routine is called during the completion interrupt * service routine for a disk write (from the procedure called * by the device driver to inform the filesystem caches of * a request completion). It should be called early in this * procedure, before the block is made available to other * processes or other routines are called. * * bioops callback - hold io_token * * Parameters: * bp: describes the completed disk write */ static void softdep_disk_write_complete(struct buf *bp) { struct worklist *wk; struct workhead reattach; struct newblk *newblk; struct allocindir *aip; struct allocdirect *adp; struct indirdep *indirdep; struct inodedep *inodedep; struct bmsafemap *bmsafemap; ACQUIRE_LOCK(&lk); LIST_INIT(&reattach); while ((wk = LIST_FIRST(&bp->b_dep)) != NULL) { WORKLIST_REMOVE(wk); switch (wk->wk_type) { case D_PAGEDEP: if (handle_written_filepage(WK_PAGEDEP(wk), bp)) WORKLIST_INSERT(&reattach, wk); continue; case D_INODEDEP: if (handle_written_inodeblock(WK_INODEDEP(wk), bp)) WORKLIST_INSERT(&reattach, wk); continue; case D_BMSAFEMAP: bmsafemap = WK_BMSAFEMAP(wk); while ((newblk = LIST_FIRST(&bmsafemap->sm_newblkhd))) { newblk->nb_state |= DEPCOMPLETE; newblk->nb_bmsafemap = NULL; LIST_REMOVE(newblk, nb_deps); } while ((adp = LIST_FIRST(&bmsafemap->sm_allocdirecthd))) { adp->ad_state |= DEPCOMPLETE; adp->ad_buf = NULL; LIST_REMOVE(adp, ad_deps); handle_allocdirect_partdone(adp); } while ((aip = LIST_FIRST(&bmsafemap->sm_allocindirhd))) { aip->ai_state |= DEPCOMPLETE; aip->ai_buf = NULL; LIST_REMOVE(aip, ai_deps); handle_allocindir_partdone(aip); } while ((inodedep = LIST_FIRST(&bmsafemap->sm_inodedephd)) != NULL) { inodedep->id_state |= DEPCOMPLETE; LIST_REMOVE(inodedep, id_deps); inodedep->id_buf = NULL; } WORKITEM_FREE(bmsafemap, D_BMSAFEMAP); continue; case D_MKDIR: handle_written_mkdir(WK_MKDIR(wk), MKDIR_BODY); continue; case D_ALLOCDIRECT: adp = WK_ALLOCDIRECT(wk); adp->ad_state |= COMPLETE; handle_allocdirect_partdone(adp); continue; case D_ALLOCINDIR: aip = WK_ALLOCINDIR(wk); aip->ai_state |= COMPLETE; handle_allocindir_partdone(aip); continue; case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); if (indirdep->ir_state & GOINGAWAY) { panic("disk_write_complete: indirdep gone"); } bcopy(indirdep->ir_saveddata, bp->b_data, bp->b_bcount); kfree(indirdep->ir_saveddata, M_INDIRDEP); indirdep->ir_saveddata = NULL; indirdep->ir_state &= ~UNDONE; indirdep->ir_state |= ATTACHED; while ((aip = LIST_FIRST(&indirdep->ir_donehd)) != NULL) { handle_allocindir_partdone(aip); if (aip == LIST_FIRST(&indirdep->ir_donehd)) { panic("disk_write_complete: not gone"); } } WORKLIST_INSERT(&reattach, wk); if ((bp->b_flags & B_DELWRI) == 0) stat_indir_blk_ptrs++; bdirty(bp); continue; default: panic("handle_disk_write_complete: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } /* * Reattach any requests that must be redone. */ while ((wk = LIST_FIRST(&reattach)) != NULL) { WORKLIST_REMOVE(wk); WORKLIST_INSERT_BP(bp, wk); } FREE_LOCK(&lk); } /* * Called from within softdep_disk_write_complete above. Note that * this routine is always called from interrupt level with further * splbio interrupts blocked. * * Parameters: * adp: the completed allocdirect */ static void handle_allocdirect_partdone(struct allocdirect *adp) { struct allocdirect *listadp; struct inodedep *inodedep; long bsize; if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE) return; if (adp->ad_buf != NULL) panic("handle_allocdirect_partdone: dangling dep"); /* * The on-disk inode cannot claim to be any larger than the last * fragment that has been written. Otherwise, the on-disk inode * might have fragments that were not the last block in the file * which would corrupt the filesystem. Thus, we cannot free any * allocdirects after one whose ad_oldblkno claims a fragment as * these blocks must be rolled back to zero before writing the inode. * We check the currently active set of allocdirects in id_inoupdt. */ inodedep = adp->ad_inodedep; bsize = inodedep->id_fs->fs_bsize; TAILQ_FOREACH(listadp, &inodedep->id_inoupdt, ad_next) { /* found our block */ if (listadp == adp) break; /* continue if ad_oldlbn is not a fragment */ if (listadp->ad_oldsize == 0 || listadp->ad_oldsize == bsize) continue; /* hit a fragment */ return; } /* * If we have reached the end of the current list without * finding the just finished dependency, then it must be * on the future dependency list. Future dependencies cannot * be freed until they are moved to the current list. */ if (listadp == NULL) { #ifdef DEBUG TAILQ_FOREACH(listadp, &inodedep->id_newinoupdt, ad_next) /* found our block */ if (listadp == adp) break; if (listadp == NULL) panic("handle_allocdirect_partdone: lost dep"); #endif /* DEBUG */ return; } /* * If we have found the just finished dependency, then free * it along with anything that follows it that is complete. */ for (; adp; adp = listadp) { listadp = TAILQ_NEXT(adp, ad_next); if ((adp->ad_state & ALLCOMPLETE) != ALLCOMPLETE) return; free_allocdirect(&inodedep->id_inoupdt, adp, 1); } } /* * Called from within softdep_disk_write_complete above. Note that * this routine is always called from interrupt level with further * splbio interrupts blocked. * * Parameters: * aip: the completed allocindir */ static void handle_allocindir_partdone(struct allocindir *aip) { struct indirdep *indirdep; if ((aip->ai_state & ALLCOMPLETE) != ALLCOMPLETE) return; if (aip->ai_buf != NULL) panic("handle_allocindir_partdone: dangling dependency"); indirdep = aip->ai_indirdep; if (indirdep->ir_state & UNDONE) { LIST_REMOVE(aip, ai_next); LIST_INSERT_HEAD(&indirdep->ir_donehd, aip, ai_next); return; } ((ufs_daddr_t *)indirdep->ir_savebp->b_data)[aip->ai_offset] = aip->ai_newblkno; LIST_REMOVE(aip, ai_next); if (aip->ai_freefrag != NULL) add_to_worklist(&aip->ai_freefrag->ff_list); WORKITEM_FREE(aip, D_ALLOCINDIR); } /* * Called from within softdep_disk_write_complete above to restore * in-memory inode block contents to their most up-to-date state. Note * that this routine is always called from interrupt level with further * splbio interrupts blocked. * * Parameters: * bp: buffer containing the inode block */ static int handle_written_inodeblock(struct inodedep *inodedep, struct buf *bp) { struct worklist *wk, *filefree; struct allocdirect *adp, *nextadp; struct ufs1_dinode *dp; int hadchanges; if ((inodedep->id_state & IOSTARTED) == 0) panic("handle_written_inodeblock: not started"); inodedep->id_state &= ~IOSTARTED; dp = (struct ufs1_dinode *)bp->b_data + ino_to_fsbo(inodedep->id_fs, inodedep->id_ino); /* * If we had to rollback the inode allocation because of * bitmaps being incomplete, then simply restore it. * Keep the block dirty so that it will not be reclaimed until * all associated dependencies have been cleared and the * corresponding updates written to disk. */ if (inodedep->id_savedino != NULL) { *dp = *inodedep->id_savedino; kfree(inodedep->id_savedino, M_INODEDEP); inodedep->id_savedino = NULL; if ((bp->b_flags & B_DELWRI) == 0) stat_inode_bitmap++; bdirty(bp); return (1); } inodedep->id_state |= COMPLETE; /* * Roll forward anything that had to be rolled back before * the inode could be updated. */ hadchanges = 0; for (adp = TAILQ_FIRST(&inodedep->id_inoupdt); adp; adp = nextadp) { nextadp = TAILQ_NEXT(adp, ad_next); if (adp->ad_state & ATTACHED) panic("handle_written_inodeblock: new entry"); if (adp->ad_lbn < UFS_NDADDR) { if (dp->di_db[adp->ad_lbn] != adp->ad_oldblkno) { panic("%s: %s #%ld mismatch %d != %d", "handle_written_inodeblock", "direct pointer", adp->ad_lbn, dp->di_db[adp->ad_lbn], adp->ad_oldblkno); } dp->di_db[adp->ad_lbn] = adp->ad_newblkno; } else { if (dp->di_ib[adp->ad_lbn - UFS_NDADDR] != 0) { panic("%s: %s #%ld allocated as %d", "handle_written_inodeblock", "indirect pointer", adp->ad_lbn - UFS_NDADDR, dp->di_ib[adp->ad_lbn - UFS_NDADDR]); } dp->di_ib[adp->ad_lbn - UFS_NDADDR] = adp->ad_newblkno; } adp->ad_state &= ~UNDONE; adp->ad_state |= ATTACHED; hadchanges = 1; } if (hadchanges && (bp->b_flags & B_DELWRI) == 0) stat_direct_blk_ptrs++; /* * Reset the file size to its most up-to-date value. */ if (inodedep->id_savedsize == -1) { panic("handle_written_inodeblock: bad size"); } if (dp->di_size != inodedep->id_savedsize) { dp->di_size = inodedep->id_savedsize; hadchanges = 1; } inodedep->id_savedsize = -1; /* * If there were any rollbacks in the inode block, then it must be * marked dirty so that its will eventually get written back in * its correct form. */ if (hadchanges) bdirty(bp); /* * Process any allocdirects that completed during the update. */ if ((adp = TAILQ_FIRST(&inodedep->id_inoupdt)) != NULL) handle_allocdirect_partdone(adp); /* * Process deallocations that were held pending until the * inode had been written to disk. Freeing of the inode * is delayed until after all blocks have been freed to * avoid creation of new <vfsid, inum, lbn> triples * before the old ones have been deleted. */ filefree = NULL; while ((wk = LIST_FIRST(&inodedep->id_bufwait)) != NULL) { WORKLIST_REMOVE(wk); switch (wk->wk_type) { case D_FREEFILE: /* * We defer adding filefree to the worklist until * all other additions have been made to ensure * that it will be done after all the old blocks * have been freed. */ if (filefree != NULL) { panic("handle_written_inodeblock: filefree"); } filefree = wk; continue; case D_MKDIR: handle_written_mkdir(WK_MKDIR(wk), MKDIR_PARENT); continue; case D_DIRADD: diradd_inode_written(WK_DIRADD(wk), inodedep); continue; case D_FREEBLKS: wk->wk_state |= COMPLETE; if ((wk->wk_state & ALLCOMPLETE) != ALLCOMPLETE) continue; /* -- fall through -- */ case D_FREEFRAG: case D_DIRREM: add_to_worklist(wk); continue; default: panic("handle_written_inodeblock: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } if (filefree != NULL) { if (free_inodedep(inodedep) == 0) { panic("handle_written_inodeblock: live inodedep"); } add_to_worklist(filefree); return (0); } /* * If no outstanding dependencies, free it. */ if (free_inodedep(inodedep) || TAILQ_FIRST(&inodedep->id_inoupdt) == NULL) return (0); return (hadchanges); } /* * Process a diradd entry after its dependent inode has been written. * This routine must be called with splbio interrupts blocked. */ static void diradd_inode_written(struct diradd *dap, struct inodedep *inodedep) { struct pagedep *pagedep; dap->da_state |= COMPLETE; if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) { if (dap->da_state & DIRCHG) pagedep = dap->da_previous->dm_pagedep; else pagedep = dap->da_pagedep; LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); } WORKLIST_INSERT(&inodedep->id_pendinghd, &dap->da_list); } /* * Handle the completion of a mkdir dependency. */ static void handle_written_mkdir(struct mkdir *mkdir, int type) { struct diradd *dap; struct pagedep *pagedep; if (mkdir->md_state != type) { panic("handle_written_mkdir: bad type"); } dap = mkdir->md_diradd; dap->da_state &= ~type; if ((dap->da_state & (MKDIR_PARENT | MKDIR_BODY)) == 0) dap->da_state |= DEPCOMPLETE; if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) { if (dap->da_state & DIRCHG) pagedep = dap->da_previous->dm_pagedep; else pagedep = dap->da_pagedep; LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); } LIST_REMOVE(mkdir, md_mkdirs); WORKITEM_FREE(mkdir, D_MKDIR); } /* * Called from within softdep_disk_write_complete above. * A write operation was just completed. Removed inodes can * now be freed and associated block pointers may be committed. * Note that this