Commit | Line | Data |
---|---|---|
460b9539 | 1 | /* |
2 | * This file is part of DisOrder. | |
964e027d | 3 | * Copyright (C) 2004, 2005, 2007, 2008 Richard Kettlewell |
460b9539 | 4 | * |
e7eb3a27 | 5 | * This program is free software: you can redistribute it and/or modify |
460b9539 | 6 | * it under the terms of the GNU General Public License as published by |
e7eb3a27 | 7 | * the Free Software Foundation, either version 3 of the License, or |
460b9539 | 8 | * (at your option) any later version. |
e7eb3a27 RK |
9 | * |
10 | * This program is distributed in the hope that it will be useful, | |
11 | * but WITHOUT ANY WARRANTY; without even the implied warranty of | |
12 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the | |
13 | * GNU General Public License for more details. | |
14 | * | |
460b9539 | 15 | * You should have received a copy of the GNU General Public License |
e7eb3a27 | 16 | * along with this program. If not, see <http://www.gnu.org/licenses/>. |
460b9539 | 17 | */ |
768d7355 RK |
18 | /** @file lib/event.c |
19 | * @brief DisOrder event loop | |
20 | */ | |
460b9539 | 21 | |
05b75f8d | 22 | #include "common.h" |
460b9539 | 23 | |
24 | #include <unistd.h> | |
25 | #include <fcntl.h> | |
26 | #include <sys/time.h> | |
27 | #include <sys/types.h> | |
28 | #include <sys/resource.h> | |
29 | #include <sys/wait.h> | |
46bd8db4 | 30 | #include <sys/stat.h> |
460b9539 | 31 | #include <unistd.h> |
460b9539 | 32 | #include <signal.h> |
33 | #include <errno.h> | |
460b9539 | 34 | #include <sys/socket.h> |
35 | #include <netinet/in.h> | |
36 | #include <sys/un.h> | |
460b9539 | 37 | #include "event.h" |
38 | #include "mem.h" | |
39 | #include "log.h" | |
40 | #include "syscalls.h" | |
41 | #include "printf.h" | |
42 | #include "sink.h" | |
768d7355 | 43 | #include "vector.h" |
3af7813d RK |
44 | #include "timeval.h" |
45 | #include "heap.h" | |
460b9539 | 46 | |
768d7355 | 47 | /** @brief A timeout */ |
460b9539 | 48 | struct timeout { |
49 | struct timeout *next; | |
50 | struct timeval when; | |
51 | ev_timeout_callback *callback; | |
52 | void *u; | |
3af7813d | 53 | int active; |
460b9539 | 54 | }; |
55 | ||
3af7813d RK |
56 | /** @brief Comparison function for timeouts */ |
57 | static int timeout_lt(const struct timeout *a, | |
58 | const struct timeout *b) { | |
59 | return tvlt(&a->when, &b->when); | |
60 | } | |
61 | ||
62 | HEAP_TYPE(timeout_heap, struct timeout *, timeout_lt); | |
63 | HEAP_DEFINE(timeout_heap, struct timeout *, timeout_lt); | |
64 | ||
768d7355 | 65 | /** @brief A file descriptor in one mode */ |
460b9539 | 66 | struct fd { |
67 | int fd; | |
68 | ev_fd_callback *callback; | |
69 | void *u; | |
e8c92ba7 | 70 | const char *what; |
460b9539 | 71 | }; |
72 | ||
768d7355 | 73 | /** @brief All the file descriptors in a given mode */ |
460b9539 | 74 | struct fdmode { |
768d7355 | 75 | /** @brief Mask of active file descriptors passed to @c select() */ |
460b9539 | 76 | fd_set enabled; |
768d7355 RK |
77 | |
78 | /** @brief File descriptor mask returned from @c select() */ | |
460b9539 | 79 | fd_set tripped; |
768d7355 RK |
80 | |
81 | /** @brief Number of file descriptors in @p fds */ | |
82 | int nfds; | |
83 | ||
84 | /** @brief Number of slots in @p fds */ | |
85 | int fdslots; | |
86 | ||
87 | /** @brief Array of all active file descriptors */ | |
460b9539 | 88 | struct fd *fds; |
768d7355 RK |
89 | |
90 | /** @brief Highest-numbered file descriptor or 0 */ | |
460b9539 | 91 | int maxfd; |
92 | }; | |
93 | ||
768d7355 | 94 | /** @brief A signal handler */ |
460b9539 | 95 | struct signal { |
96 | struct sigaction oldsa; | |
97 | ev_signal_callback *callback; | |
98 | void *u; | |
99 | }; | |
100 | ||
768d7355 | 101 | /** @brief A child process */ |
460b9539 | 102 | struct child { |
103 | pid_t pid; | |
104 | int options; | |
105 | ev_child_callback *callback; | |
106 | void *u; | |
107 | }; | |
108 | ||
768d7355 | 109 | /** @brief An event loop */ |
460b9539 | 110 | struct ev_source { |
768d7355 | 111 | /** @brief File descriptors, per mode */ |
460b9539 | 112 | struct fdmode mode[ev_nmodes]; |
768d7355 | 113 | |
3af7813d RK |
114 | /** @brief Heap of timeouts */ |
115 | struct timeout_heap timeouts[1]; | |
768d7355 RK |
116 | |
117 | /** @brief Array of handled signals */ | |
460b9539 | 118 | struct signal signals[NSIG]; |
768d7355 RK |
119 | |
120 | /** @brief Mask of handled signals */ | |
460b9539 | 121 | sigset_t sigmask; |
768d7355 RK |
122 | |
123 | /** @brief Escape early from handling of @c select() results | |
124 | * | |
125 | * This is set if any of the file descriptor arrays are invalidated, since | |
126 | * it's then not safe for processing of them to continue. | |
127 | */ | |
460b9539 | 128 | int escape; |
768d7355 RK |
129 | |
130 | /** @brief Signal handling pipe | |
131 | * | |
132 | * The signal handle writes signal numbers down this pipe. | |
133 | */ | |
460b9539 | 134 | int sigpipe[2]; |
768d7355 RK |
135 | |
136 | /** @brief Number of child processes in @p children */ | |
137 | int nchildren; | |
138 | ||
139 | /** @brief Number of slots in @p children */ | |
140 | int nchildslots; | |
141 | ||
142 | /** @brief Array of child processes */ | |
460b9539 | 143 | struct child *children; |
144 | }; | |
145 | ||
768d7355 | 146 | /** @brief Names of file descriptor modes */ |
460b9539 | 147 | static const char *modenames[] = { "read", "write", "except" }; |
148 | ||
149 | /* utilities ******************************************************************/ | |
150 | ||
460b9539 | 151 | /* creation *******************************************************************/ |
152 | ||
768d7355 | 153 | /** @brief Create a new event loop */ |
460b9539 | 154 | ev_source *ev_new(void) { |
155 | ev_source *ev = xmalloc(sizeof *ev); | |
156 | int n; | |
157 | ||
158 | memset(ev, 0, sizeof *ev); | |
159 | for(n = 0; n < ev_nmodes; ++n) | |
160 | FD_ZERO(&ev->mode[n].enabled); | |
161 | ev->sigpipe[0] = ev->sigpipe[1] = -1; | |
162 | sigemptyset(&ev->sigmask); | |
3af7813d | 163 | timeout_heap_init(ev->timeouts); |
460b9539 | 164 | return ev; |
165 | } | |
166 | ||
167 | /* event loop *****************************************************************/ | |
168 | ||
768d7355 RK |
169 | /** @brief Run the event loop |
170 | * @return -1 on error, non-0 if any callback returned non-0 | |
171 | */ | |
460b9539 | 172 | int ev_run(ev_source *ev) { |
173 | for(;;) { | |
174 | struct timeval now; | |
175 | struct timeval delta; | |
176 | int n, mode; | |
177 | int ret; | |
178 | int maxfd; | |
3af7813d | 179 | struct timeout *timeouts, *t, **tt; |
e8c92ba7 | 180 | struct stat sb; |
460b9539 | 181 | |
182 | xgettimeofday(&now, 0); | |
183 | /* Handle timeouts. We don't want to handle any timeouts that are added | |
184 | * while we're handling them (otherwise we'd have to break out of infinite | |
185 | * loops, preferrably without starving better-behaved subsystems). Hence | |
186 | * the slightly complicated two-phase approach here. */ | |
3af7813d RK |
187 | /* First we read those timeouts that have triggered out of the heap. We |
188 | * keep them in the same order they came out of the heap in. */ | |
189 | tt = &timeouts; | |
190 | while(timeout_heap_count(ev->timeouts) | |
191 | && tvle(&timeout_heap_first(ev->timeouts)->when, &now)) { | |
192 | /* This timeout has reached its trigger time; provided it has not been | |
193 | * cancelled we add it to the timeouts list. */ | |
