1 /*-
2 * SPDX-License-Identifier: BSD-2-Clause
3 *
4 * Copyright (c) 1997 John S. Dyson. All rights reserved.
5 *
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
8 * are met:
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. John S. Dyson's name may not be used to endorse or promote products
12 * derived from this software without specific prior written permission.
13 *
14 * DISCLAIMER: This code isn't warranted to do anything useful. Anything
15 * bad that happens because of using this software isn't the responsibility
16 * of the author. This software is distributed AS-IS.
17 */
18
19 /*
20 * This file contains support for the POSIX 1003.1B AIO/LIO facility.
21 */
22
23 #include <sys/cdefs.h>
24 #include <sys/param.h>
25 #include <sys/systm.h>
26 #include <sys/malloc.h>
27 #include <sys/bio.h>
28 #include <sys/buf.h>
29 #include <sys/capsicum.h>
30 #include <sys/eventhandler.h>
31 #include <sys/sysproto.h>
32 #include <sys/filedesc.h>
33 #include <sys/kernel.h>
34 #include <sys/module.h>
35 #include <sys/kthread.h>
36 #include <sys/fcntl.h>
37 #include <sys/file.h>
38 #include <sys/limits.h>
39 #include <sys/lock.h>
40 #include <sys/mutex.h>
41 #include <sys/unistd.h>
42 #include <sys/posix4.h>
43 #include <sys/proc.h>
44 #include <sys/resourcevar.h>
45 #include <sys/signalvar.h>
46 #include <sys/syscallsubr.h>
47 #include <sys/protosw.h>
48 #include <sys/rwlock.h>
49 #include <sys/sema.h>
50 #include <sys/socket.h>
51 #include <sys/socketvar.h>
52 #include <sys/syscall.h>
53 #include <sys/sysctl.h>
54 #include <sys/syslog.h>
55 #include <sys/sx.h>
56 #include <sys/taskqueue.h>
57 #include <sys/vnode.h>
58 #include <sys/conf.h>
59 #include <sys/event.h>
60 #include <sys/mount.h>
61 #include <geom/geom.h>
62
63 #include <machine/atomic.h>
64
65 #include <vm/vm.h>
66 #include <vm/vm_page.h>
67 #include <vm/vm_extern.h>
68 #include <vm/pmap.h>
69 #include <vm/vm_map.h>
70 #include <vm/vm_object.h>
71 #include <vm/vnode_pager.h>
72 #include <vm/uma.h>
73 #include <sys/aio.h>
74
75 /*
76 * Counter for allocating reference ids to new jobs. Wrapped to 1 on
77 * overflow. (XXX will be removed soon.)
78 */
79 static u_long jobrefid;
80
81 /*
82 * Counter for aio_fsync.
83 */
84 static uint64_t jobseqno;
85
86 #ifndef MAX_AIO_PER_PROC
87 #define MAX_AIO_PER_PROC 32
88 #endif
89
90 #ifndef MAX_AIO_QUEUE_PER_PROC
91 #define MAX_AIO_QUEUE_PER_PROC 256
92 #endif
93
94 #ifndef MAX_AIO_QUEUE
95 #define MAX_AIO_QUEUE 1024 /* Bigger than MAX_AIO_QUEUE_PER_PROC */
96 #endif
97
98 #ifndef MAX_BUF_AIO
99 #define MAX_BUF_AIO 16
100 #endif
101
102 FEATURE(aio, "Asynchronous I/O");
103 SYSCTL_DECL(_p1003_1b);
104
105 static MALLOC_DEFINE(M_LIO, "lio", "listio aio control block list");
106 static MALLOC_DEFINE(M_AIO, "aio", "structures for asynchronous I/O");
107
108 static SYSCTL_NODE(_vfs, OID_AUTO, aio, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
109 "Async IO management");
110
111 static int enable_aio_unsafe = 0;
112 SYSCTL_INT(_vfs_aio, OID_AUTO, enable_unsafe, CTLFLAG_RW, &enable_aio_unsafe, 0,
113 "Permit asynchronous IO on all file types, not just known-safe types");
114
115 static unsigned int unsafe_warningcnt = 1;
116 SYSCTL_UINT(_vfs_aio, OID_AUTO, unsafe_warningcnt, CTLFLAG_RW,
117 &unsafe_warningcnt, 0,
118 "Warnings that will be triggered upon failed IO requests on unsafe files");
119
120 static int max_aio_procs = MAX_AIO_PROCS;
121 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_procs, CTLFLAG_RW, &max_aio_procs, 0,
122 "Maximum number of kernel processes to use for handling async IO ");
123
124 static int num_aio_procs = 0;
125 SYSCTL_INT(_vfs_aio, OID_AUTO, num_aio_procs, CTLFLAG_RD, &num_aio_procs, 0,
126 "Number of presently active kernel processes for async IO");
127
128 /*
129 * The code will adjust the actual number of AIO processes towards this
130 * number when it gets a chance.
131 */
132 static int target_aio_procs = TARGET_AIO_PROCS;
133 SYSCTL_INT(_vfs_aio, OID_AUTO, target_aio_procs, CTLFLAG_RW, &target_aio_procs,
134 0,
135 "Preferred number of ready kernel processes for async IO");
136
137 static int max_queue_count = MAX_AIO_QUEUE;
138 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue, CTLFLAG_RW, &max_queue_count, 0,
139 "Maximum number of aio requests to queue, globally");
140
141 static int num_queue_count = 0;
142 SYSCTL_INT(_vfs_aio, OID_AUTO, num_queue_count, CTLFLAG_RD, &num_queue_count, 0,
143 "Number of queued aio requests");
144
145 static int num_buf_aio = 0;
146 SYSCTL_INT(_vfs_aio, OID_AUTO, num_buf_aio, CTLFLAG_RD, &num_buf_aio, 0,
147 "Number of aio requests presently handled by the buf subsystem");
148
149 static int num_unmapped_aio = 0;
150 SYSCTL_INT(_vfs_aio, OID_AUTO, num_unmapped_aio, CTLFLAG_RD, &num_unmapped_aio,
151 0,
152 "Number of aio requests presently handled by unmapped I/O buffers");
153
154 /* Number of async I/O processes in the process of being started */
155 /* XXX This should be local to aio_aqueue() */
156 static int num_aio_resv_start = 0;
157
158 static int aiod_lifetime;
159 SYSCTL_INT(_vfs_aio, OID_AUTO, aiod_lifetime, CTLFLAG_RW, &aiod_lifetime, 0,
160 "Maximum lifetime for idle aiod");
161
162 static int max_aio_per_proc = MAX_AIO_PER_PROC;
163 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_per_proc, CTLFLAG_RW, &max_aio_per_proc,
164 0,
165 "Maximum active aio requests per process");
166
167 static int max_aio_queue_per_proc = MAX_AIO_QUEUE_PER_PROC;
168 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue_per_proc, CTLFLAG_RW,
169 &max_aio_queue_per_proc, 0,
170 "Maximum queued aio requests per process");
171
172 static int max_buf_aio = MAX_BUF_AIO;
173 SYSCTL_INT(_vfs_aio, OID_AUTO, max_buf_aio, CTLFLAG_RW, &max_buf_aio, 0,
174 "Maximum buf aio requests per process");
175
176 /*
177 * Though redundant with vfs.aio.max_aio_queue_per_proc, POSIX requires
178 * sysconf(3) to support AIO_LISTIO_MAX, and we implement that with
179 * vfs.aio.aio_listio_max.
180 */
181 SYSCTL_INT(_p1003_1b, CTL_P1003_1B_AIO_LISTIO_MAX, aio_listio_max,
182 CTLFLAG_RD | CTLFLAG_CAPRD, &max_aio_queue_per_proc,
183 0, "Maximum aio requests for a single lio_listio call");
184
185 #ifdef COMPAT_FREEBSD6
186 typedef struct oaiocb {
187 int aio_fildes; /* File descriptor */
188 off_t aio_offset; /* File offset for I/O */
189 volatile void *aio_buf; /* I/O buffer in process space */
190 size_t aio_nbytes; /* Number of bytes for I/O */
191 struct osigevent aio_sigevent; /* Signal to deliver */
192 int aio_lio_opcode; /* LIO opcode */
193 int aio_reqprio; /* Request priority -- ignored */
194 struct __aiocb_private _aiocb_private;
195 } oaiocb_t;
196 #endif
197
198 /*
199 * Below is a key of locks used to protect each member of struct kaiocb
200 * aioliojob and kaioinfo and any backends.
201 *
202 * * - need not protected
203 * a - locked by kaioinfo lock
204 * b - locked by backend lock, the backend lock can be null in some cases,
205 * for example, BIO belongs to this type, in this case, proc lock is
206 * reused.
207 * c - locked by aio_job_mtx, the lock for the generic file I/O backend.
208 */
209
210 /*
211 * If the routine that services an AIO request blocks while running in an
212 * AIO kernel process it can starve other I/O requests. BIO requests
213 * queued via aio_qbio() complete asynchronously and do not use AIO kernel
214 * processes at all. Socket I/O requests use a separate pool of
215 * kprocs and also force non-blocking I/O. Other file I/O requests
216 * use the generic fo_read/fo_write operations which can block. The
217 * fsync and mlock operations can also block while executing. Ideally
218 * none of these requests would block while executing.
219 *
220 * Note that the service routines cannot toggle O_NONBLOCK in the file
221 * structure directly while handling a request due to races with
222 * userland threads.
223 */
224
225 /* jobflags */
226 #define KAIOCB_QUEUEING 0x01
227 #define KAIOCB_CANCELLED 0x02
228 #define KAIOCB_CANCELLING 0x04
229 #define KAIOCB_CHECKSYNC 0x08
230 #define KAIOCB_CLEARED 0x10
231 #define KAIOCB_FINISHED 0x20
232
233 /*
234 * AIO process info
235 */
236 #define AIOP_FREE 0x1 /* proc on free queue */
237
238 struct aioproc {
239 int aioprocflags; /* (c) AIO proc flags */
240 TAILQ_ENTRY(aioproc) list; /* (c) list of processes */
241 struct proc *aioproc; /* (*) the AIO proc */
242 };
243
244 /*
245 * data-structure for lio signal management
246 */
247 struct aioliojob {
248 int lioj_flags; /* (a) listio flags */
249 int lioj_count; /* (a) count of jobs */
250 int lioj_finished_count; /* (a) count of finished jobs */
251 struct sigevent lioj_signal; /* (a) signal on all I/O done */
252 TAILQ_ENTRY(aioliojob) lioj_list; /* (a) lio list */
253 struct knlist klist; /* (a) list of knotes */
254 ksiginfo_t lioj_ksi; /* (a) Realtime signal info */
255 };
256
257 #define LIOJ_SIGNAL 0x1 /* signal on all done (lio) */
258 #define LIOJ_SIGNAL_POSTED 0x2 /* signal has been posted */
259 #define LIOJ_KEVENT_POSTED 0x4 /* kevent triggered */
260
261 /*
262 * per process aio data structure
263 */
264 struct kaioinfo {
265 struct mtx kaio_mtx; /* the lock to protect this struct */
266 int kaio_flags; /* (a) per process kaio flags */
267 int kaio_active_count; /* (c) number of currently used AIOs */
268 int kaio_count; /* (a) size of AIO queue */
269 int kaio_buffer_count; /* (a) number of bio buffers */
270 TAILQ_HEAD(,kaiocb) kaio_all; /* (a) all AIOs in a process */
271 TAILQ_HEAD(,kaiocb) kaio_done; /* (a) done queue for process */
272 TAILQ_HEAD(,aioliojob) kaio_liojoblist; /* (a) list of lio jobs */
273 TAILQ_HEAD(,kaiocb) kaio_jobqueue; /* (a) job queue for process */
274 TAILQ_HEAD(,kaiocb) kaio_syncqueue; /* (a) queue for aio_fsync */
275 TAILQ_HEAD(,kaiocb) kaio_syncready; /* (a) second q for aio_fsync */
276 struct task kaio_task; /* (*) task to kick aio processes */
277 struct task kaio_sync_task; /* (*) task to schedule fsync jobs */
278 };
279
280 #define AIO_LOCK(ki) mtx_lock(&(ki)->kaio_mtx)
281 #define AIO_UNLOCK(ki) mtx_unlock(&(ki)->kaio_mtx)
282 #define AIO_LOCK_ASSERT(ki, f) mtx_assert(&(ki)->kaio_mtx, (f))
283 #define AIO_MTX(ki) (&(ki)->kaio_mtx)
284
285 #define KAIO_RUNDOWN 0x1 /* process is being run down */
286 #define KAIO_WAKEUP 0x2 /* wakeup process when AIO completes */
287
288 /*
289 * Operations used to interact with userland aio control blocks.
290 * Different ABIs provide their own operations.
291 */
292 struct aiocb_ops {
293 int (*aio_copyin)(struct aiocb *ujob, struct kaiocb *kjob, int ty);
294 long (*fetch_status)(struct aiocb *ujob);
295 long (*fetch_error)(struct aiocb *ujob);
296 int (*store_status)(struct aiocb *ujob, long status);
297 int (*store_error)(struct aiocb *ujob, long error);
298 int (*store_kernelinfo)(struct aiocb *ujob, long jobref);
299 int (*store_aiocb)(struct aiocb **ujobp, struct aiocb *ujob);
300 };
301
302 static TAILQ_HEAD(,aioproc) aio_freeproc; /* (c) Idle daemons */
303 static struct sema aio_newproc_sem;
304 static struct mtx aio_job_mtx;
305 static TAILQ_HEAD(,kaiocb) aio_jobs; /* (c) Async job list */
306 static struct unrhdr *aiod_unr;
307
308 static void aio_biocleanup(struct bio *bp);
309 void aio_init_aioinfo(struct proc *p);
310 static int aio_onceonly(void);
311 static int aio_free_entry(struct kaiocb *job);
312 static void aio_process_rw(struct kaiocb *job);
313 static void aio_process_sync(struct kaiocb *job);
314 static void aio_process_mlock(struct kaiocb *job);
315 static void aio_schedule_fsync(void *context, int pending);
316 static int aio_newproc(int *);
317 int aio_aqueue(struct thread *td, struct aiocb *ujob,
318 struct aioliojob *lio, int type, struct aiocb_ops *ops);
319 static int aio_queue_file(struct file *fp, struct kaiocb *job);
320 static void aio_biowakeup(struct bio *bp);
321 static void aio_proc_rundown(void *arg, struct proc *p);
322 static void aio_proc_rundown_exec(void *arg, struct proc *p,
323 struct image_params *imgp);
324 static int aio_qbio(struct proc *p, struct kaiocb *job);
325 static void aio_daemon(void *param);
326 static void aio_bio_done_notify(struct proc *userp, struct kaiocb *job);
327 static bool aio_clear_cancel_function_locked(struct kaiocb *job);
328 static int aio_kick(struct proc *userp);
329 static void aio_kick_nowait(struct proc *userp);
330 static void aio_kick_helper(void *context, int pending);
331 static int filt_aioattach(struct knote *kn);
332 static void filt_aiodetach(struct knote *kn);
333 static int filt_aio(struct knote *kn, long hint);
334 static int filt_lioattach(struct knote *kn);
335 static void filt_liodetach(struct knote *kn);
336 static int filt_lio(struct knote *kn, long hint);
337
338 /*
339 * Zones for:
340 * kaio Per process async io info
341 * aiocb async io jobs
342 * aiolio list io jobs
343 */
344 static uma_zone_t kaio_zone, aiocb_zone, aiolio_zone;
345
346 /* kqueue filters for aio */
347 static struct filterops aio_filtops = {
348 .f_isfd = 0,
349 .f_attach = filt_aioattach,
350 .f_detach = filt_aiodetach,
351 .f_event = filt_aio,
352 };
353 static struct filterops lio_filtops = {
354 .f_isfd = 0,
355 .f_attach = filt_lioattach,
356 .f_detach = filt_liodetach,
357 .f_event = filt_lio
358 };
359
360 static eventhandler_tag exit_tag, exec_tag;
361
362 TASKQUEUE_DEFINE_THREAD(aiod_kick);
363
364 /*
365 * Main operations function for use as a kernel module.
