xref: /netbsd-src/sys/kern/sys_aio.c (revision 0920b4f20b78ab1ccd9f2312fbe10deaf000cbf3)
1 /*	$NetBSD: sys_aio.c,v 1.6 2007/07/11 00:40:42 rmind Exp $	*/
2 
3 /*
4  * Copyright (c) 2007, Mindaugas Rasiukevicius <rmind at NetBSD org>
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. Redistributions in binary form must reproduce the above copyright
12  *    notice, this list of conditions and the following disclaimer in the
13  *    documentation and/or other materials provided with the distribution.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
16  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
17  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
18  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
19  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
20  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
21  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
22  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
23  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
24  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
25  * POSSIBILITY OF SUCH DAMAGE.
26  */
27 
28 /*
29  * TODO:
30  *   1. Additional work for VCHR and maybe VBLK devices.
31  *   2. Consider making the job-finding O(n) per one file descriptor.
32  */
33 
34 #include <sys/cdefs.h>
35 __KERNEL_RCSID(0, "$NetBSD: sys_aio.c,v 1.6 2007/07/11 00:40:42 rmind Exp $");
36 
37 #include "opt_ddb.h"
38 
39 #include <sys/param.h>
40 #include <sys/condvar.h>
41 #include <sys/file.h>
42 #include <sys/filedesc.h>
43 #include <sys/kernel.h>
44 #include <sys/kmem.h>
45 #include <sys/lwp.h>
46 #include <sys/mutex.h>
47 #include <sys/pool.h>
48 #include <sys/proc.h>
49 #include <sys/queue.h>
50 #include <sys/signal.h>
51 #include <sys/signalvar.h>
52 #include <sys/syscallargs.h>
53 #include <sys/sysctl.h>
54 #include <sys/systm.h>
55 #include <sys/types.h>
56 #include <sys/vnode.h>
57 
58 #include <uvm/uvm_extern.h>
59 
60 /*
61  * System-wide limits and counter of AIO operations.
62  * XXXSMP: We should spin-lock it, or modify atomically.
63  */
64 static u_int aio_listio_max = AIO_LISTIO_MAX;
65 static u_int aio_max = AIO_MAX;
66 static u_int aio_jobs_count;
67 
68 static struct pool aio_job_pool;
69 static struct pool aio_lio_pool;
70 
71 /* Prototypes */
72 void aio_worker(void *);
73 static void aio_process(struct aio_job *);
74 static void aio_sendsig(struct proc *, struct sigevent *);
75 static int aio_enqueue_job(int, void *, struct lio_req *);
76 
77 /*
78  * Initialize the AIO system.
79  */
80 void
81 aio_sysinit(void)
82 {
83 
84 	pool_init(&aio_job_pool, sizeof(struct aio_job), 0, 0, 0,
85 	    "aio_jobs_pool", &pool_allocator_nointr, IPL_NONE);
86 	pool_init(&aio_lio_pool, sizeof(struct lio_req), 0, 0, 0,
87 	    "aio_lio_pool", &pool_allocator_nointr, IPL_NONE);
88 }
89 
90 /*
91  * Initialize Asynchronous I/O data structures for the process.
92  */
93 int
94 aio_init(struct proc *p)
95 {
96 	struct aioproc *aio;
97 	struct lwp *l;
98 	bool inmem;
99 	vaddr_t uaddr;
100 
101 	/* Allocate and initialize AIO structure */
102 	aio = kmem_zalloc(sizeof(struct aioproc), KM_NOSLEEP);
103 	if (aio == NULL)
104 		return EAGAIN;
105 
106 	/* Initialize queue and their synchronization structures */
107 	mutex_init(&aio->aio_mtx, MUTEX_DEFAULT, IPL_NONE);
108 	cv_init(&aio->aio_worker_cv, "aiowork");
109 	cv_init(&aio->done_cv, "aiodone");
110 	TAILQ_INIT(&aio->jobs_queue);
111 
112 	/*
113 	 * Create an AIO worker thread.
114 	 * XXX: Currently, AIO thread is not protected against user's actions.
115 	 */
116 	inmem = uvm_uarea_alloc(&uaddr);
117 	if (uaddr == 0) {
118 		aio_exit(p, aio);
119 		return EAGAIN;
120 	}
121 	if (newlwp(curlwp, p, uaddr, inmem, 0, NULL, 0,
122 	    aio_worker, NULL, &l)) {
123 		uvm_uarea_free(uaddr);
124 		aio_exit(p, aio);
125 		return EAGAIN;
126 	}
127 
128 	/* Recheck if we are really first */
129 	mutex_enter(&p->p_mutex);
130 	if (p->p_aio) {
131 		mutex_exit(&p->p_mutex);
132 		aio_exit(p, aio);
133 		lwp_exit(l);
134 		return 0;
135 	}
136 	p->p_aio = aio;
137 	mutex_exit(&p->p_mutex);
138 
139 	/* Complete the initialization of thread, and run it */
140 	mutex_enter(&p->p_smutex);
141 	aio->aio_worker = l;
142 	p->p_nrlwps++;
143 	lwp_lock(l);
144 	l->l_stat = LSRUN;
145 	l->l_usrpri = PUSER - 1; /* XXX */
146 	sched_enqueue(l, false);
147 	lwp_unlock(l);
148 	mutex_exit(&p->p_smutex);
149 
150 	return 0;
151 }
152 
153 /*
154  * Exit of Asynchronous I/O subsystem of process.
