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 enum {
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 with 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
22 enum 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 IO_URING_F_TASK_DEAD = (1 << 13),
41 };
42
43 struct io_wq_work_node {
44 struct io_wq_work_node *next;
45 };
46
47 struct io_wq_work_list {
48 struct io_wq_work_node *first;
49 struct io_wq_work_node *last;
50 };
51
52 struct io_wq_work {
53 struct io_wq_work_node list;
54 atomic_t flags;
55 /* place it here instead of io_kiocb as it fills padding and saves 4B */
56 int cancel_seq;
57 };
58
59 struct io_rsrc_data {
60 unsigned int nr;
61 struct io_rsrc_node **nodes;
62 };
63
64 struct io_file_table {
65 struct io_rsrc_data data;
66 unsigned long *bitmap;
67 unsigned int alloc_hint;
68 };
69
70 struct io_hash_bucket {
71 struct hlist_head list;
72 } ____cacheline_aligned_in_smp;
73
74 struct io_hash_table {
75 struct io_hash_bucket *hbs;
76 unsigned hash_bits;
77 };
78
79 struct io_mapped_region {
80 struct page **pages;
81 void *vmap_ptr;
82 size_t nr_pages;
83 };
84
85 /*
86 * Arbitrary limit, can be raised if need be
87 */
88 #define IO_RINGFD_REG_MAX 16
89
90 struct io_uring_task {
91 /* submission side */
92 int cached_refs;
93 const struct io_ring_ctx *last;
94 struct task_struct *task;
95 struct io_wq *io_wq;
96 struct file *registered_rings[IO_RINGFD_REG_MAX];
97
98 struct xarray xa;
99 struct wait_queue_head wait;
100 atomic_t in_cancel;
101 atomic_t inflight_tracked;
102 struct percpu_counter inflight;
103
104 struct { /* task_work */
105 struct llist_head task_list;
106 struct callback_head task_work;
107 } ____cacheline_aligned_in_smp;
108 };
109
110 struct io_uring {
111 u32 head;
112 u32 tail;
113 };
114
115 /*
116 * This data is shared with the application through the mmap at offsets
117 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
118 *
119 * The offsets to the member fields are published through struct
120 * io_sqring_offsets when calling io_uring_setup.
121 */
122 struct io_rings {
123 /*
124 * Head and tail offsets into the ring; the offsets need to be
125 * masked to get valid indices.
126 *
127 * The kernel controls head of the sq ring and the tail of the cq ring,
128 * and the application controls tail of the sq ring and the head of the
129 * cq ring.
130 */
131 struct io_uring sq, cq;
132 /*
133 * Bitmasks to apply to head and tail offsets (constant, equals
134 * ring_entries - 1)
135 */
136 u32 sq_ring_mask, cq_ring_mask;
137 /* Ring sizes (constant, power of 2) */
138 u32 sq_ring_entries, cq_ring_entries;
139 /*
140 * Number of invalid entries dropped by the kernel due to
141 * invalid index stored in array
142 *
143 * Written by the kernel, shouldn't be modified by the
144 * application (i.e. get number of "new events" by comparing to
145 * cached value).
146 *
147 * After a new SQ head value was read by the application this
148 * counter includes all submissions that were dropped reaching
149 * the new SQ head (and possibly more).
150 */
151 u32 sq_dropped;
152 /*
153 * Runtime SQ flags
154 *
155 * Written by the kernel, shouldn't be modified by the
156 * application.
157 *
158 * The application needs a full memory barrier before checking
159 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
160 */
161 atomic_t sq_flags;
162 /*
163 * Runtime CQ flags
164 *
165 * Written by the application, shouldn't be modified by the
166 * kernel.
167 */
168 u32 cq_flags;
169 /*
170 * Number of completion events lost because the queue was full;
171 * this should be avoided by the application by making sure
172 * there are not more requests pending than there is space in
173 * the completion queue.
174 *
175 * Written by the kernel, shouldn't be modified by the
176 * application (i.e. get number of "new events" by comparing to
177 * cached value).
