1 #ifndef IO_URING_TYPES_H
2 #define IO_URING_TYPES_H
3
4 #include <linux/blk_plug.h>
5 #include <linux/hashtable.h>
6 #include <linux/task_work.h>
7 #include <linux/bitmap.h>
8 #include <linux/llist.h>
9 #include <linux/uio.h>
10 #include <uapi/linux/io_uring.h>
11
12 struct iou_loop_params;
13 struct io_uring_bpf_ops;
14
15 enum {
16 /*
17 * A hint to not wake right away but delay until there are enough of
18 * tw's queued to match the number of CQEs the task is waiting for.
19 *
20 * Must not be used with requests generating more than one CQE.
21 * It's also ignored unless IORING_SETUP_DEFER_TASKRUN is set.
22 */
23 IOU_F_TWQ_LAZY_WAKE = 1,
24
25 /*
26 * Set when task_work is queued from a waitqueue wakeup handler, where
27 * an arbitrary provider waitqueue lock is held. Signaling the CQ ring
28 * eventfd inline from there can recurse back into that lock through
29 * epoll, so the eventfd signal must be deferred.
30 */
31 IOU_F_TWQ_IN_WAKE = 2,
32 };
33
34 enum io_uring_cmd_flags {
35 IO_URING_F_COMPLETE_DEFER = 1,
36 IO_URING_F_UNLOCKED = 2,
37 /* the request is executed from poll, it should not be freed */
38 IO_URING_F_MULTISHOT = 4,
39 /* executed by io-wq */
40 IO_URING_F_IOWQ = 8,
41 /* executed inline from syscall */
42 IO_URING_F_INLINE = 16,
43 /* int's last bit, sign checks are usually faster than a bit test */
44 IO_URING_F_NONBLOCK = INT_MIN,
45
46 /* ctx state flags, for URING_CMD */
47 IO_URING_F_SQE128 = (1 << 8),
48 IO_URING_F_CQE32 = (1 << 9),
49 IO_URING_F_IOPOLL = (1 << 10),
50
51 /* set when uring wants to cancel a previously issued command */
52 IO_URING_F_CANCEL = (1 << 11),
53 IO_URING_F_COMPAT = (1 << 12),
54 };
55
56 enum {
57 IO_BUF_DEST = 1 << ITER_DEST,
58 IO_BUF_SOURCE = 1 << ITER_SOURCE,
59 };
60
61 struct iou_loop_params;
62
63 struct io_wq_work_node {
64 struct io_wq_work_node *next;
65 };
66
67 struct io_wq_work_list {
68 struct io_wq_work_node *first;
69 struct io_wq_work_node *last;
70 };
71
72 /*
73 * Lockless multi-producer, single-consumer FIFO queue, see
74 * io_uring/mpscq.h for the implementation and rules. Defined here so
75 * that it can be embedded in io_ring_ctx. This is the producer side
76 * only - the consumer cursor is kept separately, on a cacheline that
77 * isn't dirtied by the producers.
78 */
79 struct mpscq {
80 struct llist_node *tail; /* producers */
81 struct llist_node stub;
82 };
83
84 struct io_wq_work {
85 struct io_wq_work_node list;
86 atomic_t flags;
87 /* place it here instead of io_kiocb as it fills padding and saves 4B */
88 int cancel_seq;
89 };
90
91 struct io_rsrc_data {
92 unsigned int nr;
93 struct io_rsrc_node **nodes;
94 };
95
96 struct io_file_table {
97 struct io_rsrc_data data;
98 unsigned long *bitmap;
99 unsigned int alloc_hint;
100 };
101
102 struct io_hash_bucket {
103 struct hlist_head list;
104 } ____cacheline_aligned_in_smp;
105
106 struct io_hash_table {
107 struct io_hash_bucket *hbs;
108 unsigned hash_bits;
109 };
110
111 struct io_mapped_region {
112 struct page **pages;
113 void *ptr;
114 unsigned nr_pages;
115 unsigned flags;
116 };
117
118 /*
119 * Return value from io_buffer_list selection, to avoid stashing it in
120 * struct io_kiocb. For legacy/classic provided buffers, keeping a reference
121 * across execution contexts are fine. But for ring provided buffers, the
122 * list may go away as soon as ->uring_lock is dropped. As the io_kiocb
123 * persists, it's better to just keep the buffer local for those cases.
