]> git.ipfire.org Git - thirdparty/squid.git/blame - src/comm/ModEpoll.cc
Reduce cache_effective_user was leaking $HOME memory
[thirdparty/squid.git] / src / comm / ModEpoll.cc
CommitLineData
6039b729 1/*
b510f3a1 2 * DEBUG: section 05 Socket Functions
6039b729 3 *
4 * SQUID Web Proxy Cache http://www.squid-cache.org/
5 * ----------------------------------------------------------
6 *
7 * Squid is the result of efforts by numerous individuals from
8 * the Internet community; see the CONTRIBUTORS file for full
9 * details. Many organizations have provided support for Squid's
10 * development; see the SPONSORS file for full details. Squid is
11 * Copyrighted (C) 2001 by the Regents of the University of
12 * California; see the COPYRIGHT file for full details. Squid
13 * incorporates software developed and/or copyrighted by other
14 * sources; see the CREDITS file for full details.
15 *
16 * This program is free software; you can redistribute it and/or modify
17 * it under the terms of the GNU General Public License as published by
18 * the Free Software Foundation; either version 2 of the License, or
19 * (at your option) any later version.
20 *
21 * This program is distributed in the hope that it will be useful,
22 * but WITHOUT ANY WARRANTY; without even the implied warranty of
23 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
24 * GNU General Public License for more details.
25 *
26 * You should have received a copy of the GNU General Public License
27 * along with this program; if not, write to the Free Software
28 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111, USA.
29 *
30 */
31
32/*
33 * The idea for this came from these two websites:
34 * http://www.xmailserver.org/linux-patches/nio-improve.html
35 * http://www.kegel.com/c10k.html
36 *
37 * This is to support the epoll sysctl being added to the linux 2.5
38 * kernel tree. The new sys_epoll is an event based poller without
39 * most of the fuss of rtsignals.
40 *
41 * -- David Nicklay <dnicklay@web.turner.com>
42 */
43
44/*
45 * XXX Currently not implemented / supported by this module XXX
46 *
47 * - delay pools
48 * - deferred reads
49 *
50 */
51
f7f3304a 52#include "squid.h"
d841c88d
AJ
53
54#if USE_EPOLL
55
d841c88d 56#include "comm/Loops.h"
1287b9e1 57#include "fde.h"
582c2af2 58#include "globals.h"
d841c88d 59#include "mgr/Registration.h"
582c2af2 60#include "profiler/Profiler.h"
348697ca 61#include "SquidTime.h"
e1656dc4 62#include "StatCounters.h"
00a7574e 63#include "StatHist.h"
d841c88d 64#include "Store.h"
8a02a7f8 65
6039b729 66#define DEBUG_EPOLL 0
67
ad32c661 68#if HAVE_SYS_EPOLL_H
6039b729 69#include <sys/epoll.h>
ad32c661 70#endif
21d845b1
FC
71#if HAVE_ERRNO_H
72#include <errno.h>
73#endif
6039b729 74
75static int kdpfd;
76static int max_poll_time = 1000;
77
78static struct epoll_event *pevents;
79
5acc9f37 80static void commEPollRegisterWithCacheManager(void);
6039b729 81
6039b729 82/* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */
83/* Public functions */
84
6039b729 85/*
6039b729 86 * This is a needed exported function which will be called to initialise
87 * the network loop code.
88 */
89void
d841c88d 90Comm::SelectLoopInit(void)
6039b729 91{
6039b729 92 pevents = (struct epoll_event *) xmalloc(SQUID_MAXFD * sizeof(struct epoll_event));
93
94 if (!pevents) {
95 fatalf("comm_select_init: xmalloc() failed: %s\n",xstrerror());
96 }
97
98 kdpfd = epoll_create(SQUID_MAXFD);
99
100 if (kdpfd < 0) {
101 fatalf("comm_select_init: epoll_create(): %s\n",xstrerror());
102 }
8cb183ec
FC
103
104 commEPollRegisterWithCacheManager();
6039b729 105}
106
751406fe 107static const char* epolltype_atoi(int x)
108{
26ac0430 109 switch (x) {
751406fe 110
111 case EPOLL_CTL_ADD:
112 return "EPOLL_CTL_ADD";
113
114 case EPOLL_CTL_DEL:
115 return "EPOLL_CTL_DEL";
116
117 case EPOLL_CTL_MOD:
118 return "EPOLL_CTL_MOD";
119
120 default:
121 return "UNKNOWN_EPOLLCTL_OP";
122 }
123}
124
d841c88d 125/**
6039b729 126 * This is a needed exported function which will be called to register
127 * and deregister interest in a pending IO state for a given FD.
