FreeBSD kernel kern code
vfs_aio.c
Go to the documentation of this file.
1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
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/*
20 * This file contains support for the POSIX 1003.1B AIO/LIO facility.
21 */
22
23#include <sys/cdefs.h>
24__FBSDID("$FreeBSD$");
25
26#include <sys/param.h>
27#include <sys/systm.h>
28#include <sys/malloc.h>
29#include <sys/bio.h>
30#include <sys/buf.h>
31#include <sys/capsicum.h>
32#include <sys/eventhandler.h>
33#include <sys/sysproto.h>
34#include <sys/filedesc.h>
35#include <sys/kernel.h>
36#include <sys/module.h>
37#include <sys/kthread.h>
38#include <sys/fcntl.h>
39#include <sys/file.h>
40#include <sys/limits.h>
41#include <sys/lock.h>
42#include <sys/mutex.h>
43#include <sys/unistd.h>
44#include <sys/posix4.h>
45#include <sys/proc.h>
46#include <sys/resourcevar.h>
47#include <sys/signalvar.h>
48#include <sys/syscallsubr.h>
49#include <sys/protosw.h>
50#include <sys/rwlock.h>
51#include <sys/sema.h>
52#include <sys/socket.h>
53#include <sys/socketvar.h>
54#include <sys/syscall.h>
55#include <sys/sysent.h>
56#include <sys/sysctl.h>
57#include <sys/syslog.h>
58#include <sys/sx.h>
59#include <sys/taskqueue.h>
60#include <sys/vnode.h>
61#include <sys/conf.h>
62#include <sys/event.h>
63#include <sys/mount.h>
64#include <geom/geom.h>
65
66#include <machine/atomic.h>
67
68#include <vm/vm.h>
69#include <vm/vm_page.h>
70#include <vm/vm_extern.h>
71#include <vm/pmap.h>
72#include <vm/vm_map.h>
73#include <vm/vm_object.h>
74#include <vm/uma.h>
75#include <sys/aio.h>
76
77/*
78 * Counter for allocating reference ids to new jobs. Wrapped to 1 on
79 * overflow. (XXX will be removed soon.)
80 */
81static u_long jobrefid;
82
83/*
84 * Counter for aio_fsync.
85 */
86static uint64_t jobseqno;
87
88#ifndef MAX_AIO_PER_PROC
89#define MAX_AIO_PER_PROC 32
90#endif
91
92#ifndef MAX_AIO_QUEUE_PER_PROC
93#define MAX_AIO_QUEUE_PER_PROC 256
94#endif
95
96#ifndef MAX_AIO_QUEUE
97#define MAX_AIO_QUEUE 1024 /* Bigger than MAX_AIO_QUEUE_PER_PROC */
98#endif
99
100#ifndef MAX_BUF_AIO
101#define MAX_BUF_AIO 16
102#endif
103
104FEATURE(aio, "Asynchronous I/O");
105SYSCTL_DECL(_p1003_1b);
106
107static MALLOC_DEFINE(M_LIO, "lio", "listio aio control block list");
108static MALLOC_DEFINE(M_AIOS, "aios", "aio_suspend aio control block list");
109
110static SYSCTL_NODE(_vfs, OID_AUTO, aio, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
111 "Async IO management");
112
113static int enable_aio_unsafe = 0;
114SYSCTL_INT(_vfs_aio, OID_AUTO, enable_unsafe, CTLFLAG_RW, &enable_aio_unsafe, 0,
115 "Permit asynchronous IO on all file types, not just known-safe types");
116
117static unsigned int unsafe_warningcnt = 1;
118SYSCTL_UINT(_vfs_aio, OID_AUTO, unsafe_warningcnt, CTLFLAG_RW,
119 &unsafe_warningcnt, 0,
120 "Warnings that will be triggered upon failed IO requests on unsafe files");
121
122static int max_aio_procs = MAX_AIO_PROCS;
123SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_procs, CTLFLAG_RW, &max_aio_procs, 0,
124 "Maximum number of kernel processes to use for handling async IO ");
125
126static int num_aio_procs = 0;
127SYSCTL_INT(_vfs_aio, OID_AUTO, num_aio_procs, CTLFLAG_RD, &num_aio_procs, 0,
128 "Number of presently active kernel processes for async IO");
129
130/*
131 * The code will adjust the actual number of AIO processes towards this
132 * number when it gets a chance.
133 */
134static int target_aio_procs = TARGET_AIO_PROCS;
135SYSCTL_INT(_vfs_aio, OID_AUTO, target_aio_procs, CTLFLAG_RW, &target_aio_procs,
136 0,
137 "Preferred number of ready kernel processes for async IO");
138
139static int max_queue_count = MAX_AIO_QUEUE;
140SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue, CTLFLAG_RW, &max_queue_count, 0,
141 "Maximum number of aio requests to queue, globally");
142
143static int num_queue_count = 0;
144SYSCTL_INT(_vfs_aio, OID_AUTO, num_queue_count, CTLFLAG_RD, &num_queue_count, 0,
145 "Number of queued aio requests");
146
147static int num_buf_aio = 0;
148SYSCTL_INT(_vfs_aio, OID_AUTO, num_buf_aio, CTLFLAG_RD, &num_buf_aio, 0,
149 "Number of aio requests presently handled by the buf subsystem");
150
151static int num_unmapped_aio = 0;
152SYSCTL_INT(_vfs_aio, OID_AUTO, num_unmapped_aio, CTLFLAG_RD, &num_unmapped_aio,
153 0,
154 "Number of aio requests presently handled by unmapped I/O buffers");
155
156/* Number of async I/O processes in the process of being started */
157/* XXX This should be local to aio_aqueue() */
158static int num_aio_resv_start = 0;
159
160static int aiod_lifetime;
161SYSCTL_INT(_vfs_aio, OID_AUTO, aiod_lifetime, CTLFLAG_RW, &aiod_lifetime, 0,
162 "Maximum lifetime for idle aiod");
163
164static int max_aio_per_proc = MAX_AIO_PER_PROC;
165SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_per_proc, CTLFLAG_RW, &max_aio_per_proc,
166 0,
167 "Maximum active aio requests per process");
168
169static int max_aio_queue_per_proc = MAX_AIO_QUEUE_PER_PROC;
170SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue_per_proc, CTLFLAG_RW,
171 &max_aio_queue_per_proc, 0,
172 "Maximum queued aio requests per process");
173
174static int max_buf_aio = MAX_BUF_AIO;
175SYSCTL_INT(_vfs_aio, OID_AUTO, max_buf_aio, CTLFLAG_RW, &max_buf_aio, 0,
176 "Maximum buf aio requests per process");
177
178/*
179 * Though redundant with vfs.aio.max_aio_queue_per_proc, POSIX requires
180 * sysconf(3) to support AIO_LISTIO_MAX, and we implement that with
181 * vfs.aio.aio_listio_max.
182 */
183SYSCTL_INT(_p1003_1b, CTL_P1003_1B_AIO_LISTIO_MAX, aio_listio_max,
184 CTLFLAG_RD | CTLFLAG_CAPRD, &max_aio_queue_per_proc,
185 0, "Maximum aio requests for a single lio_listio call");
186
187#ifdef COMPAT_FREEBSD6
188typedef struct oaiocb {
189 int aio_fildes; /* File descriptor */
190 off_t aio_offset; /* File offset for I/O */
191 volatile void *aio_buf; /* I/O buffer in process space */
192 size_t aio_nbytes; /* Number of bytes for I/O */
193 struct osigevent aio_sigevent; /* Signal to deliver */
194 int aio_lio_opcode; /* LIO opcode */
195 int aio_reqprio; /* Request priority -- ignored */
196 struct __aiocb_private _aiocb_private;
197} oaiocb_t;
198#endif
199
200/*
201 * Below is a key of locks used to protect each member of struct kaiocb
202 * aioliojob and kaioinfo and any backends.
203 *
204 * * - need not protected
205 * a - locked by kaioinfo lock
206 * b - locked by backend lock, the backend lock can be null in some cases,
207 * for example, BIO belongs to this type, in this case, proc lock is
208 * reused.
209 * c - locked by aio_job_mtx, the lock for the generic file I/O backend.
210 */
211
212/*
213 * If the routine that services an AIO request blocks while running in an
214 * AIO kernel process it can starve other I/O requests. BIO requests
215 * queued via aio_qbio() complete asynchronously and do not use AIO kernel
216 * processes at all. Socket I/O requests use a separate pool of
217 * kprocs and also force non-blocking I/O. Other file I/O requests
218 * use the generic fo_read/fo_write operations which can block. The
219 * fsync and mlock operations can also block while executing. Ideally
220 * none of these requests would block while executing.
221 *
222 * Note that the service routines cannot toggle O_NONBLOCK in the file
223 * structure directly while handling a request due to races with
224 * userland threads.
225 */
226
227/* jobflags */
228#define KAIOCB_QUEUEING 0x01
229#define KAIOCB_CANCELLED 0x02
230#define KAIOCB_CANCELLING 0x04
231#define KAIOCB_CHECKSYNC 0x08
232#define KAIOCB_CLEARED 0x10
233#define KAIOCB_FINISHED 0x20
234
235/*
236 * AIO process info
237 */
238#define AIOP_FREE 0x1 /* proc on free queue */
239
240struct aioproc {
241 int aioprocflags; /* (c) AIO proc flags */
242 TAILQ_ENTRY(aioproc) list; /* (c) list of processes */
243 struct proc *aioproc; /* (*) the AIO proc */
244};
245
246/*
247 * data-structure for lio signal management
248 */
249struct aioliojob {
250 int lioj_flags; /* (a) listio flags */
251 int lioj_count; /* (a) count of jobs */
252 int lioj_finished_count; /* (a) count of finished jobs */
253 struct sigevent lioj_signal; /* (a) signal on all I/O done */
254 TAILQ_ENTRY(aioliojob) lioj_list; /* (a) lio list */
255 struct knlist klist; /* (a) list of knotes */
256 ksiginfo_t lioj_ksi; /* (a) Realtime signal info */
257};
258
259#define LIOJ_SIGNAL 0x1 /* signal on all done (lio) */
260#define LIOJ_SIGNAL_POSTED 0x2 /* signal has been posted */
261#define LIOJ_KEVENT_POSTED 0x4 /* kevent triggered */
262
263/*
264 * per process aio data structure
265 */
266struct kaioinfo {
267 struct mtx kaio_mtx; /* the lock to protect this struct */
268 int kaio_flags; /* (a) per process kaio flags */
269 int kaio_active_count; /* (c) number of currently used AIOs */
270 int kaio_count; /* (a) size of AIO queue */
271 int kaio_buffer_count; /* (a) number of bio buffers */
272 TAILQ_HEAD(,kaiocb) kaio_all; /* (a) all AIOs in a process */
273 TAILQ_HEAD(,kaiocb) kaio_done; /* (a) done queue for process */
274 TAILQ_HEAD(,aioliojob) kaio_liojoblist; /* (a) list of lio jobs */
275 TAILQ_HEAD(,kaiocb) kaio_jobqueue; /* (a) job queue for process */
276 TAILQ_HEAD(,kaiocb) kaio_syncqueue; /* (a) queue for aio_fsync */
277 TAILQ_HEAD(,kaiocb) kaio_syncready; /* (a) second q for aio_fsync */
278 struct task kaio_task; /* (*) task to kick aio processes */
279 struct task kaio_sync_task; /* (*) task to schedule fsync jobs */
280};
281
282#define AIO_LOCK(ki) mtx_lock(&(ki)->kaio_mtx)
283#define AIO_UNLOCK(ki) mtx_unlock(&(ki)->kaio_mtx)
284#define AIO_LOCK_ASSERT(ki, f) mtx_assert(&(ki)->kaio_mtx, (f))
285#define AIO_MTX(ki) (&(ki)->kaio_mtx)
286
287#define KAIO_RUNDOWN 0x1 /* process is being run down */
288#define KAIO_WAKEUP 0x2 /* wakeup process when AIO completes */
289
290/*
291 * Operations used to interact with userland aio control blocks.
292 * Different ABIs provide their own operations.
293 */
294struct aiocb_ops {
295 int (*aio_copyin)(struct aiocb *ujob, struct kaiocb *kjob, int ty);
296 long (*fetch_status)(struct aiocb *ujob);
297 long (*fetch_error)(struct aiocb *ujob);
298 int (*store_status)(struct aiocb *ujob, long status);
299 int (*store_error)(struct aiocb *ujob, long error);
300 int (*store_kernelinfo)(struct aiocb *ujob, long jobref);
301 int (*store_aiocb)(struct aiocb **ujobp, struct aiocb *ujob);
302};
303
304static TAILQ_HEAD(,aioproc) aio_freeproc; /* (c) Idle daemons */
305static struct sema aio_newproc_sem;
306static struct mtx aio_job_mtx;
307static TAILQ_HEAD(,kaiocb) aio_jobs; /* (c) Async job list */
308static struct unrhdr *aiod_unr;
309
310static void aio_biocleanup(struct bio *bp);
311void aio_init_aioinfo(struct proc *p);
312static int aio_onceonly(void);
313static int aio_free_entry(struct kaiocb *job);
314static void aio_process_rw(struct kaiocb *job);
315static void aio_process_sync(struct kaiocb *job);
316static void aio_process_mlock(struct kaiocb *job);
317static void aio_schedule_fsync(void *context, int pending);
318static int aio_newproc(int *);
319int aio_aqueue(struct thread *td, struct aiocb *ujob,
320 struct aioliojob *lio, int type, struct aiocb_ops *ops);
321static int aio_queue_file(struct file *fp, struct kaiocb *job);
322static void aio_biowakeup(struct bio *bp);
323static void aio_proc_rundown(void *arg, struct proc *p);
324static void aio_proc_rundown_exec(void *arg, struct proc *p,
325 struct image_params *imgp);
326static int aio_qbio(struct proc *p, struct kaiocb *job);
327static void aio_daemon(void *param);
328static void aio_bio_done_notify(struct proc *userp, struct kaiocb *job);
329static bool aio_clear_cancel_function_locked(struct kaiocb *job);
330static int aio_kick(struct proc *userp);
331static void aio_kick_nowait(struct proc *userp);
332static void aio_kick_helper(void *context, int pending);
333static int filt_aioattach(struct knote *kn);
334static void filt_aiodetach(struct knote *kn);
335static int filt_aio(struct knote *kn, long hint);
336static int filt_lioattach(struct knote *kn);
337static void filt_liodetach(struct knote *kn);
338static int filt_lio(struct knote *kn, long hint);
339
340/*
341 * Zones for:
342 * kaio Per process async io info
343 * aiop async io process data
344 * aiocb async io jobs
345 * aiolio list io jobs
346 */
347static uma_zone_t kaio_zone, aiop_zone, aiocb_zone, aiolio_zone;
348
349/* kqueue filters for aio */
350static struct filterops aio_filtops = {
351 .f_isfd = 0,
352 .f_attach = filt_aioattach,
353 .f_detach = filt_aiodetach,
354 .f_event = filt_aio,
355};
356static struct filterops lio_filtops = {
357 .f_isfd = 0,
358 .f_attach = filt_lioattach,
359 .f_detach = filt_liodetach,
360 .f_event = filt_lio
361};
362
363static eventhandler_tag exit_tag, exec_tag;
364
365TASKQUEUE_DEFINE_THREAD(aiod_kick);
366
367/*
368 * Main operations function for use as a kernel module.
369 */
370static int
371aio_modload(struct module *module, int cmd, void *arg)
372{
373 int error = 0;
374
375 switch (cmd) {
376 case MOD_LOAD:
377 aio_onceonly();
378 break;
379 case MOD_SHUTDOWN:
380 break;
381 default:
382 error = EOPNOTSUPP;
383 break;
384 }
385 return (error);
386}
387
388static moduledata_t aio_mod = {
389 "aio",
390 &aio_modload,
391 NULL
392};
393
394DECLARE_MODULE(aio, aio_mod, SI_SUB_VFS, SI_ORDER_ANY);
395MODULE_VERSION(aio, 1);
396
397/*
398 * Startup initialization
399 */
400static int
401aio_onceonly(void)
402{
403
404 exit_tag = EVENTHANDLER_REGISTER(process_exit, aio_proc_rundown, NULL,
405 EVENTHANDLER_PRI_ANY);
406 exec_tag = EVENTHANDLER_REGISTER(process_exec, aio_proc_rundown_exec,
407 NULL, EVENTHANDLER_PRI_ANY);
408 kqueue_add_filteropts(EVFILT_AIO, &aio_filtops);
409 kqueue_add_filteropts(EVFILT_LIO, &lio_filtops);
410 TAILQ_INIT(&aio_freeproc);
411 sema_init(&aio_newproc_sem, 0, "aio_new_proc");
412 mtx_init(&aio_job_mtx, "aio_job", NULL, MTX_DEF);
413 TAILQ_INIT(&aio_jobs);
414 aiod_unr = new_unrhdr(1, INT_MAX, NULL);
415 kaio_zone = uma_zcreate("AIO", sizeof(struct kaioinfo), NULL, NULL,
416 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
417 aiop_zone = uma_zcreate("AIOP", sizeof(struct aioproc), NULL,
418 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
419 aiocb_zone = uma_zcreate("AIOCB", sizeof(struct kaiocb), NULL, NULL,
420 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
421 aiolio_zone = uma_zcreate("AIOLIO", sizeof(struct aioliojob), NULL,
422 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE);
423 aiod_lifetime = AIOD_LIFETIME_DEFAULT;
424 jobrefid = 1;
425 p31b_setcfg(CTL_P1003_1B_ASYNCHRONOUS_IO, _POSIX_ASYNCHRONOUS_IO);
426 p31b_setcfg(CTL_P1003_1B_AIO_MAX, MAX_AIO_QUEUE);
427 p31b_setcfg(CTL_P1003_1B_AIO_PRIO_DELTA_MAX, 0);
428
429 return (0);
430}
431
432/*
433 * Init the per-process aioinfo structure. The aioinfo limits are set
434 * per-process for user limit (resource) management.
