| 1 | /* |
| 2 | * Copyright (C) 2011 - David Goulet <david.goulet@polymtl.ca> |
| 3 | * |
| 4 | * This program is free software; you can redistribute it and/or modify |
| 5 | * it under the terms of the GNU General Public License, version 2 only, |
| 6 | * as published by the Free Software Foundation. |
| 7 | * |
| 8 | * This program is distributed in the hope that it will be useful, but WITHOUT |
| 9 | * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
| 10 | * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for |
| 11 | * more details. |
| 12 | * |
| 13 | * You should have received a copy of the GNU General Public License along |
| 14 | * with this program; if not, write to the Free Software Foundation, Inc., |
| 15 | * 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA. |
| 16 | */ |
| 17 | |
| 18 | #ifndef _LTT_POLL_H |
| 19 | #define _LTT_POLL_H |
| 20 | |
| 21 | #include <assert.h> |
| 22 | #include <string.h> |
| 23 | #include <unistd.h> |
| 24 | |
| 25 | #include <common/common.h> |
| 26 | |
| 27 | /* |
| 28 | * Maximum number of fd we can monitor. |
| 29 | * |
| 30 | * For epoll(7), /proc/sys/fs/epoll/max_user_watches (since Linux 2.6.28) will |
| 31 | * be used for the maximum size of the poll set. If this interface is not |
| 32 | * available, according to the manpage, the max_user_watches value is 1/25 (4%) |
| 33 | * of the available low memory divided by the registration cost in bytes which |
| 34 | * is 90 bytes on a 32-bit kernel and 160 bytes on a 64-bit kernel. |
| 35 | * |
| 36 | * For poll(2), the max fds must not exceed RLIMIT_NOFILE given by |
| 37 | * getrlimit(2). |
| 38 | */ |
| 39 | extern unsigned int poll_max_size; |
| 40 | |
| 41 | /* |
| 42 | * Used by lttng_poll_clean to free the events structure in a lttng_poll_event. |
| 43 | */ |
| 44 | static inline void __lttng_poll_free(void *events) |
| 45 | { |
| 46 | free(events); |
| 47 | } |
| 48 | |
| 49 | /* |
| 50 | * epoll(7) implementation. |
| 51 | */ |
| 52 | #ifdef HAVE_EPOLL |
| 53 | #include <sys/epoll.h> |
| 54 | #include <stdio.h> |
| 55 | #include <features.h> |
| 56 | #include <common/compat/fcntl.h> |
| 57 | |
| 58 | /* See man epoll(7) for this define path */ |
| 59 | #define COMPAT_EPOLL_PROC_PATH "/proc/sys/fs/epoll/max_user_watches" |
| 60 | |
| 61 | enum { |
| 62 | /* Polling variables compatibility for epoll */ |
| 63 | LPOLLIN = EPOLLIN, |
| 64 | LPOLLPRI = EPOLLPRI, |
| 65 | LPOLLOUT = EPOLLOUT, |
| 66 | LPOLLRDNORM = EPOLLRDNORM, |
| 67 | LPOLLRDBAND = EPOLLRDBAND, |
| 68 | LPOLLWRNORM = EPOLLWRNORM, |
| 69 | LPOLLWRBAND = EPOLLWRBAND, |
| 70 | LPOLLMSG = EPOLLMSG, |
| 71 | LPOLLERR = EPOLLERR, |
| 72 | LPOLLHUP = EPOLLHUP, |
| 73 | LPOLLNVAL = EPOLLHUP, |
| 74 | LPOLLRDHUP = EPOLLRDHUP, |
| 75 | /* Close on exec feature of epoll */ |
| 76 | #if defined(HAVE_EPOLL_CREATE1) && defined(EPOLL_CLOEXEC) |
| 77 | LTTNG_CLOEXEC = EPOLL_CLOEXEC, |
| 78 | #else |
| 79 | /* |
| 80 | * EPOLL_CLOEXEC was added in glibc 2.8 (usually used in conjunction with |
| 81 | * epoll_create1(..)), but since neither EPOLL_CLOEXEC exists nor |
| 82 | * epoll_create1(..), we set it to FD_CLOEXEC so that we can pass it |
| 83 | * directly to fcntl(..) instead. |
| 84 | */ |
| 85 | LTTNG_CLOEXEC = FD_CLOEXEC, |
| 86 | #endif |
| 87 | }; |
| 88 | |
| 89 | struct compat_epoll_event { |
| 90 | int epfd; |
| 91 | uint32_t nb_fd; /* Current number of fd in events */ |
| 92 | uint32_t alloc_size; /* Size of events array */ |
| 93 | uint32_t init_size; /* Initial size of events array */ |