366 */
367 static int
aio_modload(struct module * module,int cmd,void * arg)368 aio_modload(struct module *module, int cmd, void *arg)
369 {
370 int error = 0;
371
372 switch (cmd) {
373 case MOD_LOAD:
374 aio_onceonly();
375 break;
376 case MOD_SHUTDOWN:
377 break;
378 default:
379 error = EOPNOTSUPP;
380 break;
381 }
382 return (error);
383 }
384
385 static moduledata_t aio_mod = {
386 "aio",
387 &aio_modload,
388 NULL
389 };
390
391 DECLARE_MODULE(aio, aio_mod, SI_SUB_VFS, SI_ORDER_ANY);
392 MODULE_VERSION(aio, 1);
393
394 /*
395 * Startup initialization
396 */
397 static int
aio_onceonly(void)398 aio_onceonly(void)
399 {
400
401 exit_tag = EVENTHANDLER_REGISTER(process_exit, aio_proc_rundown, NULL,
402 EVENTHANDLER_PRI_ANY);
403 exec_tag = EVENTHANDLER_REGISTER(process_exec, aio_proc_rundown_exec,
404 NULL, EVENTHANDLER_PRI_ANY);
405 kqueue_add_filteropts(EVFILT_AIO, &aio_filtops);
406 kqueue_add_filteropts(EVFILT_LIO, &lio_filtops);
407 TAILQ_INIT(&aio_freeproc);
408 sema_init(&aio_newproc_sem, 0, "aio_new_proc");
409 mtx_init(&aio_job_mtx, "aio_job", NULL, MTX_DEF);
410 TAILQ_INIT(&aio_jobs);
411 aiod_unr = new_unrhdr(1, INT_MAX, NULL);
412 kaio_zone = uma_zcreate("AIO", sizeof(struct kaioinfo), NULL, NULL,
413 NULL, NULL, UMA_ALIGN_PTR, 0);
414 aiocb_zone = uma_zcreate("AIOCB", sizeof(struct kaiocb), NULL, NULL,
415 NULL, NULL, UMA_ALIGN_PTR, 0);
416 aiolio_zone = uma_zcreate("AIOLIO", sizeof(struct aioliojob), NULL,
417 NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
418 aiod_lifetime = AIOD_LIFETIME_DEFAULT;
419 jobrefid = 1;
420 p31b_setcfg(CTL_P1003_1B_ASYNCHRONOUS_IO, _POSIX_ASYNCHRONOUS_IO);
421 p31b_setcfg(CTL_P1003_1B_AIO_MAX, MAX_AIO_QUEUE);
422 p31b_setcfg(CTL_P1003_1B_AIO_PRIO_DELTA_MAX, 0);
423
424 return (0);
425 }
426
427 /*
428 * Init the per-process aioinfo structure. The aioinfo limits are set
429 * per-process for user limit (resource) management.
430 */
431 void
aio_init_aioinfo(struct proc * p)432 aio_init_aioinfo(struct proc *p)
433 {
434 struct kaioinfo *ki;
435
436 ki = uma_zalloc(kaio_zone, M_WAITOK);
437 mtx_init(&ki->kaio_mtx, "aiomtx", NULL, MTX_DEF | MTX_NEW);
438 ki->kaio_flags = 0;
439 ki->kaio_active_count = 0;
440 ki->kaio_count = 0;
441 ki->kaio_buffer_count = 0;
442 TAILQ_INIT(&ki->kaio_all);
443 TAILQ_INIT(&ki->kaio_done);
444 TAILQ_INIT(&ki->kaio_jobqueue);
445 TAILQ_INIT(&ki->kaio_liojoblist);
446 TAILQ_INIT(&ki->kaio_syncqueue);
447 TAILQ_INIT(&ki->kaio_syncready);
448 TASK_INIT(&ki->kaio_task, 0, aio_kick_helper, p);
449 TASK_INIT(&ki->kaio_sync_task, 0, aio_schedule_fsync, ki);
450 PROC_LOCK(p);
451 if (p->p_aioinfo == NULL) {
452 p->p_aioinfo = ki;
453 PROC_UNLOCK(p);
454 } else {
455 PROC_UNLOCK(p);
456 mtx_destroy(&ki->kaio_mtx);
457 uma_zfree(kaio_zone, ki);
458 }
459
460 while (num_aio_procs < MIN(target_aio_procs, max_aio_procs))
461 aio_newproc(NULL);
462 }
463
464 static int
aio_sendsig(struct proc * p,struct sigevent * sigev,ksiginfo_t * ksi,bool ext)465 aio_sendsig(struct proc *p, struct sigevent *sigev, ksiginfo_t *ksi, bool ext)
466 {
467 struct thread *td;
468 int error;
469
470 error = sigev_findtd(p, sigev, &td);
471 if (error)
472 return (error);
473 if (!KSI_ONQ(ksi)) {
474 ksiginfo_set_sigev(ksi, sigev);
475 ksi->ksi_code = SI_ASYNCIO;
476 ksi->ksi_flags |= ext ? (KSI_EXT | KSI_INS) : 0;
477 tdsendsignal(p, td, ksi->ksi_signo, ksi);
478 }
479 PROC_UNLOCK(p);
480 return (error);
481 }
482
483 /*
484 * Free a job entry. Wait for completion if it is currently active, but don't
485 * delay forever. If we delay, we return a flag that says that we have to
486 * restart the queue scan.
487 */
488 static int
aio_free_entry(struct kaiocb * job)489 aio_free_entry(struct kaiocb *job)
490 {
491 struct kaioinfo *ki;
492 struct aioliojob *lj;
493 struct proc *p;
494
495 p = job->userproc;
496 MPASS(curproc == p);
497 ki = p->p_aioinfo;
498 MPASS(ki != NULL);
499
500 AIO_LOCK_ASSERT(ki, MA_OWNED);
501 MPASS(job->jobflags & KAIOCB_FINISHED);
502
503 atomic_subtract_int(&num_queue_count, 1);
504
505 ki->kaio_count--;
506 MPASS(ki->kaio_count >= 0);
507
508 TAILQ_REMOVE(&ki->kaio_done, job, plist);
509 TAILQ_REMOVE(&ki->kaio_all, job, allist);
510
511 lj = job->lio;
512 if (lj) {
513 lj->lioj_count--;
514 lj->lioj_finished_count--;
515
516 if (lj->lioj_count == 0) {
517 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list);
518 /* lio is going away, we need to destroy any knotes */
519 knlist_delete(&lj->klist, curthread, 1);
520 PROC_LOCK(p);
521 sigqueue_take(&lj->lioj_ksi);
522 PROC_UNLOCK(p);
523 uma_zfree(aiolio_zone, lj);
524 }
525 }
526
527 /* job is going away, we need to destroy any knotes */
528 knlist_delete(&job->klist, curthread, 1);
529 PROC_LOCK(p);
530 sigqueue_take(&job->ksi);
531 PROC_UNLOCK(p);
532
533 AIO_UNLOCK(ki);
534
535 /*
536 * The thread argument here is used to find the owning process
537 * and is also passed to fo_close() which may pass it to various
538 * places such as devsw close() routines. Because of that, we
539 * need a thread pointer from the process owning the job that is
540 * persistent and won't disappear out from under us or move to
541 * another process.
542 *
543 * Currently, all the callers of this function call it to remove
544 * a kaiocb from the current process' job list either via a
545 * syscall or due to the current process calling exit() or
546 * execve(). Thus, we know that p == curproc. We also know that
547 * curthread can't exit since we are curthread.
548 *
549 * Therefore, we use curthread as the thread to pass to
550 * knlist_delete(). This does mean that it is possible for the
551 * thread pointer at close time to differ from the thread pointer
552 * at open time, but this is already true of file descriptors in
553 * a multithreaded process.
554 */
555 if (job->fd_file)
556 fdrop(job->fd_file, curthread);
557 crfree(job->cred);
558 if (job->uiop != &job->uio)
559 free(job->uiop, M_IOV);
560 uma_zfree(aiocb_zone, job);
561 AIO_LOCK(ki);
562
563 return (0);
564 }
565
566 static void
aio_proc_rundown_exec(void * arg,struct proc * p,struct image_params * imgp __unused)567 aio_proc_rundown_exec(void *arg, struct proc *p,
568 struct image_params *imgp __unused)
569 {
570 aio_proc_rundown(arg, p);
571 }
572
573 static int
aio_cancel_job(struct proc * p,struct kaioinfo * ki,struct kaiocb * job)574 aio_cancel_job(struct proc *p, struct kaioinfo *ki, struct kaiocb *job)
575 {
576 aio_cancel_fn_t *func;
577 int cancelled;
578
579 AIO_LOCK_ASSERT(ki, MA_OWNED);
580 if (job->jobflags & (KAIOCB_CANCELLED | KAIOCB_FINISHED))
581 return (0);
582 MPASS((job->jobflags & KAIOCB_CANCELLING) == 0);
583 job->jobflags |= KAIOCB_CANCELLED;
584
585 func = job->cancel_fn;
586
587 /*
588 * If there is no cancel routine, just leave the job marked as
589 * cancelled. The job should be in active use by a caller who
590 * should complete it normally or when it fails to install a
591 * cancel routine.
592 */
593 if (func == NULL)
594 return (0);
595
596 /*
597 * Set the CANCELLING flag so that aio_complete() will defer
598 * completions of this job. This prevents the job from being
599 * freed out from under the cancel callback. After the
600 * callback any deferred completion (whether from the callback
601 * or any other source) will be completed.
602 */
603 job->jobflags |= KAIOCB_CANCELLING;
604 AIO_UNLOCK(ki);
605 func(job);
606 AIO_LOCK(ki);
607 job->jobflags &= ~KAIOCB_CANCELLING;
608 if (job->jobflags & KAIOCB_FINISHED) {
609 cancelled = job->uaiocb._aiocb_private.error == ECANCELED;
610 TAILQ_REMOVE(&ki->kaio_jobqueue, job, plist);
611 aio_bio_done_notify(p, job);
612 } else {
613 /*
614 * The cancel callback might have scheduled an
615 * operation to cancel this request, but it is
616 * only counted as cancelled if the request is
617 * cancelled when the callback returns.
618 */
619 cancelled = 0;
620 }
621 return (cancelled);
622 }
623
624 /*
625 * Rundown the jobs for a given process.
626 */
627 static void
aio_proc_rundown(void * arg,struct proc * p)628 aio_proc_rundown(void *arg, struct proc *p)
629 {
630 struct kaioinfo *ki;
631 struct aioliojob *lj;
632 struct kaiocb *job, *jobn;
633
634 KASSERT(curthread->td_proc == p,
635 ("%s: called on non-curproc", __func__));
636 ki = p->p_aioinfo;
637 if (ki == NULL)
638 return;
639
640 AIO_LOCK(ki);
641 ki->kaio_flags |= KAIO_RUNDOWN;
642
643 restart:
644
645 /*
646 * Try to cancel all pending requests. This code simulates
647 * aio_cancel on all pending I/O requests.
648 */
649 TAILQ_FOREACH_SAFE(job, &ki->kaio_jobqueue, plist, jobn) {
650 aio_cancel_job(p, ki, job);
651 }
652
653 /* Wait for all running I/O to be finished */
654 if (TAILQ_FIRST(&ki->kaio_jobqueue) || ki->kaio_active_count != 0) {
655 ki->kaio_flags |= KAIO_WAKEUP;
656 msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO, "aioprn", hz);
657 goto restart;
658 }
659
660 /* Free all completed I/O requests. */
661 while ((job = TAILQ_FIRST(&ki->kaio_done)) != NULL)
662 aio_free_entry(job);
663
664 while ((lj = TAILQ_FIRST(&ki->kaio_liojoblist)) != NULL) {
665 if (lj->lioj_count == 0) {
666 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list);
667 knlist_delete(&lj->klist, curthread, 1);
668 PROC_LOCK(p);
669 sigqueue_take(&lj->lioj_ksi);
670 PROC_UNLOCK(p);
671 uma_zfree(aiolio_zone, lj);
672 } else {
673 panic("LIO job not cleaned up: C:%d, FC:%d\n",
674 lj->lioj_count, lj->lioj_finished_count);
675 }
676 }
677 AIO_UNLOCK(ki);
678 taskqueue_drain(taskqueue_aiod_kick, &ki->kaio_task);
679 taskqueue_drain(taskqueue_aiod_kick, &ki->kaio_sync_task);
680 mtx_destroy(&ki->kaio_mtx);
681 uma_zfree(kaio_zone, ki);
682 p->p_aioinfo = NULL;
683 }
684
685 /*
686 * Select a job to run (called by an AIO daemon).
687 */
688 static struct kaiocb *
aio_selectjob(struct aioproc * aiop)689 aio_selectjob(struct aioproc *aiop)
690 {
691 struct kaiocb *job;
692 struct kaioinfo *ki;
693 struct proc *userp;
694
695 mtx_assert(&aio_job_mtx, MA_OWNED);
696 restart:
697 TAILQ_FOREACH(job, &aio_jobs, list) {
698 userp = job->userproc;
699 ki = userp->p_aioinfo;
700
701 if (ki->kaio_active_count < max_aio_per_proc) {
702 TAILQ_REMOVE(&aio_jobs, job, list);
703 if (!aio_clear_cancel_function(job))
704 goto restart;
705
706 /* Account for currently active jobs. */
707 ki->kaio_active_count++;
708 break;
709 }
710 }
711 return (job);
712 }
713
714 /*
715 * Move all data to a permanent storage device. This code
716 * simulates the fsync and fdatasync syscalls.
717 */
718 static int
aio_fsync_vnode(struct thread * td,struct vnode * vp,int op)719 aio_fsync_vnode(struct thread *td, struct vnode *vp, int op)
720 {
721 struct mount *mp;
722 int error;
723
724 for (;;) {
725 error = vn_start_write(vp, &mp, V_WAIT | PCATCH);
726 if (error != 0)
727 break;
728 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY);
729 vnode_pager_clean_async(vp);
730 if (op == LIO_DSYNC)
731 error = VOP_FDATASYNC(vp, td);
732 else
733 error = VOP_FSYNC(vp, MNT_WAIT, td);
734
735 VOP_UNLOCK(vp);
736 vn_finished_write(mp);
737 if (error != ERELOOKUP)
738 break;
739 }
740 return (error);
741 }
742
743 /*
744 * The AIO processing activity for LIO_READ/LIO_WRITE. This is the code that
745 * does the I/O request for the non-bio version of the operations. The normal
746 * vn operations are used, and this code should work in all instances for every
747 * type of file, including pipes, sockets, fifos, and regular files.
