| 1 | /* |
| 2 | * Copyright 2008-2018 Aerospike, Inc. |
| 3 | * |
| 4 | * Portions may be licensed to Aerospike, Inc. under one or more contributor |
| 5 | * license agreements. |
| 6 | * |
| 7 | * Licensed under the Apache License, Version 2.0 (the "License"); you may not |
| 8 | * use this file except in compliance with the License. You may obtain a copy of |
| 9 | * the License at http://www.apache.org/licenses/LICENSE-2.0 |
| 10 | * |
| 11 | * Unless required by applicable law or agreed to in writing, software |
| 12 | * distributed under the License is distributed on an "AS IS" BASIS, WITHOUT |
| 13 | * WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the |
| 14 | * License for the specific language governing permissions and limitations under |
| 15 | * the License. |
| 16 | */ |
| 17 | #include <citrusleaf/cf_queue.h> |
| 18 | #include <citrusleaf/cf_clock.h> |
| 19 | #include <citrusleaf/alloc.h> |
| 20 | #include <string.h> |
| 21 | |
| 22 | /****************************************************************************** |
| 23 | * FUNCTIONS |
| 24 | ******************************************************************************/ |
| 25 | |
| 26 | bool |
| 27 | cf_queue_init(cf_queue *q, size_t element_sz, uint32_t capacity, |
| 28 | bool threadsafe) |
| 29 | { |
| 30 | q->alloc_sz = capacity; |
| 31 | q->write_offset = q->read_offset = 0; |
| 32 | q->element_sz = element_sz; |
| 33 | q->threadsafe = threadsafe; |
| 34 | q->free_struct = false; |
| 35 | |
| 36 | q->elements = (uint8_t*)cf_malloc(capacity * element_sz); |
| 37 | |
| 38 | if (! q->elements) { |
| 39 | return false; |
| 40 | } |
| 41 | |
| 42 | if (! q->threadsafe) { |
| 43 | return q; |
| 44 | } |
| 45 | |
| 46 | if (0 != pthread_mutex_init(&q->LOCK, NULL)) { |
| 47 | cf_free(q->elements); |
| 48 | return false; |
| 49 | } |
| 50 | |
| 51 | if (0 != pthread_cond_init(&q->CV, NULL)) { |
| 52 | pthread_mutex_destroy(&q->LOCK); |
| 53 | cf_free(q->elements); |
| 54 | return false; |
| 55 | } |
| 56 | |
| 57 | return true; |
| 58 | } |
| 59 | |
| 60 | cf_queue * |
| 61 | cf_queue_create(size_t element_sz, bool threadsafe) |
| 62 | { |
| 63 | cf_queue *q = (cf_queue*)cf_malloc(sizeof(cf_queue)); |
| 64 | |
| 65 | if (! q) { |
| 66 | return NULL; |
| 67 | } |
| 68 | |
| 69 | if (! cf_queue_init(q, element_sz, CF_QUEUE_ALLOCSZ, threadsafe)) { |
| 70 | cf_free(q); |
| 71 | return NULL; |
| 72 | } |
| 73 | |
| 74 | q->free_struct = true; |
| 75 | |
| 76 | return q; |
| 77 | } |
| 78 | |
| 79 | void |
| 80 | cf_queue_destroy(cf_queue *q) |
| 81 | { |
| 82 | if (q->threadsafe) { |
| 83 | pthread_cond_destroy(&q->CV); |
| 84 | pthread_mutex_destroy(&q->LOCK); |
| 85 | } |
| 86 | |
| 87 | cf_free(q->elements); |
| 88 | |
| 89 | if (q->free_struct) { |
| 90 | memset(q, 0, sizeof(cf_queue)); |
| 91 | cf_free(q); |
| 92 | } |
| 93 | } |
| 94 | |
| 95 | static inline void |
| 96 | cf_queue_lock(cf_queue *q) |
| 97 | { |
