Tristan Matthews | 0a329cc | 2013-07-17 13:20:14 -0400 | [diff] [blame] | 1 | /* $Id$ */ |
| 2 | /* |
| 3 | * Copyright (C) 2008-2011 Teluu Inc. (http://www.teluu.com) |
| 4 | * Copyright (C) 2003-2008 Benny Prijono <benny@prijono.org> |
| 5 | * |
| 6 | * This program is free software; you can redistribute it and/or modify |
| 7 | * it under the terms of the GNU General Public License as published by |
| 8 | * the Free Software Foundation; either version 2 of the License, or |
| 9 | * (at your option) any later version. |
| 10 | * |
| 11 | * This program is distributed in the hope that it will be useful, |
| 12 | * but WITHOUT ANY WARRANTY; without even the implied warranty of |
| 13 | * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the |
| 14 | * GNU General Public License for more details. |
| 15 | * |
| 16 | * You should have received a copy of the GNU General Public License |
| 17 | * along with this program; if not, write to the Free Software |
| 18 | * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA |
| 19 | */ |
| 20 | |
| 21 | /* |
| 22 | * sock_select.c |
| 23 | * |
| 24 | * This is the implementation of IOQueue using pj_sock_select(). |
| 25 | * It runs anywhere where pj_sock_select() is available (currently |
| 26 | * Win32, Linux, Linux kernel, etc.). |
| 27 | */ |
| 28 | |
| 29 | #include <pj/ioqueue.h> |
| 30 | #include <pj/os.h> |
| 31 | #include <pj/lock.h> |
| 32 | #include <pj/log.h> |
| 33 | #include <pj/list.h> |
| 34 | #include <pj/pool.h> |
| 35 | #include <pj/string.h> |
| 36 | #include <pj/assert.h> |
| 37 | #include <pj/sock.h> |
| 38 | #include <pj/compat/socket.h> |
| 39 | #include <pj/sock_select.h> |
| 40 | #include <pj/sock_qos.h> |
| 41 | #include <pj/errno.h> |
| 42 | #include <pj/rand.h> |
| 43 | |
| 44 | /* Now that we have access to OS'es <sys/select>, lets check again that |
| 45 | * PJ_IOQUEUE_MAX_HANDLES is not greater than FD_SETSIZE |
| 46 | */ |
| 47 | #if PJ_IOQUEUE_MAX_HANDLES > FD_SETSIZE |
| 48 | # error "PJ_IOQUEUE_MAX_HANDLES cannot be greater than FD_SETSIZE" |
| 49 | #endif |
| 50 | |
| 51 | |
| 52 | /* |
| 53 | * Include declaration from common abstraction. |
| 54 | */ |
| 55 | #include "ioqueue_common_abs.h" |
| 56 | |
| 57 | /* |
| 58 | * ISSUES with ioqueue_select() |
| 59 | * |
| 60 | * EAGAIN/EWOULDBLOCK error in recv(): |
| 61 | * - when multiple threads are working with the ioqueue, application |
| 62 | * may receive EAGAIN or EWOULDBLOCK in the receive callback. |
| 63 | * This error happens because more than one thread is watching for |
| 64 | * the same descriptor set, so when all of them call recv() or recvfrom() |
| 65 | * simultaneously, only one will succeed and the rest will get the error. |
| 66 | * |
| 67 | */ |
| 68 | #define THIS_FILE "ioq_select" |
| 69 | |
| 70 | /* |
| 71 | * The select ioqueue relies on socket functions (pj_sock_xxx()) to return |
| 72 | * the correct error code. |
| 73 | */ |
| 74 | #if PJ_RETURN_OS_ERROR(100) != PJ_STATUS_FROM_OS(100) |
| 75 | # error "Error reporting must be enabled for this function to work!" |
| 76 | #endif |
| 77 | |
| 78 | /* |
| 79 | * During debugging build, VALIDATE_FD_SET is set. |
| 80 | * This will check the validity of the fd_sets. |
| 81 | */ |
| 82 | /* |
| 83 | #if defined(PJ_DEBUG) && PJ_DEBUG != 0 |
| 84 | # define VALIDATE_FD_SET 1 |
| 85 | #else |
| 86 | # define VALIDATE_FD_SET 0 |
| 87 | #endif |
| 88 | */ |
| 89 | #define VALIDATE_FD_SET 0 |
| 90 | |
| 91 | #if 0 |
| 92 | # define TRACE__(args) PJ_LOG(3,args) |
| 93 | #else |
| 94 | # define TRACE__(args) |
| 95 | #endif |
| 96 | |
| 97 | /* |
| 98 | * This describes each key. |
| 99 | */ |
| 100 | struct pj_ioqueue_key_t |
| 101 | { |
| 102 | DECLARE_COMMON_KEY |
| 103 | }; |
| 104 | |
| 105 | /* |
| 106 | * This describes the I/O queue itself. |
| 107 | */ |
| 108 | struct pj_ioqueue_t |
| 109 | { |
| 110 | DECLARE_COMMON_IOQUEUE |
| 111 | |
| 112 | unsigned max, count; /* Max and current key count */ |
| 113 | int nfds; /* The largest fd value (for select)*/ |
| 114 | pj_ioqueue_key_t active_list; /* List of active keys. */ |
| 115 | pj_fd_set_t rfdset; |
| 116 | pj_fd_set_t wfdset; |
| 117 | #if PJ_HAS_TCP |