routine is always called from interrupt level * with further splbio interrupts blocked. * * Parameters: * bp: buffer containing the written page */ static int handle_written_filepage(struct pagedep *pagedep, struct buf *bp) { struct dirrem *dirrem; struct diradd *dap, *nextdap; struct direct *ep; int i, chgs; if ((pagedep->pd_state & IOSTARTED) == 0) { panic("handle_written_filepage: not started"); } pagedep->pd_state &= ~IOSTARTED; /* * Process any directory removals that have been committed. */ while ((dirrem = LIST_FIRST(&pagedep->pd_dirremhd)) != NULL) { LIST_REMOVE(dirrem, dm_next); dirrem->dm_dirinum = pagedep->pd_ino; add_to_worklist(&dirrem->dm_list); } /* * Free any directory additions that have been committed. */ while ((dap = LIST_FIRST(&pagedep->pd_pendinghd)) != NULL) free_diradd(dap); /* * Uncommitted directory entries must be restored. */ for (chgs = 0, i = 0; i < DAHASHSZ; i++) { for (dap = LIST_FIRST(&pagedep->pd_diraddhd[i]); dap; dap = nextdap) { nextdap = LIST_NEXT(dap, da_pdlist); if (dap->da_state & ATTACHED) { panic("handle_written_filepage: attached"); } ep = (struct direct *) ((char *)bp->b_data + dap->da_offset); ep->d_ino = dap->da_newinum; dap->da_state &= ~UNDONE; dap->da_state |= ATTACHED; chgs = 1; /* * If the inode referenced by the directory has * been written out, then the dependency can be * moved to the pending list. */ if ((dap->da_state & ALLCOMPLETE) == ALLCOMPLETE) { LIST_REMOVE(dap, da_pdlist); LIST_INSERT_HEAD(&pagedep->pd_pendinghd, dap, da_pdlist); } } } /* * If there were any rollbacks in the directory, then it must be * marked dirty so that its will eventually get written back in * its correct form. */ if (chgs) { if ((bp->b_flags & B_DELWRI) == 0) stat_dir_entry++; bdirty(bp); } /* * If no dependencies remain, the pagedep will be freed. * Otherwise it will remain to update the page before it * is written back to disk. */ if (LIST_FIRST(&pagedep->pd_pendinghd) == NULL) { for (i = 0; i < DAHASHSZ; i++) if (LIST_FIRST(&pagedep->pd_diraddhd[i]) != NULL) break; if (i == DAHASHSZ) { LIST_REMOVE(pagedep, pd_hash); WORKITEM_FREE(pagedep, D_PAGEDEP); return (0); } } return (1); } /* * Writing back in-core inode structures. * * The filesystem only accesses an inode's contents when it occupies an * "in-core" inode structure. These "in-core" structures are separate from * the page frames used to cache inode blocks. Only the latter are * transferred to/from the disk. So, when the updated contents of the * "in-core" inode structure are copied to the corresponding in-memory inode * block, the dependencies are also transferred. The following procedure is * called when copying a dirty "in-core" inode to a cached inode block. */ /* * Called when an inode is loaded from disk. If the effective link count * differed from the actual link count when it was last flushed, then we * need to ensure that the correct effective link count is put back. * * Parameters: * ip: the "in_core" copy of the inode */ void softdep_load_inodeblock(struct inode *ip) { struct inodedep *inodedep; /* * Check for alternate nlink count. */ ip->i_effnlink = ip->i_nlink; ACQUIRE_LOCK(&lk); if (inodedep_lookup(ip->i_fs, ip->i_number, 0, &inodedep) == 0) { FREE_LOCK(&lk); return; } ip->i_effnlink -= inodedep->id_nlinkdelta; FREE_LOCK(&lk); } /* * This routine is called just before the "in-core" inode * information is to be copied to the in-memory inode block. * Recall that an inode block contains several inodes. If * the force flag is set, then the dependencies will be * cleared so that the update can always be made. Note that * the buffer is locked when this routine is called, so we * will never be in the middle of writing the inode block * to disk. * * Parameters: * ip: the "in_core" copy of the inode * bp: the buffer containing the inode block * waitfor: nonzero => update must be allowed */ void softdep_update_inodeblock(struct inode *ip, struct buf *bp, int waitfor) { struct inodedep *inodedep; struct worklist *wk; struct buf *ibp; int error, gotit; /* * If the effective link count is not equal to the actual link * count, then we must track the difference in an inodedep while * the inode is (potentially) tossed out of the cache. Otherwise, * if there is no existing inodedep, then there are no dependencies * to track. */ ACQUIRE_LOCK(&lk); if (inodedep_lookup(ip->i_fs, ip->i_number, 0, &inodedep) == 0) { FREE_LOCK(&lk); if (ip->i_effnlink != ip->i_nlink) panic("softdep_update_inodeblock: bad link count"); return; } if (inodedep->id_nlinkdelta != ip->i_nlink - ip->i_effnlink) { panic("softdep_update_inodeblock: bad delta"); } /* * Changes have been initiated. Anything depending on these * changes cannot occur until this inode has been written. */ inodedep->id_state &= ~COMPLETE; if ((inodedep->id_state & ONWORKLIST) == 0) WORKLIST_INSERT_BP(bp, &inodedep->id_list); /* * Any new dependencies associated with the incore inode must * now be moved to the list associated with the buffer holding * the in-memory copy of the inode. Once merged process any * allocdirects that are completed by the merger. */ merge_inode_lists(inodedep); if (TAILQ_FIRST(&inodedep->id_inoupdt) != NULL) handle_allocdirect_partdone(TAILQ_FIRST(&inodedep->id_inoupdt)); /* * Now that the inode has been pushed into the buffer, the * operations dependent on the inode being written to disk * can be moved to the id_bufwait so that they will be * processed when the buffer I/O completes. */ while ((wk = LIST_FIRST(&inodedep->id_inowait)) != NULL) { WORKLIST_REMOVE(wk); WORKLIST_INSERT(&inodedep->id_bufwait, wk); } /* * Newly allocated inodes cannot be written until the bitmap * that allocates them have been written (indicated by * DEPCOMPLETE being set in id_state). If we are