| 94 | struct epoll_event *events; |
| 95 | }; |
| 96 | #define lttng_poll_event compat_epoll_event |
| 97 | |
| 98 | static inline int __lttng_epoll_get_prev_fd(struct lttng_poll_event *events, |
| 99 | int index, uint32_t nb_fd) |
| 100 | { |
| 101 | assert(events); |
| 102 | assert(index != nb_fd); |
| 103 | |
| 104 | if (index == 0 || nb_fd == 0) { |
| 105 | return -1; |
| 106 | } else { |
| 107 | return events->events[index - 1].data.fd; |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | /* |
| 112 | * For the following calls, consider 'e' to be a lttng_poll_event pointer and i |
| 113 | * being the index of the events array. |
| 114 | */ |
| 115 | #define LTTNG_POLL_GETFD(e, i) LTTNG_REF(e)->events[i].data.fd |
| 116 | #define LTTNG_POLL_GETEV(e, i) LTTNG_REF(e)->events[i].events |
| 117 | #define LTTNG_POLL_GETNB(e) LTTNG_REF(e)->nb_fd |
| 118 | #define LTTNG_POLL_GETSZ(e) LTTNG_REF(e)->events_size |
| 119 | #define LTTNG_POLL_GET_PREV_FD(e, i, nb_fd) \ |
| 120 | __lttng_epoll_get_prev_fd(LTTNG_REF(e), i, nb_fd) |
| 121 | |
| 122 | /* |
| 123 | * Create the epoll set. No memory allocation is done here. |
| 124 | */ |
| 125 | extern int compat_epoll_create(struct lttng_poll_event *events, |
| 126 | int size, int flags); |
| 127 | #define lttng_poll_create(events, size, flags) \ |
| 128 | compat_epoll_create(events, size, flags) |
| 129 | |
| 130 | #if defined(HAVE_EPOLL_CREATE1) && defined(EPOLL_CLOEXEC) |
| 131 | static inline int compat_glibc_epoll_create(int size __attribute__((unused)), |
| 132 | int flags) |
| 133 | { |
| 134 | return epoll_create1(flags); |
| 135 | } |
| 136 | #else |
| 137 | static inline int compat_glibc_epoll_create(int size, int flags) |
| 138 | { |
| 139 | /* |
| 140 | * epoll_create1 was added in glibc 2.9, but unfortunatly reverting to |
| 141 | * epoll_create(..) also means that we lose the possibility to |
| 142 | * directly set the EPOLL_CLOEXEC, so try and do it anyway but through |
| 143 | * fcntl(..). |
| 144 | */ |
| 145 | int efd = epoll_create(size); |
| 146 | assert(fcntl(efd, F_SETFD, flags) != -1); |
| 147 | return efd; |
| 148 | } |
| 149 | #endif |
| 150 | |
| 151 | /* |
| 152 | * Wait on epoll set with the number of fd registered to the lttng_poll_event |
| 153 | * data structure (events). |
| 154 | */ |
| 155 | extern int compat_epoll_wait(struct lttng_poll_event *events, int timeout); |
| 156 | #define lttng_poll_wait(events, timeout) \ |
| 157 | compat_epoll_wait(events, timeout) |
| 158 | |
| 159 | /* |
| 160 | * Add a fd to the epoll set and resize the epoll_event structure if needed. |
| 161 | */ |
| 162 | extern int compat_epoll_add(struct lttng_poll_event *events, |
| 163 | int fd, uint32_t req_events); |
| 164 | #define lttng_poll_add(events, fd, req_events) \ |
| 165 | compat_epoll_add(events, fd, req_events) |
| 166 | |
| 167 | /* |
| 168 | * Remove a fd from the epoll set. |
| 169 | */ |
| 170 | extern int compat_epoll_del(struct lttng_poll_event *events, int fd); |
| 171 | #define lttng_poll_del(events, fd) \ |
| 172 | compat_epoll_del(events, fd) |
| 173 | |
| 174 | /* |
| 175 | * Set up the poll set limits variable poll_max_size |
| 176 | */ |
| 177 | extern int compat_epoll_set_max_size(void); |
| 178 | #define lttng_poll_set_max_size() \ |
| 179 | compat_epoll_set_max_size() |
| 180 | |
| 181 | /* |
| 182 | * This function memset with zero the structure since it can be reused at each |
| 183 | * round of a main loop. Being in a loop and using a non static number of fds, |