| 118 | pj_fd_set_t xfdset; |
| 119 | #endif |
| 120 | |
| 121 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 122 | pj_mutex_t *ref_cnt_mutex; |
| 123 | pj_ioqueue_key_t closing_list; |
| 124 | pj_ioqueue_key_t free_list; |
| 125 | #endif |
| 126 | }; |
| 127 | |
| 128 | /* Proto */ |
| 129 | #if defined(PJ_IPHONE_OS_HAS_MULTITASKING_SUPPORT) && \ |
| 130 | PJ_IPHONE_OS_HAS_MULTITASKING_SUPPORT!=0 |
| 131 | static pj_status_t replace_udp_sock(pj_ioqueue_key_t *h); |
| 132 | #endif |
| 133 | |
| 134 | /* Include implementation for common abstraction after we declare |
| 135 | * pj_ioqueue_key_t and pj_ioqueue_t. |
| 136 | */ |
| 137 | #include "ioqueue_common_abs.c" |
| 138 | |
| 139 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 140 | /* Scan closing keys to be put to free list again */ |
| 141 | static void scan_closing_keys(pj_ioqueue_t *ioqueue); |
| 142 | #endif |
| 143 | |
| 144 | /* |
| 145 | * pj_ioqueue_name() |
| 146 | */ |
| 147 | PJ_DEF(const char*) pj_ioqueue_name(void) |
| 148 | { |
| 149 | return "select"; |
| 150 | } |
| 151 | |
| 152 | /* |
| 153 | * Scan the socket descriptor sets for the largest descriptor. |
| 154 | * This value is needed by select(). |
| 155 | */ |
| 156 | #if defined(PJ_SELECT_NEEDS_NFDS) && PJ_SELECT_NEEDS_NFDS!=0 |
| 157 | static void rescan_fdset(pj_ioqueue_t *ioqueue) |
| 158 | { |
| 159 | pj_ioqueue_key_t *key = ioqueue->active_list.next; |
| 160 | int max = 0; |
| 161 | |
| 162 | while (key != &ioqueue->active_list) { |
| 163 | if (key->fd > max) |
| 164 | max = key->fd; |
| 165 | key = key->next; |
| 166 | } |
| 167 | |
| 168 | ioqueue->nfds = max; |
| 169 | } |
| 170 | #else |
| 171 | static void rescan_fdset(pj_ioqueue_t *ioqueue) |
| 172 | { |
| 173 | ioqueue->nfds = FD_SETSIZE-1; |
| 174 | } |
| 175 | #endif |
| 176 | |
| 177 | |
| 178 | /* |
| 179 | * pj_ioqueue_create() |
| 180 | * |
| 181 | * Create select ioqueue. |
| 182 | */ |
| 183 | PJ_DEF(pj_status_t) pj_ioqueue_create( pj_pool_t *pool, |
| 184 | pj_size_t max_fd, |
| 185 | pj_ioqueue_t **p_ioqueue) |
| 186 | { |
| 187 | pj_ioqueue_t *ioqueue; |
| 188 | pj_lock_t *lock; |
| 189 | unsigned i; |
| 190 | pj_status_t rc; |
| 191 | |
| 192 | /* Check that arguments are valid. */ |
| 193 | PJ_ASSERT_RETURN(pool != NULL && p_ioqueue != NULL && |
| 194 | max_fd > 0 && max_fd <= PJ_IOQUEUE_MAX_HANDLES, |
| 195 | PJ_EINVAL); |
| 196 | |
| 197 | /* Check that size of pj_ioqueue_op_key_t is sufficient */ |
| 198 | PJ_ASSERT_RETURN(sizeof(pj_ioqueue_op_key_t)-sizeof(void*) >= |
| 199 | sizeof(union operation_key), PJ_EBUG); |
| 200 | |
| 201 | /* Create and init common ioqueue stuffs */ |
| 202 | ioqueue = PJ_POOL_ALLOC_T(pool, pj_ioqueue_t); |
| 203 | ioqueue_init(ioqueue); |
| 204 | |
| 205 | ioqueue->max = (unsigned)max_fd; |
| 206 | ioqueue->count = 0; |
| 207 | PJ_FD_ZERO(&ioqueue->rfdset); |
| 208 | PJ_FD_ZERO(&ioqueue->wfdset); |
| 209 | #if PJ_HAS_TCP |
| 210 | PJ_FD_ZERO(&ioqueue->xfdset); |
| 211 | #endif |
| 212 | pj_list_init(&ioqueue->active_list); |
| 213 | |
| 214 | rescan_fdset(ioqueue); |
| 215 | |
| 216 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 217 | /* When safe unregistration is used (the default), we pre-create |
| 218 | * all keys and put them in the free list. |
| 219 | */ |
| 220 | |
| 221 | /* Mutex to protect key's reference counter |
| 222 | * We don't want to use key's mutex or ioqueue's mutex because |
| 223 | * that would create deadlock situation in some cases. |
| 224 | */ |
| 225 | rc = pj_mutex_create_simple(pool, NULL, &ioqueue->ref_cnt_mutex); |
| 226 | if (rc != PJ_SUCCESS) |
| 227 | return rc; |
| 228 | |
| 229 | |
| 230 | /* Init key list */ |
| 231 | pj_list_init(&ioqueue->free_list); |
| 232 | pj_list_init(&ioqueue->closing_list); |
| 233 | |
| 234 | |
| 235 | /* Pre-create all keys according to max_fd */ |
| 236 | for (i=0; i<max_fd; ++i) { |
| 237 | pj_ioqueue_key_t *key; |
| 238 | |
| 239 | key = PJ_POOL_ALLOC_T(pool, pj_ioqueue_key_t); |
| 240 | key->ref_count = 0; |
| 241 | rc = pj_lock_create_recursive_mutex(pool, NULL, &key->lock); |
| 242 | if (rc != PJ_SUCCESS) { |
| 243 | key = ioqueue->free_list.next; |
| 244 | while (key != &ioqueue->free_list) { |
