1 /* 2 * Copyright (c) 1997 John S. Dyson. All rights reserved. 3 * 4 * Redistribution and use in source and binary forms, with or without 5 * modification, are permitted provided that the following conditions 6 * are met: 7 * 1. Redistributions of source code must retain the above copyright 8 * notice, this list of conditions and the following disclaimer. 9 * 2. John S. Dyson's name may not be used to endorse or promote products 10 * derived from this software without specific prior written permission. 11 * 12 * DISCLAIMER: This code isn't warranted to do anything useful. Anything 13 * bad that happens because of using this software isn't the responsibility 14 * of the author. This software is distributed AS-IS. 15 */ 16 17 /* 18 * This file contains support for the POSIX 1003.1B AIO/LIO facility. 19 */ 20 21 #include <sys/cdefs.h> 22 __FBSDID("$FreeBSD$"); 23 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/eventhandler.h> 30 #include <sys/sysproto.h> 31 #include <sys/filedesc.h> 32 #include <sys/kernel.h> 33 #include <sys/module.h> 34 #include <sys/kthread.h> 35 #include <sys/fcntl.h> 36 #include <sys/file.h> 37 #include <sys/limits.h> 38 #include <sys/lock.h> 39 #include <sys/mutex.h> 40 #include <sys/unistd.h> 41 #include <sys/proc.h> 42 #include <sys/resourcevar.h> 43 #include <sys/signalvar.h> 44 #include <sys/protosw.h> 45 #include <sys/socketvar.h> 46 #include <sys/syscall.h> 47 #include <sys/sysent.h> 48 #include <sys/sysctl.h> 49 #include <sys/sx.h> 50 #include <sys/vnode.h> 51 #include <sys/conf.h> 52 #include <sys/event.h> 53 54 #include <posix4/posix4.h> 55 #include <vm/vm.h> 56 #include <vm/vm_extern.h> 57 #include <vm/pmap.h> 58 #include <vm/vm_map.h> 59 #include <vm/uma.h> 60 #include <sys/aio.h> 61 62 #include "opt_vfs_aio.h" 63 64 NET_NEEDS_GIANT("aio"); 65 66 /* 67 * Counter for allocating reference ids to new jobs. Wrapped to 1 on 68 * overflow. 69 */ 70 static long jobrefid; 71 72 #define JOBST_NULL 0x0 73 #define JOBST_JOBQGLOBAL 0x2 74 #define JOBST_JOBRUNNING 0x3 75 #define JOBST_JOBFINISHED 0x4 76 #define JOBST_JOBQBUF 0x5 77 #define JOBST_JOBBFINISHED 0x6 78 79 #ifndef MAX_AIO_PER_PROC 80 #define MAX_AIO_PER_PROC 32 81 #endif 82 83 #ifndef MAX_AIO_QUEUE_PER_PROC 84 #define MAX_AIO_QUEUE_PER_PROC 256 /* Bigger than AIO_LISTIO_MAX */ 85 #endif 86 87 #ifndef MAX_AIO_PROCS 88 #define MAX_AIO_PROCS 32 89 #endif 90 91 #ifndef MAX_AIO_QUEUE 92 #define MAX_AIO_QUEUE 1024 /* Bigger than AIO_LISTIO_MAX */ 93 #endif 94 95 #ifndef TARGET_AIO_PROCS 96 #define TARGET_AIO_PROCS 4 97 #endif 98 99 #ifndef MAX_BUF_AIO 100 #define MAX_BUF_AIO 16 101 #endif 102 103 #ifndef AIOD_TIMEOUT_DEFAULT 104 #define AIOD_TIMEOUT_DEFAULT (10 * hz) 105 #endif 106 107 #ifndef AIOD_LIFETIME_DEFAULT 108 #define AIOD_LIFETIME_DEFAULT (30 * hz) 109 #endif 110 111 SYSCTL_NODE(_vfs, OID_AUTO, aio, CTLFLAG_RW, 0, "Async IO management"); 112 113 static int max_aio_procs = MAX_AIO_PROCS; 114 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_procs, 115 CTLFLAG_RW, &max_aio_procs, 0, 116 "Maximum number of kernel threads to use for handling async IO "); 117 118 static int num_aio_procs = 0; 119 SYSCTL_INT(_vfs_aio, OID_AUTO, num_aio_procs, 120 CTLFLAG_RD, &num_aio_procs, 0, 121 "Number of presently active kernel threads for async IO"); 122 123 /* 124 * The code will adjust the actual number of AIO processes towards this 125 * number when it gets a chance. 