124 */
125 struct io_br_sel {
126 struct io_buffer_list *buf_list;
127 /*
128 * Some selection parts return the user address, others return an error.
129 */
130 union {
131 void __user *addr;
132 ssize_t val;
133 };
134 };
135
136
137 /*
138 * Arbitrary limit, can be raised if need be
139 */
140 #define IO_RINGFD_REG_MAX 16
141
142 struct io_uring_task {
143 /* submission side */
144 int cached_refs;
145 const struct io_ring_ctx *last;
146 struct task_struct *task;
147 struct io_wq *io_wq;
148 /*
149 * Consumer cursor for ->task_list. Only popped by the task itself,
150 * or by ->fallback_work once the task can no longer run task_work.
151 */
152 struct llist_node *task_head;
153 struct file *registered_rings[IO_RINGFD_REG_MAX];
154
155 struct xarray xa;
156 struct wait_queue_head wait;
157 atomic_t in_cancel;
158 atomic_t inflight_tracked;
159 struct percpu_counter inflight;
160
161 /* drains ->task_list once the task can no longer run task_work */
162 struct work_struct fallback_work;
163
164 struct { /* task_work */
165 struct mpscq task_list;
166 struct callback_head task_work;
167 } ____cacheline_aligned_in_smp;
168 };
169
170 struct iou_vec {
171 union {
172 struct iovec *iovec;
173 struct bio_vec *bvec;
174 };
175 unsigned nr; /* number of struct iovec it can hold */
176 };
177
178 struct io_uring {
179 u32 head;
180 u32 tail;
181 };
182
183 /*
184 * This data is shared with the application through the mmap at offsets
185 * IORING_OFF_SQ_RING and IORING_OFF_CQ_RING.
186 *
187 * The offsets to the member fields are published through struct
188 * io_sqring_offsets when calling io_uring_setup.
189 */
190 struct io_rings {
191 /*
192 * Head and tail offsets into the ring; the offsets need to be
193 * masked to get valid indices.
194 *
195 * The kernel controls head of the sq ring and the tail of the cq ring,
196 * and the application controls tail of the sq ring and the head of the
197 * cq ring.
198 */
199 struct io_uring sq, cq;
200 /*
201 * Bitmasks to apply to head and tail offsets (constant, equals
202 * ring_entries - 1)
203 */
204 u32 sq_ring_mask, cq_ring_mask;
205 /* Ring sizes (constant, power of 2) */
206 u32 sq_ring_entries, cq_ring_entries;
207 /*
208 * Number of invalid entries dropped by the kernel due to
209 * invalid index stored in array
210 *
211 * Written by the kernel, shouldn't be modified by the
212 * application (i.e. get number of "new events" by comparing to
213 * cached value).
214 *
215 * After a new SQ head value was read by the application this
216 * counter includes all submissions that were dropped reaching
217 * the new SQ head (and possibly more).
218 */
219 u32 sq_dropped;
220 /*
221 * Runtime SQ flags
222 *
223 * Written by the kernel, shouldn't be modified by the
224 * application.
225 *
226 * The application needs a full memory barrier before checking
227 * for IORING_SQ_NEED_WAKEUP after updating the sq tail.
228 */
229 atomic_t sq_flags;
230 /*
231 * Runtime CQ flags
232 *
233 * Written by the application, shouldn't be modified by the
234 * kernel.
235 */
236 u32 cq_flags;
237 /*
238 * Number of completion events lost because the queue was full;
239 * this should be avoided by the application by making sure
240 * there are not more requests pending than there is space in
241 * the completion queue.