| 184 | * this function must be called to insure coherent events with associted fds. |
| 185 | */ |
| 186 | static inline void lttng_poll_reset(struct lttng_poll_event *events) |
| 187 | { |
| 188 | if (events && events->events) { |
| 189 | memset(events->events, 0, |
| 190 | events->nb_fd * sizeof(struct epoll_event)); |
| 191 | } |
| 192 | } |
| 193 | |
| 194 | /* |
| 195 | * Initialize an already allocated poll event data structure. For epoll(), the |
| 196 | * epfd is set to -1 to indicate that it's not usable. |
| 197 | */ |
| 198 | static inline void lttng_poll_init(struct lttng_poll_event *events) |
| 199 | { |
| 200 | memset(events, 0, sizeof(struct lttng_poll_event)); |
| 201 | /* Set fd to -1 so if clean before created, we don't close 0. */ |
| 202 | events->epfd = -1; |
| 203 | } |
| 204 | |
| 205 | /* |
| 206 | * Clean the events structure of a lttng_poll_event. It's the caller |
| 207 | * responsability to free the lttng_poll_event memory. |
| 208 | */ |
| 209 | static inline void lttng_poll_clean(struct lttng_poll_event *events) |
| 210 | { |
| 211 | int ret; |
| 212 | |
| 213 | if (!events) { |
| 214 | return; |
| 215 | } |
| 216 | |
| 217 | if (events->epfd >= 0) { |
| 218 | ret = close(events->epfd); |
| 219 | if (ret) { |
| 220 | PERROR("close"); |
| 221 | } |
| 222 | } |
| 223 | |
| 224 | __lttng_poll_free((void *) events->events); |
| 225 | } |
| 226 | |
| 227 | #else /* HAVE_EPOLL */ |
| 228 | /* |
| 229 | * Fallback on poll(2) API |
| 230 | */ |
| 231 | |
| 232 | /* Needed for some poll event values */ |
| 233 | #ifndef __USE_XOPEN |
| 234 | #define __USE_XOPEN |
| 235 | #endif |
| 236 | |
| 237 | /* Needed for some poll event values */ |
| 238 | #ifndef __USE_GNU |
| 239 | #define __USE_GNU |
| 240 | #endif |
| 241 | |
| 242 | #include <poll.h> |
| 243 | #include <stdint.h> |
| 244 | |
| 245 | enum { |
| 246 | /* Polling variables compatibility for poll */ |
| 247 | LPOLLIN = POLLIN, |
| 248 | LPOLLPRI = POLLPRI, |
| 249 | LPOLLOUT = POLLOUT, |
| 250 | LPOLLRDNORM = POLLRDNORM, |
| 251 | LPOLLRDBAND = POLLRDBAND, |
| 252 | LPOLLWRNORM = POLLWRNORM, |
| 253 | LPOLLWRBAND = POLLWRBAND, |
| 254 | #if __linux__ |
| 255 | LPOLLMSG = POLLMSG, |
| 256 | LPOLLRDHUP = POLLRDHUP, |
| 257 | #elif (defined(__FreeBSD__) || defined(__CYGWIN__) || defined(__sun__)) |
| 258 | LPOLLMSG = 0, |
| 259 | LPOLLRDHUP = 0, |
| 260 | #else |
| 261 | #error "Please add support for your OS." |
| 262 | #endif /* __linux__ */ |
| 263 | LPOLLERR = POLLERR, |
| 264 | LPOLLHUP = POLLHUP | POLLNVAL, |
| 265 | /* Close on exec feature does not exist for poll(2) */ |
| 266 | LTTNG_CLOEXEC = 0xdead, |
| 267 | }; |
| 268 | |
| 269 | struct compat_poll_event_array { |
| 270 | uint32_t nb_fd; /* Current number of fd in events */ |
| 271 | uint32_t alloc_size; /* Size of events array */ |
| 272 | /* Initial size of the pollset. We never shrink below that. */ |
| 273 | uint32_t init_size; |
| 274 | struct pollfd *events; |
| 275 | }; |
| 276 | |
| 277 | struct compat_poll_event { |
| 278 | /* |
| 279 | * Modified by the wait action. Updated using current fields if the |
| 280 | * need_update flag is set. |
| 281 | */ |
| 282 | struct compat_poll_event_array wait; |
| 283 | /* |
| 284 | * This is modified by add/del actions being the _current_ flow of |
| 285 | * execution before a poll wait is done. |
| 286 | */ |
| 287 | struct compat_poll_event_array current; |
| 288 | |
| 289 | /* Indicate if wait.events need to be updated from current. */ |
| 290 | int need_update:1; |
| 291 | }; |