435 */
436void
437aio_init_aioinfo(struct proc *p)
438{
439 struct kaioinfo *ki;
440
441 ki = uma_zalloc(kaio_zone, M_WAITOK);
442 mtx_init(&ki->kaio_mtx, "aiomtx", NULL, MTX_DEF | MTX_NEW);
443 ki->kaio_flags = 0;
444 ki->kaio_active_count = 0;
445 ki->kaio_count = 0;
446 ki->kaio_buffer_count = 0;
447 TAILQ_INIT(&ki->kaio_all);
448 TAILQ_INIT(&ki->kaio_done);
449 TAILQ_INIT(&ki->kaio_jobqueue);
450 TAILQ_INIT(&ki->kaio_liojoblist);
451 TAILQ_INIT(&ki->kaio_syncqueue);
452 TAILQ_INIT(&ki->kaio_syncready);
453 TASK_INIT(&ki->kaio_task, 0, aio_kick_helper, p);
454 TASK_INIT(&ki->kaio_sync_task, 0, aio_schedule_fsync, ki);
455 PROC_LOCK(p);
456 if (p->p_aioinfo == NULL) {
457 p->p_aioinfo = ki;
458 PROC_UNLOCK(p);
459 } else {
460 PROC_UNLOCK(p);
461 mtx_destroy(&ki->kaio_mtx);
462 uma_zfree(kaio_zone, ki);
463 }
464
465 while (num_aio_procs < MIN(target_aio_procs, max_aio_procs))
466 aio_newproc(NULL);
467}
468
469static int
470aio_sendsig(struct proc *p, struct sigevent *sigev, ksiginfo_t *ksi, bool ext)
471{
472 struct thread *td;
473 int error;
474
475 error = sigev_findtd(p, sigev, &td);
476 if (error)
477 return (error);
478 if (!KSI_ONQ(ksi)) {
479 ksiginfo_set_sigev(ksi, sigev);
480 ksi->ksi_code = SI_ASYNCIO;
481 ksi->ksi_flags |= ext ? (KSI_EXT | KSI_INS) : 0;
482 tdsendsignal(p, td, ksi->ksi_signo, ksi);
483 }
484 PROC_UNLOCK(p);
485 return (error);
486}
487
488/*
489 * Free a job entry. Wait for completion if it is currently active, but don't
490 * delay forever. If we delay, we return a flag that says that we have to
491 * restart the queue scan.
492 */
493static int
494aio_free_entry(struct kaiocb *job)
495{
496 struct kaioinfo *ki;
497 struct aioliojob *lj;
498 struct proc *p;
499
500 p = job->userproc;
501 MPASS(curproc == p);
502 ki = p->p_aioinfo;
503 MPASS(ki != NULL);
504
505 AIO_LOCK_ASSERT(ki, MA_OWNED);
506 MPASS(job->jobflags & KAIOCB_FINISHED);
507
508 atomic_subtract_int(&num_queue_count, 1);
509
510 ki->kaio_count--;
511 MPASS(ki->kaio_count >= 0);
512
513 TAILQ_REMOVE(&ki->kaio_done, job, plist);
514 TAILQ_REMOVE(&ki->kaio_all, job, allist);
515
516 lj = job->lio;
517 if (lj) {
518 lj->lioj_count--;
519 lj->lioj_finished_count--;
520
521 if (lj->lioj_count == 0) {
522 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list);
523 /* lio is going away, we need to destroy any knotes */
524 knlist_delete(&lj->klist, curthread, 1);
525 PROC_LOCK(p);
526 sigqueue_take(&lj->lioj_ksi);
527 PROC_UNLOCK(p);
528 uma_zfree(aiolio_zone, lj);
529 }
530 }
531
532 /* job is going away, we need to destroy any knotes */
533 knlist_delete(&job->klist, curthread, 1);
534 PROC_LOCK(p);
535 sigqueue_take(&job->ksi);
536 PROC_UNLOCK(p);
537
538 AIO_UNLOCK(ki);
539
540 /*
541 * The thread argument here is used to find the owning process
542 * and is also passed to fo_close() which may pass it to various
543 * places such as devsw close() routines. Because of that, we
544 * need a thread pointer from the process owning the job that is
545 * persistent and won't disappear out from under us or move to
546 * another process.
547 *
548 * Currently, all the callers of this function call it to remove
549 * a kaiocb from the current process' job list either via a
550 * syscall or due to the current process calling exit() or
551 * execve(). Thus, we know that p == curproc. We also know that
552 * curthread can't exit since we are curthread.
553 *
554 * Therefore, we use curthread as the thread to pass to
555 * knlist_delete(). This does mean that it is possible for the
556 * thread pointer at close time to differ from the thread pointer
557 * at open time, but this is already true of file descriptors in
558 * a multithreaded process.
559 */
560 if (job->fd_file)
561 fdrop(job->fd_file, curthread);
562 crfree(job->cred);
563 if (job->uiop != &job->uio)
564 free(job->uiop, M_IOV);
565 uma_zfree(aiocb_zone, job);
566 AIO_LOCK(ki);
567
568 return (0);
569}
570
571static void
572aio_proc_rundown_exec(void *arg, struct proc *p,
573 struct image_params *imgp __unused)
574{
575 aio_proc_rundown(arg, p);
576}
577
578static int
579aio_cancel_job(struct proc *p, struct kaioinfo *ki, struct kaiocb *job)
580{
581 aio_cancel_fn_t *func;
582 int cancelled;
583
584 AIO_LOCK_ASSERT(ki, MA_OWNED);
585 if (job->jobflags & (KAIOCB_CANCELLED | KAIOCB_FINISHED))
586 return (0);
587 MPASS((job->jobflags & KAIOCB_CANCELLING) == 0);
588 job->jobflags |= KAIOCB_CANCELLED;
589
590 func = job->cancel_fn;
591
592 /*
593 * If there is no cancel routine, just leave the job marked as
594 * cancelled. The job should be in active use by a caller who
595 * should complete it normally or when it fails to install a
596 * cancel routine.
597 */
598 if (func == NULL)
599 return (0);
600
601 /*
602 * Set the CANCELLING flag so that aio_complete() will defer
603 * completions of this job. This prevents the job from being
604 * freed out from under the cancel callback. After the
605 * callback any deferred completion (whether from the callback
606 * or any other source) will be completed.
607 */
608 job->jobflags |= KAIOCB_CANCELLING;
609 AIO_UNLOCK(ki);
610 func(job);
611 AIO_LOCK(ki);
612 job->jobflags &= ~KAIOCB_CANCELLING;
613 if (job->jobflags & KAIOCB_FINISHED) {
614 cancelled = job->uaiocb._aiocb_private.error == ECANCELED;
615 TAILQ_REMOVE(&ki->kaio_jobqueue, job, plist);
616 aio_bio_done_notify(p, job);
617 } else {
618 /*
619 * The cancel callback might have scheduled an
620 * operation to cancel this request, but it is
621 * only counted as cancelled if the request is
622 * cancelled when the callback returns.
623 */
624 cancelled = 0;
625 }
626 return (cancelled);
627}
628
629/*
630 * Rundown the jobs for a given process.
631 */
632static void
633aio_proc_rundown(void *arg, struct proc *p)
634{
635 struct kaioinfo *ki;
636 struct aioliojob *lj;
637 struct kaiocb *job, *jobn;
638
639 KASSERT(curthread->td_proc == p,
640 ("%s: called on non-curproc", __func__));
641 ki = p->p_aioinfo;
642 if (ki == NULL)
643 return;
644
645 AIO_LOCK(ki);
646 ki->kaio_flags |= KAIO_RUNDOWN;
647
648restart:
649
650 /*
651 * Try to cancel all pending requests. This code simulates
652 * aio_cancel on all pending I/O requests.
653 */
654 TAILQ_FOREACH_SAFE(job, &ki->kaio_jobqueue, plist, jobn) {
655 aio_cancel_job(p, ki, job);
656 }
657
658 /* Wait for all running I/O to be finished */
659 if (TAILQ_FIRST(&ki->kaio_jobqueue) || ki->kaio_active_count != 0) {
660 ki->kaio_flags |= KAIO_WAKEUP;
661 msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO, "aioprn", hz);
662 goto restart;
663 }
664
665 /* Free all completed I/O requests. */
666 while ((job = TAILQ_FIRST(&ki->kaio_done)) != NULL)
667 aio_free_entry(job);
668
669 while ((lj = TAILQ_FIRST(&ki->kaio_liojoblist)) != NULL) {
670 if (lj->lioj_count == 0) {
671 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list);
672 knlist_delete(&lj->klist, curthread, 1);
673 PROC_LOCK(p);
674 sigqueue_take(&lj->lioj_ksi);
675 PROC_UNLOCK(p);
676 uma_zfree(aiolio_zone, lj);
677 } else {
678 panic("LIO job not cleaned up: C:%d, FC:%d\n",
679 lj->lioj_count, lj->lioj_finished_count);
680 }
681 }
682 AIO_UNLOCK(ki);
683 taskqueue_drain(taskqueue_aiod_kick, &ki->kaio_task);
684 taskqueue_drain(taskqueue_aiod_kick, &ki->kaio_sync_task);
685 mtx_destroy(&ki->kaio_mtx);
686 uma_zfree(kaio_zone, ki);
687 p->p_aioinfo = NULL;
688}
689
690/*
691 * Select a job to run (called by an AIO daemon).
692 */
693static struct kaiocb *
694aio_selectjob(struct aioproc *aiop)
695{
696 struct kaiocb *job;
697 struct kaioinfo *ki;
698 struct proc *userp;
699
700 mtx_assert(&aio_job_mtx, MA_OWNED);
701restart:
702 TAILQ_FOREACH(job, &aio_jobs, list) {
703 userp = job->userproc;
704 ki = userp->p_aioinfo;
705
706 if (ki->kaio_active_count < max_aio_per_proc) {
707 TAILQ_REMOVE(&aio_jobs, job, list);
708 if (!aio_clear_cancel_function(job))
709 goto restart;
710
711 /* Account for currently active jobs. */
712 ki->kaio_active_count++;
713 break;
714 }
715 }
716 return (job);
717}
718
719/*
720 * Move all data to a permanent storage device. This code
721 * simulates the fsync and fdatasync syscalls.
722 */
723static int
724aio_fsync_vnode(struct thread *td, struct vnode *vp, int op)
725{
726 struct mount *mp;
727 vm_object_t obj;
728 int error;
729
730 for (;;) {
731 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
732 if (error != 0)
733 break;
734 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
735 obj = vp->v_object;
736 if (obj != NULL) {
737 VM_OBJECT_WLOCK(obj);
738 vm_object_page_clean(obj, 0, 0, 0);
739 VM_OBJECT_WUNLOCK(obj);
740 }
741 if (op == LIO_DSYNC)
742 error = VOP_FDATASYNC(vp, td);
743 else
744 error = VOP_FSYNC(vp, MNT_WAIT, td);
745
746 VOP_UNLOCK(vp);
747 vn_finished_write(mp);
748 if (error != ERELOOKUP)
749 break;
750 }
751 return (error);
752}
753
754/*
755 * The AIO processing activity for LIO_READ/LIO_WRITE. This is the code that
756 * does the I/O request for the non-bio version of the operations. The normal
757 * vn operations are used, and this code should work in all instances for every
758 * type of file, including pipes, sockets, fifos, and regular files.
759 *
760 * XXX I don't think it works well for socket, pipe, and fifo.
761 */
762static void
763aio_process_rw(struct kaiocb *job)
764{
765 struct ucred *td_savedcred;
766 struct thread *td;
767 struct file *fp;
768 ssize_t cnt;
769 long msgsnd_st, msgsnd_end;
770 long msgrcv_st, msgrcv_end;
771 long oublock_st, oublock_end;
772 long inblock_st, inblock_end;
773 int error, opcode;
774
775 KASSERT(job->uaiocb.aio_lio_opcode == LIO_READ ||
776 job->uaiocb.aio_lio_opcode == LIO_READV ||
777 job->uaiocb.aio_lio_opcode == LIO_WRITE ||
778 job->uaiocb.aio_lio_opcode == LIO_WRITEV,
779 ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode));
780
781 aio_switch_vmspace(job);
782 td = curthread;
783 td_savedcred = td->td_ucred;
784 td->td_ucred = job->cred;
785 job->uiop->uio_td = td;
786 fp = job->fd_file;
787
788 opcode = job->uaiocb.aio_lio_opcode;
789 cnt = job->uiop->uio_resid;
790
791 msgrcv_st = td->td_ru.ru_msgrcv;
792 msgsnd_st = td->td_ru.ru_msgsnd;
793 inblock_st = td->td_ru.ru_inblock;
794 oublock_st = td->td_ru.ru_oublock;
795
796 /*
797 * aio_aqueue() acquires a reference to the file that is
798 * released in aio_free_entry().
