xref: /linux/include/linux/io_uring_types.h (revision 35c8711c)
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 };
41 
42 struct io_wq_work_node {
43 	struct io_wq_work_node *next;
44 };
45 
46 struct io_wq_work_list {
47 	struct io_wq_work_node *first;
48 	struct io_wq_work_node *last;
49 };
50 
51 struct 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 
58 struct io_fixed_file {
59 	/* file * with additional FFS_* flags */
60 	unsigned long file_ptr;
61 };
62 
63 struct io_file_table {
64 	struct io_fixed_file *files;
65 	unsigned long *bitmap;
66 	unsigned int alloc_hint;
67 };
68 
69 struct io_hash_bucket {
70 	spinlock_t		lock;
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 /*
80  * Arbitrary limit, can be raised if need be
81  */
82 #define IO_RINGFD_REG_MAX 16
83 
84 struct 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 
103 struct 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  */
115 struct 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 
186 struct 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 
194 struct io_submit_link {
195 	struct io_kiocb		*head;
196 	struct io_kiocb		*last;
197 };
198 
199 struct 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 	bool			cq_flush;
209 	unsigned short		submit_nr;
210 	unsigned int		cqes_count;
211 	struct blk_plug		plug;
212 };
213 
214 struct io_ev_fd {
215 	struct eventfd_ctx	*cq_ev_fd;
216 	unsigned int		eventfd_async: 1;
217 	struct rcu_head		rcu;
218 	atomic_t		refs;
219 	atomic_t		ops;
220 };
221 
222 struct io_alloc_cache {
223 	void			**entries;
224 	unsigned int		nr_cached;
225 	unsigned int		max_cached;
226 	size_t			elem_size;
227 };
228 
229 struct io_ring_ctx {
230 	/* const or read-mostly hot data */
231 	struct {
232 		unsigned int		flags;
233 		unsigned int		drain_next: 1;
234 		unsigned int		restricted: 1;
235 		unsigned int		off_timeout_used: 1;
236 		unsigned int		drain_active: 1;
237 		unsigned int		has_evfd: 1;
238 		/* all CQEs should be posted only by the submitter task */
239 		unsigned int		task_complete: 1;
240 		unsigned int		lockless_cq: 1;
241 		unsigned int		syscall_iopoll: 1;
242 		unsigned int		poll_activated: 1;
243 		unsigned int		drain_disabled: 1;
244 		unsigned int		compat: 1;
245 		unsigned int		iowq_limits_set : 1;
246 
247 		struct task_struct	*submitter_task;
248 		struct io_rings		*rings;
249 		struct percpu_ref	refs;
250 
251 		enum task_work_notify_mode	notify_method;
252 		unsigned			sq_thread_idle;
253 	} ____cacheline_aligned_in_smp;
254 
255 	/* submission data */
256 	struct {
257 		struct mutex		uring_lock;
258 
259 		/*
260 		 * Ring buffer of indices into array of io_uring_sqe, which is
261 		 * mmapped by the application using the IORING_OFF_SQES offset.
262 		 *
263 		 * This indirection could e.g. be used to assign fixed
264 		 * io_uring_sqe entries to operations and only submit them to
265 		 * the queue when needed.
266 		 *
267 		 * The kernel modifies neither the indices array nor the entries
268 		 * array.
269 		 */
270 		u32			*sq_array;
271 		struct io_uring_sqe	*sq_sqes;
272 		unsigned		cached_sq_head;
273 		unsigned		sq_entries;
274 
275 		/*
276 		 * Fixed resources fast path, should be accessed only under
277 		 * uring_lock, and updated through io_uring_register(2)
278 		 */
279 		struct io_rsrc_node	*rsrc_node;
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_mapped_ubuf	**user_bufs;
293 		unsigned		nr_user_files;
294 		unsigned		nr_user_bufs;
295 
296 		struct io_submit_state	submit_state;
297 
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 		struct io_alloc_cache	rw_cache;
304 		struct io_alloc_cache	uring_cache;
305 
306 		/*
307 		 * Any cancelable uring_cmd is added to this list in
308 		 * ->uring_cmd() by io_uring_cmd_insert_cancelable()
309 		 */
310 		struct hlist_head	cancelable_uring_cmd;
311 	} ____cacheline_aligned_in_smp;
312 
313 	struct {
314 		/*
315 		 * We cache a range of free CQEs we can use, once exhausted it
316 		 * should go through a slower range setup, see __io_get_cqe()
317 		 */
318 		struct io_uring_cqe	*cqe_cached;
319 		struct io_uring_cqe	*cqe_sentinel;
320 
321 		unsigned		cached_cq_tail;
322 		unsigned		cq_entries;
323 		struct io_ev_fd	__rcu	*io_ev_fd;
324 		unsigned		cq_extra;
325 	} ____cacheline_aligned_in_smp;
326 
327 	/*
328 	 * task_work and async notification delivery cacheline. Expected to
329 	 * regularly bounce b/w CPUs.
