| 1 | /* |
| 2 | * Copyright (c) 1998-2000 Apple Computer, Inc. All rights reserved. |
| 3 | * |
| 4 | * @APPLE_OSREFERENCE_LICENSE_HEADER_START@ |
| 5 | * |
| 6 | * This file contains Original Code and/or Modifications of Original Code |
| 7 | * as defined in and that are subject to the Apple Public Source License |
| 8 | * Version 2.0 (the 'License'). You may not use this file except in |
| 9 | * compliance with the License. The rights granted to you under the License |
| 10 | * may not be used to create, or enable the creation or redistribution of, |
| 11 | * unlawful or unlicensed copies of an Apple operating system, or to |
| 12 | * circumvent, violate, or enable the circumvention or violation of, any |
| 13 | * terms of an Apple operating system software license agreement. |
| 14 | * |
| 15 | * Please obtain a copy of the License at |
| 16 | * http://www.opensource.apple.com/apsl/ and read it before using this file. |
| 17 | * |
| 18 | * The Original Code and all software distributed under the License are |
| 19 | * distributed on an 'AS IS' basis, WITHOUT WARRANTY OF ANY KIND, EITHER |
| 20 | * EXPRESS OR IMPLIED, AND APPLE HEREBY DISCLAIMS ALL SUCH WARRANTIES, |
| 21 | * INCLUDING WITHOUT LIMITATION, ANY WARRANTIES OF MERCHANTABILITY, |
| 22 | * FITNESS FOR A PARTICULAR PURPOSE, QUIET ENJOYMENT OR NON-INFRINGEMENT. |
| 23 | * Please see the License for the specific language governing rights and |
| 24 | * limitations under the License. |
| 25 | * |
| 26 | * @APPLE_OSREFERENCE_LICENSE_HEADER_END@ |
| 27 | */ |
| 28 | |
| 29 | #define IOKIT_ENABLE_SHARED_PTR |
| 30 | |
| 31 | #include <IOKit/IOSharedDataQueue.h> |
| 32 | #include <IOKit/IODataQueueShared.h> |
| 33 | #include <IOKit/IOLib.h> |
| 34 | #include <IOKit/IOMemoryDescriptor.h> |
| 35 | #include <libkern/c++/OSSharedPtr.h> |
| 36 | |
| 37 | #ifdef enqueue |
| 38 | #undef enqueue |
| 39 | #endif |
| 40 | |
| 41 | #ifdef dequeue |
| 42 | #undef dequeue |
| 43 | #endif |
| 44 | |
| 45 | #define super IODataQueue |
| 46 | |
| 47 | OSDefineMetaClassAndStructors(IOSharedDataQueue, IODataQueue) |
| 48 | |
| 49 | OSSharedPtr<IOSharedDataQueue> |
| 50 | IOSharedDataQueue::withCapacity(UInt32 size) |
| 51 | { |
| 52 | OSSharedPtr<IOSharedDataQueue> dataQueue = OSMakeShared<IOSharedDataQueue>(); |
| 53 | |
| 54 | if (dataQueue) { |
| 55 | if (!dataQueue->initWithCapacity(size)) { |
| 56 | return nullptr; |
| 57 | } |
| 58 | } |
| 59 | |
| 60 | return dataQueue; |
| 61 | } |
| 62 | |
| 63 | OSSharedPtr<IOSharedDataQueue> |
| 64 | IOSharedDataQueue::withEntries(UInt32 numEntries, UInt32 entrySize) |
| 65 | { |
| 66 | OSSharedPtr<IOSharedDataQueue> dataQueue = OSMakeShared<IOSharedDataQueue>(); |
| 67 | |
| 68 | if (dataQueue) { |
| 69 | if (!dataQueue->initWithEntries(numEntries, entrySize)) { |
| 70 | return nullptr; |
| 71 | } |
| 72 | } |
| 73 | |
| 74 | return dataQueue; |
| 75 | } |
| 76 | |
| 77 | Boolean |
| 78 | IOSharedDataQueue::initWithCapacity(UInt32 size) |
| 79 | { |
| 80 | IODataQueueAppendix * appendix; |
| 81 | vm_size_t allocSize; |
| 82 | kern_return_t kr; |
| 83 | |
| 84 | if (!super::init()) { |
| 85 | return false; |
| 86 | } |
| 87 | |
| 88 | _reserved = IOMallocType(ExpansionData); |