799 */
800 if (opcode == LIO_READ || opcode == LIO_READV) {
801 if (job->uiop->uio_resid == 0)
802 error = 0;
803 else
804 error = fo_read(fp, job->uiop, fp->f_cred, FOF_OFFSET,
805 td);
806 } else {
807 if (fp->f_type == DTYPE_VNODE)
808 bwillwrite();
809 error = fo_write(fp, job->uiop, fp->f_cred, FOF_OFFSET, td);
810 }
811 msgrcv_end = td->td_ru.ru_msgrcv;
812 msgsnd_end = td->td_ru.ru_msgsnd;
813 inblock_end = td->td_ru.ru_inblock;
814 oublock_end = td->td_ru.ru_oublock;
815
816 job->msgrcv = msgrcv_end - msgrcv_st;
817 job->msgsnd = msgsnd_end - msgsnd_st;
818 job->inblock = inblock_end - inblock_st;
819 job->outblock = oublock_end - oublock_st;
820
821 if (error != 0 && job->uiop->uio_resid != cnt) {
822 if (error == ERESTART || error == EINTR || error == EWOULDBLOCK)
823 error = 0;
824 if (error == EPIPE && (opcode & LIO_WRITE)) {
825 PROC_LOCK(job->userproc);
826 kern_psignal(job->userproc, SIGPIPE);
827 PROC_UNLOCK(job->userproc);
828 }
829 }
830
831 cnt -= job->uiop->uio_resid;
832 td->td_ucred = td_savedcred;
833 if (error)
834 aio_complete(job, -1, error);
835 else
836 aio_complete(job, cnt, 0);
837}
838
839static void
840aio_process_sync(struct kaiocb *job)
841{
842 struct thread *td = curthread;
843 struct ucred *td_savedcred = td->td_ucred;
844 struct file *fp = job->fd_file;
845 int error = 0;
846
847 KASSERT(job->uaiocb.aio_lio_opcode & LIO_SYNC,
848 ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode));
849
850 td->td_ucred = job->cred;
851 if (fp->f_vnode != NULL) {
852 error = aio_fsync_vnode(td, fp->f_vnode,
853 job->uaiocb.aio_lio_opcode);
854 }
855 td->td_ucred = td_savedcred;
856 if (error)
857 aio_complete(job, -1, error);
858 else
859 aio_complete(job, 0, 0);
860}
861
862static void
863aio_process_mlock(struct kaiocb *job)
864{
865 struct aiocb *cb = &job->uaiocb;
866 int error;
867
868 KASSERT(job->uaiocb.aio_lio_opcode == LIO_MLOCK,
869 ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode));
870
871 aio_switch_vmspace(job);
872 error = kern_mlock(job->userproc, job->cred,
873 __DEVOLATILE(uintptr_t, cb->aio_buf), cb->aio_nbytes);
874 aio_complete(job, error != 0 ? -1 : 0, error);
875}
876
877static void
878aio_bio_done_notify(struct proc *userp, struct kaiocb *job)
879{
880 struct aioliojob *lj;
881 struct kaioinfo *ki;
882 struct kaiocb *sjob, *sjobn;
883 int lj_done;
884 bool schedule_fsync;
885
886 ki = userp->p_aioinfo;
887 AIO_LOCK_ASSERT(ki, MA_OWNED);
888 lj = job->lio;
889 lj_done = 0;
890 if (lj) {
891 lj->lioj_finished_count++;
892 if (lj->lioj_count == lj->lioj_finished_count)
893 lj_done = 1;
894 }
895 TAILQ_INSERT_TAIL(&ki->kaio_done, job, plist);
896 MPASS(job->jobflags & KAIOCB_FINISHED);
897
898 if (ki->kaio_flags & KAIO_RUNDOWN)
899 goto notification_done;
900
901 if (job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL ||
902 job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID)
903 aio_sendsig(userp, &job->uaiocb.aio_sigevent, &job->ksi, true);
904
905 KNOTE_LOCKED(&job->klist, 1);
906
907 if (lj_done) {
908 if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) {
909 lj->lioj_flags |= LIOJ_KEVENT_POSTED;
910 KNOTE_LOCKED(&lj->klist, 1);
911 }
912 if ((lj->lioj_flags & (LIOJ_SIGNAL | LIOJ_SIGNAL_POSTED))
913 == LIOJ_SIGNAL &&
914 (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL ||
915 lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID)) {
916 aio_sendsig(userp, &lj->lioj_signal, &lj->lioj_ksi,
917 true);
918 lj->lioj_flags |= LIOJ_SIGNAL_POSTED;
919 }
920 }
921
922notification_done:
923 if (job->jobflags & KAIOCB_CHECKSYNC) {
924 schedule_fsync = false;
925 TAILQ_FOREACH_SAFE(sjob, &ki->kaio_syncqueue, list, sjobn) {
926 if (job->fd_file != sjob->fd_file ||
927 job->seqno >= sjob->seqno)
928 continue;
929 if (--sjob->pending > 0)
930 continue;
931 TAILQ_REMOVE(&ki->kaio_syncqueue, sjob, list);
932 if (!aio_clear_cancel_function_locked(sjob))
933 continue;
934 TAILQ_INSERT_TAIL(&ki->kaio_syncready, sjob, list);
935 schedule_fsync = true;
936 }
937 if (schedule_fsync)
938 taskqueue_enqueue(taskqueue_aiod_kick,
939 &ki->kaio_sync_task);
940 }
941 if (ki->kaio_flags & KAIO_WAKEUP) {
942 ki->kaio_flags &= ~KAIO_WAKEUP;
943 wakeup(&userp->p_aioinfo);
944 }
945}
946
947static void
948aio_schedule_fsync(void *context, int pending)
949{
950 struct kaioinfo *ki;
951 struct kaiocb *job;
952
953 ki = context;
954 AIO_LOCK(ki);
955 while (!TAILQ_EMPTY(&ki->kaio_syncready)) {
956 job = TAILQ_FIRST(&ki->kaio_syncready);
957 TAILQ_REMOVE(&ki->kaio_syncready, job, list);
958 AIO_UNLOCK(ki);
959 aio_schedule(job, aio_process_sync);
960 AIO_LOCK(ki);
961 }
962 AIO_UNLOCK(ki);
963}
964
965bool
966aio_cancel_cleared(struct kaiocb *job)
967{
968
969 /*
970 * The caller should hold the same queue lock held when
971 * aio_clear_cancel_function() was called and set this flag
972 * ensuring this check sees an up-to-date value. However,
973 * there is no way to assert that.
974 */
975 return ((job->jobflags & KAIOCB_CLEARED) != 0);
976}
977
978static bool
979aio_clear_cancel_function_locked(struct kaiocb *job)
980{
981
982 AIO_LOCK_ASSERT(job->userproc->p_aioinfo, MA_OWNED);
983 MPASS(job->cancel_fn != NULL);
984 if (job->jobflags & KAIOCB_CANCELLING) {
985 job->jobflags |= KAIOCB_CLEARED;
986 return (false);
987 }
988 job->cancel_fn = NULL;
989 return (true);
990}
991
992bool
993aio_clear_cancel_function(struct kaiocb *job)
994{
995 struct kaioinfo *ki;
996 bool ret;
997
998 ki = job->userproc->p_aioinfo;
999 AIO_LOCK(ki);
1000 ret = aio_clear_cancel_function_locked(job);
1001 AIO_UNLOCK(ki);
1002 return (ret);
1003}
1004
1005static bool
1006aio_set_cancel_function_locked(struct kaiocb *job, aio_cancel_fn_t *func)
1007{
1008
1009 AIO_LOCK_ASSERT(job->userproc->p_aioinfo, MA_OWNED);
1010 if (job->jobflags & KAIOCB_CANCELLED)
1011 return (false);
1012 job->cancel_fn = func;
1013 return (true);
1014}
1015
1016bool
1017aio_set_cancel_function(struct kaiocb *job, aio_cancel_fn_t *func)
1018{
1019 struct kaioinfo *ki;
1020 bool ret;
1021
1022 ki = job->userproc->p_aioinfo;
1023 AIO_LOCK(ki);
1024 ret = aio_set_cancel_function_locked(job, func);
1025 AIO_UNLOCK(ki);
1026 return (ret);
1027}
1028
1029void
1030aio_complete(struct kaiocb *job, long status, int error)
1031{
1032 struct kaioinfo *ki;
1033 struct proc *userp;
1034
1035 job->uaiocb._aiocb_private.error = error;
1036 job->uaiocb._aiocb_private.status = status;
1037
1038 userp = job->userproc;
1039 ki = userp->p_aioinfo;
1040
1041 AIO_LOCK(ki);
1042 KASSERT(!(job->jobflags & KAIOCB_FINISHED),
1043 ("duplicate aio_complete"));
1044 job->jobflags |= KAIOCB_FINISHED;
1045 if ((job->jobflags & (KAIOCB_QUEUEING | KAIOCB_CANCELLING)) == 0) {
1046 TAILQ_REMOVE(&ki->kaio_jobqueue, job, plist);
1047 aio_bio_done_notify(userp, job);
1048 }
1049 AIO_UNLOCK(ki);
1050}
1051
1052void
1053aio_cancel(struct kaiocb *job)
1054{
1055
1056 aio_complete(job, -1, ECANCELED);
1057}
1058
1059void
1060aio_switch_vmspace(struct kaiocb *job)
1061{
1062
1063 vmspace_switch_aio(job->userproc->p_vmspace);
1064}
1065
1066/*
1067 * The AIO daemon, most of the actual work is done in aio_process_*,
1068 * but the setup (and address space mgmt) is done in this routine.
1069 */
1070static void
1071aio_daemon(void *_id)
1072{
1073 struct kaiocb *job;
1074 struct aioproc *aiop;
1075 struct kaioinfo *ki;
1076 struct proc *p;
1077 struct vmspace *myvm;
1078 struct thread *td = curthread;
1079 int id = (intptr_t)_id;
1080
1081 /*
1082 * Grab an extra reference on the daemon's vmspace so that it
1083 * doesn't get freed by jobs that switch to a different
1084 * vmspace.
1085 */
1086 p = td->td_proc;
1087 myvm = vmspace_acquire_ref(p);
1088
1089 KASSERT(p->p_textvp == NULL, ("kthread has a textvp"));
1090
1091 /*
1092 * Allocate and ready the aio control info. There is one aiop structure
1093 * per daemon.
1094 */
1095 aiop = uma_zalloc(aiop_zone, M_WAITOK);
1096 aiop->aioproc = p;
1097 aiop->aioprocflags = 0;
1098
1099 /*
1100 * Wakeup parent process. (Parent sleeps to keep from blasting away
1101 * and creating too many daemons.)
1102 */
1103 sema_post(&aio_newproc_sem);
1104
1105 mtx_lock(&aio_job_mtx);
1106 for (;;) {
1107 /*
1108 * Take daemon off of free queue
1109 */
1110 if (aiop->aioprocflags & AIOP_FREE) {
1111 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1112 aiop->aioprocflags &= ~AIOP_FREE;
1113 }
1114
1115 /*
1116 * Check for jobs.
1117 */
1118 while ((job = aio_selectjob(aiop)) != NULL) {
1119 mtx_unlock(&aio_job_mtx);
1120
1121 ki = job->userproc->p_aioinfo;
1122 job->handle_fn(job);
1123
1124 mtx_lock(&aio_job_mtx);
1125 /* Decrement the active job count. */
1126 ki->kaio_active_count--;
1127 }
1128
1129 /*
1130 * Disconnect from user address space.
1131 */
1132 if (p->p_vmspace != myvm) {
1133 mtx_unlock(&aio_job_mtx);
1134 vmspace_switch_aio(myvm);
1135 mtx_lock(&aio_job_mtx);
1136 /*
1137 * We have to restart to avoid race, we only sleep if
1138 * no job can be selected.
1139 */
1140 continue;
1141 }
1142
1143 mtx_assert(&aio_job_mtx, MA_OWNED);
1144
1145 TAILQ_INSERT_HEAD(&aio_freeproc, aiop, list);
1146 aiop->aioprocflags |= AIOP_FREE;
1147
1148 /*
1149 * If daemon is inactive for a long time, allow it to exit,
1150 * thereby freeing resources.
1151 */
1152 if (msleep(p, &aio_job_mtx, PRIBIO, "aiordy",
1153 aiod_lifetime) == EWOULDBLOCK && TAILQ_EMPTY(&aio_jobs) &&
1154 (aiop->aioprocflags & AIOP_FREE) &&
1155 num_aio_procs > target_aio_procs)
1156 break;
1157 }
1158 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1159 num_aio_procs--;
1160 mtx_unlock(&aio_job_mtx);
1161 uma_zfree(aiop_zone, aiop);
1162 free_unr(aiod_unr, id);
1163 vmspace_free(myvm);
1164
1165 KASSERT(p->p_vmspace == myvm,
1166 ("AIOD: bad vmspace for exiting daemon"));
1167 KASSERT(refcount_load(&myvm->vm_refcnt) > 1,
1168 ("AIOD: bad vm refcnt for exiting daemon: %d",
1169 refcount_load(&myvm->vm_refcnt)));
1170 kproc_exit(0);
1171}
1172
1173/*
1174 * Create a new AIO daemon. This is mostly a kernel-thread fork routine. The
1175 * AIO daemon modifies its environment itself.
1176 */
1177static int
1178aio_newproc(int *start)
1179{
1180 int error;
1181 struct proc *p;
1182 int id;
1183
1184 id = alloc_unr(aiod_unr);
1185 error = kproc_create(aio_daemon, (void *)(intptr_t)id, &p,
1186 RFNOWAIT, 0, "aiod%d", id);
1187 if (error == 0) {
1188 /*
1189 * Wait until daemon is started.
1190 */
1191 sema_wait(&aio_newproc_sem);
1192 mtx_lock(&aio_job_mtx);
1193 num_aio_procs++;
1194 if (start != NULL)
1195 (*start)--;
1196 mtx_unlock(&aio_job_mtx);
1197 } else {
1198 free_unr(aiod_unr, id);
1199 }
1200 return (error);
1201}
1202
1203/*
1204 * Try the high-performance, low-overhead bio method for eligible
1205 * VCHR devices. This method doesn't use an aio helper thread, and
1206 * thus has very low overhead.
1207 *
1208 * Assumes that the caller, aio_aqueue(), has incremented the file
1209 * structure's reference count, preventing its deallocation for the
1210 * duration of this call.
1211 */
1212static int
1213aio_qbio(struct proc *p, struct kaiocb *job)
1214{
1215 struct aiocb *cb;
1216 struct file *fp;
1217 struct buf *pbuf;
1218 struct vnode *vp;
1219 struct cdevsw *csw;
1220 struct cdev *dev;
1221 struct kaioinfo *ki;
1222 struct bio **bios = NULL;
1223 off_t offset;
1224 int bio_cmd, error, i, iovcnt, opcode, poff, ref;
1225 vm_prot_t prot;
1226 bool use_unmapped;
1227
1228 cb = &job->uaiocb;
1229 fp = job->fd_file;
1230 opcode = cb->aio_lio_opcode;
1231
1232 if (!(opcode == LIO_WRITE || opcode == LIO_WRITEV ||
1233 opcode == LIO_READ || opcode == LIO_READV))
1234 return (-1);
1235 if (fp == NULL || fp->f_type != DTYPE_VNODE)
1236 return (-1);
1237
1238 vp = fp->f_vnode;
1239 if (vp->v_type != VCHR)
1240 return (-1);
1241 if (vp->v_bufobj.bo_bsize == 0)
1242 return (-1);
1243
1244 bio_cmd = (opcode & LIO_WRITE) ? BIO_WRITE : BIO_READ;
1245 iovcnt = job->uiop->uio_iovcnt;
1246 if (iovcnt > max_buf_aio)
1247 return (-1);
1248 for (i = 0; i < iovcnt; i++) {
1249 if (job->uiop->uio_iov[i].iov_len % vp->v_bufobj.bo_bsize != 0)
1250 return (-1);
1251 if (job->uiop->uio_iov[i].iov_len > maxphys) {
1252 error = -1;
1253 return (-1);
1254 }
1255 }
1256 offset = cb->aio_offset;
1257
1258 ref = 0;
1259 csw = devvn_refthread(vp, &dev, &ref);
1260 if (csw == NULL)
1261 return (ENXIO);
1262
1263 if ((csw->d_flags & D_DISK) == 0) {
1264 error = -1;
1265 goto unref;
1266 }
1267 if (job->uiop->uio_resid > dev->si_iosize_max) {
1268 error = -1;
1269 goto unref;
1270 }
1271
1272 ki = p->p_aioinfo;
1273 job->error = 0;
1274
1275 use_unmapped = (dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed;
1276 if (!use_unmapped) {
1277 AIO_LOCK(ki);
1278 if (ki->kaio_buffer_count + iovcnt > max_buf_aio) {
1279 AIO_UNLOCK(ki);
1280 error = EAGAIN;
1281 goto unref;
1282 }
1283 ki->kaio_buffer_count += iovcnt;
1284 AIO_UNLOCK(ki);
1285 }
1286
1287 bios = malloc(sizeof(struct bio *) * iovcnt, M_TEMP, M_WAITOK);
1288 atomic_store_int(&job->nbio, iovcnt);
1289 for (i = 0; i < iovcnt; i++) {
1290 struct vm_page** pages;
1291 struct bio *bp;
1292 void *buf;
1293 size_t nbytes;
1294 int npages;
1295
1296 buf = job->uiop->uio_iov[i].iov_base;
1297 nbytes = job->uiop->uio_iov[i].iov_len;
1298
1299 bios[i] = g_alloc_bio();
1300 bp = bios[i];
1301
1302 poff = (vm_offset_t)buf & PAGE_MASK;
1303 if (use_unmapped) {
1304 pbuf = NULL;
1305 pages = malloc(sizeof(vm_page_t) * (atop(round_page(
1306 nbytes)) + 1), M_TEMP, M_WAITOK | M_ZERO);
1307 } else {
1308 pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
1309 BUF_KERNPROC(pbuf);
1310 pages = pbuf->b_pages;
1311 }
1312
1313 bp->bio_length = nbytes;
1314 bp->bio_bcount = nbytes;
1315 bp->bio_done = aio_biowakeup;
1316 bp->bio_offset = offset;
1317 bp->bio_cmd = bio_cmd;
1318 bp->bio_dev = dev;
1319 bp->bio_caller1 = job;
1320 bp->bio_caller2 = pbuf;
1321
1322 prot = VM_PROT_READ;
1323 if (opcode == LIO_READ || opcode == LIO_READV)
1324 prot |= VM_PROT_WRITE; /* Less backwards than it looks */
1325 npages = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map,
1326 (vm_offset_t)buf, bp->bio_length, prot, pages,
1327 atop(maxphys) + 1);
1328 if (npages < 0) {
1329 if (pbuf != NULL)
1330 uma_zfree(pbuf_zone, pbuf);
1331 else
1332 free(pages, M_TEMP);
1333 error = EFAULT;
1334 g_destroy_bio(bp);
1335 i--;
1336 goto destroy_bios;
1337 }
1338 if (pbuf != NULL) {
1339 pmap_qenter((vm_offset_t)pbuf->b_data, pages, npages);
1340 bp->bio_data = pbuf->b_data + poff;
1341 pbuf->b_npages = npages;
1342 atomic_add_int(&num_buf_aio, 1);
1343 } else {
1344 bp->bio_ma = pages;
1345 bp->bio_ma_n = npages;
1346 bp->bio_ma_offset = poff;
1347 bp->bio_data = unmapped_buf;
1348 bp->bio_flags |= BIO_UNMAPPED;
1349 atomic_add_int(&num_unmapped_aio, 1);
1350 }
1351
1352 offset += nbytes;
1353 }
1354
1355 /* Perform transfer. */
1356 for (i = 0; i < iovcnt; i++)
1357 csw->d_strategy(bios[i]);
1358 free(bios, M_TEMP);
1359
1360 dev_relthread(dev, ref);
1361 return (0);
1362
1363destroy_bios:
1364 for (; i >= 0; i--)
1365 aio_biocleanup(bios[i]);
1366 free(bios, M_TEMP);
1367unref:
1368 dev_relthread(dev, ref);
1369 return (error);
1370}
1371
1372#ifdef COMPAT_FREEBSD6
1373static int
1374convert_old_sigevent(struct osigevent *osig, struct sigevent *nsig)
1375{
1376
1377 /*
1378 * Only SIGEV_NONE, SIGEV_SIGNAL, and SIGEV_KEVENT are
1379 * supported by AIO with the old sigevent structure.
