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