| 89 | if (!_reserved) { |
| 90 | return false; |
| 91 | } |
| 92 | |
| 93 | if (size > UINT32_MAX - DATA_QUEUE_MEMORY_HEADER_SIZE - DATA_QUEUE_MEMORY_APPENDIX_SIZE) { |
| 94 | return false; |
| 95 | } |
| 96 | |
| 97 | allocSize = round_page(x: size + DATA_QUEUE_MEMORY_HEADER_SIZE + DATA_QUEUE_MEMORY_APPENDIX_SIZE); |
| 98 | |
| 99 | if (allocSize < size) { |
| 100 | return false; |
| 101 | } |
| 102 | |
| 103 | kr = kmem_alloc(map: kernel_map, addrp: (vm_offset_t *)&dataQueue, size: allocSize, |
| 104 | flags: (kma_flags_t)(KMA_DATA | KMA_ZERO), tag: IOMemoryTag(map: kernel_map)); |
| 105 | if (kr != KERN_SUCCESS) { |
| 106 | return false; |
| 107 | } |
| 108 | |
| 109 | dataQueue->queueSize = size; |
| 110 | // dataQueue->head = 0; |
| 111 | // dataQueue->tail = 0; |
| 112 | |
| 113 | if (!setQueueSize(size)) { |
| 114 | return false; |
| 115 | } |
| 116 | |
| 117 | appendix = (IODataQueueAppendix *)((UInt8 *)dataQueue + size + DATA_QUEUE_MEMORY_HEADER_SIZE); |
| 118 | appendix->version = 0; |
| 119 | |
| 120 | if (!notifyMsg) { |
| 121 | notifyMsg = IOMallocType(mach_msg_header_t); |
| 122 | if (!notifyMsg) { |
| 123 | return false; |
| 124 | } |
| 125 | } |
| 126 | bzero(s: notifyMsg, n: sizeof(mach_msg_header_t)); |
| 127 | |
| 128 | setNotificationPort(MACH_PORT_NULL); |
| 129 | |
| 130 | return true; |
| 131 | } |
| 132 | |
| 133 | void |
| 134 | IOSharedDataQueue::free() |
| 135 | { |
| 136 | if (dataQueue) { |
| 137 | kmem_free(map: kernel_map, addr: (vm_offset_t)dataQueue, size: round_page(x: getQueueSize() + |
| 138 | DATA_QUEUE_MEMORY_HEADER_SIZE + DATA_QUEUE_MEMORY_APPENDIX_SIZE)); |
| 139 | dataQueue = NULL; |
| 140 | if (notifyMsg) { |
| 141 | IOFreeType(notifyMsg, mach_msg_header_t); |
| 142 | notifyMsg = NULL; |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | if (_reserved) { |
| 147 | IOFreeType(_reserved, ExpansionData); |
| 148 | _reserved = NULL; |
| 149 | } |
| 150 | |
| 151 | super::free(); |
| 152 | } |
| 153 | |
| 154 | OSSharedPtr<IOMemoryDescriptor> |
| 155 | IOSharedDataQueue::getMemoryDescriptor() |
| 156 | { |
| 157 | OSSharedPtr<IOMemoryDescriptor> descriptor; |
| 158 | |
| 159 | if (dataQueue != NULL) { |
| 160 | descriptor = IOMemoryDescriptor::withAddress(address: dataQueue, withLength: getQueueSize() + DATA_QUEUE_MEMORY_HEADER_SIZE + DATA_QUEUE_MEMORY_APPENDIX_SIZE, withDirection: kIODirectionOutIn); |
| 161 | } |
| 162 | |
| 163 | return descriptor; |
| 164 | } |
| 165 | |
| 166 | |
| 167 | IODataQueueEntry * |
| 168 | IOSharedDataQueue::peek() |
| 169 | { |
| 170 | IODataQueueEntry *entry = NULL; |
| 171 | UInt32 headOffset; |
| 172 | UInt32 tailOffset; |
| 173 | |
| 174 | if (!dataQueue) { |
| 175 | return NULL; |
| 176 | } |
| 177 | |
| 178 | // Read head and tail with acquire barrier |
| 179 | // See rdar://problem/40780584 for an explanation of relaxed/acquire barriers |
| 180 | headOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_RELAXED); |
| 181 | tailOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->tail, __ATOMIC_ACQUIRE); |
| 182 | |
| 183 | if (headOffset != tailOffset) { |
| 184 | volatile IODataQueueEntry * head = NULL; |
| 185 | UInt32 headSize = 0; |
| 186 | UInt32 headOffset = dataQueue->head; |
