1 #ifndef IO_URING_TYPES_H 2 #define IO_URING_TYPES_H 3 4 #include <linux/blkdev.h> 5 #include <linux/hashtable.h> 6 #include <linux/task_work.h> 7 #include <linux/bitmap.h> 8 #include <linux/llist.h> 9 #include <uapi/linux/io_uring.h> 10 11 struct iou_loop_params; 12 struct io_uring_bpf_ops; 13 14 enum { 15 /* 16 * A hint to not wake right away but delay until there are enough of 17 * tw's queued to match the number of CQEs the task is waiting for. 18 * 19 * Must not be used with requests generating more than one CQE. 20 * It's also ignored unless IORING_SETUP_DEFER_TASKRUN is set. 21 */ 22 IOU_F_TWQ_LAZY_WAKE = 1, 23 }; 24 25 enum io_uring_cmd_flags { 26 IO_URING_F_COMPLETE_DEFER = 1, 27 IO_URING_F_UNLOCKED = 2, 28 /* the request is executed from poll, it should not be freed */ 29 IO_URING_F_MULTISHOT = 4, 30 /* executed by io-wq */ 31 IO_URING_F_IOWQ = 8, 32 /* executed inline from syscall */ 33 IO_URING_F_INLINE = 16, 34 /* int's last bit, sign checks are usually faster than a bit test */ 35 IO_URING_F_NONBLOCK = INT_MIN, 36 37 /* ctx state flags, for URING_CMD */ 38 IO_URING_F_SQE128 = (1 << 8), 39 IO_URING_F_CQE32 = (1 << 9), 40 IO_URING_F_IOPOLL = (1 << 10), 41 42 /* set when uring wants to cancel a previously issued command */ 43 IO_URING_F_CANCEL = (1 << 11), 44 IO_URING_F_COMPAT = (1 << 12), 45 }; 46 47 struct iou_loop_params; 48 49 struct io_wq_work_node { 50 struct io_wq_work_node *next; 51 }; 52 53 struct io_wq_work_list { 54 struct io_wq_work_node *first; 55 struct io_wq_work_node *last; 56 }; 57 58 /* 59 * Lockless multi-producer, single-consumer FIFO queue, see 60 * io_uring/mpscq.h for the implementation and rules. Defined here so 61 * that it can be embedded in io_ring_ctx. This is the producer side 62 * only - the consumer cursor is kept separately, on a cacheline that 63 * isn't dirtied by the producers. 64 */ 65 struct mpscq { 66 struct llist_node *tail; /* producers */ 67 struct llist_node stub; 68 }; 69 70 struct io_wq_work { 71 struct io_wq_work_node list; 72 atomic_t flags; 73 /* place it here instead of io_kiocb as it fills padding and saves 4B */ 74 int cancel_seq; 75 }; 76 77 struct io_rsrc_data { 78 unsigned int nr; 79 struct io_rsrc_node **nodes; 80 }; 81 82 struct io_file_table { 83 struct io_rsrc_data data; 84 unsigned long *bitmap; 85 unsigned int alloc_hint; 86 }; 87 88 struct io_hash_bucket { 89 struct hlist_head list; 90 } ____cacheline_aligned_in_smp; 91 92 struct io_hash_table { 93 struct io_hash_bucket *hbs; 94 unsigned hash_bits; 95 }; 96 97 struct io_mapped_region { 98 struct page **pages; 99 void *ptr; 100 unsigned nr_pages; 101 unsigned flags; 102 }; 103 104 /* 105 * Return value from io_buffer_list selection, to avoid stashing it in 106 * struct io_kiocb. For legacy/classic provided buffers, keeping a reference 107 * across execution contexts are fine. But for ring provided buffers, the 108 * list may go away as soon as ->uring_lock is dropped. As the io_kiocb 109 * persists, it's better to just keep the buffer local for those cases. 110 */ 111 struct io_br_sel { 112 struct io_buffer_list *buf_list; 113 /* 114 * Some selection parts return the user address, others return an error. 