blob: 3591f1eb1f134b649c2b9e2cbeb3df9a844f7c37 [file] [log] [blame]
Willy Tarreau4f60f162007-04-08 16:39:58 +02001/*
2 * FD polling functions for generic select()
3 *
Willy Tarreauf817e9f2014-01-10 16:58:45 +01004 * Copyright 2000-2014 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 Tarreau60b639c2018-08-02 10:16:17 +020019#include <common/hathreads.h>
Willy Tarreau0c303ee2008-07-07 00:09:58 +020020#include <common/ticks.h>
Willy Tarreau4f60f162007-04-08 16:39:58 +020021#include <common/time.h>
22
Willy Tarreau4f60f162007-04-08 16:39:58 +020023#include <types/global.h>
24
Willy Tarreau609aad92018-11-22 08:31:09 +010025#include <proto/activity.h>
Willy Tarreau4f60f162007-04-08 16:39:58 +020026#include <proto/fd.h>
Willy Tarreau4f60f162007-04-08 16:39:58 +020027
28
Christopher Fauletd4604ad2017-05-29 10:40:41 +020029/* private data */
Willy Tarreau173d9952018-01-26 21:48:23 +010030static int maxfd; /* # of the highest fd + 1 */
Willy Tarreaud51a5072018-01-25 16:48:46 +010031static unsigned int *fd_evts[2];
Christopher Fauletd4604ad2017-05-29 10:40:41 +020032static THREAD_LOCAL fd_set *tmp_evts[2];
Willy Tarreau4f60f162007-04-08 16:39:58 +020033
Willy Tarreau4d31fb22012-11-11 16:53:50 +010034/* Immediately remove the entry upon close() */
35REGPRM1 static void __fd_clo(int fd)
Willy Tarreau4f60f162007-04-08 16:39:58 +020036{
Willy Tarreaud51a5072018-01-25 16:48:46 +010037 hap_fd_clr(fd, fd_evts[DIR_RD]);
38 hap_fd_clr(fd, fd_evts[DIR_WR]);
Willy Tarreau4f60f162007-04-08 16:39:58 +020039}
40
Olivier Houchard6b96f722018-04-25 16:58:25 +020041static void _update_fd(int fd, int *max_add_fd)
42{
43 int en;
44
45 en = fdtab[fd].state;
46
47 /* we have a single state for all threads, which is why we
48 * don't check the tid_bit. First thread to see the update
49 * takes it for every other one.
50 */
51 if (!(en & FD_EV_POLLED_RW)) {
Olivier Houchardcb92f5c2018-04-26 14:23:07 +020052 if (!polled_mask[fd]) {
Olivier Houchard6b96f722018-04-25 16:58:25 +020053 /* fd was not watched, it's still not */
54 return;
55 }
56 /* fd totally removed from poll list */
57 hap_fd_clr(fd, fd_evts[DIR_RD]);
58 hap_fd_clr(fd, fd_evts[DIR_WR]);
Olivier Houchardcb6c9272019-03-08 18:49:54 +010059 _HA_ATOMIC_AND(&polled_mask[fd], 0);
Olivier Houchard6b96f722018-04-25 16:58:25 +020060 }
61 else {
62 /* OK fd has to be monitored, it was either added or changed */
63 if (!(en & FD_EV_POLLED_R))
64 hap_fd_clr(fd, fd_evts[DIR_RD]);
65 else
66 hap_fd_set(fd, fd_evts[DIR_RD]);
67
68 if (!(en & FD_EV_POLLED_W))
69 hap_fd_clr(fd, fd_evts[DIR_WR]);
70 else
71 hap_fd_set(fd, fd_evts[DIR_WR]);
72
Olivier Houchardcb6c9272019-03-08 18:49:54 +010073 _HA_ATOMIC_OR(&polled_mask[fd], tid_bit);
Olivier Houchard6b96f722018-04-25 16:58:25 +020074 if (fd > *max_add_fd)
75 *max_add_fd = fd;
76 }
77}
78
Willy Tarreau4f60f162007-04-08 16:39:58 +020079/*
80 * Select() poller
81 */
Willy Tarreau0c303ee2008-07-07 00:09:58 +020082REGPRM2 static void _do_poll(struct poller *p, int exp)