| 187 | UInt32 queueSize = getQueueSize(); |
| 188 | |
| 189 | if (headOffset > queueSize) { |
| 190 | return NULL; |
| 191 | } |
| 192 | |
| 193 | head = (IODataQueueEntry *)((char *)dataQueue->queue + headOffset); |
| 194 | headSize = head->size; |
| 195 | |
| 196 | // Check if there's enough room before the end of the queue for a header. |
| 197 | // If there is room, check if there's enough room to hold the header and |
| 198 | // the data. |
| 199 | |
| 200 | if ((headOffset > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) || |
| 201 | (headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize) || |
| 202 | (headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > UINT32_MAX - headSize) || |
| 203 | (headOffset + headSize + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize)) { |
| 204 | // No room for the header or the data, wrap to the beginning of the queue. |
| 205 | // Note: wrapping even with the UINT32_MAX checks, as we have to support |
| 206 | // queueSize of UINT32_MAX |
| 207 | entry = dataQueue->queue; |
| 208 | } else { |
| 209 | entry = (IODataQueueEntry *)head; |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | return entry; |
| 214 | } |
| 215 | |
| 216 | Boolean |
| 217 | IOSharedDataQueue::enqueue(void * data, UInt32 dataSize) |
| 218 | { |
| 219 | UInt32 head; |
| 220 | UInt32 tail; |
| 221 | UInt32 newTail; |
| 222 | const UInt32 entrySize = dataSize + DATA_QUEUE_ENTRY_HEADER_SIZE; |
| 223 | IODataQueueEntry * entry; |
| 224 | |
| 225 | // Force a single read of head and tail |
| 226 | // See rdar://problem/40780584 for an explanation of relaxed/acquire barriers |
| 227 | tail = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->tail, __ATOMIC_RELAXED); |
| 228 | head = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_ACQUIRE); |
| 229 | |
| 230 | // Check for overflow of entrySize |
| 231 | if (dataSize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) { |
| 232 | return false; |
| 233 | } |
| 234 | // Check for underflow of (getQueueSize() - tail) |
| 235 | if (getQueueSize() < tail || getQueueSize() < head) { |
| 236 | return false; |
| 237 | } |
| 238 | |
| 239 | if (tail >= head) { |
| 240 | // Is there enough room at the end for the entry? |
| 241 | if ((entrySize <= UINT32_MAX - tail) && |
| 242 | ((tail + entrySize) <= getQueueSize())) { |
| 243 | entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail); |
| 244 | |
| 245 | entry->size = dataSize; |
| 246 | __nochk_memcpy(dst: &entry->data, src: data, n: dataSize); |
| 247 | |
| 248 | // The tail can be out of bound when the size of the new entry |
| 249 | // exactly matches the available space at the end of the queue. |
| 250 | // The tail can range from 0 to dataQueue->queueSize inclusive. |
| 251 | |
| 252 | newTail = tail + entrySize; |
| 253 | } else if (head > entrySize) { // Is there enough room at the beginning? |
| 254 | // Wrap around to the beginning, but do not allow the tail to catch |
| 255 | // up to the head. |
| 256 | |
| 257 | dataQueue->queue->size = dataSize; |
| 258 | |
| 259 | // We need to make sure that there is enough room to set the size before |
| 260 | // doing this. The user client checks for this and will look for the size |