330 	 */
331 	struct {
332 		struct llist_head	work_llist;
333 		unsigned long		check_cq;
334 		atomic_t		cq_wait_nr;
335 		atomic_t		cq_timeouts;
336 		struct wait_queue_head	cq_wait;
337 	} ____cacheline_aligned_in_smp;
338 
339 	/* timeouts */
340 	struct {
341 		spinlock_t		timeout_lock;
342 		struct list_head	timeout_list;
343 		struct list_head	ltimeout_list;
344 		unsigned		cq_last_tm_flush;
345 	} ____cacheline_aligned_in_smp;
346 
347 	spinlock_t		completion_lock;
348 
349 	struct list_head	io_buffers_comp;
350 	struct list_head	cq_overflow_list;
351 	struct io_hash_table	cancel_table;
352 
353 	struct hlist_head	waitid_list;
354 
355 #ifdef CONFIG_FUTEX
356 	struct hlist_head	futex_list;
357 	struct io_alloc_cache	futex_cache;
358 #endif
359 
360 	const struct cred	*sq_creds;	/* cred used for __io_sq_thread() */
361 	struct io_sq_data	*sq_data;	/* if using sq thread polling */
362 
363 	struct wait_queue_head	sqo_sq_wait;
364 	struct list_head	sqd_list;
365 
366 	unsigned int		file_alloc_start;
367 	unsigned int		file_alloc_end;
368 
369 	struct list_head	io_buffers_cache;
370 
371 	/* Keep this last, we don't need it for the fast path */
372 	struct wait_queue_head		poll_wq;
373 	struct io_restriction		restrictions;
374 
375 	/* slow path rsrc auxilary data, used by update/register */
376 	struct io_mapped_ubuf		*dummy_ubuf;
377 	struct io_rsrc_data		*file_data;
378 	struct io_rsrc_data		*buf_data;
379 
380 	/* protected by ->uring_lock */
381 	struct list_head		rsrc_ref_list;
382 	struct io_alloc_cache		rsrc_node_cache;
383 	struct wait_queue_head		rsrc_quiesce_wq;
384 	unsigned			rsrc_quiesce;
385 
386 	u32			pers_next;
387 	struct xarray		personalities;
388 
389 	/* hashed buffered write serialization */
390 	struct io_wq_hash		*hash_map;
391 
392 	/* Only used for accounting purposes */
393 	struct user_struct		*user;
394 	struct mm_struct		*mm_account;
395 
396 	/* ctx exit and cancelation */
397 	struct llist_head		fallback_llist;
398 	struct delayed_work		fallback_work;
399 	struct work_struct		exit_work;
400 	struct list_head		tctx_list;
401 	struct completion		ref_comp;
402 
403 	/* io-wq management, e.g. thread count */
404 	u32				iowq_limits[2];
405 
406 	struct callback_head		poll_wq_task_work;
407 	struct list_head		defer_list;
408 
409 #ifdef CONFIG_NET_RX_BUSY_POLL
410 	struct list_head	napi_list;	/* track busy poll napi_id */
411 	spinlock_t		napi_lock;	/* napi_list lock */
412 
413 	/* napi busy poll default timeout */
414 	unsigned int		napi_busy_poll_to;
415 	bool			napi_prefer_busy_poll;
416 	bool			napi_enabled;
417 
418 	DECLARE_HASHTABLE(napi_ht, 4);
419 #endif
420 
421 	/* protected by ->completion_lock */
422 	unsigned			evfd_last_cq_tail;
423 
424 	/*
425 	 * If IORING_SETUP_NO_MMAP is used, then the below holds
426 	 * the gup'ed pages for the two rings, and the sqes.