748 *
749 * XXX I don't think it works well for socket, pipe, and fifo.
750 */
751 static void
aio_process_rw(struct kaiocb * job)752 aio_process_rw(struct kaiocb *job)
753 {
754 struct ucred *td_savedcred;
755 struct thread *td;
756 struct file *fp;
757 ssize_t cnt;
758 long msgsnd_st, msgsnd_end;
759 long msgrcv_st, msgrcv_end;
760 long oublock_st, oublock_end;
761 long inblock_st, inblock_end;
762 int error, opcode;
763
764 KASSERT(job->uaiocb.aio_lio_opcode == LIO_READ ||
765 job->uaiocb.aio_lio_opcode == LIO_READV ||
766 job->uaiocb.aio_lio_opcode == LIO_WRITE ||
767 job->uaiocb.aio_lio_opcode == LIO_WRITEV,
768 ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode));
769
770 aio_switch_vmspace(job);
771 td = curthread;
772 td_savedcred = td->td_ucred;
773 td->td_ucred = job->cred;
774 job->uiop->uio_td = td;
775 fp = job->fd_file;
776
777 opcode = job->uaiocb.aio_lio_opcode;
778 cnt = job->uiop->uio_resid;
779
780 msgrcv_st = td->td_ru.ru_msgrcv;
781 msgsnd_st = td->td_ru.ru_msgsnd;
782 inblock_st = td->td_ru.ru_inblock;
783 oublock_st = td->td_ru.ru_oublock;
784
785 /*
786 * aio_aqueue() acquires a reference to the file that is
787 * released in aio_free_entry().
788 */
789 if (opcode == LIO_READ || opcode == LIO_READV) {
790 if (job->uiop->uio_resid == 0)
791 error = 0;
792 else
793 error = fo_read(fp, job->uiop, fp->f_cred, FOF_OFFSET,
794 td);
795 } else {
796 if (fp->f_type == DTYPE_VNODE)
797 bwillwrite();
798 error = fo_write(fp, job->uiop, fp->f_cred, FOF_OFFSET, td);
799 }
800 msgrcv_end = td->td_ru.ru_msgrcv;
801 msgsnd_end = td->td_ru.ru_msgsnd;
802 inblock_end = td->td_ru.ru_inblock;
803 oublock_end = td->td_ru.ru_oublock;
804
805 job->msgrcv = msgrcv_end - msgrcv_st;
806 job->msgsnd = msgsnd_end - msgsnd_st;
807 job->inblock = inblock_end - inblock_st;
808 job->outblock = oublock_end - oublock_st;
809
810 if (error != 0 && job->uiop->uio_resid != cnt) {
811 if (error == ERESTART || error == EINTR || error == EWOULDBLOCK)
812 error = 0;
813 if (error == EPIPE && (opcode & LIO_WRITE)) {
814 PROC_LOCK(job->userproc);
815 kern_psignal(job->userproc, SIGPIPE);
816 PROC_UNLOCK(job->userproc);
817 }
818 }
819
820 cnt -= job->uiop->uio_resid;
821 td->td_ucred = td_savedcred;
822 if (error)
823 aio_complete(job, -1, error);
824 else
825 aio_complete(job, cnt, 0);
826 }
827
828 static void
aio_process_sync(struct kaiocb * job)829 aio_process_sync(struct kaiocb *job)
830 {
831 struct thread *td = curthread;
832 struct ucred *td_savedcred = td->td_ucred;
833 struct file *fp = job->fd_file;
834 int error = 0;
835
836 KASSERT(job->uaiocb.aio_lio_opcode & LIO_SYNC,
837 ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode));
838
839 td->td_ucred = job->cred;
840 if (fp->f_vnode != NULL) {
841 error = aio_fsync_vnode(td, fp->f_vnode,
842 job->uaiocb.aio_lio_opcode);
843 }
844 td->td_ucred = td_savedcred;
845 if (error)
846 aio_complete(job, -1, error);
847 else
848 aio_complete(job, 0, 0);
849 }
850
851 static void
aio_process_mlock(struct kaiocb * job)852 aio_process_mlock(struct kaiocb *job)
853 {
854 struct aiocb *cb = &job->uaiocb;
855 int error;
856
857 KASSERT(job->uaiocb.aio_lio_opcode == LIO_MLOCK,
858 ("%s: opcode %d", __func__, job->uaiocb.aio_lio_opcode));
859
860 aio_switch_vmspace(job);
861 error = kern_mlock(job->userproc, job->cred,
862 __DEVOLATILE(uintptr_t, cb->aio_buf), cb->aio_nbytes);
863 aio_complete(job, error != 0 ? -1 : 0, error);
864 }
865
866 static void
aio_bio_done_notify(struct proc * userp,struct kaiocb * job)867 aio_bio_done_notify(struct proc *userp, struct kaiocb *job)
868 {
869 struct aioliojob *lj;
870 struct kaioinfo *ki;
871 struct kaiocb *sjob, *sjobn;
872 int lj_done;
873 bool schedule_fsync;
874
875 ki = userp->p_aioinfo;
876 AIO_LOCK_ASSERT(ki, MA_OWNED);
877 lj = job->lio;
878 lj_done = 0;
879 if (lj) {
880 lj->lioj_finished_count++;
881 if (lj->lioj_count == lj->lioj_finished_count)
882 lj_done = 1;
883 }
884 TAILQ_INSERT_TAIL(&ki->kaio_done, job, plist);
885 MPASS(job->jobflags & KAIOCB_FINISHED);
886
887 if (ki->kaio_flags & KAIO_RUNDOWN)
888 goto notification_done;
889
890 if (job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL ||
891 job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID)
892 aio_sendsig(userp, &job->uaiocb.aio_sigevent, &job->ksi, true);
893
894 KNOTE_LOCKED(&job->klist, 1);
895
896 if (lj_done) {
897 if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) {
898 lj->lioj_flags |= LIOJ_KEVENT_POSTED;
899 KNOTE_LOCKED(&lj->klist, 1);
900 }
901 if ((lj->lioj_flags & (LIOJ_SIGNAL | LIOJ_SIGNAL_POSTED))
902 == LIOJ_SIGNAL &&
903 (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL ||
904 lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID)) {
905 aio_sendsig(userp, &lj->lioj_signal, &lj->lioj_ksi,
906 true);
907 lj->lioj_flags |= LIOJ_SIGNAL_POSTED;
908 }
909 }
910
911 notification_done:
912 if (job->jobflags & KAIOCB_CHECKSYNC) {
913 schedule_fsync = false;
914 TAILQ_FOREACH_SAFE(sjob, &ki->kaio_syncqueue, list, sjobn) {
915 if (job->fd_file != sjob->fd_file ||
916 job->seqno >= sjob->seqno)
917 continue;
918 if (--sjob->pending > 0)
919 continue;
920 TAILQ_REMOVE(&ki->kaio_syncqueue, sjob, list);
921 if (!aio_clear_cancel_function_locked(sjob))
922 continue;
923 TAILQ_INSERT_TAIL(&ki->kaio_syncready, sjob, list);
924 schedule_fsync = true;
925 }
926 if (schedule_fsync)
927 taskqueue_enqueue(taskqueue_aiod_kick,
928 &ki->kaio_sync_task);
929 }
930 if (ki->kaio_flags & KAIO_WAKEUP) {
931 ki->kaio_flags &= ~KAIO_WAKEUP;
932 wakeup(&userp->p_aioinfo);
933 }
934 }
935
936 static void
aio_schedule_fsync(void * context,int pending)937 aio_schedule_fsync(void *context, int pending)
938 {
939 struct kaioinfo *ki;
940 struct kaiocb *job;
941
942 ki = context;
943 AIO_LOCK(ki);
944 while (!TAILQ_EMPTY(&ki->kaio_syncready)) {
945 job = TAILQ_FIRST(&ki->kaio_syncready);
946 TAILQ_REMOVE(&ki->kaio_syncready, job, list);
947 AIO_UNLOCK(ki);
948 aio_schedule(job, aio_process_sync);
949 AIO_LOCK(ki);
950 }
951 AIO_UNLOCK(ki);
952 }
953
954 bool
aio_cancel_cleared(struct kaiocb * job)955 aio_cancel_cleared(struct kaiocb *job)
956 {
957
958 /*
959 * The caller should hold the same queue lock held when
960 * aio_clear_cancel_function() was called and set this flag
961 * ensuring this check sees an up-to-date value. However,
962 * there is no way to assert that.
963 */
964 return ((job->jobflags & KAIOCB_CLEARED) != 0);
965 }
966
967 static bool
aio_clear_cancel_function_locked(struct kaiocb * job)968 aio_clear_cancel_function_locked(struct kaiocb *job)
969 {
970
971 AIO_LOCK_ASSERT(job->userproc->p_aioinfo, MA_OWNED);
972 MPASS(job->cancel_fn != NULL);
973 if (job->jobflags & KAIOCB_CANCELLING) {
974 job->jobflags |= KAIOCB_CLEARED;
975 return (false);
976 }
977 job->cancel_fn = NULL;
978 return (true);
979 }
980
981 bool
aio_clear_cancel_function(struct kaiocb * job)982 aio_clear_cancel_function(struct kaiocb *job)
983 {
984 struct kaioinfo *ki;
985 bool ret;
986
987 ki = job->userproc->p_aioinfo;
988 AIO_LOCK(ki);
989 ret = aio_clear_cancel_function_locked(job);
990 AIO_UNLOCK(ki);
991 return (ret);
992 }
993
994 static bool
aio_set_cancel_function_locked(struct kaiocb * job,aio_cancel_fn_t * func)995 aio_set_cancel_function_locked(struct kaiocb *job, aio_cancel_fn_t *func)
996 {
997
998 AIO_LOCK_ASSERT(job->userproc->p_aioinfo, MA_OWNED);
999 if (job->jobflags & KAIOCB_CANCELLED)
1000 return (false);
1001 job->cancel_fn = func;
1002 return (true);
1003 }
1004
1005 bool
aio_set_cancel_function(struct kaiocb * job,aio_cancel_fn_t * func)1006 aio_set_cancel_function(struct kaiocb *job, aio_cancel_fn_t *func)
1007 {
1008 struct kaioinfo *ki;
1009 bool ret;
1010
1011 ki = job->userproc->p_aioinfo;
1012 AIO_LOCK(ki);
1013 ret = aio_set_cancel_function_locked(job, func);
1014 AIO_UNLOCK(ki);
1015 return (ret);
1016 }
1017
1018 void
aio_complete(struct kaiocb * job,long status,int error)1019 aio_complete(struct kaiocb *job, long status, int error)
1020 {
1021 struct kaioinfo *ki;
1022 struct proc *userp;
1023
1024 job->uaiocb._aiocb_private.error = error;
1025 job->uaiocb._aiocb_private.status = status;
1026
1027 userp = job->userproc;
1028 ki = userp->p_aioinfo;
1029
1030 AIO_LOCK(ki);
1031 KASSERT(!(job->jobflags & KAIOCB_FINISHED),
1032 ("duplicate aio_complete"));
1033 job->jobflags |= KAIOCB_FINISHED;
1034 if ((job->jobflags & (KAIOCB_QUEUEING | KAIOCB_CANCELLING)) == 0) {
1035 TAILQ_REMOVE(&ki->kaio_jobqueue, job, plist);
1036 aio_bio_done_notify(userp, job);
1037 }
1038 AIO_UNLOCK(ki);
1039 }
1040
1041 void
aio_cancel(struct kaiocb * job)1042 aio_cancel(struct kaiocb *job)
1043 {
1044
1045 aio_complete(job, -1, ECANCELED);
1046 }
1047
1048 void
aio_switch_vmspace(struct kaiocb * job)1049 aio_switch_vmspace(struct kaiocb *job)
1050 {
1051
1052 vmspace_switch_aio(job->userproc->p_vmspace);
1053 }
1054
1055 /*
1056 * The AIO daemon, most of the actual work is done in aio_process_*,
1057 * but the setup (and address space mgmt) is done in this routine.
1058 */
1059 static void
aio_daemon(void * _id)1060 aio_daemon(void *_id)
1061 {
1062 struct kaiocb *job;
1063 struct aioproc *aiop;
1064 struct kaioinfo *ki;
1065 struct proc *p;
1066 struct vmspace *myvm;
1067 struct thread *td = curthread;
1068 int id = (intptr_t)_id;
1069
1070 /*
1071 * Grab an extra reference on the daemon's vmspace so that it
1072 * doesn't get freed by jobs that switch to a different
1073 * vmspace.
1074 */
1075 p = td->td_proc;
1076 myvm = vmspace_acquire_ref(p);
1077
1078 KASSERT(p->p_textvp == NULL, ("kthread has a textvp"));
1079
1080 /*
1081 * Allocate and ready the aio control info. There is one aiop structure
1082 * per daemon.
1083 */
1084 aiop = malloc(sizeof(*aiop), M_AIO, M_WAITOK);
1085 aiop->aioproc = p;
1086 aiop->aioprocflags = 0;
1087
1088 /*
1089 * Wakeup parent process. (Parent sleeps to keep from blasting away
1090 * and creating too many daemons.)
1091 */
1092 sema_post(&aio_newproc_sem);
1093
1094 mtx_lock(&aio_job_mtx);
1095 for (;;) {
1096 /*
1097 * Take daemon off of free queue
1098 */
1099 if (aiop->aioprocflags & AIOP_FREE) {
1100 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1101 aiop->aioprocflags &= ~AIOP_FREE;
1102 }
1103
1104 /*
1105 * Check for jobs.
1106 */
1107 while ((job = aio_selectjob(aiop)) != NULL) {
1108 mtx_unlock(&aio_job_mtx);
1109
1110 ki = job->userproc->p_aioinfo;
1111 job->handle_fn(job);
1112
1113 mtx_lock(&aio_job_mtx);
1114 /* Decrement the active job count. */
1115 ki->kaio_active_count--;
1116 }
1117
1118 /*
1119 * Disconnect from user address space.
1120 */
1121 if (p->p_vmspace != myvm) {
1122 mtx_unlock(&aio_job_mtx);
1123 vmspace_switch_aio(myvm);
1124 mtx_lock(&aio_job_mtx);
1125 /*
1126 * We have to restart to avoid race, we only sleep if
1127 * no job can be selected.
1128 */
1129 continue;
1130 }
1131
1132 mtx_assert(&aio_job_mtx, MA_OWNED);
1133
1134 TAILQ_INSERT_HEAD(&aio_freeproc, aiop, list);
1135 aiop->aioprocflags |= AIOP_FREE;
1136
1137 /*
1138 * If daemon is inactive for a long time, allow it to exit,
1139 * thereby freeing resources.
1140 */
1141 if (msleep(p, &aio_job_mtx, PRIBIO, "aiordy",
1142 aiod_lifetime) == EWOULDBLOCK && TAILQ_EMPTY(&aio_jobs) &&
1143 (aiop->aioprocflags & AIOP_FREE) &&
1144 num_aio_procs > target_aio_procs)
1145 break;
1146 }
1147 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1148 num_aio_procs--;
1149 mtx_unlock(&aio_job_mtx);
1150 free(aiop, M_AIO);
1151 free_unr(aiod_unr, id);
1152 vmspace_free(myvm);
1153
1154 KASSERT(p->p_vmspace == myvm,
1155 ("AIOD: bad vmspace for exiting daemon"));
1156 KASSERT(refcount_load(&myvm->vm_refcnt) > 1,
1157 ("AIOD: bad vm refcnt for exiting daemon: %d",
1158 refcount_load(&myvm->vm_refcnt)));
1159 kproc_exit(0);
1160 }
1161
1162 /*
1163 * Create a new AIO daemon. This is mostly a kernel-thread fork routine. The
1164 * AIO daemon modifies its environment itself.