126 */ 127 static int target_aio_procs = TARGET_AIO_PROCS; 128 SYSCTL_INT(_vfs_aio, OID_AUTO, target_aio_procs, CTLFLAG_RW, &target_aio_procs, 129 0, "Preferred number of ready kernel threads for async IO"); 130 131 static int max_queue_count = MAX_AIO_QUEUE; 132 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue, CTLFLAG_RW, &max_queue_count, 0, 133 "Maximum number of aio requests to queue, globally"); 134 135 static int num_queue_count = 0; 136 SYSCTL_INT(_vfs_aio, OID_AUTO, num_queue_count, CTLFLAG_RD, &num_queue_count, 0, 137 "Number of queued aio requests"); 138 139 static int num_buf_aio = 0; 140 SYSCTL_INT(_vfs_aio, OID_AUTO, num_buf_aio, CTLFLAG_RD, &num_buf_aio, 0, 141 "Number of aio requests presently handled by the buf subsystem"); 142 143 /* Number of async I/O thread in the process of being started */ 144 /* XXX This should be local to _aio_aqueue() */ 145 static int num_aio_resv_start = 0; 146 147 static int aiod_timeout; 148 SYSCTL_INT(_vfs_aio, OID_AUTO, aiod_timeout, CTLFLAG_RW, &aiod_timeout, 0, 149 "Timeout value for synchronous aio operations"); 150 151 static int aiod_lifetime; 152 SYSCTL_INT(_vfs_aio, OID_AUTO, aiod_lifetime, CTLFLAG_RW, &aiod_lifetime, 0, 153 "Maximum lifetime for idle aiod"); 154 155 static int unloadable = 0; 156 SYSCTL_INT(_vfs_aio, OID_AUTO, unloadable, CTLFLAG_RW, &unloadable, 0, 157 "Allow unload of aio (not recommended)"); 158 159 160 static int max_aio_per_proc = MAX_AIO_PER_PROC; 161 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_per_proc, CTLFLAG_RW, &max_aio_per_proc, 162 0, "Maximum active aio requests per process (stored in the process)"); 163 164 static int max_aio_queue_per_proc = MAX_AIO_QUEUE_PER_PROC; 165 SYSCTL_INT(_vfs_aio, OID_AUTO, max_aio_queue_per_proc, CTLFLAG_RW, 166 &max_aio_queue_per_proc, 0, 167 "Maximum queued aio requests per process (stored in the process)"); 168 169 static int max_buf_aio = MAX_BUF_AIO; 170 SYSCTL_INT(_vfs_aio, OID_AUTO, max_buf_aio, CTLFLAG_RW, &max_buf_aio, 0, 171 "Maximum buf aio requests per process (stored in the process)"); 172 173 struct aiocblist { 174 TAILQ_ENTRY(aiocblist) list; /* List of jobs */ 175 TAILQ_ENTRY(aiocblist) plist; /* List of jobs for proc */ 176 int jobflags; 177 int jobstate; 178 int inputcharge; 179 int outputcharge; 180 struct callout_handle timeouthandle; 181 struct buf *bp; /* Buffer pointer */ 182 struct proc *userproc; /* User process */ /* Not td! */ 183 struct ucred *cred; /* Active credential when created */ 184 struct file *fd_file; /* Pointer to file structure */ 185 struct aio_liojob *lio; /* Optional lio job */ 186 struct aiocb *uuaiocb; /* Pointer in userspace of aiocb */ 187 struct knlist klist; /* list of knotes */ 188 struct aiocb uaiocb; /* Kernel