427 	 */
428 	unsigned short			n_ring_pages;
429 	unsigned short			n_sqe_pages;
430 	struct page			**ring_pages;
431 	struct page			**sqe_pages;
432 };
433 
434 struct io_tw_state {
435 };
436 
437 enum {
438 	REQ_F_FIXED_FILE_BIT	= IOSQE_FIXED_FILE_BIT,
439 	REQ_F_IO_DRAIN_BIT	= IOSQE_IO_DRAIN_BIT,
440 	REQ_F_LINK_BIT		= IOSQE_IO_LINK_BIT,
441 	REQ_F_HARDLINK_BIT	= IOSQE_IO_HARDLINK_BIT,
442 	REQ_F_FORCE_ASYNC_BIT	= IOSQE_ASYNC_BIT,
443 	REQ_F_BUFFER_SELECT_BIT	= IOSQE_BUFFER_SELECT_BIT,
444 	REQ_F_CQE_SKIP_BIT	= IOSQE_CQE_SKIP_SUCCESS_BIT,
445 
446 	/* first byte is taken by user flags, shift it to not overlap */
447 	REQ_F_FAIL_BIT		= 8,
448 	REQ_F_INFLIGHT_BIT,
449 	REQ_F_CUR_POS_BIT,
450 	REQ_F_NOWAIT_BIT,
451 	REQ_F_LINK_TIMEOUT_BIT,
452 	REQ_F_NEED_CLEANUP_BIT,
453 	REQ_F_POLLED_BIT,
454 	REQ_F_BUFFER_SELECTED_BIT,
455 	REQ_F_BUFFER_RING_BIT,
456 	REQ_F_REISSUE_BIT,
457 	REQ_F_CREDS_BIT,
458 	REQ_F_REFCOUNT_BIT,
459 	REQ_F_ARM_LTIMEOUT_BIT,
460 	REQ_F_ASYNC_DATA_BIT,
461 	REQ_F_SKIP_LINK_CQES_BIT,
462 	REQ_F_SINGLE_POLL_BIT,
463 	REQ_F_DOUBLE_POLL_BIT,
464 	REQ_F_APOLL_MULTISHOT_BIT,
465 	REQ_F_CLEAR_POLLIN_BIT,
466 	REQ_F_HASH_LOCKED_BIT,
467 	/* keep async read/write and isreg together and in order */
468 	REQ_F_SUPPORT_NOWAIT_BIT,
469 	REQ_F_ISREG_BIT,
470 	REQ_F_POLL_NO_LAZY_BIT,
471 	REQ_F_CANCEL_SEQ_BIT,
472 	REQ_F_CAN_POLL_BIT,
473 	REQ_F_BL_EMPTY_BIT,
474 	REQ_F_BL_NO_RECYCLE_BIT,
475 	REQ_F_BUFFERS_COMMIT_BIT,
476 
477 	/* not a real bit, just to check we're not overflowing the space */
478 	__REQ_F_LAST_BIT,
479 };
480 
481 typedef u64 __bitwise io_req_flags_t;
482 #define IO_REQ_FLAG(bitno)	((__force io_req_flags_t) BIT_ULL((bitno)))
483 
484 enum {
485 	/* ctx owns file */
486 	REQ_F_FIXED_FILE	= IO_REQ_FLAG(REQ_F_FIXED_FILE_BIT),
487 	/* drain existing IO first */
488 	REQ_F_IO_DRAIN		= IO_REQ_FLAG(REQ_F_IO_DRAIN_BIT),
489 	/* linked sqes */
490 	REQ_F_LINK		= IO_REQ_FLAG(REQ_F_LINK_BIT),
491 	/* doesn't sever on completion < 0 */
492 	REQ_F_HARDLINK		= IO_REQ_FLAG(REQ_F_HARDLINK_BIT),
493 	/* IOSQE_ASYNC */
494 	REQ_F_FORCE_ASYNC	= IO_REQ_FLAG(REQ_F_FORCE_ASYNC_BIT),
495 	/* IOSQE_BUFFER_SELECT */
496 	REQ_F_BUFFER_SELECT	= IO_REQ_FLAG(REQ_F_BUFFER_SELECT_BIT),
497 	/* IOSQE_CQE_SKIP_SUCCESS */
498 	REQ_F_CQE_SKIP		= IO_REQ_FLAG(REQ_F_CQE_SKIP_BIT),
499 
500 	/* fail rest of links */
501 	REQ_F_FAIL		= IO_REQ_FLAG(REQ_F_FAIL_BIT),
502 	/* on inflight list, should be cancelled and waited on exit reliably */
503 	REQ_F_INFLIGHT		= IO_REQ_FLAG(REQ_F_INFLIGHT_BIT),
504 	/* read/write uses file position */
505 	REQ_F_CUR_POS		= IO_REQ_FLAG(REQ_F_CUR_POS_BIT),
506 	/* must not punt to workers */
507 	REQ_F_NOWAIT		= IO_REQ_FLAG(REQ_F_NOWAIT_BIT),
508 	/* has or had linked timeout */
509 	REQ_F_LINK_TIMEOUT	= IO_REQ_FLAG(REQ_F_LINK_TIMEOUT_BIT),
510 	/* needs cleanup */
