1/*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1997 John S. Dyson.  All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 *    notice, this list of conditions and the following disclaimer.
11 * 2. John S. Dyson's name may not be used to endorse or promote products
12 *    derived from this software without specific prior written permission.
13 *
14 * DISCLAIMER:  This code isn't warranted to do anything useful.  Anything
15 * bad that happens because of using this software isn't the responsibility
16 * of the author.  This software is distributed AS-IS.
17 */
18
19#ifndef _SYS_AIO_H_
20#define	_SYS_AIO_H_
21
22#include <sys/types.h>
23#include <sys/signal.h>
24#ifdef _KERNEL
25#include <sys/queue.h>
26#include <sys/event.h>
27#include <sys/signalvar.h>
28#include <sys/uio.h>
29#endif
30
31/*
32 * Returned by aio_cancel:
33 */
34#define	AIO_CANCELED		0x1
35#define	AIO_NOTCANCELED		0x2
36#define	AIO_ALLDONE		0x3
37
38/*
39 * LIO opcodes
40 */
41#define	LIO_NOP			0x0
42#define LIO_WRITE		0x1
43#define	LIO_READ		0x2
44#if __BSD_VISIBLE
45#define	LIO_VECTORED		0x4
46#define	LIO_WRITEV		(LIO_WRITE | LIO_VECTORED)
47#define	LIO_READV		(LIO_READ | LIO_VECTORED)
48#endif
49#if defined(_KERNEL) || defined(_WANT_ALL_LIO_OPCODES)
50#define	LIO_SYNC		0x8
51#define	LIO_DSYNC		(0x10 | LIO_SYNC)
52#define	LIO_MLOCK		0x20
53#endif
54#if __BSD_VISIBLE
55#define	LIO_FOFFSET		0x40
56#endif
57
58/* aio_read2/aio_write2 flags */
59#if __BSD_VISIBLE
60#define	AIO_OP2_FOFFSET		0x00000001
61#define	AIO_OP2_VECTORED	0x00000002
62#endif
63
64/*
65 * LIO modes
66 */
67#define	LIO_NOWAIT		0x0
68#define	LIO_WAIT		0x1
69
70/*
71 * Maximum number of operations in a single lio_listio call
72 */
73#define	AIO_LISTIO_MAX		16
74
75#ifdef _KERNEL
76
77/* Default values of tunables for the AIO worker pool. */
78
79#ifndef MAX_AIO_PROCS
80#define MAX_AIO_PROCS		32
81#endif
82
83#ifndef TARGET_AIO_PROCS
84#define TARGET_AIO_PROCS	4
85#endif
86
87#ifndef AIOD_LIFETIME_DEFAULT
88#define AIOD_LIFETIME_DEFAULT	(30 * hz)
89#endif
90
91#endif
92
93/*
94 * Private members for aiocb -- don't access
95 * directly.
96 */
97struct __aiocb_private {
98	long	status;
99	long	error;
100	void	*kernelinfo;
101};
102
103/*
104 * I/O control block
105 */
106typedef struct aiocb {
107	int	aio_fildes;		/* File descriptor */
108	off_t	aio_offset;		/* File offset for I/O */
109	volatile void *aio_buf;		/* I/O buffer in process space */
110	size_t	aio_nbytes;		/* Number of bytes for I/O */
111	int	__spare__[2];
112	void	*__spare2__;
113	int	aio_lio_opcode;		/* LIO opcode */
114	int	aio_reqprio;		/* Request priority -- ignored */
115	struct	__aiocb_private	_aiocb_private;
116	struct	sigevent aio_sigevent;	/* Signal to deliver */
117} aiocb_t;
118
119#define	aio_iov	aio_buf			/* I/O scatter/gather list */
120#define	aio_iovcnt	aio_nbytes	/* Length of aio_iov */
121
122#ifdef _KERNEL
123
124typedef void aio_cancel_fn_t(struct kaiocb *);
125typedef void aio_handle_fn_t(struct kaiocb *);
126
127/*
128 * Kernel version of an I/O control block.