115 */ 116 union { 117 void __user *addr; 118 ssize_t val; 119 }; 120 }; 121 122 123 /* 124 * Arbitrary limit, can be raised if need be 125 */ 126 #define IO_RINGFD_REG_MAX 16 127 128 struct io_uring_task { 129 /* submission side */ 130 int cached_refs; 131 const struct io_ring_ctx *last; 132 struct task_struct *task; 133 struct io_wq *io_wq; 134 /* 135 * Consumer cursor for ->task_list. Only popped by the task itself, 136 * or by ->fallback_work once the task can no longer run task_work. 137 */ 138 struct llist_node *task_head; 139 struct file *registered_rings[IO_RINGFD_REG_MAX]; 140 141 struct xarray xa; 142 struct wait_queue_head wait; 143 atomic_t in_cancel; 144 atomic_t inflight_tracked; 145 struct percpu_counter inflight; 146 147 /* drains ->task_list once the task can no longer run task_work */ 148 struct work_struct fallback_work; 149 150 struct { /* task_work */ 151 struct mpscq task_list; 152 /* BIT(0) guards adding tw only once */ 153 unsigned long tw_pending; 154 struct callback_head task_work; 155 } ____cacheline_aligned_in_smp; 156 }; 157 158 struct iou_vec { 159 union { 160 struct iovec *iovec; 161 struct bio_vec *bvec; 162 }; 163 unsigned nr; /* number of struct iovec it can hold */ 164 }; 165 166 struct io_uring { 167 u32 head; 168 u32 tail; 169 }; 170 171 /* 172 * This data is shared with the application through the mmap at offsets 173 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. 174 * 175 * The offsets to the member fields are published through struct 176 * io_sqring_offsets when calling io_uring_setup. 177 */ 178 struct io_rings { 179 /* 180 * Head and tail offsets into the ring; the offsets need to be 181 * masked to get valid indices. 182 * 183 * The kernel controls head of the sq ring and the tail of the cq ring, 184 * and the application controls tail of the sq ring and the head of the 185 * cq ring. 186 */ 187 struct io_uring sq, cq; 188 /* 189 * Bitmasks to apply to head and tail offsets (constant, equals 190 * ring_entries - 1) 191 */ 192 u32 sq_ring_mask, cq_ring_mask; 193 /* Ring sizes (constant, power of 2) */ 194 u32 sq_ring_entries, cq_ring_entries; 195 /* 196 * Number of invalid entries dropped by the kernel due to 197 * invalid index stored in array 198 * 199 * Written by the kernel, shouldn't be modified by the 200 * application (i.e. get number of "new events" by comparing to 201 * cached value). 202 * 203 * After a new SQ head value was read by the application this 204 * counter includes all submissions that were dropped reaching 205 * the new SQ head (and possibly more). 206 */ 207 u32 sq_dropped; 208 /* 209 * Runtime SQ flags 210 * 211 * Written by the kernel, shouldn't be modified by the 212 * application. 213 * 214 * The application needs a full memory barrier before checking 215 * for IORING_SQ_NEED_WAKEUP after updating the sq tail. 216 */ 217 atomic_t sq_flags; 218 /* 219 * Runtime CQ flags 220 * 221 * Written by the application, shouldn't be modified by the 222 * kernel. 223 */ 224 u32 cq_flags; 225 /* 226 * Number of completion events lost because the queue was full; 227 * this should be avoided by the application by making sure 228 * there are not more requests pending than there is space in 229 * the completion queue. 230 * 231 * Written by the kernel, shouldn't be modified by the 232 * application (i.e. get number of "new events" by comparing to 233 * cached value). 234 * 235 * As completion events come in out of order this counter is not 236 * ordered with any other data. 237 */ 238 u32 cq_overflow; 239 /* 240 * Ring buffer of completion events. 241 * 242 * The kernel writes completion events fresh every time they are 243 * produced, so the application is allowed to modify pending 244 * entries. 