]> git.ipfire.org Git - thirdparty/squid.git/blame - src/comm/ModEpoll.cc
Source Format Enforcement (#763)
[thirdparty/squid.git] / src / comm / ModEpoll.cc
CommitLineData
6039b729 1/*
f70aedc4 2 * Copyright (C) 1996-2021 The Squid Software Foundation and contributors
6039b729 3 *
bbc27441
AJ
4 * Squid software is distributed under GPLv2+ license and includes
5 * contributions from numerous individuals and organizations.
6 * Please see the COPYING and CONTRIBUTORS files for details.
6039b729 7 */
8
bbc27441
AJ
9/* DEBUG: section 05 Socket Functions */
10
6039b729 11/*
12 * The idea for this came from these two websites:
13 * http://www.xmailserver.org/linux-patches/nio-improve.html
14 * http://www.kegel.com/c10k.html
15 *
16 * This is to support the epoll sysctl being added to the linux 2.5
17 * kernel tree. The new sys_epoll is an event based poller without
18 * most of the fuss of rtsignals.
19 *
20 * -- David Nicklay <dnicklay@web.turner.com>
21 */
22
23/*
24 * XXX Currently not implemented / supported by this module XXX
25 *
26 * - delay pools
27 * - deferred reads
28 *
29 */
30
f7f3304a 31#include "squid.h"
d841c88d
AJ
32
33#if USE_EPOLL
34
ccfbe8f4 35#include "base/CodeContext.h"
d841c88d 36#include "comm/Loops.h"
1287b9e1 37#include "fde.h"
582c2af2 38#include "globals.h"
d841c88d 39#include "mgr/Registration.h"
582c2af2 40#include "profiler/Profiler.h"
348697ca 41#include "SquidTime.h"
e1656dc4 42#include "StatCounters.h"
00a7574e 43#include "StatHist.h"
d841c88d 44#include "Store.h"
8a02a7f8 45
6039b729 46#define DEBUG_EPOLL 0
47
1a30fdf5 48#include <cerrno>
ad32c661 49#if HAVE_SYS_EPOLL_H
6039b729 50#include <sys/epoll.h>
ad32c661 51#endif
6039b729 52
612fc474 53static int kdpfd = -1;
6039b729 54static int max_poll_time = 1000;
55
56static struct epoll_event *pevents;
57
5acc9f37 58static void commEPollRegisterWithCacheManager(void);
6039b729 59
6039b729 60/* XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX */
61/* Public functions */
62
6039b729 63/*
6039b729 64 * This is a needed exported function which will be called to initialise
65 * the network loop code.
66 */
67void
d841c88d 68Comm::SelectLoopInit(void)
6039b729 69{
6039b729 70 pevents = (struct epoll_event *) xmalloc(SQUID_MAXFD * sizeof(struct epoll_event));
71
72 if (!pevents) {
b69e9ffa
AJ
73 int xerrno = errno;
74 fatalf("comm_select_init: xmalloc() failed: %s\n", xstrerr(xerrno));
6039b729 75 }
76
77 kdpfd = epoll_create(SQUID_MAXFD);
78
79 if (kdpfd < 0) {
b69e9ffa
AJ
80 int xerrno = errno;
81 fatalf("comm_select_init: epoll_create(): %s\n", xstrerr(xerrno));
6039b729 82 }
8cb183ec
FC
83
84 commEPollRegisterWithCacheManager();
6039b729 85}
86
751406fe 87static const char* epolltype_atoi(int x)
88{
26ac0430 89 switch (x) {
751406fe 90
91 case EPOLL_CTL_ADD:
92 return "EPOLL_CTL_ADD";
93
94 case EPOLL_CTL_DEL:
95 return "EPOLL_CTL_DEL";
96
97 case EPOLL_CTL_MOD:
98 return "EPOLL_CTL_MOD";
99
100 default:
101 return "UNKNOWN_EPOLLCTL_OP";
102 }
103}
104
d841c88d 105/**
6039b729 106 * This is a needed exported function which will be called to register
107 * and deregister interest in a pending IO state for a given FD.