| 98 | if (q->threadsafe) { |
| 99 | pthread_mutex_lock(&q->LOCK); |
| 100 | } |
| 101 | } |
| 102 | |
| 103 | static inline void |
| 104 | cf_queue_unlock(cf_queue *q) |
| 105 | { |
| 106 | if (q->threadsafe) { |
| 107 | pthread_mutex_unlock(&q->LOCK); |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | int |
| 112 | cf_queue_sz(cf_queue *q) |
| 113 | { |
| 114 | cf_queue_lock(q); |
| 115 | int rv = CF_Q_SZ(q); |
| 116 | cf_queue_unlock(q); |
| 117 | |
| 118 | return rv; |
| 119 | } |
| 120 | |
| 121 | // |
| 122 | // Internal function. Call with new size with lock held. This function only |
| 123 | // works on full queues. |
| 124 | // |
| 125 | static int |
| 126 | cf_queue_resize(cf_queue *q, uint32_t new_sz) |
| 127 | { |
| 128 | // Check if queue is not full. |
| 129 | if (CF_Q_SZ(q) != q->alloc_sz) { |
| 130 | return CF_QUEUE_ERR; |
| 131 | } |
| 132 | |
| 133 | // The rare case where the queue is not fragmented, and realloc makes sense |
| 134 | // and none of the offsets need to move. |
| 135 | if (0 == q->read_offset % q->alloc_sz) { |
| 136 | q->elements = (uint8_t*)cf_realloc(q->elements, new_sz * q->element_sz); |
| 137 | |
| 138 | if (! q->elements) { |
| 139 | return CF_QUEUE_ERR; |
| 140 | } |
| 141 | |
| 142 | q->read_offset = 0; |
| 143 | q->write_offset = q->alloc_sz; |
| 144 | } |
| 145 | else { |
| 146 | uint8_t *newq = (uint8_t*)cf_malloc(new_sz * q->element_sz); |
| 147 | |
| 148 | if (! newq) { |
| 149 | return CF_QUEUE_ERR; |
| 150 | } |
| 151 | |
| 152 | // end_sz is used bytes in old queue from insert point to end. |
| 153 | size_t end_sz = |
| 154 | (q->alloc_sz - (q->read_offset % q->alloc_sz)) * q->element_sz; |
| 155 | |
| 156 | memcpy(&newq[0], CF_Q_ELEM_PTR(q, q->read_offset), end_sz); |
| 157 | memcpy(&newq[end_sz], &q->elements[0], |
| 158 | (q->alloc_sz * q->element_sz) - end_sz); |
| 159 | |
| 160 | cf_free(q->elements); |
| 161 | q->elements = newq; |
| 162 | |
| 163 | q->write_offset = q->alloc_sz; |
| 164 | q->read_offset = 0; |
| 165 | } |
| 166 | |
| 167 | q->alloc_sz = new_sz; |
| 168 | |
| 169 | return CF_QUEUE_OK; |
| 170 | } |
| 171 | |
| 172 | // |
| 173 | // We have to guard against wrap-around, call this occasionally. We really |
| 174 | // expect this will never get called, however it can be a symptom of a queue |
| 175 | // getting really, really deep. |
| 176 | // |
| 177 | static inline void |
| 178 | cf_queue_unwrap(cf_queue *q) |
| 179 | { |
| 180 | if ((q->write_offset & 0xC0000000) != 0) { |
| 181 | int sz = CF_Q_SZ(q); |
| 182 | |
| 183 | q->read_offset %= q->alloc_sz; |
| 184 | q->write_offset = q->read_offset + sz; |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | int |
| 189 | cf_queue_push(cf_queue *q, const void *ptr) |
| 190 | { |
| 191 | cf_queue_lock(q); |
| 192 | |
| 193 | // Check queue length. |
| 194 | if (CF_Q_SZ(q) == q->alloc_sz) { |
| 195 | if (0 != cf_queue_resize(q, q->alloc_sz * 2)) { |