245 */ 246 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; 247 }; 248 249 struct io_bpf_filter; 250 struct io_bpf_filters { 251 refcount_t refs; /* ref for ->bpf_filters */ 252 spinlock_t lock; /* protects ->bpf_filters modifications */ 253 struct io_bpf_filter __rcu **filters; 254 struct rcu_head rcu_head; 255 }; 256 257 struct io_restriction { 258 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); 259 DECLARE_BITMAP(sqe_op, IORING_OP_LAST); 260 struct io_bpf_filters *bpf_filters; 261 /* ->bpf_filters needs COW on modification */ 262 bool bpf_filters_cow; 263 u8 sqe_flags_allowed; 264 u8 sqe_flags_required; 265 /* IORING_OP_* restrictions exist */ 266 bool op_registered; 267 /* IORING_REGISTER_* restrictions exist */ 268 bool reg_registered; 269 }; 270 271 struct io_submit_link { 272 struct io_kiocb *head; 273 struct io_kiocb *last; 274 }; 275 276 struct io_submit_state { 277 /* inline/task_work completion list, under ->uring_lock */ 278 struct io_wq_work_node free_list; 279 /* batch completion logic */ 280 struct io_wq_work_list compl_reqs; 281 struct io_submit_link link; 282 283 bool plug_started; 284 bool need_plug; 285 bool cq_flush; 286 unsigned short submit_nr; 287 struct blk_plug plug; 288 }; 289 290 struct io_alloc_cache { 291 void **entries; 292 unsigned int nr_cached; 293 unsigned int max_cached; 294 unsigned int elem_size; 295 unsigned int init_clear; 296 }; 297 298 enum { 299 IO_RING_F_DRAIN_NEXT = BIT(0), 300 IO_RING_F_OP_RESTRICTED = BIT(1), 301 IO_RING_F_REG_RESTRICTED = BIT(2), 302 IO_RING_F_OFF_TIMEOUT_USED = BIT(3), 303 IO_RING_F_DRAIN_ACTIVE = BIT(4), 304 IO_RING_F_HAS_EVFD = BIT(5), 305 /* all CQEs should be posted only by the submitter task */ 306 IO_RING_F_TASK_COMPLETE = BIT(6), 307 IO_RING_F_LOCKLESS_CQ = BIT(7), 308 IO_RING_F_SYSCALL_IOPOLL = BIT(8), 309 IO_RING_F_POLL_ACTIVATED = BIT(9), 310 IO_RING_F_DRAIN_DISABLED = BIT(10), 311 IO_RING_F_COMPAT = BIT(11), 312 IO_RING_F_IOWQ_LIMITS_SET = BIT(12), 313 }; 314 315 struct iou_ctx {}; 316 317 struct io_ring_ctx { 318 /* const or read-mostly hot data */ 319 struct { 320 /* ring setup flags */ 321 unsigned int flags; 322 /* internal state flags IO_RING_F_* flags , mostly read-only */ 323 unsigned int int_flags; 324 325 struct task_struct *submitter_task; 326 struct io_rings *rings; 327 /* cache of ->restrictions.bpf_filters->filters */ 328 struct io_bpf_filter __rcu **bpf_filters; 329 struct percpu_ref refs; 330 331 clockid_t clockid; 332 enum tk_offsets clock_offset; 333 334 enum task_work_notify_mode notify_method; 335 unsigned sq_thread_idle; 336 } ____cacheline_aligned_in_smp; 337 338 /* submission data */ 339 struct { 340 struct mutex uring_lock; 341 342 /* 343 * Ring buffer of indices into array of io_uring_sqe, which is 344 * mmapped by the application using the IORING_OFF_SQES offset. 345 * 346 * This indirection could e.g. be used to assign fixed 347 * io_uring_sqe entries to operations and only submit them to 348 * the queue when needed. 349 * 350 * The kernel modifies neither the indices array nor the entries 351 * array. 352 */ 353 u32 *sq_array; 354 struct io_uring_sqe *sq_sqes; 355 unsigned cached_sq_head; 356 unsigned sq_entries; 357 358 /* 359 * Fixed resources fast path, should be accessed only under 360 * uring_lock, and updated through io_uring_register(2) 361 */ 362 atomic_t cancel_seq; 363 364 /* 365 * ->iopoll_list is protected by the ctx->uring_lock for 366 * io_uring instances that don't use IORING_SETUP_SQPOLL. 367 * For SQPOLL, only the single threaded io_sq_thread() will 368 * manipulate the list, hence no extra locking is needed there. 