6039b729 108 */
109void
d841c88d 110Comm::SetSelect(int fd, unsigned int type, PF * handler, void *client_data, time_t timeout)
6039b729 111{
112 fde *F = &fd_table[fd];
751406fe 113 int epoll_ctl_type = 0;
6039b729 114
6039b729 115 assert(fd >= 0);
48e7baac 116 debugs(5, 5, HERE << "FD " << fd << ", type=" << type <<
d841c88d
AJ
117 ", handler=" << handler << ", client_data=" << client_data <<
118 ", timeout=" << timeout);
6039b729 119
612fc474
AJ
120 struct epoll_event ev;
121 memset(&ev, 0, sizeof(ev));
751406fe 122 ev.data.fd = fd;
6039b729 123
a1727eb7 124 if (!F->flags.open) {
125 epoll_ctl(kdpfd, EPOLL_CTL_DEL, fd, &ev);
126 return;
127 }
128
751406fe 129 // If read is an interest
6039b729 130
131 if (type & COMM_SELECT_READ) {
0a080ce3 132 if (handler) {
26ac0430
AJ
133 // Hack to keep the events flowing if there is data immediately ready
134 if (F->flags.read_pending)
135 ev.events |= EPOLLOUT;
751406fe 136 ev.events |= EPOLLIN;
26ac0430 137 }
6039b729 138
139 F->read_handler = handler;
140
141 F->read_data = client_data;
751406fe 142
143 // Otherwise, use previously stored value
144 } else if (F->epoll_state & EPOLLIN) {
145 ev.events |= EPOLLIN;
6039b729 146 }
147
751406fe 148 // If write is an interest
6039b729 149 if (type & COMM_SELECT_WRITE) {
751406fe 150 if (handler)
151 ev.events |= EPOLLOUT;
6039b729 152
153 F->write_handler = handler;
154
155 F->write_data = client_data;
6039b729 156
751406fe 157 // Otherwise, use previously stored value
158 } else if (F->epoll_state & EPOLLOUT) {
159 ev.events |= EPOLLOUT;
160 }
6039b729 161
751406fe 162 if (ev.events)
163 ev.events |= EPOLLHUP | EPOLLERR;
6039b729 164
751406fe 165 if (ev.events != F->epoll_state) {
166 if (F->epoll_state) // already monitoring something.
167 epoll_ctl_type = ev.events ? EPOLL_CTL_MOD : EPOLL_CTL_DEL;
168 else
169 epoll_ctl_type = EPOLL_CTL_ADD;
6039b729 170
751406fe 171 F->epoll_state = ev.events;
6039b729 172
751406fe 173 if (epoll_ctl(kdpfd, epoll_ctl_type, fd, &ev) < 0) {
b69e9ffa
AJ
174 int xerrno = errno;
175 debugs(5, DEBUG_EPOLL ? 0 : 8, "epoll_ctl(," << epolltype_atoi(epoll_ctl_type) <<
176 ",,): failed on FD " << fd << ": " << xstrerr(xerrno));
6039b729 177 }
178 }
179
180 if (timeout)
181 F->timeout = squid_curtime + timeout;
ccfbe8f4
AR
182
183 if (timeout || handler) // all non-cleanup requests
184 F->codeContext = CodeContext::Current(); // TODO: Avoid clearing if set?
185 else if (!ev.events) // full cleanup: no more FD-associated work expected
186 F->codeContext = nullptr;
187 // else: direction-specific/timeout cleanup requests preserve F->codeContext
6039b729 188}
189
610ee341 190static void commIncomingStats(StoreEntry * sentry);
191
5acc9f37 192static void
dbeed9ea 193commEPollRegisterWithCacheManager(void)
610ee341 194{
8822ebee 195 Mgr::RegisterAction("comm_epoll_incoming",
d9fc6862
A
196 "comm_incoming() stats",
197 commIncomingStats, 0, 1);
610ee341 198}
199
200static void
201commIncomingStats(StoreEntry * sentry)
202{
203 StatCounters *f = &statCounter;
0a515876 204 storeAppendPrintf(sentry, "Total number of epoll(2) loops: %ld\n", statCounter.select_loops);
610ee341 205 storeAppendPrintf(sentry, "Histogram of returned filedescriptors\n");
96886986 206 f->select_fds_hist.dump(sentry, statHistIntDumper);
610ee341 207}
208
d841c88d 209/**
6039b729 210 * Check all connections for new connections and input data that is to be
211 * processed. Also check for connections with data queued and whether we can
212 * write it out.