194 | t = timeout_heap_remove(ev->timeouts); | |
195 | if(t->active) { | |
196 | *tt = t; | |
197 | tt = &t->next; | |
198 | } | |
199 | } | |
200 | *tt = 0; | |
201 | /* Now we can run the callbacks for those timeouts. They might add further | |
202 | * timeouts that are already in the past but they won't trigger until the | |
203 | * next time round the event loop. */ | |
204 | for(t = timeouts; t; t = t->next) { | |
460b9539 | 205 | D(("calling timeout for %ld.%ld callback %p %p", |
206 | (long)t->when.tv_sec, (long)t->when.tv_usec, | |
207 | (void *)t->callback, t->u)); | |
208 | ret = t->callback(ev, &now, t->u); | |
209 | if(ret) | |
210 | return ret; | |
211 | } | |
460b9539 | 212 | maxfd = 0; |
213 | for(mode = 0; mode < ev_nmodes; ++mode) { | |
214 | ev->mode[mode].tripped = ev->mode[mode].enabled; | |
215 | if(ev->mode[mode].maxfd > maxfd) | |
216 | maxfd = ev->mode[mode].maxfd; | |
217 | } | |
218 | xsigprocmask(SIG_UNBLOCK, &ev->sigmask, 0); | |
219 | do { | |
3af7813d RK |
220 | if(timeout_heap_count(ev->timeouts)) { |
221 | t = timeout_heap_first(ev->timeouts); | |
460b9539 | 222 | xgettimeofday(&now, 0); |
3af7813d RK |
223 | delta.tv_sec = t->when.tv_sec - now.tv_sec; |
224 | delta.tv_usec = t->when.tv_usec - now.tv_usec; | |
460b9539 | 225 | if(delta.tv_usec < 0) { |
226 | delta.tv_usec += 1000000; | |
227 | --delta.tv_sec; | |
228 | } | |
229 | if(delta.tv_sec < 0) | |
230 | delta.tv_sec = delta.tv_usec = 0; | |
231 | n = select(maxfd + 1, | |
232 | &ev->mode[ev_read].tripped, | |
233 | &ev->mode[ev_write].tripped, | |
234 | &ev->mode[ev_except].tripped, | |
235 | &delta); | |
236 | } else { | |
237 | n = select(maxfd + 1, | |
238 | &ev->mode[ev_read].tripped, | |
239 | &ev->mode[ev_write].tripped, | |
240 | &ev->mode[ev_except].tripped, | |
241 | 0); | |
242 | } | |
243 | } while(n < 0 && errno == EINTR); | |
244 | xsigprocmask(SIG_BLOCK, &ev->sigmask, 0); | |
245 | if(n < 0) { | |
2e9ba080 | 246 | disorder_error(errno, "error calling select"); |
e8c92ba7 RK |
247 | if(errno == EBADF) { |
248 | /* If there's a bad FD in the mix then check them all and log what we | |
249 | * find, to ease debugging */ | |
250 | for(mode = 0; mode < ev_nmodes; ++mode) { | |
251 | for(n = 0; n < ev->mode[mode].nfds; ++n) { | |
252 | const int fd = ev->mode[mode].fds[n].fd; | |
253 | ||
254 | if(FD_ISSET(fd, &ev->mode[mode].enabled) | |
255 | && fstat(fd, &sb) < 0) | |
2e9ba080 RK |
256 | disorder_error(errno, "mode %s fstat %d (%s)", |
257 | modenames[mode], fd, ev->mode[mode].fds[n].what); | |
e8c92ba7 | 258 | } |
7958ad2f | 259 | for(n = 0; n <= maxfd; ++n) |
34a3e246 RK |
260 | if(FD_ISSET(n, &ev->mode[mode].enabled) |
261 | && fstat(n, &sb) < 0) | |
2e9ba080 | 262 | disorder_error(errno, "mode %s fstat %d", modenames[mode], n); |
e8c92ba7 RK |
263 | } |
264 | } | |
460b9539 | 265 | return -1; |
266 | } | |
267 | if(n > 0) { | |
268 | /* if anything deranges the meaning of an fd, or re-orders the | |
269 | * fds[] tables, we'd better give up; such operations will | |
270 | * therefore set @escape@. */ | |
271 | ev->escape = 0; | |
272 | for(mode = 0; mode < ev_nmodes && !ev->escape; ++mode) | |
273 | for(n = 0; n < ev->mode[mode].nfds && !ev->escape; ++n) { | |
274 | int fd = ev->mode[mode].fds[n].fd; | |
275 | if(FD_ISSET(fd, &ev->mode[mode].tripped)) { | |
276 | D(("calling %s fd %d callback %p %p", modenames[mode], fd, | |
277 | (void *)ev->mode[mode].fds[n].callback, | |
278 | ev->mode[mode].fds[n].u)); | |
279 | ret = ev->mode[mode].fds[n].callback(ev, fd, | |
280 | ev->mode[mode].fds[n].u); | |
281 | if(ret) | |
282 | return ret; | |
283 | } | |
284 | } | |
285 | } | |
286 | /* we'll pick up timeouts back round the loop */ | |
287 | } | |
288 | } | |
289 | ||
290 | /* file descriptors ***********************************************************/ | |
291 | ||
768d7355 RK |
292 | /** @brief Register a file descriptor |
293 | * @param ev Event loop | |
294 | * @param mode @c ev_read or @c ev_write | |
295 | * @param fd File descriptor | |
296 | * @param callback Called when @p is readable/writable | |
297 | * @param u Passed to @p callback | |
298 | * @param what Text description | |
299 | * @return 0 on success, non-0 on error | |
300 | * | |
301 | * Sets @ref ev_source::escape, so no further processing of file descriptors | |
302 | * will occur this time round the event loop. | |
303 | */ | |
460b9539 | 304 | int ev_fd(ev_source *ev, |
305 | ev_fdmode mode, | |
306 | int fd, | |
307 | ev_fd_callback *callback, | |
e8c92ba7 RK |
308 | void *u, |
309 | const char *what) { | |
460b9539 | 310 | int n; |
311 | ||
312 | D(("registering %s fd %d callback %p %p", modenames[mode], fd, | |
313 | (void *)callback, u)); | |
3bd1c92f RK |
314 | /* FreeBSD defines FD_SETSIZE as 1024u for some reason */ |
315 | if((unsigned)fd >= FD_SETSIZE) | |
31e2a93e | 316 | return -1; |
460b9539 | 317 | assert(mode < ev_nmodes); |
318 | if(ev->mode[mode].nfds >= ev->mode[mode].fdslots) { | |
319 | ev->mode[mode].fdslots = (ev->mode[mode].fdslots | |
320 | ? 2 * ev->mode[mode].fdslots : 16); | |
321 | D(("expanding %s fd table to %d entries", modenames[mode], | |
322 | ev->mode[mode].fdslots)); | |
323 | ev->mode[mode].fds = xrealloc(ev->mode[mode].fds, | |
324 | ev->mode[mode].fdslots * sizeof (struct fd)); | |
325 | } | |
326 | n = ev->mode[mode].nfds++; | |
327 | FD_SET(fd, &ev->mode[mode].enabled); | |
328 | ev->mode[mode].fds[n].fd = fd; | |
329 | ev->mode[mode].fds[n].callback = callback; | |
330 | ev->mode[mode].fds[n].u = u; | |
e8c92ba7 | 331 | ev->mode[mode].fds[n].what = what; |
460b9539 | 332 | if(fd > ev->mode[mode].maxfd) |
333 | ev->mode[mode].maxfd = fd; | |
334 | ev->escape = 1; | |
335 | return 0; | |
336 | } | |
337 | ||
768d7355 RK |
338 | /** @brief Cancel a file descriptor |
339 | * @param ev Event loop | |
340 | * @param mode @c ev_read or @c ev_write | |
341 | * @param fd File descriptor | |
342 | * @return 0 on success, non-0 on error | |
343 | * | |
344 | * Sets @ref ev_source::escape, so no further processing of file descriptors | |
345 | * will occur this time round the event loop. | |
346 | */ | |
460b9539 | 347 | int ev_fd_cancel(ev_source *ev, ev_fdmode mode, int fd) { |
348 | int n; | |
349 | int maxfd; | |
350 | ||
351 | D(("cancelling mode %s fd %d", modenames[mode], fd)); | |
352 | /* find the right struct fd */ | |
353 | for(n = 0; n < ev->mode[mode].nfds && fd != ev->mode[mode].fds[n].fd; ++n) | |
354 | ; | |
355 | assert(n < ev->mode[mode].nfds); | |
356 | /* swap in the last fd and reduce the count */ | |
357 | if(n != ev->mode[mode].nfds - 1) | |
358 | ev->mode[mode].fds[n] = ev->mode[mode].fds[ev->mode[mode].nfds - 1]; | |
359 | --ev->mode[mode].nfds; | |
360 | /* if that was the biggest fd, find the new biggest one */ | |
361 | if(fd == ev->mode[mode].maxfd) { | |
362 | maxfd = 0; | |
363 | for(n = 0; n < ev->mode[mode].nfds; ++n) | |
364 | if(ev->mode[mode].fds[n].fd > maxfd) | |
365 | maxfd = ev->mode[mode].fds[n].fd; | |
366 | ev->mode[mode].maxfd = maxfd; | |
367 | } | |
368 | /* don't tell select about this fd any more */ | |
369 | FD_CLR(fd, &ev->mode[mode].enabled); | |
370 | ev->escape = 1; | |
371 | return 0; | |
372 | } | |
373 | ||
768d7355 RK |
374 | /** @brief Re-enable a file descriptor |
375 | * @param ev Event loop | |
376 | * @param mode @c ev_read or @c ev_write | |
377 | * @param fd File descriptor | |
378 | * @return 0 on success, non-0 on error | |
379 | * | |
380 | * It is harmless if @p fd is currently disabled, but it must not have been | |