| 245 | pj_lock_destroy(key->lock); |
| 246 | key = key->next; |
| 247 | } |
| 248 | pj_mutex_destroy(ioqueue->ref_cnt_mutex); |
| 249 | return rc; |
| 250 | } |
| 251 | |
| 252 | pj_list_push_back(&ioqueue->free_list, key); |
| 253 | } |
| 254 | #endif |
| 255 | |
| 256 | /* Create and init ioqueue mutex */ |
| 257 | rc = pj_lock_create_simple_mutex(pool, "ioq%p", &lock); |
| 258 | if (rc != PJ_SUCCESS) |
| 259 | return rc; |
| 260 | |
| 261 | rc = pj_ioqueue_set_lock(ioqueue, lock, PJ_TRUE); |
| 262 | if (rc != PJ_SUCCESS) |
| 263 | return rc; |
| 264 | |
| 265 | PJ_LOG(4, ("pjlib", "select() I/O Queue created (%p)", ioqueue)); |
| 266 | |
| 267 | *p_ioqueue = ioqueue; |
| 268 | return PJ_SUCCESS; |
| 269 | } |
| 270 | |
| 271 | /* |
| 272 | * pj_ioqueue_destroy() |
| 273 | * |
| 274 | * Destroy ioqueue. |
| 275 | */ |
| 276 | PJ_DEF(pj_status_t) pj_ioqueue_destroy(pj_ioqueue_t *ioqueue) |
| 277 | { |
| 278 | pj_ioqueue_key_t *key; |
| 279 | |
| 280 | PJ_ASSERT_RETURN(ioqueue, PJ_EINVAL); |
| 281 | |
| 282 | pj_lock_acquire(ioqueue->lock); |
| 283 | |
| 284 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 285 | /* Destroy reference counters */ |
| 286 | key = ioqueue->active_list.next; |
| 287 | while (key != &ioqueue->active_list) { |
| 288 | pj_lock_destroy(key->lock); |
| 289 | key = key->next; |
| 290 | } |
| 291 | |
| 292 | key = ioqueue->closing_list.next; |
| 293 | while (key != &ioqueue->closing_list) { |
| 294 | pj_lock_destroy(key->lock); |
| 295 | key = key->next; |
| 296 | } |
| 297 | |
| 298 | key = ioqueue->free_list.next; |
| 299 | while (key != &ioqueue->free_list) { |
| 300 | pj_lock_destroy(key->lock); |
| 301 | key = key->next; |
| 302 | } |
| 303 | |
| 304 | pj_mutex_destroy(ioqueue->ref_cnt_mutex); |
| 305 | #endif |
| 306 | |
| 307 | return ioqueue_destroy(ioqueue); |
| 308 | } |
| 309 | |
| 310 | |
| 311 | /* |
| 312 | * pj_ioqueue_register_sock() |
| 313 | * |
| 314 | * Register socket handle to ioqueue. |
| 315 | */ |
| 316 | PJ_DEF(pj_status_t) pj_ioqueue_register_sock2(pj_pool_t *pool, |
| 317 | pj_ioqueue_t *ioqueue, |
| 318 | pj_sock_t sock, |
| 319 | pj_grp_lock_t *grp_lock, |
| 320 | void *user_data, |
| 321 | const pj_ioqueue_callback *cb, |
| 322 | pj_ioqueue_key_t **p_key) |
| 323 | { |
| 324 | pj_ioqueue_key_t *key = NULL; |
| 325 | #if defined(PJ_WIN32) && PJ_WIN32!=0 || \ |
| 326 | defined(PJ_WIN64) && PJ_WIN64 != 0 || \ |
| 327 | defined(PJ_WIN32_WINCE) && PJ_WIN32_WINCE!=0 |
| 328 | u_long value; |
| 329 | #else |
| 330 | pj_uint32_t value; |
| 331 | #endif |
| 332 | pj_status_t rc = PJ_SUCCESS; |
| 333 | |
| 334 | PJ_ASSERT_RETURN(pool && ioqueue && sock != PJ_INVALID_SOCKET && |
| 335 | cb && p_key, PJ_EINVAL); |
| 336 | |
| 337 | pj_lock_acquire(ioqueue->lock); |
| 338 | |
| 339 | if (ioqueue->count >= ioqueue->max) { |
| 340 | rc = PJ_ETOOMANY; |
| 341 | goto on_return; |
| 342 | } |
| 343 | |
| 344 | /* If safe unregistration (PJ_IOQUEUE_HAS_SAFE_UNREG) is used, get |
| 345 | * the key from the free list. Otherwise allocate a new one. |
| 346 | */ |
| 347 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 348 | |
| 349 | /* Scan closing_keys first to let them come back to free_list */ |
| 350 | scan_closing_keys(ioqueue); |
| 351 | |
| 352 | pj_assert(!pj_list_empty(&ioqueue->free_list)); |
| 353 | if (pj_list_empty(&ioqueue->free_list)) { |
| 354 | rc = PJ_ETOOMANY; |
| 355 | goto on_return; |
| 356 | } |
| 357 | |
| 358 | key = ioqueue->free_list.next; |
| 359 | pj_list_erase(key); |
| 360 | #else |
| 361 | key = (pj_ioqueue_key_t*)pj_pool_zalloc(pool, sizeof(pj_ioqueue_key_t)); |
| 362 | #endif |
| 363 | |
| 364 | rc = ioqueue_init_key(pool, ioqueue, key, sock, grp_lock, user_data, cb); |
| 365 | if (rc != PJ_SUCCESS) { |
| 366 | key = NULL; |
| 367 | goto on_return; |
| 368 | } |
| 369 | |
| 370 | /* Set socket to nonblocking. */ |
| 371 | value = 1; |
| 372 | #if defined(PJ_WIN32) && PJ_WIN32!=0 || \ |
| 373 | defined(PJ_WIN64) && PJ_WIN64 != 0 || \ |
| 374 | defined(PJ_WIN32_WINCE) && PJ_WIN32_WINCE!=0 |
| 375 | if (ioctlsocket(sock, FIONBIO, &value)) { |
| 376 | #else |
| 377 | if (ioctl(sock, FIONBIO, &value)) { |