6039b729 128 */
129void
d841c88d 130Comm::SetSelect(int fd, unsigned int type, PF * handler, void *client_data, time_t timeout)
6039b729 131{
132 fde *F = &fd_table[fd];
751406fe 133 int epoll_ctl_type = 0;
6039b729 134
135 struct epoll_event ev;
136 assert(fd >= 0);
48e7baac 137 debugs(5, 5, HERE << "FD " << fd << ", type=" << type <<
d841c88d
AJ
138 ", handler=" << handler << ", client_data=" << client_data <<
139 ", timeout=" << timeout);
6039b729 140
b4bab919 141 if (RUNNING_ON_VALGRIND) {
26ac0430
AJ
142 /* Keep valgrind happy.. complains about uninitialized bytes otherwise */
143 memset(&ev, 0, sizeof(ev));
b4bab919 144 }
751406fe 145 ev.events = 0;
146 ev.data.fd = fd;
6039b729 147
a1727eb7 148 if (!F->flags.open) {
149 epoll_ctl(kdpfd, EPOLL_CTL_DEL, fd, &ev);
150 return;
151 }
152
751406fe 153 // If read is an interest
6039b729 154
155 if (type & COMM_SELECT_READ) {
0a080ce3 156 if (handler) {
26ac0430
AJ
157 // Hack to keep the events flowing if there is data immediately ready
158 if (F->flags.read_pending)
159 ev.events |= EPOLLOUT;
751406fe 160 ev.events |= EPOLLIN;
26ac0430 161 }
6039b729 162
163 F->read_handler = handler;
164
165 F->read_data = client_data;
751406fe 166
167 // Otherwise, use previously stored value
168 } else if (F->epoll_state & EPOLLIN) {
169 ev.events |= EPOLLIN;
6039b729 170 }
171
751406fe 172 // If write is an interest
6039b729 173 if (type & COMM_SELECT_WRITE) {
751406fe 174 if (handler)
175 ev.events |= EPOLLOUT;
6039b729 176
177 F->write_handler = handler;
178
179 F->write_data = client_data;
6039b729 180
751406fe 181 // Otherwise, use previously stored value
182 } else if (F->epoll_state & EPOLLOUT) {
183 ev.events |= EPOLLOUT;
184 }
6039b729 185
751406fe 186 if (ev.events)
187 ev.events |= EPOLLHUP | EPOLLERR;
6039b729 188
751406fe 189 if (ev.events != F->epoll_state) {
190 if (F->epoll_state) // already monitoring something.
191 epoll_ctl_type = ev.events ? EPOLL_CTL_MOD : EPOLL_CTL_DEL;
192 else
193 epoll_ctl_type = EPOLL_CTL_ADD;
6039b729 194
751406fe 195 F->epoll_state = ev.events;
6039b729 196
751406fe 197 if (epoll_ctl(kdpfd, epoll_ctl_type, fd, &ev) < 0) {
d841c88d 198 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "epoll_ctl(," << epolltype_atoi(epoll_ctl_type) <<
bf8fe701 199 ",,): failed on FD " << fd << ": " << xstrerror());
6039b729 200 }
201 }
202
203 if (timeout)
204 F->timeout = squid_curtime + timeout;
205}
206
3a5a4930 207void
d841c88d 208Comm::ResetSelect(int fd)
3a5a4930 209{
210 fde *F = &fd_table[fd];
211 F->epoll_state = 0;
d841c88d 212 SetSelect(fd, 0, NULL, NULL, 0);
3a5a4930 213}
214
610ee341 215static void commIncomingStats(StoreEntry * sentry);
216
5acc9f37 217static void
dbeed9ea 218commEPollRegisterWithCacheManager(void)
610ee341 219{
8822ebee 220 Mgr::RegisterAction("comm_epoll_incoming",
d9fc6862
A
221 "comm_incoming() stats",
222 commIncomingStats, 0, 1);
610ee341 223}
224
225static void
226commIncomingStats(StoreEntry * sentry)
227{
228 StatCounters *f = &statCounter;
0a515876 229 storeAppendPrintf(sentry, "Total number of epoll(2) loops: %ld\n", statCounter.select_loops);
610ee341 230 storeAppendPrintf(sentry, "Histogram of returned filedescriptors\n");
96886986 231 f->select_fds_hist.dump(sentry, statHistIntDumper);
610ee341 232}
233
d841c88d 234/**
6039b729 235 * Check all connections for new connections and input data that is to be
236 * processed. Also check for connections with data queued and whether we can
237 * write it out.
