1/* $NetBSD: sys_aio.c,v 1.41 2016/07/07 06:55:43 msaitoh Exp $ */
2
3/*
4 * Copyright (c) 2007 Mindaugas Rasiukevicius <rmind at NetBSD org>
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 * notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 * notice, this list of conditions and the following disclaimer in the
14 * documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29/*
30 * Implementation of POSIX asynchronous I/O.
31 * Defined in the Base Definitions volume of IEEE Std 1003.1-2001.
32 */
33
34#include <sys/cdefs.h>
35__KERNEL_RCSID(0, "$NetBSD: sys_aio.c,v 1.41 2016/07/07 06:55:43 msaitoh Exp $");
36
37#ifdef _KERNEL_OPT
38#include "opt_ddb.h"
39#endif
40
41#include <sys/param.h>
42#include <sys/condvar.h>
43#include <sys/file.h>
44#include <sys/filedesc.h>
45#include <sys/kernel.h>
46#include <sys/kmem.h>
47#include <sys/lwp.h>
48#include <sys/mutex.h>
49#include <sys/pool.h>
50#include <sys/proc.h>
51#include <sys/queue.h>
52#include <sys/signal.h>
53#include <sys/signalvar.h>
54#include <sys/syscall.h>
55#include <sys/syscallargs.h>
56#include <sys/syscallvar.h>
57#include <sys/sysctl.h>
58#include <sys/systm.h>
59#include <sys/types.h>
60#include <sys/vnode.h>
61#include <sys/atomic.h>
62#include <sys/module.h>
63#include <sys/buf.h>
64
65#include <uvm/uvm_extern.h>
66
67MODULE(MODULE_CLASS_MISC, aio, NULL);
68
69/*
70 * System-wide limits and counter of AIO operations.
71 */
72u_int aio_listio_max = AIO_LISTIO_MAX;
73static u_int aio_max = AIO_MAX;
74static u_int aio_jobs_count;
75
76static struct sysctllog *aio_sysctl;
77static struct pool aio_job_pool;
78static struct pool aio_lio_pool;
79static void * aio_ehook;
80
81static void aio_worker(void *);
82static void aio_process(struct aio_job *);
83static void aio_sendsig(struct proc *, struct sigevent *);
84static int aio_enqueue_job(int, void *, struct lio_req *);
85static void aio_exit(proc_t *, void *);
86
87static int sysctl_aio_listio_max(SYSCTLFN_PROTO);
88static int sysctl_aio_max(SYSCTLFN_PROTO);
89static int sysctl_aio_init(void);
90
91static const struct syscall_package aio_syscalls[] = {
92 { SYS_aio_cancel, 0, (sy_call_t *)sys_aio_cancel },
93 { SYS_aio_error, 0, (sy_call_t *)sys_aio_error },
94 { SYS_aio_fsync, 0, (sy_call_t *)sys_aio_fsync },
95 { SYS_aio_read, 0, (sy_call_t *)sys_aio_read },
96 { SYS_aio_return, 0, (sy_call_t *)sys_aio_return },
97 { SYS___aio_suspend50, 0, (sy_call_t *)sys___aio_suspend50 },
98 { SYS_aio_write, 0, (sy_call_t *)sys_aio_write },
99 { SYS_lio_listio, 0, (sy_call_t *)sys_lio_listio },
100 { 0, 0, NULL },
101};
102
103/*
104 * Tear down all AIO state.
105 */
106static int
107aio_fini(bool interface)
108{
109 int error;
110 proc_t *p;
111
112 if (interface) {
113 /* Stop syscall activity. */
114 error = syscall_disestablish(NULL, aio_syscalls);
115 if (error != 0)
116 return error;
117 /* Abort if any processes are using AIO. */
118 mutex_enter(proc_lock);
119 PROCLIST_FOREACH(p, &allproc) {
120 if (p->p_aio != NULL)
121 break;
122 }
123 mutex_exit(proc_lock);
124 if (p != NULL) {
125 error = syscall_establish(NULL, aio_syscalls);
126 KASSERT(error == 0);
127 return EBUSY;
128 }
129 }
130 if (aio_sysctl != NULL)
131 sysctl_teardown(&aio_sysctl);
132
133 KASSERT(aio_jobs_count == 0);
134 exithook_disestablish(aio_ehook);
135 pool_destroy(&aio_job_pool);
136 pool_destroy(&aio_lio_pool);
137 return 0;
138}
139
140/*
141 * Initialize global AIO state.
142 */
143static int
144aio_init(void)
145{
146 int error;
147
148 pool_init(&aio_job_pool, sizeof(struct aio_job), 0, 0, 0,
149 "aio_jobs_pool", &pool_allocator_nointr, IPL_NONE);
150 pool_init(&aio_lio_pool, sizeof(struct lio_req), 0, 0, 0,
151 "aio_lio_pool", &pool_allocator_nointr, IPL_NONE);
152 aio_ehook = exithook_establish(aio_exit, NULL);
153
154 error = sysctl_aio_init();
155 if (error != 0) {
156 (void)aio_fini(false);
157 return error;
158 }
159 error = syscall_establish(NULL, aio_syscalls);
160 if (error != 0)
161 (void)aio_fini(false);
162 return error;
163}
164
165/*
166 * Module interface.