| 378 | #endif |
| 379 | rc = pj_get_netos_error(); |
| 380 | goto on_return; |
| 381 | } |
| 382 | |
| 383 | |
| 384 | /* Put in active list. */ |
| 385 | pj_list_insert_before(&ioqueue->active_list, key); |
| 386 | ++ioqueue->count; |
| 387 | |
| 388 | /* Rescan fdset to get max descriptor */ |
| 389 | rescan_fdset(ioqueue); |
| 390 | |
| 391 | on_return: |
| 392 | /* On error, socket may be left in non-blocking mode. */ |
| 393 | if (rc != PJ_SUCCESS) { |
| 394 | if (key && key->grp_lock) |
| 395 | pj_grp_lock_dec_ref_dbg(key->grp_lock, "ioqueue", 0); |
| 396 | } |
| 397 | *p_key = key; |
| 398 | pj_lock_release(ioqueue->lock); |
| 399 | |
| 400 | return rc; |
| 401 | } |
| 402 | |
| 403 | PJ_DEF(pj_status_t) pj_ioqueue_register_sock( pj_pool_t *pool, |
| 404 | pj_ioqueue_t *ioqueue, |
| 405 | pj_sock_t sock, |
| 406 | void *user_data, |
| 407 | const pj_ioqueue_callback *cb, |
| 408 | pj_ioqueue_key_t **p_key) |
| 409 | { |
| 410 | return pj_ioqueue_register_sock2(pool, ioqueue, sock, NULL, user_data, |
| 411 | cb, p_key); |
| 412 | } |
| 413 | |
| 414 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 415 | /* Increment key's reference counter */ |
| 416 | static void increment_counter(pj_ioqueue_key_t *key) |
| 417 | { |
| 418 | pj_mutex_lock(key->ioqueue->ref_cnt_mutex); |
| 419 | ++key->ref_count; |
| 420 | pj_mutex_unlock(key->ioqueue->ref_cnt_mutex); |
| 421 | } |
| 422 | |
| 423 | /* Decrement the key's reference counter, and when the counter reach zero, |
| 424 | * destroy the key. |
| 425 | * |
| 426 | * Note: MUST NOT CALL THIS FUNCTION WHILE HOLDING ioqueue's LOCK. |
| 427 | */ |
| 428 | static void decrement_counter(pj_ioqueue_key_t *key) |
| 429 | { |
| 430 | pj_lock_acquire(key->ioqueue->lock); |
| 431 | pj_mutex_lock(key->ioqueue->ref_cnt_mutex); |
| 432 | --key->ref_count; |
| 433 | if (key->ref_count == 0) { |
| 434 | |
| 435 | pj_assert(key->closing == 1); |
| 436 | pj_gettickcount(&key->free_time); |
| 437 | key->free_time.msec += PJ_IOQUEUE_KEY_FREE_DELAY; |
| 438 | pj_time_val_normalize(&key->free_time); |
| 439 | |
| 440 | pj_list_erase(key); |
| 441 | pj_list_push_back(&key->ioqueue->closing_list, key); |
| 442 | /* Rescan fdset to get max descriptor */ |
| 443 | rescan_fdset(key->ioqueue); |
| 444 | } |
| 445 | pj_mutex_unlock(key->ioqueue->ref_cnt_mutex); |
| 446 | pj_lock_release(key->ioqueue->lock); |
| 447 | } |
| 448 | #endif |
| 449 | |
| 450 | |
| 451 | /* |
| 452 | * pj_ioqueue_unregister() |
| 453 | * |
| 454 | * Unregister handle from ioqueue. |
| 455 | */ |
| 456 | PJ_DEF(pj_status_t) pj_ioqueue_unregister( pj_ioqueue_key_t *key) |
| 457 | { |
| 458 | pj_ioqueue_t *ioqueue; |
| 459 | |
| 460 | PJ_ASSERT_RETURN(key, PJ_EINVAL); |
| 461 | |
| 462 | ioqueue = key->ioqueue; |
| 463 | |
| 464 | /* Lock the key to make sure no callback is simultaneously modifying |
| 465 | * the key. We need to lock the key before ioqueue here to prevent |
| 466 | * deadlock. |
| 467 | */ |
| 468 | pj_ioqueue_lock_key(key); |
| 469 | |
| 470 | /* Also lock ioqueue */ |
| 471 | pj_lock_acquire(ioqueue->lock); |
| 472 | |
| 473 | pj_assert(ioqueue->count > 0); |
| 474 | --ioqueue->count; |
| 475 | #if !PJ_IOQUEUE_HAS_SAFE_UNREG |
| 476 | /* Ticket #520, key will be erased more than once */ |
| 477 | pj_list_erase(key); |
| 478 | #endif |
| 479 | PJ_FD_CLR(key->fd, &ioqueue->rfdset); |
| 480 | PJ_FD_CLR(key->fd, &ioqueue->wfdset); |
| 481 | #if PJ_HAS_TCP |
| 482 | PJ_FD_CLR(key->fd, &ioqueue->xfdset); |
| 483 | #endif |
| 484 | |
| 485 | /* Close socket. */ |
| 486 | pj_sock_close(key->fd); |
| 487 | |
| 488 | /* Clear callback */ |
| 489 | key->cb.on_accept_complete = NULL; |
| 490 | key->cb.on_connect_complete = NULL; |
| 491 | key->cb.on_read_complete = NULL; |
| 492 | key->cb.on_write_complete = NULL; |
| 493 | |
| 494 | /* Must release ioqueue lock first before decrementing counter, to |
| 495 | * prevent deadlock. |
| 496 | */ |
| 497 | pj_lock_release(ioqueue->lock); |
| 498 | |
| 499 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 500 | /* Mark key is closing. */ |
| 501 | key->closing = 1; |
| 502 | |
| 503 | /* Decrement counter. */ |
| 504 | decrement_counter(key); |
| 505 | |