Willy Tarreau4f60f162007-04-08 16:39:58 +020083{
84 int status;
85 int fd, i;
86 struct timeval delta;
Willy Tarreaub0b37bc2008-06-23 14:00:57 +020087 int delta_ms;
Willy Tarreau4f60f162007-04-08 16:39:58 +020088 int fds;
Olivier Houchard6b96f722018-04-25 16:58:25 +020089 int updt_idx;
Willy Tarreau4f60f162007-04-08 16:39:58 +020090 char count;
Christopher Fauletd4604ad2017-05-29 10:40:41 +020091 int readnotnull, writenotnull;
Willy Tarreau173d9952018-01-26 21:48:23 +010092 int old_maxfd, new_maxfd, max_add_fd;
Olivier Houchard6b96f722018-04-25 16:58:25 +020093 int old_fd;
Willy Tarreau173d9952018-01-26 21:48:23 +010094
95 max_add_fd = -1;
Christopher Fauletd4604ad2017-05-29 10:40:41 +020096
Willy Tarreau4d31fb22012-11-11 16:53:50 +010097 /* first, scan the update list to find changes */
98 for (updt_idx = 0; updt_idx < fd_nbupdt; updt_idx++) {
99 fd = fd_updt[updt_idx];
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100100
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100101 _HA_ATOMIC_AND(&fdtab[fd].update_mask, ~tid_bit);
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100102 if (!fdtab[fd].owner) {
103 activity[tid].poll_drop++;
Willy Tarreauf817e9f2014-01-10 16:58:45 +0100104 continue;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100105 }
Olivier Houchard6b96f722018-04-25 16:58:25 +0200106 _update_fd(fd, &max_add_fd);
107 }
108 /* Now scan the global update list */
109 for (old_fd = fd = update_list.first; fd != -1; fd = fdtab[fd].update.next) {
110 if (fd == -2) {
111 fd = old_fd;
112 continue;
Willy Tarreauf817e9f2014-01-10 16:58:45 +0100113 }
Olivier Houchard6b96f722018-04-25 16:58:25 +0200114 else if (fd <= -3)
115 fd = -fd -4;
116 if (fd == -1)
117 break;
118 if (fdtab[fd].update_mask & tid_bit) {
119 /* Cheat a bit, as the state is global to all pollers
120 * we don't need every thread ot take care of the
121 * update.
122 */
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100123 _HA_ATOMIC_AND(&fdtab[fd].update_mask, ~all_threads_mask);
Olivier Houchard6b96f722018-04-25 16:58:25 +0200124 done_update_polling(fd);
125 } else
126 continue;
127 if (!fdtab[fd].owner)
128 continue;
129 _update_fd(fd, &max_add_fd);
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100130 }
Willy Tarreau173d9952018-01-26 21:48:23 +0100131
Olivier Houchard6b96f722018-04-25 16:58:25 +0200132
Willy Tarreau173d9952018-01-26 21:48:23 +0100133 /* maybe we added at least one fd larger than maxfd */
134 for (old_maxfd = maxfd; old_maxfd <= max_add_fd; ) {
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100135 if (_HA_ATOMIC_CAS(&maxfd, &old_maxfd, max_add_fd + 1))
Willy Tarreau173d9952018-01-26 21:48:23 +0100136 break;
137 }
138
139 /* maxfd doesn't need to be precise but it needs to cover *all* active
140 * FDs. Thus we only shrink it if we have such an opportunity. The algo
141 * is simple : look for the previous used place, try to update maxfd to
142 * point to it, abort if maxfd changed in the mean time.