242 *
243 * Written by the kernel, shouldn't be modified by the
244 * application (i.e. get number of "new events" by comparing to
245 * cached value).
246 *
247 * As completion events come in out of order this counter is not
248 * ordered with any other data.
249 */
250 u32 cq_overflow;
251 /*
252 * Ring buffer of completion events.
253 *
254 * The kernel writes completion events fresh every time they are
255 * produced, so the application is allowed to modify pending
256 * entries.
257 */
258 struct io_uring_cqe cqes[] ____cacheline_aligned_in_smp;
259 };
260
261 struct io_bpf_filter;
262 struct io_bpf_filters {
263 refcount_t refs; /* ref for ->bpf_filters */
264 spinlock_t lock; /* protects ->bpf_filters modifications */
265 struct io_bpf_filter __rcu **filters;
266 struct rcu_head rcu_head;
267 };
268
269 struct io_restriction {
270 DECLARE_BITMAP(register_op, IORING_REGISTER_LAST);
271 DECLARE_BITMAP(sqe_op, IORING_OP_LAST);
272 struct io_bpf_filters *bpf_filters;
273 /* ->bpf_filters needs COW on modification */
274 bool bpf_filters_cow;
275 u8 sqe_flags_allowed;
276 u8 sqe_flags_required;
277 /* IORING_OP_* restrictions exist */
278 bool op_registered;
279 /* IORING_REGISTER_* restrictions exist */
280 bool reg_registered;
281 };
282
283 struct io_submit_link {
284 struct io_kiocb *head;
285 struct io_kiocb *last;
286 };
287
288 struct io_submit_state {
289 /* inline/task_work completion list, under ->uring_lock */
290 struct io_wq_work_node free_list;
291 /* batch completion logic */
292 struct io_wq_work_list compl_reqs;
293 struct io_submit_link link;
294
295 bool plug_started;
296 bool need_plug;
297 bool cq_flush;
298 unsigned short submit_nr;
299 struct blk_plug plug;
300 };
301
302 struct io_alloc_cache {
303 void **entries;
304 unsigned int nr_cached;
305 unsigned int max_cached;
306 unsigned int elem_size;
307 unsigned int init_clear;
308 };
309
310 enum {
311 IO_RING_F_DRAIN_NEXT = BIT(0),
312 IO_RING_F_OP_RESTRICTED = BIT(1),
313 IO_RING_F_REG_RESTRICTED = BIT(2),
314 IO_RING_F_OFF_TIMEOUT_USED = BIT(3),
315 IO_RING_F_DRAIN_ACTIVE = BIT(4),
316 IO_RING_F_HAS_EVFD = BIT(5),
317 /* all CQEs should be posted only by the submitter task */
318 IO_RING_F_TASK_COMPLETE = BIT(6),
319 IO_RING_F_LOCKLESS_CQ = BIT(7),
320 IO_RING_F_SYSCALL_IOPOLL = BIT(8),
321 IO_RING_F_POLL_ACTIVATED = BIT(9),
322 IO_RING_F_DRAIN_DISABLED = BIT(10),
323 IO_RING_F_COMPAT = BIT(11),
324 IO_RING_F_IOWQ_LIMITS_SET = BIT(12),
325 /* restrictions were inherited from the task that created the ring */
326 IO_RING_F_RESTRICT_INHERITED = BIT(13),
327 };
328
329 struct iou_ctx {};
330
331 struct io_ring_ctx {
332 /* const or read-mostly hot data */
333 struct {
334 /* ring setup flags */
335 unsigned int flags;
336 /* internal state flags IO_RING_F_* flags , mostly read-only */
337 unsigned int int_flags;
338
339 struct task_struct *submitter_task;
340 struct io_rings *rings;
341 /* cache of ->restrictions.bpf_filters->filters */
342 struct io_bpf_filter __rcu **bpf_filters;
343 struct percpu_ref refs;
344
345 clockid_t clockid;
346 enum tk_offsets clock_offset;
347
348 enum task_work_notify_mode notify_method;
349 unsigned sq_thread_idle;
350 } ____cacheline_aligned_in_smp;
351
352 /* submission data */
353 struct {
354 struct mutex uring_lock;
355
356 /*
357 * Ring buffer of indices into array of io_uring_sqe, which is
358 * mmapped by the application using the IORING_OFF_SQES offset.