155  */
156 void
157 aio_exit(struct proc *p, struct aioproc *aio)
158 {
159 	struct aio_job *a_job;
160 
161 	if (aio == NULL)
162 		return;
163 
164 	/* Free AIO queue */
165 	while (!TAILQ_EMPTY(&aio->jobs_queue)) {
166 		a_job = TAILQ_FIRST(&aio->jobs_queue);
167 		TAILQ_REMOVE(&aio->jobs_queue, a_job, list);
168 		pool_put(&aio_job_pool, a_job);
169 		aio_jobs_count--; /* XXXSMP */
170 	}
171 
172 	/* Destroy and free the entire AIO data structure */
173 	cv_destroy(&aio->aio_worker_cv);
174 	cv_destroy(&aio->done_cv);
175 	mutex_destroy(&aio->aio_mtx);
176 	kmem_free(aio, sizeof(struct aioproc));
177 }
178 
179 /*
180  * AIO worker thread and processor.
181  */
182 void
183 aio_worker(void *arg)
184 {
185 	struct proc *p = curlwp->l_proc;
186 	struct aioproc *aio = p->p_aio;
187 	struct aio_job *a_job;
188 	struct lio_req *lio;
189 	sigset_t oss, nss;
190 	int error, refcnt;
191 
192 	/*
193 	 * Make an empty signal mask, so it
194 	 * handles only SIGKILL and SIGSTOP.
195 	 */
196 	sigfillset(&nss);
197 	mutex_enter(&p->p_smutex);
198 	error = sigprocmask1(curlwp, SIG_SETMASK, &nss, &oss);
199 	mutex_exit(&p->p_smutex);
200 	KASSERT(error == 0);
201 
202 	for (;;) {
203 		/*
204 		 * Loop for each job in the queue.  If there
205 		 * are no jobs then sleep.
206 		 */
207 		mutex_enter(&aio->aio_mtx);
208 		while ((a_job = TAILQ_FIRST(&aio->jobs_queue)) == NULL) {
209 			if (cv_wait_sig(&aio->aio_worker_cv, &aio->aio_mtx)) {
210 				/*
211 				 * Thread was interrupted - check for
212 				 * pending exit or suspend.
213 				 */
214 				mutex_exit(&aio->aio_mtx);
215 				lwp_userret(curlwp);
216 				mutex_enter(&aio->aio_mtx);
217 			}
218 		}
219 
220 		/* Take the job from the queue */
221 		aio->curjob = a_job;
222 		TAILQ_REMOVE(&aio->jobs_queue, a_job, list);
223 
224 		aio_jobs_count--; /* XXXSMP */
225 		aio->jobs_count--;
226 
227 		mutex_exit(&aio->aio_mtx);
228 
229 		/* Process an AIO operation */
230 		aio_process(a_job);
231 
232 		/* Copy data structure back to the user-space */
233 		(void)copyout(&a_job->aiocbp, a_job->aiocb_uptr,
234 		    sizeof(struct aiocb));
235 
236 		mutex_enter(&aio->aio_mtx);
237 		aio->curjob = NULL;
238 
239 		/* Decrease a reference counter, if there is a LIO structure */
240 		lio = a_job->lio;
241 		refcnt = (lio != NULL ? --lio->refcnt : -1);
242 
243 		/* Notify all suspenders */
244 		cv_broadcast(&aio->done_cv);
245 		mutex_exit(&aio->aio_mtx);
246 
247 		/* Send a signal, if any */
248 		aio_sendsig(p, &a_job->aiocbp.aio_sigevent);
249 
250 		/* Destroy the LIO structure */
251 		if (refcnt == 0) {
252 			aio_sendsig(p, &lio->sig);
253 			pool_put(&aio_lio_pool, lio);
254 		}
255 
256 		/* Destroy the the job */
257 		pool_put(&aio_job_pool, a_job);
258 	}
259 
260 	/* NOTREACHED */
261 }
262 
263 static void
264 aio_process(struct aio_job *a_job)
265 {
266 	struct proc *p = curlwp->l_proc;
267 	struct aiocb *aiocbp = &a_job->aiocbp;
268 	struct file *fp;
269 	struct filedesc	*fdp = p->p_fd;
270 	int fd = aiocbp->aio_fildes;
271 	int error = 0;
272 
273 	KASSERT(fdp != NULL);
274 	KASSERT(a_job->aio_op != 0);
275 
276 	if ((a_job->aio_op & (AIO_READ | AIO_WRITE)) != 0) {
277 		struct iovec aiov;
278 		struct uio auio;
279 