178 *
179 * As completion events come in out of order this counter is not
180 * ordered with any other data.
181 */
182 u32 cq_overflow;
183 /*
184 * Ring buffer of completion events.
185 *
186 * The kernel writes completion events fresh every time they are
187 * produced, so the application is allowed to modify pending
188 * entries.
189 */
190 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
191 };
192
193 struct io_restriction {
194 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
195 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
196 u8 sqe_flags_allowed;
197 u8 sqe_flags_required;
198 bool registered;
199 };
200
201 struct io_submit_link {
202 struct io_kiocb *head;
203 struct io_kiocb *last;
204 };
205
206 struct io_submit_state {
207 /* inline/task_work completion list, under ->uring_lock */
208 struct io_wq_work_node free_list;
209 /* batch completion logic */
210 struct io_wq_work_list compl_reqs;
211 struct io_submit_link link;
212
213 bool plug_started;
214 bool need_plug;
215 bool cq_flush;
216 unsigned short submit_nr;
217 struct blk_plug plug;
218 };
219
220 struct io_alloc_cache {
221 void **entries;
222 unsigned int nr_cached;
223 unsigned int max_cached;
224 size_t elem_size;
225 };
226
227 struct io_ring_ctx {
228 /* const or read-mostly hot data */
229 struct {
230 unsigned int flags;
231 unsigned int drain_next: 1;
232 unsigned int restricted: 1;
233 unsigned int off_timeout_used: 1;
234 unsigned int drain_active: 1;
235 unsigned int has_evfd: 1;
236 /* all CQEs should be posted only by the submitter task */
237 unsigned int task_complete: 1;
238 unsigned int lockless_cq: 1;
239 unsigned int syscall_iopoll: 1;
240 unsigned int poll_activated: 1;
241 unsigned int drain_disabled: 1;
242 unsigned int compat: 1;
243 unsigned int iowq_limits_set : 1;
244
245 struct task_struct *submitter_task;
246 struct io_rings *rings;
247 struct percpu_ref refs;
248
249 clockid_t clockid;
250 enum tk_offsets clock_offset;
251
252 enum task_work_notify_mode notify_method;
253 unsigned sq_thread_idle;
254 } ____cacheline_aligned_in_smp;
255
256 /* submission data */
257 struct {
258 struct mutex uring_lock;
259
260 /*
261 * Ring buffer of indices into array of io_uring_sqe, which is
262 * mmapped by the application using the IORING_OFF_SQES offset.
263 *
264 * This indirection could e.g. be used to assign fixed
265 * io_uring_sqe entries to operations and only submit them to
266 * the queue when needed.
267 *
268 * The kernel modifies neither the indices array nor the entries
269 * array.
270 */
271 u32 *sq_array;
272 struct io_uring_sqe *sq_sqes;
273 unsigned cached_sq_head;
274 unsigned sq_entries;
275
276 /*
277 * Fixed resources fast path, should be accessed only under
278 * uring_lock, and updated through io_uring_register(2)
279 */
280 atomic_t cancel_seq;
281
282 /*
283 * ->iopoll_list is protected by the ctx->uring_lock for
284 * io_uring instances that don't use IORING_SETUP_SQPOLL.
285 * For SQPOLL, only the single threaded io_sq_thread() will
286 * manipulate the list, hence no extra locking is needed there.
287 */
288 bool poll_multi_queue;
289 struct io_wq_work_list iopoll_list;
290
291 struct io_file_table file_table;
292 struct io_rsrc_data buf_table;
293
294 struct io_submit_state submit_state;
295
296 struct xarray io_bl_xa;
297
298 struct io_hash_table cancel_table;
299 struct io_alloc_cache apoll_cache;
300 struct io_alloc_cache netmsg_cache;
301 struct io_alloc_cache rw_cache;
302 struct io_alloc_cache uring_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 /*
310 * For Hybrid IOPOLL, runtime in hybrid polling, without
311 * scheduling time
312 */
313 u64 hybrid_poll_time;
314 } ____cacheline_aligned_in_smp;
315
316 struct {
317 /*
318 * We cache a range of free CQEs we can use, once exhausted it
319 * should go through a slower range setup, see __io_get_cqe()
320 */
321 struct io_uring_cqe *cqe_cached;
322 struct io_uring_cqe *cqe_sentinel;
323
324 unsigned cached_cq_tail;
325 unsigned cq_entries;
326 struct io_ev_fd __rcu *io_ev_fd;
327 unsigned cq_extra;
328
329 void *cq_wait_arg;
330 size_t cq_wait_size;
331 } ____cacheline_aligned_in_smp;
332
333 /*
334 * task_work and async notification delivery cacheline. Expected to
335 * regularly bounce b/w CPUs.
