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 11enum { 12 /* 13 * A hint to not wake right away but delay until there are enough of 14 * tw's queued to match the number of CQEs the task is waiting for. 15 * 16 * Must not be used wirh requests generating more than one CQE. 17 * It's also ignored unless IORING_SETUP_DEFER_TASKRUN is set. 18 */ 19 IOU_F_TWQ_LAZY_WAKE = 1, 20}; 21 22enum io_uring_cmd_flags { 23 IO_URING_F_COMPLETE_DEFER = 1, 24 IO_URING_F_UNLOCKED = 2, 25 /* the request is executed from poll, it should not be freed */ 26 IO_URING_F_MULTISHOT = 4, 27 /* executed by io-wq */ 28 IO_URING_F_IOWQ = 8, 29 /* int's last bit, sign checks are usually faster than a bit test */ 30 IO_URING_F_NONBLOCK = INT_MIN, 31 32 /* ctx state flags, for URING_CMD */ 33 IO_URING_F_SQE128 = (1 << 8), 34 IO_URING_F_CQE32 = (1 << 9), 35 IO_URING_F_IOPOLL = (1 << 10), 36 37 /* set when uring wants to cancel a previously issued command */ 38 IO_URING_F_CANCEL = (1 << 11), 39 IO_URING_F_COMPAT = (1 << 12), 40}; 41 42struct io_wq_work_node { 43 struct io_wq_work_node *next; 44}; 45 46struct io_wq_work_list { 47 struct io_wq_work_node *first; 48 struct io_wq_work_node *last; 49}; 50 51struct io_wq_work { 52 struct io_wq_work_node list; 53 unsigned flags; 54 /* place it here instead of io_kiocb as it fills padding and saves 4B */ 55 int cancel_seq; 56}; 57 58struct io_fixed_file { 59 /* file * with additional FFS_* flags */ 60 unsigned long file_ptr; 61}; 62 63struct io_file_table { 64 struct io_fixed_file *files; 65 unsigned long *bitmap; 66 unsigned int alloc_hint; 67}; 68 69struct io_hash_bucket { 70 spinlock_t lock; 71 struct hlist_head list; 72} ____cacheline_aligned_in_smp; 73 74struct io_hash_table { 75 struct io_hash_bucket *hbs; 76 unsigned hash_bits; 77}; 78 79/* 80 * Arbitrary limit, can be raised if need be 81 */ 82#define IO_RINGFD_REG_MAX 16 83 84struct io_uring_task { 85 /* submission side */ 86 int cached_refs; 87 const struct io_ring_ctx *last; 88 struct io_wq *io_wq; 89 struct file *registered_rings[IO_RINGFD_REG_MAX]; 90 91 struct xarray xa; 92 struct wait_queue_head wait; 93 atomic_t in_cancel; 94 atomic_t inflight_tracked; 95 struct percpu_counter inflight; 96 97 struct { /* task_work */ 98 struct llist_head task_list; 99 struct callback_head task_work; 100 } ____cacheline_aligned_in_smp; 101}; 102 103struct io_uring { 104 u32 head; 105 u32 tail; 106}; 107 108/* 109 * This data is shared with the application through the mmap at offsets 110 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING. 111 * 112 * The offsets to the member fields are published through struct 113 * io_sqring_offsets when calling io_uring_setup. 114 */ 115struct io_rings { 116 /* 117 * Head and tail offsets into the ring; the offsets need to be 118 * masked to get valid indices. 119 * 120 * The kernel controls head of the sq ring and the tail of the cq ring, 121 * and the application controls tail of the sq ring and the head of the 122 * cq ring. 123 */ 124 struct io_uring sq, cq; 125 /* 126 * Bitmasks to apply to head and tail offsets (constant, equals 127 * ring_entries - 1) 128 */ 129 u32 sq_ring_mask, cq_ring_mask; 130 /* Ring sizes (constant, power of 2) */ 131 u32 sq_ring_entries, cq_ring_entries; 132 /* 133 * Number of invalid entries dropped by the kernel due to 134 * invalid index stored in array 135 * 136 * Written by the kernel, shouldn't be modified by the 137 * application (i.e. get number of "new events" by comparing to 138 * cached value). 