1380 */
1381 nsig->sigev_notify = osig->sigev_notify;
1382 switch (nsig->sigev_notify) {
1383 case SIGEV_NONE:
1384 break;
1385 case SIGEV_SIGNAL:
1386 nsig->sigev_signo = osig->__sigev_u.__sigev_signo;
1387 break;
1388 case SIGEV_KEVENT:
1389 nsig->sigev_notify_kqueue =
1390 osig->__sigev_u.__sigev_notify_kqueue;
1391 nsig->sigev_value.sival_ptr = osig->sigev_value.sival_ptr;
1392 break;
1393 default:
1394 return (EINVAL);
1395 }
1396 return (0);
1397}
1398
1399static int
1400aiocb_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob,
1401 int type __unused)
1402{
1403 struct oaiocb *ojob;
1404 struct aiocb *kcb = &kjob->uaiocb;
1405 int error;
1406
1407 bzero(kcb, sizeof(struct aiocb));
1408 error = copyin(ujob, kcb, sizeof(struct oaiocb));
1409 if (error)
1410 return (error);
1411 /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */
1412 ojob = (struct oaiocb *)kcb;
1413 return (convert_old_sigevent(&ojob->aio_sigevent, &kcb->aio_sigevent));
1414}
1415#endif
1416
1417static int
1418aiocb_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type)
1419{
1420 struct aiocb *kcb = &kjob->uaiocb;
1421 int error;
1422
1423 error = copyin(ujob, kcb, sizeof(struct aiocb));
1424 if (error)
1425 return (error);
1426 if (type == LIO_NOP)
1427 type = kcb->aio_lio_opcode;
1428 if (type & LIO_VECTORED) {
1429 /* malloc a uio and copy in the iovec */
1430 error = copyinuio(__DEVOLATILE(struct iovec*, kcb->aio_iov),
1431 kcb->aio_iovcnt, &kjob->uiop);
1432 }
1433
1434 return (error);
1435}
1436
1437static long
1438aiocb_fetch_status(struct aiocb *ujob)
1439{
1440
1441 return (fuword(&ujob->_aiocb_private.status));
1442}
1443
1444static long
1445aiocb_fetch_error(struct aiocb *ujob)
1446{
1447
1448 return (fuword(&ujob->_aiocb_private.error));
1449}
1450
1451static int
1452aiocb_store_status(struct aiocb *ujob, long status)
1453{
1454
1455 return (suword(&ujob->_aiocb_private.status, status));
1456}
1457
1458static int
1459aiocb_store_error(struct aiocb *ujob, long error)
1460{
1461
1462 return (suword(&ujob->_aiocb_private.error, error));
1463}
1464
1465static int
1466aiocb_store_kernelinfo(struct aiocb *ujob, long jobref)
1467{
1468
1469 return (suword(&ujob->_aiocb_private.kernelinfo, jobref));
1470}
1471
1472static int
1473aiocb_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob)
1474{
1475
1476 return (suword(ujobp, (long)ujob));
1477}
1478
1479static struct aiocb_ops aiocb_ops = {
1480 .aio_copyin = aiocb_copyin,
1481 .fetch_status = aiocb_fetch_status,
1482 .fetch_error = aiocb_fetch_error,
1483 .store_status = aiocb_store_status,
1484 .store_error = aiocb_store_error,
1485 .store_kernelinfo = aiocb_store_kernelinfo,
1486 .store_aiocb = aiocb_store_aiocb,
1487};
1488
1489#ifdef COMPAT_FREEBSD6
1490static struct aiocb_ops aiocb_ops_osigevent = {
1491 .aio_copyin = aiocb_copyin_old_sigevent,
1492 .fetch_status = aiocb_fetch_status,
1493 .fetch_error = aiocb_fetch_error,
1494 .store_status = aiocb_store_status,
1495 .store_error = aiocb_store_error,
1496 .store_kernelinfo = aiocb_store_kernelinfo,
1497 .store_aiocb = aiocb_store_aiocb,
1498};
1499#endif
1500
1501/*
1502 * Queue a new AIO request. Choosing either the threaded or direct bio VCHR
1503 * technique is done in this code.
1504 */
1505int
1506aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj,
1507 int type, struct aiocb_ops *ops)
1508{
1509 struct proc *p = td->td_proc;
1510 struct file *fp = NULL;
1511 struct kaiocb *job;
1512 struct kaioinfo *ki;
1513 struct kevent kev;
1514 int opcode;
1515 int error;
1516 int fd, kqfd;
1517 int jid;
1518 u_short evflags;
1519
1520 if (p->p_aioinfo == NULL)
1521 aio_init_aioinfo(p);
1522
1523 ki = p->p_aioinfo;
1524
1525 ops->store_status(ujob, -1);
1526 ops->store_error(ujob, 0);
1527 ops->store_kernelinfo(ujob, -1);
1528
1529 if (num_queue_count >= max_queue_count ||
1530 ki->kaio_count >= max_aio_queue_per_proc) {
1531 error = EAGAIN;
1532 goto err1;
1533 }
1534
1535 job = uma_zalloc(aiocb_zone, M_WAITOK | M_ZERO);
1536 knlist_init_mtx(&job->klist, AIO_MTX(ki));
1537
1538 error = ops->aio_copyin(ujob, job, type);
1539 if (error)
1540 goto err2;
1541
1542 if (job->uaiocb.aio_nbytes > IOSIZE_MAX) {
1543 error = EINVAL;
1544 goto err2;
1545 }
1546
1547 if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT &&
1548 job->uaiocb.aio_sigevent.sigev_notify != SIGEV_SIGNAL &&
1549 job->uaiocb.aio_sigevent.sigev_notify != SIGEV_THREAD_ID &&
1550 job->uaiocb.aio_sigevent.sigev_notify != SIGEV_NONE) {
1551 error = EINVAL;
1552 goto err2;
1553 }
1554
1555 if ((job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL ||
1556 job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID) &&
1557 !_SIG_VALID(job->uaiocb.aio_sigevent.sigev_signo)) {
1558 error = EINVAL;
1559 goto err2;
1560 }
1561
1562 /* Get the opcode. */
1563 if (type == LIO_NOP) {
1564 switch (job->uaiocb.aio_lio_opcode) {
1565 case LIO_WRITE:
1566 case LIO_WRITEV:
1567 case LIO_NOP:
1568 case LIO_READ:
1569 case LIO_READV:
1570 opcode = job->uaiocb.aio_lio_opcode;
1571 break;
1572 default:
1573 error = EINVAL;
1574 goto err2;
1575 }
1576 } else
1577 opcode = job->uaiocb.aio_lio_opcode = type;
1578
1579 ksiginfo_init(&job->ksi);
1580
1581 /* Save userspace address of the job info. */
1582 job->ujob = ujob;
1583
1584 /*
1585 * Validate the opcode and fetch the file object for the specified
1586 * file descriptor.
1587 *
1588 * XXXRW: Moved the opcode validation up here so that we don't
1589 * retrieve a file descriptor without knowing what the capabiltity
1590 * should be.
1591 */
1592 fd = job->uaiocb.aio_fildes;
1593 switch (opcode) {
1594 case LIO_WRITE:
1595 case LIO_WRITEV:
1596 error = fget_write(td, fd, &cap_pwrite_rights, &fp);
1597 break;
1598 case LIO_READ:
1599 case LIO_READV:
1600 error = fget_read(td, fd, &cap_pread_rights, &fp);
1601 break;
1602 case LIO_SYNC:
1603 case LIO_DSYNC:
1604 error = fget(td, fd, &cap_fsync_rights, &fp);
1605 break;
1606 case LIO_MLOCK:
1607 break;
1608 case LIO_NOP:
1609 error = fget(td, fd, &cap_no_rights, &fp);
1610 break;
1611 default:
1612 error = EINVAL;
1613 }
1614 if (error)
1615 goto err3;
1616
1617 if ((opcode & LIO_SYNC) && fp->f_vnode == NULL) {
1618 error = EINVAL;
1619 goto err3;
1620 }
1621
1622 if ((opcode == LIO_READ || opcode == LIO_READV ||
1623 opcode == LIO_WRITE || opcode == LIO_WRITEV) &&
1624 job->uaiocb.aio_offset < 0 &&
1625 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) {
1626 error = EINVAL;
1627 goto err3;
1628 }
1629
1630 if (fp != NULL && fp->f_ops == &path_fileops) {
1631 error = EBADF;
1632 goto err3;
1633 }
1634
1635 job->fd_file = fp;
1636
1637 mtx_lock(&aio_job_mtx);
1638 jid = jobrefid++;
1639 job->seqno = jobseqno++;
1640 mtx_unlock(&aio_job_mtx);
1641 error = ops->store_kernelinfo(ujob, jid);
1642 if (error) {
1643 error = EINVAL;
1644 goto err3;
1645 }
1646 job->uaiocb._aiocb_private.kernelinfo = (void *)(intptr_t)jid;
1647
1648 if (opcode == LIO_NOP) {
1649 fdrop(fp, td);
1650 MPASS(job->uiop == &job->uio || job->uiop == NULL);
1651 uma_zfree(aiocb_zone, job);
1652 return (0);
1653 }
1654
1655 if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT)
1656 goto no_kqueue;
1657 evflags = job->uaiocb.aio_sigevent.sigev_notify_kevent_flags;
1658 if ((evflags & ~(EV_CLEAR | EV_DISPATCH | EV_ONESHOT)) != 0) {
1659 error = EINVAL;
1660 goto err3;
1661 }
1662 kqfd = job->uaiocb.aio_sigevent.sigev_notify_kqueue;
1663 memset(&kev, 0, sizeof(kev));
1664 kev.ident = (uintptr_t)job->ujob;
1665 kev.filter = EVFILT_AIO;
1666 kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1 | evflags;
1667 kev.data = (intptr_t)job;
1668 kev.udata = job->uaiocb.aio_sigevent.sigev_value.sival_ptr;
1669 error = kqfd_register(kqfd, &kev, td, M_WAITOK);
1670 if (error)
1671 goto err3;
1672
1673no_kqueue:
1674
1675 ops->store_error(ujob, EINPROGRESS);
1676 job->uaiocb._aiocb_private.error = EINPROGRESS;
1677 job->userproc = p;
1678 job->cred = crhold(td->td_ucred);
1679 job->jobflags = KAIOCB_QUEUEING;
1680 job->lio = lj;
1681
1682 if (opcode & LIO_VECTORED) {
1683 /* Use the uio copied in by aio_copyin */
1684 MPASS(job->uiop != &job->uio && job->uiop != NULL);
1685 } else {
1686 /* Setup the inline uio */
1687 job->iov[0].iov_base = (void *)(uintptr_t)job->uaiocb.aio_buf;
1688 job->iov[0].iov_len = job->uaiocb.aio_nbytes;
1689 job->uio.uio_iov = job->iov;
1690 job->uio.uio_iovcnt = 1;
1691 job->uio.uio_resid = job->uaiocb.aio_nbytes;
1692 job->uio.uio_segflg = UIO_USERSPACE;
1693 job->uiop = &job->uio;
1694 }
1695 switch (opcode & (LIO_READ | LIO_WRITE)) {
1696 case LIO_READ:
1697 job->uiop->uio_rw = UIO_READ;
1698 break;
1699 case LIO_WRITE:
1700 job->uiop->uio_rw = UIO_WRITE;
1701 break;
1702 }
1703 job->uiop->uio_offset = job->uaiocb.aio_offset;
1704 job->uiop->uio_td = td;
1705
1706 if (opcode == LIO_MLOCK) {
1707 aio_schedule(job, aio_process_mlock);
1708 error = 0;
1709 } else if (fp->f_ops->fo_aio_queue == NULL)
1710 error = aio_queue_file(fp, job);
1711 else
1712 error = fo_aio_queue(fp, job);
1713 if (error)
1714 goto err4;
1715
1716 AIO_LOCK(ki);
1717 job->jobflags &= ~KAIOCB_QUEUEING;
1718 TAILQ_INSERT_TAIL(&ki->kaio_all, job, allist);
1719 ki->kaio_count++;
1720 if (lj)
1721 lj->lioj_count++;
1722 atomic_add_int(&num_queue_count, 1);
1723 if (job->jobflags & KAIOCB_FINISHED) {
1724 /*
1725 * The queue callback completed the request synchronously.
1726 * The bulk of the completion is deferred in that case
1727 * until this point.
1728 */
1729 aio_bio_done_notify(p, job);
1730 } else
1731 TAILQ_INSERT_TAIL(&ki->kaio_jobqueue, job, plist);
1732 AIO_UNLOCK(ki);
1733 return (0);
1734
1735err4:
1736 crfree(job->cred);
1737err3:
1738 if (fp)
1739 fdrop(fp, td);
1740 knlist_delete(&job->klist, curthread, 0);
1741err2:
1742 if (job->uiop != &job->uio)
1743 free(job->uiop, M_IOV);
1744 uma_zfree(aiocb_zone, job);
1745err1:
1746 ops->store_error(ujob, error);
1747 return (error);
1748}
1749
1750static void
1751aio_cancel_daemon_job(struct kaiocb *job)
1752{
1753
1754 mtx_lock(&aio_job_mtx);
1755 if (!aio_cancel_cleared(job))
1756 TAILQ_REMOVE(&aio_jobs, job, list);
1757 mtx_unlock(&aio_job_mtx);
1758 aio_cancel(job);
1759}
1760
1761void
1762aio_schedule(struct kaiocb *job, aio_handle_fn_t *func)
1763{
1764
1765 mtx_lock(&aio_job_mtx);
1766 if (!aio_set_cancel_function(job, aio_cancel_daemon_job)) {
1767 mtx_unlock(&aio_job_mtx);
1768 aio_cancel(job);
1769 return;
1770 }
1771 job->handle_fn = func;
1772 TAILQ_INSERT_TAIL(&aio_jobs, job, list);
1773 aio_kick_nowait(job->userproc);
1774 mtx_unlock(&aio_job_mtx);
1775}
1776
1777static void
1778aio_cancel_sync(struct kaiocb *job)
1779{
1780 struct kaioinfo *ki;
1781
1782 ki = job->userproc->p_aioinfo;
1783 AIO_LOCK(ki);
1784 if (!aio_cancel_cleared(job))
1785 TAILQ_REMOVE(&ki->kaio_syncqueue, job, list);
1786 AIO_UNLOCK(ki);
1787 aio_cancel(job);
1788}
1789
1790int
1791aio_queue_file(struct file *fp, struct kaiocb *job)
1792{
1793 struct kaioinfo *ki;
1794 struct kaiocb *job2;
1795 struct vnode *vp;
1796 struct mount *mp;
1797 int error;
1798 bool safe;
1799
1800 ki = job->userproc->p_aioinfo;
1801 error = aio_qbio(job->userproc, job);
1802 if (error >= 0)
1803 return (error);
1804 safe = false;
1805 if (fp->f_type == DTYPE_VNODE) {
1806 vp = fp->f_vnode;
1807 if (vp->v_type == VREG || vp->v_type == VDIR) {
1808 mp = fp->f_vnode->v_mount;
1809 if (mp == NULL || (mp->mnt_flag & MNT_LOCAL) != 0)
1810 safe = true;
1811 }
1812 }
1813 if (!(safe || enable_aio_unsafe)) {
1814 counted_warning(&unsafe_warningcnt,
1815 "is attempting to use unsafe AIO requests");
1816 return (EOPNOTSUPP);
1817 }
1818
1819 if (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
1820 aio_schedule(job, aio_process_rw);
1821 error = 0;
1822 } else if (job->uaiocb.aio_lio_opcode & LIO_SYNC) {
1823 AIO_LOCK(ki);
1824 TAILQ_FOREACH(job2, &ki->kaio_jobqueue, plist) {
1825 if (job2->fd_file == job->fd_file &&
1826 ((job2->uaiocb.aio_lio_opcode & LIO_SYNC) == 0) &&
1827 job2->seqno < job->seqno) {
1828 job2->jobflags |= KAIOCB_CHECKSYNC;
1829 job->pending++;
1830 }
1831 }
1832 if (job->pending != 0) {
1833 if (!aio_set_cancel_function_locked(job,
1834 aio_cancel_sync)) {
1835 AIO_UNLOCK(ki);
1836 aio_cancel(job);
1837 return (0);
1838 }
1839 TAILQ_INSERT_TAIL(&ki->kaio_syncqueue, job, list);
1840 AIO_UNLOCK(ki);
1841 return (0);
1842 }
1843 AIO_UNLOCK(ki);
1844 aio_schedule(job, aio_process_sync);
1845 error = 0;
1846 } else {
1847 error = EINVAL;
1848 }
1849 return (error);
1850}
1851
1852static void
1853aio_kick_nowait(struct proc *userp)
1854{
1855 struct kaioinfo *ki = userp->p_aioinfo;
1856 struct aioproc *aiop;
1857
1858 mtx_assert(&aio_job_mtx, MA_OWNED);
1859 if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) {
1860 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1861 aiop->aioprocflags &= ~AIOP_FREE;
1862 wakeup(aiop->aioproc);
1863 } else if (num_aio_resv_start + num_aio_procs < max_aio_procs &&
1864 ki->kaio_active_count + num_aio_resv_start < max_aio_per_proc) {
1865 taskqueue_enqueue(taskqueue_aiod_kick, &ki->kaio_task);
1866 }
1867}
1868
1869static int
1870aio_kick(struct proc *userp)
1871{
1872 struct kaioinfo *ki = userp->p_aioinfo;
1873 struct aioproc *aiop;
1874 int error, ret = 0;
1875
1876 mtx_assert(&aio_job_mtx, MA_OWNED);
1877retryproc:
1878 if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) {
1879 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1880 aiop->aioprocflags &= ~AIOP_FREE;
1881 wakeup(aiop->aioproc);
1882 } else if (num_aio_resv_start + num_aio_procs < max_aio_procs &&
1883 ki->kaio_active_count + num_aio_resv_start < max_aio_per_proc) {
1884 num_aio_resv_start++;
1885 mtx_unlock(&aio_job_mtx);
1886 error = aio_newproc(&num_aio_resv_start);
1887 mtx_lock(&aio_job_mtx);
1888 if (error) {
1889 num_aio_resv_start--;
1890 goto retryproc;
1891 }
1892 } else {
1893 ret = -1;
1894 }
1895 return (ret);
1896}
1897
1898static void
1899aio_kick_helper(void *context, int pending)
1900{
1901 struct proc *userp = context;
1902
1903 mtx_lock(&aio_job_mtx);
1904 while (--pending >= 0) {
1905 if (aio_kick(userp))
1906 break;
1907 }
1908 mtx_unlock(&aio_job_mtx);
1909}
1910
1911/*
1912 * Support the aio_return system call, as a side-effect, kernel resources are
1913 * released.