280 		if (aiocbp->aio_nbytes > SSIZE_MAX) {
281 			error = EINVAL;
282 			goto done;
283 		}
284 
285 		fp = fd_getfile(fdp, fd);
286 		if (fp == NULL) {
287 			error = EBADF;
288 			goto done;
289 		}
290 
291 		aiov.iov_base = (void *)(uintptr_t)aiocbp->aio_buf;
292 		aiov.iov_len = aiocbp->aio_nbytes;
293 		auio.uio_iov = &aiov;
294 		auio.uio_iovcnt = 1;
295 		auio.uio_resid = aiocbp->aio_nbytes;
296 		auio.uio_vmspace = p->p_vmspace;
297 
298 		FILE_USE(fp);
299 		if (a_job->aio_op & AIO_READ) {
300 			/*
301 			 * Perform a Read operation
302 			 */
303 			KASSERT((a_job->aio_op & AIO_WRITE) == 0);
304 
305 			if ((fp->f_flag & FREAD) == 0) {
306 				FILE_UNUSE(fp, curlwp);
307 				error = EBADF;
308 				goto done;
309 			}
310 			auio.uio_rw = UIO_READ;
311 			error = (*fp->f_ops->fo_read)(fp, &aiocbp->aio_offset,
312 			    &auio, fp->f_cred, FOF_UPDATE_OFFSET);
313 		} else {
314 			/*
315 			 * Perform a Write operation
316 			 */
317 			KASSERT(a_job->aio_op & AIO_WRITE);
318 
319 			if ((fp->f_flag & FWRITE) == 0) {
320 				FILE_UNUSE(fp, curlwp);
321 				error = EBADF;
322 				goto done;
323 			}
324 			auio.uio_rw = UIO_WRITE;
325 			error = (*fp->f_ops->fo_write)(fp, &aiocbp->aio_offset,
326 			    &auio, fp->f_cred, FOF_UPDATE_OFFSET);
327 		}
328 		FILE_UNUSE(fp, curlwp);
329 
330 		/* Store the result value */
331 		a_job->aiocbp.aio_nbytes -= auio.uio_resid;
332 		a_job->aiocbp._retval = (error == 0) ?
333 		    a_job->aiocbp.aio_nbytes : -1;
334 
335 	} else if ((a_job->aio_op & (AIO_SYNC | AIO_DSYNC)) != 0) {
336 		/*
337 		 * Perform a file Sync operation
338 		 */
339 		struct vnode *vp;
340 
341 		if ((error = getvnode(fdp, fd, &fp)) != 0)
342 			goto done;
343 
344 		if ((fp->f_flag & FWRITE) == 0) {
345 			FILE_UNUSE(fp, curlwp);
346 			error = EBADF;
347 			goto done;
348 		}
349 
350 		vp = (struct vnode *)fp->f_data;
351 		vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
352 		if (a_job->aio_op & AIO_DSYNC) {
353 			error = VOP_FSYNC(vp, fp->f_cred,
354 			    FSYNC_WAIT | FSYNC_DATAONLY, 0, 0, curlwp);
355 		} else if (a_job->aio_op & AIO_SYNC) {
356 			error = VOP_FSYNC(vp, fp->f_cred,
357 			    FSYNC_WAIT, 0, 0, curlwp);
358 			if (error == 0 && bioops.io_fsync != NULL &&
359 			    vp->v_mount &&
360 			    (vp->v_mount->mnt_flag & MNT_SOFTDEP))
361 			    (*bioops.io_fsync)(vp, 0);
362 		}
363 		VOP_UNLOCK(vp, 0);
364 		FILE_UNUSE(fp, curlwp);
365 
366 		/* Store the result value */
367 		a_job->aiocbp._retval = (error == 0) ? 0 : -1;
368 
369 	} else
370 		panic("aio_process: invalid operation code\n");
371 
372 done:
373 	/* Job is done, set the error, if any */
374 	a_job->aiocbp._errno = error;
375 	a_job->aiocbp._state = JOB_DONE;
376 }
377 
378 /*
379  * Send AIO signal.
380  */
381 static void
382 aio_sendsig(struct proc *p, struct sigevent *sig)
383 {
384 	ksiginfo_t ksi;
385 
386 	if (sig->sigev_signo == 0 || sig->sigev_notify == SIGEV_NONE)
387 		return;
388 
389 	KSI_INIT(&ksi);
390 	ksi.ksi_signo = sig->sigev_signo;
391 	ksi.ksi_code = SI_ASYNCIO;
392 	ksi.ksi_value = sig->sigev_value;
393 	mutex_enter(&proclist_mutex);
394 	kpsignal(p, &ksi, NULL);
395 	mutex_exit(&proclist_mutex);
396 }
397 
398 /*
399  * Enqueue the job.