359 *
360 * This indirection could e.g. be used to assign fixed
361 * io_uring_sqe entries to operations and only submit them to
362 * the queue when needed.
363 *
364 * The kernel modifies neither the indices array nor the entries
365 * array.
366 */
367 u32 *sq_array;
368 struct io_uring_sqe *sq_sqes;
369 unsigned cached_sq_head;
370 unsigned sq_entries;
371
372 /*
373 * Fixed resources fast path, should be accessed only under
374 * uring_lock, and updated through io_uring_register(2)
375 */
376 atomic_t cancel_seq;
377
378 /*
379 * ->iopoll_list is protected by the ctx->uring_lock for
380 * io_uring instances that don't use IORING_SETUP_SQPOLL.
381 * For SQPOLL, only the single threaded io_sq_thread() will
382 * manipulate the list, hence no extra locking is needed there.
383 */
384 bool poll_multi_queue;
385 struct list_head iopoll_list;
386
387 /*
388 * Consumer cursor for ->work_list, protected by ->uring_lock.
389 * Deliberately kept away from the producer side of the queue,
390 * as it's written for every popped entry, and the producer
391 * cacheline is contended enough as it is.
392 */
393 struct llist_node *work_head;
394
395 struct io_file_table file_table;
396 struct io_rsrc_data buf_table;
397 struct io_alloc_cache node_cache;
398 struct io_alloc_cache imu_cache;
399
400 struct io_submit_state submit_state;
401
402 /*
403 * Modifications are protected by ->uring_lock and ->mmap_lock.
404 * The buffer list's io mapped region should be stable once
405 * published.
406 */
407 struct xarray io_bl_xa;
408
409 struct io_hash_table cancel_table;
410 struct io_alloc_cache apoll_cache;
411 struct io_alloc_cache netmsg_cache;
412 struct io_alloc_cache rw_cache;
413 struct io_alloc_cache cmd_cache;
414
415 int (*loop_step)(struct iou_ctx *,
416 struct iou_loop_params *);
417
418 /*
419 * Any cancelable uring_cmd is added to this list in
420 * ->uring_cmd() by io_uring_cmd_insert_cancelable()
421 */
422 struct hlist_head cancelable_uring_cmd;
423 /*
424 * For Hybrid IOPOLL, runtime in hybrid polling, without
425 * scheduling time
426 */
427 u64 hybrid_poll_time;
428 } ____cacheline_aligned_in_smp;
429
430 struct {
431 /*
432 * We cache a range of free CQEs we can use, once exhausted it
433 * should go through a slower range setup, see __io_get_cqe()
434 */
435 struct io_uring_cqe *cqe_cached;
436 struct io_uring_cqe *cqe_sentinel;
437
438 unsigned cached_cq_tail;
439 unsigned cq_entries;
440 struct io_ev_fd __rcu *io_ev_fd;
441
442 void *cq_wait_arg;
443 size_t cq_wait_size;
444 } ____cacheline_aligned_in_smp;
445
446 /*
447 * task_work and async notification delivery cacheline. Expected to
448 * regularly bounce b/w CPUs.