doing a * forced sync (e.g., an fsync on a file), we force the bitmap * to be written so that the update can be done. */ if (waitfor == 0) { FREE_LOCK(&lk); return; } retry: if ((inodedep->id_state & DEPCOMPLETE) != 0) { FREE_LOCK(&lk); return; } gotit = getdirtybuf(&inodedep->id_buf, MNT_WAIT); if (gotit == 0) { if (inodedep_lookup(ip->i_fs, ip->i_number, 0, &inodedep) != 0) goto retry; FREE_LOCK(&lk); return; } ibp = inodedep->id_buf; FREE_LOCK(&lk); if ((error = bwrite(ibp)) != 0) softdep_error("softdep_update_inodeblock: bwrite", error); } /* * Merge the new inode dependency list (id_newinoupdt) into the old * inode dependency list (id_inoupdt). This routine must be called * with splbio interrupts blocked. */ static void merge_inode_lists(struct inodedep *inodedep) { struct allocdirect *listadp, *newadp; newadp = TAILQ_FIRST(&inodedep->id_newinoupdt); for (listadp = TAILQ_FIRST(&inodedep->id_inoupdt); listadp && newadp;) { if (listadp->ad_lbn < newadp->ad_lbn) { listadp = TAILQ_NEXT(listadp, ad_next); continue; } TAILQ_REMOVE(&inodedep->id_newinoupdt, newadp, ad_next); TAILQ_INSERT_BEFORE(listadp, newadp, ad_next); if (listadp->ad_lbn == newadp->ad_lbn) { allocdirect_merge(&inodedep->id_inoupdt, newadp, listadp); listadp = newadp; } newadp = TAILQ_FIRST(&inodedep->id_newinoupdt); } while ((newadp = TAILQ_FIRST(&inodedep->id_newinoupdt)) != NULL) { TAILQ_REMOVE(&inodedep->id_newinoupdt, newadp, ad_next); TAILQ_INSERT_TAIL(&inodedep->id_inoupdt, newadp, ad_next); } } /* * If we are doing an fsync, then we must ensure that any directory * entries for the inode have been written after the inode gets to disk. * * bioops callback - hold io_token * * Parameters: * vp: the "in_core" copy of the inode */ static int softdep_fsync(struct vnode *vp) { struct inodedep *inodedep; struct pagedep *pagedep; struct worklist *wk; struct diradd *dap; struct mount *mnt; struct vnode *pvp; struct inode *ip; struct buf *bp; struct fs *fs; int error, flushparent; ino_t parentino; ufs_lbn_t lbn; /* * Move check from original kernel code, possibly not needed any * more with the per-mount bioops. */ if ((vp->v_mount->mnt_flag & MNT_SOFTDEP) == 0) return (0); ip = VTOI(vp); fs = ip->i_fs; ACQUIRE_LOCK(&lk); if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) == 0) { FREE_LOCK(&lk); return (0); } if (LIST_FIRST(&inodedep->id_inowait) != NULL || LIST_FIRST(&inodedep->id_bufwait) != NULL || TAILQ_FIRST(&inodedep->id_inoupdt) != NULL || TAILQ_FIRST(&inodedep->id_newinoupdt) != NULL) { panic("softdep_fsync: pending ops"); } for (error = 0, flushparent = 0; ; ) { if ((wk = LIST_FIRST(&inodedep->id_pendinghd)) == NULL) break; if (wk->wk_type != D_DIRADD) { panic("softdep_fsync: Unexpected type %s", TYPENAME(wk->wk_type)); } dap = WK_DIRADD(wk); /* * Flush our parent if this directory entry * has a MKDIR_PARENT dependency. */ if (dap->da_state & DIRCHG) pagedep = dap->da_previous->dm_pagedep; else pagedep = dap->da_pagedep; mnt = pagedep->pd_mnt; parentino = pagedep->pd_ino; lbn = pagedep->pd_lbn; if ((dap->da_state & (MKDIR_BODY | COMPLETE)) != COMPLETE) { panic("softdep_fsync: dirty"); } flushparent = dap->da_state & MKDIR_PARENT; /* * If we are being fsync'ed as part of vgone'ing this vnode, * then we will not be able to release and recover the * vnode below, so we just have to give up on writing its * directory entry out. It will eventually be written, just * not now, but then the user was not asking to have it * written, so we are not breaking any promises. */ if (vp->v_flag & VRECLAIMED) break; /* * We prevent deadlock by always fetching inodes from the * root, moving down the directory tree. Thus, when fetching * our parent directory, we must unlock ourselves before * requesting the lock on our parent. See the comment in * ufs_lookup for details on possible races. */ FREE_LOCK(&lk); vn_unlock(vp); error = VFS_VGET(mnt, NULL, parentino, &pvp); vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); if (error != 0) { return (error); } if (flushparent) { if ((error = ffs_update(pvp, 1)) != 0) { vput(pvp); return (error); } } /* * Flush directory page containing the inode's name. */ error = bread(pvp, lblktodoff(fs, lbn), blksize(fs, VTOI(pvp), lbn), &bp); if (error == 0) error = bwrite(bp); vput(pvp); if (error != 0) { return (error); } ACQUIRE_LOCK(&lk); if (inodedep_lookup(fs, ip->i_number, 0, &inodedep) == 0) break; } FREE_LOCK(&lk); return (0); } /* * Flush all the dirty bitmaps associated with the block device * before flushing the rest of the dirty blocks so as to reduce * the number of dependencies that will have to be rolled back. */ static int softdep_fsync_mountdev_bp(struct buf *bp, void *data); void softdep_fsync_mountdev(struct vnode *vp) { if (!vn_isdisk(vp, NULL)) panic("softdep_fsync_mountdev: vnode not a disk"); ACQUIRE_LOCK(&lk); lwkt_gettoken(&vp->v_token); RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, softdep_fsync_mountdev_bp, vp); lwkt_reltoken(&vp->v_token); drain_output(vp, 1); FREE_LOCK(&lk); } static int softdep_fsync_mountdev_bp(struct buf *bp, void *data) { struct worklist *wk; struct vnode *vp = data; /* * If it is already scheduled, skip to the next buffer. */ if (BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT)) return(0); if (bp->b_vp != vp || (bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); kprintf("softdep_fsync_mountdev_bp: warning, buffer %p ripped out from under vnode %p\n", bp, vp); return(0); } /* * We are only interested in bitmaps with outstanding * dependencies. */ if ((wk = LIST_FIRST(&bp->b_dep)) == NULL || wk->wk_type != D_BMSAFEMAP) { BUF_UNLOCK(bp); return(0); } bremfree(bp); FREE_LOCK(&lk); (void) bawrite(bp); ACQUIRE_LOCK(&lk); return(0); } /* * This routine is called when we are