167 */
168static int
169aio_modcmd(modcmd_t cmd, void *arg)
170{
171
172 switch (cmd) {
173 case MODULE_CMD_INIT:
174 return aio_init();
175 case MODULE_CMD_FINI:
176 return aio_fini(true);
177 default:
178 return ENOTTY;
179 }
180}
181
182/*
183 * Initialize Asynchronous I/O data structures for the process.
184 */
185static int
186aio_procinit(struct proc *p)
187{
188 struct aioproc *aio;
189 struct lwp *l;
190 int error;
191 vaddr_t uaddr;
192
193 /* Allocate and initialize AIO structure */
194 aio = kmem_zalloc(sizeof(struct aioproc), KM_SLEEP);
195 if (aio == NULL)
196 return EAGAIN;
197
198 /* Initialize queue and their synchronization structures */
199 mutex_init(&aio->aio_mtx, MUTEX_DEFAULT, IPL_NONE);
200 cv_init(&aio->aio_worker_cv, "aiowork");
201 cv_init(&aio->done_cv, "aiodone");
202 TAILQ_INIT(&aio->jobs_queue);
203
204 /*
205 * Create an AIO worker thread.
206 * XXX: Currently, AIO thread is not protected against user's actions.
207 */
208 uaddr = uvm_uarea_alloc();
209 if (uaddr == 0) {
210 aio_exit(p, aio);
211 return EAGAIN;
212 }
213 error = lwp_create(curlwp, p, uaddr, 0, NULL, 0, aio_worker,
214 NULL, &l, curlwp->l_class);
215 if (error != 0) {
216 uvm_uarea_free(uaddr);
217 aio_exit(p, aio);
218 return error;
219 }
220
221 /* Recheck if we are really first */
222 mutex_enter(p->p_lock);
223 if (p->p_aio) {
224 mutex_exit(p->p_lock);
225 aio_exit(p, aio);
226 lwp_exit(l);
227 return 0;
228 }
229 p->p_aio = aio;
230
231 /* Complete the initialization of thread, and run it */
232 aio->aio_worker = l;
233 lwp_lock(l);
234 l->l_stat = LSRUN;
235 l->l_priority = MAXPRI_USER;
236 sched_enqueue(l, false);
237 lwp_unlock(l);
238 mutex_exit(p->p_lock);
239
240 return 0;
241}
242
243/*
244 * Exit of Asynchronous I/O subsystem of process.
245 */
246static void
247aio_exit(struct proc *p, void *cookie)
248{
249 struct aio_job *a_job;
250 struct aioproc *aio;
251
252 if (cookie != NULL)
253 aio = cookie;
254 else if ((aio = p->p_aio) == NULL)
255 return;
256
257 /* Free AIO queue */
258 while (!TAILQ_EMPTY(&aio->jobs_queue)) {
259 a_job = TAILQ_FIRST(&aio->jobs_queue);
260 TAILQ_REMOVE(&aio->jobs_queue, a_job, list);
261 pool_put(&aio_job_pool, a_job);
262 atomic_dec_uint(&aio_jobs_count);
263 }
264
265 /* Destroy and free the entire AIO data structure */
266 cv_destroy(&aio->aio_worker_cv);
267 cv_destroy(&aio->done_cv);
268 mutex_destroy(&aio->aio_mtx);
269 kmem_free(aio, sizeof(struct aioproc));
270}
271
272/*
273 * AIO worker thread and processor.
274 */
275static void
276aio_worker(void *arg)
277{
278 struct proc *p = curlwp->l_proc;
279 struct aioproc *aio = p->p_aio;
280 struct aio_job *a_job;
281 struct lio_req *lio;
282 sigset_t oss, nss;
283 int error __diagused, refcnt;
284
285 /*
286 * Make an empty signal mask, so it
287 * handles only SIGKILL and SIGSTOP.
288 */
289 sigfillset(&nss);
290 mutex_enter(p->p_lock);
291 error = sigprocmask1(curlwp, SIG_SETMASK, &nss, &oss);
292 mutex_exit(p->p_lock);
293 KASSERT(error == 0);
294
295 for (;;) {
296 /*
297 * Loop for each job in the queue. If there
298 * are no jobs then sleep.
299 */
300 mutex_enter(&aio->aio_mtx);
301 while ((a_job = TAILQ_FIRST(&aio->jobs_queue)) == NULL) {
302 if (cv_wait_sig(&aio->aio_worker_cv, &aio->aio_mtx)) {
303 /*
304 * Thread was interrupted - check for
305 * pending exit or suspend.