449 */
450 struct {
451 struct io_rings __rcu *rings_rcu;
452 struct mpscq work_list;
453 unsigned long check_cq;
454 atomic_t cq_wait_nr;
455 atomic_t cq_timeouts;
456 struct wait_queue_head cq_wait;
457 } ____cacheline_aligned_in_smp;
458
459 /* timeouts */
460 struct {
461 raw_spinlock_t timeout_lock;
462 struct list_head timeout_list;
463 struct list_head ltimeout_list;
464 unsigned cq_last_tm_flush;
465 } ____cacheline_aligned_in_smp;
466
467 spinlock_t completion_lock;
468
469 struct list_head cq_overflow_list;
470
471 struct hlist_head waitid_list;
472
473 #ifdef CONFIG_FUTEX
474 struct hlist_head futex_list;
475 struct io_alloc_cache futex_cache;
476 #endif
477
478 const struct cred *sq_creds; /* cred used for __io_sq_thread() */
479 struct io_sq_data *sq_data; /* if using sq thread polling */
480
481 struct wait_queue_head sqo_sq_wait;
482 struct list_head sqd_list;
483
484 unsigned int file_alloc_start;
485 unsigned int file_alloc_end;
486
487 /* Keep this last, we don't need it for the fast path */
488 struct wait_queue_head poll_wq;
489 struct io_restriction restrictions;
490
491 /* Stores zcrx object pointers of type struct io_zcrx_ifq */
492 struct xarray zcrx_ctxs;
493
494 /* Used for accounting references on pages in registered buffers */
495 struct xarray hpage_acct;
496
497 u32 pers_next;
498 struct xarray personalities;
499
500 /* hashed buffered write serialization */
501 struct io_wq_hash *hash_map;
502
503 /* Only used for accounting purposes */
504 struct user_struct *user;
505 struct mm_struct *mm_account;
506
507 /*
508 * List of tctx nodes for this ctx, protected by tctx_lock. For
509 * cancelation purposes, nests under uring_lock.
510 */
511 struct list_head tctx_list;
512 struct mutex tctx_lock;
513
514 /* ctx exit and cancelation */
515 struct work_struct exit_work;
516 struct completion ref_comp;
517
518 /* io-wq management, e.g. thread count */
519 u32 iowq_limits[2];
520
521 struct callback_head poll_wq_task_work;
522 struct list_head defer_list;
523 unsigned nr_drained;
524
525 /* protected by ->completion_lock */
526 unsigned nr_req_allocated;
527
528 #ifdef CONFIG_NET_RX_BUSY_POLL
529 struct list_head napi_list; /* track busy poll napi_id */
530 spinlock_t napi_lock; /* napi_list lock */
531
532 /* napi busy poll default timeout */
533 ktime_t napi_busy_poll_dt;
534 bool napi_prefer_busy_poll;
535 u8 napi_track_mode;
536
537 DECLARE_HASHTABLE(napi_ht, 4);
538 #endif
539
540 struct io_uring_bpf_ops *bpf_ops;
541
542 /*
543 * Protection for resize vs mmap races - both the mmap and resize
544 * side will need to grab this lock, to prevent either side from
545 * being run concurrently with the other.
546 */
547 struct mutex mmap_lock;
548
549 struct io_mapped_region sq_region;
550 struct io_mapped_region ring_region;
551 /* used for optimised request parameter and wait argument passing */
552 struct io_mapped_region param_region;
553
554 struct kcov_common_handle_id kcov_handle;
555 };
556
557 /*
558 * Token indicating function is called in task work context:
559 * ctx->uring_lock is held and any completions generated will be flushed.
560 * ONLY core io_uring.c should instantiate this struct.