trying to synchronously flush a * file. This routine must eliminate any filesystem metadata dependencies * so that the syncing routine can succeed by pushing the dirty blocks * associated with the file. If any I/O errors occur, they are returned. */ struct softdep_sync_metadata_info { struct vnode *vp; int waitfor; }; static int softdep_sync_metadata_bp(struct buf *bp, void *data); int softdep_sync_metadata(struct vnode *vp, struct thread *td) { struct softdep_sync_metadata_info info; int error, waitfor; /* * Check whether this vnode is involved in a filesystem * that is doing soft dependency processing. */ if (!vn_isdisk(vp, NULL)) { if (!DOINGSOFTDEP(vp)) return (0); } else if (vp->v_rdev->si_mountpoint == NULL || (vp->v_rdev->si_mountpoint->mnt_flag & MNT_SOFTDEP) == 0) return (0); /* * Ensure that any direct block dependencies have been cleared. */ ACQUIRE_LOCK(&lk); if ((error = flush_inodedep_deps(VTOI(vp)->i_fs, VTOI(vp)->i_number))) { FREE_LOCK(&lk); return (error); } /* * For most files, the only metadata dependencies are the * cylinder group maps that allocate their inode or blocks. * The block allocation dependencies can be found by traversing * the dependency lists for any buffers that remain on their * dirty buffer list. The inode allocation dependency will * be resolved when the inode is updated with MNT_WAIT. * This work is done in two passes. The first pass grabs most * of the buffers and begins asynchronously writing them. The * only way to wait for these asynchronous writes is to sleep * on the filesystem vnode which may stay busy for a long time * if the filesystem is active. So, instead, we make a second * pass over the dependencies blocking on each write. In the * usual case we will be blocking against a write that we * initiated, so when it is done the dependency will have been * resolved. Thus the second pass is expected to end quickly. */ waitfor = MNT_NOWAIT; top: /* * We must wait for any I/O in progress to finish so that * all potential buffers on the dirty list will be visible. */ drain_output(vp, 1); info.vp = vp; info.waitfor = waitfor; lwkt_gettoken(&vp->v_token); error = RB_SCAN(buf_rb_tree, &vp->v_rbdirty_tree, NULL, softdep_sync_metadata_bp, &info); lwkt_reltoken(&vp->v_token); if (error < 0) { FREE_LOCK(&lk); return(-error); /* error code */ } /* * The brief unlock is to allow any pent up dependency * processing to be done. Then proceed with the second pass. */ if (waitfor & MNT_NOWAIT) { waitfor = MNT_WAIT; FREE_LOCK(&lk); ACQUIRE_LOCK(&lk); goto top; } /* * If we have managed to get rid of all the dirty buffers, * then we are done. For certain directories and block * devices, we may need to do further work. * * We must wait for any I/O in progress to finish so that * all potential buffers on the dirty list will be visible. */ drain_output(vp, 1); if (RB_EMPTY(&vp->v_rbdirty_tree)) { FREE_LOCK(&lk); return (0); } FREE_LOCK(&lk); /* * If we are trying to sync a block device, some of its buffers may * contain metadata that cannot be written until the contents of some * partially written files have been written to disk. The only easy * way to accomplish this is to sync the entire filesystem (luckily * this happens rarely). */ if (vn_isdisk(vp, NULL) && vp->v_rdev && vp->v_rdev->si_mountpoint && !vn_islocked(vp) && (error = VFS_SYNC(vp->v_rdev->si_mountpoint, MNT_WAIT)) != 0) return (error); return (0); } static int softdep_sync_metadata_bp(struct buf *bp, void *data) { struct softdep_sync_metadata_info *info = data; struct pagedep *pagedep; struct allocdirect *adp; struct allocindir *aip; struct worklist *wk; struct buf *nbp; int error; int i; if (getdirtybuf(&bp, MNT_WAIT) == 0) { kprintf("softdep_sync_metadata_bp(1): caught buf %p going away\n", bp); return (1); } if (bp->b_vp != info->vp || (bp->b_flags & B_DELWRI) == 0) { kprintf("softdep_sync_metadata_bp(2): caught buf %p going away vp %p\n", bp, info->vp); BUF_UNLOCK(bp); return(1); } /* * As we hold the buffer locked, none of its dependencies * will disappear. */ LIST_FOREACH(wk, &bp->b_dep, wk_list) { switch (wk->wk_type) { case D_ALLOCDIRECT: adp = WK_ALLOCDIRECT(wk); if (adp->ad_state & DEPCOMPLETE) break; nbp = adp->ad_buf; if (getdirtybuf(&nbp, info->waitfor) == 0) break; FREE_LOCK(&lk); if (info->waitfor & MNT_NOWAIT) { bawrite(nbp); } else if ((error = bwrite(nbp)) != 0) { bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } ACQUIRE_LOCK(&lk); break; case D_ALLOCINDIR: aip = WK_ALLOCINDIR(wk); if (aip->ai_state & DEPCOMPLETE) break; nbp = aip->ai_buf; if (getdirtybuf(&nbp, info->waitfor) == 0) break; FREE_LOCK(&lk); if (info->waitfor & MNT_NOWAIT) { bawrite(nbp); } else if ((error = bwrite(nbp)) != 0) { bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } ACQUIRE_LOCK(&lk); break; case D_INDIRDEP: restart: LIST_FOREACH(aip, &WK_INDIRDEP(wk)->ir_deplisthd, ai_next) { if (aip->ai_state & DEPCOMPLETE) continue; nbp = aip->ai_buf; if (getdirtybuf(&nbp, MNT_WAIT) == 0) goto restart; FREE_LOCK(&lk); if ((error = bwrite(nbp)) != 0) { bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } ACQUIRE_LOCK(&lk); goto restart; } break; case D_INODEDEP: if ((error = flush_inodedep_deps(WK_INODEDEP(wk)->id_fs, WK_INODEDEP(wk)->id_ino)) != 0) { FREE_LOCK(&lk); bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } break; case D_PAGEDEP: /* * We are trying to sync a directory that may * have dependencies on both its own metadata * and/or dependencies on the inodes of any * recently allocated files. We walk its diradd * lists pushing out the associated inode. */ pagedep = WK_PAGEDEP(wk); for (i = 0; i < DAHASHSZ; i++) { if (LIST_FIRST(&pagedep->pd_diraddhd[i]) == NULL) continue; if ((error = flush_pagedep_deps(info->vp, pagedep->pd_mnt, &pagedep->pd_diraddhd[i]))) { FREE_LOCK(&lk); bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } } break; case D_MKDIR: /* * This case should never happen if the vnode has * been properly sync'ed. However, if this function * is used at a place where the vnode has not yet * been sync'ed, this dependency can show up. So, * rather than panic, just flush it. */ nbp = WK_MKDIR(wk)->md_buf; if (getdirtybuf(&nbp, info->waitfor) == 0) break; FREE_LOCK(&lk); if (info->waitfor & MNT_NOWAIT) { bawrite(nbp); } else if ((error = bwrite(nbp)) != 0) { bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } ACQUIRE_LOCK(&lk); break; case D_BMSAFEMAP: /* * This case should never happen if the vnode has * been properly sync'ed. However, if this function * is used at a place where the vnode has not yet * been sync'ed, this dependency can show up. So, * rather than panic, just flush it. * * nbp can wind up == bp if a device node for the * same filesystem is being fsynced at the same time, * leading to a panic if we don't catch the case. */ nbp = WK_BMSAFEMAP(wk)->sm_buf; if (nbp == bp) break; if (getdirtybuf(&nbp, info->waitfor) == 0) break; FREE_LOCK(&lk); if (info->waitfor & MNT_NOWAIT) { bawrite(nbp); } else if ((error = bwrite(nbp)) != 0) { bawrite(bp); ACQUIRE_LOCK(&lk); return (-error); } ACQUIRE_LOCK(&lk); break; default: panic("softdep_sync_metadata: Unknown type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } FREE_LOCK(&lk); bawrite(bp); ACQUIRE_LOCK(&lk); return(0); } /* * Flush the dependencies associated with an inodedep. * Called with splbio blocked. */ static int flush_inodedep_deps(struct fs *fs, ino_t ino) { struct inodedep *inodedep; struct allocdirect *adp; int error, waitfor; struct buf *bp; /* * This work is done in two passes. The first pass grabs most * of the buffers and begins asynchronously writing them. The * only way to wait for these asynchronous writes is to sleep * on the filesystem vnode which may stay busy for a long time * if the filesystem is active. So, instead, we make a second * pass over the dependencies blocking on each write. In the * usual case we will be blocking against a write that we * initiated, so when it is done the dependency will have been * resolved. Thus the second pass is expected to end quickly. * We give a brief window at the top of the loop to allow * any pending I/O to complete. */ for (waitfor = MNT_NOWAIT; ; ) { FREE_LOCK(&lk); ACQUIRE_LOCK(&lk); if (inodedep_lookup(fs, ino, 0, &inodedep) == 0) return (0); TAILQ_FOREACH(adp, &inodedep->id_inoupdt, ad_next) { if (adp->ad_state & DEPCOMPLETE) continue; bp = adp->ad_buf; if (getdirtybuf(&bp, waitfor) == 0) { if (waitfor & MNT_NOWAIT) continue; break; } FREE_LOCK(&lk); if (waitfor & MNT_NOWAIT) { bawrite(bp); } else if ((error = bwrite(bp)) != 0) { ACQUIRE_LOCK(&lk); return (error); } ACQUIRE_LOCK(&lk); break; } if (adp != NULL) continue; TAILQ_FOREACH(adp, &inodedep->id_newinoupdt, ad_next) { if (adp->ad_state & DEPCOMPLETE) continue; bp = adp->ad_buf; if (getdirtybuf(&bp, waitfor) == 0) { if (waitfor & MNT_NOWAIT) continue; break; } FREE_LOCK(&lk); if (waitfor & MNT_NOWAIT) { bawrite(bp); } else if ((error = bwrite(bp)) != 0) { ACQUIRE_LOCK(&lk); return (error); } ACQUIRE_LOCK(&lk); break; } if (adp != NULL) continue; /* * If pass2, we are done, otherwise do pass 2. */ if (waitfor == MNT_WAIT) break; waitfor = MNT_WAIT; } /* * Try freeing inodedep in case all dependencies have been removed. */ if (inodedep_lookup(fs, ino, 0, &inodedep) != 0) (void) free_inodedep(inodedep); return (0); } /* * Eliminate a pagedep dependency by flushing out all its diradd dependencies. * Called with splbio blocked. */ static int flush_pagedep_deps(struct vnode *pvp, struct mount *mp, struct diraddhd *diraddhdp) { struct inodedep *inodedep; struct ufsmount *ump; struct diradd *dap; struct worklist *wk; struct vnode *vp; int gotit, error = 0; struct buf *bp; ino_t inum; ump = VFSTOUFS(mp); while ((dap = LIST_FIRST(diraddhdp)) != NULL) { /* * Flush ourselves if this directory entry * has a MKDIR_PARENT dependency. */ if (dap->da_state & MKDIR_PARENT) { FREE_LOCK(&lk); if ((error = ffs_update(pvp, 1)) != 0) break; ACQUIRE_LOCK(&lk); /* * If that cleared dependencies, go on to next. */ if (dap != LIST_FIRST(diraddhdp)) continue; if (dap->da_state & MKDIR_PARENT) { panic("flush_pagedep_deps: MKDIR_PARENT"); } } /* * A newly allocated directory must have its "." and * ".." entries written out before its name can be * committed in its parent. We do not want or need * the full semantics of a synchronous VOP_FSYNC as * that may end up here again, once for each directory * level in the filesystem. Instead, we push the blocks * and wait for them to clear. We have to fsync twice * because the first call may choose to defer blocks * that still have dependencies, but deferral will * happen at most once. */ inum = dap->da_newinum; if (dap->da_state & MKDIR_BODY) { FREE_LOCK(&lk); if ((error = VFS_VGET(mp, NULL, inum, &vp)) != 0) break; if ((error=VOP_FSYNC(vp, MNT_NOWAIT, 0)) || (error=VOP_FSYNC(vp, MNT_NOWAIT, 0))) { vput(vp); break; } drain_output(vp, 0); /* * If first block is still dirty with a D_MKDIR * dependency then it needs to be written now. */ error = 0; ACQUIRE_LOCK(&lk); bp = findblk(vp, 0, FINDBLK_TEST); if (bp == NULL) { FREE_LOCK(&lk); goto mkdir_body_continue; } LIST_FOREACH(wk, &bp->b_dep, wk_list) if (wk->wk_type == D_MKDIR) { gotit = getdirtybuf(&bp, MNT_WAIT); FREE_LOCK(&lk); if (gotit && (error = bwrite(bp)) != 0) goto mkdir_body_continue; break; } if (wk == NULL) FREE_LOCK(&lk); mkdir_body_continue: vput(vp); /* Flushing of first block failed. */ if (error) break; ACQUIRE_LOCK(&lk); /* * If that cleared dependencies, go on to next. */ if (dap != LIST_FIRST(diraddhdp)) continue; if (dap->da_state & MKDIR_BODY) { panic("flush_pagedep_deps: %p MKDIR_BODY", dap); } } /* * Flush the inode on which the directory entry depends. * Having accounted for MKDIR_PARENT and MKDIR_BODY above, * the only remaining dependency is that the updated inode * count must get pushed to disk. The inode has already * been pushed into its inode buffer (via VOP_UPDATE) at * the time of the reference count change. So we need only * locate that buffer, ensure that there will be no rollback * caused by a bitmap dependency, then write the inode buffer. */ retry_lookup: if (inodedep_lookup(ump->um_fs, inum, 0, &inodedep) == 0) { panic("flush_pagedep_deps: lost inode"); } /* * If the inode still has bitmap dependencies, * push them to disk. */ if ((inodedep->id_state & DEPCOMPLETE) == 0) { gotit = getdirtybuf(&inodedep->id_buf, MNT_WAIT); if (gotit == 0) goto retry_lookup; FREE_LOCK(&lk); if (gotit && (error = bwrite(inodedep->id_buf)) != 0) break; ACQUIRE_LOCK(&lk); if (dap != LIST_FIRST(diraddhdp)) continue; } /* * If the inode is still sitting in a buffer waiting * to be written, push it to disk. */ FREE_LOCK(&lk); if ((error = bread(ump->um_devvp, fsbtodoff(ump->um_fs, ino_to_fsba(ump->um_fs, inum)), (int)ump->um_fs->fs_bsize, &bp)) != 0) break; if ((error = bwrite(bp)) != 0) break; ACQUIRE_LOCK(&lk); /* * If we have failed to get rid of all the dependencies * then something is seriously wrong. */ if (dap == LIST_FIRST(diraddhdp)) { panic("flush_pagedep_deps: flush failed"); } } if (error) ACQUIRE_LOCK(&lk); return (error); } /* * A large burst of file addition or deletion activity can drive the * memory load excessively high. First attempt to slow things down * using the techniques below. If that fails, this routine requests * the offending operations to fall back to running synchronously * until the memory load returns to a reasonable level. */ int softdep_slowdown(struct vnode *vp) { int max_softdeps_hard; max_softdeps_hard = max_softdeps * 11 / 10; if (num_dirrem < max_softdeps_hard / 2 && num_inodedep < max_softdeps_hard) return (0); stat_sync_limit_hit += 1; return (1); } /* * If memory utilization has gotten too high, deliberately slow things * down and speed up the I/O processing. */ static int request_cleanup(int resource) { struct thread *td = curthread; /* XXX */ KKASSERT(lock_held(&lk)); /* * We never hold up the filesystem syncer process. */ if (td == filesys_syncer) return (0); /* * First check to see if the work list has gotten backlogged. * If it has, co-opt this process to help clean up two entries. * Because this process may hold inodes locked, we cannot * handle any remove requests that might block on a locked * inode as that could lead to deadlock. */ if (num_on_worklist > max_softdeps / 10) { process_worklist_item(NULL, LK_NOWAIT); process_worklist_item(NULL, LK_NOWAIT); stat_worklist_push += 2; return(1); } /* * If we are resource constrained on inode dependencies, try * flushing some dirty inodes. Otherwise, we are constrained * by file deletions, so try accelerating flushes of directories * with removal dependencies. We would like to do the cleanup * here, but we probably hold an inode locked at this point and * that might deadlock against one that we try to clean. So, * the best that we can do is request the syncer daemon to do * the cleanup for us. */ switch (resource) { case FLUSH_INODES: stat_ino_limit_push += 1; req_clear_inodedeps += 1; stat_countp = &stat_ino_limit_hit; break; case FLUSH_REMOVE: stat_blk_limit_push += 1; req_clear_remove += 1; stat_countp = &stat_blk_limit_hit; break; default: panic("request_cleanup: unknown type"); } /* * Hopefully the syncer daemon will catch up and awaken us. * We wait at most tickdelay before proceeding in any case. */ lksleep(&proc_waiting, &lk, 0, "softupdate", tickdelay > 2 ? tickdelay : 2); return (1); } /* * Flush out a directory with at least one removal dependency in an effort to * reduce the number of dirrem, freefile, and freeblks dependency structures. */ static void clear_remove(struct thread *td) { struct pagedep_hashhead *pagedephd; struct pagedep *pagedep; static int next = 0; struct mount *mp; struct vnode *vp; int error, cnt; ino_t ino; ACQUIRE_LOCK(&lk); for (cnt = 0; cnt < pagedep_hash; cnt++) { pagedephd = &pagedep_hashtbl[next++]; if (next >= pagedep_hash) next = 0; LIST_FOREACH(pagedep, pagedephd, pd_hash) { if (LIST_FIRST(&pagedep->pd_dirremhd) == NULL) continue; mp = pagedep->pd_mnt; ino = pagedep->pd_ino; FREE_LOCK(&lk); if ((error = VFS_VGET(mp, NULL, ino, &vp)) != 0) { softdep_error("clear_remove: vget", error); return; } if ((error = VOP_FSYNC(vp, MNT_NOWAIT, 0))) softdep_error("clear_remove: fsync", error); drain_output(vp, 0); vput(vp); return; } } FREE_LOCK(&lk); } /* * Clear out a block of dirty inodes in an effort to reduce * the number of inodedep dependency structures. */ struct clear_inodedeps_info { struct fs *fs; struct mount *mp; }; static int clear_inodedeps_mountlist_callback(struct mount *mp, void *data) { struct clear_inodedeps_info *info = data; if ((mp->mnt_flag & MNT_SOFTDEP) && info->fs == VFSTOUFS(mp)->um_fs) { info->mp = mp; return(-1); } return(0); } static void clear_inodedeps(struct thread *td) { struct clear_inodedeps_info info; struct inodedep_hashhead *inodedephd; struct inodedep *inodedep; static int next = 0; struct vnode *vp; struct fs *fs; int error, cnt; ino_t firstino, lastino, ino; ACQUIRE_LOCK(&lk); /* * Pick a random inode dependency to be cleared. * We will then gather up all the inodes in its block * that have dependencies and flush them out. */ inodedep = NULL; /* avoid gcc warnings */ for (cnt = 0; cnt < inodedep_hash; cnt++) { inodedephd = &inodedep_hashtbl[next++]; if (next >= inodedep_hash) next = 0; if ((inodedep = LIST_FIRST(inodedephd)) != NULL) break; } if (inodedep == NULL) { FREE_LOCK(&lk); return; } /* * Ugly code to find mount point given pointer to superblock. */ fs = inodedep->id_fs; info.mp = NULL; info.fs = fs; mountlist_scan(clear_inodedeps_mountlist_callback, &info, MNTSCAN_FORWARD|MNTSCAN_NOBUSY); /* * Find the last inode in the block with dependencies. */ firstino = rounddown2(inodedep->id_ino, INOPB(fs)); for (lastino = firstino + INOPB(fs) - 1; lastino > firstino; lastino--) if (inodedep_lookup(fs, lastino, 0, &inodedep) != 0) break; /* * Asynchronously push all but the last inode with dependencies. * Synchronously push the last inode with dependencies to ensure * that the inode block gets written to free up the inodedeps. */ for (ino = firstino; ino <= lastino; ino++) { if (inodedep_lookup(fs, ino, 0, &inodedep) == 0) continue; FREE_LOCK(&lk); if ((error = VFS_VGET(info.mp, NULL, ino, &vp)) != 0) { softdep_error("clear_inodedeps: vget", error); return; } if (ino == lastino) { if ((error = VOP_FSYNC(vp, MNT_WAIT, 0))) softdep_error("clear_inodedeps: fsync1", error); } else { if ((error = VOP_FSYNC(vp, MNT_NOWAIT, 0))) softdep_error("clear_inodedeps: fsync2", error); drain_output(vp, 0); } vput(vp); ACQUIRE_LOCK(&lk); } FREE_LOCK(&lk); } /* * Function to determine if the buffer has outstanding dependencies * that will cause a roll-back if the buffer is written. If wantcount * is set, return number of dependencies, otherwise just yes or no. * * bioops callback - hold io_token */ static int softdep_count_dependencies(struct buf *bp, int wantcount) { struct worklist *wk; struct inodedep *inodedep; struct indirdep *indirdep; struct allocindir *aip; struct pagedep *pagedep; struct diradd *dap; int i, retval; retval = 0; ACQUIRE_LOCK(&lk); LIST_FOREACH(wk, &bp->b_dep, wk_list) { switch (wk->wk_type) { case D_INODEDEP: inodedep = WK_INODEDEP(wk); if ((inodedep->id_state & DEPCOMPLETE) == 0) { /* bitmap allocation dependency */ retval += 1; if (!wantcount) goto out; } if (TAILQ_FIRST(&inodedep->id_inoupdt)) { /* direct block pointer dependency */ retval += 1; if (!wantcount) goto out; } continue; case D_INDIRDEP: indirdep = WK_INDIRDEP(wk); LIST_FOREACH(aip, &indirdep->ir_deplisthd, ai_next) { /* indirect block pointer dependency */ retval += 1; if (!wantcount) goto out; } continue; case D_PAGEDEP: pagedep = WK_PAGEDEP(wk); for (i = 0; i < DAHASHSZ; i++) { LIST_FOREACH(dap, &pagedep->pd_diraddhd[i], da_pdlist) { /* directory entry dependency */ retval += 1; if (!wantcount) goto out; } } continue; case D_BMSAFEMAP: case D_ALLOCDIRECT: case D_ALLOCINDIR: case D_MKDIR: /* never a dependency on these blocks */ continue; default: panic("softdep_check_for_rollback: Unexpected type %s", TYPENAME(wk->wk_type)); /* NOTREACHED */ } } out: FREE_LOCK(&lk); return retval; } /* * Acquire exclusive access to a buffer. Requires softdep lock * to be held on entry. If waitfor is MNT_WAIT, may release/reacquire * softdep lock. * * Returns 1 if the buffer was locked, 0 if it was not locked or * if we had to block. * * NOTE! In order to return 1 we must acquire the buffer lock prior * to any release of &lk. Once we release &lk it's all over. * We may still have to block on the (type-stable) bp in that * case, but we must then unlock it and return 0. */ static int getdirtybuf(struct buf **bpp, int waitfor) { struct buf *bp; int error; /* * If the contents of *bpp is NULL the caller presumably lost a race. */ bp = *bpp; if (bp == NULL) return (0); /* * Try to obtain the buffer lock without deadlocking on &lk. */ KKASSERT(lock_held(&lk)); error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_NOWAIT); if (error == 0) { /* * If the buffer is no longer dirty the OS already wrote it * out, return failure. */ if ((bp->b_flags & B_DELWRI) == 0) { BUF_UNLOCK(bp); return (0); } /* * Finish nominal buffer locking sequence return success. * * Since we are not using a normal getblk(), and UFS * isn't KVABIO aware, we must make sure that the bp * is synchronized before returning it. */ bremfree(bp); bkvasync_all(bp); return (1); } /* * Failure case. * * If we are not being asked to wait, return 0 immediately. */ if (waitfor != MNT_WAIT) return (0); /* * Once we release the softdep lock we can never return success, * but we still have to block on the type-stable buf for the caller * to be able to retry without livelocking the system. * * The caller will normally retry in this case. */ FREE_LOCK(&lk); error = BUF_LOCK(bp, LK_EXCLUSIVE | LK_SLEEPFAIL); ACQUIRE_LOCK(&lk); if (error == 0) BUF_UNLOCK(bp); return (0); } /* * Wait for pending output on a vnode to complete. * Must be called with vnode locked. */ static void drain_output(struct vnode *vp, int islocked) { if (!islocked) ACQUIRE_LOCK(&lk); while (bio_track_active(&vp->v_track_write)) { FREE_LOCK(&lk); bio_track_wait(&vp->v_track_write, 0, 0); ACQUIRE_LOCK(&lk); } if (!islocked) FREE_LOCK(&lk); } /* * Called whenever a buffer that is being invalidated or reallocated * contains dependencies. This should only happen if an I/O error has * occurred. The routine is called with the buffer locked. * * bioops callback - hold io_token */ static void softdep_deallocate_dependencies(struct buf *bp) { /* nothing to do, mp lock not needed */ if ((bp->b_flags & B_ERROR) == 0) panic("softdep_deallocate_dependencies: dangling deps"); softdep_error(bp->b_vp->v_mount->mnt_stat.f_mntfromname, bp->b_error); panic("softdep_deallocate_dependencies: unrecovered I/O error"); } /* * Function to handle asynchronous write errors in the filesystem. */ void softdep_error(char *func, int error) { /* XXX should do something better! */ kprintf("%s: got error %d while accessing filesystem\n", func, error); } |