400  */
401 static int
402 aio_enqueue_job(int op, void *aiocb_uptr, struct lio_req *lio)
403 {
404 	struct proc *p = curlwp->l_proc;
405 	struct aioproc *aio;
406 	struct aio_job *a_job;
407 	struct aiocb aiocbp;
408 	struct sigevent *sig;
409 	int error;
410 
411 	/* Check for the limit */
412 	if (aio_jobs_count + 1 > aio_max) /* XXXSMP */
413 		return EAGAIN;
414 
415 	/* Get the data structure from user-space */
416 	error = copyin(aiocb_uptr, &aiocbp, sizeof(struct aiocb));
417 	if (error)
418 		return error;
419 
420 	/* Check if signal is set, and validate it */
421 	sig = &aiocbp.aio_sigevent;
422 	if (sig->sigev_signo < 0 || sig->sigev_signo >= NSIG ||
423 	    sig->sigev_notify < SIGEV_NONE || sig->sigev_notify > SIGEV_SA)
424 		return EINVAL;
425 
426 	/* Buffer and byte count */
427 	if (((AIO_SYNC | AIO_DSYNC) & op) == 0)
428 		if (aiocbp.aio_buf == NULL || aiocbp.aio_nbytes > SSIZE_MAX)
429 			return EINVAL;
430 
431 	/* Check the opcode, if LIO_NOP - simply ignore */
432 	if (op == AIO_LIO) {
433 		KASSERT(lio != NULL);
434 		if (aiocbp.aio_lio_opcode == LIO_WRITE)
435 			op = AIO_WRITE;
436 		else if (aiocbp.aio_lio_opcode == LIO_READ)
437 			op = AIO_READ;
438 		else
439 			return (aiocbp.aio_lio_opcode == LIO_NOP) ? 0 : EINVAL;
440 	} else {
441 		KASSERT(lio == NULL);
442 	}
443 
444 	/*
445 	 * Look for already existing job.  If found - the job is in-progress.
446 	 * According to POSIX this is invalid, so return the error.
447 	 */
448 	aio = p->p_aio;
449 	if (aio) {
450 		mutex_enter(&aio->aio_mtx);
451 		if (aio->curjob) {
452 			a_job = aio->curjob;
453 			if (a_job->aiocb_uptr == aiocb_uptr) {
454 				mutex_exit(&aio->aio_mtx);
455 				return EINVAL;
456 			}
457 		}
458 		TAILQ_FOREACH(a_job, &aio->jobs_queue, list) {
459 			if (a_job->aiocb_uptr != aiocb_uptr)
460 				continue;
461 			mutex_exit(&aio->aio_mtx);
462 			return EINVAL;
463 		}
464 		mutex_exit(&aio->aio_mtx);
465 	}
466 
467 	/*
468 	 * Check if AIO structure is initialized, if not - initialize it.
469 	 * In LIO case, we did that already.  We will recheck this with
470 	 * the lock in aio_init().
471 	 */
472 	if (lio == NULL && p->p_aio == NULL)
473 		if (aio_init(p))
474 			return EAGAIN;
475 	aio = p->p_aio;
476 
477 	/*
478 	 * Set the state with errno, and copy data
479 	 * structure back to the user-space.
480 	 */
481 	aiocbp._state = JOB_WIP;
482 	aiocbp._errno = EINPROGRESS;
483 	aiocbp._retval = -1;
484 	error = copyout(&aiocbp, aiocb_uptr, sizeof(struct aiocb));
485 	if (error)
486 		return error;
487 
488 	/* Allocate and initialize a new AIO job */
489 	a_job = pool_get(&aio_job_pool, PR_WAITOK);
490 	memset(a_job, 0, sizeof(struct aio_job));
491 
492 	/*
493 	 * Set the data.
494 	 * Store the user-space pointer for searching.  Since we
495 	 * are storing only per proc pointers - it is safe.
496 	 */
497 	memcpy(&a_job->aiocbp, &aiocbp, sizeof(struct aiocb));
498 	a_job->aiocb_uptr = aiocb_uptr;
499 	a_job->aio_op |= op;
500 	a_job->lio = lio;
501 
502 	/*
503 	 * Add the job to the queue, update the counters, and
504 	 * notify the AIO worker thread to handle the job.
505 	 */
506 	mutex_enter(&aio->aio_mtx);
507 
508 	/* Fail, if the limit was reached */
509 	if (aio->jobs_count >= aio_listio_max) {
510 		mutex_exit(&aio->aio_mtx);
511 		pool_put(&aio_job_pool, a_job);
512 		return EAGAIN;
513 	}
514 
515 	TAILQ_INSERT_TAIL(&aio->jobs_queue, a_job, list);
516 	aio_jobs_count++; /* XXXSMP */
517 	aio->jobs_count++;
518 	if (lio)
519 		lio->refcnt++;
520 	cv_signal(&aio->aio_worker_cv);
521 
522 	mutex_exit(&aio->aio_mtx);
523 
524 	/*
525 	 * One would handle the errors only with aio_error() function.
526 	 * This way is appropriate according to POSIX.
527 	 */
528 	return 0;
529 }
530 
531 /*
532  * Syscall functions.
