You can not select more than 25 topics Topics must start with a letter or number, can include dashes ('-') and can be up to 35 characters long.

279 lines
9.6 KiB

11 years ago
  1. /*
  2. * libev epoll fd activity backend
  3. *
  4. * Copyright (c) 2007,2008,2009,2010,2011 Marc Alexander Lehmann <libev@schmorp.de>
  5. * All rights reserved.
  6. *
  7. * Redistribution and use in source and binary forms, with or without modifica-
  8. * tion, are permitted provided that the following conditions are met:
  9. *
  10. * 1. Redistributions of source code must retain the above copyright notice,
  11. * this list of conditions and the following disclaimer.
  12. *
  13. * 2. Redistributions in binary form must reproduce the above copyright
  14. * notice, this list of conditions and the following disclaimer in the
  15. * documentation and/or other materials provided with the distribution.
  16. *
  17. * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR IMPLIED
  18. * WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MER-
  19. * CHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO
  20. * EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPE-
  21. * CIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
  22. * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS;
  23. * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY,
  24. * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTH-
  25. * ERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED
  26. * OF THE POSSIBILITY OF SUCH DAMAGE.
  27. *
  28. * Alternatively, the contents of this file may be used under the terms of
  29. * the GNU General Public License ("GPL") version 2 or any later version,
  30. * in which case the provisions of the GPL are applicable instead of
  31. * the above. If you wish to allow the use of your version of this file
  32. * only under the terms of the GPL and not to allow others to use your
  33. * version of this file under the BSD license, indicate your decision
  34. * by deleting the provisions above and replace them with the notice
  35. * and other provisions required by the GPL. If you do not delete the
  36. * provisions above, a recipient may use your version of this file under
  37. * either the BSD or the GPL.
  38. */
  39. /*
  40. * general notes about epoll:
  41. *
  42. * a) epoll silently removes fds from the fd set. as nothing tells us
  43. * that an fd has been removed otherwise, we have to continually
  44. * "rearm" fds that we suspect *might* have changed (same
  45. * problem with kqueue, but much less costly there).
  46. * b) the fact that ADD != MOD creates a lot of extra syscalls due to a)
  47. * and seems not to have any advantage.
  48. * c) the inability to handle fork or file descriptors (think dup)
  49. * limits the applicability over poll, so this is not a generic
  50. * poll replacement.
  51. * d) epoll doesn't work the same as select with many file descriptors
  52. * (such as files). while not critical, no other advanced interface
  53. * seems to share this (rather non-unixy) limitation.
  54. * e) epoll claims to be embeddable, but in practise you never get
  55. * a ready event for the epoll fd (broken: <=2.6.26, working: >=2.6.32).
  56. * f) epoll_ctl returning EPERM means the fd is always ready.
  57. *
  58. * lots of "weird code" and complication handling in this file is due
  59. * to these design problems with epoll, as we try very hard to avoid
  60. * epoll_ctl syscalls for common usage patterns and handle the breakage
  61. * ensuing from receiving events for closed and otherwise long gone
  62. * file descriptors.
  63. */
  64. #include <sys/epoll.h>
  65. #define EV_EMASK_EPERM 0x80
  66. static void
  67. epoll_modify (EV_P_ int fd, int oev, int nev)
  68. {
  69. struct epoll_event ev;
  70. unsigned char oldmask;
  71. /*
  72. * we handle EPOLL_CTL_DEL by ignoring it here
  73. * on the assumption that the fd is gone anyways
  74. * if that is wrong, we have to handle the spurious
  75. * event in epoll_poll.
  76. * if the fd is added again, we try to ADD it, and, if that
  77. * fails, we assume it still has the same eventmask.
  78. */
  79. if (!nev)
  80. return;
  81. oldmask = anfds [fd].emask;
  82. anfds [fd].emask = nev;
  83. /* store the generation counter in the upper 32 bits, the fd in the lower 32 bits */
  84. ev.data.u64 = (uint64_t)(uint32_t)fd
  85. | ((uint64_t)(uint32_t)++anfds [fd].egen << 32);
  86. ev.events = (nev & EV_READ ? EPOLLIN : 0)
  87. | (nev & EV_WRITE ? EPOLLOUT : 0);
  88. if (expect_true (!epoll_ctl (backend_fd, oev && oldmask != nev ? EPOLL_CTL_MOD : EPOLL_CTL_ADD, fd, &ev)))
  89. return;
  90. if (expect_true (errno == ENOENT))
  91. {
  92. /* if ENOENT then the fd went away, so try to do the right thing */
  93. if (!nev)
  94. goto dec_egen;
  95. if (!epoll_ctl (backend_fd, EPOLL_CTL_ADD, fd, &ev))
  96. return;
  97. }
  98. else if (expect_true (errno == EEXIST))
  99. {
  100. /* EEXIST means we ignored a previous DEL, but the fd is still active */
  101. /* if the kernel mask is the same as the new mask, we assume it hasn't changed */
  102. if (oldmask == nev)