I/O control block */ 189 }; 190 191 /* jobflags */ 192 #define AIOCBLIST_RUNDOWN 0x4 193 #define AIOCBLIST_DONE 0x10 194 195 /* 196 * AIO process info 197 */ 198 #define AIOP_FREE 0x1 /* proc on free queue */ 199 #define AIOP_SCHED 0x2 /* proc explicitly scheduled */ 200 201 struct aiothreadlist { 202 int aiothreadflags; /* AIO proc flags */ 203 TAILQ_ENTRY(aiothreadlist) list; /* List of processes */ 204 struct thread *aiothread; /* The AIO thread */ 205 }; 206 207 /* 208 * data-structure for lio signal management 209 */ 210 struct aio_liojob { 211 int lioj_flags; 212 int lioj_buffer_count; 213 int lioj_buffer_finished_count; 214 int lioj_queue_count; 215 int lioj_queue_finished_count; 216 struct sigevent lioj_signal; /* signal on all I/O done */ 217 TAILQ_ENTRY(aio_liojob) lioj_list; 218 struct kaioinfo *lioj_ki; 219 }; 220 #define LIOJ_SIGNAL 0x1 /* signal on all done (lio) */ 221 #define LIOJ_SIGNAL_POSTED 0x2 /* signal has been posted */ 222 223 /* 224 * per process aio data structure 225 */ 226 struct kaioinfo { 227 int kaio_flags; /* per process kaio flags */ 228 int kaio_maxactive_count; /* maximum number of AIOs */ 229 int kaio_active_count; /* number of currently used AIOs */ 230 int kaio_qallowed_count; /* maxiumu size of AIO queue */ 231 int kaio_queue_count; /* size of AIO queue */ 232 int kaio_ballowed_count; /* maximum number of buffers */ 233 int kaio_queue_finished_count; /* number of daemon jobs finished */ 234 int kaio_buffer_count; /* number of physio buffers */ 235 int kaio_buffer_finished_count; /* count of I/O done */ 236 struct proc *kaio_p; /* process that uses this kaio block */ 237 TAILQ_HEAD(,aio_liojob) kaio_liojoblist; /* list of lio jobs */ 238 TAILQ_HEAD(,aiocblist) kaio_jobqueue; /* job queue for process */ 239 TAILQ_HEAD(,aiocblist) kaio_jobdone; /* done queue for process */ 240 TAILQ_HEAD(,aiocblist) kaio_bufqueue; /* buffer job queue for process */ 241 TAILQ_HEAD(,aiocblist) kaio_bufdone; /* buffer done queue for process */ 242 TAILQ_HEAD(,aiocblist) kaio_sockqueue; /* queue for aios waiting on sockets */ 243 }; 244 245 #define KAIO_RUNDOWN 0x1 /* process is being run down */ 246 #define KAIO_WAKEUP 0x2 /* wakeup process when there is a significant event */ 247 248 static TAILQ_HEAD(,aiothreadlist) aio_activeproc; /* Active daemons */ 249 static TAILQ_HEAD(,aiothreadlist) aio_freeproc; /* Idle daemons */ 250 static TAILQ_HEAD(,aiocblist) aio_jobs; /* Async job list */ 251 static TAILQ_HEAD(,aiocblist) aio_bufjobs; /* Phys I/O job list */ 252 253 static void aio_init_aioinfo(struct proc *p); 254 static void aio_onceonly(void); 255 static int aio_free_entry(struct aiocblist *aiocbe); 256 static void aio_process(struct aiocblist *aiocbe); 257 static int aio_newproc(void); 258 static int aio_aqueue(struct thread *td, struct aiocb *job, int type); 259 static void aio_physwakeup(struct buf *bp); 260 static void aio_proc_rundown(void *arg, struct proc *p); 261 static int aio_fphysio(struct aiocblist *aiocbe); 262 static int aio_qphysio(struct proc *p, struct aiocblist *iocb); 263 