blob: ab021b97f78e5ee793e478d151d5d1e9c466095d [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() */
Willy Tarreau03e78532020-02-25 07:38:05 +010035static 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 */
Willy Tarreau5bee3e22019-09-04 09:52:57 +020051 if (!(en & FD_EV_ACTIVE_RW)) {
Olivier Houchard53055052019-07-25 14:00:18 +000052 if (!(polled_mask[fd].poll_recv | polled_mask[fd].poll_send)) {
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 Houchard53055052019-07-25 14:00:18 +000059 _HA_ATOMIC_AND(&polled_mask[fd].poll_recv, 0);
Olivier Houcharda3a8ea22019-08-05 23:54:37 +020060 _HA_ATOMIC_AND(&polled_mask[fd].poll_send, 0);
Olivier Houchard6b96f722018-04-25 16:58:25 +020061 }
62 else {
63 /* OK fd has to be monitored, it was either added or changed */
Willy Tarreau5bee3e22019-09-04 09:52:57 +020064 if (!(en & FD_EV_ACTIVE_R)) {
Olivier Houchard6b96f722018-04-25 16:58:25 +020065 hap_fd_clr(fd, fd_evts[DIR_RD]);
Olivier Houchard53055052019-07-25 14:00:18 +000066 if (polled_mask[fd].poll_recv & tid_bit)
67 _HA_ATOMIC_AND(&polled_mask[fd].poll_recv, ~tid_bit);
68 } else {
Olivier Houchard6b96f722018-04-25 16:58:25 +020069 hap_fd_set(fd, fd_evts[DIR_RD]);
Olivier Houchard53055052019-07-25 14:00:18 +000070 if (!(polled_mask[fd].poll_recv & tid_bit))
71 _HA_ATOMIC_OR(&polled_mask[fd].poll_recv, tid_bit);
72 }
Olivier Houchard6b96f722018-04-25 16:58:25 +020073
Willy Tarreau5bee3e22019-09-04 09:52:57 +020074 if (!(en & FD_EV_ACTIVE_W)) {
Olivier Houchard6b96f722018-04-25 16:58:25 +020075 hap_fd_clr(fd, fd_evts[DIR_WR]);
Olivier Houchard53055052019-07-25 14:00:18 +000076 if (polled_mask[fd].poll_send & tid_bit)
77 _HA_ATOMIC_AND(&polled_mask[fd].poll_send, ~tid_bit);
78 } else {
Olivier Houchard6b96f722018-04-25 16:58:25 +020079 hap_fd_set(fd, fd_evts[DIR_WR]);
Olivier Houchard53055052019-07-25 14:00:18 +000080 if (!(polled_mask[fd].poll_send & tid_bit))
81 _HA_ATOMIC_OR(&polled_mask[fd].poll_send, tid_bit);
82 }
Olivier Houchard6b96f722018-04-25 16:58:25 +020083
Olivier Houchard6b96f722018-04-25 16:58:25 +020084 if (fd > *max_add_fd)
85 *max_add_fd = fd;
86 }
87}
88
Willy Tarreau4f60f162007-04-08 16:39:58 +020089/*
90 * Select() poller
91 */
Willy Tarreau03e78532020-02-25 07:38:05 +010092static void _do_poll(struct poller *p, int exp, int wake)
Willy Tarreau4f60f162007-04-08 16:39:58 +020093{
94 int status;
95 int fd, i;
96 struct timeval delta;
Willy Tarreaub0b37bc2008-06-23 14:00:57 +020097 int delta_ms;
Willy Tarreau4f60f162007-04-08 16:39:58 +020098 int fds;
Olivier Houchard6b96f722018-04-25 16:58:25 +020099 int updt_idx;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200100 char count;
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200101 int readnotnull, writenotnull;
Willy Tarreau173d9952018-01-26 21:48:23 +0100102 int old_maxfd, new_maxfd, max_add_fd;
Olivier Houchard6b96f722018-04-25 16:58:25 +0200103 int old_fd;
Willy Tarreau173d9952018-01-26 21:48:23 +0100104
105 max_add_fd = -1;
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200106
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100107 /* first, scan the update list to find changes */
108 for (updt_idx = 0; updt_idx < fd_nbupdt; updt_idx++) {
109 fd = fd_updt[updt_idx];
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100110
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100111 _HA_ATOMIC_AND(&fdtab[fd].update_mask, ~tid_bit);
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100112 if (!fdtab[fd].owner) {
113 activity[tid].poll_drop++;
Willy Tarreauf817e9f2014-01-10 16:58:45 +0100114 continue;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100115 }
Olivier Houchard6b96f722018-04-25 16:58:25 +0200116 _update_fd(fd, &max_add_fd);
117 }
118 /* Now scan the global update list */
119 for (old_fd = fd = update_list.first; fd != -1; fd = fdtab[fd].update.next) {
120 if (fd == -2) {
121 fd = old_fd;
122 continue;
Willy Tarreauf817e9f2014-01-10 16:58:45 +0100123 }
Olivier Houchard6b96f722018-04-25 16:58:25 +0200124 else if (fd <= -3)
125 fd = -fd -4;
126 if (fd == -1)
127 break;
128 if (fdtab[fd].update_mask & tid_bit) {
129 /* Cheat a bit, as the state is global to all pollers
130 * we don't need every thread ot take care of the
131 * update.