  103. goto dec_egen;
  104. if (!epoll_ctl (backend_fd, EPOLL_CTL_MOD, fd, &ev))
  105. return;
  106. }
  107. else if (expect_true (errno == EPERM))
  108. {
  109. /* EPERM means the fd is always ready, but epoll is too snobbish */
  110. /* to handle it, unlike select or poll. */
  111. anfds [fd].emask = EV_EMASK_EPERM;
  112. /* add fd to epoll_eperms, if not already inside */
  113. if (!(oldmask & EV_EMASK_EPERM))
  114. {
  115. array_needsize (int, epoll_eperms, epoll_epermmax, epoll_epermcnt + 1, EMPTY2);
  116. epoll_eperms [epoll_epermcnt++] = fd;
  117. }
  118. return;
  119. }
  120. fd_kill (EV_A_ fd);
  121. dec_egen:
  122. /* we didn't successfully call epoll_ctl, so decrement the generation counter again */
  123. --anfds [fd].egen;
  124. }
  125. static void
  126. epoll_poll (EV_P_ ev_tstamp timeout)
  127. {
  128. int i;
  129. int eventcnt;
  130. if (expect_false (epoll_epermcnt))
  131. timeout = 0.;
  132. /* epoll wait times cannot be larger than (LONG_MAX - 999UL) / HZ msecs, which is below */
  133. /* the default libev max wait time, however. */
  134. EV_RELEASE_CB;
  135. eventcnt = epoll_wait (backend_fd, epoll_events, epoll_eventmax, timeout * 1e3);
  136. EV_ACQUIRE_CB;
  137. if (expect_false (eventcnt < 0))
  138. {
  139. if (errno != EINTR)
  140. ev_syserr ("(libev) epoll_wait");
  141. return;
  142. }
  143. for (i = 0; i < eventcnt; ++i)
  144. {
  145. struct epoll_event *ev = epoll_events + i;
  146. int fd = (uint32_t)ev->data.u64; /* mask out the lower 32 bits */
  147. int want = anfds [fd].events;
  148. int got = (ev->events & (EPOLLOUT | EPOLLERR | EPOLLHUP) ? EV_WRITE : 0)
  149. | (ev->events & (EPOLLIN | EPOLLERR | EPOLLHUP) ? EV_READ : 0);
  150. /*
  151. * check for spurious notification.
  152. * this only finds spurious notifications on egen updates
  153. * other spurious notifications will be found by epoll_ctl, below
  154. * we assume that fd is always in range, as we never shrink the anfds array
  155. */
  156. if (expect_false ((uint32_t)anfds [fd].egen != (uint32_t)(ev->data.u64 >> 32)))
  157. {
  158. /* recreate kernel state */
  159. postfork = 1;
  160. continue;
  161. }
  162. if (expect_false (got & ~want))
  163. {
  164. anfds [fd].emask = want;
  165. /*
  166. * we received an event but are not interested in it, try mod or del
  167. * this often happens because we optimistically do not unregister fds
  168. * when we are no longer interested in them, but also when we get spurious
  169. * notifications for fds from another process. this is partially handled
  170. * above with the gencounter check (== our fd is not the event fd), and
  171. * partially here, when epoll_ctl returns an error (== a child has the fd
  172. * but we closed it).
  173. */
  174. ev->events = (want & EV_READ ? EPOLLIN : 0)
  175. | (want & EV_WRITE ? EPOLLOUT : 0);
  176. /* pre-2.6.9 kernels require a non-null pointer with EPOLL_CTL_DEL, */
  177. /* which is fortunately easy to do for us. */
  178. if (epoll_ctl (backend_fd, want ? EPOLL_CTL_MOD : EPOLL_CTL_DEL, fd, ev))
  179. {
  180. postfork = 1; /* an error occurred, recreate kernel state */
  181. continue;
  182. }
  183. }
  184. fd_event (EV_A_ fd, got);
  185. }
  186. /* if the receive array was full, increase its size */
  187. if (expect_false (eventcnt == epoll_eventmax))
  188. {
  189. ev_free (epoll_events);
  190. epoll_eventmax = array_nextsize (sizeof (struct epoll_event), epoll_eventmax, epoll_eventmax + 1);
  191. epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
  192. }
  193. /* now synthesize events for all fds where epoll fails, while select works... */
  194. for (i = epoll_epermcnt; i--; )
  195. {
  196. int fd = epoll_eperms [i];
  197. unsigned char events = anfds [fd].events & (EV_READ | EV_WRITE);
  198. if (anfds [fd].emask & EV_EMASK_EPERM && events)
  199. fd_event (EV_A_ fd, events);
  200. else
  201. epoll_eperms [i] = epoll_eperms [--epoll_epermcnt];
  202. }
  203. }
  204. int inline_size
  205. epoll_init (EV_P_ int flags)
  206. {
  207. #ifdef EPOLL_CLOEXEC
  208. backend_fd = epoll_create1 (EPOLL_CLOEXEC);
  209. if (backend_fd < 0 && (errno == EINVAL || errno == ENOSYS))
  210. #endif
  211. backend_fd = epoll_create (256);
  212. if (backend_fd < 0)
  213. return 0;
  214. fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
  215. backend_mintime = 1e-3; /* epoll does sometimes return early, this is just to avoid the worst */
  216. backend_modify = epoll_modify;
  217. backend_poll = epoll_poll;
  218. epoll_eventmax = 64; /* initial number of events receivable per poll */
  219. epoll_events = (struct epoll_event *)ev_malloc (sizeof (struct epoll_event) * epoll_eventmax);
  220. return EVBACKEND_EPOLL;
  221. }
  222. void inline_size
  223. epoll_destroy (EV_P)
  224. {
  225. ev_free (epoll_events);
  226. array_free (epoll_eperm, EMPTY);
  227. }
  228. void inline_size
  229. epoll_fork (EV_P)
  230. {
  231. close (backend_fd);
  232. while ((backend_fd = epoll_create (256)) < 0)
  233. ev_syserr ("(libev) epoll_create");
  234. fcntl (backend_fd, F_SETFD, FD_CLOEXEC);
  235. fd_rearm_all (EV_A);
  236. }