1914 */
1915static int
1916kern_aio_return(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops)
1917{
1918 struct proc *p = td->td_proc;
1919 struct kaiocb *job;
1920 struct kaioinfo *ki;
1921 long status, error;
1922
1923 ki = p->p_aioinfo;
1924 if (ki == NULL)
1925 return (EINVAL);
1926 AIO_LOCK(ki);
1927 TAILQ_FOREACH(job, &ki->kaio_done, plist) {
1928 if (job->ujob == ujob)
1929 break;
1930 }
1931 if (job != NULL) {
1932 MPASS(job->jobflags & KAIOCB_FINISHED);
1933 status = job->uaiocb._aiocb_private.status;
1934 error = job->uaiocb._aiocb_private.error;
1935 td->td_retval[0] = status;
1936 td->td_ru.ru_oublock += job->outblock;
1937 td->td_ru.ru_inblock += job->inblock;
1938 td->td_ru.ru_msgsnd += job->msgsnd;
1939 td->td_ru.ru_msgrcv += job->msgrcv;
1940 aio_free_entry(job);
1941 AIO_UNLOCK(ki);
1942 ops->store_error(ujob, error);
1943 ops->store_status(ujob, status);
1944 } else {
1945 error = EINVAL;
1946 AIO_UNLOCK(ki);
1947 }
1948 return (error);
1949}
1950
1951int
1952sys_aio_return(struct thread *td, struct aio_return_args *uap)
1953{
1954
1955 return (kern_aio_return(td, uap->aiocbp, &aiocb_ops));
1956}
1957
1958/*
1959 * Allow a process to wakeup when any of the I/O requests are completed.
1960 */
1961static int
1962kern_aio_suspend(struct thread *td, int njoblist, struct aiocb **ujoblist,
1963 struct timespec *ts)
1964{
1965 struct proc *p = td->td_proc;
1966 struct timeval atv;
1967 struct kaioinfo *ki;
1968 struct kaiocb *firstjob, *job;
1969 int error, i, timo;
1970
1971 timo = 0;
1972 if (ts) {
1973 if (ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1974 return (EINVAL);
1975
1976 TIMESPEC_TO_TIMEVAL(&atv, ts);
1977 if (itimerfix(&atv))
1978 return (EINVAL);
1979 timo = tvtohz(&atv);
1980 }
1981
1982 ki = p->p_aioinfo;
1983 if (ki == NULL)
1984 return (EAGAIN);
1985
1986 if (njoblist == 0)
1987 return (0);
1988
1989 AIO_LOCK(ki);
1990 for (;;) {
1991 firstjob = NULL;
1992 error = 0;
1993 TAILQ_FOREACH(job, &ki->kaio_all, allist) {
1994 for (i = 0; i < njoblist; i++) {
1995 if (job->ujob == ujoblist[i]) {
1996 if (firstjob == NULL)
1997 firstjob = job;
1998 if (job->jobflags & KAIOCB_FINISHED)
1999 goto RETURN;
2000 }
2001 }
2002 }
2003 /* All tasks were finished. */
2004 if (firstjob == NULL)
2005 break;
2006
2007 ki->kaio_flags |= KAIO_WAKEUP;
2008 error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH,
2009 "aiospn", timo);
2010 if (error == ERESTART)
2011 error = EINTR;
2012 if (error)
2013 break;
2014 }
2015RETURN:
2016 AIO_UNLOCK(ki);
2017 return (error);
2018}
2019
2020int
2021sys_aio_suspend(struct thread *td, struct aio_suspend_args *uap)
2022{
2023 struct timespec ts, *tsp;
2024 struct aiocb **ujoblist;
2025 int error;
2026
2027 if (uap->nent < 0 || uap->nent > max_aio_queue_per_proc)
2028 return (EINVAL);
2029
2030 if (uap->timeout) {
2031 /* Get timespec struct. */
2032 if ((error = copyin(uap->timeout, &ts, sizeof(ts))) != 0)
2033 return (error);
2034 tsp = &ts;
2035 } else
2036 tsp = NULL;
2037
2038 ujoblist = malloc(uap->nent * sizeof(ujoblist[0]), M_AIOS, M_WAITOK);
2039 error = copyin(uap->aiocbp, ujoblist, uap->nent * sizeof(ujoblist[0]));
2040 if (error == 0)
2041 error = kern_aio_suspend(td, uap->nent, ujoblist, tsp);
2042 free(ujoblist, M_AIOS);
2043 return (error);
2044}
2045
2046/*
2047 * aio_cancel cancels any non-bio aio operations not currently in progress.
2048 */
2049int
2050sys_aio_cancel(struct thread *td, struct aio_cancel_args *uap)
2051{
2052 struct proc *p = td->td_proc;
2053 struct kaioinfo *ki;
2054 struct kaiocb *job, *jobn;
2055 struct file *fp;
2056 int error;
2057 int cancelled = 0;
2058 int notcancelled = 0;
2059 struct vnode *vp;
2060
2061 /* Lookup file object. */
2062 error = fget(td, uap->fd, &cap_no_rights, &fp);
2063 if (error)
2064 return (error);
2065
2066 ki = p->p_aioinfo;
2067 if (ki == NULL)
2068 goto done;
2069
2070 if (fp->f_type == DTYPE_VNODE) {
2071 vp = fp->f_vnode;
2072 if (vn_isdisk(vp)) {
2073 fdrop(fp, td);
2074 td->td_retval[0] = AIO_NOTCANCELED;
2075 return (0);
2076 }
2077 }
2078
2079 AIO_LOCK(ki);
2080 TAILQ_FOREACH_SAFE(job, &ki->kaio_jobqueue, plist, jobn) {
2081 if ((uap->fd == job->uaiocb.aio_fildes) &&
2082 ((uap->aiocbp == NULL) ||
2083 (uap->aiocbp == job->ujob))) {
2084 if (aio_cancel_job(p, ki, job)) {
2085 cancelled++;
2086 } else {
2087 notcancelled++;
2088 }
2089 if (uap->aiocbp != NULL)
2090 break;
2091 }
2092 }
2093 AIO_UNLOCK(ki);
2094
2095done:
2096 fdrop(fp, td);
2097
2098 if (uap->aiocbp != NULL) {
2099 if (cancelled) {
2100 td->td_retval[0] = AIO_CANCELED;
2101 return (0);
2102 }
2103 }
2104
2105 if (notcancelled) {
2106 td->td_retval[0] = AIO_NOTCANCELED;
2107 return (0);
2108 }
2109
2110 if (cancelled) {
2111 td->td_retval[0] = AIO_CANCELED;
2112 return (0);
2113 }
2114
2115 td->td_retval[0] = AIO_ALLDONE;
2116
2117 return (0);
2118}
2119
2120/*
2121 * aio_error is implemented in the kernel level for compatibility purposes
2122 * only. For a user mode async implementation, it would be best to do it in
2123 * a userland subroutine.
2124 */
2125static int
2126kern_aio_error(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops)
2127{
2128 struct proc *p = td->td_proc;
2129 struct kaiocb *job;
2130 struct kaioinfo *ki;
2131 int status;
2132
2133 ki = p->p_aioinfo;
2134 if (ki == NULL) {
2135 td->td_retval[0] = EINVAL;
2136 return (0);
2137 }
2138
2139 AIO_LOCK(ki);
2140 TAILQ_FOREACH(job, &ki->kaio_all, allist) {
2141 if (job->ujob == ujob) {
2142 if (job->jobflags & KAIOCB_FINISHED)
2143 td->td_retval[0] =
2144 job->uaiocb._aiocb_private.error;
2145 else
2146 td->td_retval[0] = EINPROGRESS;
2147 AIO_UNLOCK(ki);
2148 return (0);
2149 }
2150 }
2151 AIO_UNLOCK(ki);
2152
2153 /*
2154 * Hack for failure of aio_aqueue.
2155 */
2156 status = ops->fetch_status(ujob);
2157 if (status == -1) {
2158 td->td_retval[0] = ops->fetch_error(ujob);
2159 return (0);
2160 }
2161
2162 td->td_retval[0] = EINVAL;
2163 return (0);
2164}
2165
2166int
2167sys_aio_error(struct thread *td, struct aio_error_args *uap)
2168{
2169
2170 return (kern_aio_error(td, uap->aiocbp, &aiocb_ops));
2171}
2172
2173/* syscall - asynchronous read from a file (REALTIME) */
2174#ifdef COMPAT_FREEBSD6
2175int
2176freebsd6_aio_read(struct thread *td, struct freebsd6_aio_read_args *uap)
2177{
2178
2179 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ,
2180 &aiocb_ops_osigevent));
2181}
2182#endif
2183
2184int
2185sys_aio_read(struct thread *td, struct aio_read_args *uap)
2186{
2187
2188 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READ, &aiocb_ops));
2189}
2190
2191int
2192sys_aio_readv(struct thread *td, struct aio_readv_args *uap)
2193{
2194
2195 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READV, &aiocb_ops));
2196}
2197
2198/* syscall - asynchronous write to a file (REALTIME) */
2199#ifdef COMPAT_FREEBSD6
2200int
2201freebsd6_aio_write(struct thread *td, struct freebsd6_aio_write_args *uap)
2202{
2203
2204 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE,
2205 &aiocb_ops_osigevent));
2206}
2207#endif
2208
2209int
2210sys_aio_write(struct thread *td, struct aio_write_args *uap)
2211{
2212
2213 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITE, &aiocb_ops));
2214}
2215
2216int
2217sys_aio_writev(struct thread *td, struct aio_writev_args *uap)
2218{
2219
2220 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITEV, &aiocb_ops));
2221}
2222
2223int
2224sys_aio_mlock(struct thread *td, struct aio_mlock_args *uap)
2225{
2226
2227 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_MLOCK, &aiocb_ops));
2228}
2229
2230static int
2231kern_lio_listio(struct thread *td, int mode, struct aiocb * const *uacb_list,
2232 struct aiocb **acb_list, int nent, struct sigevent *sig,
2233 struct aiocb_ops *ops)
2234{
2235 struct proc *p = td->td_proc;
2236 struct aiocb *job;
2237 struct kaioinfo *ki;
2238 struct aioliojob *lj;
2239 struct kevent kev;
2240 int error;
2241 int nagain, nerror;
2242 int i;
2243
2244 if ((mode != LIO_NOWAIT) && (mode != LIO_WAIT))
2245 return (EINVAL);
2246
2247 if (nent < 0 || nent > max_aio_queue_per_proc)
2248 return (EINVAL);
2249
2250 if (p->p_aioinfo == NULL)
2251 aio_init_aioinfo(p);
2252
2253 ki = p->p_aioinfo;
2254
2255 lj = uma_zalloc(aiolio_zone, M_WAITOK);
2256 lj->lioj_flags = 0;
2257 lj->lioj_count = 0;
2258 lj->lioj_finished_count = 0;
2259 lj->lioj_signal.sigev_notify = SIGEV_NONE;
2260 knlist_init_mtx(&lj->klist, AIO_MTX(ki));
2261 ksiginfo_init(&lj->lioj_ksi);
2262
2263 /*
2264 * Setup signal.
2265 */
2266 if (sig && (mode == LIO_NOWAIT)) {
2267 bcopy(sig, &lj->lioj_signal, sizeof(lj->lioj_signal));
2268 if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) {
2269 /* Assume only new style KEVENT */
2270 memset(&kev, 0, sizeof(kev));
2271 kev.filter = EVFILT_LIO;
2272 kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1;
2273 kev.ident = (uintptr_t)uacb_list; /* something unique */
2274 kev.data = (intptr_t)lj;
2275 /* pass user defined sigval data */
2276 kev.udata = lj->lioj_signal.sigev_value.sival_ptr;
2277 error = kqfd_register(
2278 lj->lioj_signal.sigev_notify_kqueue, &kev, td,
2279 M_WAITOK);
2280 if (error) {
2281 uma_zfree(aiolio_zone, lj);
2282 return (error);
2283 }
2284 } else if (lj->lioj_signal.sigev_notify == SIGEV_NONE) {
2285 ;
2286 } else if (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL ||
2287 lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID) {
2288 if (!_SIG_VALID(lj->lioj_signal.sigev_signo)) {
2289 uma_zfree(aiolio_zone, lj);
2290 return EINVAL;
2291 }
2292 lj->lioj_flags |= LIOJ_SIGNAL;
2293 } else {
2294 uma_zfree(aiolio_zone, lj);
2295 return EINVAL;
2296 }
2297 }
2298
2299 AIO_LOCK(ki);
2300 TAILQ_INSERT_TAIL(&ki->kaio_liojoblist, lj, lioj_list);
2301 /*
2302 * Add extra aiocb count to avoid the lio to be freed
2303 * by other threads doing aio_waitcomplete or aio_return,
2304 * and prevent event from being sent until we have queued
2305 * all tasks.
2306 */
2307 lj->lioj_count = 1;
2308 AIO_UNLOCK(ki);
2309
2310 /*
2311 * Get pointers to the list of I/O requests.