| 506 | /* Done. */ |
| 507 | if (key->grp_lock) { |
| 508 | /* just dec_ref and unlock. we will set grp_lock to NULL |
| 509 | * elsewhere */ |
| 510 | pj_grp_lock_t *grp_lock = key->grp_lock; |
| 511 | // Don't set grp_lock to NULL otherwise the other thread |
| 512 | // will crash. Just leave it as dangling pointer, but this |
| 513 | // should be safe |
| 514 | //key->grp_lock = NULL; |
| 515 | pj_grp_lock_dec_ref_dbg(grp_lock, "ioqueue", 0); |
| 516 | pj_grp_lock_release(grp_lock); |
| 517 | } else { |
| 518 | pj_ioqueue_unlock_key(key); |
| 519 | } |
| 520 | #else |
| 521 | if (key->grp_lock) { |
| 522 | /* set grp_lock to NULL and unlock */ |
| 523 | pj_grp_lock_t *grp_lock = key->grp_lock; |
| 524 | // Don't set grp_lock to NULL otherwise the other thread |
| 525 | // will crash. Just leave it as dangling pointer, but this |
| 526 | // should be safe |
| 527 | //key->grp_lock = NULL; |
| 528 | pj_grp_lock_dec_ref_dbg(grp_lock, "ioqueue", 0); |
| 529 | pj_grp_lock_release(grp_lock); |
| 530 | } else { |
| 531 | pj_ioqueue_unlock_key(key); |
| 532 | } |
| 533 | |
| 534 | pj_lock_destroy(key->lock); |
| 535 | #endif |
| 536 | |
| 537 | return PJ_SUCCESS; |
| 538 | } |
| 539 | |
| 540 | |
| 541 | /* This supposed to check whether the fd_set values are consistent |
| 542 | * with the operation currently set in each key. |
| 543 | */ |
| 544 | #if VALIDATE_FD_SET |
| 545 | static void validate_sets(const pj_ioqueue_t *ioqueue, |
| 546 | const pj_fd_set_t *rfdset, |
| 547 | const pj_fd_set_t *wfdset, |
| 548 | const pj_fd_set_t *xfdset) |
| 549 | { |
| 550 | pj_ioqueue_key_t *key; |
| 551 | |
| 552 | /* |
| 553 | * This basicly would not work anymore. |
| 554 | * We need to lock key before performing the check, but we can't do |
| 555 | * so because we're holding ioqueue mutex. If we acquire key's mutex |
| 556 | * now, the will cause deadlock. |
| 557 | */ |
| 558 | pj_assert(0); |
| 559 | |
| 560 | key = ioqueue->active_list.next; |
| 561 | while (key != &ioqueue->active_list) { |
| 562 | if (!pj_list_empty(&key->read_list) |
| 563 | #if defined(PJ_HAS_TCP) && PJ_HAS_TCP != 0 |
| 564 | || !pj_list_empty(&key->accept_list) |
| 565 | #endif |
| 566 | ) |
| 567 | { |
| 568 | pj_assert(PJ_FD_ISSET(key->fd, rfdset)); |
| 569 | } |
| 570 | else { |
| 571 | pj_assert(PJ_FD_ISSET(key->fd, rfdset) == 0); |
| 572 | } |
| 573 | if (!pj_list_empty(&key->write_list) |
| 574 | #if defined(PJ_HAS_TCP) && PJ_HAS_TCP != 0 |
| 575 | || key->connecting |
| 576 | #endif |
| 577 | ) |
| 578 | { |
| 579 | pj_assert(PJ_FD_ISSET(key->fd, wfdset)); |
| 580 | } |
| 581 | else { |
| 582 | pj_assert(PJ_FD_ISSET(key->fd, wfdset) == 0); |
| 583 | } |
| 584 | #if defined(PJ_HAS_TCP) && PJ_HAS_TCP != 0 |
| 585 | if (key->connecting) |
| 586 | { |
| 587 | pj_assert(PJ_FD_ISSET(key->fd, xfdset)); |
| 588 | } |
| 589 | else { |
| 590 | pj_assert(PJ_FD_ISSET(key->fd, xfdset) == 0); |
| 591 | } |
| 592 | #endif /* PJ_HAS_TCP */ |
| 593 | |
| 594 | key = key->next; |
| 595 | } |
| 596 | } |
| 597 | #endif /* VALIDATE_FD_SET */ |
| 598 | |
| 599 | |
| 600 | /* ioqueue_remove_from_set() |
| 601 | * This function is called from ioqueue_dispatch_event() to instruct |
| 602 | * the ioqueue to remove the specified descriptor from ioqueue's descriptor |
| 603 | * set for the specified event. |
| 604 | */ |
| 605 | static void ioqueue_remove_from_set( pj_ioqueue_t *ioqueue, |
| 606 | pj_ioqueue_key_t *key, |
| 607 | enum ioqueue_event_type event_type) |
| 608 | { |
| 609 | pj_lock_acquire(ioqueue->lock); |
| 610 | |
| 611 | if (event_type == READABLE_EVENT) |
| 612 | PJ_FD_CLR((pj_sock_t)key->fd, &ioqueue->rfdset); |
| 613 | else if (event_type == WRITEABLE_EVENT) |
| 614 | PJ_FD_CLR((pj_sock_t)key->fd, &ioqueue->wfdset); |
| 615 | #if defined(PJ_HAS_TCP) && PJ_HAS_TCP!=0 |
| 616 | else if (event_type == EXCEPTION_EVENT) |
| 617 | PJ_FD_CLR((pj_sock_t)key->fd, &ioqueue->xfdset); |
| 618 | #endif |
| 619 | else |
| 620 | pj_assert(0); |
| 621 | |
| 622 | pj_lock_release(ioqueue->lock); |
| 623 | } |
| 624 | |
| 625 | /* |
| 626 | * ioqueue_add_to_set() |
| 627 | * This function is called from pj_ioqueue_recv(), pj_ioqueue_send() etc |