533  */
534 
535 int
536 sys_aio_cancel(struct lwp *l, void *v, register_t *retval)
537 {
538 	struct sys_aio_cancel_args /* {
539 		syscallarg(int) fildes;
540 		syscallarg(struct aiocb *) aiocbp;
541 	} */ *uap = v;
542 	struct proc *p = l->l_proc;
543 	struct aioproc *aio;
544 	struct aio_job *a_job;
545 	struct aiocb *aiocbp_ptr;
546 	struct lio_req *lio;
547 	struct filedesc	*fdp = p->p_fd;
548 	unsigned int cn, errcnt, fildes;
549 
550 	TAILQ_HEAD(, aio_job) tmp_jobs_list;
551 
552 	/* Check for invalid file descriptor */
553 	fildes = (unsigned int)SCARG(uap, fildes);
554 	if (fildes >= fdp->fd_nfiles || fdp->fd_ofiles[fildes] == NULL)
555 		return EBADF;
556 
557 	/* Check if AIO structure is initialized */
558 	if (p->p_aio == NULL) {
559 		*retval = AIO_NOTCANCELED;
560 		return 0;
561 	}
562 
563 	aio = p->p_aio;
564 	aiocbp_ptr = (struct aiocb *)SCARG(uap, aiocbp);
565 
566 	mutex_enter(&aio->aio_mtx);
567 
568 	/* Cancel the jobs, and remove them from the queue */
569 	cn = 0;
570 	TAILQ_INIT(&tmp_jobs_list);
571 	TAILQ_FOREACH(a_job, &aio->jobs_queue, list) {
572 		if (aiocbp_ptr) {
573 			if (aiocbp_ptr != a_job->aiocb_uptr)
574 				continue;
575 			if (fildes != a_job->aiocbp.aio_fildes) {
576 				mutex_exit(&aio->aio_mtx);
577 				return EBADF;
578 			}
579 		} else if (a_job->aiocbp.aio_fildes != fildes)
580 			continue;
581 
582 		TAILQ_REMOVE(&aio->jobs_queue, a_job, list);
583 		TAILQ_INSERT_TAIL(&tmp_jobs_list, a_job, list);
584 
585 		/* Decrease the counters */
586 		aio_jobs_count--; /* XXXSMP */
587 		aio->jobs_count--;
588 		lio = a_job->lio;
589 		if (lio != NULL && --lio->refcnt != 0)
590 			a_job->lio = NULL;
591 
592 		cn++;
593 		if (aiocbp_ptr)
594 			break;
595 	}
596 
597 	/* There are canceled jobs */
598 	if (cn)
599 		*retval = AIO_CANCELED;
600 
601 	/* We cannot cancel current job */
602 	a_job = aio->curjob;
603 	if (a_job && ((a_job->aiocbp.aio_fildes == fildes) ||
604 	    (a_job->aiocb_uptr == aiocbp_ptr)))
605 		*retval = AIO_NOTCANCELED;
606 
607 	mutex_exit(&aio->aio_mtx);
608 
609 	/* Free the jobs after the lock */
610 	errcnt = 0;
611 	while (!TAILQ_EMPTY(&tmp_jobs_list)) {
612 		a_job = TAILQ_FIRST(&tmp_jobs_list);
613 		TAILQ_REMOVE(&tmp_jobs_list, a_job, list);
614 		/* Set the errno and copy structures back to the user-space */
615 		a_job->aiocbp._errno = ECANCELED;
616 		a_job->aiocbp._state = JOB_DONE;
617 		if (copyout(&a_job->aiocbp, a_job->aiocb_uptr,
618 		    sizeof(struct aiocb)))
619 			errcnt++;
620 		/* Send a signal if any */
621 		aio_sendsig(p, &a_job->aiocbp.aio_sigevent);
622 		if (a_job->lio) {
623 			lio = a_job->lio;
624 			aio_sendsig(p, &lio->sig);
625 			pool_put(&aio_lio_pool, lio);
626 		}
627 		pool_put(&aio_job_pool, a_job);
628 	}
629 
630 	if (errcnt)
631 		return EFAULT;
632 
633 	/* Set a correct return value */
634 	if (*retval == 0)
635 		*retval = AIO_ALLDONE;
636 
637 	return 0;
638 }
639 
640 int
641 sys_aio_error(struct lwp *l, void *v, register_t *retval)
642 {
643 	struct sys_aio_error_args /* {
644 		syscallarg(const struct aiocb *) aiocbp;
645 	} */ *uap = v;
646 	struct proc *p = l->l_proc;
647 	struct aioproc *aio = p->p_aio;
648 	struct aiocb aiocbp;
649 	int error;
650 
651 	if (aio == NULL)
652 		return EINVAL;
653 
654 	error = copyin(SCARG(uap, aiocbp), &aiocbp, sizeof(struct aiocb));
655 	if (error)
656 		return error;
657 
658 	if (aiocbp._state == JOB_NONE)
659 		return EINVAL;
660 
661 	*retval = aiocbp._errno;
662 
663 	return 0;
664 }
665 