1165 */
1166 static int
aio_newproc(int * start)1167 aio_newproc(int *start)
1168 {
1169 int error;
1170 struct proc *p;
1171 int id;
1172
1173 id = alloc_unr(aiod_unr);
1174 error = kproc_create(aio_daemon, (void *)(intptr_t)id, &p,
1175 RFNOWAIT, 0, "aiod%d", id);
1176 if (error == 0) {
1177 /*
1178 * Wait until daemon is started.
1179 */
1180 sema_wait(&aio_newproc_sem);
1181 mtx_lock(&aio_job_mtx);
1182 num_aio_procs++;
1183 if (start != NULL)
1184 (*start)--;
1185 mtx_unlock(&aio_job_mtx);
1186 } else {
1187 free_unr(aiod_unr, id);
1188 }
1189 return (error);
1190 }
1191
1192 /*
1193 * Try the high-performance, low-overhead bio method for eligible
1194 * VCHR devices. This method doesn't use an aio helper thread, and
1195 * thus has very low overhead.
1196 *
1197 * Assumes that the caller, aio_aqueue(), has incremented the file
1198 * structure's reference count, preventing its deallocation for the
1199 * duration of this call.
1200 */
1201 static int
aio_qbio(struct proc * p,struct kaiocb * job)1202 aio_qbio(struct proc *p, struct kaiocb *job)
1203 {
1204 struct aiocb *cb;
1205 struct file *fp;
1206 struct buf *pbuf;
1207 struct vnode *vp;
1208 struct cdevsw *csw;
1209 struct cdev *dev;
1210 struct kaioinfo *ki;
1211 struct bio **bios = NULL;
1212 off_t offset;
1213 int bio_cmd, error, i, iovcnt, opcode, poff, ref;
1214 vm_prot_t prot;
1215 bool use_unmapped;
1216
1217 cb = &job->uaiocb;
1218 fp = job->fd_file;
1219 opcode = cb->aio_lio_opcode;
1220
1221 if (!(opcode == LIO_WRITE || opcode == LIO_WRITEV ||
1222 opcode == LIO_READ || opcode == LIO_READV))
1223 return (-1);
1224 if (fp == NULL || fp->f_type != DTYPE_VNODE)
1225 return (-1);
1226
1227 vp = fp->f_vnode;
1228 if (vp->v_type != VCHR)
1229 return (-1);
1230 if (vp->v_bufobj.bo_bsize == 0)
1231 return (-1);
1232
1233 bio_cmd = (opcode & LIO_WRITE) ? BIO_WRITE : BIO_READ;
1234 iovcnt = job->uiop->uio_iovcnt;
1235 if (iovcnt > max_buf_aio)
1236 return (-1);
1237 for (i = 0; i < iovcnt; i++) {
1238 if (job->uiop->uio_iov[i].iov_len % vp->v_bufobj.bo_bsize != 0)
1239 return (-1);
1240 if (job->uiop->uio_iov[i].iov_len > maxphys) {
1241 error = -1;
1242 return (-1);
1243 }
1244 }
1245 offset = cb->aio_offset;
1246
1247 ref = 0;
1248 csw = devvn_refthread(vp, &dev, &ref);
1249 if (csw == NULL)
1250 return (ENXIO);
1251
1252 if ((csw->d_flags & D_DISK) == 0) {
1253 error = -1;
1254 goto unref;
1255 }
1256 if (job->uiop->uio_resid > dev->si_iosize_max) {
1257 error = -1;
1258 goto unref;
1259 }
1260
1261 ki = p->p_aioinfo;
1262 job->error = 0;
1263
1264 use_unmapped = (dev->si_flags & SI_UNMAPPED) && unmapped_buf_allowed;
1265 if (!use_unmapped) {
1266 AIO_LOCK(ki);
1267 if (ki->kaio_buffer_count + iovcnt > max_buf_aio) {
1268 AIO_UNLOCK(ki);
1269 error = EAGAIN;
1270 goto unref;
1271 }
1272 ki->kaio_buffer_count += iovcnt;
1273 AIO_UNLOCK(ki);
1274 }
1275
1276 bios = malloc(sizeof(struct bio *) * iovcnt, M_TEMP, M_WAITOK);
1277 refcount_init(&job->nbio, iovcnt);
1278 for (i = 0; i < iovcnt; i++) {
1279 struct vm_page** pages;
1280 struct bio *bp;
1281 void *buf;
1282 size_t nbytes;
1283 int npages;
1284
1285 buf = job->uiop->uio_iov[i].iov_base;
1286 nbytes = job->uiop->uio_iov[i].iov_len;
1287
1288 bios[i] = g_alloc_bio();
1289 bp = bios[i];
1290
1291 poff = (vm_offset_t)buf & PAGE_MASK;
1292 if (use_unmapped) {
1293 pbuf = NULL;
1294 pages = malloc(sizeof(vm_page_t) * (atop(round_page(
1295 nbytes)) + 1), M_TEMP, M_WAITOK | M_ZERO);
1296 } else {
1297 pbuf = uma_zalloc(pbuf_zone, M_WAITOK);
1298 BUF_KERNPROC(pbuf);
1299 pages = pbuf->b_pages;
1300 }
1301
1302 bp->bio_length = nbytes;
1303 bp->bio_bcount = nbytes;
1304 bp->bio_done = aio_biowakeup;
1305 bp->bio_offset = offset;
1306 bp->bio_cmd = bio_cmd;
1307 bp->bio_dev = dev;
1308 bp->bio_caller1 = job;
1309 bp->bio_caller2 = pbuf;
1310
1311 prot = VM_PROT_READ;
1312 if (opcode == LIO_READ || opcode == LIO_READV)
1313 prot |= VM_PROT_WRITE; /* Less backwards than it looks */
1314 npages = vm_fault_quick_hold_pages(&curproc->p_vmspace->vm_map,
1315 (vm_offset_t)buf, bp->bio_length, prot, pages,
1316 atop(maxphys) + 1);
1317 if (npages < 0) {
1318 if (pbuf != NULL)
1319 uma_zfree(pbuf_zone, pbuf);
1320 else
1321 free(pages, M_TEMP);
1322 error = EFAULT;
1323 g_destroy_bio(bp);
1324 i--;
1325 goto destroy_bios;
1326 }
1327 if (pbuf != NULL) {
1328 pmap_qenter((vm_offset_t)pbuf->b_data, pages, npages);
1329 bp->bio_data = pbuf->b_data + poff;
1330 pbuf->b_npages = npages;
1331 atomic_add_int(&num_buf_aio, 1);
1332 } else {
1333 bp->bio_ma = pages;
1334 bp->bio_ma_n = npages;
1335 bp->bio_ma_offset = poff;
1336 bp->bio_data = unmapped_buf;
1337 bp->bio_flags |= BIO_UNMAPPED;
1338 atomic_add_int(&num_unmapped_aio, 1);
1339 }
1340
1341 offset += nbytes;
1342 }
1343
1344 /* Perform transfer. */
1345 for (i = 0; i < iovcnt; i++)
1346 csw->d_strategy(bios[i]);
1347 free(bios, M_TEMP);
1348
1349 dev_relthread(dev, ref);
1350 return (0);
1351
1352 destroy_bios:
1353 for (; i >= 0; i--)
1354 aio_biocleanup(bios[i]);
1355 free(bios, M_TEMP);
1356 unref:
1357 dev_relthread(dev, ref);
1358 return (error);
1359 }
1360
1361 #ifdef COMPAT_FREEBSD6
1362 static int
convert_old_sigevent(struct osigevent * osig,struct sigevent * nsig)1363 convert_old_sigevent(struct osigevent *osig, struct sigevent *nsig)
1364 {
1365
1366 /*
1367 * Only SIGEV_NONE, SIGEV_SIGNAL, and SIGEV_KEVENT are
1368 * supported by AIO with the old sigevent structure.
1369 */
1370 nsig->sigev_notify = osig->sigev_notify;
1371 switch (nsig->sigev_notify) {
1372 case SIGEV_NONE:
1373 break;
1374 case SIGEV_SIGNAL:
1375 nsig->sigev_signo = osig->__sigev_u.__sigev_signo;
1376 break;
1377 case SIGEV_KEVENT:
1378 nsig->sigev_notify_kqueue =
1379 osig->__sigev_u.__sigev_notify_kqueue;
1380 nsig->sigev_value.sival_ptr = osig->sigev_value.sival_ptr;
1381 break;
1382 default:
1383 return (EINVAL);
1384 }
1385 return (0);
1386 }
1387
1388 static int
aiocb_copyin_old_sigevent(struct aiocb * ujob,struct kaiocb * kjob,int type __unused)1389 aiocb_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob,
1390 int type __unused)
1391 {
1392 struct oaiocb *ojob;
1393 struct aiocb *kcb = &kjob->uaiocb;
1394 int error;
1395
1396 bzero(kcb, sizeof(struct aiocb));
1397 error = copyin(ujob, kcb, sizeof(struct oaiocb));
1398 if (error)
1399 return (error);
1400 /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */
1401 ojob = (struct oaiocb *)kcb;
1402 return (convert_old_sigevent(&ojob->aio_sigevent, &kcb->aio_sigevent));
1403 }
1404 #endif
1405
1406 static int
aiocb_copyin(struct aiocb * ujob,struct kaiocb * kjob,int type)1407 aiocb_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type)
1408 {
1409 struct aiocb *kcb = &kjob->uaiocb;
1410 int error;
1411
1412 error = copyin(ujob, kcb, sizeof(struct aiocb));
1413 if (error)
1414 return (error);
1415 if (type == LIO_NOP)
1416 type = kcb->aio_lio_opcode;
1417 if (type & LIO_VECTORED) {
1418 /* malloc a uio and copy in the iovec */
1419 error = copyinuio(__DEVOLATILE(struct iovec*, kcb->aio_iov),
1420 kcb->aio_iovcnt, &kjob->uiop);
1421 }
1422
1423 return (error);
1424 }
1425
1426 static long
aiocb_fetch_status(struct aiocb * ujob)1427 aiocb_fetch_status(struct aiocb *ujob)
1428 {
1429
1430 return (fuword(&ujob->_aiocb_private.status));
1431 }
1432
1433 static long
aiocb_fetch_error(struct aiocb * ujob)1434 aiocb_fetch_error(struct aiocb *ujob)
1435 {
1436
1437 return (fuword(&ujob->_aiocb_private.error));
1438 }
1439
1440 static int
aiocb_store_status(struct aiocb * ujob,long status)1441 aiocb_store_status(struct aiocb *ujob, long status)
1442 {
1443
1444 return (suword(&ujob->_aiocb_private.status, status));
1445 }
1446
1447 static int
aiocb_store_error(struct aiocb * ujob,long error)1448 aiocb_store_error(struct aiocb *ujob, long error)
1449 {
1450
1451 return (suword(&ujob->_aiocb_private.error, error));
1452 }
1453
1454 static int
aiocb_store_kernelinfo(struct aiocb * ujob,long jobref)1455 aiocb_store_kernelinfo(struct aiocb *ujob, long jobref)
1456 {
1457
1458 return (suword(&ujob->_aiocb_private.kernelinfo, jobref));
1459 }
1460
1461 static int
aiocb_store_aiocb(struct aiocb ** ujobp,struct aiocb * ujob)1462 aiocb_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob)
1463 {
1464
1465 return (suword(ujobp, (long)ujob));
1466 }
1467
1468 static struct aiocb_ops aiocb_ops = {
1469 .aio_copyin = aiocb_copyin,
1470 .fetch_status = aiocb_fetch_status,
1471 .fetch_error = aiocb_fetch_error,
1472 .store_status = aiocb_store_status,
1473 .store_error = aiocb_store_error,
1474 .store_kernelinfo = aiocb_store_kernelinfo,
1475 .store_aiocb = aiocb_store_aiocb,
1476 };
1477
1478 #ifdef COMPAT_FREEBSD6
1479 static struct aiocb_ops aiocb_ops_osigevent = {
1480 .aio_copyin = aiocb_copyin_old_sigevent,
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 #endif
1489
1490 /*
1491 * Queue a new AIO request. Choosing either the threaded or direct bio VCHR
1492 * technique is done in this code.
1493 */
1494 int
aio_aqueue(struct thread * td,struct aiocb * ujob,struct aioliojob * lj,int type,struct aiocb_ops * ops)1495 aio_aqueue(struct thread *td, struct aiocb *ujob, struct aioliojob *lj,
1496 int type, struct aiocb_ops *ops)
1497 {
1498 struct proc *p = td->td_proc;
1499 struct file *fp = NULL;
1500 struct kaiocb *job;
1501 struct kaioinfo *ki;
1502 struct kevent kev;
1503 int opcode;
1504 int error;
1505 int fd, kqfd;
1506 int jid;
1507 u_short evflags;
1508
1509 if (p->p_aioinfo == NULL)
1510 aio_init_aioinfo(p);
1511
1512 ki = p->p_aioinfo;
1513
1514 ops->store_status(ujob, -1);
1515 ops->store_error(ujob, 0);
1516 ops->store_kernelinfo(ujob, -1);
1517
1518 if (num_queue_count >= max_queue_count ||
1519 ki->kaio_count >= max_aio_queue_per_proc) {
1520 error = EAGAIN;
1521 goto err1;
1522 }
1523
1524 job = uma_zalloc(aiocb_zone, M_WAITOK | M_ZERO);
1525 knlist_init_mtx(&job->klist, AIO_MTX(ki));
1526
1527 error = ops->aio_copyin(ujob, job, type);
1528 if (error)
1529 goto err2;
1530
1531 if (job->uaiocb.aio_nbytes > IOSIZE_MAX) {
1532 error = EINVAL;
1533 goto err2;
1534 }
1535
1536 if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT &&
1537 job->uaiocb.aio_sigevent.sigev_notify != SIGEV_SIGNAL &&
1538 job->uaiocb.aio_sigevent.sigev_notify != SIGEV_THREAD_ID &&
1539 job->uaiocb.aio_sigevent.sigev_notify != SIGEV_NONE) {
1540 error = EINVAL;
1541 goto err2;
1542 }
1543
1544 if ((job->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL ||
1545 job->uaiocb.aio_sigevent.sigev_notify == SIGEV_THREAD_ID) &&
1546 !_SIG_VALID(job->uaiocb.aio_sigevent.sigev_signo)) {
1547 error = EINVAL;
1548 goto err2;
1549 }
1550
1551 /* Get the opcode. */
1552 if (type == LIO_NOP) {
1553 switch (job->uaiocb.aio_lio_opcode) {
1554 case LIO_WRITE:
1555 case LIO_WRITEV:
1556 case LIO_NOP:
1557 case LIO_READ:
1558 case LIO_READV:
1559 opcode = job->uaiocb.aio_lio_opcode;
1560 break;
1561 default:
1562 error = EINVAL;
1563 goto err2;
1564 }
1565 } else
1566 opcode = job->uaiocb.aio_lio_opcode = type;
1567
1568 ksiginfo_init(&job->ksi);
1569
1570 /* Save userspace address of the job info. */
1571 job->ujob = ujob;
1572
1573 /*
1574 * Validate the opcode and fetch the file object for the specified
1575 * file descriptor.