381 | * cancelled. | |
382 | */ | |
460b9539 | 383 | int ev_fd_enable(ev_source *ev, ev_fdmode mode, int fd) { |
38b8221f | 384 | assert(fd >= 0); |
460b9539 | 385 | D(("enabling mode %s fd %d", modenames[mode], fd)); |
386 | FD_SET(fd, &ev->mode[mode].enabled); | |
387 | return 0; | |
388 | } | |
389 | ||
768d7355 RK |
390 | /** @brief Temporarily disable a file descriptor |
391 | * @param ev Event loop | |
392 | * @param mode @c ev_read or @c ev_write | |
393 | * @param fd File descriptor | |
394 | * @return 0 on success, non-0 on error | |
395 | * | |
396 | * Re-enable with ev_fd_enable(). It is harmless if @p fd is already disabled, | |
397 | * but it must not have been cancelled. | |
398 | */ | |
460b9539 | 399 | int ev_fd_disable(ev_source *ev, ev_fdmode mode, int fd) { |
400 | D(("disabling mode %s fd %d", modenames[mode], fd)); | |
401 | FD_CLR(fd, &ev->mode[mode].enabled); | |
402 | FD_CLR(fd, &ev->mode[mode].tripped); | |
75d64210 RK |
403 | /* Suppress any pending callbacks */ |
404 | ev->escape = 1; | |
460b9539 | 405 | return 0; |
406 | } | |
407 | ||
768d7355 RK |
408 | /** @brief Log a report of file descriptor state */ |
409 | void ev_report(ev_source *ev) { | |
410 | int n, fd; | |
411 | ev_fdmode mode; | |
412 | struct dynstr d[1]; | |
413 | char b[4096]; | |
414 | ||
0fa83caa RK |
415 | if(!debugging) |
416 | return; | |
768d7355 RK |
417 | dynstr_init(d); |
418 | for(mode = 0; mode < ev_nmodes; ++mode) { | |
0fa83caa | 419 | D(("mode %s maxfd %d", modenames[mode], ev->mode[mode].maxfd)); |
768d7355 RK |
420 | for(n = 0; n < ev->mode[mode].nfds; ++n) { |
421 | fd = ev->mode[mode].fds[n].fd; | |
0fa83caa RK |
422 | D(("fd %s %d%s%s (%s)", modenames[mode], fd, |
423 | FD_ISSET(fd, &ev->mode[mode].enabled) ? " enabled" : "", | |
424 | FD_ISSET(fd, &ev->mode[mode].tripped) ? " tripped" : "", | |
425 | ev->mode[mode].fds[n].what)); | |
768d7355 RK |
426 | } |
427 | d->nvec = 0; | |
428 | for(fd = 0; fd <= ev->mode[mode].maxfd; ++fd) { | |
429 | if(!FD_ISSET(fd, &ev->mode[mode].enabled)) | |
430 | continue; | |
431 | for(n = 0; n < ev->mode[mode].nfds; ++n) { | |
432 | if(ev->mode[mode].fds[n].fd == fd) | |
433 | break; | |
434 | } | |
435 | if(n < ev->mode[mode].nfds) | |
34a3e246 | 436 | snprintf(b, sizeof b, "%d(%s)", fd, ev->mode[mode].fds[n].what); |
768d7355 | 437 | else |
34a3e246 | 438 | snprintf(b, sizeof b, "%d", fd); |
768d7355 RK |
439 | dynstr_append(d, ' '); |
440 | dynstr_append_string(d, b); | |
441 | } | |
442 | dynstr_terminate(d); | |
0fa83caa | 443 | D(("%s enabled:%s", modenames[mode], d->vec)); |
768d7355 RK |
444 | } |
445 | } | |
446 | ||
460b9539 | 447 | /* timeouts *******************************************************************/ |
448 | ||
768d7355 RK |
449 | /** @brief Register a timeout |
450 | * @param ev Event source | |
3149c1e2 | 451 | * @param handlep Where to store timeout handle, or @c NULL |
768d7355 RK |
452 | * @param when Earliest time to call @p callback, or @c NULL |
453 | * @param callback Function to call at or after @p when | |
454 | * @param u Passed to @p callback | |
455 | * @return 0 on success, non-0 on error | |
456 | * | |
457 | * If @p when is a null pointer then a time of 0 is assumed. The effect is to | |
458 | * call the timeout handler from ev_run() next time around the event loop. | |
459 | * This is used internally to schedule various operations if it is not | |
460 | * convenient to call them from the current place in the call stack, or | |
461 | * externally to ensure that other clients of the event loop get a look in when | |
462 | * performing some lengthy operation. | |
463 | */ | |
460b9539 | 464 | int ev_timeout(ev_source *ev, |
465 | ev_timeout_handle *handlep, | |
466 | const struct timeval *when, | |
467 | ev_timeout_callback *callback, | |
468 | void *u) { | |
3af7813d | 469 | struct timeout *t; |
460b9539 | 470 | |
471 | D(("registering timeout at %ld.%ld callback %p %p", | |
472 | when ? (long)when->tv_sec : 0, when ? (long)when->tv_usec : 0, | |
473 | (void *)callback, u)); | |
474 | t = xmalloc(sizeof *t); | |
475 | if(when) | |
476 | t->when = *when; | |
477 | t->callback = callback; | |
478 | t->u = u; | |
3af7813d RK |
479 | t->active = 1; |
480 | timeout_heap_insert(ev->timeouts, t); | |
460b9539 | 481 | if(handlep) |
482 | *handlep = t; | |
483 | return 0; | |
484 | } | |
485 | ||
768d7355 RK |
486 | /** @brief Cancel a timeout |
487 | * @param ev Event loop | |
cb9a695c | 488 | * @param handle Handle returned from ev_timeout(), or 0 |
768d7355 | 489 | * @return 0 on success, non-0 on error |
cb9a695c RK |
490 | * |
491 | * If @p handle is 0 then this is a no-op. | |
768d7355 | 492 | */ |
3af7813d | 493 | int ev_timeout_cancel(ev_source attribute((unused)) *ev, |
460b9539 | 494 | ev_timeout_handle handle) { |
3af7813d | 495 | struct timeout *t = handle; |
460b9539 | 496 | |
3af7813d RK |
497 | if(t) |
498 | t->active = 0; | |
499 | return 0; | |
460b9539 | 500 | } |
501 | ||
502 | /* signals ********************************************************************/ | |
503 | ||
768d7355 RK |
504 | /** @brief Mapping of signals to pipe write ends |
505 | * | |
506 | * The pipes are per-event loop, it's possible in theory for there to be | |
507 | * multiple event loops (e.g. in different threads), although in fact DisOrder | |
508 | * does not do this. | |
509 | */ | |
460b9539 | 510 | static int sigfd[NSIG]; |
511 | ||
768d7355 RK |
512 | /** @brief The signal handler |
513 | * @param s Signal number | |
514 | * | |
515 | * Writes to @c sigfd[s]. | |
516 | */ | |
460b9539 | 517 | static void sighandler(int s) { |
518 | unsigned char sc = s; | |
519 | static const char errmsg[] = "error writing to signal pipe"; | |
520 | ||
521 | /* probably the reader has stopped listening for some reason */ | |
522 | if(write(sigfd[s], &sc, 1) < 0) { | |
918393ff RK |
523 | /* do the best we can as we're about to abort; shut _up_, gcc */ |
524 | int _ignore = write(2, errmsg, sizeof errmsg - 1); | |
525 | (void)_ignore; | |
460b9539 | 526 | abort(); |
527 | } | |
528 | } | |
529 | ||
768d7355 | 530 | /** @brief Read callback for signals */ |
460b9539 | 531 | static int signal_read(ev_source *ev, |
532 | int attribute((unused)) fd, | |
533 | void attribute((unused)) *u) { | |
534 | unsigned char s; | |
535 | int n; | |
536 | int ret; | |
537 | ||
538 | if((n = read(ev->sigpipe[0], &s, 1)) == 1) | |
539 | if((ret = ev->signals[s].callback(ev, s, ev->signals[s].u))) | |
540 | return ret; | |
541 | assert(n != 0); | |
542 | if(n < 0 && (errno != EINTR && errno != EAGAIN)) { | |
2e9ba080 | 543 | disorder_error(errno, "error reading from signal pipe %d", ev->sigpipe[0]); |
460b9539 | 544 | return -1; |
545 | } | |
546 | return 0; | |
547 | } | |
548 | ||
768d7355 | 549 | /** @brief Close the signal pipe */ |
460b9539 | 550 | static void close_sigpipe(ev_source *ev) { |
551 | int save_errno = errno; | |
552 | ||
553 | xclose(ev->sigpipe[0]); | |
554 | xclose(ev->sigpipe[1]); | |
555 | ev->sigpipe[0] = ev->sigpipe[1] = -1; | |
556 | errno = save_errno; | |
557 | } | |
558 | ||
768d7355 RK |
559 | /** @brief Register a signal handler |
560 | * @param ev Event loop | |
561 | * @param sig Signal to handle | |
562 | * @param callback Called when signal is delivered | |
563 | * @param u Passed to @p callback | |
564 | * @return 0 on success, non-0 on error | |
565 | * | |
566 | * Note that @p callback is called from inside ev_run(), not from inside the | |
