blob: dba5928c362b312a295be4b8eac800b7551d77cc [file] [log] [blame]
Willy Tarreau4f60f162007-04-08 16:39:58 +02001/*
2 * FD polling functions for generic select()
3 *
Willy Tarreau4d31fb22012-11-11 16:53:50 +01004 * Copyright 2000-2012 Willy Tarreau <w@1wt.eu>
Willy Tarreau4f60f162007-04-08 16:39:58 +02005 *
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>
Willy Tarreau0c303ee2008-07-07 00:09:58 +020019#include <common/ticks.h>
Willy Tarreau4f60f162007-04-08 16:39:58 +020020#include <common/time.h>
21
Willy Tarreau4f60f162007-04-08 16:39:58 +020022#include <types/global.h>
23
24#include <proto/fd.h>
Willy Tarreau332740d2009-05-10 09:57:21 +020025#include <proto/signal.h>
Willy Tarreau4f60f162007-04-08 16:39:58 +020026#include <proto/task.h>
27
28
Willy Tarreau28d86862007-04-08 17:42:27 +020029static fd_set *fd_evts[2];
30static fd_set *tmp_evts[2];
Willy Tarreau4f60f162007-04-08 16:39:58 +020031
Willy Tarreau4d31fb22012-11-11 16:53:50 +010032/* Immediately remove the entry upon close() */
33REGPRM1 static void __fd_clo(int fd)
Willy Tarreau4f60f162007-04-08 16:39:58 +020034{
Willy Tarreau28d86862007-04-08 17:42:27 +020035 FD_CLR(fd, fd_evts[DIR_RD]);
36 FD_CLR(fd, fd_evts[DIR_WR]);
Willy Tarreau4f60f162007-04-08 16:39:58 +020037}
38
Willy Tarreau4f60f162007-04-08 16:39:58 +020039/*
40 * Select() poller
41 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +020042REGPRM2 static void _do_poll(struct poller *p, int exp)
Willy Tarreau4f60f162007-04-08 16:39:58 +020043{
44 int status;
45 int fd, i;
46 struct timeval delta;
Willy Tarreaub0b37bc2008-06-23 14:00:57 +020047 int delta_ms;
Willy Tarreau4f60f162007-04-08 16:39:58 +020048 int readnotnull, writenotnull;
49 int fds;
Willy Tarreau4d31fb22012-11-11 16:53:50 +010050 int updt_idx, en, eo;
Willy Tarreau4f60f162007-04-08 16:39:58 +020051 char count;
52
Willy Tarreau4d31fb22012-11-11 16:53:50 +010053 /* first, scan the update list to find changes */
54 for (updt_idx = 0; updt_idx < fd_nbupdt; updt_idx++) {
55 fd = fd_updt[updt_idx];
56 en = fdtab[fd].spec_e & 15; /* new events */
57 eo = fdtab[fd].spec_e >> 4; /* previous events */
58
59 if (fdtab[fd].owner && (eo ^ en)) {
60 if ((eo ^ en) & FD_EV_POLLED_RW) {
61 /* poll status changed, update the lists */
62 if ((eo & ~en) & FD_EV_POLLED_R)
63 FD_CLR(fd, fd_evts[DIR_RD]);
64 else if ((en & ~eo) & FD_EV_POLLED_R)
65 FD_SET(fd, fd_evts[DIR_RD]);
66
67 if ((eo & ~en) & FD_EV_POLLED_W)
68 FD_CLR(fd, fd_evts[DIR_WR]);
69 else if ((en & ~eo) & FD_EV_POLLED_W)
70 FD_SET(fd, fd_evts[DIR_WR]);
71 }
72
73 fdtab[fd].spec_e = (en << 4) + en; /* save new events */
74
75 if (!(en & FD_EV_ACTIVE_RW)) {
76 /* This fd doesn't use any active entry anymore, we can
77 * kill its entry.