2312 */
2313 nagain = 0;
2314 nerror = 0;
2315 for (i = 0; i < nent; i++) {
2316 job = acb_list[i];
2317 if (job != NULL) {
2318 error = aio_aqueue(td, job, lj, LIO_NOP, ops);
2319 if (error == EAGAIN)
2320 nagain++;
2321 else if (error != 0)
2322 nerror++;
2323 }
2324 }
2325
2326 error = 0;
2327 AIO_LOCK(ki);
2328 if (mode == LIO_WAIT) {
2329 while (lj->lioj_count - 1 != lj->lioj_finished_count) {
2330 ki->kaio_flags |= KAIO_WAKEUP;
2331 error = msleep(&p->p_aioinfo, AIO_MTX(ki),
2332 PRIBIO | PCATCH, "aiospn", 0);
2333 if (error == ERESTART)
2334 error = EINTR;
2335 if (error)
2336 break;
2337 }
2338 } else {
2339 if (lj->lioj_count - 1 == lj->lioj_finished_count) {
2340 if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) {
2341 lj->lioj_flags |= LIOJ_KEVENT_POSTED;
2342 KNOTE_LOCKED(&lj->klist, 1);
2343 }
2344 if ((lj->lioj_flags & (LIOJ_SIGNAL |
2345 LIOJ_SIGNAL_POSTED)) == LIOJ_SIGNAL &&
2346 (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL ||
2347 lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID)) {
2348 aio_sendsig(p, &lj->lioj_signal, &lj->lioj_ksi,
2349 lj->lioj_count != 1);
2350 lj->lioj_flags |= LIOJ_SIGNAL_POSTED;
2351 }
2352 }
2353 }
2354 lj->lioj_count--;
2355 if (lj->lioj_count == 0) {
2356 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list);
2357 knlist_delete(&lj->klist, curthread, 1);
2358 PROC_LOCK(p);
2359 sigqueue_take(&lj->lioj_ksi);
2360 PROC_UNLOCK(p);
2361 AIO_UNLOCK(ki);
2362 uma_zfree(aiolio_zone, lj);
2363 } else
2364 AIO_UNLOCK(ki);
2365
2366 if (nerror)
2367 return (EIO);
2368 else if (nagain)
2369 return (EAGAIN);
2370 else
2371 return (error);
2372}
2373
2374/* syscall - list directed I/O (REALTIME) */
2375#ifdef COMPAT_FREEBSD6
2376int
2377freebsd6_lio_listio(struct thread *td, struct freebsd6_lio_listio_args *uap)
2378{
2379 struct aiocb **acb_list;
2380 struct sigevent *sigp, sig;
2381 struct osigevent osig;
2382 int error, nent;
2383
2384 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
2385 return (EINVAL);
2386
2387 nent = uap->nent;
2388 if (nent < 0 || nent > max_aio_queue_per_proc)
2389 return (EINVAL);
2390
2391 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
2392 error = copyin(uap->sig, &osig, sizeof(osig));
2393 if (error)
2394 return (error);
2395 error = convert_old_sigevent(&osig, &sig);
2396 if (error)
2397 return (error);
2398 sigp = &sig;
2399 } else
2400 sigp = NULL;
2401
2402 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
2403 error = copyin(uap->acb_list, acb_list, nent * sizeof(acb_list[0]));
2404 if (error == 0)
2405 error = kern_lio_listio(td, uap->mode,
2406 (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp,
2407 &aiocb_ops_osigevent);
2408 free(acb_list, M_LIO);
2409 return (error);
2410}
2411#endif
2412
2413/* syscall - list directed I/O (REALTIME) */
2414int
2415sys_lio_listio(struct thread *td, struct lio_listio_args *uap)
2416{
2417 struct aiocb **acb_list;
2418 struct sigevent *sigp, sig;
2419 int error, nent;
2420
2421 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
2422 return (EINVAL);
2423
2424 nent = uap->nent;
2425 if (nent < 0 || nent > max_aio_queue_per_proc)
2426 return (EINVAL);
2427
2428 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
2429 error = copyin(uap->sig, &sig, sizeof(sig));
2430 if (error)
2431 return (error);
2432 sigp = &sig;
2433 } else
2434 sigp = NULL;
2435
2436 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
2437 error = copyin(uap->acb_list, acb_list, nent * sizeof(acb_list[0]));
2438 if (error == 0)
2439 error = kern_lio_listio(td, uap->mode, uap->acb_list, acb_list,
2440 nent, sigp, &aiocb_ops);
2441 free(acb_list, M_LIO);
2442 return (error);
2443}
2444
2445static void
2446aio_biocleanup(struct bio *bp)
2447{
2448 struct kaiocb *job = (struct kaiocb *)bp->bio_caller1;
2449 struct kaioinfo *ki;
2450 struct buf *pbuf = (struct buf *)bp->bio_caller2;
2451
2452 /* Release mapping into kernel space. */
2453 if (pbuf != NULL) {
2454 MPASS(pbuf->b_npages <= atop(maxphys) + 1);
2455 pmap_qremove((vm_offset_t)pbuf->b_data, pbuf->b_npages);
2456 vm_page_unhold_pages(pbuf->b_pages, pbuf->b_npages);
2457 uma_zfree(pbuf_zone, pbuf);
2458 atomic_subtract_int(&num_buf_aio, 1);
2459 ki = job->userproc->p_aioinfo;
2460 AIO_LOCK(ki);
2461 ki->kaio_buffer_count--;
2462 AIO_UNLOCK(ki);
2463 } else {
2464 MPASS(bp->bio_ma_n <= atop(maxphys) + 1);
2465 vm_page_unhold_pages(bp->bio_ma, bp->bio_ma_n);
2466 free(bp->bio_ma, M_TEMP);
2467 atomic_subtract_int(&num_unmapped_aio, 1);
2468 }
2469 g_destroy_bio(bp);
2470}
2471
2472static void
2473aio_biowakeup(struct bio *bp)
2474{
2475 struct kaiocb *job = (struct kaiocb *)bp->bio_caller1;
2476 size_t nbytes;
2477 long bcount = bp->bio_bcount;
2478 long resid = bp->bio_resid;
2479 int opcode, nblks;
2480 int bio_error = bp->bio_error;
2481 uint16_t flags = bp->bio_flags;
2482
2483 opcode = job->uaiocb.aio_lio_opcode;
2484
2485 aio_biocleanup(bp);
2486
2487 nbytes =bcount - resid;
2488 atomic_add_acq_long(&job->nbytes, nbytes);
2489 nblks = btodb(nbytes);
2490 /*
2491 * If multiple bios experienced an error, the job will reflect the
2492 * error of whichever failed bio completed last.
2493 */
2494 if (flags & BIO_ERROR)
2495 atomic_set_int(&job->error, bio_error);
2496 if (opcode & LIO_WRITE)
2497 atomic_add_int(&job->outblock, nblks);
2498 else
2499 atomic_add_int(&job->inblock, nblks);
2500 atomic_subtract_int(&job->nbio, 1);
2501
2502
2503 if (atomic_load_int(&job->nbio) == 0) {
2504 if (atomic_load_int(&job->error))
2505 aio_complete(job, -1, job->error);
2506 else
2507 aio_complete(job, atomic_load_long(&job->nbytes), 0);
2508 }
2509}
2510
2511/* syscall - wait for the next completion of an aio request */
2512static int
2513kern_aio_waitcomplete(struct thread *td, struct aiocb **ujobp,
2514 struct timespec *ts, struct aiocb_ops *ops)
2515{
2516 struct proc *p = td->td_proc;
2517 struct timeval atv;
2518 struct kaioinfo *ki;
2519 struct kaiocb *job;
2520 struct aiocb *ujob;
2521 long error, status;
2522 int timo;
2523
2524 ops->store_aiocb(ujobp, NULL);
2525
2526 if (ts == NULL) {
2527 timo = 0;
2528 } else if (ts->tv_sec == 0 && ts->tv_nsec == 0) {
2529 timo = -1;
2530 } else {
2531 if ((ts->tv_nsec < 0) || (ts->tv_nsec >= 1000000000))
2532 return (EINVAL);
2533
2534 TIMESPEC_TO_TIMEVAL(&atv, ts);
2535 if (itimerfix(&atv))
2536 return (EINVAL);
2537 timo = tvtohz(&atv);
2538 }
2539
2540 if (p->p_aioinfo == NULL)
2541 aio_init_aioinfo(p);
2542 ki = p->p_aioinfo;
2543
2544 error = 0;
2545 job = NULL;
2546 AIO_LOCK(ki);
2547 while ((job = TAILQ_FIRST(&ki->kaio_done)) == NULL) {
2548 if (timo == -1) {
2549 error = EWOULDBLOCK;
2550 break;
2551 }
2552 ki->kaio_flags |= KAIO_WAKEUP;
2553 error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH,
2554 "aiowc", timo);
2555 if (timo && error == ERESTART)
2556 error = EINTR;
2557 if (error)
2558 break;
2559 }
2560
2561 if (job != NULL) {
2562 MPASS(job->jobflags & KAIOCB_FINISHED);
2563 ujob = job->ujob;
2564 status = job->uaiocb._aiocb_private.status;
2565 error = job->uaiocb._aiocb_private.error;
2566 td->td_retval[0] = status;
2567 td->td_ru.ru_oublock += job->outblock;
2568 td->td_ru.ru_inblock += job->inblock;
2569 td->td_ru.ru_msgsnd += job->msgsnd;
2570 td->td_ru.ru_msgrcv += job->msgrcv;
2571 aio_free_entry(job);
2572 AIO_UNLOCK(ki);
2573 ops->store_aiocb(ujobp, ujob);
2574 ops->store_error(ujob, error);
2575 ops->store_status(ujob, status);
2576 } else
2577 AIO_UNLOCK(ki);
2578
2579 return (error);
2580}
2581
2582int
2583sys_aio_waitcomplete(struct thread *td, struct aio_waitcomplete_args *uap)
2584{
2585 struct timespec ts, *tsp;
2586 int error;
2587
2588 if (uap->timeout) {
2589 /* Get timespec struct. */
2590 error = copyin(uap->timeout, &ts, sizeof(ts));
2591 if (error)
2592 return (error);
2593 tsp = &ts;
2594 } else
2595 tsp = NULL;
2596
2597 return (kern_aio_waitcomplete(td, uap->aiocbp, tsp, &aiocb_ops));
2598}
2599
2600static int
2601kern_aio_fsync(struct thread *td, int op, struct aiocb *ujob,
2602 struct aiocb_ops *ops)
2603{
2604 int listop;
2605
2606 switch (op) {
2607 case O_SYNC:
2608 listop = LIO_SYNC;
2609 break;
2610 case O_DSYNC:
2611 listop = LIO_DSYNC;
2612 break;
2613 default:
2614 return (EINVAL);
2615 }
2616
2617 return (aio_aqueue(td, ujob, NULL, listop, ops));
2618}
2619
2620int
2621sys_aio_fsync(struct thread *td, struct aio_fsync_args *uap)
2622{
2623
2624 return (kern_aio_fsync(td, uap->op, uap->aiocbp, &aiocb_ops));
2625}
2626
2627/* kqueue attach function */
2628static int
2629filt_aioattach(struct knote *kn)
2630{
2631 struct kaiocb *job;
2632
2633 job = (struct kaiocb *)(uintptr_t)kn->kn_sdata;
2634
2635 /*
2636 * The job pointer must be validated before using it, so
2637 * registration is restricted to the kernel; the user cannot
2638 * set EV_FLAG1.
2639 */
2640 if ((kn->kn_flags & EV_FLAG1) == 0)
2641 return (EPERM);
2642 kn->kn_ptr.p_aio = job;
2643 kn->kn_flags &= ~EV_FLAG1;
2644
2645 knlist_add(&job->klist, kn, 0);
2646
2647 return (0);
2648}
2649
2650/* kqueue detach function */
2651static void
2652filt_aiodetach(struct knote *kn)
2653{
2654 struct knlist *knl;
2655
2656 knl = &kn->kn_ptr.p_aio->klist;
2657 knl->kl_lock(knl->kl_lockarg);
2658 if (!knlist_empty(knl))
2659 knlist_remove(knl, kn, 1);
2660 knl->kl_unlock(knl->kl_lockarg);
2661}
2662
2663/* kqueue filter function */
2664/*ARGSUSED*/
2665static int
2666filt_aio(struct knote *kn, long hint)
2667{
2668 struct kaiocb *job = kn->kn_ptr.p_aio;
2669
2670 kn->kn_data = job->uaiocb._aiocb_private.error;
2671 if (!(job->jobflags & KAIOCB_FINISHED))
2672 return (0);
2673 kn->kn_flags |= EV_EOF;
2674 return (1);
2675}
2676
2677/* kqueue attach function */
2678static int
2679filt_lioattach(struct knote *kn)
2680{
2681 struct aioliojob *lj;
2682
2683 lj = (struct aioliojob *)(uintptr_t)kn->kn_sdata;
2684
2685 /*
2686 * The aioliojob pointer must be validated before using it, so
2687 * registration is restricted to the kernel; the user cannot
2688 * set EV_FLAG1.
2689 */
2690 if ((kn->kn_flags & EV_FLAG1) == 0)
2691 return (EPERM);
2692 kn->kn_ptr.p_lio = lj;
2693 kn->kn_flags &= ~EV_FLAG1;
2694
2695 knlist_add(&lj->klist, kn, 0);
2696
2697 return (0);
2698}
2699
2700/* kqueue detach function */
2701static void
2702filt_liodetach(struct knote *kn)
2703{
2704 struct knlist *knl;
2705
2706 knl = &kn->kn_ptr.p_lio->klist;
2707 knl->kl_lock(knl->kl_lockarg);
2708 if (!knlist_empty(knl))
2709 knlist_remove(knl, kn, 1);
2710 knl->kl_unlock(knl->kl_lockarg);
2711}
2712
2713/* kqueue filter function */
2714/*ARGSUSED*/
2715static int
2716filt_lio(struct knote *kn, long hint)
2717{
2718 struct aioliojob * lj = kn->kn_ptr.p_lio;
2719
2720 return (lj->lioj_flags & LIOJ_KEVENT_POSTED);
2721}
2722
2723#ifdef COMPAT_FREEBSD32
2724#include <sys/mount.h>
2725#include <sys/socket.h>
2726#include <compat/freebsd32/freebsd32.h>
2727#include <compat/freebsd32/freebsd32_proto.h>
2728#include <compat/freebsd32/freebsd32_signal.h>
2729#include <compat/freebsd32/freebsd32_syscall.h>
2730#include <compat/freebsd32/freebsd32_util.h>
2731
2732struct __aiocb_private32 {
2733 int32_t status;
2734 int32_t error;
2735 uint32_t kernelinfo;
2736};
2737
2738#ifdef COMPAT_FREEBSD6
2739typedef struct oaiocb32 {
2740 int aio_fildes; /* File descriptor */
2741 uint64_t aio_offset __packed; /* File offset for I/O */
2742 uint32_t aio_buf; /* I/O buffer in process space */
2743 uint32_t aio_nbytes; /* Number of bytes for I/O */
2744 struct osigevent32 aio_sigevent; /* Signal to deliver */
2745 int aio_lio_opcode; /* LIO opcode */
2746 int aio_reqprio; /* Request priority -- ignored */
2747 struct __aiocb_private32 _aiocb_private;
2748} oaiocb32_t;
2749#endif
2750
2751typedef struct aiocb32 {
2752 int32_t aio_fildes; /* File descriptor */
2753 uint64_t aio_offset __packed; /* File offset for I/O */
2754 uint32_t aio_buf; /* I/O buffer in process space */
2755 uint32_t aio_nbytes; /* Number of bytes for I/O */
2756 int __spare__[2];
2757 uint32_t __spare2__;
2758 int aio_lio_opcode; /* LIO opcode */
2759 int aio_reqprio; /* Request priority -- ignored */
2760 struct __aiocb_private32 _aiocb_private;
2761 struct sigevent32 aio_sigevent; /* Signal to deliver */
2762} aiocb32_t;
2763
2764#ifdef COMPAT_FREEBSD6
2765static int
2766convert_old_sigevent32(struct osigevent32 *osig, struct sigevent *nsig)
2767{
2768
2769 /*
2770 * Only SIGEV_NONE, SIGEV_SIGNAL, and SIGEV_KEVENT are
2771 * supported by AIO with the old sigevent structure.
