Willy Tarreau | 4f60f16 | 2007-04-08 16:39:58 +0200 | [diff] [blame] | 1 | /* |
| 2 | * FD polling functions for generic select() |
| 3 | * |
| 4 | * Copyright 2000-2007 Willy Tarreau <w@1wt.eu> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or |
| 7 | * modify it under the terms of the GNU General Public License |
| 8 | * as published by the Free Software Foundation; either version |
| 9 | * 2 of the License, or (at your option) any later version. |
| 10 | * |
| 11 | */ |
| 12 | |
| 13 | #include <unistd.h> |
| 14 | #include <sys/time.h> |
| 15 | #include <sys/types.h> |
| 16 | |
| 17 | #include <common/compat.h> |
| 18 | #include <common/config.h> |
| 19 | #include <common/time.h> |
| 20 | |
| 21 | #include <types/fd.h> |
| 22 | #include <types/global.h> |
| 23 | |
| 24 | #include <proto/fd.h> |
| 25 | #include <proto/polling.h> |
| 26 | #include <proto/task.h> |
| 27 | |
| 28 | |
| 29 | static fd_set *ReadEvent, *WriteEvent; |
| 30 | static fd_set *StaticReadEvent, *StaticWriteEvent; |
| 31 | |
| 32 | |
| 33 | /* |
| 34 | * Benchmarks performed on a Pentium-M notebook show that using functions |
| 35 | * instead of the usual macros improve the FD_* performance by about 80%, |
| 36 | * and that marking them regparm(2) adds another 20%. |
| 37 | */ |
| 38 | REGPRM2 static int __fd_isset(const int fd, const int dir) |
| 39 | { |
| 40 | fd_set *ev; |
| 41 | if (dir == DIR_RD) |
| 42 | ev = StaticReadEvent; |
| 43 | else |
| 44 | ev = StaticWriteEvent; |
| 45 | |
| 46 | return FD_ISSET(fd, ev); |
| 47 | } |
| 48 | |
| 49 | REGPRM2 static void __fd_set(const int fd, const int dir) |
| 50 | { |
| 51 | fd_set *ev; |
| 52 | if (dir == DIR_RD) |
| 53 | ev = StaticReadEvent; |
| 54 | else |
| 55 | ev = StaticWriteEvent; |
| 56 | |
| 57 | FD_SET(fd, ev); |
| 58 | } |
| 59 | |
| 60 | REGPRM2 static void __fd_clr(const int fd, const int dir) |
| 61 | { |
| 62 | fd_set *ev; |
| 63 | if (dir == DIR_RD) |
| 64 | ev = StaticReadEvent; |
| 65 | else |
| 66 | ev = StaticWriteEvent; |
| 67 | |
| 68 | FD_CLR(fd, ev); |
| 69 | } |
| 70 | |
| 71 | REGPRM2 static int __fd_cond_s(const int fd, const int dir) |
| 72 | { |
| 73 | int ret; |
| 74 | fd_set *ev; |
| 75 | if (dir == DIR_RD) |
| 76 | ev = StaticReadEvent; |
| 77 | else |
| 78 | ev = StaticWriteEvent; |
| 79 | |
| 80 | ret = !FD_ISSET(fd, ev); |
| 81 | if (ret) |
| 82 | FD_SET(fd, ev); |
| 83 | return ret; |
| 84 | } |
| 85 | |
| 86 | REGPRM2 static int __fd_cond_c(const int fd, const int dir) |
| 87 | { |
| 88 | int ret; |
| 89 | fd_set *ev; |
| 90 | if (dir == DIR_RD) |
| 91 | ev = StaticReadEvent; |
| 92 | else |
| 93 | ev = StaticWriteEvent; |
| 94 | |
| 95 | ret = FD_ISSET(fd, ev); |
| 96 | if (ret) |
| 97 | FD_CLR(fd, ev); |
| 98 | return ret; |
| 99 | } |
| 100 | |
| 101 | REGPRM1 static void __fd_rem(const int fd) |
| 102 | { |
| 103 | FD_CLR(fd, StaticReadEvent); |
| 104 | FD_CLR(fd, StaticWriteEvent); |
| 105 | } |
| 106 | |
| 107 | |
| 108 | /* |
| 109 | * Initialization of the select() poller. |
| 110 | * Returns 0 in case of failure, non-zero in case of success. If it fails, it |
| 111 | * disables the poller by setting its pref to 0. |
| 112 | */ |
| 113 | REGPRM1 static int select_init(struct poller *p) |
| 114 | { |
| 115 | __label__ fail_swevt, fail_srevt, fail_wevt, fail_revt; |
| 116 | int fd_set_bytes; |
| 117 | |
| 118 | p->private = NULL; |
| 119 | fd_set_bytes = sizeof(fd_set) * (global.maxsock + FD_SETSIZE - 1) / FD_SETSIZE; |
| 120 | |
| 121 | if ((ReadEvent = (fd_set *)calloc(1, fd_set_bytes)) == NULL) |
| 122 | goto fail_revt; |
| 123 | |
| 124 | if ((WriteEvent = (fd_set *)calloc(1, fd_set_bytes)) == NULL) |
| 125 | goto fail_wevt; |
| 126 | |
| 127 | if ((StaticReadEvent = (fd_set *)calloc(1, fd_set_bytes)) == NULL) |
| 128 | goto fail_srevt; |
| 129 | |
| 130 | if ((StaticWriteEvent = (fd_set *)calloc(1, fd_set_bytes)) == NULL) |
| 131 | goto fail_swevt; |
| 132 | |
| 133 | return 1; |
| 134 | |
| 135 | fail_swevt: |