369 */ 370 bool poll_multi_queue; 371 struct list_head iopoll_list; 372 373 /* 374 * Consumer cursor for ->work_list, protected by ->uring_lock. 375 * Deliberately kept away from the producer side of the queue, 376 * as it's written for every popped entry, and the producer 377 * cacheline is contended enough as it is. 378 */ 379 struct llist_node *work_head; 380 381 struct io_file_table file_table; 382 struct io_rsrc_data buf_table; 383 struct io_alloc_cache node_cache; 384 struct io_alloc_cache imu_cache; 385 386 struct io_submit_state submit_state; 387 388 /* 389 * Modifications are protected by ->uring_lock and ->mmap_lock. 390 * The buffer list's io mapped region should be stable once 391 * published. 392 */ 393 struct xarray io_bl_xa; 394 395 struct io_hash_table cancel_table; 396 struct io_alloc_cache apoll_cache; 397 struct io_alloc_cache netmsg_cache; 398 struct io_alloc_cache rw_cache; 399 struct io_alloc_cache cmd_cache; 400 401 int (*loop_step)(struct iou_ctx *, 402 struct iou_loop_params *); 403 404 /* 405 * Any cancelable uring_cmd is added to this list in 406 * ->uring_cmd() by io_uring_cmd_insert_cancelable() 407 */ 408 struct hlist_head cancelable_uring_cmd; 409 /* 410 * For Hybrid IOPOLL, runtime in hybrid polling, without 411 * scheduling time 412 */ 413 u64 hybrid_poll_time; 414 } ____cacheline_aligned_in_smp; 415 416 struct { 417 /* 418 * We cache a range of free CQEs we can use, once exhausted it 419 * should go through a slower range setup, see __io_get_cqe() 420 */ 421 struct io_uring_cqe *cqe_cached; 422 struct io_uring_cqe *cqe_sentinel; 423 424 unsigned cached_cq_tail; 425 unsigned cq_entries; 426 struct io_ev_fd __rcu *io_ev_fd; 427 428 void *cq_wait_arg; 429 size_t cq_wait_size; 430 } ____cacheline_aligned_in_smp; 431 432 /* 433 * task_work and async notification delivery cacheline. Expected to 434 * regularly bounce b/w CPUs. 435 */ 436 struct { 437 struct io_rings __rcu *rings_rcu; 438 struct mpscq work_list; 439 unsigned long check_cq; 440 atomic_t cq_wait_nr; 441 atomic_t cq_timeouts; 442 struct wait_queue_head cq_wait; 443 } ____cacheline_aligned_in_smp; 444 445 /* timeouts */ 446 struct { 447 raw_spinlock_t timeout_lock; 448 struct list_head timeout_list; 449 struct list_head ltimeout_list; 450 unsigned cq_last_tm_flush; 451 } ____cacheline_aligned_in_smp; 452 453 spinlock_t completion_lock; 454 455 struct list_head cq_overflow_list; 456 457 struct hlist_head waitid_list; 458 459 #ifdef CONFIG_FUTEX 460 struct hlist_head futex_list; 461 struct io_alloc_cache futex_cache; 462 #endif 463 464 const struct cred *sq_creds; /* cred used for __io_sq_thread() */ 465 struct io_sq_data *sq_data; /* if using sq thread polling */ 466 467 struct wait_queue_head sqo_sq_wait; 468 struct list_head sqd_list; 469 470 unsigned int file_alloc_start; 471 unsigned int file_alloc_end; 472 473 /* Keep this last, we don't need it for the fast path */ 474 struct wait_queue_head poll_wq; 475 struct io_restriction restrictions; 476 477 /* Stores zcrx object pointers of type struct io_zcrx_ifq */ 478 struct xarray zcrx_ctxs; 479 480 /* Used for accounting references on pages in registered buffers */ 481 struct xarray hpage_acct; 482 483 u32 pers_next; 484 struct xarray personalities; 485 486 /* hashed buffered write serialization */ 487 struct io_wq_hash *hash_map; 488 489 /* Only used for accounting purposes */ 490 struct user_struct *user; 491 struct mm_struct *mm_account; 492 493 /* 494 * List of tctx nodes for this ctx, protected by tctx_lock. For 495 * cancelation purposes, nests under uring_lock. 496 */ 497 struct list_head tctx_list; 498 struct mutex tctx_lock; 499 500 /* ctx exit and cancelation */ 501 struct work_struct exit_work; 502 struct completion ref_comp; 503 504 /* io-wq management, e.g. thread count */ 505 u32 iowq_limits[2]; 506 507 struct callback_head poll_wq_task_work; 508 struct list_head