| 261 | // at the beginning if there isn't room for it at the end. |
| 262 | |
| 263 | if ((getQueueSize() - tail) >= DATA_QUEUE_ENTRY_HEADER_SIZE) { |
| 264 | ((IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail))->size = dataSize; |
| 265 | } |
| 266 | |
| 267 | __nochk_memcpy(dst: &dataQueue->queue->data, src: data, n: dataSize); |
| 268 | newTail = entrySize; |
| 269 | } else { |
| 270 | return false; // queue is full |
| 271 | } |
| 272 | } else { |
| 273 | // Do not allow the tail to catch up to the head when the queue is full. |
| 274 | // That's why the comparison uses a '>' rather than '>='. |
| 275 | |
| 276 | if ((head - tail) > entrySize) { |
| 277 | entry = (IODataQueueEntry *)((UInt8 *)dataQueue->queue + tail); |
| 278 | |
| 279 | entry->size = dataSize; |
| 280 | __nochk_memcpy(dst: &entry->data, src: data, n: dataSize); |
| 281 | newTail = tail + entrySize; |
| 282 | } else { |
| 283 | return false; // queue is full |
| 284 | } |
| 285 | } |
| 286 | |
| 287 | // Publish the data we just enqueued |
| 288 | __c11_atomic_store((_Atomic UInt32 *)&dataQueue->tail, newTail, __ATOMIC_RELEASE); |
| 289 | |
| 290 | if (tail != head) { |
| 291 | // |
| 292 | // The memory barrier below paris with the one in ::dequeue |
| 293 | // so that either our store to the tail cannot be missed by |
| 294 | // the next dequeue attempt, or we will observe the dequeuer |
| 295 | // making the queue empty. |
| 296 | // |
| 297 | // Of course, if we already think the queue is empty, |
| 298 | // there's no point paying this extra cost. |
| 299 | // |
| 300 | __c11_atomic_thread_fence(__ATOMIC_SEQ_CST); |
| 301 | head = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_RELAXED); |
| 302 | } |
| 303 | |
| 304 | if (tail == head) { |
| 305 | // Send notification (via mach message) that data is now available. |
| 306 | sendDataAvailableNotification(); |
| 307 | } |
| 308 | return true; |
| 309 | } |
| 310 | |
| 311 | Boolean |
| 312 | IOSharedDataQueue::dequeue(void *data, UInt32 *dataSize) |
| 313 | { |
| 314 | Boolean retVal = TRUE; |
| 315 | volatile IODataQueueEntry * entry = NULL; |
| 316 | UInt32 entrySize = 0; |
| 317 | UInt32 headOffset = 0; |
| 318 | UInt32 tailOffset = 0; |
| 319 | UInt32 newHeadOffset = 0; |
| 320 | |
| 321 | if (!dataQueue || (data && !dataSize)) { |
| 322 | return false; |
| 323 | } |
| 324 | |
| 325 | // Read head and tail with acquire barrier |
| 326 | // See rdar://problem/40780584 for an explanation of relaxed/acquire barriers |
| 327 | headOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->head, __ATOMIC_RELAXED); |
| 328 | tailOffset = __c11_atomic_load((_Atomic UInt32 *)&dataQueue->tail, __ATOMIC_ACQUIRE); |
| 329 | |
| 330 | if (headOffset != tailOffset) { |
| 331 | volatile IODataQueueEntry * head = NULL; |
| 332 | UInt32 headSize = 0; |
| 333 | UInt32 queueSize = getQueueSize(); |
| 334 | |
| 335 | if (headOffset > queueSize) { |
| 336 | return false; |
| 337 | } |
| 338 | |
| 339 | head = (IODataQueueEntry *)((char *)dataQueue->queue + headOffset); |
| 340 | headSize = head->size; |
| 341 | |
| 342 | // we wrapped around to beginning, so read from there |
| 343 | // either there was not even room for the header |