| 136 | free(StaticReadEvent); |
| 137 | fail_srevt: |
| 138 | free(WriteEvent); |
| 139 | fail_wevt: |
| 140 | free(ReadEvent); |
| 141 | fail_revt: |
| 142 | p->pref = 0; |
| 143 | return 0; |
| 144 | } |
| 145 | |
| 146 | /* |
| 147 | * Termination of the select() poller. |
| 148 | * Memory is released and the poller is marked as unselectable. |
| 149 | */ |
| 150 | REGPRM1 static void select_term(struct poller *p) |
| 151 | { |
| 152 | if (StaticWriteEvent) |
| 153 | free(StaticWriteEvent); |
| 154 | if (StaticReadEvent) |
| 155 | free(StaticReadEvent); |
| 156 | if (WriteEvent) |
| 157 | free(WriteEvent); |
| 158 | if (ReadEvent) |
| 159 | free(ReadEvent); |
| 160 | p->private = NULL; |
| 161 | p->pref = 0; |
| 162 | } |
| 163 | |
| 164 | /* |
| 165 | * Select() poller |
| 166 | */ |
| 167 | REGPRM2 static void select_poll(struct poller *p, int wait_time) |
| 168 | { |
| 169 | int status; |
| 170 | int fd, i; |
| 171 | struct timeval delta; |
| 172 | int readnotnull, writenotnull; |
| 173 | int fds; |
| 174 | char count; |
| 175 | |
| 176 | /* allow select to return immediately when needed */ |
| 177 | delta.tv_sec = delta.tv_usec = 0; |
| 178 | if (wait_time > 0) { /* FIXME */ |
| 179 | /* Convert to timeval */ |
| 180 | /* to avoid eventual select loops due to timer precision */ |
| 181 | wait_time += SCHEDULER_RESOLUTION; |
| 182 | delta.tv_sec = wait_time / 1000; |
| 183 | delta.tv_usec = (wait_time % 1000) * 1000; |
| 184 | } |
| 185 | |
| 186 | /* let's restore fdset state */ |
| 187 | |
| 188 | readnotnull = 0; writenotnull = 0; |
| 189 | for (i = 0; i < (maxfd + FD_SETSIZE - 1)/(8*sizeof(int)); i++) { |
| 190 | readnotnull |= (*(((int*)ReadEvent)+i) = *(((int*)StaticReadEvent)+i)) != 0; |
| 191 | writenotnull |= (*(((int*)WriteEvent)+i) = *(((int*)StaticWriteEvent)+i)) != 0; |
| 192 | } |
| 193 | |
| 194 | // /* just a verification code, needs to be removed for performance */ |
| 195 | // for (i=0; i<maxfd; i++) { |
| 196 | // if (FD_ISSET(i, ReadEvent) != FD_ISSET(i, StaticReadEvent)) |
| 197 | // abort(); |
| 198 | // if (FD_ISSET(i, WriteEvent) != FD_ISSET(i, StaticWriteEvent)) |
| 199 | // abort(); |
| 200 | // |
| 201 | // } |
| 202 | |
| 203 | status = select(maxfd, |
| 204 | readnotnull ? ReadEvent : NULL, |
| 205 | writenotnull ? WriteEvent : NULL, |
| 206 | NULL, |
| 207 | (wait_time >= 0) ? &delta : NULL); |
| 208 | |
| 209 | tv_now(&now); |
| 210 | |
| 211 | if (status <= 0) |
| 212 | return; |
| 213 | |
| 214 | for (fds = 0; (fds << INTBITS) < maxfd; fds++) { |
| 215 | if ((((int *)(ReadEvent))[fds] | ((int *)(WriteEvent))[fds]) == 0) |
| 216 | continue; |
| 217 | |
| 218 | for (count = 1<<INTBITS, fd = fds << INTBITS; count && fd < maxfd; count--, fd++) { |
| 219 | /* if we specify read first, the accepts and zero reads will be |
| 220 | * seen first. Moreover, system buffers will be flushed faster. |
| 221 | */ |
| 222 | if (FD_ISSET(fd, ReadEvent)) { |
| 223 | if (fdtab[fd].state == FD_STCLOSE) |
| 224 | continue; |
| 225 | fdtab[fd].cb[DIR_RD].f(fd); |
| 226 | } |
| 227 | |
| 228 | if (FD_ISSET(fd, WriteEvent)) { |
| 229 | if (fdtab[fd].state == FD_STCLOSE) |
| 230 | continue; |
| 231 | fdtab[fd].cb[DIR_WR].f(fd); |
| 232 | } |
| 233 | } |
| 234 | } |
| 235 | } |
| 236 | |
| 237 | /* |
| 238 | * The only exported function. Returns 1. |
| 239 | */ |
| 240 | int select_register(struct poller *p) |
| 241 | { |
| 242 | p->name = "select"; |
| 243 | p->pref = 150; |
| 244 | p->private = NULL; |
| 245 | |
| 246 | p->init = select_init; |
| 247 | p->term = select_term; |
| 248 | p->poll = select_poll; |
| 249 | p->isset = __fd_isset; |
| 250 | p->set = __fd_set; |
| 251 | p->clr = __fd_clr; |
| 252 | p->clo = p->rem = __fd_rem; |
| 253 | p->cond_s = __fd_cond_s; |
| 254 | p->cond_c = __fd_cond_c; |
| 255 | return 1; |
| 256 | } |
| 257 | |
| 258 | |
| 259 | /* |
| 260 | * Local variables: |
| 261 | * c-indent-level: 8 |
| 262 | * c-basic-offset: 8 |
| 263 | * End: |
| 264 | */ |