| 196 | cf_queue_unlock(q); |
| 197 | return CF_QUEUE_ERR; |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | // TODO - if queues are power of 2, this can be a shift. |
| 202 | memcpy(CF_Q_ELEM_PTR(q, q->write_offset), ptr, q->element_sz); |
| 203 | q->write_offset++; |
| 204 | cf_queue_unwrap(q); |
| 205 | |
| 206 | if (q->threadsafe) { |
| 207 | pthread_cond_signal(&q->CV); |
| 208 | } |
| 209 | |
| 210 | cf_queue_unlock(q); |
| 211 | return CF_QUEUE_OK; |
| 212 | } |
| 213 | |
| 214 | // |
| 215 | // Push element on the queue only if size < limit. |
| 216 | // |
| 217 | bool |
| 218 | cf_queue_push_limit(cf_queue *q, const void *ptr, uint32_t limit) |
| 219 | { |
| 220 | cf_queue_lock(q); |
| 221 | |
| 222 | uint32_t size = CF_Q_SZ(q); |
| 223 | |
| 224 | if (size >= limit) { |
| 225 | cf_queue_unlock(q); |
| 226 | return false; |
| 227 | } |
| 228 | |
| 229 | if (size == q->alloc_sz) { |
| 230 | if (0 != cf_queue_resize(q, q->alloc_sz * 2)) { |
| 231 | cf_queue_unlock(q); |
| 232 | return false; |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | // TODO - if queues are power of 2, this can be a shift. |
| 237 | memcpy(CF_Q_ELEM_PTR(q, q->write_offset), ptr, q->element_sz); |
| 238 | q->write_offset++; |
| 239 | cf_queue_unwrap(q); |
| 240 | |
| 241 | if (q->threadsafe) { |
| 242 | pthread_cond_signal(&q->CV); |
| 243 | } |
| 244 | |
| 245 | cf_queue_unlock(q); |
| 246 | return true; |
| 247 | } |
| 248 | |
| 249 | // |
| 250 | // Same as cf_queue_push() except it's a no-op if element is already queued. |
| 251 | // |
| 252 | int |
| 253 | cf_queue_push_unique(cf_queue *q, const void *ptr) |
| 254 | { |
| 255 | cf_queue_lock(q); |
| 256 | |
| 257 | // Check if element is already queued. |
| 258 | if (CF_Q_SZ(q) != 0) { |
| 259 | for (uint32_t i = q->read_offset; i < q->write_offset; i++) { |
| 260 | if (0 == memcmp(CF_Q_ELEM_PTR(q, i), ptr, q->element_sz)) { |
| 261 | // Element is already queued. |
| 262 | // TODO - return 0 if all callers regard this as normal? |
| 263 | cf_queue_unlock(q); |
| 264 | return -2; |
| 265 | } |
| 266 | } |
| 267 | } |
| 268 | |
| 269 | if (CF_Q_SZ(q) == q->alloc_sz) { |
| 270 | if (0 != cf_queue_resize(q, q->alloc_sz * 2)) { |
| 271 | cf_queue_unlock(q); |
| 272 | return CF_QUEUE_ERR; |
| 273 | } |
| 274 | } |
| 275 | |
| 276 | // TODO - if queues are power of 2, this can be a shift. |
| 277 | memcpy(CF_Q_ELEM_PTR(q, q->write_offset), ptr, q->element_sz); |
| 278 | q->write_offset++; |
| 279 | cf_queue_unwrap(q); |
| 280 | |
| 281 | if (q->threadsafe) { |
| 282 | pthread_cond_signal(&q->CV); |
| 283 | } |
| 284 | |
| 285 | cf_queue_unlock(q); |
| 286 | return CF_QUEUE_OK; |
| 287 | } |
| 288 | |
| 289 | // |
| 290 | // Push to the front of the queue. |
| 291 | // |
| 292 | int |
| 293 | cf_queue_push_head(cf_queue *q, const void *ptr) |
| 294 | { |
| 295 | cf_queue_lock(q); |
| 296 | |
| 297 | if (CF_Q_SZ(q) == q->alloc_sz) { |