561 */
562 struct io_tw_state {
563 bool cancel;
564 };
565 /* Alias to use in code that doesn't instantiate struct io_tw_state */
566 typedef struct io_tw_state io_tw_token_t;
567
568 enum {
569 REQ_F_FIXED_FILE_BIT = IOSQE_FIXED_FILE_BIT,
570 REQ_F_IO_DRAIN_BIT = IOSQE_IO_DRAIN_BIT,
571 REQ_F_LINK_BIT = IOSQE_IO_LINK_BIT,
572 REQ_F_HARDLINK_BIT = IOSQE_IO_HARDLINK_BIT,
573 REQ_F_FORCE_ASYNC_BIT = IOSQE_ASYNC_BIT,
574 REQ_F_BUFFER_SELECT_BIT = IOSQE_BUFFER_SELECT_BIT,
575 REQ_F_CQE_SKIP_BIT = IOSQE_CQE_SKIP_SUCCESS_BIT,
576
577 /* first byte is taken by user flags, shift it to not overlap */
578 REQ_F_FAIL_BIT = 8,
579 REQ_F_INFLIGHT_BIT,
580 REQ_F_CUR_POS_BIT,
581 REQ_F_NOWAIT_BIT,
582 REQ_F_LINK_TIMEOUT_BIT,
583 REQ_F_NEED_CLEANUP_BIT,
584 REQ_F_POLLED_BIT,
585 REQ_F_HYBRID_IOPOLL_STATE_BIT,
586 REQ_F_BUFFER_SELECTED_BIT,
587 REQ_F_BUFFER_RING_BIT,
588 REQ_F_REISSUE_BIT,
589 REQ_F_CREDS_BIT,
590 REQ_F_REFCOUNT_BIT,
591 REQ_F_ARM_LTIMEOUT_BIT,
592 REQ_F_ASYNC_DATA_BIT,
593 REQ_F_SKIP_LINK_CQES_BIT,
594 REQ_F_SINGLE_POLL_BIT,
595 REQ_F_DOUBLE_POLL_BIT,
596 REQ_F_MULTISHOT_BIT,
597 REQ_F_APOLL_MULTISHOT_BIT,
598 REQ_F_CLEAR_POLLIN_BIT,
599 /* keep async read/write and isreg together and in order */
600 REQ_F_SUPPORT_NOWAIT_BIT,
601 REQ_F_ISREG_BIT,
602 REQ_F_POLL_NO_LAZY_BIT,
603 REQ_F_CAN_POLL_BIT,
604 REQ_F_BL_EMPTY_BIT,
605 REQ_F_BL_NO_RECYCLE_BIT,
606 REQ_F_BUFFERS_COMMIT_BIT,
607 REQ_F_BUF_NODE_BIT,
608 REQ_F_BUF_MORE_BIT,
609 REQ_F_HAS_METADATA_BIT,
610 REQ_F_IMPORT_BUFFER_BIT,
611 REQ_F_SQE_COPIED_BIT,
612 REQ_F_IOPOLL_BIT,
613
614 /* not a real bit, just to check we're not overflowing the space */
615 __REQ_F_LAST_BIT,
616 };
617
618 typedef u64 __bitwise io_req_flags_t;
619 #define IO_REQ_FLAG(bitno) ((__force io_req_flags_t) BIT_ULL((bitno)))
620
621 enum {
622 /* ctx owns file */
623 REQ_F_FIXED_FILE = IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT),
624 /* drain existing IO first */
625 REQ_F_IO_DRAIN = IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT),
626 /* linked sqes */
627 REQ_F_LINK = IO_REQ_FLAG(REQ_F_LINK_BIT),
628 /* doesn't sever on completion < 0 */
629 REQ_F_HARDLINK = IO_REQ_FLAG(REQ_F_HARDLINK_BIT),
630 /* IOSQE_ASYNC */
631 REQ_F_FORCE_ASYNC = IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT),
632 /* IOSQE_BUFFER_SELECT */
633 REQ_F_BUFFER_SELECT = IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
634 /* IOSQE_CQE_SKIP_SUCCESS */
635 REQ_F_CQE_SKIP = IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
636
637 /* fail rest of links */
638 REQ_F_FAIL = IO_REQ_FLAG(REQ_F_FAIL_BIT),
639 /* on inflight list, should be cancelled and waited on exit reliably */
640 REQ_F_INFLIGHT = IO_REQ_FLAG(REQ_F_INFLIGHT_BIT),
641 /* read/write uses file position */
642 REQ_F_CUR_POS = IO_REQ_FLAG(REQ_F_CUR_POS_BIT),
643 /* must not punt to workers */
644 REQ_F_NOWAIT = IO_REQ_FLAG(REQ_F_NOWAIT_BIT),
645 /* has or had linked timeout */
646 REQ_F_LINK_TIMEOUT = IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT),
647 /* needs cleanup */