static void aio_daemon(void *uproc); 264 static void aio_swake_cb(struct socket *, struct sockbuf *); 265 static int aio_unload(void); 266 static void process_signal(void *aioj); 267 static int filt_aioattach(struct knote *kn); 268 static void filt_aiodetach(struct knote *kn); 269 static int filt_aio(struct knote *kn, long hint); 270 271 /* 272 * Zones for: 273 * kaio Per process async io info 274 * aiop async io thread data 275 * aiocb async io jobs 276 * aiol list io job pointer - internal to aio_suspend XXX 277 * aiolio list io jobs 278 */ 279 static uma_zone_t kaio_zone, aiop_zone, aiocb_zone, aiol_zone, aiolio_zone; 280 281 /* kqueue filters for aio */ 282 static struct filterops aio_filtops = 283 { 0, filt_aioattach, filt_aiodetach, filt_aio }; 284 285 static eventhandler_tag exit_tag, exec_tag; 286 287 /* 288 * Main operations function for use as a kernel module. 289 */ 290 static int 291 aio_modload(struct module *module, int cmd, void *arg) 292 { 293 int error = 0; 294 295 switch (cmd) { 296 case MOD_LOAD: 297 aio_onceonly(); 298 break; 299 case MOD_UNLOAD: 300 error = aio_unload(); 301 break; 302 case MOD_SHUTDOWN: 303 break; 304 default: 305 error = EINVAL; 306 break; 307 } 308 return (error); 309 } 310 311 static moduledata_t aio_mod = { 312 "aio", 313 &aio_modload, 314 NULL 315 }; 316 317 SYSCALL_MODULE_HELPER(aio_return); 318 SYSCALL_MODULE_HELPER(aio_suspend); 319 SYSCALL_MODULE_HELPER(aio_cancel); 320 SYSCALL_MODULE_HELPER(aio_error); 321 SYSCALL_MODULE_HELPER(aio_read); 322 SYSCALL_MODULE_HELPER(aio_write); 323 SYSCALL_MODULE_HELPER(aio_waitcomplete); 324 SYSCALL_MODULE_HELPER(lio_listio); 325 326 DECLARE_MODULE(aio, aio_mod, 327 SI_SUB_VFS, SI_ORDER_ANY); 328 MODULE_VERSION(aio, 1); 329 330 /* 331 * Startup initialization 332 */ 333 static void 334 aio_onceonly(void) 335 { 336 337 /* XXX: should probably just use so->callback */ 338 aio_swake = &aio_swake_cb; 339 exit_tag = EVENTHANDLER_REGISTER(process_exit, aio_proc_rundown, NULL, 340 EVENTHANDLER_PRI_ANY); 341 exec_tag = EVENTHANDLER_REGISTER(process_exec, aio_proc_rundown, NULL, 342 EVENTHANDLER_PRI_ANY); 343 kqueue_add_filteropts(EVFILT_AIO, &aio_filtops); 344 TAILQ_INIT(&aio_freeproc); 345 TAILQ_INIT(&aio_activeproc); 346 TAILQ_INIT(&aio_jobs); 347 TAILQ_INIT(&aio_bufjobs); 348 kaio_zone = uma_zcreate("AIO", sizeof(struct kaioinfo), NULL, NULL, 349 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 350 aiop_zone = uma_zcreate("AIOP", sizeof(struct aiothreadlist), NULL, 351 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 352 aiocb_zone = uma_zcreate("AIOCB", sizeof(struct aiocblist), NULL, NULL, 353 NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 354 aiol_zone = uma_zcreate("AIOL", AIO_LISTIO_MAX*sizeof(intptr_t) , NULL, 355 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 356 aiolio_zone = uma_zcreate("AIOLIO", sizeof(struct aio_liojob), NULL, 357 NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 358 aiod_timeout = AIOD_TIMEOUT_DEFAULT; 359 aiod_lifetime = AIOD_LIFETIME_DEFAULT; 360 jobrefid = 1; 361 async_io_version = _POSIX_VERSION; 362 p31b_setcfg(CTL_P1003_1B_AIO_LISTIO_MAX, AIO_LISTIO_MAX); 363 p31b_setcfg(CTL_P1003_1B_AIO_MAX, MAX_AIO_QUEUE); 364 p31b_setcfg(CTL_P1003_1B_AIO_PRIO_DELTA_MAX, 0); 365 } 366 367 /* 368 * Callback for unload of AIO when used as a module. 