78 */
79 release_spec_entry(fd);
80 }
81 else if ((en & ~eo) & FD_EV_ACTIVE_RW) {
82 /* we need a new spec entry now */
83 alloc_spec_entry(fd);
84 }
85
86 }
87 fdtab[fd].updated = 0;
88 fdtab[fd].new = 0;
89 }
90 fd_nbupdt = 0;
91
Willy Tarreau0c303ee2008-07-07 00:09:58 +020092 delta_ms = 0;
93 delta.tv_sec = 0;
94 delta.tv_usec = 0;
95
Willy Tarreau4d31fb22012-11-11 16:53:50 +010096 if (!fd_nbspec && !run_queue && !signal_queue_len) {
Willy Tarreau0c303ee2008-07-07 00:09:58 +020097 if (!exp) {
98 delta_ms = MAX_DELAY_MS;
99 delta.tv_sec = (MAX_DELAY_MS / 1000);
100 delta.tv_usec = (MAX_DELAY_MS % 1000) * 1000;
Willy Tarreaub0b37bc2008-06-23 14:00:57 +0200101 }
Willy Tarreau0c303ee2008-07-07 00:09:58 +0200102 else if (!tick_is_expired(exp, now_ms)) {
103 delta_ms = TICKS_TO_MS(tick_remain(now_ms, exp)) + SCHEDULER_RESOLUTION;
104 if (delta_ms > MAX_DELAY_MS)
Willy Tarreaub0b37bc2008-06-23 14:00:57 +0200105 delta_ms = MAX_DELAY_MS;
Willy Tarreau0c303ee2008-07-07 00:09:58 +0200106 delta.tv_sec = (delta_ms / 1000);
107 delta.tv_usec = (delta_ms % 1000) * 1000;
Willy Tarreaud825eef2007-05-12 22:35:00 +0200108 }
Willy Tarreau4f60f162007-04-08 16:39:58 +0200109 }
110
111 /* let's restore fdset state */
112
113 readnotnull = 0; writenotnull = 0;
114 for (i = 0; i < (maxfd + FD_SETSIZE - 1)/(8*sizeof(int)); i++) {
Willy Tarreau28d86862007-04-08 17:42:27 +0200115 readnotnull |= (*(((int*)tmp_evts[DIR_RD])+i) = *(((int*)fd_evts[DIR_RD])+i)) != 0;
116 writenotnull |= (*(((int*)tmp_evts[DIR_WR])+i) = *(((int*)fd_evts[DIR_WR])+i)) != 0;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200117 }
118
119 // /* just a verification code, needs to be removed for performance */
120 // for (i=0; i<maxfd; i++) {
Willy Tarreau28d86862007-04-08 17:42:27 +0200121 // if (FD_ISSET(i, tmp_evts[DIR_RD]) != FD_ISSET(i, fd_evts[DIR_RD]))
Willy Tarreau4f60f162007-04-08 16:39:58 +0200122 // abort();
Willy Tarreau28d86862007-04-08 17:42:27 +0200123 // if (FD_ISSET(i, tmp_evts[DIR_WR]) != FD_ISSET(i, fd_evts[DIR_WR]))
Willy Tarreau4f60f162007-04-08 16:39:58 +0200124 // abort();
125 //
126 // }
127
Willy Tarreau45a12512011-09-10 16:56:42 +0200128 gettimeofday(&before_poll, NULL);
Willy Tarreau4f60f162007-04-08 16:39:58 +0200129 status = select(maxfd,
Willy Tarreau28d86862007-04-08 17:42:27 +0200130 readnotnull ? tmp_evts[DIR_RD] : NULL,
131 writenotnull ? tmp_evts[DIR_WR] : NULL,
Willy Tarreau4f60f162007-04-08 16:39:58 +0200132 NULL,
Willy Tarreaub0b37bc2008-06-23 14:00:57 +0200133 &delta);
Willy Tarreau4f60f162007-04-08 16:39:58 +0200134
Willy Tarreaub0b37bc2008-06-23 14:00:57 +0200135 tv_update_date(delta_ms, status);
Willy Tarreau45a12512011-09-10 16:56:42 +0200136 measure_idle();
Willy Tarreau4f60f162007-04-08 16:39:58 +0200137
138 if (status <= 0)
139 return;
140
Willy Tarreau177e2b02008-07-15 00:36:31 +0200141 for (fds = 0; (fds * BITS_PER_INT) < maxfd; fds++) {
Willy Tarreau28d86862007-04-08 17:42:27 +0200142 if ((((int *)(tmp_evts[DIR_RD]))[fds] | ((int *)(tmp_evts[DIR_WR]))[fds]) == 0)
Willy Tarreau4f60f162007-04-08 16:39:58 +0200143 continue;
144
Willy Tarreau177e2b02008-07-15 00:36:31 +0200145 for (count = BITS_PER_INT, fd = fds * BITS_PER_INT; count && fd < maxfd; count--, fd++) {
Willy Tarreau4f60f162007-04-08 16:39:58 +0200146 /* if we specify read first, the accepts and zero reads will be
147 * seen first. Moreover, system buffers will be flushed faster.