336 */
337 struct {
338 struct llist_head work_llist;
339 struct llist_head retry_llist;
340 unsigned long check_cq;
341 atomic_t cq_wait_nr;
342 atomic_t cq_timeouts;
343 struct wait_queue_head cq_wait;
344 } ____cacheline_aligned_in_smp;
345
346 /* timeouts */
347 struct {
348 raw_spinlock_t timeout_lock;
349 struct list_head timeout_list;
350 struct list_head ltimeout_list;
351 unsigned cq_last_tm_flush;
352 } ____cacheline_aligned_in_smp;
353
354 spinlock_t completion_lock;
355
356 struct list_head io_buffers_comp;
357 struct list_head cq_overflow_list;
358
359 struct hlist_head waitid_list;
360
361 #ifdef CONFIG_FUTEX
362 struct hlist_head futex_list;
363 struct io_alloc_cache futex_cache;
364 #endif
365
366 const struct cred *sq_creds; /* cred used for __io_sq_thread() */
367 struct io_sq_data *sq_data; /* if using sq thread polling */
368
369 struct wait_queue_head sqo_sq_wait;
370 struct list_head sqd_list;
371
372 unsigned int file_alloc_start;
373 unsigned int file_alloc_end;
374
375 struct list_head io_buffers_cache;
376
377 /* Keep this last, we don't need it for the fast path */
378 struct wait_queue_head poll_wq;
379 struct io_restriction restrictions;
380
381 u32 pers_next;
382 struct xarray personalities;
383
384 /* hashed buffered write serialization */
385 struct io_wq_hash *hash_map;
386
387 /* Only used for accounting purposes */
388 struct user_struct *user;
389 struct mm_struct *mm_account;
390
391 /* ctx exit and cancelation */
392 struct llist_head fallback_llist;
393 struct delayed_work fallback_work;
394 struct work_struct exit_work;
395 struct list_head tctx_list;
396 struct completion ref_comp;
397
398 /* io-wq management, e.g. thread count */
399 u32 iowq_limits[2];
400
401 struct callback_head poll_wq_task_work;
402 struct list_head defer_list;
403
404 struct io_alloc_cache msg_cache;
405 spinlock_t msg_lock;
406
407 #ifdef CONFIG_NET_RX_BUSY_POLL
408 struct list_head napi_list; /* track busy poll napi_id */
409 spinlock_t napi_lock; /* napi_list lock */
410
411 /* napi busy poll default timeout */
412 ktime_t napi_busy_poll_dt;
413 bool napi_prefer_busy_poll;
414 u8 napi_track_mode;
415
416 DECLARE_HASHTABLE(napi_ht, 4);
417 #endif
418
419 /* protected by ->completion_lock */
420 unsigned evfd_last_cq_tail;
421
422 /*
423 * Protection for resize vs mmap races - both the mmap and resize
424 * side will need to grab this lock, to prevent either side from
425 * being run concurrently with the other.
426 */
427 struct mutex resize_lock;
428
429 /*
430 * If IORING_SETUP_NO_MMAP is used, then the below holds
431 * the gup'ed pages for the two rings, and the sqes.