567 | * signal handler, so the usual restrictions on signal handlers do not apply. | |
568 | */ | |
460b9539 | 569 | int ev_signal(ev_source *ev, |
570 | int sig, | |
571 | ev_signal_callback *callback, | |
572 | void *u) { | |
573 | int n; | |
574 | struct sigaction sa; | |
575 | ||
576 | D(("registering signal %d handler callback %p %p", sig, (void *)callback, u)); | |
577 | assert(sig > 0); | |
578 | assert(sig < NSIG); | |
579 | assert(sig <= UCHAR_MAX); | |
580 | if(ev->sigpipe[0] == -1) { | |
581 | D(("creating signal pipe")); | |
582 | xpipe(ev->sigpipe); | |
583 | D(("signal pipe is %d, %d", ev->sigpipe[0], ev->sigpipe[1])); | |
584 | for(n = 0; n < 2; ++n) { | |
585 | nonblock(ev->sigpipe[n]); | |
586 | cloexec(ev->sigpipe[n]); | |
587 | } | |
e8c92ba7 | 588 | if(ev_fd(ev, ev_read, ev->sigpipe[0], signal_read, 0, "sigpipe read")) { |
460b9539 | 589 | close_sigpipe(ev); |
590 | return -1; | |
591 | } | |
592 | } | |
593 | sigaddset(&ev->sigmask, sig); | |
594 | xsigprocmask(SIG_BLOCK, &ev->sigmask, 0); | |
595 | sigfd[sig] = ev->sigpipe[1]; | |
596 | ev->signals[sig].callback = callback; | |
597 | ev->signals[sig].u = u; | |
598 | sa.sa_handler = sighandler; | |
599 | sigfillset(&sa.sa_mask); | |
600 | sa.sa_flags = SA_RESTART; | |
601 | xsigaction(sig, &sa, &ev->signals[sig].oldsa); | |
602 | ev->escape = 1; | |
603 | return 0; | |
604 | } | |
605 | ||
768d7355 RK |
606 | /** @brief Cancel a signal handler |
607 | * @param ev Event loop | |
608 | * @param sig Signal to cancel | |
609 | * @return 0 on success, non-0 on error | |
610 | */ | |
460b9539 | 611 | int ev_signal_cancel(ev_source *ev, |
612 | int sig) { | |
613 | sigset_t ss; | |
614 | ||
615 | xsigaction(sig, &ev->signals[sig].oldsa, 0); | |
616 | ev->signals[sig].callback = 0; | |
617 | ev->escape = 1; | |
618 | sigdelset(&ev->sigmask, sig); | |
619 | sigemptyset(&ss); | |
620 | sigaddset(&ss, sig); | |
621 | xsigprocmask(SIG_UNBLOCK, &ss, 0); | |
622 | return 0; | |
623 | } | |
624 | ||
768d7355 RK |
625 | /** @brief Clean up signal handling |
626 | * @param ev Event loop | |
627 | * | |
628 | * This function can be called from inside a fork. It restores signal | |
629 | * handlers, unblocks the signals, and closes the signal pipe for @p ev. | |
630 | */ | |
460b9539 | 631 | void ev_signal_atfork(ev_source *ev) { |
632 | int sig; | |
633 | ||
634 | if(ev->sigpipe[0] != -1) { | |
635 | /* revert any handled signals to their original state */ | |
636 | for(sig = 1; sig < NSIG; ++sig) { | |
637 | if(ev->signals[sig].callback != 0) | |
638 | xsigaction(sig, &ev->signals[sig].oldsa, 0); | |
639 | } | |
640 | /* and then unblock them */ | |
641 | xsigprocmask(SIG_UNBLOCK, &ev->sigmask, 0); | |
642 | /* don't want a copy of the signal pipe open inside the fork */ | |
643 | xclose(ev->sigpipe[0]); | |
644 | xclose(ev->sigpipe[1]); | |
645 | } | |
646 | } | |
647 | ||
648 | /* child processes ************************************************************/ | |
649 | ||
768d7355 | 650 | /** @brief Called on SIGCHLD */ |
460b9539 | 651 | static int sigchld_callback(ev_source *ev, |
652 | int attribute((unused)) sig, | |
653 | void attribute((unused)) *u) { | |
654 | struct rusage ru; | |
655 | pid_t r; | |
656 | int status, n, ret, revisit; | |
657 | ||
658 | do { | |
659 | revisit = 0; | |
660 | for(n = 0; n < ev->nchildren; ++n) { | |
661 | r = wait4(ev->children[n].pid, | |
662 | &status, | |
663 | ev->children[n].options | WNOHANG, | |
664 | &ru); | |
665 | if(r > 0) { | |
666 | ev_child_callback *c = ev->children[n].callback; | |
667 | void *cu = ev->children[n].u; | |
668 | ||
669 | if(WIFEXITED(status) || WIFSIGNALED(status)) | |
670 | ev_child_cancel(ev, r); | |
671 | revisit = 1; | |
672 | if((ret = c(ev, r, status, &ru, cu))) | |
673 | return ret; | |
674 | } else if(r < 0) { | |
675 | /* We should "never" get an ECHILD but it can in fact happen. For | |
676 | * instance on Linux 2.4.31, and probably other versions, if someone | |
677 | * straces a child process and then a different child process | |
678 | * terminates, when we wait4() the trace process we will get ECHILD | |
679 | * because it has been reparented to strace. Obviously this is a | |
680 | * hopeless design flaw in the tracing infrastructure, but we don't | |
681 | * want the disorder server to bomb out because of it. So we just log | |
682 | * the problem and ignore it. | |
683 | */ | |
2e9ba080 RK |
684 | disorder_error(errno, "error calling wait4 for PID %lu (broken ptrace?)", |
685 | (unsigned long)ev->children[n].pid); | |
460b9539 | 686 | if(errno != ECHILD) |
687 | return -1; | |
688 | } | |
689 | } | |
690 | } while(revisit); | |
691 | return 0; | |
692 | } | |
693 | ||
768d7355 RK |
694 | /** @brief Configure event loop for child process handling |
695 | * @return 0 on success, non-0 on error | |
696 | * | |
697 | * Currently at most one event loop can handle child processes and it must be | |
698 | * distinguished from others by calling this function on it. This could be | |
699 | * fixed but since no process ever makes use of more than one event loop there | |
700 | * is no need. | |
701 | */ | |
460b9539 | 702 | int ev_child_setup(ev_source *ev) { |
703 | D(("installing SIGCHLD handler")); | |
704 | return ev_signal(ev, SIGCHLD, sigchld_callback, 0); | |
705 | } | |
706 | ||
768d7355 RK |
707 | /** @brief Wait for a child process to terminate |
708 | * @param ev Event loop | |
709 | * @param pid Process ID of child | |
710 | * @param options Options to pass to @c wait4() | |
711 | * @param callback Called when child terminates (or possibly when it stops) | |
712 | * @param u Passed to @p callback | |
713 | * @return 0 on success, non-0 on error | |
714 | * | |
715 | * You must have called ev_child_setup() on @p ev once first. | |
716 | */ | |
460b9539 | 717 | int ev_child(ev_source *ev, |
718 | pid_t pid, | |
719 | int options, | |
720 | ev_child_callback *callback, | |
721 | void *u) { | |
722 | int n; | |
723 | ||
724 | D(("registering child handling %ld options %d callback %p %p", | |
725 | (long)pid, options, (void *)callback, u)); | |
726 | assert(ev->signals[SIGCHLD].callback == sigchld_callback); | |
727 | if(ev->nchildren >= ev->nchildslots) { | |
728 | ev->nchildslots = ev->nchildslots ? 2 * ev->nchildslots : 16; | |
729 | ev->children = xrealloc(ev->children, | |
730 | ev->nchildslots * sizeof (struct child)); | |
731 | } | |
732 | n = ev->nchildren++; | |
733 | ev->children[n].pid = pid; | |
734 | ev->children[n].options = options; | |
735 | ev->children[n].callback = callback; | |
736 | ev->children[n].u = u; | |
737 | return 0; | |
738 | } | |
739 | ||
768d7355 RK |
740 | /** @brief Stop waiting for a child process |
741 | * @param ev Event loop | |
742 | * @param pid Child process ID | |
743 | * @return 0 on success, non-0 on error | |
744 | */ | |
460b9539 | 745 | int ev_child_cancel(ev_source *ev, |
746 | pid_t pid) { | |
747 | int n; | |
748 | ||
749 | for(n = 0; n < ev->nchildren && ev->children[n].pid != pid; ++n) | |
750 | ; | |
751 | assert(n < ev->nchildren); | |
752 | if(n != ev->nchildren - 1) | |
753 | ev->children[n] = ev->children[ev->nchildren - 1]; | |
754 | --ev->nchildren; | |
755 | return 0; | |
756 | } | |
757 | ||
18ed984a RK |
758 | /** @brief Terminate and wait for all child processes |
759 | * @param ev Event loop | |
760 | * | |
761 | * Does *not* call the completion callbacks. Only used during teardown. | |
762 | */ | |