369 */ 370 static int 371 aio_unload(void) 372 { 373 int error; 374 375 /* 376 * XXX: no unloads by default, it's too dangerous. 377 * perhaps we could do it if locked out callers and then 378 * did an aio_proc_rundown() on each process. 379 */ 380 if (!unloadable) 381 return (EOPNOTSUPP); 382 383 error = kqueue_del_filteropts(EVFILT_AIO); 384 if (error) 385 return error; 386 387 async_io_version = 0; 388 aio_swake = NULL; 389 EVENTHANDLER_DEREGISTER(process_exit, exit_tag); 390 EVENTHANDLER_DEREGISTER(process_exec, exec_tag); 391 p31b_setcfg(CTL_P1003_1B_AIO_LISTIO_MAX, -1); 392 p31b_setcfg(CTL_P1003_1B_AIO_MAX, -1); 393 p31b_setcfg(CTL_P1003_1B_AIO_PRIO_DELTA_MAX, -1); 394 return (0); 395 } 396 397 /* 398 * Init the per-process aioinfo structure. The aioinfo limits are set 399 * per-process for user limit (resource) management. 400 */ 401 static void 402 aio_init_aioinfo(struct proc *p) 403 { 404 struct kaioinfo *ki; 405 406 if (p->p_aioinfo == NULL) { 407 ki = uma_zalloc(kaio_zone, M_WAITOK); 408 p->p_aioinfo = ki; 409 ki->kaio_flags = 0; 410 ki->kaio_maxactive_count = max_aio_per_proc; 411 ki->kaio_active_count = 0; 412 ki->kaio_qallowed_count = max_aio_queue_per_proc; 413 ki->kaio_queue_count = 0; 414 ki->kaio_ballowed_count = max_buf_aio; 415 ki->kaio_buffer_count = 0; 416 ki->kaio_buffer_finished_count = 0; 417 ki->kaio_p = p; 418 TAILQ_INIT(&ki->kaio_jobdone); 419 TAILQ_INIT(&ki->kaio_jobqueue); 420 TAILQ_INIT(&ki->kaio_bufdone); 421 TAILQ_INIT(&ki->kaio_bufqueue); 422 TAILQ_INIT(&ki->kaio_liojoblist); 423 TAILQ_INIT(&ki->kaio_sockqueue); 424 } 425 426 while (num_aio_procs < target_aio_procs) 427 aio_newproc(); 428 } 429 430 /* 431 * Free a job entry. Wait for completion if it is currently active, but don't 432 * delay forever. If we delay, we return a flag that says that we have to 433 * restart the queue scan. 434 */ 435 static int 436 aio_free_entry(struct aiocblist *aiocbe) 437 { 438 struct kaioinfo *ki; 439 struct aio_liojob *lj; 440 struct proc *p; 441 int error; 442 int s; 443 444 if (aiocbe->jobstate == JOBST_NULL) 445 panic("aio_free_entry: freeing already free job"); 446 447 p = aiocbe->userproc; 448 ki = p->p_aioinfo; 449 lj = aiocbe->lio; 450 if (ki == NULL) 451 panic("aio_free_entry: missing p->p_aioinfo"); 452 453 while (aiocbe->jobstate == JOBST_JOBRUNNING) { 454 aiocbe->jobflags |= AIOCBLIST_RUNDOWN; 455 tsleep(aiocbe, PRIBIO, "jobwai", 0); 456 } 457 if (aiocbe->bp == NULL) { 458 if (ki->kaio_queue_count <= 0) 459 panic("aio_free_entry: process queue size <= 0"); 460 if (num_queue_count <= 0) 461 panic("aio_free_entry: system wide queue size <= 0"); 462 463 if (lj) { 464 lj->lioj_queue_count--; 465 if (aiocbe->jobflags & AIOCBLIST_DONE) 466 lj->lioj_queue_finished_count--; 467 } 468 ki->kaio_queue_count--; 469 if (aiocbe->jobflags & AIOCBLIST_DONE) 