648 REQ_F_NEED_CLEANUP = IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
649 /* already went through poll handler */
650 REQ_F_POLLED = IO_REQ_FLAG(REQ_F_POLLED_BIT),
651 /* every req only blocks once in hybrid poll */
652 REQ_F_IOPOLL_STATE = IO_REQ_FLAG(REQ_F_HYBRID_IOPOLL_STATE_BIT),
653 /* buffer already selected */
654 REQ_F_BUFFER_SELECTED = IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
655 /* buffer selected from ring, needs commit */
656 REQ_F_BUFFER_RING = IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
657 /* caller should reissue async */
658 REQ_F_REISSUE = IO_REQ_FLAG(REQ_F_REISSUE_BIT),
659 /* supports async reads/writes */
660 REQ_F_SUPPORT_NOWAIT = IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
661 /* regular file */
662 REQ_F_ISREG = IO_REQ_FLAG(REQ_F_ISREG_BIT),
663 /* has creds assigned */
664 REQ_F_CREDS = IO_REQ_FLAG(REQ_F_CREDS_BIT),
665 /* skip refcounting if not set */
666 REQ_F_REFCOUNT = IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
667 /* there is a linked timeout that has to be armed */
668 REQ_F_ARM_LTIMEOUT = IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
669 /* ->async_data allocated */
670 REQ_F_ASYNC_DATA = IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
671 /* don't post CQEs while failing linked requests */
672 REQ_F_SKIP_LINK_CQES = IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
673 /* single poll may be active */
674 REQ_F_SINGLE_POLL = IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
675 /* double poll may active */
676 REQ_F_DOUBLE_POLL = IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
677 /* request posts multiple completions, should be set at prep time */
678 REQ_F_MULTISHOT = IO_REQ_FLAG(REQ_F_MULTISHOT_BIT),
679 /* fast poll multishot mode */
680 REQ_F_APOLL_MULTISHOT = IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
681 /* recvmsg special flag, clear EPOLLIN */
682 REQ_F_CLEAR_POLLIN = IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT),
683 /* don't use lazy poll wake for this request */
684 REQ_F_POLL_NO_LAZY = IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
685 /* file is pollable */
686 REQ_F_CAN_POLL = IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
687 /* buffer list was empty after selection of buffer */
688 REQ_F_BL_EMPTY = IO_REQ_FLAG(REQ_F_BL_EMPTY_BIT),
689 /* don't recycle provided buffers for this request */
690 REQ_F_BL_NO_RECYCLE = IO_REQ_FLAG(REQ_F_BL_NO_RECYCLE_BIT),
691 /* buffer ring head needs incrementing on put */
692 REQ_F_BUFFERS_COMMIT = IO_REQ_FLAG(REQ_F_BUFFERS_COMMIT_BIT),
693 /* buf node is valid */
694 REQ_F_BUF_NODE = IO_REQ_FLAG(REQ_F_BUF_NODE_BIT),
695 /* incremental buffer consumption, more space available */
696 REQ_F_BUF_MORE = IO_REQ_FLAG(REQ_F_BUF_MORE_BIT),
697 /* request has read/write metadata assigned */
698 REQ_F_HAS_METADATA = IO_REQ_FLAG(REQ_F_HAS_METADATA_BIT),
699 /*
700 * For vectored fixed buffers, resolve iovec to registered buffers.
701 * For SEND_ZC, whether to import buffers (i.e. the first issue).