763 | void ev_child_killall(ev_source *ev) { | |
764 | int n, rc, w; | |
765 | ||
766 | for(n = 0; n < ev->nchildren; ++n) { | |
767 | if(kill(ev->children[n].pid, SIGTERM) < 0) { | |
2e9ba080 RK |
768 | disorder_error(errno, "sending SIGTERM to pid %lu", |
769 | (unsigned long)ev->children[n].pid); | |
18ed984a | 770 | ev->children[n].pid = -1; |
77b521b0 | 771 | } |
18ed984a RK |
772 | } |
773 | for(n = 0; n < ev->nchildren; ++n) { | |
774 | if(ev->children[n].pid == -1) | |
775 | continue; | |
776 | do { | |
777 | rc = waitpid(ev->children[n].pid, &w, 0); | |
778 | } while(rc < 0 && errno == EINTR); | |
779 | if(rc < 0) { | |
2e9ba080 RK |
780 | disorder_error(errno, "waiting for pid %lu", |
781 | (unsigned long)ev->children[n].pid); | |
18ed984a RK |
782 | continue; |
783 | } | |
18ed984a RK |
784 | } |
785 | ev->nchildren = 0; | |
786 | } | |
787 | ||
460b9539 | 788 | /* socket listeners ***********************************************************/ |
789 | ||
768d7355 | 790 | /** @brief State for a socket listener */ |
460b9539 | 791 | struct listen_state { |
792 | ev_listen_callback *callback; | |
793 | void *u; | |
794 | }; | |
795 | ||
768d7355 | 796 | /** @brief Called when a listenign socket is readable */ |
460b9539 | 797 | static int listen_callback(ev_source *ev, int fd, void *u) { |
798 | const struct listen_state *l = u; | |
799 | int newfd; | |
800 | union { | |
801 | struct sockaddr_in in; | |
802 | #if HAVE_STRUCT_SOCKADDR_IN6 | |
803 | struct sockaddr_in6 in6; | |
804 | #endif | |
805 | struct sockaddr_un un; | |
806 | struct sockaddr sa; | |
807 | } addr; | |
808 | socklen_t addrlen; | |
809 | int ret; | |
810 | ||
811 | D(("callback for listener fd %d", fd)); | |
812 | while((addrlen = sizeof addr), | |
813 | (newfd = accept(fd, &addr.sa, &addrlen)) >= 0) { | |
814 | if((ret = l->callback(ev, newfd, &addr.sa, addrlen, l->u))) | |
815 | return ret; | |
816 | } | |
817 | switch(errno) { | |
818 | case EINTR: | |
819 | case EAGAIN: | |
820 | break; | |
821 | #ifdef ECONNABORTED | |
822 | case ECONNABORTED: | |
2e9ba080 | 823 | disorder_error(errno, "error calling accept"); |
460b9539 | 824 | break; |
825 | #endif | |
826 | #ifdef EPROTO | |
827 | case EPROTO: | |
828 | /* XXX on some systems EPROTO should be fatal, but we don't know if | |
829 | * we're running on one of them */ | |
2e9ba080 | 830 | disorder_error(errno, "error calling accept"); |
460b9539 | 831 | break; |
832 | #endif | |
833 | default: | |
2e9ba080 | 834 | disorder_fatal(errno, "error calling accept"); |
460b9539 | 835 | break; |
836 | } | |
837 | if(errno != EINTR && errno != EAGAIN) | |
2e9ba080 | 838 | disorder_error(errno, "error calling accept"); |
460b9539 | 839 | return 0; |
840 | } | |
841 | ||
768d7355 RK |
842 | /** @brief Listen on a socket for inbound stream connections |
843 | * @param ev Event source | |
844 | * @param fd File descriptor of socket | |
845 | * @param callback Called when a new connection arrives | |
846 | * @param u Passed to @p callback | |
847 | * @param what Text description of socket | |
848 | * @return 0 on success, non-0 on error | |
849 | */ | |
460b9539 | 850 | int ev_listen(ev_source *ev, |
851 | int fd, | |
852 | ev_listen_callback *callback, | |
e8c92ba7 RK |
853 | void *u, |
854 | const char *what) { | |
460b9539 | 855 | struct listen_state *l = xmalloc(sizeof *l); |
856 | ||
857 | D(("registering listener fd %d callback %p %p", fd, (void *)callback, u)); | |
858 | l->callback = callback; | |
859 | l->u = u; | |
e8c92ba7 | 860 | return ev_fd(ev, ev_read, fd, listen_callback, l, what); |
460b9539 | 861 | } |
862 | ||
768d7355 RK |
863 | /** @brief Stop listening on a socket |
864 | * @param ev Event loop | |
865 | * @param fd File descriptor of socket | |
866 | * @return 0 on success, non-0 on error | |
867 | */ | |
460b9539 | 868 | int ev_listen_cancel(ev_source *ev, int fd) { |
869 | D(("cancelling listener fd %d", fd)); | |
870 | return ev_fd_cancel(ev, ev_read, fd); | |
871 | } | |
872 | ||
873 | /* buffer *********************************************************************/ | |
874 | ||
768d7355 | 875 | /** @brief Buffer structure */ |
460b9539 | 876 | struct buffer { |
877 | char *base, *start, *end, *top; | |
878 | }; | |
879 | ||
768d7355 | 880 | /* @brief Make sure there is @p bytes available at @c b->end */ |
460b9539 | 881 | static void buffer_space(struct buffer *b, size_t bytes) { |
882 | D(("buffer_space %p %p %p %p want %lu", | |
883 | (void *)b->base, (void *)b->start, (void *)b->end, (void *)b->top, | |
884 | (unsigned long)bytes)); | |
885 | if(b->start == b->end) | |
886 | b->start = b->end = b->base; | |
887 | if((size_t)(b->top - b->end) < bytes) { | |
888 | if((size_t)((b->top - b->end) + (b->start - b->base)) < bytes) { | |
889 | size_t newspace = b->end - b->start + bytes, n; | |
890 | char *newbase; | |
891 | ||
892 | for(n = 16; n < newspace; n *= 2) | |
893 | ; | |
894 | newbase = xmalloc_noptr(n); | |
895 | memcpy(newbase, b->start, b->end - b->start); | |
896 | b->base = newbase; | |
897 | b->end = newbase + (b->end - b->start); | |
898 | b->top = newbase + n; | |
899 | b->start = newbase; /* must be last */ | |
900 | } else { | |
901 | memmove(b->base, b->start, b->end - b->start); | |
902 | b->end = b->base + (b->end - b->start); | |
903 | b->start = b->base; | |
904 | } | |
905 | } | |
906 | D(("result %p %p %p %p", | |
907 | (void *)b->base, (void *)b->start, (void *)b->end, (void *)b->top)); | |
908 | } | |
909 | ||
75d64210 | 910 | /* readers and writers *******************************************************/ |
460b9539 | 911 | |
768d7355 | 912 | /** @brief State structure for a buffered writer */ |
460b9539 | 913 | struct ev_writer { |
75d64210 | 914 | /** @brief Sink used for writing to the buffer */ |
460b9539 | 915 | struct sink s; |
75d64210 RK |
916 | |
917 | /** @brief Output buffer */ | |
460b9539 | 918 | struct buffer b; |
75d64210 RK |
919 | |
920 | /** @brief File descriptor to write to */ | |
460b9539 | 921 | int fd; |
75d64210 RK |
922 | |
923 | /** @brief Set if there'll be no more output */ | |
460b9539 | 924 | int eof; |
75d64210 RK |
925 | |
926 | /** @brief Error/termination callback */ | |
460b9539 | 927 | ev_error_callback *callback; |
75d64210 RK |
928 | |
929 | /** @brief Passed to @p callback */ | |
460b9539 | 930 | void *u; |
75d64210 RK |
931 | |
932 | /** @brief Parent event source */ | |
460b9539 | 933 | ev_source *ev; |
cb9a695c RK |
934 | |
935 | /** @brief Maximum amount of time between succesful writes, 0 = don't care */ | |
936 | int timebound; | |
937 | /** @brief Maximum amount of data to buffer, 0 = don't care */ | |
938 | int spacebound; | |
75d64210 RK |
939 | /** @brief Error code to pass to @p callback (see writer_shutdown()) */ |
940 | int error; | |
cb9a695c RK |
941 | /** @brief Timeout handle for @p timebound (or 0) */ |
942 | ev_timeout_handle timeout; | |
943 | ||
75d64210 | 944 | /** @brief Description of this writer */ |
cb9a695c | 945 | const char *what; |
75d64210 RK |
946 | |
947 | /** @brief Tied reader or 0 */ | |
948 | ev_reader *reader; | |
d742bb47 RK |
949 | |
950 | /** @brief Set when abandoned */ | |
951 | int abandoned; | |
460b9539 | 952 | }; |
953 | ||
75d64210 RK |
954 | /** @brief State structure for a buffered reader */ |
955 | struct ev_reader { | |
956 | /** @brief Input buffer */ | |
957 | struct buffer b; | |
958 | /** @brief File descriptor read from */ | |
959 | int fd; | |