1576 *
1577 * XXXRW: Moved the opcode validation up here so that we don't
1578 * retrieve a file descriptor without knowing what the capabiltity
1579 * should be.
1580 */
1581 fd = job->uaiocb.aio_fildes;
1582 switch (opcode) {
1583 case LIO_WRITE:
1584 case LIO_WRITEV:
1585 error = fget_write(td, fd, &cap_pwrite_rights, &fp);
1586 break;
1587 case LIO_READ:
1588 case LIO_READV:
1589 error = fget_read(td, fd, &cap_pread_rights, &fp);
1590 break;
1591 case LIO_SYNC:
1592 case LIO_DSYNC:
1593 error = fget(td, fd, &cap_fsync_rights, &fp);
1594 break;
1595 case LIO_MLOCK:
1596 break;
1597 case LIO_NOP:
1598 error = fget(td, fd, &cap_no_rights, &fp);
1599 break;
1600 default:
1601 error = EINVAL;
1602 }
1603 if (error)
1604 goto err3;
1605
1606 if ((opcode & LIO_SYNC) && fp->f_vnode == NULL) {
1607 error = EINVAL;
1608 goto err3;
1609 }
1610
1611 if ((opcode == LIO_READ || opcode == LIO_READV ||
1612 opcode == LIO_WRITE || opcode == LIO_WRITEV) &&
1613 job->uaiocb.aio_offset < 0 &&
1614 (fp->f_vnode == NULL || fp->f_vnode->v_type != VCHR)) {
1615 error = EINVAL;
1616 goto err3;
1617 }
1618
1619 if (fp != NULL && fp->f_ops == &path_fileops) {
1620 error = EBADF;
1621 goto err3;
1622 }
1623
1624 job->fd_file = fp;
1625
1626 mtx_lock(&aio_job_mtx);
1627 jid = jobrefid++;
1628 job->seqno = jobseqno++;
1629 mtx_unlock(&aio_job_mtx);
1630 error = ops->store_kernelinfo(ujob, jid);
1631 if (error) {
1632 error = EINVAL;
1633 goto err3;
1634 }
1635 job->uaiocb._aiocb_private.kernelinfo = (void *)(intptr_t)jid;
1636
1637 if (opcode == LIO_NOP) {
1638 fdrop(fp, td);
1639 MPASS(job->uiop == &job->uio || job->uiop == NULL);
1640 uma_zfree(aiocb_zone, job);
1641 return (0);
1642 }
1643
1644 if (job->uaiocb.aio_sigevent.sigev_notify != SIGEV_KEVENT)
1645 goto no_kqueue;
1646 evflags = job->uaiocb.aio_sigevent.sigev_notify_kevent_flags;
1647 if ((evflags & ~(EV_CLEAR | EV_DISPATCH | EV_ONESHOT)) != 0) {
1648 error = EINVAL;
1649 goto err3;
1650 }
1651 kqfd = job->uaiocb.aio_sigevent.sigev_notify_kqueue;
1652 memset(&kev, 0, sizeof(kev));
1653 kev.ident = (uintptr_t)job->ujob;
1654 kev.filter = EVFILT_AIO;
1655 kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1 | evflags;
1656 kev.data = (intptr_t)job;
1657 kev.udata = job->uaiocb.aio_sigevent.sigev_value.sival_ptr;
1658 error = kqfd_register(kqfd, &kev, td, M_WAITOK);
1659 if (error)
1660 goto err3;
1661
1662 no_kqueue:
1663
1664 ops->store_error(ujob, EINPROGRESS);
1665 job->uaiocb._aiocb_private.error = EINPROGRESS;
1666 job->userproc = p;
1667 job->cred = crhold(td->td_ucred);
1668 job->jobflags = KAIOCB_QUEUEING;
1669 job->lio = lj;
1670
1671 if (opcode & LIO_VECTORED) {
1672 /* Use the uio copied in by aio_copyin */
1673 MPASS(job->uiop != &job->uio && job->uiop != NULL);
1674 } else {
1675 /* Setup the inline uio */
1676 job->iov[0].iov_base = (void *)(uintptr_t)job->uaiocb.aio_buf;
1677 job->iov[0].iov_len = job->uaiocb.aio_nbytes;
1678 job->uio.uio_iov = job->iov;
1679 job->uio.uio_iovcnt = 1;
1680 job->uio.uio_resid = job->uaiocb.aio_nbytes;
1681 job->uio.uio_segflg = UIO_USERSPACE;
1682 job->uiop = &job->uio;
1683 }
1684 switch (opcode & (LIO_READ | LIO_WRITE)) {
1685 case LIO_READ:
1686 job->uiop->uio_rw = UIO_READ;
1687 break;
1688 case LIO_WRITE:
1689 job->uiop->uio_rw = UIO_WRITE;
1690 break;
1691 }
1692 job->uiop->uio_offset = job->uaiocb.aio_offset;
1693 job->uiop->uio_td = td;
1694
1695 if (opcode == LIO_MLOCK) {
1696 aio_schedule(job, aio_process_mlock);
1697 error = 0;
1698 } else if (fp->f_ops->fo_aio_queue == NULL)
1699 error = aio_queue_file(fp, job);
1700 else
1701 error = fo_aio_queue(fp, job);
1702 if (error)
1703 goto err4;
1704
1705 AIO_LOCK(ki);
1706 job->jobflags &= ~KAIOCB_QUEUEING;
1707 TAILQ_INSERT_TAIL(&ki->kaio_all, job, allist);
1708 ki->kaio_count++;
1709 if (lj)
1710 lj->lioj_count++;
1711 atomic_add_int(&num_queue_count, 1);
1712 if (job->jobflags & KAIOCB_FINISHED) {
1713 /*
1714 * The queue callback completed the request synchronously.
1715 * The bulk of the completion is deferred in that case
1716 * until this point.
1717 */
1718 aio_bio_done_notify(p, job);
1719 } else
1720 TAILQ_INSERT_TAIL(&ki->kaio_jobqueue, job, plist);
1721 AIO_UNLOCK(ki);
1722 return (0);
1723
1724 err4:
1725 crfree(job->cred);
1726 err3:
1727 if (fp)
1728 fdrop(fp, td);
1729 knlist_delete(&job->klist, curthread, 0);
1730 err2:
1731 if (job->uiop != &job->uio)
1732 free(job->uiop, M_IOV);
1733 uma_zfree(aiocb_zone, job);
1734 err1:
1735 ops->store_error(ujob, error);
1736 return (error);
1737 }
1738
1739 static void
aio_cancel_daemon_job(struct kaiocb * job)1740 aio_cancel_daemon_job(struct kaiocb *job)
1741 {
1742
1743 mtx_lock(&aio_job_mtx);
1744 if (!aio_cancel_cleared(job))
1745 TAILQ_REMOVE(&aio_jobs, job, list);
1746 mtx_unlock(&aio_job_mtx);
1747 aio_cancel(job);
1748 }
1749
1750 void
aio_schedule(struct kaiocb * job,aio_handle_fn_t * func)1751 aio_schedule(struct kaiocb *job, aio_handle_fn_t *func)
1752 {
1753
1754 mtx_lock(&aio_job_mtx);
1755 if (!aio_set_cancel_function(job, aio_cancel_daemon_job)) {
1756 mtx_unlock(&aio_job_mtx);
1757 aio_cancel(job);
1758 return;
1759 }
1760 job->handle_fn = func;
1761 TAILQ_INSERT_TAIL(&aio_jobs, job, list);
1762 aio_kick_nowait(job->userproc);
1763 mtx_unlock(&aio_job_mtx);
1764 }
1765
1766 static void
aio_cancel_sync(struct kaiocb * job)1767 aio_cancel_sync(struct kaiocb *job)
1768 {
1769 struct kaioinfo *ki;
1770
1771 ki = job->userproc->p_aioinfo;
1772 AIO_LOCK(ki);
1773 if (!aio_cancel_cleared(job))
1774 TAILQ_REMOVE(&ki->kaio_syncqueue, job, list);
1775 AIO_UNLOCK(ki);
1776 aio_cancel(job);
1777 }
1778
1779 int
aio_queue_file(struct file * fp,struct kaiocb * job)1780 aio_queue_file(struct file *fp, struct kaiocb *job)
1781 {
1782 struct kaioinfo *ki;
1783 struct kaiocb *job2;
1784 struct vnode *vp;
1785 struct mount *mp;
1786 int error;
1787 bool safe;
1788
1789 ki = job->userproc->p_aioinfo;
1790 error = aio_qbio(job->userproc, job);
1791 if (error >= 0)
1792 return (error);
1793 safe = false;
1794 if (fp->f_type == DTYPE_VNODE) {
1795 vp = fp->f_vnode;
1796 if (vp->v_type == VREG || vp->v_type == VDIR) {
1797 mp = fp->f_vnode->v_mount;
1798 if (mp == NULL || (mp->mnt_flag & MNT_LOCAL) != 0)
1799 safe = true;
1800 }
1801 }
1802 if (!(safe || enable_aio_unsafe)) {
1803 counted_warning(&unsafe_warningcnt,
1804 "is attempting to use unsafe AIO requests");
1805 return (EOPNOTSUPP);
1806 }
1807
1808 if (job->uaiocb.aio_lio_opcode & (LIO_WRITE | LIO_READ)) {
1809 aio_schedule(job, aio_process_rw);
1810 error = 0;
1811 } else if (job->uaiocb.aio_lio_opcode & LIO_SYNC) {
1812 AIO_LOCK(ki);
1813 TAILQ_FOREACH(job2, &ki->kaio_jobqueue, plist) {
1814 if (job2->fd_file == job->fd_file &&
1815 ((job2->uaiocb.aio_lio_opcode & LIO_SYNC) == 0) &&
1816 job2->seqno < job->seqno) {
1817 job2->jobflags |= KAIOCB_CHECKSYNC;
1818 job->pending++;
1819 }
1820 }
1821 if (job->pending != 0) {
1822 if (!aio_set_cancel_function_locked(job,
1823 aio_cancel_sync)) {
1824 AIO_UNLOCK(ki);
1825 aio_cancel(job);
1826 return (0);
1827 }
1828 TAILQ_INSERT_TAIL(&ki->kaio_syncqueue, job, list);
1829 AIO_UNLOCK(ki);
1830 return (0);
1831 }
1832 AIO_UNLOCK(ki);
1833 aio_schedule(job, aio_process_sync);
1834 error = 0;
1835 } else {
1836 error = EINVAL;
1837 }
1838 return (error);
1839 }
1840
1841 static void
aio_kick_nowait(struct proc * userp)1842 aio_kick_nowait(struct proc *userp)
1843 {
1844 struct kaioinfo *ki = userp->p_aioinfo;
1845 struct aioproc *aiop;
1846
1847 mtx_assert(&aio_job_mtx, MA_OWNED);
1848 if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) {
1849 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1850 aiop->aioprocflags &= ~AIOP_FREE;
1851 wakeup(aiop->aioproc);
1852 } else if (num_aio_resv_start + num_aio_procs < max_aio_procs &&
1853 ki->kaio_active_count + num_aio_resv_start < max_aio_per_proc) {
1854 taskqueue_enqueue(taskqueue_aiod_kick, &ki->kaio_task);
1855 }
1856 }
1857
1858 static int
aio_kick(struct proc * userp)1859 aio_kick(struct proc *userp)
1860 {
1861 struct kaioinfo *ki = userp->p_aioinfo;
1862 struct aioproc *aiop;
1863 int error, ret = 0;
1864
1865 mtx_assert(&aio_job_mtx, MA_OWNED);
1866 retryproc:
1867 if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) {
1868 TAILQ_REMOVE(&aio_freeproc, aiop, list);
1869 aiop->aioprocflags &= ~AIOP_FREE;
1870 wakeup(aiop->aioproc);
1871 } else if (num_aio_resv_start + num_aio_procs < max_aio_procs &&
1872 ki->kaio_active_count + num_aio_resv_start < max_aio_per_proc) {
1873 num_aio_resv_start++;
1874 mtx_unlock(&aio_job_mtx);
1875 error = aio_newproc(&num_aio_resv_start);
1876 mtx_lock(&aio_job_mtx);
1877 if (error) {
1878 num_aio_resv_start--;
1879 goto retryproc;
1880 }
1881 } else {
1882 ret = -1;
1883 }
1884 return (ret);
1885 }
1886
1887 static void
aio_kick_helper(void * context,int pending)1888 aio_kick_helper(void *context, int pending)
1889 {
1890 struct proc *userp = context;
1891
1892 mtx_lock(&aio_job_mtx);
1893 while (--pending >= 0) {
1894 if (aio_kick(userp))
1895 break;
1896 }
1897 mtx_unlock(&aio_job_mtx);
1898 }
1899
1900 /*
1901 * Support the aio_return system call, as a side-effect, kernel resources are
1902 * released.
1903 */
1904 static int
kern_aio_return(struct thread * td,struct aiocb * ujob,struct aiocb_ops * ops)1905 kern_aio_return(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops)
1906 {
1907 struct proc *p = td->td_proc;
1908 struct kaiocb *job;
1909 struct kaioinfo *ki;
1910 long status, error;
1911
1912 ki = p->p_aioinfo;
1913 if (ki == NULL)
1914 return (EINVAL);
1915 AIO_LOCK(ki);
1916 TAILQ_FOREACH(job, &ki->kaio_done, plist) {
1917 if (job->ujob == ujob)
1918 break;
1919 }
1920 if (job != NULL) {
1921 MPASS(job->jobflags & KAIOCB_FINISHED);
1922 status = job->uaiocb._aiocb_private.status;
1923 error = job->uaiocb._aiocb_private.error;
1924 td->td_retval[0] = status;
1925 td->td_ru.ru_oublock += job->outblock;
1926 td->td_ru.ru_inblock += job->inblock;
1927 td->td_ru.ru_msgsnd += job->msgsnd;
1928 td->td_ru.ru_msgrcv += job->msgrcv;
1929 aio_free_entry(job);
1930 AIO_UNLOCK(ki);
1931 ops->store_error(ujob, error);
1932 ops->store_status(ujob, status);
1933 } else {
1934 error = EINVAL;
1935 AIO_UNLOCK(ki);
1936 }
1937 return (error);
1938 }
1939
1940 int
sys_aio_return(struct thread * td,struct aio_return_args * uap)1941 sys_aio_return(struct thread *td, struct aio_return_args *uap)
1942 {
1943
1944 return (kern_aio_return(td, uap->aiocbp, &aiocb_ops));
1945 }
1946
1947 /*
1948 * Allow a process to wakeup when any of the I/O requests are completed.