702 */
703 REQ_F_IMPORT_BUFFER = IO_REQ_FLAG(REQ_F_IMPORT_BUFFER_BIT),
704 /* ->sqe_copy() has been called, if necessary */
705 REQ_F_SQE_COPIED = IO_REQ_FLAG(REQ_F_SQE_COPIED_BIT),
706 /* request must be iopolled to completion (set in ->issue()) */
707 REQ_F_IOPOLL = IO_REQ_FLAG(REQ_F_IOPOLL_BIT),
708 };
709
710 struct io_tw_req {
711 struct io_kiocb *req;
712 };
713
714 typedef void (*io_req_tw_func_t)(struct io_tw_req tw_req, io_tw_token_t tw);
715
716 struct io_task_work {
717 struct llist_node node;
718 io_req_tw_func_t func;
719 };
720
721 struct io_cqe {
722 __u64 user_data;
723 __s32 res;
724 /* fd initially, then cflags for completion */
725 union {
726 __u32 flags;
727 int fd;
728 };
729 };
730
731 /*
732 * Each request type overlays its private data structure on top of this one.
733 * They must not exceed this one in size.
734 */
735 struct io_cmd_data {
736 struct file *file;
737 /* each command gets 56 bytes of data */
738 __u8 data[56];
739 };
740
io_kiocb_cmd_sz_check(size_t cmd_sz)741 static inline void io_kiocb_cmd_sz_check(size_t cmd_sz)
742 {
743 BUILD_BUG_ON(cmd_sz > sizeof(struct io_cmd_data));
744 }
745 #define io_kiocb_to_cmd(req, cmd_type) ( \
746 io_kiocb_cmd_sz_check(sizeof(cmd_type)) , \
747 ((cmd_type *)&(req)->cmd) \
748 )
749
cmd_to_io_kiocb(void * ptr)750 static inline struct io_kiocb *cmd_to_io_kiocb(void *ptr)
751 {
752 return ptr;
753 }
754
755 struct io_kiocb {
756 union {
757 /*
758 * NOTE! Each of the io_kiocb union members has the file pointer
759 * as the first entry in their struct definition. So you can
760 * access the file pointer through any of the sub-structs,
761 * or directly as just 'file' in this struct.
762 */
763 struct file *file;
764 struct io_cmd_data cmd;
765 };
766
767 u8 opcode;
768 /* polled IO has completed */
769 u8 iopoll_completed;
770 /*
771 * Can be either a fixed buffer index, or used with provided buffers.
772 * For the latter, it points to the selected buffer ID.
773 */
774 u16 buf_index;
775
776 /* REQ_F_* flags */
777 io_req_flags_t flags;
778
779 struct io_cqe cqe;
780
781 struct io_ring_ctx *ctx;
782 struct io_uring_task *tctx;
783
784 union {
785 /* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */
786 struct io_buffer *kbuf;
787
788 struct io_rsrc_node *buf_node;
789 };
790
791 union {
792 /* used by request caches, completion batching and iopoll */
793 struct io_wq_work_node comp_list;
794 /* cache ->apoll->events */
795 __poll_t apoll_events;
796 };
797
798 struct io_rsrc_node *file_node;
799
800 atomic_t refs;
801 bool cancel_seq_set;
802
803 union {
804 struct io_task_work io_task_work;
805 /* For IOPOLL setup queues, with hybrid polling */
806 u64 iopoll_start;
807 };
808
809 union {
810 /*
811 * for polled requests, i.e. IORING_OP_POLL_ADD and async armed
812 * poll
813 */
814 struct hlist_node hash_node;
815 /* IOPOLL completion handling */
816 struct list_head iopoll_node;
817 /* for private io_kiocb freeing */
818 struct rcu_head rcu_head;
819 };
820 /* internal polling, see IORING_FEAT_FAST_POLL */
821 struct async_poll *apoll;
822 /* opcode allocated if it needs to store data for async defer */
823 void *async_data;
824 /* linked requests, IFF REQ_F_HARDLINK or REQ_F_LINK are set */
825 atomic_t poll_refs;
826 struct io_kiocb *link;
827 /* custom credentials, valid IFF REQ_F_CREDS is set */
828 const struct cred *creds;
829 struct io_wq_work work;
830
831 struct io_big_cqe {
832 u64 extra1;
833 u64 extra2;
834 } big_cqe;
835 };
836
837 struct io_overflow_cqe {
838 struct list_head list;
839 struct io_uring_cqe cqe;
840 };
841 #endif
842