| 298 | if (0 != cf_queue_resize(q, q->alloc_sz * 2)) { |
| 299 | cf_queue_unlock(q); |
| 300 | return CF_QUEUE_ERR; |
| 301 | } |
| 302 | } |
| 303 | |
| 304 | if (q->read_offset == 0) { |
| 305 | q->read_offset += q->alloc_sz; |
| 306 | q->write_offset += q->alloc_sz; |
| 307 | } |
| 308 | |
| 309 | q->read_offset--; |
| 310 | memcpy(CF_Q_ELEM_PTR(q, q->read_offset), ptr, q->element_sz); |
| 311 | |
| 312 | cf_queue_unwrap(q); |
| 313 | |
| 314 | if (q->threadsafe) { |
| 315 | pthread_cond_signal(&q->CV); |
| 316 | } |
| 317 | |
| 318 | cf_queue_unlock(q); |
| 319 | return CF_QUEUE_OK; |
| 320 | } |
| 321 | |
| 322 | // |
| 323 | // If ms_wait < 0, wait forever. |
| 324 | // If ms_wait = 0, don't wait at all. |
| 325 | // If ms_wait > 0, wait that number of milliseconds. |
| 326 | // |
| 327 | int |
| 328 | cf_queue_pop(cf_queue *q, void *buf, int ms_wait) |
| 329 | { |
| 330 | struct timespec tp; |
| 331 | |
| 332 | if (ms_wait > 0) { |
| 333 | cf_set_wait_timespec(ms_wait, &tp); |
| 334 | } |
| 335 | |
| 336 | cf_queue_lock(q); |
| 337 | |
| 338 | if (q->threadsafe) { |
| 339 | |
| 340 | // Note that we have to use a while() loop. The pthread_cond_signal() |
| 341 | // documentation says that AT LEAST ONE waiting thread will be awakened. |
| 342 | // If more than one are awakened, the first will get the popped element, |
| 343 | // others will find the queue empty and go back to waiting. |
| 344 | |
| 345 | while (CF_Q_EMPTY(q)) { |
| 346 | if (CF_QUEUE_FOREVER == ms_wait) { |
| 347 | pthread_cond_wait(&q->CV, &q->LOCK); |
| 348 | } |
| 349 | else if (CF_QUEUE_NOWAIT == ms_wait) { |
| 350 | pthread_mutex_unlock(&q->LOCK); |
| 351 | return CF_QUEUE_EMPTY; |
| 352 | } |
| 353 | else { |
| 354 | pthread_cond_timedwait(&q->CV, &q->LOCK, &tp); |
| 355 | |
| 356 | if (CF_Q_EMPTY(q)) { |
| 357 | pthread_mutex_unlock(&q->LOCK); |
| 358 | return CF_QUEUE_EMPTY; |
| 359 | } |
| 360 | } |
| 361 | } |
| 362 | } |
| 363 | else if (CF_Q_EMPTY(q)) { |
| 364 | return CF_QUEUE_EMPTY; |
| 365 | } |
| 366 | |
| 367 | memcpy(buf, CF_Q_ELEM_PTR(q, q->read_offset), q->element_sz); |
| 368 | q->read_offset++; |
| 369 | |
| 370 | // This probably keeps the cache fresher because the queue is fully empty. |
| 371 | if (q->read_offset == q->write_offset) { |
| 372 | q->read_offset = q->write_offset = 0; |
| 373 | } |
| 374 | |
| 375 | cf_queue_unlock(q); |
| 376 | return CF_QUEUE_OK; |
| 377 | } |
| 378 | |
| 379 | void |
| 380 | cf_queue_delete_offset(cf_queue *q, uint32_t index) |
| 381 | { |
| 382 | index %= q->alloc_sz; |
| 383 | |
| 384 | uint32_t r_index = q->read_offset % q->alloc_sz; |
| 385 | uint32_t w_index = q->write_offset % q->alloc_sz; |
| 386 | |
| 387 | // Assumes index is validated! |
| 388 | |
| 389 | // If we're deleting the one at the head, just increase the read offset. |
| 390 | if (index == r_index) { |
| 391 | q->read_offset++; |
| 392 | return; |
| 393 | } |
| 394 | |