132 */
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100133 _HA_ATOMIC_AND(&fdtab[fd].update_mask, ~all_threads_mask);
Olivier Houchard6b96f722018-04-25 16:58:25 +0200134 done_update_polling(fd);
135 } else
136 continue;
137 if (!fdtab[fd].owner)
138 continue;
139 _update_fd(fd, &max_add_fd);
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100140 }
Willy Tarreau173d9952018-01-26 21:48:23 +0100141
Olivier Houchard6b96f722018-04-25 16:58:25 +0200142
Willy Tarreau173d9952018-01-26 21:48:23 +0100143 /* maybe we added at least one fd larger than maxfd */
144 for (old_maxfd = maxfd; old_maxfd <= max_add_fd; ) {
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100145 if (_HA_ATOMIC_CAS(&maxfd, &old_maxfd, max_add_fd + 1))
Willy Tarreau173d9952018-01-26 21:48:23 +0100146 break;
147 }
148
149 /* maxfd doesn't need to be precise but it needs to cover *all* active
150 * FDs. Thus we only shrink it if we have such an opportunity. The algo
151 * is simple : look for the previous used place, try to update maxfd to
152 * point to it, abort if maxfd changed in the mean time.
153 */
154 old_maxfd = maxfd;
155 do {
156 new_maxfd = old_maxfd;
157 while (new_maxfd - 1 >= 0 && !fdtab[new_maxfd - 1].owner)
158 new_maxfd--;
159 if (new_maxfd >= old_maxfd)
160 break;
Olivier Houchardcb6c9272019-03-08 18:49:54 +0100161 } while (!_HA_ATOMIC_CAS(&maxfd, &old_maxfd, new_maxfd));
Willy Tarreau173d9952018-01-26 21:48:23 +0100162
Willy Tarreau60b639c2018-08-02 10:16:17 +0200163 thread_harmless_now();
164
Willy Tarreau4d31fb22012-11-11 16:53:50 +0100165 fd_nbupdt = 0;
166
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200167 /* let's restore fdset state */
168 readnotnull = 0; writenotnull = 0;
169 for (i = 0; i < (maxfd + FD_SETSIZE - 1)/(8*sizeof(int)); i++) {
170 readnotnull |= (*(((int*)tmp_evts[DIR_RD])+i) = *(((int*)fd_evts[DIR_RD])+i)) != 0;
171 writenotnull |= (*(((int*)tmp_evts[DIR_WR])+i) = *(((int*)fd_evts[DIR_WR])+i)) != 0;
172 }
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200173
Willy Tarreauf37ba942018-10-17 11:25:54 +0200174 /* now let's wait for events */
Willy Tarreau2ae84e42019-05-28 16:44:05 +0200175 delta_ms = wake ? 0 : compute_poll_timeout(exp);
Willy Tarreauf37ba942018-10-17 11:25:54 +0200176 delta.tv_sec = (delta_ms / 1000);
177 delta.tv_usec = (delta_ms % 1000) * 1000;
Willy Tarreau7e9c4ae2018-10-17 14:31:19 +0200178 tv_entering_poll();
Willy Tarreau609aad92018-11-22 08:31:09 +0100179 activity_count_runtime();
Willy Tarreau4f60f162007-04-08 16:39:58 +0200180 status = select(maxfd,
Willy Tarreau28d86862007-04-08 17:42:27 +0200181 readnotnull ? tmp_evts[DIR_RD] : NULL,
182 writenotnull ? tmp_evts[DIR_WR] : NULL,
Willy Tarreau4f60f162007-04-08 16:39:58 +0200183 NULL,
Willy Tarreaub0b37bc2008-06-23 14:00:57 +0200184 &delta);
Willy Tarreau48f8bc12018-11-22 18:57:37 +0100185 tv_update_date(delta_ms, status);