666 int
667 sys_aio_fsync(struct lwp *l, void *v, register_t *retval)
668 {
669 	struct sys_aio_fsync_args /* {
670 		syscallarg(int) op;
671 		syscallarg(struct aiocb *) aiocbp;
672 	} */ *uap = v;
673 	int op = SCARG(uap, op);
674 
675 	if ((op != O_DSYNC) && (op != O_SYNC))
676 		return EINVAL;
677 
678 	op = O_DSYNC ? AIO_DSYNC : AIO_SYNC;
679 
680 	return aio_enqueue_job(op, SCARG(uap, aiocbp), NULL);
681 }
682 
683 int
684 sys_aio_read(struct lwp *l, void *v, register_t *retval)
685 {
686 	struct sys_aio_read_args /* {
687 		syscallarg(struct aiocb *) aiocbp;
688 	} */ *uap = v;
689 
690 	return aio_enqueue_job(AIO_READ, SCARG(uap, aiocbp), NULL);
691 }
692 
693 int
694 sys_aio_return(struct lwp *l, void *v, register_t *retval)
695 {
696 	struct sys_aio_return_args /* {
697 		syscallarg(struct aiocb *) aiocbp;
698 	} */ *uap = v;
699 	struct proc *p = l->l_proc;
700 	struct aioproc *aio = p->p_aio;
701 	struct aiocb aiocbp;
702 	int error;
703 
704 	if (aio == NULL)
705 		return EINVAL;
706 
707 	error = copyin(SCARG(uap, aiocbp), &aiocbp, sizeof(struct aiocb));
708 	if (error)
709 		return error;
710 
711 	if (aiocbp._errno == EINPROGRESS || aiocbp._state != JOB_DONE)
712 		return EINVAL;
713 
714 	*retval = aiocbp._retval;
715 
716 	/* Reset the internal variables */
717 	aiocbp._errno = 0;
718 	aiocbp._retval = -1;
719 	aiocbp._state = JOB_NONE;
720 	error = copyout(&aiocbp, SCARG(uap, aiocbp), sizeof(struct aiocb));
721 
722 	return error;
723 }
724 
725 int
726 sys_aio_suspend(struct lwp *l, void *v, register_t *retval)
727 {
728 	struct sys_aio_suspend_args /* {
729 		syscallarg(const struct aiocb *const[]) list;
730 		syscallarg(int) nent;
731 		syscallarg(const struct timespec *) timeout;
732 	} */ *uap = v;
733 	struct proc *p = l->l_proc;
734 	struct aioproc *aio;
735 	struct aio_job *a_job;
736 	struct aiocb **aiocbp_list;
737 	struct timespec ts;
738 	int i, error, nent, timo;
739 
740 	if (p->p_aio == NULL)
741 		return EAGAIN;
742 	aio = p->p_aio;
743 
744 	nent = SCARG(uap, nent);
745 	if (nent <= 0 || nent > aio_listio_max)
746 		return EAGAIN;
747 
748 	if (SCARG(uap, timeout)) {
749 		/* Convert timespec to ticks */
750 		error = copyin(SCARG(uap, timeout), &ts,
751 		    sizeof(struct timespec));
752 		if (error)
753 			return error;
754 		timo = mstohz((ts.tv_sec * 1000) + (ts.tv_nsec / 1000000));
755 		if (timo == 0 && ts.tv_sec == 0 && ts.tv_nsec > 0)
756 			timo = 1;
757 		if (timo <= 0)
758 			return EAGAIN;
759 	} else
760 		timo = 0;
761 
762 	/* Get the list from user-space */
763 	aiocbp_list = kmem_zalloc(nent * sizeof(struct aio_job), KM_SLEEP);
764 	error = copyin(SCARG(uap, list), aiocbp_list,
765 	    nent * sizeof(struct aiocb));
766 	if (error) {
767 		kmem_free(aiocbp_list, nent * sizeof(struct aio_job));
768 		return error;
769 	}
770 
771 	mutex_enter(&aio->aio_mtx);
772 	for (;;) {
773 
774 		for (i = 0; i < nent; i++) {
775 
776 			/* Skip NULL entries */
777 			if (aiocbp_list[i] == NULL)
778 				continue;
779 
780 			/* Skip current job */
781 			if (aio->curjob) {
782 				a_job = aio->curjob;
783 				if (a_job->aiocb_uptr == aiocbp_list[i])
784 					continue;
785 			}
786 
787 			/* Look for a job in the queue */
788 			TAILQ_FOREACH(a_job, &aio->jobs_queue, list)
789 				if (a_job->aiocb_uptr == aiocbp_list[i])
790 					break;
791 
792 			if (a_job == NULL) {
793 				struct aiocb aiocbp;
794 
795 				mutex_exit(&aio->aio_mtx);
796 
797 				error = copyin(aiocbp_list[i], &aiocbp,
798 				    sizeof(struct aiocb));