| 395 | // If we're deleting the tail just decrease the write offset. |
| 396 | if (w_index && (index == w_index - 1)) { |
| 397 | q->write_offset--; |
| 398 | return; |
| 399 | } |
| 400 | |
| 401 | if (index > r_index) { |
| 402 | // The memory copy is overlapping, so must use memmove(). |
| 403 | memmove(&q->elements[(r_index + 1) * q->element_sz], |
| 404 | &q->elements[r_index * q->element_sz], |
| 405 | (index - r_index) * q->element_sz); |
| 406 | q->read_offset++; |
| 407 | return; |
| 408 | } |
| 409 | |
| 410 | if (index < w_index) { |
| 411 | // The memory copy is overlapping, so must use memmove(). |
| 412 | memmove(&q->elements[index * q->element_sz], |
| 413 | &q->elements[(index + 1) * q->element_sz], |
| 414 | (w_index - index - 1) * q->element_sz); |
| 415 | q->write_offset--; |
| 416 | } |
| 417 | } |
| 418 | |
| 419 | // |
| 420 | // Iterate over all queue members, calling the callback. |
| 421 | // |
| 422 | int |
| 423 | cf_queue_reduce(cf_queue *q, cf_queue_reduce_fn cb, void *udata) |
| 424 | { |
| 425 | cf_queue_lock(q); |
| 426 | |
| 427 | if (CF_Q_SZ(q) != 0) { |
| 428 | for (uint32_t i = q->read_offset; i < q->write_offset; i++) { |
| 429 | int rv = cb(CF_Q_ELEM_PTR(q, i), udata); |
| 430 | |
| 431 | if (rv == 0) { |
| 432 | continue; |
| 433 | } |
| 434 | |
| 435 | if (rv == -1) { |
| 436 | // Found what it was looking for, stop reducing. |
| 437 | break; |
| 438 | } |
| 439 | |
| 440 | if (rv == -2) { |
| 441 | // Delete, and stop reducing. |
| 442 | cf_queue_delete_offset(q, i); |
| 443 | break; |
| 444 | } |
| 445 | } |
| 446 | } |
| 447 | |
| 448 | cf_queue_unlock(q); |
| 449 | return CF_QUEUE_OK; |
| 450 | } |
| 451 | |
| 452 | // |
| 453 | // Iterate over all queue members, calling the callback. Pop element (or not) |
| 454 | // based on callback return value. |
| 455 | // |
| 456 | int |
| 457 | cf_queue_reduce_pop(cf_queue *q, void *buf, int ms_wait, cf_queue_reduce_fn cb, |
| 458 | void *udata) |
| 459 | { |
| 460 | struct timespec tp; |
| 461 | |
| 462 | if (ms_wait > 0) { |
| 463 | cf_set_wait_timespec(ms_wait, &tp); |
| 464 | } |
| 465 | |
| 466 | cf_queue_lock(q); |
| 467 | |
| 468 | if (q->threadsafe) { |
| 469 | |
| 470 | // Note that we have to use a while() loop. The pthread_cond_signal() |
| 471 | // documentation says that AT LEAST ONE waiting thread will be awakened. |
| 472 | // If more than one are awakened, the first will get the popped element, |
| 473 | // others will find the queue empty and go back to waiting. |
| 474 | |
| 475 | while (CF_Q_EMPTY(q)) { |
| 476 | if (CF_QUEUE_FOREVER == ms_wait) { |
| 477 | pthread_cond_wait(&q->CV, &q->LOCK); |
| 478 | } |
| 479 | else if (CF_QUEUE_NOWAIT == ms_wait) { |
| 480 | pthread_mutex_unlock(&q->LOCK); |
| 481 | return CF_QUEUE_EMPTY; |
| 482 | } |
| 483 | else { |
| 484 | pthread_cond_timedwait(&q->CV, &q->LOCK, &tp); |
| 485 | |
| 486 | if (CF_Q_EMPTY(q)) { |
| 487 | pthread_mutex_unlock(&q->LOCK); |
| 488 | return CF_QUEUE_EMPTY; |