Willy Tarreau7e9c4ae2018-10-17 14:31:19 +0200186 tv_leaving_poll(delta_ms, status);
Willy Tarreau4f60f162007-04-08 16:39:58 +0200187
Willy Tarreau60b639c2018-08-02 10:16:17 +0200188 thread_harmless_end();
Olivier Houchard305d5ab2019-07-24 18:07:06 +0200189 if (sleeping_thread_mask & tid_bit)
190 _HA_ATOMIC_AND(&sleeping_thread_mask, ~tid_bit);
Willy Tarreau60b639c2018-08-02 10:16:17 +0200191
Willy Tarreau4f60f162007-04-08 16:39:58 +0200192 if (status <= 0)
193 return;
194
Willy Tarreau177e2b02008-07-15 00:36:31 +0200195 for (fds = 0; (fds * BITS_PER_INT) < maxfd; fds++) {
Willy Tarreau28d86862007-04-08 17:42:27 +0200196 if ((((int *)(tmp_evts[DIR_RD]))[fds] | ((int *)(tmp_evts[DIR_WR]))[fds]) == 0)
Willy Tarreau4f60f162007-04-08 16:39:58 +0200197 continue;
198
Willy Tarreau177e2b02008-07-15 00:36:31 +0200199 for (count = BITS_PER_INT, fd = fds * BITS_PER_INT; count && fd < maxfd; count--, fd++) {
Christopher Fauletab62f512017-08-30 10:34:36 +0200200 unsigned int n = 0;
201
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100202 if (!fdtab[fd].owner) {
203 activity[tid].poll_dead++;
204 continue;
205 }
206
207 if (!(fdtab[fd].thread_mask & tid_bit)) {
208 activity[tid].poll_skip++;
Willy Tarreau076be252012-07-06 16:02:29 +0200209 continue;
Willy Tarreaud80cb4e2018-01-20 19:30:13 +0100210 }
Willy Tarreau076be252012-07-06 16:02:29 +0200211
Willy Tarreau076be252012-07-06 16:02:29 +0200212 if (FD_ISSET(fd, tmp_evts[DIR_RD]))
Willy Tarreau6b308982019-09-06 19:05:50 +0200213 n |= FD_EV_READY_R;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200214
Willy Tarreau076be252012-07-06 16:02:29 +0200215 if (FD_ISSET(fd, tmp_evts[DIR_WR]))
Willy Tarreau6b308982019-09-06 19:05:50 +0200216 n |= FD_EV_READY_W;
Willy Tarreau5be2f352014-11-19 19:43:05 +0100217
Christopher Fauletab62f512017-08-30 10:34:36 +0200218 fd_update_events(fd, n);
Willy Tarreau4f60f162007-04-08 16:39:58 +0200219 }
220 }
221}
222
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200223static int init_select_per_thread()
224{
225 int fd_set_bytes;
226
227 fd_set_bytes = sizeof(fd_set) * (global.maxsock + FD_SETSIZE - 1) / FD_SETSIZE;
228 if ((tmp_evts[DIR_RD] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
229 goto fail;
230 if ((tmp_evts[DIR_WR] = (fd_set *)calloc(1, fd_set_bytes)) == NULL)
231 goto fail;
232 return 1;
233 fail:
234 free(tmp_evts[DIR_RD]);
235 free(tmp_evts[DIR_WR]);
236 return 0;
237}
238
239static void deinit_select_per_thread()
240{
Christopher Fauletcd7879a2017-10-27 13:53:47 +0200241 free(tmp_evts[DIR_WR]); tmp_evts[DIR_WR] = NULL;
242 free(tmp_evts[DIR_RD]); tmp_evts[DIR_RD] = NULL;
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200243}
244
Willy Tarreau4f60f162007-04-08 16:39:58 +0200245/*
Willy Tarreaue54e9172007-04-09 09:23:31 +0200246 * Initialization of the select() poller.
247 * Returns 0 in case of failure, non-zero in case of success. If it fails, it
248 * disables the poller by setting its pref to 0.