470 ki->kaio_queue_finished_count--; 471 num_queue_count--; 472 } else { 473 if (lj) { 474 lj->lioj_buffer_count--; 475 if (aiocbe->jobflags & AIOCBLIST_DONE) 476 lj->lioj_buffer_finished_count--; 477 } 478 if (aiocbe->jobflags & AIOCBLIST_DONE) 479 ki->kaio_buffer_finished_count--; 480 ki->kaio_buffer_count--; 481 num_buf_aio--; 482 } 483 484 /* aiocbe is going away, we need to destroy any knotes */ 485 /* XXXKSE Note the thread here is used to eventually find the 486 * owning process again, but it is also used to do a fo_close 487 * and that requires the thread. (but does it require the 488 * OWNING thread? (or maybe the running thread?) 489 * There is a semantic problem here... 490 */ 491 knlist_clear(&aiocbe->klist, 0); /* XXXKSE */ 492 493 if ((ki->kaio_flags & KAIO_WAKEUP) || ((ki->kaio_flags & KAIO_RUNDOWN) 494 && ((ki->kaio_buffer_count == 0) && (ki->kaio_queue_count == 0)))) { 495 ki->kaio_flags &= ~KAIO_WAKEUP; 496 wakeup(p); 497 } 498 499 if (aiocbe->jobstate == JOBST_JOBQBUF) { 500 if ((error = aio_fphysio(aiocbe)) != 0) 501 return (error); 502 if (aiocbe->jobstate != JOBST_JOBBFINISHED) 503 panic("aio_free_entry: invalid physio finish-up state"); 504 s = splbio(); 505 TAILQ_REMOVE(&ki->kaio_bufdone, aiocbe, plist); 506 splx(s); 507 } else if (aiocbe->jobstate == JOBST_JOBQGLOBAL) { 508 s = splnet(); 509 TAILQ_REMOVE(&aio_jobs, aiocbe, list); 510 TAILQ_REMOVE(&ki->kaio_jobqueue, aiocbe, plist); 511 splx(s); 512 } else if (aiocbe->jobstate == JOBST_JOBFINISHED) 513 TAILQ_REMOVE(&ki->kaio_jobdone, aiocbe, plist); 514 else if (aiocbe->jobstate == JOBST_JOBBFINISHED) { 515 s = splbio(); 516 TAILQ_REMOVE(&ki->kaio_bufdone, aiocbe, plist); 517 splx(s); 518 if (aiocbe->bp) { 519 vunmapbuf(aiocbe->bp); 520 relpbuf(aiocbe->bp, NULL); 521 aiocbe->bp = NULL; 522 } 523 } 524 if (lj && (lj->lioj_buffer_count == 0) && (lj->lioj_queue_count == 0)) { 525 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list); 526 uma_zfree(aiolio_zone, lj); 527 } 528 aiocbe->jobstate = JOBST_NULL; 529 untimeout(process_signal, aiocbe, aiocbe->timeouthandle); 530 fdrop(aiocbe->fd_file, curthread); 531 crfree(aiocbe->cred); 532 uma_zfree(aiocb_zone, aiocbe); 533 return (0); 534 } 535 536 /* 537 * Rundown the jobs for a given process. 538 */ 539 static void 540 aio_proc_rundown(void *arg, struct proc *p) 541 { 542 int s; 543 struct kaioinfo *ki; 544 struct aio_liojob *lj, *ljn; 545 struct aiocblist *aiocbe, *aiocbn; 546 struct file *fp; 547 struct socket *so; 548 549 ki = p->p_aioinfo; 550 if (ki == NULL) 551 return; 552 553 mtx_lock(&Giant); 554 ki->kaio_flags |= LIOJ_SIGNAL_POSTED; 555 while ((ki->kaio_active_count > 0) || (ki->kaio_buffer_count > 556 ki->kaio_buffer_finished_count)) { 557 ki->kaio_flags |= KAIO_RUNDOWN; 558 if (tsleep(p, PRIBIO, "kaiowt", aiod_timeout)) 559 break; 560 } 561 562 /* 563 * Move any aio ops that are waiting on socket I/O to the normal job 564 * queues so they are cleaned up with any others. 