| 292 | #define lttng_poll_event compat_poll_event |
| 293 | |
| 294 | static inline int __lttng_poll_get_prev_fd(struct lttng_poll_event *events, |
| 295 | int index, uint32_t nb_fd) |
| 296 | { |
| 297 | assert(events); |
| 298 | assert(index != nb_fd); |
| 299 | |
| 300 | if (index == 0 || nb_fd == 0) { |
| 301 | return -1; |
| 302 | } else { |
| 303 | return events->current.events[index - 1].fd; |
| 304 | } |
| 305 | } |
| 306 | |
| 307 | /* |
| 308 | * For the following calls, consider 'e' to be a lttng_poll_event pointer and i |
| 309 | * being the index of the events array. |
| 310 | */ |
| 311 | #define LTTNG_POLL_GETFD(e, i) LTTNG_REF(e)->wait.events[i].fd |
| 312 | #define LTTNG_POLL_GETEV(e, i) LTTNG_REF(e)->wait.events[i].revents |
| 313 | #define LTTNG_POLL_GETNB(e) LTTNG_REF(e)->wait.nb_fd |
| 314 | #define LTTNG_POLL_GETSZ(e) LTTNG_REF(e)->wait.events_size |
| 315 | #define LTTNG_POLL_GET_PREV_FD(e, i, nb_fd) \ |
| 316 | __lttng_poll_get_prev_fd(LTTNG_REF(e), i, nb_fd) |
| 317 | |
| 318 | /* |
| 319 | * Create a pollfd structure of size 'size'. |
| 320 | */ |
| 321 | extern int compat_poll_create(struct lttng_poll_event *events, int size); |
| 322 | #define lttng_poll_create(events, size, flags) \ |
| 323 | compat_poll_create(events, size) |
| 324 | |
| 325 | /* |
| 326 | * Wait on poll(2) event with nb_fd registered to the lttng_poll_event data |
| 327 | * structure. |
| 328 | */ |
| 329 | extern int compat_poll_wait(struct lttng_poll_event *events, int timeout); |
| 330 | #define lttng_poll_wait(events, timeout) \ |
| 331 | compat_poll_wait(events, timeout) |
| 332 | |
| 333 | /* |
| 334 | * Add the fd to the pollfd structure. Resize if needed. |
| 335 | */ |
| 336 | extern int compat_poll_add(struct lttng_poll_event *events, |
| 337 | int fd, uint32_t req_events); |
| 338 | #define lttng_poll_add(events, fd, req_events) \ |
| 339 | compat_poll_add(events, fd, req_events) |
| 340 | |
| 341 | /* |
| 342 | * Remove the fd from the pollfd. Memory allocation is done to recreate a new |
| 343 | * pollfd, data is copied from the old pollfd to the new and, finally, the old |
| 344 | * one is freed(). |
| 345 | */ |
| 346 | extern int compat_poll_del(struct lttng_poll_event *events, int fd); |
| 347 | #define lttng_poll_del(events, fd) \ |
| 348 | compat_poll_del(events, fd) |
| 349 | |
| 350 | /* |
| 351 | * Set up the poll set limits variable poll_max_size |
| 352 | */ |
| 353 | extern int compat_poll_set_max_size(void); |
| 354 | #define lttng_poll_set_max_size() \ |
| 355 | compat_poll_set_max_size() |
| 356 | |
| 357 | /* |
| 358 | * No need to reset a pollfd structure for poll(2) |
| 359 | */ |
| 360 | static inline void lttng_poll_reset(struct lttng_poll_event *events) |
| 361 | {} |
| 362 | |
| 363 | /* |
| 364 | * Initialize an already allocated poll event data structure. |
| 365 | */ |
| 366 | static inline void lttng_poll_init(struct lttng_poll_event *events) |
| 367 | { |
| 368 | memset(events, 0, sizeof(struct lttng_poll_event)); |
| 369 | } |
| 370 | |
| 371 | /* |
| 372 | * Clean the events structure of a lttng_poll_event. It's the caller |
| 373 | * responsability to free the lttng_poll_event memory. |
| 374 | */ |
| 375 | static inline void lttng_poll_clean(struct lttng_poll_event *events) |
| 376 | { |
| 377 | if (events) { |
| 378 | __lttng_poll_free((void *) events->wait.events); |
| 379 | __lttng_poll_free((void *) events->current.events); |
| 380 | } |
| 381 | } |
| 382 | |
| 383 | #endif /* HAVE_EPOLL */ |
| 384 | |
| 385 | #endif /* _LTT_POLL_H */ |