249 */
Willy Tarreau03e78532020-02-25 07:38:05 +0100250static int _do_init(struct poller *p)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200251{
Christopher Fauletd4604ad2017-05-29 10:40:41 +0200252 __label__ fail_swevt, fail_srevt, fail_revt;
Willy Tarreaue54e9172007-04-09 09:23:31 +0200253 int fd_set_bytes;
254
255 p->private = NULL;
Willy Tarreau3fa87b12013-03-31 14:41:15 +0200256
257 if (global.maxsock > FD_SETSIZE)
258 goto fail_revt;
259
Willy Tarreaue54e9172007-04-09 09:23:31 +0200260 fd_set_bytes = sizeof(fd_set) * (global.maxsock + FD_SETSIZE - 1) / FD_SETSIZE;
Willy Tarreaue54e9172007-04-09 09:23:31 +0200261
Willy Tarreaud51a5072018-01-25 16:48:46 +0100262 if ((fd_evts[DIR_RD] = calloc(1, fd_set_bytes)) == NULL)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200263 goto fail_srevt;
Willy Tarreaud51a5072018-01-25 16:48:46 +0100264 if ((fd_evts[DIR_WR] = calloc(1, fd_set_bytes)) == NULL)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200265 goto fail_swevt;
266
Christopher Fauletcd7879a2017-10-27 13:53:47 +0200267 hap_register_per_thread_init(init_select_per_thread);
268 hap_register_per_thread_deinit(deinit_select_per_thread);
269
Willy Tarreaue54e9172007-04-09 09:23:31 +0200270 return 1;
271
272 fail_swevt:
273 free(fd_evts[DIR_RD]);
274 fail_srevt:
275 free(tmp_evts[DIR_WR]);
Willy Tarreaue54e9172007-04-09 09:23:31 +0200276 free(tmp_evts[DIR_RD]);
277 fail_revt:
278 p->pref = 0;
279 return 0;
280}
281
282/*
283 * Termination of the select() poller.
284 * Memory is released and the poller is marked as unselectable.
285 */
Willy Tarreau03e78532020-02-25 07:38:05 +0100286static void _do_term(struct poller *p)
Willy Tarreaue54e9172007-04-09 09:23:31 +0200287{
Willy Tarreaua534fea2008-08-03 12:19:50 +0200288 free(fd_evts[DIR_WR]);
289 free(fd_evts[DIR_RD]);
Willy Tarreaue54e9172007-04-09 09:23:31 +0200290 p->private = NULL;
291 p->pref = 0;
292}
293
294/*
Willy Tarreau2ff76222007-04-09 19:29:56 +0200295 * Check that the poller works.
296 * Returns 1 if OK, otherwise 0.
297 */
Willy Tarreau03e78532020-02-25 07:38:05 +0100298static int _do_test(struct poller *p)
Willy Tarreau2ff76222007-04-09 19:29:56 +0200299{
Willy Tarreau3fa87b12013-03-31 14:41:15 +0200300 if (global.maxsock > FD_SETSIZE)
301 return 0;
302
Willy Tarreau2ff76222007-04-09 19:29:56 +0200303 return 1;
304}
305
306/*
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200307 * It is a constructor, which means that it will automatically be called before
308 * main(). This is GCC-specific but it works at least since 2.95.
309 * Special care must be taken so that it does not need any uninitialized data.
Willy Tarreau4f60f162007-04-08 16:39:58 +0200310 */
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200311__attribute__((constructor))
312static void _do_register(void)
Willy Tarreau4f60f162007-04-08 16:39:58 +0200313{
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200314 struct poller *p;
315
316 if (nbpollers >= MAX_POLLERS)
317 return;
318 p = &pollers[nbpollers++];
319
Willy Tarreau4f60f162007-04-08 16:39:58 +0200320 p->name = "select";
321 p->pref = 150;
Willy Tarreau5a767692017-03-13 11:38:28 +0100322 p->flags = 0;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200323 p->private = NULL;
324
Willy Tarreau70c6fd82012-11-11 21:02:34 +0100325 p->clo = __fd_clo;
Willy Tarreauef1d1f82007-04-16 00:25:25 +0200326 p->test = _do_test;
327 p->init = _do_init;
328 p->term = _do_term;
329 p->poll = _do_poll;
Willy Tarreau4f60f162007-04-08 16:39:58 +0200330}
331
332
333/*
334 * Local variables:
335 * c-indent-level: 8
336 * c-basic-offset: 8
337 * End:
338 */