| 489 | } |
| 490 | } |
| 491 | } |
| 492 | } |
| 493 | else if (CF_Q_EMPTY(q)) { |
| 494 | return CF_QUEUE_EMPTY; |
| 495 | } |
| 496 | |
| 497 | uint32_t best_index = q->read_offset; |
| 498 | |
| 499 | for (uint32_t i = q->read_offset; i < q->write_offset; i++) { |
| 500 | int rv = cb(CF_Q_ELEM_PTR(q, i), udata); |
| 501 | |
| 502 | if (rv == 0) { |
| 503 | continue; |
| 504 | } |
| 505 | |
| 506 | if (rv == -1) { |
| 507 | // Found what it was looking for, so break. |
| 508 | best_index = i; |
| 509 | break; |
| 510 | } |
| 511 | |
| 512 | if (rv == -2) { |
| 513 | // Found new candidate, but keep looking for a better one. |
| 514 | best_index = i; |
| 515 | } |
| 516 | } |
| 517 | |
| 518 | memcpy(buf, CF_Q_ELEM_PTR(q, best_index), q->element_sz); |
| 519 | cf_queue_delete_offset(q, best_index); |
| 520 | |
| 521 | cf_queue_unlock(q); |
| 522 | |
| 523 | return CF_QUEUE_OK; |
| 524 | } |
| 525 | |
| 526 | // |
| 527 | // Iterate over all queue members starting from the tail, calling the callback. |
| 528 | // |
| 529 | int |
| 530 | cf_queue_reduce_reverse(cf_queue *q, cf_queue_reduce_fn cb, void *udata) |
| 531 | { |
| 532 | cf_queue_lock(q); |
| 533 | |
| 534 | if (CF_Q_SZ(q) != 0) { |
| 535 | for (int i = (int)q->write_offset - 1; i >= (int)q->read_offset; i--) { |
| 536 | int rv = cb(CF_Q_ELEM_PTR(q, i), udata); |
| 537 | |
| 538 | if (rv == 0) { |
| 539 | continue; |
| 540 | } |
| 541 | |
| 542 | if (rv == -1) { |
| 543 | // Found what it was looking for, stop reducing. |
| 544 | break; |
| 545 | } |
| 546 | |
| 547 | if (rv == -2) { |
| 548 | // Delete, and stop reducing. |
| 549 | cf_queue_delete_offset(q, i); |
| 550 | break; |
| 551 | } |
| 552 | } |
| 553 | } |
| 554 | |
| 555 | cf_queue_unlock(q); |
| 556 | return CF_QUEUE_OK; |
| 557 | } |
| 558 | |
| 559 | // |
| 560 | // Delete element(s) from the queue. Pass 'only_one' as true if there can be |
| 561 | // only one element with this value on the queue. |
| 562 | // |
| 563 | int |
| 564 | cf_queue_delete(cf_queue *q, const void *ptr, bool only_one) |
| 565 | { |
| 566 | cf_queue_lock(q); |
| 567 | |
| 568 | bool found = false; |
| 569 | |
| 570 | if (CF_Q_SZ(q) != 0) { |
| 571 | for (uint32_t i = q->read_offset; i < q->write_offset; i++) { |
| 572 | int rv = 0; |
| 573 | |
| 574 | // If buf is null, rv is always 0 and we delete all elements. |
| 575 | if (ptr) { |
| 576 | rv = memcmp(CF_Q_ELEM_PTR(q, i), ptr, q->element_sz); |
| 577 | } |
| 578 | |
| 579 | if (rv == 0) { |
| 580 | cf_queue_delete_offset(q, i); |
| 581 | found = true; |
| 582 | |
| 583 | if (only_one) { |
| 584 | break; |
| 585 | } |
| 586 | } |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | cf_queue_unlock(q); |
| 591 | return found ? CF_QUEUE_OK : CF_QUEUE_EMPTY; |
| 592 | } |
| 593 | |
| 594 | int |
| 595 | cf_queue_delete_all(cf_queue *q) |
| 596 | { |
| 597 | return cf_queue_delete(q, NULL, false); |
| 598 | } |
| 599 | |