139 * 140 * After a new SQ head value was read by the application this 141 * counter includes all submissions that were dropped reaching 142 * the new SQ head (and possibly more). 143 */ 144 u32 sq_dropped; 145 /* 146 * Runtime SQ flags 147 * 148 * Written by the kernel, shouldn't be modified by the 149 * application. 150 * 151 * The application needs a full memory barrier before checking 152 * for IORING_SQ_NEED_WAKEUP after updating the sq tail. 153 */ 154 atomic_t sq_flags; 155 /* 156 * Runtime CQ flags 157 * 158 * Written by the application, shouldn't be modified by the 159 * kernel. 160 */ 161 u32 cq_flags; 162 /* 163 * Number of completion events lost because the queue was full; 164 * this should be avoided by the application by making sure 165 * there are not more requests pending than there is space in 166 * the completion queue. 167 * 168 * Written by the kernel, shouldn't be modified by the 169 * application (i.e. get number of "new events" by comparing to 170 * cached value). 171 * 172 * As completion events come in out of order this counter is not 173 * ordered with any other data. 174 */ 175 u32 cq_overflow; 176 /* 177 * Ring buffer of completion events. 178 * 179 * The kernel writes completion events fresh every time they are 180 * produced, so the application is allowed to modify pending 181 * entries. 182 */ 183 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp; 184}; 185 186struct io_restriction { 187 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST); 188 DECLARE_BITMAP(sqe_op, IORING_OP_LAST); 189 u8 sqe_flags_allowed; 190 u8 sqe_flags_required; 191 bool registered; 192}; 193 194struct io_submit_link { 195 struct io_kiocb *head; 196 struct io_kiocb *last; 197}; 198 199struct io_submit_state { 200 /* inline/task_work completion list, under ->uring_lock */ 201 struct io_wq_work_node free_list; 202 /* batch completion logic */ 203 struct io_wq_work_list compl_reqs; 204 struct io_submit_link link; 205 206 bool plug_started; 207 bool need_plug; 208 unsigned short submit_nr; 209 unsigned int cqes_count; 210 struct blk_plug plug; 211}; 212 213struct io_ev_fd { 214 struct eventfd_ctx *cq_ev_fd; 215 unsigned int eventfd_async: 1; 216 struct rcu_head rcu; 217 atomic_t refs; 218 atomic_t ops; 219}; 220 221struct io_alloc_cache { 222 struct io_wq_work_node list; 223 unsigned int nr_cached; 224 unsigned int max_cached; 225 size_t elem_size; 226}; 227 228struct io_ring_ctx { 229 /* const or read-mostly hot data */ 230 struct { 231 unsigned int flags; 232 unsigned int drain_next: 1; 233 unsigned int restricted: 1; 234 unsigned int off_timeout_used: 1; 235 unsigned int drain_active: 1; 236 unsigned int has_evfd: 1; 237 /* all CQEs should be posted only by the submitter task */ 238 unsigned int task_complete: 1; 239 unsigned int lockless_cq: 1; 240 unsigned int syscall_iopoll: 1; 241 unsigned int poll_activated: 1; 242 unsigned int drain_disabled: 1; 243 unsigned int compat: 1; 244 unsigned int iowq_limits_set : 1; 245 246 struct task_struct *submitter_task; 247 struct io_rings *rings; 248 struct percpu_ref refs; 249 250 enum task_work_notify_mode notify_method; 251 unsigned sq_thread_idle; 252 } ____cacheline_aligned_in_smp; 253 254 /* submission data */ 255 struct { 256 struct mutex uring_lock; 257 258 /* 259 * Ring buffer of indices into array of io_uring_sqe, which is 260 * mmapped by the application using the IORING_OFF_SQES offset. 