432 */
433 unsigned short n_ring_pages;
434 unsigned short n_sqe_pages;
435 struct page **ring_pages;
436 struct page **sqe_pages;
437
438 /* used for optimised request parameter and wait argument passing */
439 struct io_mapped_region param_region;
440 };
441
442 struct io_tw_state {
443 };
444
445 enum {
446 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
447 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
448 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
449 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
450 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
451 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
452 REQ_F_CQE_SKIP_BIT = IOSQE_CQE_SKIP_SUCCESS_BIT,
453
454 /* first byte is taken by user flags, shift it to not overlap */
455 REQ_F_FAIL_BIT = 8,
456 REQ_F_INFLIGHT_BIT,
457 REQ_F_CUR_POS_BIT,
458 REQ_F_NOWAIT_BIT,
459 REQ_F_LINK_TIMEOUT_BIT,
460 REQ_F_NEED_CLEANUP_BIT,
461 REQ_F_POLLED_BIT,
462 REQ_F_HYBRID_IOPOLL_STATE_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 /* keep async read/write and isreg together and in order */
476 REQ_F_SUPPORT_NOWAIT_BIT,
477 REQ_F_ISREG_BIT,
478 REQ_F_POLL_NO_LAZY_BIT,
479 REQ_F_CAN_POLL_BIT,
480 REQ_F_BL_EMPTY_BIT,
481 REQ_F_BL_NO_RECYCLE_BIT,
482 REQ_F_BUFFERS_COMMIT_BIT,
483 REQ_F_BUF_NODE_BIT,
484
485 /* not a real bit, just to check we're not overflowing the space */
486 __REQ_F_LAST_BIT,
487 };
488
489 typedef u64 __bitwise io_req_flags_t;
490 #define IO_REQ_FLAG(bitno) ((__force io_req_flags_t) BIT_ULL((bitno)))
491
492 enum {
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 /* every req only blocks once in hybrid poll */
523 REQ_F_IOPOLL_STATE = IO_REQ_FLAG(REQ_F_HYBRID_IOPOLL_STATE_BIT),
524 /* buffer already selected */
525 REQ_F_BUFFER_SELECTED = IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
526 /* buffer selected from ring, needs commit */
527 REQ_F_BUFFER_RING = IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
528 /* caller should reissue async */
529 REQ_F_REISSUE = IO_REQ_FLAG(REQ_F_REISSUE_BIT),
530 /* supports async reads/writes */
531 REQ_F_SUPPORT_NOWAIT = IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
532 /* regular file */
533 REQ_F_ISREG = IO_REQ_FLAG(REQ_F_ISREG_BIT),
534 /* has creds assigned */
535 REQ_F_CREDS = IO_REQ_FLAG(REQ_F_CREDS_BIT),
536 /* skip refcounting if not set */
537 REQ_F_REFCOUNT = IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
538 /* there is a linked timeout that has to be armed */
539 REQ_F_ARM_LTIMEOUT = IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
540 /* ->async_data allocated */
541 REQ_F_ASYNC_DATA = IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
542 /* don't post CQEs while failing linked requests */
543 REQ_F_SKIP_LINK_CQES = IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
544 /* single poll may be active */
545 REQ_F_SINGLE_POLL = IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
546 /* double poll may active */
547 REQ_F_DOUBLE_POLL = IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
548 /* fast poll multishot mode */
549 REQ_F_APOLL_MULTISHOT = IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
550 /* recvmsg special flag, clear EPOLLIN */
551 REQ_F_CLEAR_POLLIN = IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_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 /* file is pollable */
555 REQ_F_CAN_POLL = IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
556 /* buffer list was empty after selection of buffer */
557 REQ_F_BL_EMPTY = IO_REQ_FLAG(REQ_F_BL_EMPTY_BIT),
558 /* don't recycle provided buffers for this request */
559 REQ_F_BL_NO_RECYCLE = IO_REQ_FLAG(REQ_F_BL_NO_RECYCLE_BIT),
560 /* buffer ring head needs incrementing on put */
561 REQ_F_BUFFERS_COMMIT = IO_REQ_FLAG(REQ_F_BUFFERS_COMMIT_BIT),
562 /* buf node is valid */
563 REQ_F_BUF_NODE = IO_REQ_FLAG(REQ_F_BUF_NODE_BIT),
564 };
565
566 typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
567
568 struct io_task_work {
569 struct llist_node node;
570 io_req_tw_func_t func;
571 };
572
573 struct io_cqe {
574 __u64 user_data;
575 __s32 res;
576 /* fd initially, then cflags for completion */
577 union {
578 __u32 flags;
579 int fd;
580 };
581 };
582
583 /*
584 * Each request type overlays its private data structure on top of this one.