960 | /** @brief Called when new data is available */ | |
961 | ev_reader_callback *callback; | |
962 | /** @brief Called on error and shutdown */ | |
963 | ev_error_callback *error_callback; | |
964 | /** @brief Passed to @p callback and @p error_callback */ | |
965 | void *u; | |
966 | /** @brief Parent event loop */ | |
967 | ev_source *ev; | |
968 | /** @brief Set when EOF is detected */ | |
969 | int eof; | |
970 | /** @brief Error code to pass to error callback */ | |
971 | int error; | |
972 | /** @brief Tied writer or NULL */ | |
973 | ev_writer *writer; | |
974 | }; | |
975 | ||
976 | /* buffered writer ************************************************************/ | |
977 | ||
978 | /** @brief Shut down the writer | |
979 | * | |
980 | * This is called to shut down a writer. The error callback is not called | |
981 | * through any other path. Also we do not cancel @p fd from anywhere else, | |
982 | * though we might disable it. | |
983 | * | |
984 | * It has the signature of a timeout callback so that it can be called from a | |
985 | * time=0 timeout. | |
cb9a695c RK |
986 | * |
987 | * Calls @p callback with @p w->syntherr as the error code (which might be 0). | |
988 | */ | |
989 | static int writer_shutdown(ev_source *ev, | |
990 | const attribute((unused)) struct timeval *now, | |
991 | void *u) { | |
992 | ev_writer *w = u; | |
993 | ||
e4a9c7c5 | 994 | if(w->fd == -1) |
75d64210 | 995 | return 0; /* already shut down */ |
0fa83caa | 996 | D(("writer_shutdown fd=%d error=%d", w->fd, w->error)); |
cb9a695c | 997 | ev_timeout_cancel(ev, w->timeout); |
75d64210 | 998 | ev_fd_cancel(ev, ev_write, w->fd); |
cb9a695c | 999 | w->timeout = 0; |
75d64210 | 1000 | if(w->reader) { |
0fa83caa | 1001 | D(("found a tied reader")); |
75d64210 RK |
1002 | /* If there is a reader still around we just untie it */ |
1003 | w->reader->writer = 0; | |
1004 | shutdown(w->fd, SHUT_WR); /* there'll be no more writes */ | |
1005 | } else { | |
0fa83caa | 1006 | D(("no tied reader")); |
75d64210 RK |
1007 | /* There's no reader so we are free to close the FD */ |
1008 | xclose(w->fd); | |
1009 | } | |
e4a9c7c5 | 1010 | w->fd = -1; |
75d64210 | 1011 | return w->callback(ev, w->error, w->u); |
cb9a695c RK |
1012 | } |
1013 | ||
1014 | /** @brief Called when a writer's @p timebound expires */ | |
1015 | static int writer_timebound_exceeded(ev_source *ev, | |
75d64210 | 1016 | const struct timeval *now, |
cb9a695c RK |
1017 | void *u) { |
1018 | ev_writer *const w = u; | |
1019 | ||
d742bb47 RK |
1020 | if(!w->abandoned) { |
1021 | w->abandoned = 1; | |
2e9ba080 RK |
1022 | disorder_error(0, "abandoning writer '%s' because no writes within %ds", |
1023 | w->what, w->timebound); | |
d742bb47 RK |
1024 | w->error = ETIMEDOUT; |
1025 | } | |
75d64210 | 1026 | return writer_shutdown(ev, now, u); |
cb9a695c RK |
1027 | } |
1028 | ||
1029 | /** @brief Set the time bound callback (if not set already) */ | |
1030 | static void writer_set_timebound(ev_writer *w) { | |
1031 | if(w->timebound && !w->timeout) { | |
1032 | struct timeval when; | |
1033 | ev_source *const ev = w->ev; | |
1034 | ||
1035 | xgettimeofday(&when, 0); | |
1036 | when.tv_sec += w->timebound; | |
1037 | ev_timeout(ev, &w->timeout, &when, writer_timebound_exceeded, w); | |
1038 | } | |
1039 | } | |
1040 | ||
768d7355 | 1041 | /** @brief Called when a writer's file descriptor is writable */ |
460b9539 | 1042 | static int writer_callback(ev_source *ev, int fd, void *u) { |
cb9a695c | 1043 | ev_writer *const w = u; |
460b9539 | 1044 | int n; |
1045 | ||
1046 | n = write(fd, w->b.start, w->b.end - w->b.start); | |
1047 | D(("callback for writer fd %d, %ld bytes, n=%d, errno=%d", | |
1048 | fd, (long)(w->b.end - w->b.start), n, errno)); | |
1049 | if(n >= 0) { | |
75d64210 | 1050 | /* Consume bytes from the buffer */ |
460b9539 | 1051 | w->b.start += n; |
75d64210 | 1052 | /* Suppress any outstanding timeout */ |
cb9a695c RK |
1053 | ev_timeout_cancel(ev, w->timeout); |
1054 | w->timeout = 0; | |
460b9539 | 1055 | if(w->b.start == w->b.end) { |
75d64210 | 1056 | /* The buffer is empty */ |
460b9539 | 1057 | if(w->eof) { |
75d64210 RK |
1058 | /* We're done, we can shut down this writer */ |
1059 | w->error = 0; | |
1060 | return writer_shutdown(ev, 0, w); | |
460b9539 | 1061 | } else |
75d64210 RK |
1062 | /* There might be more to come but we don't need writer_callback() to |
1063 | * be called for the time being */ | |
460b9539 | 1064 | ev_fd_disable(ev, ev_write, fd); |
cb9a695c | 1065 | } else |
75d64210 RK |
1066 | /* The buffer isn't empty, set a timeout so we give up if we don't manage |
1067 | * to write some more within a reasonable time */ | |
cb9a695c | 1068 | writer_set_timebound(w); |
460b9539 | 1069 | } else { |
1070 | switch(errno) { | |
1071 | case EINTR: | |
1072 | case EAGAIN: | |
1073 | break; | |
1074 | default: | |
75d64210 RK |
1075 | w->error = errno; |
1076 | return writer_shutdown(ev, 0, w); | |
460b9539 | 1077 | } |
1078 | } | |
1079 | return 0; | |
1080 | } | |
1081 | ||
768d7355 RK |
1082 | /** @brief Write bytes to a writer's buffer |
1083 | * | |
1084 | * This is the sink write callback. | |
1085 | * | |
1086 | * Calls ev_fd_enable() if necessary (i.e. if the buffer was empty but | |
1087 | * now is not). | |
1088 | */ | |
460b9539 | 1089 | static int ev_writer_write(struct sink *sk, const void *s, int n) { |
1090 | ev_writer *w = (ev_writer *)sk; | |
cb9a695c RK |
1091 | |
1092 | if(!n) | |
1093 | return 0; /* avoid silliness */ | |
38b8221f | 1094 | if(w->fd == -1) |
2e9ba080 | 1095 | disorder_error(0, "ev_writer_write on %s after shutdown", w->what); |
75d64210 RK |
1096 | if(w->spacebound && w->b.end - w->b.start + n > w->spacebound) { |
1097 | /* The new buffer contents will exceed the space bound. We assume that the | |
cb9a695c RK |
1098 | * remote client has gone away and TCP hasn't noticed yet, or that it's got |
1099 | * hopelessly stuck. */ | |
d742bb47 RK |
1100 | if(!w->abandoned) { |
1101 | w->abandoned = 1; | |
2e9ba080 RK |
1102 | disorder_error(0, "abandoning writer '%s' because buffer has reached %td bytes", |
1103 | w->what, w->b.end - w->b.start); | |
d742bb47 RK |
1104 | ev_fd_disable(w->ev, ev_write, w->fd); |
1105 | w->error = EPIPE; | |
1106 | return ev_timeout(w->ev, 0, 0, writer_shutdown, w); | |
1107 | } else | |
1108 | return 0; | |
cb9a695c | 1109 | } |
75d64210 RK |
1110 | /* Make sure there is space */ |
1111 | buffer_space(&w->b, n); | |
1112 | /* If the buffer was formerly empty then we'll need to re-enable the FD */ | |
1113 | if(w->b.start == w->b.end) | |
1114 | ev_fd_enable(w->ev, ev_write, w->fd); | |
1115 | memcpy(w->b.end, s, n); | |
1116 | w->b.end += n; | |
1117 | /* Arrange a timeout if there wasn't one set already */ | |
cb9a695c | 1118 | writer_set_timebound(w); |
460b9539 | 1119 | return 0; |
1120 | } | |
1121 | ||
768d7355 RK |
1122 | /** @brief Create a new buffered writer |
1123 | * @param ev Event loop | |
1124 | * @param fd File descriptor to write to | |
1125 | * @param callback Called if an error occurs and when finished | |
1126 | * @param u Passed to @p callback | |
1127 | * @param what Text description | |
1128 | * @return New writer or @c NULL | |
75d64210 RK |
1129 | * |
1130 | * Writers own their file descriptor and close it when they have finished with | |
1131 | * it. | |
1132 | * | |
1133 | * If you pass the same fd to a reader and writer, you must tie them together | |
1134 | * with ev_tie(). | |
768d7355 | 1135 | */ |