261 * 262 * This indirection could e.g. be used to assign fixed 263 * io_uring_sqe entries to operations and only submit them to 264 * the queue when needed. 265 * 266 * The kernel modifies neither the indices array nor the entries 267 * array. 268 */ 269 u32 *sq_array; 270 struct io_uring_sqe *sq_sqes; 271 unsigned cached_sq_head; 272 unsigned sq_entries; 273 274 /* 275 * Fixed resources fast path, should be accessed only under 276 * uring_lock, and updated through io_uring_register(2) 277 */ 278 struct io_rsrc_node *rsrc_node; 279 atomic_t cancel_seq; 280 281 /* 282 * ->iopoll_list is protected by the ctx->uring_lock for 283 * io_uring instances that don't use IORING_SETUP_SQPOLL. 284 * For SQPOLL, only the single threaded io_sq_thread() will 285 * manipulate the list, hence no extra locking is needed there. 286 */ 287 bool poll_multi_queue; 288 struct io_wq_work_list iopoll_list; 289 290 struct io_file_table file_table; 291 struct io_mapped_ubuf **user_bufs; 292 unsigned nr_user_files; 293 unsigned nr_user_bufs; 294 295 struct io_submit_state submit_state; 296 297 struct io_buffer_list *io_bl; 298 struct xarray io_bl_xa; 299 300 struct io_hash_table cancel_table_locked; 301 struct io_alloc_cache apoll_cache; 302 struct io_alloc_cache netmsg_cache; 303 304 /* 305 * Any cancelable uring_cmd is added to this list in 306 * ->uring_cmd() by io_uring_cmd_insert_cancelable() 307 */ 308 struct hlist_head cancelable_uring_cmd; 309 } ____cacheline_aligned_in_smp; 310 311 struct { 312 /* 313 * We cache a range of free CQEs we can use, once exhausted it 314 * should go through a slower range setup, see __io_get_cqe() 315 */ 316 struct io_uring_cqe *cqe_cached; 317 struct io_uring_cqe *cqe_sentinel; 318 319 unsigned cached_cq_tail; 320 unsigned cq_entries; 321 struct io_ev_fd __rcu *io_ev_fd; 322 unsigned cq_extra; 323 } ____cacheline_aligned_in_smp; 324 325 /* 326 * task_work and async notification delivery cacheline. Expected to 327 * regularly bounce b/w CPUs. 328 */ 329 struct { 330 struct llist_head work_llist; 331 unsigned long check_cq; 332 atomic_t cq_wait_nr; 333 atomic_t cq_timeouts; 334 struct wait_queue_head cq_wait; 335 } ____cacheline_aligned_in_smp; 336 337 /* timeouts */ 338 struct { 339 spinlock_t timeout_lock; 340 struct list_head timeout_list; 341 struct list_head ltimeout_list; 342 unsigned cq_last_tm_flush; 343 } ____cacheline_aligned_in_smp; 344 345 struct io_uring_cqe completion_cqes[16]; 346 347 spinlock_t completion_lock; 348 349 /* IRQ completion list, under ->completion_lock */ 350 unsigned int locked_free_nr; 351 struct io_wq_work_list locked_free_list; 352 353 struct list_head io_buffers_comp; 354 struct list_head cq_overflow_list; 355 struct io_hash_table cancel_table; 356 357 struct hlist_head waitid_list; 358 359#ifdef CONFIG_FUTEX 360 struct hlist_head futex_list; 361 struct io_alloc_cache futex_cache; 362#endif 363 364 const struct cred *sq_creds; /* cred used for __io_sq_thread() */ 365 struct io_sq_data *sq_data; /* if using sq thread polling */ 366 367 struct wait_queue_head sqo_sq_wait; 368 struct list_head sqd_list; 369 370 unsigned int file_alloc_start; 371 unsigned int file_alloc_end; 372 373 struct list_head io_buffers_cache; 374 375 /* deferred free list, protected by ->uring_lock */ 376 struct hlist_head