306 */
307 mutex_exit(&aio->aio_mtx);
308 lwp_userret(curlwp);
309 mutex_enter(&aio->aio_mtx);
310 }
311 }
312
313 /* Take the job from the queue */
314 aio->curjob = a_job;
315 TAILQ_REMOVE(&aio->jobs_queue, a_job, list);
316
317 atomic_dec_uint(&aio_jobs_count);
318 aio->jobs_count--;
319
320 mutex_exit(&aio->aio_mtx);
321
322 /* Process an AIO operation */
323 aio_process(a_job);
324
325 /* Copy data structure back to the user-space */
326 (void)copyout(&a_job->aiocbp, a_job->aiocb_uptr,
327 sizeof(struct aiocb));
328
329 mutex_enter(&aio->aio_mtx);
330 KASSERT(aio->curjob == a_job);
331 aio->curjob = NULL;
332
333 /* Decrease a reference counter, if there is a LIO structure */
334 lio = a_job->lio;
335 refcnt = (lio != NULL ? --lio->refcnt : -1);
336
337 /* Notify all suspenders */
338 cv_broadcast(&aio->done_cv);
339 mutex_exit(&aio->aio_mtx);
340
341 /* Send a signal, if any */
342 aio_sendsig(p, &a_job->aiocbp.aio_sigevent);
343
344 /* Destroy the LIO structure */
345 if (refcnt == 0) {
346 aio_sendsig(p, &lio->sig);
347 pool_put(&aio_lio_pool, lio);
348 }
349
350 /* Destroy the job */
351 pool_put(&aio_job_pool, a_job);
352 }
353
354 /* NOTREACHED */
355}
356
357static void
358aio_process(struct aio_job *a_job)
359{
360 struct proc *p = curlwp->l_proc;
361 struct aiocb *aiocbp = &a_job->aiocbp;
362 struct file *fp;
363 int fd = aiocbp->aio_fildes;
364 int error = 0;
365
366 KASSERT(a_job->aio_op != 0);
367
368 if ((a_job->aio_op & (AIO_READ | AIO_WRITE)) != 0) {
369 struct iovec aiov;
370 struct uio auio;
371
372 if (aiocbp->aio_nbytes > SSIZE_MAX) {
373 error = EINVAL;
374 goto done;
375 }
376
377 fp = fd_getfile(fd);
378 if (fp == NULL) {
379 error = EBADF;
380 goto done;
381 }
382
383 aiov.iov_base = (void *)(uintptr_t)aiocbp->aio_buf;
384 aiov.iov_len = aiocbp->aio_nbytes;
385 auio.uio_iov = &aiov;
386 auio.uio_iovcnt = 1;
387 auio.uio_resid = aiocbp->aio_nbytes;
388 auio.uio_vmspace = p->p_vmspace;
389
390 if (a_job->aio_op & AIO_READ) {
391 /*
392 * Perform a Read operation
393 */
394 KASSERT((a_job->aio_op & AIO_WRITE) == 0);
395
396 if ((fp->f_flag & FREAD) == 0) {
397 fd_putfile(fd);
398 error = EBADF;
399 goto done;
400 }
401 auio.uio_rw = UIO_READ;
402 error = (*fp->f_ops->fo_read)(fp, &aiocbp->aio_offset,
403 &auio, fp->f_cred, FOF_UPDATE_OFFSET);
404 } else {
405 /*
406 * Perform a Write operation
407 */
408 KASSERT(a_job->aio_op & AIO_WRITE);
409
410 if ((fp->f_flag & FWRITE) == 0) {
411 fd_putfile(fd);
412 error = EBADF;
413 goto done;
414 }
415 auio.uio_rw = UIO_WRITE;
416 error = (*fp->f_ops->fo_write)(fp, &aiocbp->aio_offset,
417 &auio, fp->f_cred, FOF_UPDATE_OFFSET);
418 }
419 fd_putfile(fd);
420
421 /* Store the result value */
422 a_job->aiocbp.aio_nbytes -= auio.uio_resid;
423 a_job->aiocbp._retval = (error == 0) ?
424 a_job->aiocbp.aio_nbytes : -1;
425
426 } else if ((a_job->aio_op & (AIO_SYNC | AIO_DSYNC)) != 0) {
427 /*
428 * Perform a file Sync operation
429 */
430 struct vnode *vp;
431
432 if ((error = fd_getvnode(fd, &fp)) != 0)
433 goto done;
434
435 if ((fp->f_flag & FWRITE) == 0) {
436 fd_putfile(fd);
437 error = EBADF;
438 goto done;
439 }
440
441 vp = fp->f_vnode;
442 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
443 if (a_job->aio_op & AIO_DSYNC) {
444 error = VOP_FSYNC(vp, fp->f_cred,
445 FSYNC_WAIT | FSYNC_DATAONLY, 0, 0);
446 } else if (a_job->aio_op & AIO_SYNC) {
447 error = VOP_FSYNC(vp, fp->f_cred,
448 FSYNC_WAIT, 0, 0);
449 }
450 VOP_UNLOCK(vp);
451 fd_putfile(fd);
452
453 /* Store the result value */
454 a_job->aiocbp._retval = (error == 0) ? 0 : -1;
455
456 } else
457 panic("aio_process: invalid operation code\n");
458
459done:
460 /* Job is done, set the error, if any */
461 a_job->aiocbp._errno = error;
462 a_job->aiocbp._state = JOB_DONE;
463}
464
465/*
466 * Send AIO signal.