585 * They must not exceed this one in size.
586 */
587 struct io_cmd_data {
588 struct file *file;
589 /* each command gets 56 bytes of data */
590 __u8 data[56];
591 };
592
io_kiocb_cmd_sz_check(size_t cmd_sz)593 static inline void io_kiocb_cmd_sz_check(size_t cmd_sz)
594 {
595 BUILD_BUG_ON(cmd_sz > sizeof(struct io_cmd_data));
596 }
597 #define io_kiocb_to_cmd(req, cmd_type) ( \
598 io_kiocb_cmd_sz_check(sizeof(cmd_type)) , \
599 ((cmd_type *)&(req)->cmd) \
600 )
601 #define cmd_to_io_kiocb(ptr) ((struct io_kiocb *) ptr)
602
603 struct io_kiocb {
604 union {
605 /*
606 * NOTE! Each of the io_kiocb union members has the file pointer
607 * as the first entry in their struct definition. So you can
608 * access the file pointer through any of the sub-structs,
609 * or directly as just 'file' in this struct.
610 */
611 struct file *file;
612 struct io_cmd_data cmd;
613 };
614
615 u8 opcode;
616 /* polled IO has completed */
617 u8 iopoll_completed;
618 /*
619 * Can be either a fixed buffer index, or used with provided buffers.
620 * For the latter, before issue it points to the buffer group ID,
621 * and after selection it points to the buffer ID itself.
622 */
623 u16 buf_index;
624
625 unsigned nr_tw;
626
627 /* REQ_F_* flags */
628 io_req_flags_t flags;
629
630 struct io_cqe cqe;
631
632 struct io_ring_ctx *ctx;
633 struct io_uring_task *tctx;
634
635 union {
636 /* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */
637 struct io_buffer *kbuf;
638
639 /*
640 * stores buffer ID for ring provided buffers, valid IFF
641 * REQ_F_BUFFER_RING is set.
642 */
643 struct io_buffer_list *buf_list;
644
645 struct io_rsrc_node *buf_node;
646 };
647
648 union {
649 /* used by request caches, completion batching and iopoll */
650 struct io_wq_work_node comp_list;
651 /* cache ->apoll->events */
652 __poll_t apoll_events;
653 };
654
655 struct io_rsrc_node *file_node;
656
657 atomic_t refs;
658 bool cancel_seq_set;
659 struct io_task_work io_task_work;
660 union {
661 /*
662 * for polled requests, i.e. IORING_OP_POLL_ADD and async armed
663 * poll
664 */
665 struct hlist_node hash_node;
666 /* For IOPOLL setup queues, with hybrid polling */
667 u64 iopoll_start;
668 };
669 /* internal polling, see IORING_FEAT_FAST_POLL */
670 struct async_poll *apoll;
671 /* opcode allocated if it needs to store data for async defer */
672 void *async_data;
673 /* linked requests, IFF REQ_F_HARDLINK or REQ_F_LINK are set */
674 atomic_t poll_refs;
675 struct io_kiocb *link;
676 /* custom credentials, valid IFF REQ_F_CREDS is set */
677 const struct cred *creds;
678 struct io_wq_work work;
679
680 struct {
681 u64 extra1;
682 u64 extra2;
683 } big_cqe;
684 };
685
686 struct io_overflow_cqe {
687 struct list_head list;
688 struct io_uring_cqe cqe;
689 };
690
io_ctx_cqe32(struct io_ring_ctx * ctx)691 static inline bool io_ctx_cqe32(struct io_ring_ctx *ctx)
692 {
693 return ctx->flags & IORING_SETUP_CQE32;
694 }
695
696 #endif
697