defer_list; 509 unsigned nr_drained; 510 511 /* protected by ->completion_lock */ 512 unsigned nr_req_allocated; 513 514 #ifdef CONFIG_NET_RX_BUSY_POLL 515 struct list_head napi_list; /* track busy poll napi_id */ 516 spinlock_t napi_lock; /* napi_list lock */ 517 518 /* napi busy poll default timeout */ 519 ktime_t napi_busy_poll_dt; 520 bool napi_prefer_busy_poll; 521 u8 napi_track_mode; 522 523 DECLARE_HASHTABLE(napi_ht, 4); 524 #endif 525 526 struct io_uring_bpf_ops *bpf_ops; 527 528 /* 529 * Protection for resize vs mmap races - both the mmap and resize 530 * side will need to grab this lock, to prevent either side from 531 * being run concurrently with the other. 532 */ 533 struct mutex mmap_lock; 534 535 struct io_mapped_region sq_region; 536 struct io_mapped_region ring_region; 537 /* used for optimised request parameter and wait argument passing */ 538 struct io_mapped_region param_region; 539 }; 540 541 /* 542 * Token indicating function is called in task work context: 543 * ctx->uring_lock is held and any completions generated will be flushed. 544 * ONLY core io_uring.c should instantiate this struct. 545 */ 546 struct io_tw_state { 547 bool cancel; 548 }; 549 /* Alias to use in code that doesn't instantiate struct io_tw_state */ 550 typedef struct io_tw_state io_tw_token_t; 551 552 enum { 553 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, 554 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, 555 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, 556 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, 557 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, 558 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, 559 REQ_F_CQE_SKIP_BIT = IOSQE_CQE_SKIP_SUCCESS_BIT, 560 561 /* first byte is taken by user flags, shift it to not overlap */ 562 REQ_F_FAIL_BIT = 8, 563 REQ_F_INFLIGHT_BIT, 564 REQ_F_CUR_POS_BIT, 565 REQ_F_NOWAIT_BIT, 566 REQ_F_LINK_TIMEOUT_BIT, 567 REQ_F_NEED_CLEANUP_BIT, 568 REQ_F_POLLED_BIT, 569 REQ_F_HYBRID_IOPOLL_STATE_BIT, 570 REQ_F_BUFFER_SELECTED_BIT, 571 REQ_F_BUFFER_RING_BIT, 572 REQ_F_REISSUE_BIT, 573 REQ_F_CREDS_BIT, 574 REQ_F_REFCOUNT_BIT, 575 REQ_F_ARM_LTIMEOUT_BIT, 576 REQ_F_ASYNC_DATA_BIT, 577 REQ_F_SKIP_LINK_CQES_BIT, 578 REQ_F_SINGLE_POLL_BIT, 579 REQ_F_DOUBLE_POLL_BIT, 580 REQ_F_MULTISHOT_BIT, 581 REQ_F_APOLL_MULTISHOT_BIT, 582 REQ_F_CLEAR_POLLIN_BIT, 583 /* keep async read/write and isreg together and in order */ 584 REQ_F_SUPPORT_NOWAIT_BIT, 585 REQ_F_ISREG_BIT, 586 REQ_F_POLL_NO_LAZY_BIT, 587 REQ_F_CAN_POLL_BIT, 588 REQ_F_BL_EMPTY_BIT, 589 REQ_F_BL_NO_RECYCLE_BIT, 590 REQ_F_BUFFERS_COMMIT_BIT, 591 REQ_F_BUF_NODE_BIT, 592 REQ_F_BUF_MORE_BIT, 593 REQ_F_HAS_METADATA_BIT, 594 REQ_F_IMPORT_BUFFER_BIT, 595 REQ_F_SQE_COPIED_BIT, 596 REQ_F_IOPOLL_BIT, 597 598 /* not a real bit, just to check we're not overflowing the space */ 599 __REQ_F_LAST_BIT, 600 }; 601 602 typedef u64 __bitwise io_req_flags_t; 603 #define IO_REQ_FLAG(bitno) ((__force io_req_flags_t) BIT_ULL((bitno))) 604 605 enum { 606 /* ctx owns file */ 607 REQ_F_FIXED_FILE = IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT), 608 /* drain existing IO first */ 609 REQ_F_IO_DRAIN = IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT), 610 /* linked sqes */ 611 REQ_F_LINK = IO_REQ_FLAG(REQ_F_LINK_BIT), 612 /* doesn't sever on completion < 0 */ 613 REQ_F_HARDLINK = IO_REQ_FLAG(REQ_F_HARDLINK_BIT), 614 /* IOSQE_ASYNC */ 615 REQ_F_FORCE_ASYNC = IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT), 616 /* IOSQE_BUFFER_SELECT */ 617 REQ_F_BUFFER_SELECT = IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT), 