6039b729 213 *
214 * Called to do the new-style IO, courtesy of of squid (like most of this
215 * new IO code). This routine handles the stuff we've hidden in
216 * comm_setselect and fd_table[] and calls callbacks for IO ready
217 * events.
218 */
c8407295 219Comm::Flag
d841c88d 220Comm::DoSelect(int msec)
6039b729 221{
222 int num, i,fd;
223 fde *F;
224 PF *hdl;
225
226 struct epoll_event *cevents;
6039b729 227
a1727eb7 228 PROF_start(comm_check_incoming);
229
6039b729 230 if (msec > max_poll_time)
231 msec = max_poll_time;
232
233 for (;;) {
234 num = epoll_wait(kdpfd, pevents, SQUID_MAXFD, msec);
098346fd 235 ++ statCounter.select_loops;
6039b729 236
237 if (num >= 0)
238 break;
239
240 if (ignoreErrno(errno))
241 break;
242
243 getCurrentTime();
244
a1727eb7 245 PROF_stop(comm_check_incoming);
246
4ee57cbe 247 return Comm::COMM_ERROR;
6039b729 248 }
249
a1727eb7 250 PROF_stop(comm_check_incoming);
6039b729 251 getCurrentTime();
252
f30f7998 253 statCounter.select_fds_hist.count(num);
751406fe 254
6039b729 255 if (num == 0)
f53969cc 256 return Comm::TIMEOUT; /* No error.. */
751406fe 257
258 PROF_start(comm_handle_ready_fd);
6039b729 259
cbebe602 260 for (i = 0, cevents = pevents; i < num; ++i, ++cevents) {
6039b729 261 fd = cevents->data.fd;
262 F = &fd_table[fd];
ccfbe8f4 263 CodeContext::Reset(F->codeContext);
d841c88d 264 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "got FD " << fd << " events=" <<
26ac0430
AJ
265 std::hex << cevents->events << " monitoring=" << F->epoll_state <<
266 " F->read_handler=" << F->read_handler << " F->write_handler=" << F->write_handler);
751406fe 267
268 // TODO: add EPOLLPRI??
6039b729 269
0a080ce3 270 if (cevents->events & (EPOLLIN|EPOLLHUP|EPOLLERR) || F->flags.read_pending) {
a1727eb7 271 if ((hdl = F->read_handler) != NULL) {
d841c88d 272 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "Calling read handler on FD " << fd);
751406fe 273 PROF_start(comm_write_handler);
6039b729 274 F->read_handler = NULL;
275 hdl(fd, F->read_data);
751406fe 276 PROF_stop(comm_write_handler);
098346fd 277 ++ statCounter.select_fds;
751406fe 278 } else {
d841c88d 279 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "no read handler for FD " << fd);
751406fe 280 // remove interest since no handler exist for this event.
d841c88d 281 SetSelect(fd, COMM_SELECT_READ, NULL, NULL, 0);
6039b729 282 }
283 }
284
751406fe 285 if (cevents->events & (EPOLLOUT|EPOLLHUP|EPOLLERR)) {
a1727eb7 286 if ((hdl = F->write_handler) != NULL) {
d841c88d 287 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "Calling write handler on FD " << fd);
751406fe 288 PROF_start(comm_read_handler);
6039b729 289 F->write_handler = NULL;
290 hdl(fd, F->write_data);
751406fe 291 PROF_stop(comm_read_handler);
098346fd 292 ++ statCounter.select_fds;
751406fe 293 } else {
d841c88d 294 debugs(5, DEBUG_EPOLL ? 0 : 8, HERE << "no write handler for FD " << fd);
751406fe 295 // remove interest since no handler exist for this event.
d841c88d 296 SetSelect(fd, COMM_SELECT_WRITE, NULL, NULL, 0);
6039b729 297 }
298 }
299 }
300
ccfbe8f4
AR
301 CodeContext::Reset();
302
751406fe 303 PROF_stop(comm_handle_ready_fd);
304
c8407295 305 return Comm::OK;
6039b729 306}
307
308void
d841c88d 309Comm::QuickPollRequired(void)
6039b729 310{
768d3d2f 311 max_poll_time = 10;
6039b729 312}
313
8a02a7f8 314#endif /* USE_EPOLL */
f53969cc 315