799 				if (error == 0 && aiocbp._state != JOB_DONE) {
800 					mutex_enter(&aio->aio_mtx);
801 					continue;
802 				}
803 
804 				kmem_free(aiocbp_list,
805 				    nent * sizeof(struct aio_job));
806 				return error;
807 			}
808 		}
809 
810 		/* Wait for a signal or when timeout occurs */
811 		error = cv_timedwait_sig(&aio->done_cv, &aio->aio_mtx, timo);
812 		if (error) {
813 			if (error == EWOULDBLOCK)
814 				error = EAGAIN;
815 			break;
816 		}
817 	}
818 	mutex_exit(&aio->aio_mtx);
819 
820 	kmem_free(aiocbp_list, nent * sizeof(struct aio_job));
821 	return error;
822 }
823 
824 int
825 sys_aio_write(struct lwp *l, void *v, register_t *retval)
826 {
827 	struct sys_aio_write_args /* {
828 		syscallarg(struct aiocb *) aiocbp;
829 	} */ *uap = v;
830 
831 	return aio_enqueue_job(AIO_WRITE, SCARG(uap, aiocbp), NULL);
832 }
833 
834 int
835 sys_lio_listio(struct lwp *l, void *v, register_t *retval)
836 {
837 	struct sys_lio_listio_args /* {
838 		syscallarg(int) mode;
839 		syscallarg(struct aiocb *const[]) list;
840 		syscallarg(int) nent;
841 		syscallarg(struct sigevent *) sig;
842 	} */ *uap = v;
843 	struct proc *p = l->l_proc;
844 	struct aioproc *aio;
845 	struct aiocb **aiocbp_list;
846 	struct lio_req *lio;
847 	int i, error, errcnt, mode, nent;
848 
849 	mode = SCARG(uap, mode);
850 	nent = SCARG(uap, nent);
851 
852 	/* Check for the limits, and invalid values */
853 	if (nent < 1 || nent > aio_listio_max)
854 		return EINVAL;
855 	if (aio_jobs_count + nent > aio_max) /* XXXSMP */
856 		return EAGAIN;
857 
858 	/* Check if AIO structure is initialized, if not - initialize it */
859 	if (p->p_aio == NULL)
860 		if (aio_init(p))
861 			return EAGAIN;
862 	aio = p->p_aio;
863 
864 	/* Create a LIO structure */
865 	lio = pool_get(&aio_lio_pool, PR_WAITOK);
866 	lio->refcnt = 1;
867 	error = 0;
868 
869 	switch (mode) {
870 	case LIO_WAIT:
871 		memset(&lio->sig, 0, sizeof(struct sigevent));
872 		break;
873 	case LIO_NOWAIT:
874 		/* Check for signal, validate it */
875 		if (SCARG(uap, sig)) {
876 			struct sigevent *sig = &lio->sig;
877 
878 			error = copyin(SCARG(uap, sig), &lio->sig,
879 			    sizeof(struct sigevent));
880 			if (error == 0 &&
881 			    (sig->sigev_signo < 0 ||
882 			    sig->sigev_signo >= NSIG ||
883 			    sig->sigev_notify < SIGEV_NONE ||
884 			    sig->sigev_notify > SIGEV_SA))
885 				error = EINVAL;
886 		} else
887 			memset(&lio->sig, 0, sizeof(struct sigevent));
888 		break;
889 	default:
890 		error = EINVAL;
891 		break;
892 	}
893 
894 	if (error != 0) {
895 		pool_put(&aio_lio_pool, lio);
896 		return error;
897 	}
898 
899 	/* Get the list from user-space */
900 	aiocbp_list = kmem_zalloc(nent * sizeof(struct aio_job), KM_SLEEP);
901 	error = copyin(SCARG(uap, list), aiocbp_list,
902 	    nent * sizeof(struct aiocb));
903 	if (error) {
904 		mutex_enter(&aio->aio_mtx);
905 		goto err;
906 	}
907 
908 	/* Enqueue all jobs */
909 	errcnt = 0;
910 	for (i = 0; i < nent; i++) {
911 		error = aio_enqueue_job(AIO_LIO, aiocbp_list[i], lio);
912 		/*
913 		 * According to POSIX, in such error case it may
914 		 * fail with other I/O operations initiated.
915 		 */
916 		if (error)
917 			errcnt++;
918 	}
919 
920 	mutex_enter(&aio->aio_mtx);
921 
922 	/* Return an error, if any */
923 	if (errcnt) {
924 		error = EIO;
925 		goto err;
926 	}
927 
928 	if (mode == LIO_WAIT) {
929 		/*
930 		 * Wait for AIO completion.  In such case,
931 		 * the LIO structure will be freed here.