467 */
468static void
469aio_sendsig(struct proc *p, struct sigevent *sig)
470{
471 ksiginfo_t ksi;
472
473 if (sig->sigev_signo == 0 || sig->sigev_notify == SIGEV_NONE)
474 return;
475
476 KSI_INIT(&ksi);
477 ksi.ksi_signo = sig->sigev_signo;
478 ksi.ksi_code = SI_ASYNCIO;
479 ksi.ksi_value = sig->sigev_value;
480 mutex_enter(proc_lock);
481 kpsignal(p, &ksi, NULL);
482 mutex_exit(proc_lock);
483}
484
485/*
486 * Enqueue the job.
487 */
488static int
489aio_enqueue_job(int op, void *aiocb_uptr, struct lio_req *lio)
490{
491 struct proc *p = curlwp->l_proc;
492 struct aioproc *aio;
493 struct aio_job *a_job;
494 struct aiocb aiocbp;
495 struct sigevent *sig;
496 int error;
497
498 /* Non-accurate check for the limit */
499 if (aio_jobs_count + 1 > aio_max)
500 return EAGAIN;
501
502 /* Get the data structure from user-space */
503 error = copyin(aiocb_uptr, &aiocbp, sizeof(struct aiocb));
504 if (error)
505 return error;
506
507 /* Check if signal is set, and validate it */
508 sig = &aiocbp.aio_sigevent;
509 if (sig->sigev_signo < 0 || sig->sigev_signo >= NSIG ||
510 sig->sigev_notify < SIGEV_NONE || sig->sigev_notify > SIGEV_SA)
511 return EINVAL;
512
513 /* Buffer and byte count */
514 if (((AIO_SYNC | AIO_DSYNC) & op) == 0)
515 if (aiocbp.aio_buf == NULL || aiocbp.aio_nbytes > SSIZE_MAX)
516 return EINVAL;
517
518 /* Check the opcode, if LIO_NOP - simply ignore */
519 if (op == AIO_LIO) {
520 KASSERT(lio != NULL);
521 if (aiocbp.aio_lio_opcode == LIO_WRITE)
522 op = AIO_WRITE;
523 else if (aiocbp.aio_lio_opcode == LIO_READ)
524 op = AIO_READ;
525 else
526 return (aiocbp.aio_lio_opcode == LIO_NOP) ? 0 : EINVAL;
527 } else {
528 KASSERT(lio == NULL);
529 }
530
531 /*
532 * Look for already existing job. If found - the job is in-progress.
533 * According to POSIX this is invalid, so return the error.
534 */
535 aio = p->p_aio;
536 if (aio) {
537 mutex_enter(&aio->aio_mtx);
538 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) {
539 if (a_job->aiocb_uptr != aiocb_uptr)
540 continue;
541 mutex_exit(&aio->aio_mtx);
542 return EINVAL;
543 }
544 mutex_exit(&aio->aio_mtx);
545 }
546
547 /*
548 * Check if AIO structure is initialized, if not - initialize it.
549 * In LIO case, we did that already. We will recheck this with
550 * the lock in aio_procinit().
551 */
552 if (lio == NULL && p->p_aio == NULL)
553 if (aio_procinit(p))
554 return EAGAIN;
555 aio = p->p_aio;
556
557 /*
558 * Set the state with errno, and copy data
559 * structure back to the user-space.
560 */
561 aiocbp._state = JOB_WIP;
562 aiocbp._errno = EINPROGRESS;
563 aiocbp._retval = -1;
564 error = copyout(&aiocbp, aiocb_uptr, sizeof(struct aiocb));
565 if (error)
566 return error;
567
568 /* Allocate and initialize a new AIO job */
569 a_job = pool_get(&aio_job_pool, PR_WAITOK);
570 memset(a_job, 0, sizeof(struct aio_job));
571
572 /*
573 * Set the data.
574 * Store the user-space pointer for searching. Since we
575 * are storing only per proc pointers - it is safe.
576 */
577 memcpy(&a_job->aiocbp, &aiocbp, sizeof(struct aiocb));
578 a_job->aiocb_uptr = aiocb_uptr;
579 a_job->aio_op |= op;
580 a_job->lio = lio;
581
582 /*
583 * Add the job to the queue, update the counters, and
584 * notify the AIO worker thread to handle the job.
585 */
586 mutex_enter(&aio->aio_mtx);
587
588 /* Fail, if the limit was reached */
589 if (atomic_inc_uint_nv(&aio_jobs_count) > aio_max ||
590 aio->jobs_count >= aio_listio_max) {
591 atomic_dec_uint(&aio_jobs_count);
592 mutex_exit(&aio->aio_mtx);
593 pool_put(&aio_job_pool, a_job);
594 return EAGAIN;
595 }
596
597 TAILQ_INSERT_TAIL(&aio->jobs_queue, a_job, list);
598 aio->jobs_count++;
599 if (lio)
600 lio->refcnt++;
601 cv_signal(&aio->aio_worker_cv);
602
603 mutex_exit(&aio->aio_mtx);
604
605 /*
606 * One would handle the errors only with aio_error() function.