| 628 | * to instruct the ioqueue to add the specified handle to ioqueue's descriptor |
| 629 | * set for the specified event. |
| 630 | */ |
| 631 | static void ioqueue_add_to_set( pj_ioqueue_t *ioqueue, |
| 632 | pj_ioqueue_key_t *key, |
| 633 | enum ioqueue_event_type event_type ) |
| 634 | { |
| 635 | pj_lock_acquire(ioqueue->lock); |
| 636 | |
| 637 | if (event_type == READABLE_EVENT) |
| 638 | PJ_FD_SET((pj_sock_t)key->fd, &ioqueue->rfdset); |
| 639 | else if (event_type == WRITEABLE_EVENT) |
| 640 | PJ_FD_SET((pj_sock_t)key->fd, &ioqueue->wfdset); |
| 641 | #if defined(PJ_HAS_TCP) && PJ_HAS_TCP!=0 |
| 642 | else if (event_type == EXCEPTION_EVENT) |
| 643 | PJ_FD_SET((pj_sock_t)key->fd, &ioqueue->xfdset); |
| 644 | #endif |
| 645 | else |
| 646 | pj_assert(0); |
| 647 | |
| 648 | pj_lock_release(ioqueue->lock); |
| 649 | } |
| 650 | |
| 651 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 652 | /* Scan closing keys to be put to free list again */ |
| 653 | static void scan_closing_keys(pj_ioqueue_t *ioqueue) |
| 654 | { |
| 655 | pj_time_val now; |
| 656 | pj_ioqueue_key_t *h; |
| 657 | |
| 658 | pj_gettickcount(&now); |
| 659 | h = ioqueue->closing_list.next; |
| 660 | while (h != &ioqueue->closing_list) { |
| 661 | pj_ioqueue_key_t *next = h->next; |
| 662 | |
| 663 | pj_assert(h->closing != 0); |
| 664 | |
| 665 | if (PJ_TIME_VAL_GTE(now, h->free_time)) { |
| 666 | pj_list_erase(h); |
| 667 | // Don't set grp_lock to NULL otherwise the other thread |
| 668 | // will crash. Just leave it as dangling pointer, but this |
| 669 | // should be safe |
| 670 | //h->grp_lock = NULL; |
| 671 | pj_list_push_back(&ioqueue->free_list, h); |
| 672 | } |
| 673 | h = next; |
| 674 | } |
| 675 | } |
| 676 | #endif |
| 677 | |
| 678 | #if defined(PJ_IPHONE_OS_HAS_MULTITASKING_SUPPORT) && \ |
| 679 | PJ_IPHONE_OS_HAS_MULTITASKING_SUPPORT!=0 |
| 680 | static pj_status_t replace_udp_sock(pj_ioqueue_key_t *h) |
| 681 | { |
| 682 | enum flags { |
| 683 | HAS_PEER_ADDR = 1, |
| 684 | HAS_QOS = 2 |
| 685 | }; |
| 686 | pj_sock_t old_sock, new_sock = PJ_INVALID_SOCKET; |
| 687 | pj_sockaddr local_addr, rem_addr; |
| 688 | int val, addr_len; |
| 689 | pj_fd_set_t *fds[3]; |
| 690 | unsigned i, fds_cnt, flags=0; |
| 691 | pj_qos_params qos_params; |
| 692 | unsigned msec; |
| 693 | pj_status_t status; |
| 694 | |
| 695 | pj_lock_acquire(h->ioqueue->lock); |
| 696 | |
| 697 | old_sock = h->fd; |
| 698 | |
| 699 | /* Can only replace UDP socket */ |
| 700 | pj_assert(h->fd_type == pj_SOCK_DGRAM()); |
| 701 | |
| 702 | PJ_LOG(4,(THIS_FILE, "Attempting to replace UDP socket %d", old_sock)); |
| 703 | |
| 704 | /* Investigate the old socket */ |
| 705 | addr_len = sizeof(local_addr); |
| 706 | status = pj_sock_getsockname(old_sock, &local_addr, &addr_len); |
| 707 | if (status != PJ_SUCCESS) |
| 708 | goto on_error; |
| 709 | |
| 710 | addr_len = sizeof(rem_addr); |
| 711 | status = pj_sock_getpeername(old_sock, &rem_addr, &addr_len); |
| 712 | if (status == PJ_SUCCESS) |
| 713 | flags |= HAS_PEER_ADDR; |
| 714 | |
| 715 | status = pj_sock_get_qos_params(old_sock, &qos_params); |
| 716 | if (status == PJ_SUCCESS) |
| 717 | flags |= HAS_QOS; |
| 718 | |
| 719 | /* We're done with the old socket, close it otherwise we'll get |
| 720 | * error in bind() |
| 721 | */ |
| 722 | pj_sock_close(old_sock); |
| 723 | |
| 724 | /* Prepare the new socket */ |
| 725 | status = pj_sock_socket(local_addr.addr.sa_family, PJ_SOCK_DGRAM, 0, |
| 726 | &new_sock); |
| 727 | if (status != PJ_SUCCESS) |
| 728 | goto on_error; |
| 729 | |
| 730 | /* Even after the socket is closed, we'll still get "Address in use" |
| 731 | * errors, so force it with SO_REUSEADDR |
| 732 | */ |
| 733 | val = 1; |
| 734 | status = pj_sock_setsockopt(new_sock, SOL_SOCKET, SO_REUSEADDR, |
| 735 | &val, sizeof(val)); |
| 736 | if (status != PJ_SUCCESS) |
| 737 | goto on_error; |
| 738 | |
| 739 | /* The loop is silly, but what else can we do? */ |
| 740 | addr_len = pj_sockaddr_get_len(&local_addr); |
| 741 | for (msec=20; ; msec<1000? msec=msec*2 : 1000) { |
| 742 | status = pj_sock_bind(new_sock, &local_addr, addr_len); |
| 743 | if (status != PJ_STATUS_FROM_OS(EADDRINUSE)) |