618 /* IOSQE_CQE_SKIP_SUCCESS */ 619 REQ_F_CQE_SKIP = IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT), 620 621 /* fail rest of links */ 622 REQ_F_FAIL = IO_REQ_FLAG(REQ_F_FAIL_BIT), 623 /* on inflight list, should be cancelled and waited on exit reliably */ 624 REQ_F_INFLIGHT = IO_REQ_FLAG(REQ_F_INFLIGHT_BIT), 625 /* read/write uses file position */ 626 REQ_F_CUR_POS = IO_REQ_FLAG(REQ_F_CUR_POS_BIT), 627 /* must not punt to workers */ 628 REQ_F_NOWAIT = IO_REQ_FLAG(REQ_F_NOWAIT_BIT), 629 /* has or had linked timeout */ 630 REQ_F_LINK_TIMEOUT = IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT), 631 /* needs cleanup */ 632 REQ_F_NEED_CLEANUP = IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT), 633 /* already went through poll handler */ 634 REQ_F_POLLED = IO_REQ_FLAG(REQ_F_POLLED_BIT), 635 /* every req only blocks once in hybrid poll */ 636 REQ_F_IOPOLL_STATE = IO_REQ_FLAG(REQ_F_HYBRID_IOPOLL_STATE_BIT), 637 /* buffer already selected */ 638 REQ_F_BUFFER_SELECTED = IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT), 639 /* buffer selected from ring, needs commit */ 640 REQ_F_BUFFER_RING = IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT), 641 /* caller should reissue async */ 642 REQ_F_REISSUE = IO_REQ_FLAG(REQ_F_REISSUE_BIT), 643 /* supports async reads/writes */ 644 REQ_F_SUPPORT_NOWAIT = IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT), 645 /* regular file */ 646 REQ_F_ISREG = IO_REQ_FLAG(REQ_F_ISREG_BIT), 647 /* has creds assigned */ 648 REQ_F_CREDS = IO_REQ_FLAG(REQ_F_CREDS_BIT), 649 /* skip refcounting if not set */ 650 REQ_F_REFCOUNT = IO_REQ_FLAG(REQ_F_REFCOUNT_BIT), 651 /* there is a linked timeout that has to be armed */ 652 REQ_F_ARM_LTIMEOUT = IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT), 653 /* ->async_data allocated */ 654 REQ_F_ASYNC_DATA = IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT), 655 /* don't post CQEs while failing linked requests */ 656 REQ_F_SKIP_LINK_CQES = IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT), 657 /* single poll may be active */ 658 REQ_F_SINGLE_POLL = IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT), 659 /* double poll may active */ 660 REQ_F_DOUBLE_POLL = IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT), 661 /* request posts multiple completions, should be set at prep time */ 662 REQ_F_MULTISHOT = IO_REQ_FLAG(REQ_F_MULTISHOT_BIT), 663 /* fast poll multishot mode */ 664 REQ_F_APOLL_MULTISHOT = IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT), 665 /* recvmsg special flag, clear EPOLLIN */ 666 REQ_F_CLEAR_POLLIN = IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT), 667 /* don't use lazy poll wake for this request */ 668 REQ_F_POLL_NO_LAZY = IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT), 669 /* file is pollable */ 670 REQ_F_CAN_POLL = IO_REQ_FLAG(REQ_F_CAN_POLL_BIT), 671 /* buffer list was empty after selection of buffer */ 672 REQ_F_BL_EMPTY = IO_REQ_FLAG(REQ_F_BL_EMPTY_BIT), 673 /* don't recycle provided buffers for this request */ 674 REQ_F_BL_NO_RECYCLE = IO_REQ_FLAG(REQ_F_BL_NO_RECYCLE_BIT), 675 /* buffer ring head needs incrementing on put */ 676 REQ_F_BUFFERS_COMMIT = IO_REQ_FLAG(REQ_F_BUFFERS_COMMIT_BIT), 677 /* buf node is valid */ 678 REQ_F_BUF_NODE = IO_REQ_FLAG(REQ_F_BUF_NODE_BIT), 679 /* incremental buffer consumption, more space available */ 680 REQ_F_BUF_MORE = IO_REQ_FLAG(REQ_F_BUF_MORE_BIT), 681 /* request has read/write metadata assigned */ 682 REQ_F_HAS_METADATA = IO_REQ_FLAG(REQ_F_HAS_METADATA_BIT), 683 /* 684 * For vectored fixed buffers, resolve iovec to registered buffers. 