1949 */
1950 static int
kern_aio_suspend(struct thread * td,int njoblist,struct aiocb ** ujoblist,struct timespec * ts)1951 kern_aio_suspend(struct thread *td, int njoblist, struct aiocb **ujoblist,
1952 struct timespec *ts)
1953 {
1954 struct proc *p = td->td_proc;
1955 struct timeval atv;
1956 struct kaioinfo *ki;
1957 struct kaiocb *firstjob, *job;
1958 int error, i, timo;
1959
1960 timo = 0;
1961 if (ts) {
1962 if (ts->tv_nsec < 0 || ts->tv_nsec >= 1000000000)
1963 return (EINVAL);
1964
1965 TIMESPEC_TO_TIMEVAL(&atv, ts);
1966 if (itimerfix(&atv))
1967 return (EINVAL);
1968 timo = tvtohz(&atv);
1969 }
1970
1971 ki = p->p_aioinfo;
1972 if (ki == NULL)
1973 return (EAGAIN);
1974
1975 if (njoblist == 0)
1976 return (0);
1977
1978 AIO_LOCK(ki);
1979 for (;;) {
1980 firstjob = NULL;
1981 error = 0;
1982 TAILQ_FOREACH(job, &ki->kaio_all, allist) {
1983 for (i = 0; i < njoblist; i++) {
1984 if (job->ujob == ujoblist[i]) {
1985 if (firstjob == NULL)
1986 firstjob = job;
1987 if (job->jobflags & KAIOCB_FINISHED)
1988 goto RETURN;
1989 }
1990 }
1991 }
1992 /* All tasks were finished. */
1993 if (firstjob == NULL)
1994 break;
1995
1996 ki->kaio_flags |= KAIO_WAKEUP;
1997 error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH,
1998 "aiospn", timo);
1999 if (error == ERESTART)
2000 error = EINTR;
2001 if (error)
2002 break;
2003 }
2004 RETURN:
2005 AIO_UNLOCK(ki);
2006 return (error);
2007 }
2008
2009 int
sys_aio_suspend(struct thread * td,struct aio_suspend_args * uap)2010 sys_aio_suspend(struct thread *td, struct aio_suspend_args *uap)
2011 {
2012 struct timespec ts, *tsp;
2013 struct aiocb **ujoblist;
2014 int error;
2015
2016 if (uap->nent < 0 || uap->nent > max_aio_queue_per_proc)
2017 return (EINVAL);
2018
2019 if (uap->timeout) {
2020 /* Get timespec struct. */
2021 if ((error = copyin(uap->timeout, &ts, sizeof(ts))) != 0)
2022 return (error);
2023 tsp = &ts;
2024 } else
2025 tsp = NULL;
2026
2027 ujoblist = malloc(uap->nent * sizeof(ujoblist[0]), M_AIO, M_WAITOK);
2028 error = copyin(uap->aiocbp, ujoblist, uap->nent * sizeof(ujoblist[0]));
2029 if (error == 0)
2030 error = kern_aio_suspend(td, uap->nent, ujoblist, tsp);
2031 free(ujoblist, M_AIO);
2032 return (error);
2033 }
2034
2035 /*
2036 * aio_cancel cancels any non-bio aio operations not currently in progress.
2037 */
2038 int
sys_aio_cancel(struct thread * td,struct aio_cancel_args * uap)2039 sys_aio_cancel(struct thread *td, struct aio_cancel_args *uap)
2040 {
2041 struct proc *p = td->td_proc;
2042 struct kaioinfo *ki;
2043 struct kaiocb *job, *jobn;
2044 struct file *fp;
2045 int error;
2046 int cancelled = 0;
2047 int notcancelled = 0;
2048 struct vnode *vp;
2049
2050 /* Lookup file object. */
2051 error = fget(td, uap->fd, &cap_no_rights, &fp);
2052 if (error)
2053 return (error);
2054
2055 ki = p->p_aioinfo;
2056 if (ki == NULL)
2057 goto done;
2058
2059 if (fp->f_type == DTYPE_VNODE) {
2060 vp = fp->f_vnode;
2061 if (vn_isdisk(vp)) {
2062 fdrop(fp, td);
2063 td->td_retval[0] = AIO_NOTCANCELED;
2064 return (0);
2065 }
2066 }
2067
2068 AIO_LOCK(ki);
2069 TAILQ_FOREACH_SAFE(job, &ki->kaio_jobqueue, plist, jobn) {
2070 if ((uap->fd == job->uaiocb.aio_fildes) &&
2071 ((uap->aiocbp == NULL) ||
2072 (uap->aiocbp == job->ujob))) {
2073 if (aio_cancel_job(p, ki, job)) {
2074 cancelled++;
2075 } else {
2076 notcancelled++;
2077 }
2078 if (uap->aiocbp != NULL)
2079 break;
2080 }
2081 }
2082 AIO_UNLOCK(ki);
2083
2084 done:
2085 fdrop(fp, td);
2086
2087 if (uap->aiocbp != NULL) {
2088 if (cancelled) {
2089 td->td_retval[0] = AIO_CANCELED;
2090 return (0);
2091 }
2092 }
2093
2094 if (notcancelled) {
2095 td->td_retval[0] = AIO_NOTCANCELED;
2096 return (0);
2097 }
2098
2099 if (cancelled) {
2100 td->td_retval[0] = AIO_CANCELED;
2101 return (0);
2102 }
2103
2104 td->td_retval[0] = AIO_ALLDONE;
2105
2106 return (0);
2107 }
2108
2109 /*
2110 * aio_error is implemented in the kernel level for compatibility purposes
2111 * only. For a user mode async implementation, it would be best to do it in
2112 * a userland subroutine.
2113 */
2114 static int
kern_aio_error(struct thread * td,struct aiocb * ujob,struct aiocb_ops * ops)2115 kern_aio_error(struct thread *td, struct aiocb *ujob, struct aiocb_ops *ops)
2116 {
2117 struct proc *p = td->td_proc;
2118 struct kaiocb *job;
2119 struct kaioinfo *ki;
2120 int status;
2121
2122 ki = p->p_aioinfo;
2123 if (ki == NULL) {
2124 td->td_retval[0] = EINVAL;
2125 return (0);
2126 }
2127
2128 AIO_LOCK(ki);
2129 TAILQ_FOREACH(job, &ki->kaio_all, allist) {
2130 if (job->ujob == ujob) {
2131 if (job->jobflags & KAIOCB_FINISHED)
2132 td->td_retval[0] =
2133 job->uaiocb._aiocb_private.error;
2134 else
2135 td->td_retval[0] = EINPROGRESS;
2136 AIO_UNLOCK(ki);
2137 return (0);
2138 }
2139 }
2140 AIO_UNLOCK(ki);
2141
2142 /*
2143 * Hack for failure of aio_aqueue.
2144 */
2145 status = ops->fetch_status(ujob);
2146 if (status == -1) {
2147 td->td_retval[0] = ops->fetch_error(ujob);
2148 return (0);
2149 }
2150
2151 td->td_retval[0] = EINVAL;
2152 return (0);
2153 }
2154
2155 int
sys_aio_error(struct thread * td,struct aio_error_args * uap)2156 sys_aio_error(struct thread *td, struct aio_error_args *uap)
2157 {
2158
2159 return (kern_aio_error(td, uap->aiocbp, &aiocb_ops));
2160 }
2161
2162 /* syscall - asynchronous read from a file (REALTIME) */
2163 #ifdef COMPAT_FREEBSD6
2164 int
freebsd6_aio_read(struct thread * td,struct freebsd6_aio_read_args * uap)2165 freebsd6_aio_read(struct thread *td, struct freebsd6_aio_read_args *uap)
2166 {
2167
2168 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ,
2169 &aiocb_ops_osigevent));
2170 }
2171 #endif
2172
2173 int
sys_aio_read(struct thread * td,struct aio_read_args * uap)2174 sys_aio_read(struct thread *td, struct aio_read_args *uap)
2175 {
2176
2177 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READ, &aiocb_ops));
2178 }
2179
2180 int
sys_aio_readv(struct thread * td,struct aio_readv_args * uap)2181 sys_aio_readv(struct thread *td, struct aio_readv_args *uap)
2182 {
2183
2184 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_READV, &aiocb_ops));
2185 }
2186
2187 /* syscall - asynchronous write to a file (REALTIME) */
2188 #ifdef COMPAT_FREEBSD6
2189 int
freebsd6_aio_write(struct thread * td,struct freebsd6_aio_write_args * uap)2190 freebsd6_aio_write(struct thread *td, struct freebsd6_aio_write_args *uap)
2191 {
2192
2193 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE,
2194 &aiocb_ops_osigevent));
2195 }
2196 #endif
2197
2198 int
sys_aio_write(struct thread * td,struct aio_write_args * uap)2199 sys_aio_write(struct thread *td, struct aio_write_args *uap)
2200 {
2201
2202 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITE, &aiocb_ops));
2203 }
2204
2205 int
sys_aio_writev(struct thread * td,struct aio_writev_args * uap)2206 sys_aio_writev(struct thread *td, struct aio_writev_args *uap)
2207 {
2208
2209 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_WRITEV, &aiocb_ops));
2210 }
2211
2212 int
sys_aio_mlock(struct thread * td,struct aio_mlock_args * uap)2213 sys_aio_mlock(struct thread *td, struct aio_mlock_args *uap)
2214 {
2215
2216 return (aio_aqueue(td, uap->aiocbp, NULL, LIO_MLOCK, &aiocb_ops));
2217 }
2218
2219 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)2220 kern_lio_listio(struct thread *td, int mode, struct aiocb * const *uacb_list,
2221 struct aiocb **acb_list, int nent, struct sigevent *sig,
2222 struct aiocb_ops *ops)
2223 {
2224 struct proc *p = td->td_proc;
2225 struct aiocb *job;
2226 struct kaioinfo *ki;
2227 struct aioliojob *lj;
2228 struct kevent kev;
2229 int error;
2230 int nagain, nerror;
2231 int i;
2232
2233 if ((mode != LIO_NOWAIT) && (mode != LIO_WAIT))
2234 return (EINVAL);
2235
2236 if (nent < 0 || nent > max_aio_queue_per_proc)
2237 return (EINVAL);
2238
2239 if (p->p_aioinfo == NULL)
2240 aio_init_aioinfo(p);
2241
2242 ki = p->p_aioinfo;
2243
2244 lj = uma_zalloc(aiolio_zone, M_WAITOK);
2245 lj->lioj_flags = 0;
2246 lj->lioj_count = 0;
2247 lj->lioj_finished_count = 0;
2248 lj->lioj_signal.sigev_notify = SIGEV_NONE;
2249 knlist_init_mtx(&lj->klist, AIO_MTX(ki));
2250 ksiginfo_init(&lj->lioj_ksi);
2251
2252 /*
2253 * Setup signal.
2254 */
2255 if (sig && (mode == LIO_NOWAIT)) {
2256 bcopy(sig, &lj->lioj_signal, sizeof(lj->lioj_signal));
2257 if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) {
2258 /* Assume only new style KEVENT */
2259 memset(&kev, 0, sizeof(kev));
2260 kev.filter = EVFILT_LIO;
2261 kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1;
2262 kev.ident = (uintptr_t)uacb_list; /* something unique */
2263 kev.data = (intptr_t)lj;
2264 /* pass user defined sigval data */
2265 kev.udata = lj->lioj_signal.sigev_value.sival_ptr;
2266 error = kqfd_register(
2267 lj->lioj_signal.sigev_notify_kqueue, &kev, td,
2268 M_WAITOK);
2269 if (error) {
2270 uma_zfree(aiolio_zone, lj);
2271 return (error);
2272 }
2273 } else if (lj->lioj_signal.sigev_notify == SIGEV_NONE) {
2274 ;
2275 } else if (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL ||
2276 lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID) {
2277 if (!_SIG_VALID(lj->lioj_signal.sigev_signo)) {
2278 uma_zfree(aiolio_zone, lj);
2279 return EINVAL;
2280 }
2281 lj->lioj_flags |= LIOJ_SIGNAL;
2282 } else {
2283 uma_zfree(aiolio_zone, lj);
2284 return EINVAL;
2285 }
2286 }
2287
2288 AIO_LOCK(ki);
2289 TAILQ_INSERT_TAIL(&ki->kaio_liojoblist, lj, lioj_list);
2290 /*
2291 * Add extra aiocb count to avoid the lio to be freed
2292 * by other threads doing aio_waitcomplete or aio_return,
2293 * and prevent event from being sent until we have queued
2294 * all tasks.
2295 */
2296 lj->lioj_count = 1;
2297 AIO_UNLOCK(ki);
2298
2299 /*
2300 * Get pointers to the list of I/O requests.