607 * This way is appropriate according to POSIX.
608 */
609 return 0;
610}
611
612/*
613 * Syscall functions.
614 */
615
616int
617sys_aio_cancel(struct lwp *l, const struct sys_aio_cancel_args *uap,
618 register_t *retval)
619{
620 /* {
621 syscallarg(int) fildes;
622 syscallarg(struct aiocb *) aiocbp;
623 } */
624 struct proc *p = l->l_proc;
625 struct aioproc *aio;
626 struct aio_job *a_job;
627 struct aiocb *aiocbp_ptr;
628 struct lio_req *lio;
629 struct filedesc *fdp = p->p_fd;
630 unsigned int cn, errcnt, fildes;
631 fdtab_t *dt;
632
633 TAILQ_HEAD(, aio_job) tmp_jobs_list;
634
635 /* Check for invalid file descriptor */
636 fildes = (unsigned int)SCARG(uap, fildes);
637 dt = fdp->fd_dt;
638 if (fildes >= dt->dt_nfiles)
639 return EBADF;
640 if (dt->dt_ff[fildes] == NULL || dt->dt_ff[fildes]->ff_file == NULL)
641 return EBADF;
642
643 /* Check if AIO structure is initialized */
644 if (p->p_aio == NULL) {
645 *retval = AIO_NOTCANCELED;
646 return 0;
647 }
648
649 aio = p->p_aio;
650 aiocbp_ptr = (struct aiocb *)SCARG(uap, aiocbp);
651
652 mutex_enter(&aio->aio_mtx);
653
654 /* Cancel the jobs, and remove them from the queue */
655 cn = 0;
656 TAILQ_INIT(&tmp_jobs_list);
657 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) {
658 if (aiocbp_ptr) {
659 if (aiocbp_ptr != a_job->aiocb_uptr)
660 continue;
661 if (fildes != a_job->aiocbp.aio_fildes) {
662 mutex_exit(&aio->aio_mtx);
663 return EBADF;
664 }
665 } else if (a_job->aiocbp.aio_fildes != fildes)
666 continue;
667
668 TAILQ_REMOVE(&aio->jobs_queue, a_job, list);
669 TAILQ_INSERT_TAIL(&tmp_jobs_list, a_job, list);
670
671 /* Decrease the counters */
672 atomic_dec_uint(&aio_jobs_count);
673 aio->jobs_count--;
674 lio = a_job->lio;
675 if (lio != NULL && --lio->refcnt != 0)
676 a_job->lio = NULL;
677
678 cn++;
679 if (aiocbp_ptr)
680 break;
681 }
682
683 /* There are canceled jobs */
684 if (cn)
685 *retval = AIO_CANCELED;
686
687 /* We cannot cancel current job */
688 a_job = aio->curjob;
689 if (a_job && ((a_job->aiocbp.aio_fildes == fildes) ||
690 (a_job->aiocb_uptr == aiocbp_ptr)))
691 *retval = AIO_NOTCANCELED;
692
693 mutex_exit(&aio->aio_mtx);
694
695 /* Free the jobs after the lock */
696 errcnt = 0;
697 while (!TAILQ_EMPTY(&tmp_jobs_list)) {
698 a_job = TAILQ_FIRST(&tmp_jobs_list);
699 TAILQ_REMOVE(&tmp_jobs_list, a_job, list);
700 /* Set the errno and copy structures back to the user-space */
701 a_job->aiocbp._errno = ECANCELED;
702 a_job->aiocbp._state = JOB_DONE;
703 if (copyout(&a_job->aiocbp, a_job->aiocb_uptr,
704 sizeof(struct aiocb)))
705 errcnt++;
706 /* Send a signal if any */
707 aio_sendsig(p, &a_job->aiocbp.aio_sigevent);
708 if (a_job->lio) {
709 lio = a_job->lio;
710 aio_sendsig(p, &lio->sig);
711 pool_put(&aio_lio_pool, lio);
712 }
713 pool_put(&aio_job_pool, a_job);
714 }
715
716 if (errcnt)
717 return EFAULT;
718
719 /* Set a correct return value */
720 if (*retval == 0)
721 *retval = AIO_ALLDONE;
722
723 return 0;
724}
725
726int
727sys_aio_error(struct lwp *l, const struct sys_aio_error_args *uap,
728 register_t *retval)
729{
730 /* {
731 syscallarg(const struct aiocb *) aiocbp;
732 } */
733 struct proc *p = l->l_proc;
734 struct aioproc *aio = p->p_aio;
735 struct aiocb aiocbp;
736 int error;
737
738 if (aio == NULL)
739 return EINVAL;
740
741 error = copyin(SCARG(uap, aiocbp), &aiocbp, sizeof(struct aiocb));
742 if (error)