685 * For SEND_ZC, whether to import buffers (i.e. the first issue). 686 */ 687 REQ_F_IMPORT_BUFFER = IO_REQ_FLAG(REQ_F_IMPORT_BUFFER_BIT), 688 /* ->sqe_copy() has been called, if necessary */ 689 REQ_F_SQE_COPIED = IO_REQ_FLAG(REQ_F_SQE_COPIED_BIT), 690 /* request must be iopolled to completion (set in ->issue()) */ 691 REQ_F_IOPOLL = IO_REQ_FLAG(REQ_F_IOPOLL_BIT), 692 }; 693 694 struct io_tw_req { 695 struct io_kiocb *req; 696 }; 697 698 typedef void (*io_req_tw_func_t)(struct io_tw_req tw_req, io_tw_token_t tw); 699 700 struct io_task_work { 701 struct llist_node node; 702 io_req_tw_func_t func; 703 }; 704 705 struct io_cqe { 706 __u64 user_data; 707 __s32 res; 708 /* fd initially, then cflags for completion */ 709 union { 710 __u32 flags; 711 int fd; 712 }; 713 }; 714 715 /* 716 * Each request type overlays its private data structure on top of this one. 717 * They must not exceed this one in size. 718 */ 719 struct io_cmd_data { 720 struct file *file; 721 /* each command gets 56 bytes of data */ 722 __u8 data[56]; 723 }; 724 725 static inline void io_kiocb_cmd_sz_check(size_t cmd_sz) 726 { 727 BUILD_BUG_ON(cmd_sz > sizeof(struct io_cmd_data)); 728 } 729 #define io_kiocb_to_cmd(req, cmd_type) ( \ 730 io_kiocb_cmd_sz_check(sizeof(cmd_type)) , \ 731 ((cmd_type *)&(req)->cmd) \ 732 ) 733 734 static inline struct io_kiocb *cmd_to_io_kiocb(void *ptr) 735 { 736 return ptr; 737 } 738 739 struct io_kiocb { 740 union { 741 /* 742 * NOTE! Each of the io_kiocb union members has the file pointer 743 * as the first entry in their struct definition. So you can 744 * access the file pointer through any of the sub-structs, 745 * or directly as just 'file' in this struct. 746 */ 747 struct file *file; 748 struct io_cmd_data cmd; 749 }; 750 751 u8 opcode; 752 /* polled IO has completed */ 753 u8 iopoll_completed; 754 /* 755 * Can be either a fixed buffer index, or used with provided buffers. 756 * For the latter, it points to the selected buffer ID. 757 */ 758 u16 buf_index; 759 760 /* REQ_F_* flags */ 761 io_req_flags_t flags; 762 763 struct io_cqe cqe; 764 765 struct io_ring_ctx *ctx; 766 struct io_uring_task *tctx; 767 768 union { 769 /* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */ 770 struct io_buffer *kbuf; 771 772 struct io_rsrc_node *buf_node; 773 }; 774 775 union { 776 /* used by request caches, completion batching and iopoll */ 777 struct io_wq_work_node comp_list; 778 /* cache ->apoll->events */ 779 __poll_t apoll_events; 780 }; 781 782 struct io_rsrc_node *file_node; 783 784 atomic_t refs; 785 bool cancel_seq_set; 786 787 union { 788 struct io_task_work io_task_work; 789 /* For IOPOLL setup queues, with hybrid polling */ 790 u64 iopoll_start; 791 }; 792 793 union { 794 /* 795 * for polled requests, i.e. IORING_OP_POLL_ADD and async armed 796 * poll 797 */ 798 struct hlist_node hash_node; 799 /* IOPOLL completion handling */ 800 struct list_head iopoll_node; 801 /* for private io_kiocb freeing */ 802 struct rcu_head rcu_head; 803 }; 804 /* internal polling, see IORING_FEAT_FAST_POLL */ 805 struct async_poll *apoll; 806 /* opcode allocated if it needs to store data for async defer */ 807 void *async_data; 808 /* linked requests, IFF REQ_F_HARDLINK or REQ_F_LINK are set */ 809 atomic_t poll_refs; 810 struct io_kiocb *link; 811 /* custom credentials, valid IFF REQ_F_CREDS is set */ 812 const struct cred *creds; 813 struct io_wq_work work; 814 815 struct io_big_cqe { 816 u64 extra1; 817 u64 extra2; 818 } big_cqe; 819 }; 820 821 struct io_overflow_cqe { 822 struct list_head list; 823 struct io_uring_cqe cqe; 824 }; 825 #endif 826