460b9539 | 1136 | ev_writer *ev_writer_new(ev_source *ev, |
1137 | int fd, | |
1138 | ev_error_callback *callback, | |
e8c92ba7 RK |
1139 | void *u, |
1140 | const char *what) { | |
460b9539 | 1141 | ev_writer *w = xmalloc(sizeof *w); |
1142 | ||
1143 | D(("registering writer fd %d callback %p %p", fd, (void *)callback, u)); | |
1144 | w->s.write = ev_writer_write; | |
1145 | w->fd = fd; | |
1146 | w->callback = callback; | |
1147 | w->u = u; | |
1148 | w->ev = ev; | |
cb9a695c RK |
1149 | w->timebound = 10 * 60; |
1150 | w->spacebound = 512 * 1024; | |
1151 | w->what = what; | |
e8c92ba7 | 1152 | if(ev_fd(ev, ev_write, fd, writer_callback, w, what)) |
460b9539 | 1153 | return 0; |
75d64210 | 1154 | /* Buffer is initially empty so we don't want a callback */ |
460b9539 | 1155 | ev_fd_disable(ev, ev_write, fd); |
1156 | return w; | |
1157 | } | |
1158 | ||
cb9a695c RK |
1159 | /** @brief Get/set the time bound |
1160 | * @param w Writer | |
1161 | * @param new_time_bound New bound or -1 for no change | |
1162 | * @return Latest time bound | |
1163 | * | |
1164 | * If @p new_time_bound is negative then the current time bound is returned. | |
1165 | * Otherwise it is set and the new value returned. | |
1166 | * | |
1167 | * The time bound is the number of seconds allowed between writes. If it takes | |
1168 | * longer than this to flush a buffer then the peer will be assumed to be dead | |
1169 | * and an error will be synthesized. 0 means "don't care". The default time | |
1170 | * bound is 10 minutes. | |
1171 | * | |
1172 | * Note that this value does not take into account kernel buffering and | |
1173 | * timeouts. | |
1174 | */ | |
1175 | int ev_writer_time_bound(ev_writer *w, | |
1176 | int new_time_bound) { | |
1177 | if(new_time_bound >= 0) | |
1178 | w->timebound = new_time_bound; | |
1179 | return w->timebound; | |
1180 | } | |
1181 | ||
1182 | /** @brief Get/set the space bound | |
1183 | * @param w Writer | |
1184 | * @param new_space_bound New bound or -1 for no change | |
1185 | * @return Latest space bound | |
1186 | * | |
1187 | * If @p new_space_bound is negative then the current space bound is returned. | |
1188 | * Otherwise it is set and the new value returned. | |
1189 | * | |
1190 | * The space bound is the number of bytes allowed between in the buffer. If | |
1191 | * the buffer exceeds this size an error will be synthesized. 0 means "don't | |
1192 | * care". The default space bound is 512Kbyte. | |
1193 | * | |
1194 | * Note that this value does not take into account kernel buffering. | |
1195 | */ | |
1196 | int ev_writer_space_bound(ev_writer *w, | |
1197 | int new_space_bound) { | |
1198 | if(new_space_bound >= 0) | |
1199 | w->spacebound = new_space_bound; | |
1200 | return w->spacebound; | |
1201 | } | |
1202 | ||
768d7355 RK |
1203 | /** @brief Return the sink associated with a writer |
1204 | * @param w Writer | |
1205 | * @return Pointer to sink | |
1206 | * | |
1207 | * Writing to the sink will arrange for those bytes to be written to the file | |
1208 | * descriptor as and when it is writable. | |
1209 | */ | |
460b9539 | 1210 | struct sink *ev_writer_sink(ev_writer *w) { |
f6033c46 | 1211 | if(!w) |
2e9ba080 | 1212 | disorder_fatal(0, "ev_write_sink called with null writer"); |
460b9539 | 1213 | return &w->s; |
1214 | } | |
1215 | ||
768d7355 RK |
1216 | /** @brief Close a writer |
1217 | * @param w Writer to close | |
1218 | * @return 0 on success, non-0 on error | |
1219 | * | |
1220 | * Close a writer. No more bytes should be written to its sink. | |
1221 | * | |
1222 | * When the last byte has been written the callback will be called with an | |
1223 | * error code of 0. It is guaranteed that this will NOT happen before | |
1224 | * ev_writer_close() returns (although the file descriptor for the writer might | |
1225 | * be cancelled by the time it returns). | |
1226 | */ | |
460b9539 | 1227 | int ev_writer_close(ev_writer *w) { |
1228 | D(("close writer fd %d", w->fd)); | |
75d64210 RK |
1229 | if(w->eof) |
1230 | return 0; /* already closed */ | |
460b9539 | 1231 | w->eof = 1; |
1232 | if(w->b.start == w->b.end) { | |
75d64210 RK |
1233 | /* We're already finished */ |
1234 | w->error = 0; /* no error */ | |
460b9539 | 1235 | return ev_timeout(w->ev, 0, 0, writer_shutdown, w); |
1236 | } | |
1237 | return 0; | |
1238 | } | |
1239 | ||
768d7355 RK |
1240 | /** @brief Attempt to flush a writer |
1241 | * @param w Writer to flush | |
1242 | * @return 0 on success, non-0 on error | |
1243 | * | |
1244 | * Does a speculative write of any buffered data. Does not block if it cannot | |
1245 | * be written. | |
1246 | */ | |
460b9539 | 1247 | int ev_writer_flush(ev_writer *w) { |
1248 | return writer_callback(w->ev, w->fd, w); | |
1249 | } | |
1250 | ||
1251 | /* buffered reader ************************************************************/ | |
1252 | ||
49a773eb | 1253 | /** @brief Shut down a reader |
75d64210 RK |
1254 | * |
1255 | * This is the only path through which we cancel and close the file descriptor. | |
1256 | * As with the writer case it is given timeout signature to allow it be | |
1257 | * deferred to the next iteration of the event loop. | |
1258 | * | |
1259 | * We only call @p error_callback if @p error is nonzero (unlike the writer | |
1260 | * case). | |
1261 | */ | |
1262 | static int reader_shutdown(ev_source *ev, | |
1263 | const attribute((unused)) struct timeval *now, | |
1264 | void *u) { | |
1265 | ev_reader *const r = u; | |
1266 | ||
1267 | if(r->fd == -1) | |
1268 | return 0; /* already shut down */ | |
0fa83caa | 1269 | D(("reader_shutdown fd=%d", r->fd)); |
75d64210 RK |
1270 | ev_fd_cancel(ev, ev_read, r->fd); |
1271 | r->eof = 1; | |
1272 | if(r->writer) { | |
0fa83caa | 1273 | D(("found a tied writer")); |
75d64210 RK |
1274 | /* If there is a writer still around we just untie it */ |
1275 | r->writer->reader = 0; | |
1276 | shutdown(r->fd, SHUT_RD); /* there'll be no more reads */ | |
1277 | } else { | |
0fa83caa | 1278 | D(("no tied writer found")); |
75d64210 RK |
1279 | /* There's no writer so we are free to close the FD */ |
1280 | xclose(r->fd); | |
1281 | } | |
1282 | r->fd = -1; | |
1283 | if(r->error) | |
1284 | return r->error_callback(ev, r->error, r->u); | |
1285 | else | |
1286 | return 0; | |
1287 | } | |
460b9539 | 1288 | |
768d7355 | 1289 | /** @brief Called when a reader's @p fd is readable */ |
460b9539 | 1290 | static int reader_callback(ev_source *ev, int fd, void *u) { |
1291 | ev_reader *r = u; | |
1292 | int n; | |
1293 | ||
1294 | buffer_space(&r->b, 1); | |
1295 | n = read(fd, r->b.end, r->b.top - r->b.end); | |
1296 | D(("read fd %d buffer %d returned %d errno %d", | |
1297 | fd, (int)(r->b.top - r->b.end), n, errno)); | |
1298 | if(n > 0) { | |
1299 | r->b.end += n; | |
75d64210 | 1300 | return r->callback(ev, r, r->b.start, r->b.end - r->b.start, 0, r->u); |
460b9539 | 1301 | } else if(n == 0) { |
75d64210 RK |
1302 | /* No more read callbacks needed */ |
1303 | ev_fd_disable(r->ev, ev_read, r->fd); | |
1304 | ev_timeout(r->ev, 0, 0, reader_shutdown, r); | |
1305 | /* Pass the remaining data and an eof indicator to the user */ | |
1306 | return r->callback(ev, r, r->b.start, r->b.end - r->b.start, 1, r->u); | |
460b9539 | 1307 | } else { |
1308 | switch(errno) { | |
1309 | case EINTR: | |
1310 | case EAGAIN: | |
1311 | break; | |
1312 | default: | |
75d64210 RK |
1313 | /* Fatal error, kill the reader now */ |
1314 | r->error = errno; | |
1315 | return reader_shutdown(ev, 0, r); | |
460b9539 | 1316 | } |
1317 | } | |
1318 | return 0; | |
1319 | } | |
1320 | ||
768d7355 RK |
1321 | /** @brief Create a new buffered reader |
1322 | * @param ev Event loop | |
1323 | * @param fd File descriptor to read from | |
1324 | * @param callback Called when new data is available | |
1325 | * @param error_callback Called if an error occurs | |
1326 | * @param u Passed to callbacks | |
1327 | * @param what Text description | |
1328 | * @return New reader or @c NULL | |
75d64210 RK |
1329 | * |
1330 | * Readers own their fd and close it when they are finished with it. | |
1331 | * | |
1332 | * If you pass the same fd to a reader and writer, you must tie them together | |
1333 | * with ev_tie(). | |
768d7355 | 1334 | */ |
460b9539 | 1335 | ev_reader *ev_reader_new(ev_source *ev, |
1336 | int fd, | |
1337 | ev_reader_callback *callback, | |
1338 | ev_error_callback *error_callback, | |
e8c92ba7 RK |
1339 | void *u, |
1340 | const char *what) { | |
460b9539 | 1341 | ev_reader *r = xmalloc(sizeof *r); |
1342 | ||
1343 | D(("registering reader fd %d callback %p %p %p", | |
1344 | fd, (void *)callback, (void *)error_callback, u)); | |
1345 | r->fd = fd; | |
1346 | r->callback = callback; | |
1347 | r->error_callback = error_callback; | |
1348 | r->u = u; | |
1349 | r->ev = ev; | |
e8c92ba7 | 1350 | if(ev_fd(ev, ev_read, fd, reader_callback, r, what)) |
460b9539 | 1351 | return 0; |
1352 | return r; | |
1353 | } | |
1354 | ||
1355 | void ev_reader_buffer(ev_reader *r, size_t nbytes) { | |
1356 | buffer_space(&r->b, nbytes - (r->b.end - r->b.start)); | |
1357 | } | |
1358 | ||
768d7355 RK |
1359 | /** @brief Consume @p n bytes from the reader's buffer |
1360 | * @param r Reader | |
1361 | * @param n Number of bytes to consume | |
1362 | * | |
1363 | * Tells the reader than the next @p n bytes have been dealt with and can now | |
1364 | * be discarded. | |
1365 | */ | |
460b9539 | 1366 | void ev_reader_consume(ev_reader *r, size_t n) { |
1367 | r->b.start += n; | |
1368 | } | |
1369 | ||
768d7355 RK |
1370 | /** @brief Cancel a reader |
1371 | * @param r Reader | |
1372 | * @return 0 on success, non-0 on error | |
75d64210 RK |
1373 | * |
1374 | * No further callbacks will be made, and the FD will be closed (in a later | |
1375 | * iteration of the event loop). | |
768d7355 | 1376 | */ |
460b9539 | 1377 | int ev_reader_cancel(ev_reader *r) { |
1378 | D(("cancel reader fd %d", r->fd)); | |
75d64210 RK |
1379 | if(r->fd == -1) |
1380 | return 0; /* already thoroughly cancelled */ | |
1381 | ev_fd_disable(r->ev, ev_read, r->fd); | |
1382 | return ev_timeout(r->ev, 0, 0, reader_shutdown, r); | |
460b9539 | 1383 | } |
1384 | ||
768d7355 RK |
1385 | /** @brief Temporarily disable a reader |
1386 | * @param r Reader | |
1387 | * @return 0 on success, non-0 on error | |
1388 | * | |
1389 | * No further callbacks for this reader will be made. Re-enable with | |
1390 | * ev_reader_enable(). | |
1391 | */ | |
460b9539 | 1392 | int ev_reader_disable(ev_reader *r) { |
1393 | D(("disable reader fd %d", r->fd)); | |
75d64210 | 1394 | return ev_fd_disable(r->ev, ev_read, r->fd); |
460b9539 | 1395 | } |
1396 | ||
768d7355 | 1397 | /** @brief Called from ev_run() for ev_reader_incomplete() */ |
460b9539 | 1398 | static int reader_continuation(ev_source attribute((unused)) *ev, |
1399 | const attribute((unused)) struct timeval *now, | |
1400 | void *u) { | |
1401 | ev_reader *r = u; | |
1402 | ||
1403 | D(("reader continuation callback fd %d", r->fd)); | |
75d64210 RK |
1404 | /* If not at EOF turn the FD back on */ |
1405 | if(!r->eof) | |
1406 | if(ev_fd_enable(r->ev, ev_read, r->fd)) | |
1407 | return -1; | |
1408 | /* We're already in a timeout callback so there's no reason we can't call the | |
1409 | * user callback directly (compare ev_reader_enable()). */ | |
1410 | return r->callback(ev, r, r->b.start, r->b.end - r->b.start, r->eof, r->u); | |
460b9539 | 1411 | } |
1412 | ||
768d7355 RK |
1413 | /** @brief Arrange another callback |
1414 | * @param r reader | |
1415 | * @return 0 on success, non-0 on error | |
1416 | * | |
1417 | * Indicates that the reader can process more input but would like to yield to | |
1418 | * other clients of the event loop. Input will be disabled but it will be | |
1419 | * re-enabled on the next iteration of the event loop and the read callback | |
1420 | * will be called again (even if no further bytes are available). | |
1421 | */ | |
460b9539 | 1422 | int ev_reader_incomplete(ev_reader *r) { |
1423 | if(ev_fd_disable(r->ev, ev_read, r->fd)) return -1; | |
1424 | return ev_timeout(r->ev, 0, 0, reader_continuation, r); | |
1425 | } | |
1426 | ||
1427 | static int reader_enabled(ev_source *ev, | |
1428 | const attribute((unused)) struct timeval *now, | |
1429 | void *u) { | |
1430 | ev_reader *r = u; | |
1431 | ||
1432 | D(("reader enabled callback fd %d", r->fd)); | |
75d64210 | 1433 | return r->callback(ev, r, r->b.start, r->b.end - r->b.start, r->eof, r->u); |
460b9539 | 1434 | } |
1435 | ||
768d7355 RK |
1436 | /** @brief Re-enable reading |
1437 | * @param r reader | |
1438 | * @return 0 on success, non-0 on error | |
1439 | * | |
1440 | * If there is unconsumed data then you get a callback next time round the | |
1441 | * event loop even if nothing new has been read. | |
1442 | * | |
1443 | * The idea is in your read callback you come across a line (or whatever) that | |
1444 | * can't be processed immediately. So you set up processing and disable | |
1445 | * reading with ev_reader_disable(). Later when you finish processing you | |
1446 | * re-enable. You'll automatically get another callback directly from the | |
1447 | * event loop (i.e. not from inside ev_reader_enable()) so you can handle the | |
1448 | * next line (or whatever) if the whole thing has in fact already arrived. | |
75d64210 RK |
1449 | * |
1450 | * The difference between this process and calling ev_reader_incomplete() is | |
1451 | * ev_reader_incomplete() deals with the case where you can process now but | |
1452 | * would rather yield to other clients of the event loop, while using | |
1453 | * ev_reader_disable() and ev_reader_enable() deals with the case where you | |
1454 | * cannot process input yet because some other process is actually not | |
1455 | * complete. | |
768d7355 | 1456 | */ |
460b9539 | 1457 | int ev_reader_enable(ev_reader *r) { |
1458 | D(("enable reader fd %d", r->fd)); | |
75d64210 RK |
1459 | |
1460 | /* First if we're not at EOF then we re-enable reading */ | |
1461 | if(!r->eof) | |
1462 | if(ev_fd_enable(r->ev, ev_read, r->fd)) | |
1463 | return -1; | |
1464 | /* Arrange another callback next time round the event loop */ | |
1465 | return ev_timeout(r->ev, 0, 0, reader_enabled, r); | |
1466 | } | |
1467 | ||
1468 | /** @brief Tie a reader and a writer together | |
1469 | * @param r Reader | |
1470 | * @param w Writer | |
1471 | * @return 0 on success, non-0 on error | |
1472 | * | |
1473 | * This function must be called if @p r and @p w share a file descritptor. | |
1474 | */ | |
1475 | int ev_tie(ev_reader *r, ev_writer *w) { | |
1476 | assert(r->writer == 0); | |
1477 | assert(w->reader == 0); | |
1478 | r->writer = w; | |
1479 | w->reader = r; | |
1480 | return 0; | |
460b9539 | 1481 | } |
1482 | ||
1483 | /* | |
1484 | Local Variables: | |
1485 | c-basic-offset:2 | |
1486 | comment-column:40 | |
1487 | fill-column:79 | |
1488 | End: | |
1489 | */ |