511 	REQ_F_NEED_CLEANUP	= IO_REQ_FLAG(REQ_F_NEED_CLEANUP_BIT),
512 	/* already went through poll handler */
513 	REQ_F_POLLED		= IO_REQ_FLAG(REQ_F_POLLED_BIT),
514 	/* buffer already selected */
515 	REQ_F_BUFFER_SELECTED	= IO_REQ_FLAG(REQ_F_BUFFER_SELECTED_BIT),
516 	/* buffer selected from ring, needs commit */
517 	REQ_F_BUFFER_RING	= IO_REQ_FLAG(REQ_F_BUFFER_RING_BIT),
518 	/* caller should reissue async */
519 	REQ_F_REISSUE		= IO_REQ_FLAG(REQ_F_REISSUE_BIT),
520 	/* supports async reads/writes */
521 	REQ_F_SUPPORT_NOWAIT	= IO_REQ_FLAG(REQ_F_SUPPORT_NOWAIT_BIT),
522 	/* regular file */
523 	REQ_F_ISREG		= IO_REQ_FLAG(REQ_F_ISREG_BIT),
524 	/* has creds assigned */
525 	REQ_F_CREDS		= IO_REQ_FLAG(REQ_F_CREDS_BIT),
526 	/* skip refcounting if not set */
527 	REQ_F_REFCOUNT		= IO_REQ_FLAG(REQ_F_REFCOUNT_BIT),
528 	/* there is a linked timeout that has to be armed */
529 	REQ_F_ARM_LTIMEOUT	= IO_REQ_FLAG(REQ_F_ARM_LTIMEOUT_BIT),
530 	/* ->async_data allocated */
531 	REQ_F_ASYNC_DATA	= IO_REQ_FLAG(REQ_F_ASYNC_DATA_BIT),
532 	/* don't post CQEs while failing linked requests */
533 	REQ_F_SKIP_LINK_CQES	= IO_REQ_FLAG(REQ_F_SKIP_LINK_CQES_BIT),
534 	/* single poll may be active */
535 	REQ_F_SINGLE_POLL	= IO_REQ_FLAG(REQ_F_SINGLE_POLL_BIT),
536 	/* double poll may active */
537 	REQ_F_DOUBLE_POLL	= IO_REQ_FLAG(REQ_F_DOUBLE_POLL_BIT),
538 	/* fast poll multishot mode */
539 	REQ_F_APOLL_MULTISHOT	= IO_REQ_FLAG(REQ_F_APOLL_MULTISHOT_BIT),
540 	/* recvmsg special flag, clear EPOLLIN */
541 	REQ_F_CLEAR_POLLIN	= IO_REQ_FLAG(REQ_F_CLEAR_POLLIN_BIT),
542 	/* hashed into ->cancel_hash_locked, protected by ->uring_lock */
543 	REQ_F_HASH_LOCKED	= IO_REQ_FLAG(REQ_F_HASH_LOCKED_BIT),
544 	/* don't use lazy poll wake for this request */
545 	REQ_F_POLL_NO_LAZY	= IO_REQ_FLAG(REQ_F_POLL_NO_LAZY_BIT),
546 	/* cancel sequence is set and valid */
547 	REQ_F_CANCEL_SEQ	= IO_REQ_FLAG(REQ_F_CANCEL_SEQ_BIT),
548 	/* file is pollable */
549 	REQ_F_CAN_POLL		= IO_REQ_FLAG(REQ_F_CAN_POLL_BIT),
550 	/* buffer list was empty after selection of buffer */
551 	REQ_F_BL_EMPTY		= IO_REQ_FLAG(REQ_F_BL_EMPTY_BIT),
552 	/* don't recycle provided buffers for this request */
553 	REQ_F_BL_NO_RECYCLE	= IO_REQ_FLAG(REQ_F_BL_NO_RECYCLE_BIT),
554 	/* buffer ring head needs incrementing on put */
555 	REQ_F_BUFFERS_COMMIT	= IO_REQ_FLAG(REQ_F_BUFFERS_COMMIT_BIT),
556 };
557 
558 typedef void (*io_req_tw_func_t)(struct io_kiocb *req, struct io_tw_state *ts);
559 
560 struct io_task_work {
561 	struct llist_node		node;
562 	io_req_tw_func_t		func;
563 };
564 
565 struct io_cqe {
566 	__u64	user_data;
567 	__s32	res;
568 	/* fd initially, then cflags for completion */
569 	union {
570 		__u32	flags;
571 		int	fd;
572 	};
573 };
574 
575 /*
576  * Each request type overlays its private data structure on top of this one.