| 344 | if ((headOffset > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) || |
| 345 | (headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize) || |
| 346 | // or there was room for the header, but not for the data |
| 347 | (headOffset + DATA_QUEUE_ENTRY_HEADER_SIZE > UINT32_MAX - headSize) || |
| 348 | (headOffset + headSize + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize)) { |
| 349 | // Note: we have to wrap to the beginning even with the UINT32_MAX checks |
| 350 | // because we have to support a queueSize of UINT32_MAX. |
| 351 | entry = dataQueue->queue; |
| 352 | entrySize = entry->size; |
| 353 | if ((entrySize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) || |
| 354 | (entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE > queueSize)) { |
| 355 | return false; |
| 356 | } |
| 357 | newHeadOffset = entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE; |
| 358 | // else it is at the end |
| 359 | } else { |
| 360 | entry = head; |
| 361 | entrySize = entry->size; |
| 362 | if ((entrySize > UINT32_MAX - DATA_QUEUE_ENTRY_HEADER_SIZE) || |
| 363 | (entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE > UINT32_MAX - headOffset) || |
| 364 | (entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE + headOffset > queueSize)) { |
| 365 | return false; |
| 366 | } |
| 367 | newHeadOffset = headOffset + entrySize + DATA_QUEUE_ENTRY_HEADER_SIZE; |
| 368 | } |
| 369 | } else { |
| 370 | // empty queue |
| 371 | return false; |
| 372 | } |
| 373 | |
| 374 | if (data) { |
| 375 | if (entrySize > *dataSize) { |
| 376 | // not enough space |
| 377 | return false; |
| 378 | } |
| 379 | __nochk_memcpy(dst: data, src: (void *)entry->data, n: entrySize); |
| 380 | *dataSize = entrySize; |
| 381 | } |
| 382 | |
| 383 | __c11_atomic_store((_Atomic UInt32 *)&dataQueue->head, newHeadOffset, __ATOMIC_RELEASE); |
| 384 | |
| 385 | if (newHeadOffset == tailOffset) { |
| 386 | // |
| 387 | // If we are making the queue empty, then we need to make sure |
| 388 | // that either the enqueuer notices, or we notice the enqueue |
| 389 | // that raced with our making of the queue empty. |
| 390 | // |
| 391 | __c11_atomic_thread_fence(__ATOMIC_SEQ_CST); |
| 392 | } |
| 393 | |
| 394 | return retVal; |
| 395 | } |
| 396 | |
| 397 | UInt32 |
| 398 | IOSharedDataQueue::getQueueSize() |
| 399 | { |
| 400 | if (!_reserved) { |
| 401 | return 0; |
| 402 | } |
| 403 | return _reserved->queueSize; |
| 404 | } |
| 405 | |
| 406 | Boolean |
| 407 | IOSharedDataQueue::setQueueSize(UInt32 size) |
| 408 | { |
| 409 | if (!_reserved) { |
| 410 | return false; |
| 411 | } |
| 412 | _reserved->queueSize = size; |
| 413 | return true; |
| 414 | } |
| 415 | |
| 416 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 0); |
| 417 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 1); |
| 418 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 2); |
| 419 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 3); |
| 420 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 4); |
| 421 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 5); |
| 422 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 6); |
| 423 | OSMetaClassDefineReservedUnused(IOSharedDataQueue, 7); |
| 424 | |