932 		 */
933 		while (lio->refcnt > 1 && error == 0)
934 			error = cv_wait_sig(&aio->done_cv, &aio->aio_mtx);
935 		if (error)
936 			error = EINTR;
937 	}
938 
939 err:
940 	if (--lio->refcnt != 0)
941 		lio = NULL;
942 	mutex_exit(&aio->aio_mtx);
943 	if (lio != NULL) {
944 		aio_sendsig(p, &lio->sig);
945 		pool_put(&aio_lio_pool, lio);
946 	}
947 	kmem_free(aiocbp_list, nent * sizeof(struct aio_job));
948 	return error;
949 }
950 
951 /*
952  * SysCtl
953  */
954 
955 static int
956 sysctl_aio_listio_max(SYSCTLFN_ARGS)
957 {
958 	struct sysctlnode node;
959 	int error, newsize;
960 
961 	node = *rnode;
962 	node.sysctl_data = &newsize;
963 
964 	newsize = aio_listio_max;
965 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
966 	if (error || newp == NULL)
967 		return error;
968 
969 	/* XXXSMP */
970 	if (newsize < 1 || newsize > aio_max)
971 		return EINVAL;
972 	aio_listio_max = newsize;
973 
974 	return 0;
975 }
976 
977 static int
978 sysctl_aio_max(SYSCTLFN_ARGS)
979 {
980 	struct sysctlnode node;
981 	int error, newsize;
982 
983 	node = *rnode;
984 	node.sysctl_data = &newsize;
985 
986 	newsize = aio_max;
987 	error = sysctl_lookup(SYSCTLFN_CALL(&node));
988 	if (error || newp == NULL)
989 		return error;
990 
991 	/* XXXSMP */
992 	if (newsize < 1 || newsize < aio_listio_max)
993 		return EINVAL;
994 	aio_max = newsize;
995 
996 	return 0;
997 }
998 
999 SYSCTL_SETUP(sysctl_aio_setup, "sysctl aio setup")
1000 {
1001 
1002 	sysctl_createv(clog, 0, NULL, NULL,
1003 		CTLFLAG_PERMANENT,
1004 		CTLTYPE_NODE, "kern", NULL,
1005 		NULL, 0, NULL, 0,
1006 		CTL_KERN, CTL_EOL);
1007 	sysctl_createv(clog, 0, NULL, NULL,
1008 		CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE,
1009 		CTLTYPE_INT, "posix_aio",
1010 		SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
1011 			     "Asynchronous I/O option to which the "
1012 			     "system attempts to conform"),
1013 		NULL, _POSIX_ASYNCHRONOUS_IO, NULL, 0,
1014 		CTL_KERN, CTL_CREATE, CTL_EOL);
1015 	sysctl_createv(clog, 0, NULL, NULL,
1016 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1017 		CTLTYPE_INT, "aio_listio_max",
1018 		SYSCTL_DESCR("Maximum number of asynchronous I/O "
1019 			     "operations in a single list I/O call"),
1020 		sysctl_aio_listio_max, 0, &aio_listio_max, 0,
1021 		CTL_KERN, CTL_CREATE, CTL_EOL);
1022 	sysctl_createv(clog, 0, NULL, NULL,
1023 		CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1024 		CTLTYPE_INT, "aio_max",
1025 		SYSCTL_DESCR("Maximum number of asynchronous I/O "
1026 			     "operations"),
1027 		sysctl_aio_max, 0, &aio_max, 0,
1028 		CTL_KERN, CTL_CREATE, CTL_EOL);
1029 }
1030 
1031 /*
1032  * Debugging
1033  */
1034 #if defined(DDB)
1035 void
1036 aio_print_jobs(void (*pr)(const char *, ...))
1037 {
1038 	struct proc *p = (curlwp == NULL ? NULL : curlwp->l_proc);
1039 	struct aioproc *aio;
1040 	struct aio_job *a_job;
1041 	struct aiocb *aiocbp;
1042 
1043 	if (p == NULL) {
1044 		(*pr)("AIO: We are not in the processes right now.\n");
1045 		return;
1046 	}
1047 
1048 	aio = p->p_aio;
1049 	if (aio == NULL) {
1050 		(*pr)("AIO data is not initialized (PID = %d).\n", p->p_pid);
1051 		return;
1052 	}
1053 
1054 	(*pr)("AIO: PID = %d\n", p->p_pid);
1055 	(*pr)("AIO: Global count of the jobs = %u\n", aio_jobs_count);
1056 	(*pr)("AIO: Count of the jobs = %u\n", aio->jobs_count);
1057 
1058 	if (aio->curjob) {
1059 		a_job = aio->curjob;
1060 		(*pr)("\nAIO current job:\n");
1061 		(*pr)(" opcode = %d, errno = %d, state = %d, aiocb_ptr = %p\n",
1062 		    a_job->aio_op, a_job->aiocbp._errno,
1063 		    a_job->aiocbp._state, a_job->aiocb_uptr);
1064 		aiocbp = &a_job->aiocbp;
1065 		(*pr)("   fd = %d, offset = %u, buf = %p, nbytes = %u\n",
1066 		    aiocbp->aio_fildes, aiocbp->aio_offset,
1067 		    aiocbp->aio_buf, aiocbp->aio_nbytes);
1068 	}
1069 
1070 	(*pr)("\nAIO queue:\n");
1071 	TAILQ_FOREACH(a_job, &aio->jobs_queue, list) {
1072 		(*pr)(" opcode = %d, errno = %d, state = %d, aiocb_ptr = %p\n",
1073 		    a_job->aio_op, a_job->aiocbp._errno,
1074 		    a_job->aiocbp._state, a_job->aiocb_uptr);
1075 		aiocbp = &a_job->aiocbp;
1076 		(*pr)("   fd = %d, offset = %u, buf = %p, nbytes = %u\n",
1077 		    aiocbp->aio_fildes, aiocbp->aio_offset,
1078 		    aiocbp->aio_buf, aiocbp->aio_nbytes);
1079 	}
1080 }
1081 #endif /* defined(DDB) */
1082