2772 */
2773 CP(*osig, *nsig, sigev_notify);
2774 switch (nsig->sigev_notify) {
2775 case SIGEV_NONE:
2776 break;
2777 case SIGEV_SIGNAL:
2778 nsig->sigev_signo = osig->__sigev_u.__sigev_signo;
2779 break;
2780 case SIGEV_KEVENT:
2781 nsig->sigev_notify_kqueue =
2782 osig->__sigev_u.__sigev_notify_kqueue;
2783 PTRIN_CP(*osig, *nsig, sigev_value.sival_ptr);
2784 break;
2785 default:
2786 return (EINVAL);
2787 }
2788 return (0);
2789}
2790
2791static int
2792aiocb32_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob,
2793 int type __unused)
2794{
2795 struct oaiocb32 job32;
2796 struct aiocb *kcb = &kjob->uaiocb;
2797 int error;
2798
2799 bzero(kcb, sizeof(struct aiocb));
2800 error = copyin(ujob, &job32, sizeof(job32));
2801 if (error)
2802 return (error);
2803
2804 /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */
2805
2806 CP(job32, *kcb, aio_fildes);
2807 CP(job32, *kcb, aio_offset);
2808 PTRIN_CP(job32, *kcb, aio_buf);
2809 CP(job32, *kcb, aio_nbytes);
2810 CP(job32, *kcb, aio_lio_opcode);
2811 CP(job32, *kcb, aio_reqprio);
2812 CP(job32, *kcb, _aiocb_private.status);
2813 CP(job32, *kcb, _aiocb_private.error);
2814 PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo);
2815 return (convert_old_sigevent32(&job32.aio_sigevent,
2816 &kcb->aio_sigevent));
2817}
2818#endif
2819
2820static int
2821aiocb32_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type)
2822{
2823 struct aiocb32 job32;
2824 struct aiocb *kcb = &kjob->uaiocb;
2825 struct iovec32 *iov32;
2826 int error;
2827
2828 error = copyin(ujob, &job32, sizeof(job32));
2829 if (error)
2830 return (error);
2831 CP(job32, *kcb, aio_fildes);
2832 CP(job32, *kcb, aio_offset);
2833 CP(job32, *kcb, aio_lio_opcode);
2834 if (type == LIO_NOP)
2835 type = kcb->aio_lio_opcode;
2836 if (type & LIO_VECTORED) {
2837 iov32 = PTRIN(job32.aio_iov);
2838 CP(job32, *kcb, aio_iovcnt);
2839 /* malloc a uio and copy in the iovec */
2840 error = freebsd32_copyinuio(iov32,
2841 kcb->aio_iovcnt, &kjob->uiop);
2842 if (error)
2843 return (error);
2844 } else {
2845 PTRIN_CP(job32, *kcb, aio_buf);
2846 CP(job32, *kcb, aio_nbytes);
2847 }
2848 CP(job32, *kcb, aio_reqprio);
2849 CP(job32, *kcb, _aiocb_private.status);
2850 CP(job32, *kcb, _aiocb_private.error);
2851 PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo);
2852 error = convert_sigevent32(&job32.aio_sigevent, &kcb->aio_sigevent);
2853
2854 return (error);
2855}
2856
2857static long
2858aiocb32_fetch_status(struct aiocb *ujob)
2859{
2860 struct aiocb32 *ujob32;
2861
2862 ujob32 = (struct aiocb32 *)ujob;
2863 return (fuword32(&ujob32->_aiocb_private.status));
2864}
2865
2866static long
2867aiocb32_fetch_error(struct aiocb *ujob)
2868{
2869 struct aiocb32 *ujob32;
2870
2871 ujob32 = (struct aiocb32 *)ujob;
2872 return (fuword32(&ujob32->_aiocb_private.error));
2873}
2874
2875static int
2876aiocb32_store_status(struct aiocb *ujob, long status)
2877{
2878 struct aiocb32 *ujob32;
2879
2880 ujob32 = (struct aiocb32 *)ujob;
2881 return (suword32(&ujob32->_aiocb_private.status, status));
2882}
2883
2884static int
2885aiocb32_store_error(struct aiocb *ujob, long error)
2886{
2887 struct aiocb32 *ujob32;
2888
2889 ujob32 = (struct aiocb32 *)ujob;
2890 return (suword32(&ujob32->_aiocb_private.error, error));
2891}
2892
2893static int
2894aiocb32_store_kernelinfo(struct aiocb *ujob, long jobref)
2895{
2896 struct aiocb32 *ujob32;
2897
2898 ujob32 = (struct aiocb32 *)ujob;
2899 return (suword32(&ujob32->_aiocb_private.kernelinfo, jobref));
2900}
2901
2902static int
2903aiocb32_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob)
2904{
2905
2906 return (suword32(ujobp, (long)ujob));
2907}
2908
2909static struct aiocb_ops aiocb32_ops = {
2910 .aio_copyin = aiocb32_copyin,
2911 .fetch_status = aiocb32_fetch_status,
2912 .fetch_error = aiocb32_fetch_error,
2913 .store_status = aiocb32_store_status,
2914 .store_error = aiocb32_store_error,
2915 .store_kernelinfo = aiocb32_store_kernelinfo,
2916 .store_aiocb = aiocb32_store_aiocb,
2917};
2918
2919#ifdef COMPAT_FREEBSD6
2920static struct aiocb_ops aiocb32_ops_osigevent = {
2921 .aio_copyin = aiocb32_copyin_old_sigevent,
2922 .fetch_status = aiocb32_fetch_status,
2923 .fetch_error = aiocb32_fetch_error,
2924 .store_status = aiocb32_store_status,
2925 .store_error = aiocb32_store_error,
2926 .store_kernelinfo = aiocb32_store_kernelinfo,
2927 .store_aiocb = aiocb32_store_aiocb,
2928};
2929#endif
2930
2931int
2932freebsd32_aio_return(struct thread *td, struct freebsd32_aio_return_args *uap)
2933{
2934
2935 return (kern_aio_return(td, (struct aiocb *)uap->aiocbp, &aiocb32_ops));
2936}
2937
2938int
2939freebsd32_aio_suspend(struct thread *td, struct freebsd32_aio_suspend_args *uap)
2940{
2941 struct timespec32 ts32;
2942 struct timespec ts, *tsp;
2943 struct aiocb **ujoblist;
2944 uint32_t *ujoblist32;
2945 int error, i;
2946
2947 if (uap->nent < 0 || uap->nent > max_aio_queue_per_proc)
2948 return (EINVAL);
2949
2950 if (uap->timeout) {
2951 /* Get timespec struct. */
2952 if ((error = copyin(uap->timeout, &ts32, sizeof(ts32))) != 0)
2953 return (error);
2954 CP(ts32, ts, tv_sec);
2955 CP(ts32, ts, tv_nsec);
2956 tsp = &ts;
2957 } else
2958 tsp = NULL;
2959
2960 ujoblist = malloc(uap->nent * sizeof(ujoblist[0]), M_AIOS, M_WAITOK);
2961 ujoblist32 = (uint32_t *)ujoblist;
2962 error = copyin(uap->aiocbp, ujoblist32, uap->nent *
2963 sizeof(ujoblist32[0]));
2964 if (error == 0) {
2965 for (i = uap->nent - 1; i >= 0; i--)
2966 ujoblist[i] = PTRIN(ujoblist32[i]);
2967
2968 error = kern_aio_suspend(td, uap->nent, ujoblist, tsp);
2969 }
2970 free(ujoblist, M_AIOS);
2971 return (error);
2972}
2973
2974int
2975freebsd32_aio_error(struct thread *td, struct freebsd32_aio_error_args *uap)
2976{
2977
2978 return (kern_aio_error(td, (struct aiocb *)uap->aiocbp, &aiocb32_ops));
2979}
2980
2981#ifdef COMPAT_FREEBSD6
2982int
2983freebsd6_freebsd32_aio_read(struct thread *td,
2984 struct freebsd6_freebsd32_aio_read_args *uap)
2985{
2986
2987 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ,
2988 &aiocb32_ops_osigevent));
2989}
2990#endif
2991
2992int
2993freebsd32_aio_read(struct thread *td, struct freebsd32_aio_read_args *uap)
2994{
2995
2996 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ,
2997 &aiocb32_ops));
2998}
2999
3000int
3001freebsd32_aio_readv(struct thread *td, struct freebsd32_aio_readv_args *uap)
3002{
3003
3004 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READV,
3005 &aiocb32_ops));
3006}
3007
3008#ifdef COMPAT_FREEBSD6
3009int
3010freebsd6_freebsd32_aio_write(struct thread *td,
3011 struct freebsd6_freebsd32_aio_write_args *uap)
3012{
3013
3014 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE,
3015 &aiocb32_ops_osigevent));
3016}
3017#endif
3018
3019int
3020freebsd32_aio_write(struct thread *td, struct freebsd32_aio_write_args *uap)
3021{
3022
3023 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE,
3024 &aiocb32_ops));
3025}
3026
3027int
3028freebsd32_aio_writev(struct thread *td, struct freebsd32_aio_writev_args *uap)
3029{
3030
3031 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITEV,
3032 &aiocb32_ops));
3033}
3034
3035int
3036freebsd32_aio_mlock(struct thread *td, struct freebsd32_aio_mlock_args *uap)
3037{
3038
3039 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_MLOCK,
3040 &aiocb32_ops));
3041}
3042
3043int
3044freebsd32_aio_waitcomplete(struct thread *td,
3045 struct freebsd32_aio_waitcomplete_args *uap)
3046{
3047 struct timespec32 ts32;
3048 struct timespec ts, *tsp;
3049 int error;
3050
3051 if (uap->timeout) {
3052 /* Get timespec struct. */
3053 error = copyin(uap->timeout, &ts32, sizeof(ts32));
3054 if (error)
3055 return (error);
3056 CP(ts32, ts, tv_sec);
3057 CP(ts32, ts, tv_nsec);
3058 tsp = &ts;
3059 } else
3060 tsp = NULL;
3061
3062 return (kern_aio_waitcomplete(td, (struct aiocb **)uap->aiocbp, tsp,
3063 &aiocb32_ops));
3064}
3065
3066int
3067freebsd32_aio_fsync(struct thread *td, struct freebsd32_aio_fsync_args *uap)
3068{
3069
3070 return (kern_aio_fsync(td, uap->op, (struct aiocb *)uap->aiocbp,
3071 &aiocb32_ops));
3072}
3073
3074#ifdef COMPAT_FREEBSD6
3075int
3076freebsd6_freebsd32_lio_listio(struct thread *td,
3077 struct freebsd6_freebsd32_lio_listio_args *uap)
3078{
3079 struct aiocb **acb_list;
3080 struct sigevent *sigp, sig;
3081 struct osigevent32 osig;
3082 uint32_t *acb_list32;
3083 int error, i, nent;
3084
3085 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
3086 return (EINVAL);
3087
3088 nent = uap->nent;
3089 if (nent < 0 || nent > max_aio_queue_per_proc)
3090 return (EINVAL);
3091
3092 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
3093 error = copyin(uap->sig, &osig, sizeof(osig));
3094 if (error)
3095 return (error);
3096 error = convert_old_sigevent32(&osig, &sig);
3097 if (error)
3098 return (error);
3099 sigp = &sig;
3100 } else
3101 sigp = NULL;
3102
3103 acb_list32 = malloc(sizeof(uint32_t) * nent, M_LIO, M_WAITOK);
3104 error = copyin(uap->acb_list, acb_list32, nent * sizeof(uint32_t));
3105 if (error) {
3106 free(acb_list32, M_LIO);
3107 return (error);
3108 }
3109 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
3110 for (i = 0; i < nent; i++)
3111 acb_list[i] = PTRIN(acb_list32[i]);
3112 free(acb_list32, M_LIO);
3113
3114 error = kern_lio_listio(td, uap->mode,
3115 (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp,
3116 &aiocb32_ops_osigevent);
3117 free(acb_list, M_LIO);
3118 return (error);
3119}
3120#endif
3121
3122int
3123freebsd32_lio_listio(struct thread *td, struct freebsd32_lio_listio_args *uap)
3124{
3125 struct aiocb **acb_list;
3126 struct sigevent *sigp, sig;
3127 struct sigevent32 sig32;
3128 uint32_t *acb_list32;
3129 int error, i, nent;
3130
3131 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
3132 return (EINVAL);
3133
3134 nent = uap->nent;
3135 if (nent < 0 || nent > max_aio_queue_per_proc)
3136 return (EINVAL);
3137
3138 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
3139 error = copyin(uap->sig, &sig32, sizeof(sig32));
3140 if (error)
3141 return (error);
3142 error = convert_sigevent32(&sig32, &sig);
3143 if (error)
3144 return (error);
3145 sigp = &sig;
3146 } else
3147 sigp = NULL;
3148
3149 acb_list32 = malloc(sizeof(uint32_t) * nent, M_LIO, M_WAITOK);
3150 error = copyin(uap->acb_list, acb_list32, nent * sizeof(uint32_t));
3151 if (error) {
3152 free(acb_list32, M_LIO);
3153 return (error);
3154 }
3155 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
3156 for (i = 0; i < nent; i++)
3157 acb_list[i] = PTRIN(acb_list32[i]);
3158 free(acb_list32, M_LIO);
3159
3160 error = kern_lio_listio(td, uap->mode,
3161 (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp,
3162 &aiocb32_ops);
3163 free(acb_list, M_LIO);
3164 return (error);
3165}
3166
3167#endif
ts
struct timespec * ts
Definition: clock_if.m:39
type
device_property_type_t type
Definition: bus_if.m:941
tvtohz
int tvtohz(struct timeval *tv)
Definition: kern_clock.c:529
dev_relthread
void dev_relthread(struct cdev *dev, int ref)
Definition: kern_conf.c:249
devvn_refthread
struct cdevsw * devvn_refthread(struct vnode *vp, struct cdev **devp, int *ref)
Definition: kern_conf.c:205
fget_write
int fget_write(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
Definition: kern_descrip.c:3402
fget
int fget(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
Definition: kern_descrip.c:3344
fget_read
int fget_read(struct thread *td, int fd, cap_rights_t *rightsp, struct file **fpp)
Definition: kern_descrip.c:3395
path_fileops
struct fileops path_fileops
Definition: kern_descrip.c:5067
knlist_remove
void knlist_remove(struct knlist *knl, struct knote *kn, int islocked)
Definition: kern_event.c:2467
kqueue_add_filteropts
int kqueue_add_filteropts(int filt, struct filterops *filtops)
Definition: kern_event.c:1391
knlist_add
void knlist_add(struct knlist *knl, struct knote *kn, int islocked)
Definition: kern_event.c:2420
knote
void knote(struct knlist *list, long hint, int lockflags)
Definition: kern_event.c:2363
knlist_empty
int knlist_empty(struct knlist *knl)
Definition: kern_event.c:2474
kqfd_register
int kqfd_register(int fd, struct kevent *kev, struct thread *td, int mflag)
Definition: kern_event.c:2834
knlist_init_mtx
void knlist_init_mtx(struct knlist *knl, struct mtx *lock)
Definition: kern_event.c:2564
kproc_create
int kproc_create(void(*func)(void *), void *arg, struct proc **newpp, int flags, int pages, const char *fmt,...)
Definition: kern_kthread.c:84
kproc_exit
void kproc_exit(int ecode)
Definition: kern_kthread.c:148
malloc
void *() malloc(size_t size, struct malloc_type *mtp, int flags)
Definition: kern_malloc.c:632
free
void free(void *addr, struct malloc_type *mtp)
Definition: kern_malloc.c:907
crhold
struct ucred * crhold(struct ucred *cr)
Definition: kern_prot.c:2014
crfree
void crfree(struct ucred *cr)
Definition: kern_prot.c:2035
sema_init
void sema_init(struct sema *sema, int value, const char *description)
Definition: kern_sema.c:50
panic
void panic(const char *fmt,...)