io_buf_list; 377 378 /* Keep this last, we don't need it for the fast path */ 379 struct wait_queue_head poll_wq; 380 struct io_restriction restrictions; 381 382 /* slow path rsrc auxilary data, used by update/register */ 383 struct io_mapped_ubuf *dummy_ubuf; 384 struct io_rsrc_data *file_data; 385 struct io_rsrc_data *buf_data; 386 387 /* protected by ->uring_lock */ 388 struct list_head rsrc_ref_list; 389 struct io_alloc_cache rsrc_node_cache; 390 struct wait_queue_head rsrc_quiesce_wq; 391 unsigned rsrc_quiesce; 392 393 u32 pers_next; 394 struct xarray personalities; 395 396 /* hashed buffered write serialization */ 397 struct io_wq_hash *hash_map; 398 399 /* Only used for accounting purposes */ 400 struct user_struct *user; 401 struct mm_struct *mm_account; 402 403 /* ctx exit and cancelation */ 404 struct llist_head fallback_llist; 405 struct delayed_work fallback_work; 406 struct work_struct exit_work; 407 struct list_head tctx_list; 408 struct completion ref_comp; 409 410 /* io-wq management, e.g. thread count */ 411 u32 iowq_limits[2]; 412 413 struct callback_head poll_wq_task_work; 414 struct list_head defer_list; 415 416#ifdef CONFIG_NET_RX_BUSY_POLL 417 struct list_head napi_list; /* track busy poll napi_id */ 418 spinlock_t napi_lock; /* napi_list lock */ 419 420 /* napi busy poll default timeout */ 421 unsigned int napi_busy_poll_to; 422 bool napi_prefer_busy_poll; 423 bool napi_enabled; 424 425 DECLARE_HASHTABLE(napi_ht, 4); 426#endif 427 428 /* protected by ->completion_lock */ 429 unsigned evfd_last_cq_tail; 430 431 /* 432 * If IORING_SETUP_NO_MMAP is used, then the below holds 433 * the gup'ed pages for the two rings, and the sqes. 434 */ 435 unsigned short n_ring_pages; 436 unsigned short n_sqe_pages; 437 struct page **ring_pages; 438 struct page **sqe_pages; 439}; 440 441struct io_tw_state { 442 /* ->uring_lock is taken, callbacks can use io_tw_lock to lock it */ 443 bool locked; 444}; 445 446enum { 447 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT, 448 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT, 449 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT, 450 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT, 451 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT, 452 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT, 453 REQ_F_CQE_SKIP_BIT = IOSQE_CQE_SKIP_SUCCESS_BIT, 454 455 /* first byte is taken by user flags, shift it to not overlap */ 456 REQ_F_FAIL_BIT = 8, 457 REQ_F_INFLIGHT_BIT, 458 REQ_F_CUR_POS_BIT, 459 REQ_F_NOWAIT_BIT, 460 REQ_F_LINK_TIMEOUT_BIT, 461 REQ_F_NEED_CLEANUP_BIT, 462 REQ_F_POLLED_BIT, 463 REQ_F_BUFFER_SELECTED_BIT, 464 REQ_F_BUFFER_RING_BIT, 465 REQ_F_REISSUE_BIT, 466 REQ_F_CREDS_BIT, 467 REQ_F_REFCOUNT_BIT, 468 REQ_F_ARM_LTIMEOUT_BIT, 469 REQ_F_ASYNC_DATA_BIT, 470 REQ_F_SKIP_LINK_CQES_BIT, 471 REQ_F_SINGLE_POLL_BIT, 472 REQ_F_DOUBLE_POLL_BIT, 473 REQ_F_APOLL_MULTISHOT_BIT, 474 REQ_F_CLEAR_POLLIN_BIT, 475 REQ_F_HASH_LOCKED_BIT, 476 /* keep async read/write and isreg together and in order */ 477 REQ_F_SUPPORT_NOWAIT_BIT, 478 REQ_F_ISREG_BIT, 479 REQ_F_POLL_NO_LAZY_BIT, 480 REQ_F_CANCEL_SEQ_BIT, 481 REQ_F_CAN_POLL_BIT, 482 REQ_F_BL_EMPTY_BIT, 483 REQ_F_BL_NO_RECYCLE_BIT, 484 485 /* not a real bit, just to check we're not overflowing the space */ 486 __REQ_F_LAST_BIT, 487}; 488 489typedef u64 __bitwise io_req_flags_t; 490#define IO_REQ_FLAG(bitno) ((__force