| 744 | break; |
| 745 | PJ_LOG(4,(THIS_FILE, "Address is still in use, retrying..")); |
| 746 | pj_thread_sleep(msec); |
| 747 | } |
| 748 | |
| 749 | if (status != PJ_SUCCESS) |
| 750 | goto on_error; |
| 751 | |
| 752 | if (flags & HAS_QOS) { |
| 753 | status = pj_sock_set_qos_params(new_sock, &qos_params); |
| 754 | if (status != PJ_SUCCESS) |
| 755 | goto on_error; |
| 756 | } |
| 757 | |
| 758 | if (flags & HAS_PEER_ADDR) { |
| 759 | status = pj_sock_connect(new_sock, &rem_addr, addr_len); |
| 760 | if (status != PJ_SUCCESS) |
| 761 | goto on_error; |
| 762 | } |
| 763 | |
| 764 | /* Set socket to nonblocking. */ |
| 765 | val = 1; |
| 766 | #if defined(PJ_WIN32) && PJ_WIN32!=0 || \ |
| 767 | defined(PJ_WIN64) && PJ_WIN64 != 0 || \ |
| 768 | defined(PJ_WIN32_WINCE) && PJ_WIN32_WINCE!=0 |
| 769 | if (ioctlsocket(new_sock, FIONBIO, &val)) { |
| 770 | #else |
| 771 | if (ioctl(new_sock, FIONBIO, &val)) { |
| 772 | #endif |
| 773 | status = pj_get_netos_error(); |
| 774 | goto on_error; |
| 775 | } |
| 776 | |
| 777 | /* Replace the occurrence of old socket with new socket in the |
| 778 | * fd sets. |
| 779 | */ |
| 780 | fds_cnt = 0; |
| 781 | fds[fds_cnt++] = &h->ioqueue->rfdset; |
| 782 | fds[fds_cnt++] = &h->ioqueue->wfdset; |
| 783 | #if PJ_HAS_TCP |
| 784 | fds[fds_cnt++] = &h->ioqueue->xfdset; |
| 785 | #endif |
| 786 | |
| 787 | for (i=0; i<fds_cnt; ++i) { |
| 788 | if (PJ_FD_ISSET(old_sock, fds[i])) { |
| 789 | PJ_FD_CLR(old_sock, fds[i]); |
| 790 | PJ_FD_SET(new_sock, fds[i]); |
| 791 | } |
| 792 | } |
| 793 | |
| 794 | /* And finally replace the fd in the key */ |
| 795 | h->fd = new_sock; |
| 796 | |
| 797 | PJ_LOG(4,(THIS_FILE, "UDP has been replaced successfully!")); |
| 798 | |
| 799 | pj_lock_release(h->ioqueue->lock); |
| 800 | |
| 801 | return PJ_SUCCESS; |
| 802 | |
| 803 | on_error: |
| 804 | if (new_sock != PJ_INVALID_SOCKET) |
| 805 | pj_sock_close(new_sock); |
| 806 | PJ_PERROR(1,(THIS_FILE, status, "Error replacing socket")); |
| 807 | pj_lock_release(h->ioqueue->lock); |
| 808 | return status; |
| 809 | } |
| 810 | #endif |
| 811 | |
| 812 | |
| 813 | /* |
| 814 | * pj_ioqueue_poll() |
| 815 | * |
| 816 | * Few things worth written: |
| 817 | * |
| 818 | * - we used to do only one callback called per poll, but it didn't go |
| 819 | * very well. The reason is because on some situation, the write |
| 820 | * callback gets called all the time, thus doesn't give the read |
| 821 | * callback to get called. This happens, for example, when user |
| 822 | * submit write operation inside the write callback. |
| 823 | * As the result, we changed the behaviour so that now multiple |
| 824 | * callbacks are called in a single poll. It should be fast too, |
| 825 | * just that we need to be carefull with the ioqueue data structs. |
| 826 | * |
| 827 | * - to guarantee preemptiveness etc, the poll function must strictly |
| 828 | * work on fd_set copy of the ioqueue (not the original one). |
| 829 | */ |
| 830 | PJ_DEF(int) pj_ioqueue_poll( pj_ioqueue_t *ioqueue, const pj_time_val *timeout) |
| 831 | { |
| 832 | pj_fd_set_t rfdset, wfdset, xfdset; |
| 833 | int count, i, counter; |
| 834 | pj_ioqueue_key_t *h; |
| 835 | struct event |
| 836 | { |
| 837 | pj_ioqueue_key_t *key; |
| 838 | enum ioqueue_event_type event_type; |
| 839 | } event[PJ_IOQUEUE_MAX_EVENTS_IN_SINGLE_POLL]; |
| 840 | |
| 841 | PJ_ASSERT_RETURN(ioqueue, -PJ_EINVAL); |
| 842 | |
| 843 | /* Lock ioqueue before making fd_set copies */ |
| 844 | pj_lock_acquire(ioqueue->lock); |
| 845 | |
| 846 | /* We will only do select() when there are sockets to be polled. |
| 847 | * Otherwise select() will return error. |
| 848 | */ |
| 849 | if (PJ_FD_COUNT(&ioqueue->rfdset)==0 && |
| 850 | PJ_FD_COUNT(&ioqueue->wfdset)==0 |
| 851 | #if defined(PJ_HAS_TCP) && PJ_HAS_TCP!=0 |
| 852 | && PJ_FD_COUNT(&ioqueue->xfdset)==0 |
| 853 | #endif |
| 854 | ) |
| 855 | { |
| 856 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 857 | scan_closing_keys(ioqueue); |
| 858 | #endif |
| 859 | pj_lock_release(ioqueue->lock); |
| 860 | TRACE__((THIS_FILE, " poll: no fd is set")); |
| 861 | if (timeout) |
| 862 | pj_thread_sleep(PJ_TIME_VAL_MSEC(*timeout)); |