Definition: kern_shutdown.c:884
sigqueue_take
void sigqueue_take(ksiginfo_t *ksi)
Definition: kern_sig.c:380
kern_psignal
void kern_psignal(struct proc *p, int sig)
Definition: kern_sig.c:2117
tdsendsignal
int tdsendsignal(struct proc *p, struct thread *td, int sig, ksiginfo_t *ksi)
Definition: kern_sig.c:2183
sigev_findtd
int sigev_findtd(struct proc *p, struct sigevent *sigev, struct thread **ttd)
Definition: kern_sig.c:2136
wakeup
void wakeup(const void *ident)
Definition: kern_synch.c:349
itimerfix
int itimerfix(struct timeval *tv)
Definition: kern_time.c:986
start
void *** start
Definition: linker_if.m:98
p31b_setcfg
void p31b_setcfg(int num, int value)
Definition: posix4_mib.c:129
aiocb_ops
Definition: vfs_aio.c:294
aiocb_ops::fetch_status
long(* fetch_status)(struct aiocb *ujob)
Definition: vfs_aio.c:296
aiocb_ops::store_aiocb
int(* store_aiocb)(struct aiocb **ujobp, struct aiocb *ujob)
Definition: vfs_aio.c:301
aiocb_ops::store_status
int(* store_status)(struct aiocb *ujob, long status)
Definition: vfs_aio.c:298
aiocb_ops::fetch_error
long(* fetch_error)(struct aiocb *ujob)
Definition: vfs_aio.c:297
aiocb_ops::aio_copyin
int(* aio_copyin)(struct aiocb *ujob, struct kaiocb *kjob, int ty)
Definition: vfs_aio.c:295
aiocb_ops::store_kernelinfo
int(* store_kernelinfo)(struct aiocb *ujob, long jobref)
Definition: vfs_aio.c:300
aiocb_ops::store_error
int(* store_error)(struct aiocb *ujob, long error)
Definition: vfs_aio.c:299
aioliojob::lioj_finished_count
int lioj_finished_count
Definition: vfs_aio.c:252
aioliojob::lioj_flags
int lioj_flags
Definition: vfs_aio.c:250
aioliojob::lioj_count
int lioj_count
Definition: vfs_aio.c:251
aioliojob::lioj_signal
struct sigevent lioj_signal
Definition: vfs_aio.c:253
aioproc::aioprocflags
int aioprocflags
Definition: vfs_aio.c:241
kaioinfo::kaio_mtx
struct mtx kaio_mtx
Definition: vfs_aio.c:267
kaioinfo::kaio_count
int kaio_count
Definition: vfs_aio.c:270
kaioinfo::kaio_buffer_count
int kaio_buffer_count
Definition: vfs_aio.c:271
kaioinfo::kaio_flags
int kaio_flags
Definition: vfs_aio.c:268
kaioinfo::kaio_active_count
int kaio_active_count
Definition: vfs_aio.c:269
cap_no_rights
__read_mostly cap_rights_t cap_no_rights
Definition: subr_capability.c:100
cap_pread_rights
__read_mostly cap_rights_t cap_pread_rights
Definition: subr_capability.c:86
cap_pwrite_rights
__read_mostly cap_rights_t cap_pwrite_rights
Definition: subr_capability.c:87
cap_fsync_rights
__read_mostly cap_rights_t cap_fsync_rights
Definition: subr_capability.c:70
hz
int hz
Definition: subr_param.c:85
maxphys
u_long maxphys
Definition: subr_param.c:103
counted_warning
void counted_warning(unsigned *counter, const char *msg)
Definition: subr_prf.c:1261
flags
uint16_t flags
Definition: subr_stats.c:2
taskqueue_enqueue
int taskqueue_enqueue(struct taskqueue *queue, struct task *task)
Definition: subr_taskqueue.c:277
taskqueue_drain
void taskqueue_drain(struct taskqueue *queue, struct task *task)
Definition: subr_taskqueue.c:576
fuword32
int32_t fuword32(volatile const void *addr)
Definition: subr_uio.c:465
copyinuio
int copyinuio(const struct iovec *iovp, u_int iovcnt, struct uio **uiop)
Definition: subr_uio.c:365
fuword
long fuword(volatile const void *addr)
Definition: subr_uio.c:487
alloc_unr
int alloc_unr(struct unrhdr *uh)
Definition: subr_unit.c:650
new_unrhdr
struct unrhdr * new_unrhdr(int low, int high, struct mtx *mutex)
Definition: subr_unit.c:360
free_unr
void free_unr(struct unrhdr *uh, u_int item)
Definition: subr_unit.c:900
sema
static struct semid_kernel * sema
Definition: sysv_sem.c:119
mtx
struct mtx mtx
Definition: uipc_ktls.c:0
kern_aio_fsync
static int kern_aio_fsync(struct thread *td, int op, struct aiocb *ujob, struct aiocb_ops *ops)
Definition: vfs_aio.c:2601
filt_lio
static int filt_lio(struct knote *kn, long hint)
Definition: vfs_aio.c:2716
aiocb_fetch_status
static long aiocb_fetch_status(struct aiocb *ujob)
Definition: vfs_aio.c:1438
sys_aio_write
int sys_aio_write(struct thread *td, struct aio_write_args *uap)
Definition: vfs_aio.c:2210
KAIOCB_CANCELLED
#define KAIOCB_CANCELLED
Definition: vfs_aio.c:229
AIO_MTX
#define AIO_MTX(ki)
Definition: vfs_aio.c:285
jobrefid
static u_long jobrefid
Definition: vfs_aio.c:81
jobseqno
static uint64_t jobseqno
Definition: vfs_aio.c:86
kern_aio_suspend
static int kern_aio_suspend(struct thread *td, int njoblist, struct aiocb **ujoblist, struct timespec *ts)
Definition: vfs_aio.c:1962
aio_switch_vmspace
void aio_switch_vmspace(struct kaiocb *job)
Definition: vfs_aio.c:1060
KAIOCB_CANCELLING
#define KAIOCB_CANCELLING
Definition: vfs_aio.c:230
unsafe_warningcnt
static unsigned int unsafe_warningcnt
Definition: vfs_aio.c:117
aio_kick_nowait
static void aio_kick_nowait(struct proc *userp)
Definition: vfs_aio.c:1853
aio_set_cancel_function
bool aio_set_cancel_function(struct kaiocb *job, aio_cancel_fn_t *func)
Definition: vfs_aio.c:1017
aio_process_mlock
static void aio_process_mlock(struct kaiocb *job)
Definition: vfs_aio.c:863
MAX_AIO_QUEUE_PER_PROC
#define MAX_AIO_QUEUE_PER_PROC
Definition: vfs_aio.c:93
aio_free_entry
static int aio_free_entry(struct kaiocb *job)
Definition: vfs_aio.c:494
sys_aio_error
int sys_aio_error(struct thread *td, struct aio_error_args *uap)
Definition: vfs_aio.c:2167
kern_lio_listio
static int kern_lio_listio(struct thread *td, int mode, struct aiocb *const *uacb_list, struct aiocb **acb_list, int nent, struct sigevent *sig, struct aiocb_ops *ops)
Definition: vfs_aio.c:2231
aio_qbio
static int aio_qbio(struct proc *p, struct kaiocb *job)
Definition: vfs_aio.c:1213
KAIO_RUNDOWN
#define KAIO_RUNDOWN
Definition: vfs_aio.c:287
aio_proc_rundown
static void aio_proc_rundown(void *arg, struct proc *p)
Definition: vfs_aio.c:633
kern_aio_return
static int kern_aio_return(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops)
Definition: vfs_aio.c:1916
aio_biocleanup
static void aio_biocleanup(struct bio *bp)
Definition: vfs_aio.c:2446
KAIO_WAKEUP
#define KAIO_WAKEUP
Definition: vfs_aio.c:288
aio_cancel
void aio_cancel(struct kaiocb *job)
Definition: vfs_aio.c:1053
max_aio_per_proc
static int max_aio_per_proc
Definition: vfs_aio.c:164
filt_aioattach
static int filt_aioattach(struct knote *kn)
Definition: vfs_aio.c:2629
aiocb_store_status
static int aiocb_store_status(struct aiocb *ujob, long status)
Definition: vfs_aio.c:1452
aio_onceonly
static int aio_onceonly(void)
Definition: vfs_aio.c:401
KAIOCB_CHECKSYNC
#define KAIOCB_CHECKSYNC
Definition: vfs_aio.c:231
filt_aio
static int filt_aio(struct knote *kn, long hint)
Definition: vfs_aio.c:2666
sys_aio_suspend
int sys_aio_suspend(struct thread *td, struct aio_suspend_args *uap)
Definition: vfs_aio.c:2021
exec_tag
static eventhandler_tag exec_tag
Definition: vfs_aio.c:363
aio_selectjob
static struct kaiocb * aio_selectjob(struct aioproc *aiop)
Definition: vfs_aio.c:694
aio_fsync_vnode
static int aio_fsync_vnode(struct thread *td, struct vnode *vp, int op)
Definition: vfs_aio.c:724
DECLARE_MODULE
DECLARE_MODULE(aio, aio_mod, SI_SUB_VFS, SI_ORDER_ANY)
SYSCTL_INT
SYSCTL_INT(_vfs_aio, OID_AUTO, enable_unsafe, CTLFLAG_RW, &enable_aio_unsafe, 0, "Permit asynchronous IO on all file types, not just known-safe types")
aio_process_rw
static void aio_process_rw(struct kaiocb *job)
Definition: vfs_aio.c:763
aio_schedule
void aio_schedule(struct kaiocb *job, aio_handle_fn_t *func)
Definition: vfs_aio.c:1762
sys_aio_cancel
int sys_aio_cancel(struct thread *td, struct aio_cancel_args *uap)
Definition: vfs_aio.c:2050
num_buf_aio
static int num_buf_aio
Definition: vfs_aio.c:147
AIOP_FREE
#define AIOP_FREE
Definition: vfs_aio.c:238
aio_clear_cancel_function_locked
static bool aio_clear_cancel_function_locked(struct kaiocb *job)
Definition: vfs_aio.c:979
LIOJ_SIGNAL
#define LIOJ_SIGNAL
Definition: vfs_aio.c:259
kern_aio_error
static int kern_aio_error(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops)
Definition: vfs_aio.c:2126
SYSCTL_NODE
static SYSCTL_NODE(_vfs, OID_AUTO, aio, CTLFLAG_RW|CTLFLAG_MPSAFE, 0, "Async IO management")
aio_biowakeup
static void aio_biowakeup(struct bio *bp)
Definition: vfs_aio.c:2473
sys_aio_mlock
int sys_aio_mlock(struct thread *td, struct aio_mlock_args *uap)
Definition: vfs_aio.c:2224
exit_tag
static eventhandler_tag exit_tag
Definition: vfs_aio.c:363
sys_aio_fsync
int sys_aio_fsync(struct thread *td, struct aio_fsync_args *uap)
Definition: vfs_aio.c:2621
aio_clear_cancel_function
bool aio_clear_cancel_function(struct kaiocb *job)
Definition: vfs_aio.c:993
aio_set_cancel_function_locked
static bool aio_set_cancel_function_locked(struct kaiocb *job, aio_cancel_fn_t *func)
Definition: vfs_aio.c:1006
KAIOCB_QUEUEING
#define KAIOCB_QUEUEING
Definition: vfs_aio.c:228
aiocb_store_error
static int aiocb_store_error(struct aiocb *ujob, long error)
Definition: vfs_aio.c:1459
aio_cancel_job
static int aio_cancel_job(struct proc *p, struct kaioinfo *ki, struct kaiocb *job)
Definition: vfs_aio.c:579
aio_modload
static int aio_modload(struct module *module, int cmd, void *arg)
Definition: vfs_aio.c:371
aiocb_store_aiocb
static int aiocb_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob)
Definition: vfs_aio.c:1473
MODULE_VERSION
MODULE_VERSION(aio, 1)
max_queue_count
static int max_queue_count
Definition: vfs_aio.c:139
num_aio_procs
static int num_aio_procs
Definition: vfs_aio.c:126
aiocb_fetch_error
static long aiocb_fetch_error(struct aiocb *ujob)
Definition: vfs_aio.c:1445
FEATURE
FEATURE(aio, "Asynchronous I/O")
__FBSDID
__FBSDID("$FreeBSD$")
max_aio_procs
static int max_aio_procs
Definition: vfs_aio.c:122
aio_bio_done_notify
static void aio_bio_done_notify(struct proc *userp, struct kaiocb *job)
Definition: vfs_aio.c:878
filt_aiodetach
static void filt_aiodetach(struct knote *kn)
Definition: vfs_aio.c:2652
lio_filtops
static struct filterops lio_filtops
Definition: vfs_aio.c:356
sys_aio_readv
int sys_aio_readv(struct thread *td, struct aio_readv_args *uap)
Definition: vfs_aio.c:2192
aiod_lifetime
static int aiod_lifetime
Definition: vfs_aio.c:160
aio_complete
void aio_complete(struct kaiocb *job, long status, int error)
Definition: vfs_aio.c:1030
sys_lio_listio
int sys_lio_listio(struct thread *td, struct lio_listio_args *uap)
Definition: vfs_aio.c:2415
num_aio_resv_start
static int num_aio_resv_start
Definition: vfs_aio.c:158
KAIOCB_FINISHED
#define KAIOCB_FINISHED
Definition: vfs_aio.c:233
kern_aio_waitcomplete
static int kern_aio_waitcomplete(struct thread *td, struct aiocb **ujobp, struct timespec *ts, struct aiocb_ops *ops)
Definition: vfs_aio.c:2513
aio_daemon
static void aio_daemon(void *_id)
Definition: vfs_aio.c:1071
MAX_AIO_PER_PROC
#define MAX_AIO_PER_PROC
Definition: vfs_aio.c:89
MAX_AIO_QUEUE
#define MAX_AIO_QUEUE
Definition: vfs_aio.c:97
KAIOCB_CLEARED
#define KAIOCB_CLEARED
Definition: vfs_aio.c:232
LIOJ_SIGNAL_POSTED
#define LIOJ_SIGNAL_POSTED
Definition: vfs_aio.c:260
target_aio_procs
static int target_aio_procs
Definition: vfs_aio.c:134
TAILQ_HEAD
static TAILQ_HEAD(aioproc)
Definition: vfs_aio.c:304
num_queue_count
static int num_queue_count
Definition: vfs_aio.c:143
LIOJ_KEVENT_POSTED
#define LIOJ_KEVENT_POSTED
Definition: vfs_aio.c:261
sys_aio_read
int sys_aio_read(struct thread *td, struct aio_read_args *uap)
Definition: vfs_aio.c:2185
aio_init_aioinfo
void aio_init_aioinfo(struct proc *p)
Definition: vfs_aio.c:437
num_unmapped_aio
static int num_unmapped_aio
Definition: vfs_aio.c:151
aio_aqueue
int aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj, int type, struct aiocb_ops *ops)
Definition: vfs_aio.c:1506
SYSCTL_DECL
SYSCTL_DECL(_p1003_1b)
aio_schedule_fsync
static void aio_schedule_fsync(void *context, int pending)
Definition: vfs_aio.c:948
aio_cancel_daemon_job
static void aio_cancel_daemon_job(struct kaiocb *job)
Definition: vfs_aio.c:1751
SYSCTL_UINT
SYSCTL_UINT(_vfs_aio, OID_AUTO, unsafe_warningcnt, CTLFLAG_RW, &unsafe_warningcnt, 0, "Warnings that will be triggered upon failed IO requests on unsafe files")
aio_process_sync
static void aio_process_sync(struct kaiocb *job)
Definition: vfs_aio.c:840
sys_aio_waitcomplete
int sys_aio_waitcomplete(struct thread *td, struct aio_waitcomplete_args *uap)
Definition: vfs_aio.c:2583
filt_liodetach
static void filt_liodetach(struct knote *kn)
Definition: vfs_aio.c:2702
AIO_LOCK_ASSERT
#define AIO_LOCK_ASSERT(ki, f)
Definition: vfs_aio.c:284
aiocb_store_kernelinfo
static int aiocb_store_kernelinfo(struct aiocb *ujob, long jobref)
Definition: vfs_aio.c:1466
aio_mod
static moduledata_t aio_mod
Definition: vfs_aio.c:388
MAX_BUF_AIO
#define MAX_BUF_AIO
Definition: vfs_aio.c:101
MALLOC_DEFINE
static MALLOC_DEFINE(M_LIO, "lio", "listio aio control block list")
TASKQUEUE_DEFINE_THREAD
TASKQUEUE_DEFINE_THREAD(aiod_kick)
sys_aio_writev
int sys_aio_writev(struct thread *td, struct aio_writev_args *uap)
Definition: vfs_aio.c:2217
aio_newproc
static int aio_newproc(int *start)
Definition: vfs_aio.c:1178
aio_kick_helper
static void aio_kick_helper(void *context, int pending)
Definition: vfs_aio.c:1899
aio_kick
static int aio_kick(struct proc *userp)
Definition: vfs_aio.c:1870
aio_sendsig
static int aio_sendsig(struct proc *p, struct sigevent *sigev, ksiginfo_t *ksi, bool ext)
Definition: vfs_aio.c:470
AIO_LOCK
#define AIO_LOCK(ki)
Definition: vfs_aio.c:282
enable_aio_unsafe
static int enable_aio_unsafe
Definition: vfs_aio.c:113
max_aio_queue_per_proc
static int max_aio_queue_per_proc
Definition: vfs_aio.c:169
aio_proc_rundown_exec
static void aio_proc_rundown_exec(void *arg, struct proc *p, struct image_params *imgp __unused)
Definition: vfs_aio.c:572
sys_aio_return
int sys_aio_return(struct thread *td, struct aio_return_args *uap)
Definition: vfs_aio.c:1952
aio_cancel_cleared
bool aio_cancel_cleared(struct kaiocb *job)
Definition: vfs_aio.c:966
filt_lioattach
static int filt_lioattach(struct knote *kn)
Definition: vfs_aio.c:2679
aio_cancel_sync
static void aio_cancel_sync(struct kaiocb *job)
Definition: vfs_aio.c:1778
AIO_UNLOCK
#define AIO_UNLOCK(ki)
Definition: vfs_aio.c:283
aio_queue_file
int aio_queue_file(struct file *fp, struct kaiocb *job)
Definition: vfs_aio.c:1791
max_buf_aio
static int max_buf_aio
Definition: vfs_aio.c:174
aiocb_copyin
static int aiocb_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type)
Definition: vfs_aio.c:1418
unmapped_buf
caddr_t __read_mostly unmapped_buf
Definition: vfs_bio.c:161
bwillwrite
void bwillwrite(void)
Definition: vfs_bio.c:2631
nbytes
size_t nbytes
Definition: vfs_extattr.c:718
vn_isdisk
bool vn_isdisk(struct vnode *vp)
Definition: vfs_subr.c:5215
buf
struct stat * buf
Definition: vfs_syscalls.c:2423
fd
int fd
Definition: vfs_syscalls.c:2421
mode
mode_t mode
Definition: vfs_syscalls.c:2910
vn_start_write
int vn_start_write(struct vnode *vp, struct mount **mpp, int flags)
Definition: vfs_vnops.c:1901
vn_finished_write
void vn_finished_write(struct mount *mp)
Definition: vfs_vnops.c:2009