io_req_flags_t) BIT_ULL((bitno))) 491 492enum { 493 /* ctx owns file */ 494 REQ_F_FIXED_FILE = IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT), 495 /* drain existing IO first */ 496 REQ_F_IO_DRAIN = IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT), 497 /* linked sqes */ 498 REQ_F_LINK = IO_REQ_FLAG(REQ_F_LINK_BIT), 499 /* doesn't sever on completion < 0 */ 500 REQ_F_HARDLINK = IO_REQ_FLAG(REQ_F_HARDLINK_BIT), 501 /* IOSQE_ASYNC */ 502 REQ_F_FORCE_ASYNC = IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT), 503 /* IOSQE_BUFFER_SELECT */ 504 REQ_F_BUFFER_SELECT = IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT), 505 /* IOSQE_CQE_SKIP_SUCCESS */ 506 REQ_F_CQE_SKIP = IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT), 507 508 /* fail rest of links */ 509 REQ_F_FAIL = IO_REQ_FLAG(REQ_F_FAIL_BIT), 510 /* on inflight list, should be cancelled and waited on exit reliably */ 511 REQ_F_INFLIGHT = IO_REQ_FLAG(REQ_F_INFLIGHT_BIT), 512 /* read/write uses file position */ 513 REQ_F_CUR_POS = IO_REQ_FLAG(REQ_F_CUR_POS_BIT), 514 /* must not punt to workers */ 515 REQ_F_NOWAIT = IO_REQ_FLAG(REQ_F_NOWAIT_BIT), 516 /* has or had linked timeout */ 517 REQ_F_LINK_TIMEOUT = IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT), 518 /* needs cleanup */ 519 REQ_F_NEED_CLEANUP = IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT), 520 /* already went through poll handler */ 521 REQ_F_POLLED = IO_REQ_FLAG(REQ_F_POLLED_BIT), 522 /* buffer already selected */ 523 REQ_F_BUFFER_SELECTED = IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT), 524 /* buffer selected from ring, needs commit */ 525 REQ_F_BUFFER_RING = IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT), 526 /* caller should reissue async */ 527 REQ_F_REISSUE = IO_REQ_FLAG(REQ_F_REISSUE_BIT), 528 /* supports async reads/writes */ 529 REQ_F_SUPPORT_NOWAIT = IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT), 530 /* regular file */ 531 REQ_F_ISREG = IO_REQ_FLAG(REQ_F_ISREG_BIT), 532 /* has creds assigned */ 533 REQ_F_CREDS = IO_REQ_FLAG(REQ_F_CREDS_BIT), 534 /* skip refcounting if not set */ 535 REQ_F_REFCOUNT = IO_REQ_FLAG(REQ_F_REFCOUNT_BIT), 536 /* there is a linked timeout that has to be armed */ 537 REQ_F_ARM_LTIMEOUT = IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT), 538 /* ->async_data allocated */ 539 REQ_F_ASYNC_DATA = IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT), 540 /* don't post CQEs while failing linked requests */ 541 REQ_F_SKIP_LINK_CQES = IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT), 542 /* single poll may be active */ 543 REQ_F_SINGLE_POLL = IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT), 544 /* double poll may active */ 545 REQ_F_DOUBLE_POLL = IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT), 546 /* fast poll multishot mode */ 547 REQ_F_APOLL_MULTISHOT = IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT), 548 /* recvmsg special flag, clear EPOLLIN */ 549 REQ_F_CLEAR_POLLIN = IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT), 550 /* hashed into ->cancel_hash_locked, protected by ->uring_lock */ 551 REQ_F_HASH_LOCKED = IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT), 552 /* don't use lazy poll wake for this request */ 553 REQ_F_POLL_NO_LAZY = IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT), 554 /* cancel sequence is set and valid */ 555 REQ_F_CANCEL_SEQ = IO_REQ_FLAG(REQ_F_CANCEL_SEQ_BIT), 556 /* file is pollable */ 557 REQ_F_CAN_POLL = IO_REQ_FLAG(REQ_F_CAN_POLL_BIT), 558 /* buffer list was empty after selection of buffer */ 559 REQ_F_BL_EMPTY = IO_REQ_FLAG(REQ_F_BL_EMPTY_BIT), 560 /* don't recycle provided buffers for this request */ 561 REQ_F_BL_NO_RECYCLE = IO_REQ_FLAG(REQ_F_BL_NO_RECYCLE_BIT), 562}; 563 564typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts); 565 566struct io_task_work { 567 struct llist_node node; 568 io_req_tw_func_t func; 569}; 570 571struct io_cqe { 572 __u64 user_data; 573 __s32 res; 574 /* fd initially, then cflags for completion */ 575 union { 576 __u32 flags; 577 int fd; 578 }; 579}; 580 581/* 582 * Each request type overlays its private data structure on top of this one. 583 * They must not exceed this one in size. 584 */ 585struct io_cmd_data { 586 struct file *file; 587 /* each command gets 56 bytes of data */ 588 __u8 data[56]; 589}; 590 591static inline void io_kiocb_cmd_sz_check(size_t cmd_sz) 592{ 593 BUILD_BUG_ON(cmd_sz > sizeof(struct io_cmd_data)); 594} 595#define io_kiocb_to_cmd(req, cmd_type) ( \ 596 io_kiocb_cmd_sz_check(sizeof(cmd_type)) , \ 597 ((cmd_type *)&(req)->cmd) \ 598) 599#define cmd_to_io_kiocb(ptr) ((struct io_kiocb *) ptr) 600 601struct io_kiocb { 602 union { 603 /* 604 * NOTE! Each of the io_kiocb union members has the file pointer 605 * as the first entry in their struct definition. So you can 606 * access the file pointer through any of the sub-structs, 607 * or directly as just 'file' in this struct. 608 */ 609 struct file *file; 610 struct io_cmd_data cmd; 611 }; 612 613 u8 opcode; 614 /* polled IO has completed */ 615 u8 iopoll_completed; 616 /* 617 * Can be either a fixed buffer index, or used with provided buffers. 618 * For the latter, before issue it points to the buffer group ID, 619 * and after selection it points to the buffer ID itself. 620 */ 621 u16 buf_index; 622 623 unsigned nr_tw; 624 625 /* REQ_F_* flags */ 626 io_req_flags_t flags; 627 628 struct io_cqe cqe; 629 630 struct io_ring_ctx *ctx; 631 struct task_struct *task; 632 633 union { 634 /* store used ubuf, so we can prevent reloading */ 635 struct io_mapped_ubuf *imu; 636 637 /* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */ 638 struct io_buffer *kbuf; 639 640 /* 641 * stores buffer ID for ring provided buffers, valid IFF 642 * REQ_F_BUFFER_RING is set. 643 */ 644 struct io_buffer_list *buf_list; 645 }; 646 647 union { 648 /* used by request caches, completion batching and iopoll */ 649 struct io_wq_work_node comp_list; 650 /* cache ->apoll->events */ 651 __poll_t apoll_events; 652 }; 653 654 struct io_rsrc_node *rsrc_node; 655 656 atomic_t refs; 657 atomic_t poll_refs; 658 struct io_task_work io_task_work; 659 /* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */ 660 struct hlist_node hash_node; 661 /* internal polling, see IORING_FEAT_FAST_POLL */ 662 struct async_poll *apoll; 663 /* opcode allocated if it needs to store data for async defer */ 664 void *async_data; 665 /* linked requests, IFF REQ_F_HARDLINK or REQ_F_LINK are set */ 666 struct io_kiocb *link; 667 /* custom credentials, valid IFF REQ_F_CREDS is set */ 668 const struct cred *creds; 669 struct io_wq_work work; 670 671 struct { 672 u64 extra1; 673 u64 extra2; 674 } big_cqe; 675}; 676 677struct io_overflow_cqe { 678 struct list_head list; 679 struct io_uring_cqe cqe; 680}; 681 682#endif 683