743 return error;
744
745 if (aiocbp._state == JOB_NONE)
746 return EINVAL;
747
748 *retval = aiocbp._errno;
749
750 return 0;
751}
752
753int
754sys_aio_fsync(struct lwp *l, const struct sys_aio_fsync_args *uap,
755 register_t *retval)
756{
757 /* {
758 syscallarg(int) op;
759 syscallarg(struct aiocb *) aiocbp;
760 } */
761 int op = SCARG(uap, op);
762
763 if ((op != O_DSYNC) && (op != O_SYNC))
764 return EINVAL;
765
766 op = O_DSYNC ? AIO_DSYNC : AIO_SYNC;
767
768 return aio_enqueue_job(op, SCARG(uap, aiocbp), NULL);
769}
770
771int
772sys_aio_read(struct lwp *l, const struct sys_aio_read_args *uap,
773 register_t *retval)
774{
775 /* {
776 syscallarg(struct aiocb *) aiocbp;
777 } */
778
779 return aio_enqueue_job(AIO_READ, SCARG(uap, aiocbp), NULL);
780}
781
782int
783sys_aio_return(struct lwp *l, const struct sys_aio_return_args *uap,
784 register_t *retval)
785{
786 /* {
787 syscallarg(struct aiocb *) aiocbp;
788 } */
789 struct proc *p = l->l_proc;
790 struct aioproc *aio = p->p_aio;
791 struct aiocb aiocbp;
792 int error;
793
794 if (aio == NULL)
795 return EINVAL;
796
797 error = copyin(SCARG(uap, aiocbp), &aiocbp, sizeof(struct aiocb));
798 if (error)
799 return error;
800
801 if (aiocbp._errno == EINPROGRESS || aiocbp._state != JOB_DONE)
802 return EINVAL;
803
804 *retval = aiocbp._retval;
805
806 /* Reset the internal variables */
807 aiocbp._errno = 0;
808 aiocbp._retval = -1;
809 aiocbp._state = JOB_NONE;
810 error = copyout(&aiocbp, SCARG(uap, aiocbp), sizeof(struct aiocb));
811
812 return error;
813}
814
815int
816sys___aio_suspend50(struct lwp *l, const struct sys___aio_suspend50_args *uap,
817 register_t *retval)
818{
819 /* {
820 syscallarg(const struct aiocb *const[]) list;
821 syscallarg(int) nent;
822 syscallarg(const struct timespec *) timeout;
823 } */
824 struct aiocb **list;
825 struct timespec ts;
826 int error, nent;
827
828 nent = SCARG(uap, nent);
829 if (nent <= 0 || nent > aio_listio_max)
830 return EAGAIN;
831
832 if (SCARG(uap, timeout)) {
833 /* Convert timespec to ticks */
834 error = copyin(SCARG(uap, timeout), &ts,
835 sizeof(struct timespec));
836 if (error)
837 return error;
838 }
839
840 list = kmem_alloc(nent * sizeof(*list), KM_SLEEP);
841 error = copyin(SCARG(uap, list), list, nent * sizeof(*list));
842 if (error)
843 goto out;
844 error = aio_suspend1(l, list, nent, SCARG(uap, timeout) ? &ts : NULL);
845out:
846 kmem_free(list, nent * sizeof(*list));
847 return error;
848}
849
850int
851aio_suspend1(struct lwp *l, struct aiocb **aiocbp_list, int nent,
852 struct timespec *ts)
853{
854 struct proc *p = l->l_proc;
855 struct aioproc *aio;
856 struct aio_job *a_job;
857 int i, error, timo;
858
859 if (p->p_aio == NULL)
860 return EAGAIN;
861 aio = p->p_aio;
862
863 if (ts) {
864 timo = mstohz((ts->tv_sec * 1000) + (ts->tv_nsec / 1000000));
865 if (timo == 0 && ts->tv_sec == 0 && ts->tv_nsec > 0)
866 timo = 1;
867 if (timo <= 0)
868 return EAGAIN;
869 } else
870 timo = 0;
871
872 mutex_enter(&aio->aio_mtx);
873 for (;;) {
874 for (i = 0; i < nent; i++) {
875
876 /* Skip NULL entries */
877 if (aiocbp_list[i] == NULL)
878 continue;
879
880 /* Skip current job */
881 if (aio->curjob) {
882 a_job = aio->curjob;
883 if (a_job->aiocb_uptr == aiocbp_list[i])
884 continue;
885 }
886
887 /* Look for a job in the queue */
888 TAILQ_FOREACH(a_job, &aio->jobs_queue, list)