6039b729 238 *
239 * Called to do the new-style IO, courtesy of of squid (like most of this
240 * new IO code). This routine handles the stuff we've hidden in
241 * comm_setselect and fd_table[] and calls callbacks for IO ready
242 * events.
243 */
6039b729 244comm_err_t
d841c88d 245Comm::DoSelect(int msec)
6039b729 246{
247 int num, i,fd;
248 fde *F;
249 PF *hdl;
250
251 struct epoll_event *cevents;
6039b729 252
a1727eb7 253 PROF_start(comm_check_incoming);
254
6039b729 255 if (msec > max_poll_time)
256 msec = max_poll_time;
257
258 for (;;) {
259 num = epoll_wait(kdpfd, pevents, SQUID_MAXFD, msec);
098346fd 260 ++ statCounter.select_loops;
6039b729 261
262 if (num >= 0)
263 break;
264
265 if (ignoreErrno(errno))
266 break;
267
268 getCurrentTime();
269
a1727eb7 270 PROF_stop(comm_check_incoming);
271
6039b729 272 return COMM_ERROR;
273 }
274
a1727eb7 275 PROF_stop(comm_check_incoming);
6039b729 276 getCurrentTime();
277
f30f7998 278 statCounter.select_fds_hist.count(num);
751406fe 279
6039b729 280 if (num == 0)
751406fe 281 return COMM_TIMEOUT; /* No error.. */
282
283 PROF_start(comm_handle_ready_fd);
6039b729 284
cbebe602 285 for (i = 0, cevents = pevents; i < num; ++i, ++cevents) {
6039b729 286 fd = cevents->data.fd;
287 F = &fd_table[fd];
d841c88d 288 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "got FD " << fd << " events=" <<
26ac0430
AJ
289 std::hex << cevents->events << " monitoring=" << F->epoll_state <<
290 " F->read_handler=" << F->read_handler << " F->write_handler=" << F->write_handler);
751406fe 291
292 // TODO: add EPOLLPRI??
6039b729 293
0a080ce3 294 if (cevents->events & (EPOLLIN|EPOLLHUP|EPOLLERR) || F->flags.read_pending) {
a1727eb7 295 if ((hdl = F->read_handler) != NULL) {
d841c88d 296 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "Calling read handler on FD " << fd);
751406fe 297 PROF_start(comm_write_handler);
0a080ce3 298 F->flags.read_pending = 0;
6039b729 299 F->read_handler = NULL;
300 hdl(fd, F->read_data);
751406fe 301 PROF_stop(comm_write_handler);
098346fd 302 ++ statCounter.select_fds;
751406fe 303 } else {
d841c88d 304 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "no read handler for FD " << fd);
751406fe 305 // remove interest since no handler exist for this event.
d841c88d 306 SetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0);
6039b729 307 }
308 }
309
751406fe 310 if (cevents->events & (EPOLLOUT|EPOLLHUP|EPOLLERR)) {
a1727eb7 311 if ((hdl = F->write_handler) != NULL) {
d841c88d 312 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "Calling write handler on FD " << fd);
751406fe 313 PROF_start(comm_read_handler);
6039b729 314 F->write_handler = NULL;
315 hdl(fd, F->write_data);
751406fe 316 PROF_stop(comm_read_handler);
098346fd 317 ++ statCounter.select_fds;
751406fe 318 } else {
d841c88d 319 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "no write handler for FD " << fd);
751406fe 320 // remove interest since no handler exist for this event.
d841c88d 321 SetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0);
6039b729 322 }
323 }
324 }
325
751406fe 326 PROF_stop(comm_handle_ready_fd);
327
6039b729 328 return COMM_OK;
329}
330
331void
d841c88d 332Comm::QuickPollRequired(void)
6039b729 333{
768d3d2f 334 max_poll_time = 10;
6039b729 335}
336
8a02a7f8 337#endif /* USE_EPOLL */