143 */
144 old_maxfd = maxfd;
145 do {
146 new_maxfd = old_maxfd;
147 while (new_maxfd - 1 >= 0 && !fdtab[new_maxfd - 1].owner)
148 new_maxfd--;
149 if (new_maxfd >= old_maxfd)
150 break;
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100151 } while (!_HA_ATOMIC_CAS(&maxfd, &old_maxfd, new_maxfd));
Willy Tarreau173d9952018-01-26 21:48:23 +0100152
Willy Tarreau60b639c2018-08-02 10:16:17 +0200153 thread_harmless_now();
154
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100155 fd_nbupdt = 0;
156
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200157 /* let's restore fdset state */
158 readnotnull = 0; writenotnull = 0;
159 for (i = 0; i < (maxfd + FD_SETSIZE - 1)/(8*sizeof(int)); i++) {
160 readnotnull |= (*(((int*)tmp_evts[DIR_RD])+i) = *(((int*)fd_evts[DIR_RD])+i)) != 0;
161 writenotnull |= (*(((int*)tmp_evts[DIR_WR])+i) = *(((int*)fd_evts[DIR_WR])+i)) != 0;
162 }
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200163
Willy Tarreauf37ba942018-10-17 11:25:54 +0200164 /* now let's wait for events */
165 delta_ms = compute_poll_timeout(exp);
166 delta.tv_sec = (delta_ms / 1000);
167 delta.tv_usec = (delta_ms % 1000) * 1000;
Willy Tarreau7e9c4ae2018-10-17 14:31:19 +0200168 tv_entering_poll();
Willy Tarreau609aad92018-11-22 08:31:09 +0100169 activity_count_runtime();
Willy Tarreau4f60f162007-04-08 16:39:58 +0200170 status = select(maxfd,
Willy Tarreau28d86862007-04-08 17:42:27 +0200171 readnotnull ? tmp_evts[DIR_RD] : NULL,
172 writenotnull ? tmp_evts[DIR_WR] : NULL,
Willy Tarreau4f60f162007-04-08 16:39:58 +0200173 NULL,
Willy Tarreaub0b37bc2008-06-23 14:00:57 +0200174 &delta);
Willy Tarreau48f8bc12018-11-22 18:57:37 +0100175 tv_update_date(delta_ms, status);
Willy Tarreau7e9c4ae2018-10-17 14:31:19 +0200176 tv_leaving_poll(delta_ms, status);
Willy Tarreau4f60f162007-04-08 16:39:58 +0200177
Willy Tarreau60b639c2018-08-02 10:16:17 +0200178 thread_harmless_end();
179
Willy Tarreau4f60f162007-04-08 16:39:58 +0200180 if (status <= 0)
181 return;
182
Willy Tarreau177e2b02008-07-15 00:36:31 +0200183 for (fds = 0; (fds * BITS_PER_INT) < maxfd; fds++) {
Willy Tarreau28d86862007-04-08 17:42:27 +0200184 if ((((int *)(tmp_evts[DIR_RD]))[fds] | ((int *)(tmp_evts[DIR_WR]))[fds]) == 0)
Willy Tarreau4f60f162007-04-08 16:39:58 +0200185 continue;
186
Willy Tarreau177e2b02008-07-15 00:36:31 +0200187 for (count = BITS_PER_INT, fd = fds * BITS_PER_INT; count && fd < maxfd; count--, fd++) {
Christopher Fauletab62f512017-08-30 10:34:36 +0200188 unsigned int n = 0;
189
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100190 if (!fdtab[fd].owner) {
191 activity[tid].poll_dead++;
192 continue;
193 }
194
195 if (!(fdtab[fd].thread_mask & tid_bit)) {
196 activity[tid].poll_skip++;
Willy Tarreau076be252012-07-06 16:02:29 +0200197 continue;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100198 }
Willy Tarreau076be252012-07-06 16:02:29 +0200199
Willy Tarreau076be252012-07-06 16:02:29 +0200200 if (FD_ISSET(fd, tmp_evts[DIR_RD]))
Christopher Fauletab62f512017-08-30 10:34:36 +0200201 n |= FD_POLL_IN;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200202
Willy Tarreau076be252012-07-06 16:02:29 +0200203 if (FD_ISSET(fd, tmp_evts[DIR_WR]))
Christopher Fauletab62f512017-08-30 10:34:36 +0200204 n |= FD_POLL_OUT;
Willy Tarreau5be2f352014-11-19 19:43:05 +0100205
Christopher Fauletab62f512017-08-30 10:34:36 +0200206 fd_update_events(fd, n);
Willy Tarreau4f60f162007-04-08 16:39:58 +0200207 }
208 }
209}
210
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200211static int init_select_per_thread()
212{
213 int fd_set_bytes;
214
215 fd_set_bytes = sizeof(fd_set) * (global.maxsock + FD_SETSIZE - 1) / FD_SETSIZE;
216 if ((tmp_evts[DIR_RD] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
217 goto fail;
218 if ((tmp_evts[DIR_WR] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
219 goto fail;
220 return 1;
221 fail:
222 free(tmp_evts[DIR_RD]);
223 free(tmp_evts[DIR_WR]);
224 return 0;
225}
226
227static void deinit_select_per_thread()
228{
Christopher Fauletcd7879a2017-10-27 13:53:47 +0200229 free(tmp_evts[DIR_WR]); tmp_evts[DIR_WR] = NULL;
230 free(tmp_evts[DIR_RD]); tmp_evts[DIR_RD] = NULL;
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200231}
232
Willy Tarreau4f60f162007-04-08 16:39:58 +0200233/*
Willy Tarreaue54e9172007-04-09 09:23:31 +0200234 * Initialization of the select() poller.