2301 */
2302 nagain = 0;
2303 nerror = 0;
2304 for (i = 0; i < nent; i++) {
2305 job = acb_list[i];
2306 if (job != NULL) {
2307 error = aio_aqueue(td, job, lj, LIO_NOP, ops);
2308 if (error == EAGAIN)
2309 nagain++;
2310 else if (error != 0)
2311 nerror++;
2312 }
2313 }
2314
2315 error = 0;
2316 AIO_LOCK(ki);
2317 if (mode == LIO_WAIT) {
2318 while (lj->lioj_count - 1 != lj->lioj_finished_count) {
2319 ki->kaio_flags |= KAIO_WAKEUP;
2320 error = msleep(&p->p_aioinfo, AIO_MTX(ki),
2321 PRIBIO | PCATCH, "aiospn", 0);
2322 if (error == ERESTART)
2323 error = EINTR;
2324 if (error)
2325 break;
2326 }
2327 } else {
2328 if (lj->lioj_count - 1 == lj->lioj_finished_count) {
2329 if (lj->lioj_signal.sigev_notify == SIGEV_KEVENT) {
2330 lj->lioj_flags |= LIOJ_KEVENT_POSTED;
2331 KNOTE_LOCKED(&lj->klist, 1);
2332 }
2333 if ((lj->lioj_flags & (LIOJ_SIGNAL |
2334 LIOJ_SIGNAL_POSTED)) == LIOJ_SIGNAL &&
2335 (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL ||
2336 lj->lioj_signal.sigev_notify == SIGEV_THREAD_ID)) {
2337 aio_sendsig(p, &lj->lioj_signal, &lj->lioj_ksi,
2338 lj->lioj_count != 1);
2339 lj->lioj_flags |= LIOJ_SIGNAL_POSTED;
2340 }
2341 }
2342 }
2343 lj->lioj_count--;
2344 if (lj->lioj_count == 0) {
2345 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list);
2346 knlist_delete(&lj->klist, curthread, 1);
2347 PROC_LOCK(p);
2348 sigqueue_take(&lj->lioj_ksi);
2349 PROC_UNLOCK(p);
2350 AIO_UNLOCK(ki);
2351 uma_zfree(aiolio_zone, lj);
2352 } else
2353 AIO_UNLOCK(ki);
2354
2355 if (nerror)
2356 return (EIO);
2357 else if (nagain)
2358 return (EAGAIN);
2359 else
2360 return (error);
2361 }
2362
2363 /* syscall - list directed I/O (REALTIME) */
2364 #ifdef COMPAT_FREEBSD6
2365 int
freebsd6_lio_listio(struct thread * td,struct freebsd6_lio_listio_args * uap)2366 freebsd6_lio_listio(struct thread *td, struct freebsd6_lio_listio_args *uap)
2367 {
2368 struct aiocb **acb_list;
2369 struct sigevent *sigp, sig;
2370 struct osigevent osig;
2371 int error, nent;
2372
2373 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
2374 return (EINVAL);
2375
2376 nent = uap->nent;
2377 if (nent < 0 || nent > max_aio_queue_per_proc)
2378 return (EINVAL);
2379
2380 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
2381 error = copyin(uap->sig, &osig, sizeof(osig));
2382 if (error)
2383 return (error);
2384 error = convert_old_sigevent(&osig, &sig);
2385 if (error)
2386 return (error);
2387 sigp = &sig;
2388 } else
2389 sigp = NULL;
2390
2391 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
2392 error = copyin(uap->acb_list, acb_list, nent * sizeof(acb_list[0]));
2393 if (error == 0)
2394 error = kern_lio_listio(td, uap->mode,
2395 (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp,
2396 &aiocb_ops_osigevent);
2397 free(acb_list, M_LIO);
2398 return (error);
2399 }
2400 #endif
2401
2402 /* syscall - list directed I/O (REALTIME) */
2403 int
sys_lio_listio(struct thread * td,struct lio_listio_args * uap)2404 sys_lio_listio(struct thread *td, struct lio_listio_args *uap)
2405 {
2406 struct aiocb **acb_list;
2407 struct sigevent *sigp, sig;
2408 int error, nent;
2409
2410 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
2411 return (EINVAL);
2412
2413 nent = uap->nent;
2414 if (nent < 0 || nent > max_aio_queue_per_proc)
2415 return (EINVAL);
2416
2417 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
2418 error = copyin(uap->sig, &sig, sizeof(sig));
2419 if (error)
2420 return (error);
2421 sigp = &sig;
2422 } else
2423 sigp = NULL;
2424
2425 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
2426 error = copyin(uap->acb_list, acb_list, nent * sizeof(acb_list[0]));
2427 if (error == 0)
2428 error = kern_lio_listio(td, uap->mode, uap->acb_list, acb_list,
2429 nent, sigp, &aiocb_ops);
2430 free(acb_list, M_LIO);
2431 return (error);
2432 }
2433
2434 static void
aio_biocleanup(struct bio * bp)2435 aio_biocleanup(struct bio *bp)
2436 {
2437 struct kaiocb *job = (struct kaiocb *)bp->bio_caller1;
2438 struct kaioinfo *ki;
2439 struct buf *pbuf = (struct buf *)bp->bio_caller2;
2440
2441 /* Release mapping into kernel space. */
2442 if (pbuf != NULL) {
2443 MPASS(pbuf->b_npages <= atop(maxphys) + 1);
2444 pmap_qremove((vm_offset_t)pbuf->b_data, pbuf->b_npages);
2445 vm_page_unhold_pages(pbuf->b_pages, pbuf->b_npages);
2446 uma_zfree(pbuf_zone, pbuf);
2447 atomic_subtract_int(&num_buf_aio, 1);
2448 ki = job->userproc->p_aioinfo;
2449 AIO_LOCK(ki);
2450 ki->kaio_buffer_count--;
2451 AIO_UNLOCK(ki);
2452 } else {
2453 MPASS(bp->bio_ma_n <= atop(maxphys) + 1);
2454 vm_page_unhold_pages(bp->bio_ma, bp->bio_ma_n);
2455 free(bp->bio_ma, M_TEMP);
2456 atomic_subtract_int(&num_unmapped_aio, 1);
2457 }
2458 g_destroy_bio(bp);
2459 }
2460
2461 static void
aio_biowakeup(struct bio * bp)2462 aio_biowakeup(struct bio *bp)
2463 {
2464 struct kaiocb *job = (struct kaiocb *)bp->bio_caller1;
2465 size_t nbytes;
2466 long bcount = bp->bio_bcount;
2467 long resid = bp->bio_resid;
2468 int opcode, nblks;
2469 int bio_error = bp->bio_error;
2470 uint16_t flags = bp->bio_flags;
2471
2472 opcode = job->uaiocb.aio_lio_opcode;
2473
2474 aio_biocleanup(bp);
2475
2476 nbytes = bcount - resid;
2477 atomic_add_acq_long(&job->nbytes, nbytes);
2478 nblks = btodb(nbytes);
2479
2480 /*
2481 * If multiple bios experienced an error, the job will reflect the
2482 * error of whichever failed bio completed last.
2483 */
2484 if (flags & BIO_ERROR)
2485 atomic_store_int(&job->error, bio_error);
2486 if (opcode & LIO_WRITE)
2487 atomic_add_int(&job->outblock, nblks);
2488 else
2489 atomic_add_int(&job->inblock, nblks);
2490
2491 if (refcount_release(&job->nbio)) {
2492 bio_error = atomic_load_int(&job->error);
2493 if (bio_error != 0)
2494 aio_complete(job, -1, bio_error);
2495 else
2496 aio_complete(job, atomic_load_long(&job->nbytes), 0);
2497 }
2498 }
2499
2500 /* syscall - wait for the next completion of an aio request */
2501 static int
kern_aio_waitcomplete(struct thread * td,struct aiocb ** ujobp,struct timespec * ts,struct aiocb_ops * ops)2502 kern_aio_waitcomplete(struct thread *td, struct aiocb **ujobp,
2503 struct timespec *ts, struct aiocb_ops *ops)
2504 {
2505 struct proc *p = td->td_proc;
2506 struct timeval atv;
2507 struct kaioinfo *ki;
2508 struct kaiocb *job;
2509 struct aiocb *ujob;
2510 long error, status;
2511 int timo;
2512
2513 ops->store_aiocb(ujobp, NULL);
2514
2515 if (ts == NULL) {
2516 timo = 0;
2517 } else if (ts->tv_sec == 0 && ts->tv_nsec == 0) {
2518 timo = -1;
2519 } else {
2520 if ((ts->tv_nsec < 0) || (ts->tv_nsec >= 1000000000))
2521 return (EINVAL);
2522
2523 TIMESPEC_TO_TIMEVAL(&atv, ts);
2524 if (itimerfix(&atv))
2525 return (EINVAL);
2526 timo = tvtohz(&atv);
2527 }
2528
2529 if (p->p_aioinfo == NULL)
2530 aio_init_aioinfo(p);
2531 ki = p->p_aioinfo;
2532
2533 error = 0;
2534 job = NULL;
2535 AIO_LOCK(ki);
2536 while ((job = TAILQ_FIRST(&ki->kaio_done)) == NULL) {
2537 if (timo == -1) {
2538 error = EWOULDBLOCK;
2539 break;
2540 }
2541 ki->kaio_flags |= KAIO_WAKEUP;
2542 error = msleep(&p->p_aioinfo, AIO_MTX(ki), PRIBIO | PCATCH,
2543 "aiowc", timo);
2544 if (timo && error == ERESTART)
2545 error = EINTR;
2546 if (error)
2547 break;
2548 }
2549
2550 if (job != NULL) {
2551 MPASS(job->jobflags & KAIOCB_FINISHED);
2552 ujob = job->ujob;
2553 status = job->uaiocb._aiocb_private.status;
2554 error = job->uaiocb._aiocb_private.error;
2555 td->td_retval[0] = status;
2556 td->td_ru.ru_oublock += job->outblock;
2557 td->td_ru.ru_inblock += job->inblock;
2558 td->td_ru.ru_msgsnd += job->msgsnd;
2559 td->td_ru.ru_msgrcv += job->msgrcv;
2560 aio_free_entry(job);
2561 AIO_UNLOCK(ki);
2562 ops->store_aiocb(ujobp, ujob);
2563 ops->store_error(ujob, error);
2564 ops->store_status(ujob, status);
2565 } else
2566 AIO_UNLOCK(ki);
2567
2568 return (error);
2569 }
2570
2571 int
sys_aio_waitcomplete(struct thread * td,struct aio_waitcomplete_args * uap)2572 sys_aio_waitcomplete(struct thread *td, struct aio_waitcomplete_args *uap)
2573 {
2574 struct timespec ts, *tsp;
2575 int error;
2576
2577 if (uap->timeout) {
2578 /* Get timespec struct. */
2579 error = copyin(uap->timeout, &ts, sizeof(ts));
2580 if (error)
2581 return (error);
2582 tsp = &ts;
2583 } else
2584 tsp = NULL;
2585
2586 return (kern_aio_waitcomplete(td, uap->aiocbp, tsp, &aiocb_ops));
2587 }
2588
2589 static int
kern_aio_fsync(struct thread * td,int op,struct aiocb * ujob,struct aiocb_ops * ops)2590 kern_aio_fsync(struct thread *td, int op, struct aiocb *ujob,
2591 struct aiocb_ops *ops)
2592 {
2593 int listop;
2594
2595 switch (op) {
2596 case O_SYNC:
2597 listop = LIO_SYNC;
2598 break;
2599 case O_DSYNC:
2600 listop = LIO_DSYNC;
2601 break;
2602 default:
2603 return (EINVAL);
2604 }
2605
2606 return (aio_aqueue(td, ujob, NULL, listop, ops));
2607 }
2608
2609 int
sys_aio_fsync(struct thread * td,struct aio_fsync_args * uap)2610 sys_aio_fsync(struct thread *td, struct aio_fsync_args *uap)
2611 {
2612
2613 return (kern_aio_fsync(td, uap->op, uap->aiocbp, &aiocb_ops));
2614 }
2615
2616 /* kqueue attach function */
2617 static int
filt_aioattach(struct knote * kn)2618 filt_aioattach(struct knote *kn)
2619 {
2620 struct kaiocb *job;
2621
2622 job = (struct kaiocb *)(uintptr_t)kn->kn_sdata;
2623
2624 /*
2625 * The job pointer must be validated before using it, so
2626 * registration is restricted to the kernel; the user cannot
2627 * set EV_FLAG1.
2628 */
2629 if ((kn->kn_flags & EV_FLAG1) == 0)
2630 return (EPERM);
2631 kn->kn_ptr.p_aio = job;
2632 kn->kn_flags &= ~EV_FLAG1;
2633
2634 knlist_add(&job->klist, kn, 0);
2635
2636 return (0);
2637 }
2638
2639 /* kqueue detach function */
2640 static void
filt_aiodetach(struct knote * kn)2641 filt_aiodetach(struct knote *kn)
2642 {
2643 struct knlist *knl;
2644
2645 knl = &kn->kn_ptr.p_aio->klist;
2646 knl->kl_lock(knl->kl_lockarg);
2647 if (!knlist_empty(knl))
2648 knlist_remove(knl, kn, 1);
2649 knl->kl_unlock(knl->kl_lockarg);
2650 }
2651
2652 /* kqueue filter function */
2653 /*ARGSUSED*/
2654 static int
filt_aio(struct knote * kn,long hint)2655 filt_aio(struct knote *kn, long hint)
2656 {
2657 struct kaiocb *job = kn->kn_ptr.p_aio;
2658
2659 kn->kn_data = job->uaiocb._aiocb_private.error;
2660 if (!(job->jobflags & KAIOCB_FINISHED))
2661 return (0);
2662 kn->kn_flags |= EV_EOF;
2663 return (1);
2664 }
2665
2666 /* kqueue attach function */
2667 static int
filt_lioattach(struct knote * kn)2668 filt_lioattach(struct knote *kn)
2669 {
2670 struct aioliojob *lj;
2671
2672 lj = (struct aioliojob *)(uintptr_t)kn->kn_sdata;
2673
2674 /*
2675 * The aioliojob pointer must be validated before using it, so
2676 * registration is restricted to the kernel; the user cannot
2677 * set EV_FLAG1.
2678 */
2679 if ((kn->kn_flags & EV_FLAG1) == 0)
2680 return (EPERM);
2681 kn->kn_ptr.p_lio = lj;
2682 kn->kn_flags &= ~EV_FLAG1;
2683
2684 knlist_add(&lj->klist, kn, 0);
2685
2686 return (0);
2687 }
2688
2689 /* kqueue detach function */
2690 static void
filt_liodetach(struct knote * kn)2691 filt_liodetach(struct knote *kn)
2692 {
2693 struct knlist *knl;
2694
2695 knl = &kn->kn_ptr.p_lio->klist;
2696 knl->kl_lock(knl->kl_lockarg);
2697 if (!knlist_empty(knl))
2698 knlist_remove(knl, kn, 1);
2699 knl->kl_unlock(knl->kl_lockarg);
2700 }
2701
2702 /* kqueue filter function */
2703 /*ARGSUSED*/
2704 static int
filt_lio(struct knote * kn,long hint)2705 filt_lio(struct knote *kn, long hint)
2706 {
2707 struct aioliojob * lj = kn->kn_ptr.p_lio;
2708
2709 return (lj->lioj_flags & LIOJ_KEVENT_POSTED);
2710 }
2711
2712 #ifdef COMPAT_FREEBSD32
2713 #include <sys/mount.h>
2714 #include <sys/socket.h>
2715 #include <sys/sysent.h>
2716 #include <compat/freebsd32/freebsd32.h>
2717 #include <compat/freebsd32/freebsd32_proto.h>
2718 #include <compat/freebsd32/freebsd32_signal.h>
2719 #include <compat/freebsd32/freebsd32_syscall.h>
2720 #include <compat/freebsd32/freebsd32_util.h>
2721
2722 struct __aiocb_private32 {
2723 int32_t status;
2724 int32_t error;
2725 uint32_t kernelinfo;
2726 };
2727
2728 #ifdef COMPAT_FREEBSD6
2729 typedef struct oaiocb32 {
2730 int aio_fildes; /* File descriptor */
2731 uint64_t aio_offset __packed; /* File offset for I/O */
2732 uint32_t aio_buf; /* I/O buffer in process space */
2733 uint32_t aio_nbytes; /* Number of bytes for I/O */
2734 struct osigevent32 aio_sigevent; /* Signal to deliver */
2735 int aio_lio_opcode; /* LIO opcode */
2736 int aio_reqprio; /* Request priority -- ignored */
2737 struct __aiocb_private32 _aiocb_private;
2738 } oaiocb32_t;
2739 #endif
2740
2741 typedef struct aiocb32 {
2742 int32_t aio_fildes; /* File descriptor */
2743 uint64_t aio_offset __packed; /* File offset for I/O */
2744 uint32_t aio_buf; /* I/O buffer in process space */
2745 uint32_t aio_nbytes; /* Number of bytes for I/O */
2746 int __spare__[2];
2747 uint32_t __spare2__;
2748 int aio_lio_opcode; /* LIO opcode */
2749 int aio_reqprio; /* Request priority -- ignored */
2750 struct __aiocb_private32 _aiocb_private;
2751 struct sigevent32 aio_sigevent; /* Signal to deliver */
2752 } aiocb32_t;
2753
2754 #ifdef COMPAT_FREEBSD6
2755 static int
convert_old_sigevent32(struct osigevent32 * osig,struct sigevent * nsig)2756 convert_old_sigevent32(struct osigevent32 *osig, struct sigevent *nsig)
2757 {
2758
2759 /*
2760 * Only SIGEV_NONE, SIGEV_SIGNAL, and SIGEV_KEVENT are
2761 * supported by AIO with the old sigevent structure.