148 */
Willy Tarreau076be252012-07-06 16:02:29 +0200149 if (!fdtab[fd].owner)
150 continue;
151
Willy Tarreau9845e752012-07-06 11:44:28 +0200152 fdtab[fd].ev &= FD_POLL_STICKY;
Willy Tarreau076be252012-07-06 16:02:29 +0200153 if (FD_ISSET(fd, tmp_evts[DIR_RD]))
Willy Tarreau491c4982012-07-06 11:16:01 +0200154 fdtab[fd].ev |= FD_POLL_IN;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200155
Willy Tarreau076be252012-07-06 16:02:29 +0200156 if (FD_ISSET(fd, tmp_evts[DIR_WR]))
Willy Tarreau491c4982012-07-06 11:16:01 +0200157 fdtab[fd].ev |= FD_POLL_OUT;
Willy Tarreau9845e752012-07-06 11:44:28 +0200158
Willy Tarreaudb9cb0b2012-12-13 23:41:12 +0100159 if (fdtab[fd].iocb && fdtab[fd].ev) {
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100160 /* Mark the events as speculative before processing
161 * them so that if nothing can be done we don't need
162 * to poll again.
163 */
Willy Tarreau26d7cfc2012-12-07 00:09:43 +0100164 if (fdtab[fd].ev & FD_POLL_IN)
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100165 fd_ev_set(fd, DIR_RD);
166
Willy Tarreau26d7cfc2012-12-07 00:09:43 +0100167 if (fdtab[fd].ev & FD_POLL_OUT)
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100168 fd_ev_set(fd, DIR_WR);
169
Willy Tarreau39ebef82012-12-14 00:17:03 +0100170 if (fdtab[fd].spec_p) {
171 /* This fd was already scheduled for being
172 * called as a speculative I/O.
173 */
174 continue;
175 }
176
Willy Tarreau9845e752012-07-06 11:44:28 +0200177 fdtab[fd].iocb(fd);
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100178 }
Willy Tarreau4f60f162007-04-08 16:39:58 +0200179 }
180 }
181}
182
183/*
Willy Tarreaue54e9172007-04-09 09:23:31 +0200184 * Initialization of the select() poller.
185 * Returns 0 in case of failure, non-zero in case of success. If it fails, it
186 * disables the poller by setting its pref to 0.
187 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200188REGPRM1 static int _do_init(struct poller *p)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200189{
190 __label__ fail_swevt, fail_srevt, fail_wevt, fail_revt;
191 int fd_set_bytes;
192
193 p->private = NULL;
194 fd_set_bytes = sizeof(fd_set) * (global.maxsock + FD_SETSIZE - 1) / FD_SETSIZE;
195
196 if ((tmp_evts[DIR_RD] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
197 goto fail_revt;
198
199 if ((tmp_evts[DIR_WR] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
200 goto fail_wevt;
201
202 if ((fd_evts[DIR_RD] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
203 goto fail_srevt;
204
205 if ((fd_evts[DIR_WR] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
206 goto fail_swevt;
207
208 return 1;
209
210 fail_swevt:
211 free(fd_evts[DIR_RD]);
212 fail_srevt:
213 free(tmp_evts[DIR_WR]);
214 fail_wevt:
215 free(tmp_evts[DIR_RD]);
216 fail_revt:
217 p->pref = 0;
218 return 0;
219}
220
221/*
222 * Termination of the select() poller.
223 * Memory is released and the poller is marked as unselectable.
224 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200225REGPRM1 static void _do_term(struct poller *p)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200226{
Willy Tarreaua534fea2008-08-03 12:19:50 +0200227 free(fd_evts[DIR_WR]);
228 free(fd_evts[DIR_RD]);
229 free(tmp_evts[DIR_WR]);
230 free(tmp_evts[DIR_RD]);
Willy Tarreaue54e9172007-04-09 09:23:31 +0200231 p->private = NULL;
232 p->pref = 0;
233}
234
235/*
Willy Tarreau2ff76222007-04-09 19:29:56 +0200236 * Check that the poller works.
237 * Returns 1 if OK, otherwise 0.
238 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200239REGPRM1 static int _do_test(struct poller *p)
Willy Tarreau2ff76222007-04-09 19:29:56 +0200240{
241 return 1;
242}
243
244/*
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200245 * It is a constructor, which means that it will automatically be called before
246 * main(). This is GCC-specific but it works at least since 2.95.
247 * Special care must be taken so that it does not need any uninitialized data.
Willy Tarreau4f60f162007-04-08 16:39:58 +0200248 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200249__attribute__((constructor))
250static void _do_register(void)
Willy Tarreau4f60f162007-04-08 16:39:58 +0200251{
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200252 struct poller *p;
253
254 if (nbpollers >= MAX_POLLERS)
255 return;
256 p = &pollers[nbpollers++];
257
Willy Tarreau4f60f162007-04-08 16:39:58 +0200258 p->name = "select";
259 p->pref = 150;
260 p->private = NULL;
261
Willy Tarreau70c6fd82012-11-11 21:02:34 +0100262 p->clo = __fd_clo;
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200263 p->test = _do_test;
264 p->init = _do_init;
265 p->term = _do_term;
266 p->poll = _do_poll;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200267}
268
269
270/*
271 * Local variables:
272 * c-indent-level: 8
273 * c-basic-offset: 8
274 * End:
275 */