235 * Returns 0 in case of failure, non-zero in case of success. If it fails, it
236 * disables the poller by setting its pref to 0.
237 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200238REGPRM1 static int _do_init(struct poller *p)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200239{
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200240 __label__ fail_swevt, fail_srevt, fail_revt;
Willy Tarreaue54e9172007-04-09 09:23:31 +0200241 int fd_set_bytes;
242
243 p->private = NULL;
Willy Tarreau3fa87b12013-03-31 14:41:15 +0200244
245 if (global.maxsock > FD_SETSIZE)
246 goto fail_revt;
247
Willy Tarreaue54e9172007-04-09 09:23:31 +0200248 fd_set_bytes = sizeof(fd_set) * (global.maxsock + FD_SETSIZE - 1) / FD_SETSIZE;
Willy Tarreaue54e9172007-04-09 09:23:31 +0200249
Willy Tarreaud51a5072018-01-25 16:48:46 +0100250 if ((fd_evts[DIR_RD] = calloc(1, fd_set_bytes)) == NULL)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200251 goto fail_srevt;
Willy Tarreaud51a5072018-01-25 16:48:46 +0100252 if ((fd_evts[DIR_WR] = calloc(1, fd_set_bytes)) == NULL)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200253 goto fail_swevt;
254
Christopher Fauletcd7879a2017-10-27 13:53:47 +0200255 hap_register_per_thread_init(init_select_per_thread);
256 hap_register_per_thread_deinit(deinit_select_per_thread);
257
Willy Tarreaue54e9172007-04-09 09:23:31 +0200258 return 1;
259
260 fail_swevt:
261 free(fd_evts[DIR_RD]);
262 fail_srevt:
263 free(tmp_evts[DIR_WR]);
Willy Tarreaue54e9172007-04-09 09:23:31 +0200264 free(tmp_evts[DIR_RD]);
265 fail_revt:
266 p->pref = 0;
267 return 0;
268}
269
270/*
271 * Termination of the select() poller.
272 * Memory is released and the poller is marked as unselectable.
273 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200274REGPRM1 static void _do_term(struct poller *p)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200275{
Willy Tarreaua534fea2008-08-03 12:19:50 +0200276 free(fd_evts[DIR_WR]);
277 free(fd_evts[DIR_RD]);
Willy Tarreaue54e9172007-04-09 09:23:31 +0200278 p->private = NULL;
279 p->pref = 0;
280}
281
282/*
Willy Tarreau2ff76222007-04-09 19:29:56 +0200283 * Check that the poller works.
284 * Returns 1 if OK, otherwise 0.
285 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200286REGPRM1 static int _do_test(struct poller *p)
Willy Tarreau2ff76222007-04-09 19:29:56 +0200287{
Willy Tarreau3fa87b12013-03-31 14:41:15 +0200288 if (global.maxsock > FD_SETSIZE)
289 return 0;
290
Willy Tarreau2ff76222007-04-09 19:29:56 +0200291 return 1;
292}
293
294/*
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200295 * It is a constructor, which means that it will automatically be called before
296 * main(). This is GCC-specific but it works at least since 2.95.
297 * Special care must be taken so that it does not need any uninitialized data.
Willy Tarreau4f60f162007-04-08 16:39:58 +0200298 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200299__attribute__((constructor))
300static void _do_register(void)
Willy Tarreau4f60f162007-04-08 16:39:58 +0200301{
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200302 struct poller *p;
303
304 if (nbpollers >= MAX_POLLERS)
305 return;
306 p = &pollers[nbpollers++];
307
Willy Tarreau4f60f162007-04-08 16:39:58 +0200308 p->name = "select";
309 p->pref = 150;
Willy Tarreau5a767692017-03-13 11:38:28 +0100310 p->flags = 0;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200311 p->private = NULL;
312
Willy Tarreau70c6fd82012-11-11 21:02:34 +0100313 p->clo = __fd_clo;
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200314 p->test = _do_test;
315 p->init = _do_init;
316 p->term = _do_term;
317 p->poll = _do_poll;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200318}
319
320
321/*
322 * Local variables:
323 * c-indent-level: 8
324 * c-basic-offset: 8
325 * End:
326 */