2762 */
2763 CP(*osig, *nsig, sigev_notify);
2764 switch (nsig->sigev_notify) {
2765 case SIGEV_NONE:
2766 break;
2767 case SIGEV_SIGNAL:
2768 nsig->sigev_signo = osig->__sigev_u.__sigev_signo;
2769 break;
2770 case SIGEV_KEVENT:
2771 nsig->sigev_notify_kqueue =
2772 osig->__sigev_u.__sigev_notify_kqueue;
2773 PTRIN_CP(*osig, *nsig, sigev_value.sival_ptr);
2774 break;
2775 default:
2776 return (EINVAL);
2777 }
2778 return (0);
2779 }
2780
2781 static int
aiocb32_copyin_old_sigevent(struct aiocb * ujob,struct kaiocb * kjob,int type __unused)2782 aiocb32_copyin_old_sigevent(struct aiocb *ujob, struct kaiocb *kjob,
2783 int type __unused)
2784 {
2785 struct oaiocb32 job32;
2786 struct aiocb *kcb = &kjob->uaiocb;
2787 int error;
2788
2789 bzero(kcb, sizeof(struct aiocb));
2790 error = copyin(ujob, &job32, sizeof(job32));
2791 if (error)
2792 return (error);
2793
2794 /* No need to copyin aio_iov, because it did not exist in FreeBSD 6 */
2795
2796 CP(job32, *kcb, aio_fildes);
2797 CP(job32, *kcb, aio_offset);
2798 PTRIN_CP(job32, *kcb, aio_buf);
2799 CP(job32, *kcb, aio_nbytes);
2800 CP(job32, *kcb, aio_lio_opcode);
2801 CP(job32, *kcb, aio_reqprio);
2802 CP(job32, *kcb, _aiocb_private.status);
2803 CP(job32, *kcb, _aiocb_private.error);
2804 PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo);
2805 return (convert_old_sigevent32(&job32.aio_sigevent,
2806 &kcb->aio_sigevent));
2807 }
2808 #endif
2809
2810 static int
aiocb32_copyin(struct aiocb * ujob,struct kaiocb * kjob,int type)2811 aiocb32_copyin(struct aiocb *ujob, struct kaiocb *kjob, int type)
2812 {
2813 struct aiocb32 job32;
2814 struct aiocb *kcb = &kjob->uaiocb;
2815 struct iovec32 *iov32;
2816 int error;
2817
2818 error = copyin(ujob, &job32, sizeof(job32));
2819 if (error)
2820 return (error);
2821 CP(job32, *kcb, aio_fildes);
2822 CP(job32, *kcb, aio_offset);
2823 CP(job32, *kcb, aio_lio_opcode);
2824 if (type == LIO_NOP)
2825 type = kcb->aio_lio_opcode;
2826 if (type & LIO_VECTORED) {
2827 iov32 = PTRIN(job32.aio_iov);
2828 CP(job32, *kcb, aio_iovcnt);
2829 /* malloc a uio and copy in the iovec */
2830 error = freebsd32_copyinuio(iov32,
2831 kcb->aio_iovcnt, &kjob->uiop);
2832 if (error)
2833 return (error);
2834 } else {
2835 PTRIN_CP(job32, *kcb, aio_buf);
2836 CP(job32, *kcb, aio_nbytes);
2837 }
2838 CP(job32, *kcb, aio_reqprio);
2839 CP(job32, *kcb, _aiocb_private.status);
2840 CP(job32, *kcb, _aiocb_private.error);
2841 PTRIN_CP(job32, *kcb, _aiocb_private.kernelinfo);
2842 error = convert_sigevent32(&job32.aio_sigevent, &kcb->aio_sigevent);
2843
2844 return (error);
2845 }
2846
2847 static long
aiocb32_fetch_status(struct aiocb * ujob)2848 aiocb32_fetch_status(struct aiocb *ujob)
2849 {
2850 struct aiocb32 *ujob32;
2851
2852 ujob32 = (struct aiocb32 *)ujob;
2853 return (fuword32(&ujob32->_aiocb_private.status));
2854 }
2855
2856 static long
aiocb32_fetch_error(struct aiocb * ujob)2857 aiocb32_fetch_error(struct aiocb *ujob)
2858 {
2859 struct aiocb32 *ujob32;
2860
2861 ujob32 = (struct aiocb32 *)ujob;
2862 return (fuword32(&ujob32->_aiocb_private.error));
2863 }
2864
2865 static int
aiocb32_store_status(struct aiocb * ujob,long status)2866 aiocb32_store_status(struct aiocb *ujob, long status)
2867 {
2868 struct aiocb32 *ujob32;
2869
2870 ujob32 = (struct aiocb32 *)ujob;
2871 return (suword32(&ujob32->_aiocb_private.status, status));
2872 }
2873
2874 static int
aiocb32_store_error(struct aiocb * ujob,long error)2875 aiocb32_store_error(struct aiocb *ujob, long error)
2876 {
2877 struct aiocb32 *ujob32;
2878
2879 ujob32 = (struct aiocb32 *)ujob;
2880 return (suword32(&ujob32->_aiocb_private.error, error));
2881 }
2882
2883 static int
aiocb32_store_kernelinfo(struct aiocb * ujob,long jobref)2884 aiocb32_store_kernelinfo(struct aiocb *ujob, long jobref)
2885 {
2886 struct aiocb32 *ujob32;
2887
2888 ujob32 = (struct aiocb32 *)ujob;
2889 return (suword32(&ujob32->_aiocb_private.kernelinfo, jobref));
2890 }
2891
2892 static int
aiocb32_store_aiocb(struct aiocb ** ujobp,struct aiocb * ujob)2893 aiocb32_store_aiocb(struct aiocb **ujobp, struct aiocb *ujob)
2894 {
2895
2896 return (suword32(ujobp, (long)ujob));
2897 }
2898
2899 static struct aiocb_ops aiocb32_ops = {
2900 .aio_copyin = aiocb32_copyin,
2901 .fetch_status = aiocb32_fetch_status,
2902 .fetch_error = aiocb32_fetch_error,
2903 .store_status = aiocb32_store_status,
2904 .store_error = aiocb32_store_error,
2905 .store_kernelinfo = aiocb32_store_kernelinfo,
2906 .store_aiocb = aiocb32_store_aiocb,
2907 };
2908
2909 #ifdef COMPAT_FREEBSD6
2910 static struct aiocb_ops aiocb32_ops_osigevent = {
2911 .aio_copyin = aiocb32_copyin_old_sigevent,
2912 .fetch_status = aiocb32_fetch_status,
2913 .fetch_error = aiocb32_fetch_error,
2914 .store_status = aiocb32_store_status,
2915 .store_error = aiocb32_store_error,
2916 .store_kernelinfo = aiocb32_store_kernelinfo,
2917 .store_aiocb = aiocb32_store_aiocb,
2918 };
2919 #endif
2920
2921 int
freebsd32_aio_return(struct thread * td,struct freebsd32_aio_return_args * uap)2922 freebsd32_aio_return(struct thread *td, struct freebsd32_aio_return_args *uap)
2923 {
2924
2925 return (kern_aio_return(td, (struct aiocb *)uap->aiocbp, &aiocb32_ops));
2926 }
2927
2928 int
freebsd32_aio_suspend(struct thread * td,struct freebsd32_aio_suspend_args * uap)2929 freebsd32_aio_suspend(struct thread *td, struct freebsd32_aio_suspend_args *uap)
2930 {
2931 struct timespec32 ts32;
2932 struct timespec ts, *tsp;
2933 struct aiocb **ujoblist;
2934 uint32_t *ujoblist32;
2935 int error, i;
2936
2937 if (uap->nent < 0 || uap->nent > max_aio_queue_per_proc)
2938 return (EINVAL);
2939
2940 if (uap->timeout) {
2941 /* Get timespec struct. */
2942 if ((error = copyin(uap->timeout, &ts32, sizeof(ts32))) != 0)
2943 return (error);
2944 CP(ts32, ts, tv_sec);
2945 CP(ts32, ts, tv_nsec);
2946 tsp = &ts;
2947 } else
2948 tsp = NULL;
2949
2950 ujoblist = malloc(uap->nent * sizeof(ujoblist[0]), M_AIO, M_WAITOK);
2951 ujoblist32 = (uint32_t *)ujoblist;
2952 error = copyin(uap->aiocbp, ujoblist32, uap->nent *
2953 sizeof(ujoblist32[0]));
2954 if (error == 0) {
2955 for (i = uap->nent - 1; i >= 0; i--)
2956 ujoblist[i] = PTRIN(ujoblist32[i]);
2957
2958 error = kern_aio_suspend(td, uap->nent, ujoblist, tsp);
2959 }
2960 free(ujoblist, M_AIO);
2961 return (error);
2962 }
2963
2964 int
freebsd32_aio_error(struct thread * td,struct freebsd32_aio_error_args * uap)2965 freebsd32_aio_error(struct thread *td, struct freebsd32_aio_error_args *uap)
2966 {
2967
2968 return (kern_aio_error(td, (struct aiocb *)uap->aiocbp, &aiocb32_ops));
2969 }
2970
2971 #ifdef COMPAT_FREEBSD6
2972 int
freebsd6_freebsd32_aio_read(struct thread * td,struct freebsd6_freebsd32_aio_read_args * uap)2973 freebsd6_freebsd32_aio_read(struct thread *td,
2974 struct freebsd6_freebsd32_aio_read_args *uap)
2975 {
2976
2977 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ,
2978 &aiocb32_ops_osigevent));
2979 }
2980 #endif
2981
2982 int
freebsd32_aio_read(struct thread * td,struct freebsd32_aio_read_args * uap)2983 freebsd32_aio_read(struct thread *td, struct freebsd32_aio_read_args *uap)
2984 {
2985
2986 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READ,
2987 &aiocb32_ops));
2988 }
2989
2990 int
freebsd32_aio_readv(struct thread * td,struct freebsd32_aio_readv_args * uap)2991 freebsd32_aio_readv(struct thread *td, struct freebsd32_aio_readv_args *uap)
2992 {
2993
2994 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_READV,
2995 &aiocb32_ops));
2996 }
2997
2998 #ifdef COMPAT_FREEBSD6
2999 int
freebsd6_freebsd32_aio_write(struct thread * td,struct freebsd6_freebsd32_aio_write_args * uap)3000 freebsd6_freebsd32_aio_write(struct thread *td,
3001 struct freebsd6_freebsd32_aio_write_args *uap)
3002 {
3003
3004 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE,
3005 &aiocb32_ops_osigevent));
3006 }
3007 #endif
3008
3009 int
freebsd32_aio_write(struct thread * td,struct freebsd32_aio_write_args * uap)3010 freebsd32_aio_write(struct thread *td, struct freebsd32_aio_write_args *uap)
3011 {
3012
3013 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITE,
3014 &aiocb32_ops));
3015 }
3016
3017 int
freebsd32_aio_writev(struct thread * td,struct freebsd32_aio_writev_args * uap)3018 freebsd32_aio_writev(struct thread *td, struct freebsd32_aio_writev_args *uap)
3019 {
3020
3021 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_WRITEV,
3022 &aiocb32_ops));
3023 }
3024
3025 int
freebsd32_aio_mlock(struct thread * td,struct freebsd32_aio_mlock_args * uap)3026 freebsd32_aio_mlock(struct thread *td, struct freebsd32_aio_mlock_args *uap)
3027 {
3028
3029 return (aio_aqueue(td, (struct aiocb *)uap->aiocbp, NULL, LIO_MLOCK,
3030 &aiocb32_ops));
3031 }
3032
3033 int
freebsd32_aio_waitcomplete(struct thread * td,struct freebsd32_aio_waitcomplete_args * uap)3034 freebsd32_aio_waitcomplete(struct thread *td,
3035 struct freebsd32_aio_waitcomplete_args *uap)
3036 {
3037 struct timespec32 ts32;
3038 struct timespec ts, *tsp;
3039 int error;
3040
3041 if (uap->timeout) {
3042 /* Get timespec struct. */
3043 error = copyin(uap->timeout, &ts32, sizeof(ts32));
3044 if (error)
3045 return (error);
3046 CP(ts32, ts, tv_sec);
3047 CP(ts32, ts, tv_nsec);
3048 tsp = &ts;
3049 } else
3050 tsp = NULL;
3051
3052 return (kern_aio_waitcomplete(td, (struct aiocb **)uap->aiocbp, tsp,
3053 &aiocb32_ops));
3054 }
3055
3056 int
freebsd32_aio_fsync(struct thread * td,struct freebsd32_aio_fsync_args * uap)3057 freebsd32_aio_fsync(struct thread *td, struct freebsd32_aio_fsync_args *uap)
3058 {
3059
3060 return (kern_aio_fsync(td, uap->op, (struct aiocb *)uap->aiocbp,
3061 &aiocb32_ops));
3062 }
3063
3064 #ifdef COMPAT_FREEBSD6
3065 int
freebsd6_freebsd32_lio_listio(struct thread * td,struct freebsd6_freebsd32_lio_listio_args * uap)3066 freebsd6_freebsd32_lio_listio(struct thread *td,
3067 struct freebsd6_freebsd32_lio_listio_args *uap)
3068 {
3069 struct aiocb **acb_list;
3070 struct sigevent *sigp, sig;
3071 struct osigevent32 osig;
3072 uint32_t *acb_list32;
3073 int error, i, nent;
3074
3075 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
3076 return (EINVAL);
3077
3078 nent = uap->nent;
3079 if (nent < 0 || nent > max_aio_queue_per_proc)
3080 return (EINVAL);
3081
3082 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
3083 error = copyin(uap->sig, &osig, sizeof(osig));
3084 if (error)
3085 return (error);
3086 error = convert_old_sigevent32(&osig, &sig);
3087 if (error)
3088 return (error);
3089 sigp = &sig;
3090 } else
3091 sigp = NULL;
3092
3093 acb_list32 = malloc(sizeof(uint32_t) * nent, M_LIO, M_WAITOK);
3094 error = copyin(uap->acb_list, acb_list32, nent * sizeof(uint32_t));
3095 if (error) {
3096 free(acb_list32, M_LIO);
3097 return (error);
3098 }
3099 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
3100 for (i = 0; i < nent; i++)
3101 acb_list[i] = PTRIN(acb_list32[i]);
3102 free(acb_list32, M_LIO);
3103
3104 error = kern_lio_listio(td, uap->mode,
3105 (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp,
3106 &aiocb32_ops_osigevent);
3107 free(acb_list, M_LIO);
3108 return (error);
3109 }
3110 #endif
3111
3112 int
freebsd32_lio_listio(struct thread * td,struct freebsd32_lio_listio_args * uap)3113 freebsd32_lio_listio(struct thread *td, struct freebsd32_lio_listio_args *uap)
3114 {
3115 struct aiocb **acb_list;
3116 struct sigevent *sigp, sig;
3117 struct sigevent32 sig32;
3118 uint32_t *acb_list32;
3119 int error, i, nent;
3120
3121 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT))
3122 return (EINVAL);
3123
3124 nent = uap->nent;
3125 if (nent < 0 || nent > max_aio_queue_per_proc)
3126 return (EINVAL);
3127
3128 if (uap->sig && (uap->mode == LIO_NOWAIT)) {
3129 error = copyin(uap->sig, &sig32, sizeof(sig32));
3130 if (error)
3131 return (error);
3132 error = convert_sigevent32(&sig32, &sig);
3133 if (error)
3134 return (error);
3135 sigp = &sig;
3136 } else
3137 sigp = NULL;
3138
3139 acb_list32 = malloc(sizeof(uint32_t) * nent, M_LIO, M_WAITOK);
3140 error = copyin(uap->acb_list, acb_list32, nent * sizeof(uint32_t));
3141 if (error) {
3142 free(acb_list32, M_LIO);
3143 return (error);
3144 }
3145 acb_list = malloc(sizeof(struct aiocb *) * nent, M_LIO, M_WAITOK);
3146 for (i = 0; i < nent; i++)
3147 acb_list[i] = PTRIN(acb_list32[i]);
3148 free(acb_list32, M_LIO);
3149
3150 error = kern_lio_listio(td, uap->mode,
3151 (struct aiocb * const *)uap->acb_list, acb_list, nent, sigp,
3152 &aiocb32_ops);
3153 free(acb_list, M_LIO);
3154 return (error);
3155 }
3156
3157 #endif
3158