889 if (a_job->aiocb_uptr == aiocbp_list[i])
890 break;
891
892 if (a_job == NULL) {
893 struct aiocb aiocbp;
894
895 mutex_exit(&aio->aio_mtx);
896
897 /* Check if the job is done. */
898 error = copyin(aiocbp_list[i], &aiocbp,
899 sizeof(struct aiocb));
900 if (error == 0 && aiocbp._state != JOB_DONE) {
901 mutex_enter(&aio->aio_mtx);
902 continue;
903 }
904 return error;
905 }
906 }
907
908 /* Wait for a signal or when timeout occurs */
909 error = cv_timedwait_sig(&aio->done_cv, &aio->aio_mtx, timo);
910 if (error) {
911 if (error == EWOULDBLOCK)
912 error = EAGAIN;
913 break;
914 }
915 }
916 mutex_exit(&aio->aio_mtx);
917 return error;
918}
919
920int
921sys_aio_write(struct lwp *l, const struct sys_aio_write_args *uap,
922 register_t *retval)
923{
924 /* {
925 syscallarg(struct aiocb *) aiocbp;
926 } */
927
928 return aio_enqueue_job(AIO_WRITE, SCARG(uap, aiocbp), NULL);
929}
930
931int
932sys_lio_listio(struct lwp *l, const struct sys_lio_listio_args *uap,
933 register_t *retval)
934{
935 /* {
936 syscallarg(int) mode;
937 syscallarg(struct aiocb *const[]) list;
938 syscallarg(int) nent;
939 syscallarg(struct sigevent *) sig;
940 } */
941 struct proc *p = l->l_proc;
942 struct aioproc *aio;
943 struct aiocb **aiocbp_list;
944 struct lio_req *lio;
945 int i, error, errcnt, mode, nent;
946
947 mode = SCARG(uap, mode);
948 nent = SCARG(uap, nent);
949
950 /* Non-accurate checks for the limit and invalid values */
951 if (nent < 1 || nent > aio_listio_max)
952 return EINVAL;
953 if (aio_jobs_count + nent > aio_max)
954 return EAGAIN;
955
956 /* Check if AIO structure is initialized, if not - initialize it */
957 if (p->p_aio == NULL)
958 if (aio_procinit(p))
959 return EAGAIN;
960 aio = p->p_aio;
961
962 /* Create a LIO structure */
963 lio = pool_get(&aio_lio_pool, PR_WAITOK);
964 lio->refcnt = 1;
965 error = 0;
966
967 switch (mode) {
968 case LIO_WAIT:
969 memset(&lio->sig, 0, sizeof(struct sigevent));
970 break;
971 case LIO_NOWAIT:
972 /* Check for signal, validate it */
973 if (SCARG(uap, sig)) {
974 struct sigevent *sig = &lio->sig;
975
976 error = copyin(SCARG(uap, sig), &lio->sig,
977 sizeof(struct sigevent));
978 if (error == 0 &&
979 (sig->sigev_signo < 0 ||
980 sig->sigev_signo >= NSIG ||
981 sig->sigev_notify < SIGEV_NONE ||
982 sig->sigev_notify > SIGEV_SA))
983 error = EINVAL;
984 } else
985 memset(&lio->sig, 0, sizeof(struct sigevent));
986 break;
987 default:
988 error = EINVAL;
989 break;
990 }
991
992 if (error != 0) {
993 pool_put(&aio_lio_pool, lio);
994 return error;
995 }
996
997 /* Get the list from user-space */
998 aiocbp_list = kmem_alloc(nent * sizeof(*aiocbp_list), KM_SLEEP);
999 error = copyin(SCARG(uap, list), aiocbp_list,
1000 nent * sizeof(*aiocbp_list));
1001 if (error) {
1002 mutex_enter(&aio->aio_mtx);
1003 goto err;
1004 }
1005
1006 /* Enqueue all jobs */
1007 errcnt = 0;
1008 for (i = 0; i < nent; i++) {
1009 error = aio_enqueue_job(AIO_LIO, aiocbp_list[i], lio);
1010 /*
1011 * According to POSIX, in such error case it may
1012 * fail with other I/O operations initiated.
1013 */
1014 if (error)
1015 errcnt++;
1016 }
1017
1018 mutex_enter(&aio->aio_mtx);
1019
1020 /* Return an error, if any */
1021 if (errcnt) {
1022 error = EIO;
1023 goto err;
1024 }
1025
1026 if (mode == LIO_WAIT) {
1027 /*
1028 * Wait for AIO completion. In such case,
1029 * the LIO structure will be freed here.