129 *
130 * Locking key:
131 * * - need not protected
132 * a - locked by kaioinfo lock
133 * b - locked by backend lock
134 * c - locked by aio_job_mtx
135 */
136struct kaiocb {
137	TAILQ_ENTRY(kaiocb) list;	/* (b) backend-specific list of jobs */
138	TAILQ_ENTRY(kaiocb) plist;	/* (a) lists of pending / done jobs */
139	TAILQ_ENTRY(kaiocb) allist;	/* (a) list of all jobs in proc */
140	int	jobflags;		/* (a) job flags */
141	int	ioflags;		/* (*) io flags */
142	int	inblock;		/* (*) input blocks */
143	int	outblock;		/* (*) output blocks */
144	int	msgsnd;			/* (*) messages sent */
145	int	msgrcv;			/* (*) messages received */
146	struct	proc *userproc;		/* (*) user process */
147	struct	ucred *cred;		/* (*) active credential when created */
148	struct	file *fd_file;		/* (*) pointer to file structure */
149	struct	aioliojob *lio;		/* (*) optional lio job */
150	struct	aiocb *ujob;		/* (*) pointer in userspace of aiocb */
151	struct	knlist klist;		/* (a) list of knotes */
152	struct	aiocb uaiocb;		/* (*) copy of user I/O control block */
153	struct	uio uio;		/* (*) storage for non-vectored uio */
154	struct	iovec iov[1];		/* (*) storage for non-vectored uio */
155	struct	uio *uiop;		/* (*) Possibly malloced uio */
156	ksiginfo_t ksi;			/* (a) realtime signal info */
157	uint64_t seqno;			/* (*) job number */
158	aio_cancel_fn_t *cancel_fn;	/* (a) backend cancel function */
159	aio_handle_fn_t *handle_fn;	/* (c) backend handle function */
160	union {				/* Backend-specific data fields */
161		struct {		/* BIO backend */
162			volatile u_int nbio; /* Number of remaining bios */
163			int	error;	/* Worst error of all bios */
164			long	nbytes;	/* Bytes completed so far */
165		};
166		struct {		/* fsync() requests */
167			int	pending; /* (a) number of pending I/O */
168		};
169		struct {		/* socket backend */
170			void	*backend1;
171			long	backend3;
172			int	backend4;
173		};
174	};
175};
176
177struct socket;
178struct sockbuf;
179
180/*
181 * AIO backends should permit cancellation of queued requests waiting to
182 * be serviced by installing a cancel routine while the request is
183 * queued.  The cancellation routine should dequeue the request if
184 * necessary and cancel it.  Care must be used to handle races between
185 * queueing and dequeueing requests and cancellation.
186 *
187 * When queueing a request somewhere such that it can be cancelled, the
188 * caller should:
189 *
190 *  1) Acquire lock that protects the associated queue.
191 *  2) Call aio_set_cancel_function() to install the cancel routine.
192 *  3) If that fails, the request has a pending cancel and should be
193 *     cancelled via aio_cancel().
194 *  4) Queue the request.
195 *
196 * When dequeueing a request to service it or hand it off to somewhere else,
197 * the caller should:
198 *
199 *  1) Acquire the lock that protects the associated queue.
200 *  2) Dequeue the request.
201 *  3) Call aio_clear_cancel_function() to clear the cancel routine.
202 *  4) If that fails, the cancel routine is about to be called.  The
203 *     caller should ignore the request.
204 *
205 * The cancel routine should:
206 *
207 *  1) Acquire the lock that protects the associated queue.
208 *  2) Call aio_cancel_cleared() to determine if the request is already
209 *     dequeued due to a race with dequeueing thread.
210 *  3) If that fails, dequeue the request.
211 *  4) Cancel the request via aio_cancel().
212 */
213
214bool	aio_cancel_cleared(struct kaiocb *job);
215void	aio_cancel(struct kaiocb *job);
216bool	aio_clear_cancel_function(struct kaiocb *job);
217void	aio_complete(struct kaiocb *job, long status, int error);
218void	aio_schedule(struct kaiocb *job, aio_handle_fn_t *func);
219bool	aio_set_cancel_function(struct kaiocb *job, aio_cancel_fn_t *func);
220void	aio_switch_vmspace(struct kaiocb *job);
221
222#else /* !_KERNEL */
223
224struct timespec;
225
226__BEGIN_DECLS
227/*
228 * Asynchronously read from a file
229 */
230int	aio_read(struct aiocb *);
231#if __BSD_VISIBLE
232int	aio_readv(struct aiocb *);
233#endif
234
235/*
236 * Asynchronously write to file
237 */
238int	aio_write(struct aiocb *);
239#if __BSD_VISIBLE
240int	aio_writev(struct aiocb *);
241#endif
242
243/*
244 * List I/O Asynchronously/synchronously read/write to/from file
245 *	"lio_mode" specifies whether or not the I/O is synchronous.
246 *	"acb_list" is an array of "nacb_listent" I/O control blocks.
247 *	when all I/Os are complete, the optional signal "sig" is sent.
248 */
249int	lio_listio(int, struct aiocb *__restrict const *__restrict, int,
250    struct sigevent *);
251
252/*
253 * Get completion status
254 *	returns EINPROGRESS until I/O is complete.
255 *	this routine does not block.
256 */
257int	aio_error(const struct aiocb *);
258
259/*
260 * Finish up I/O, releasing I/O resources and returns the value
261 *	that would have been associated with a synchronous I/O request.
262 *	This routine must be called once and only once for each
263 *	I/O control block who has had I/O associated with it.
264 */
265ssize_t	aio_return(struct aiocb *);
266
267/*
268 * Cancel I/O
269 */
270int	aio_cancel(int, struct aiocb *);
271
272/*
273 * Suspend until all specified I/O or timeout is complete.
274 */
275int	aio_suspend(const struct aiocb * const[], int, const struct timespec *);
276
277/*
278 * Asynchronous mlock
279 */
280int	aio_mlock(struct aiocb *);
281
282#if __BSD_VISIBLE
283ssize_t	aio_waitcomplete(struct aiocb **, struct timespec *);
284int	aio_read2(struct aiocb *, int);
285int	aio_write2(struct aiocb *, int);
286#endif
287
288int	aio_fsync(int op, struct aiocb *aiocbp);
289__END_DECLS
290
291#endif /* !_KERNEL */
292
293#endif /* !_SYS_AIO_H_ */
294