577  * They must not exceed this one in size.
578  */
579 struct io_cmd_data {
580 	struct file		*file;
581 	/* each command gets 56 bytes of data */
582 	__u8			data[56];
583 };
584 
io_kiocb_cmd_sz_check(size_t cmd_sz)585 static inline void io_kiocb_cmd_sz_check(size_t cmd_sz)
586 {
587 	BUILD_BUG_ON(cmd_sz > sizeof(struct io_cmd_data));
588 }
589 #define io_kiocb_to_cmd(req, cmd_type) ( \
590 	io_kiocb_cmd_sz_check(sizeof(cmd_type)) , \
591 	((cmd_type *)&(req)->cmd) \
592 )
593 #define cmd_to_io_kiocb(ptr)	((struct io_kiocb *) ptr)
594 
595 struct io_kiocb {
596 	union {
597 		/*
598 		 * NOTE! Each of the io_kiocb union members has the file pointer
599 		 * as the first entry in their struct definition. So you can
600 		 * access the file pointer through any of the sub-structs,
601 		 * or directly as just 'file' in this struct.
602 		 */
603 		struct file		*file;
604 		struct io_cmd_data	cmd;
605 	};
606 
607 	u8				opcode;
608 	/* polled IO has completed */
609 	u8				iopoll_completed;
610 	/*
611 	 * Can be either a fixed buffer index, or used with provided buffers.
612 	 * For the latter, before issue it points to the buffer group ID,
613 	 * and after selection it points to the buffer ID itself.
614 	 */
615 	u16				buf_index;
616 
617 	unsigned			nr_tw;
618 
619 	/* REQ_F_* flags */
620 	io_req_flags_t			flags;
621 
622 	struct io_cqe			cqe;
623 
624 	struct io_ring_ctx		*ctx;
625 	struct task_struct		*task;
626 
627 	union {
628 		/* store used ubuf, so we can prevent reloading */
629 		struct io_mapped_ubuf	*imu;
630 
631 		/* stores selected buf, valid IFF REQ_F_BUFFER_SELECTED is set */
632 		struct io_buffer	*kbuf;
633 
634 		/*
635 		 * stores buffer ID for ring provided buffers, valid IFF
636 		 * REQ_F_BUFFER_RING is set.
637 		 */
638 		struct io_buffer_list	*buf_list;
639 	};
640 
641 	union {
642 		/* used by request caches, completion batching and iopoll */
643 		struct io_wq_work_node	comp_list;
644 		/* cache ->apoll->events */
645 		__poll_t apoll_events;
646 	};
647 
648 	struct io_rsrc_node		*rsrc_node;
649 
650 	atomic_t			refs;
651 	atomic_t			poll_refs;
652 	struct io_task_work		io_task_work;
653 	/* for polled requests, i.e. IORING_OP_POLL_ADD and async armed poll */
654 	struct hlist_node		hash_node;
655 	/* internal polling, see IORING_FEAT_FAST_POLL */
656 	struct async_poll		*apoll;
657 	/* opcode allocated if it needs to store data for async defer */
658 	void				*async_data;
659 	/* linked requests, IFF REQ_F_HARDLINK or REQ_F_LINK are set */
660 	struct io_kiocb			*link;
661 	/* custom credentials, valid IFF REQ_F_CREDS is set */
662 	const struct cred		*creds;
663 	struct io_wq_work		work;
664 
665 	struct {
666 		u64			extra1;
667 		u64			extra2;
668 	} big_cqe;
669 };
670 
671 struct io_overflow_cqe {
672 	struct list_head list;
673 	struct io_uring_cqe cqe;
674 };
675 
676 #endif
677