| 863 | return 0; |
| 864 | } |
| 865 | |
| 866 | /* Copy ioqueue's pj_fd_set_t to local variables. */ |
| 867 | pj_memcpy(&rfdset, &ioqueue->rfdset, sizeof(pj_fd_set_t)); |
| 868 | pj_memcpy(&wfdset, &ioqueue->wfdset, sizeof(pj_fd_set_t)); |
| 869 | #if PJ_HAS_TCP |
| 870 | pj_memcpy(&xfdset, &ioqueue->xfdset, sizeof(pj_fd_set_t)); |
| 871 | #else |
| 872 | PJ_FD_ZERO(&xfdset); |
| 873 | #endif |
| 874 | |
| 875 | #if VALIDATE_FD_SET |
| 876 | validate_sets(ioqueue, &rfdset, &wfdset, &xfdset); |
| 877 | #endif |
| 878 | |
| 879 | /* Unlock ioqueue before select(). */ |
| 880 | pj_lock_release(ioqueue->lock); |
| 881 | |
| 882 | count = pj_sock_select(ioqueue->nfds+1, &rfdset, &wfdset, &xfdset, |
| 883 | timeout); |
| 884 | |
| 885 | if (count == 0) |
| 886 | return 0; |
| 887 | else if (count < 0) |
| 888 | return -pj_get_netos_error(); |
| 889 | else if (count > PJ_IOQUEUE_MAX_EVENTS_IN_SINGLE_POLL) |
| 890 | count = PJ_IOQUEUE_MAX_EVENTS_IN_SINGLE_POLL; |
| 891 | |
| 892 | /* Scan descriptor sets for event and add the events in the event |
| 893 | * array to be processed later in this function. We do this so that |
| 894 | * events can be processed in parallel without holding ioqueue lock. |
| 895 | */ |
| 896 | pj_lock_acquire(ioqueue->lock); |
| 897 | |
| 898 | counter = 0; |
| 899 | |
| 900 | /* Scan for writable sockets first to handle piggy-back data |
| 901 | * coming with accept(). |
| 902 | */ |
| 903 | h = ioqueue->active_list.next; |
| 904 | for ( ; h!=&ioqueue->active_list && counter<count; h = h->next) { |
| 905 | |
| 906 | if ( (key_has_pending_write(h) || key_has_pending_connect(h)) |
| 907 | && PJ_FD_ISSET(h->fd, &wfdset) && !IS_CLOSING(h)) |
| 908 | { |
| 909 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 910 | increment_counter(h); |
| 911 | #endif |
| 912 | event[counter].key = h; |
| 913 | event[counter].event_type = WRITEABLE_EVENT; |
| 914 | ++counter; |
| 915 | } |
| 916 | |
| 917 | /* Scan for readable socket. */ |
| 918 | if ((key_has_pending_read(h) || key_has_pending_accept(h)) |
| 919 | && PJ_FD_ISSET(h->fd, &rfdset) && !IS_CLOSING(h) && |
| 920 | counter<count) |
| 921 | { |
| 922 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 923 | increment_counter(h); |
| 924 | #endif |
| 925 | event[counter].key = h; |
| 926 | event[counter].event_type = READABLE_EVENT; |
| 927 | ++counter; |
| 928 | } |
| 929 | |
| 930 | #if PJ_HAS_TCP |
| 931 | if (key_has_pending_connect(h) && PJ_FD_ISSET(h->fd, &xfdset) && |
| 932 | !IS_CLOSING(h) && counter<count) |
| 933 | { |
| 934 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 935 | increment_counter(h); |
| 936 | #endif |
| 937 | event[counter].key = h; |
| 938 | event[counter].event_type = EXCEPTION_EVENT; |
| 939 | ++counter; |
| 940 | } |
| 941 | #endif |
| 942 | } |
| 943 | |
| 944 | for (i=0; i<counter; ++i) { |
| 945 | if (event[i].key->grp_lock) |
| 946 | pj_grp_lock_add_ref_dbg(event[i].key->grp_lock, "ioqueue", 0); |
| 947 | } |
| 948 | |
| 949 | PJ_RACE_ME(5); |
| 950 | |
| 951 | pj_lock_release(ioqueue->lock); |
| 952 | |
| 953 | PJ_RACE_ME(5); |
| 954 | |
| 955 | count = counter; |
| 956 | |
| 957 | /* Now process all events. The dispatch functions will take care |
| 958 | * of locking in each of the key |
| 959 | */ |
| 960 | for (counter=0; counter<count; ++counter) { |
| 961 | switch (event[counter].event_type) { |
| 962 | case READABLE_EVENT: |
| 963 | ioqueue_dispatch_read_event(ioqueue, event[counter].key); |
| 964 | break; |
| 965 | case WRITEABLE_EVENT: |
| 966 | ioqueue_dispatch_write_event(ioqueue, event[counter].key); |
| 967 | break; |
| 968 | case EXCEPTION_EVENT: |
| 969 | ioqueue_dispatch_exception_event(ioqueue, event[counter].key); |
| 970 | break; |
| 971 | case NO_EVENT: |
| 972 | pj_assert(!"Invalid event!"); |
| 973 | break; |
| 974 | } |
| 975 | |
| 976 | #if PJ_IOQUEUE_HAS_SAFE_UNREG |
| 977 | decrement_counter(event[counter].key); |
| 978 | #endif |
| 979 | |
| 980 | if (event[counter].key->grp_lock) |
| 981 | pj_grp_lock_dec_ref_dbg(event[counter].key->grp_lock, |
| 982 | "ioqueue", 0); |
| 983 | } |
| 984 | |
| 985 | |
| 986 | return count; |
| 987 | } |
| 988 | |