1030 */
1031 while (lio->refcnt > 1 && error == 0)
1032 error = cv_wait_sig(&aio->done_cv, &aio->aio_mtx);
1033 if (error)
1034 error = EINTR;
1035 }
1036
1037err:
1038 if (--lio->refcnt != 0)
1039 lio = NULL;
1040 mutex_exit(&aio->aio_mtx);
1041 if (lio != NULL) {
1042 aio_sendsig(p, &lio->sig);
1043 pool_put(&aio_lio_pool, lio);
1044 }
1045 kmem_free(aiocbp_list, nent * sizeof(*aiocbp_list));
1046 return error;
1047}
1048
1049/*
1050 * SysCtl
1051 */
1052
1053static int
1054sysctl_aio_listio_max(SYSCTLFN_ARGS)
1055{
1056 struct sysctlnode node;
1057 int error, newsize;
1058
1059 node = *rnode;
1060 node.sysctl_data = &newsize;
1061
1062 newsize = aio_listio_max;
1063 error = sysctl_lookup(SYSCTLFN_CALL(&node));
1064 if (error || newp == NULL)
1065 return error;
1066
1067 if (newsize < 1 || newsize > aio_max)
1068 return EINVAL;
1069 aio_listio_max = newsize;
1070
1071 return 0;
1072}
1073
1074static int
1075sysctl_aio_max(SYSCTLFN_ARGS)
1076{
1077 struct sysctlnode node;
1078 int error, newsize;
1079
1080 node = *rnode;
1081 node.sysctl_data = &newsize;
1082
1083 newsize = aio_max;
1084 error = sysctl_lookup(SYSCTLFN_CALL(&node));
1085 if (error || newp == NULL)
1086 return error;
1087
1088 if (newsize < 1 || newsize < aio_listio_max)
1089 return EINVAL;
1090 aio_max = newsize;
1091
1092 return 0;
1093}
1094
1095static int
1096sysctl_aio_init(void)
1097{
1098 int rv;
1099
1100 aio_sysctl = NULL;
1101
1102 rv = sysctl_createv(&aio_sysctl, 0, NULL, NULL,
1103 CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE,
1104 CTLTYPE_INT, "posix_aio",
1105 SYSCTL_DESCR("Version of IEEE Std 1003.1 and its "
1106 "Asynchronous I/O option to which the "
1107 "system attempts to conform"),
1108 NULL, _POSIX_ASYNCHRONOUS_IO, NULL, 0,
1109 CTL_KERN, CTL_CREATE, CTL_EOL);
1110
1111 if (rv != 0)
1112 return rv;
1113
1114 rv = sysctl_createv(&aio_sysctl, 0, NULL, NULL,
1115 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1116 CTLTYPE_INT, "aio_listio_max",
1117 SYSCTL_DESCR("Maximum number of asynchronous I/O "
1118 "operations in a single list I/O call"),
1119 sysctl_aio_listio_max, 0, &aio_listio_max, 0,
1120 CTL_KERN, CTL_CREATE, CTL_EOL);
1121
1122 if (rv != 0)
1123 return rv;
1124
1125 rv = sysctl_createv(&aio_sysctl, 0, NULL, NULL,
1126 CTLFLAG_PERMANENT | CTLFLAG_READWRITE,
1127 CTLTYPE_INT, "aio_max",
1128 SYSCTL_DESCR("Maximum number of asynchronous I/O "
1129 "operations"),
1130 sysctl_aio_max, 0, &aio_max, 0,
1131 CTL_KERN, CTL_CREATE, CTL_EOL);
1132
1133 return rv;
1134}
1135
1136/*
1137 * Debugging
1138 */
1139#if defined(DDB)
1140void
1141aio_print_jobs(void (*pr)(const char *, ...))
1142{
1143 struct proc *p = curlwp->l_proc;
1144 struct aioproc *aio;
1145 struct aio_job *a_job;
1146 struct aiocb *aiocbp;
1147
1148 if (p == NULL) {
1149 (*pr)("AIO: We are not in the processes right now.\n");
1150 return;
1151 }
1152
1153 aio = p->p_aio;
1154 if (aio == NULL) {
1155 (*pr)("AIO data is not initialized (PID = %d).\n", p->p_pid);
1156 return;
1157 }
1158
1159 (*pr)("AIO: PID = %d\n", p->p_pid);
1160 (*pr)("AIO: Global count of the jobs = %u\n", aio_jobs_count);
1161 (*pr)("AIO: Count of the jobs = %u\n", aio->jobs_count);
1162
1163 if (aio->curjob) {
1164 a_job = aio->curjob;
1165 (*pr)("\nAIO current job:\n");
1166 (*pr)(" opcode = %d, errno = %d, state = %d, aiocb_ptr = %p\n",
1167 a_job->aio_op, a_job->aiocbp._errno,
1168 a_job->aiocbp._state, a_job->aiocb_uptr);
1169 aiocbp = &a_job->aiocbp;
1170 (*pr)(" fd = %d, offset = %u, buf = %p, nbytes = %u\n",
1171 aiocbp->aio_fildes, aiocbp->aio_offset,
1172 aiocbp->aio_buf, aiocbp->aio_nbytes);
1173 }
1174
1175 (*pr)("\nAIO queue:\n");
1176 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) {
1177 (*pr)(" opcode = %d, errno = %d, state = %d, aiocb_ptr = %p\n",
1178 a_job->aio_op, a_job->aiocbp._errno,
1179 a_job->aiocbp._state, a_job->aiocb_uptr);
1180 aiocbp = &a_job->aiocbp;
1181 (*pr)(" fd = %d, offset = %u, buf = %p, nbytes = %u\n",
1182 aiocbp->aio_fildes, aiocbp->aio_offset,
1183 aiocbp->aio_buf, aiocbp->aio_nbytes);
1184 }
1185}
1186#endif /* defined(DDB) */
1187