565 */ 566 s = splnet(); 567 for (aiocbe = TAILQ_FIRST(&ki->kaio_sockqueue); aiocbe; aiocbe = 568 aiocbn) { 569 aiocbn = TAILQ_NEXT(aiocbe, plist); 570 fp = aiocbe->fd_file; 571 if (fp != NULL) { 572 so = fp->f_data; 573 TAILQ_REMOVE(&so->so_aiojobq, aiocbe, list); 574 if (TAILQ_EMPTY(&so->so_aiojobq)) { 575 SOCKBUF_LOCK(&so->so_snd); 576 so->so_snd.sb_flags &= ~SB_AIO; 577 SOCKBUF_UNLOCK(&so->so_snd); 578 SOCKBUF_LOCK(&so->so_rcv); 579 so->so_rcv.sb_flags &= ~SB_AIO; 580 SOCKBUF_UNLOCK(&so->so_rcv); 581 } 582 } 583 TAILQ_REMOVE(&ki->kaio_sockqueue, aiocbe, plist); 584 TAILQ_INSERT_HEAD(&aio_jobs, aiocbe, list); 585 TAILQ_INSERT_HEAD(&ki->kaio_jobqueue, aiocbe, plist); 586 } 587 splx(s); 588 589 restart1: 590 for (aiocbe = TAILQ_FIRST(&ki->kaio_jobdone); aiocbe; aiocbe = aiocbn) { 591 aiocbn = TAILQ_NEXT(aiocbe, plist); 592 if (aio_free_entry(aiocbe)) 593 goto restart1; 594 } 595 596 restart2: 597 for (aiocbe = TAILQ_FIRST(&ki->kaio_jobqueue); aiocbe; aiocbe = 598 aiocbn) { 599 aiocbn = TAILQ_NEXT(aiocbe, plist); 600 if (aio_free_entry(aiocbe)) 601 goto restart2; 602 } 603 604 /* 605 * Note the use of lots of splbio here, trying to avoid splbio for long chains 606 * of I/O. Probably unnecessary. 607 */ 608 restart3: 609 s = splbio(); 610 while (TAILQ_FIRST(&ki->kaio_bufqueue)) { 611 ki->kaio_flags |= KAIO_WAKEUP; 612 tsleep(p, PRIBIO, "aioprn", 0); 613 splx(s); 614 goto restart3; 615 } 616 splx(s); 617 618 restart4: 619 s = splbio(); 620 for (aiocbe = TAILQ_FIRST(&ki->kaio_bufdone); aiocbe; aiocbe = aiocbn) { 621 aiocbn = TAILQ_NEXT(aiocbe, plist); 622 if (aio_free_entry(aiocbe)) { 623 splx(s); 624 goto restart4; 625 } 626 } 627 splx(s); 628 629 /* 630 * If we've slept, jobs might have moved from one queue to another. 631 * Retry rundown if we didn't manage to empty the queues. 632 */ 633 if (TAILQ_FIRST(&ki->kaio_jobdone) != NULL || 634 TAILQ_FIRST(&ki->kaio_jobqueue) != NULL || 635 TAILQ_FIRST(&ki->kaio_bufqueue) != NULL || 636 TAILQ_FIRST(&ki->kaio_bufdone) != NULL) 637 goto restart1; 638 639 for (lj = TAILQ_FIRST(&ki->kaio_liojoblist); lj; lj = ljn) { 640 ljn = TAILQ_NEXT(lj, lioj_list); 641 if ((lj->lioj_buffer_count == 0) && (lj->lioj_queue_count == 642 0)) { 643 TAILQ_REMOVE(&ki->kaio_liojoblist, lj, lioj_list); 644 uma_zfree(aiolio_zone, lj); 645 } else { 646 #ifdef DIAGNOSTIC 647 printf("LIO job not cleaned up: B:%d, BF:%d, Q:%d, " 648 "QF:%d\n", lj->lioj_buffer_count, 649 lj->lioj_buffer_finished_count, 650 lj->lioj_queue_count, 651 lj->lioj_queue_finished_count); 652 #endif 653 } 654 } 655 656 uma_zfree(kaio_zone, ki); 657 p->p_aioinfo = NULL; 658 mtx_unlock(&Giant); 659 } 660 661 /* 662 * Select a job to run (called by an AIO daemon). 663 */ 664 static struct aiocblist * 665 aio_selectjob(struct aiothreadlist *aiop) 666 { 667 int s; 668 struct aiocblist *aiocbe; 669 struct kaioinfo *ki; 670 struct proc *userp; 671 672 s = splnet(); 673 for (aiocbe = TAILQ_FIRST(&aio_jobs); aiocbe; aiocbe = 674 TAILQ_NEXT(aiocbe, list)) { 675 userp = aiocbe->userproc; 676 ki = userp->p_aioinfo; 677 678 if (ki->kaio_active_count < ki->kaio_maxactive_count) { 679 TAILQ_REMOVE(&aio_jobs, aiocbe, list); 680 splx(s); 681 return (aiocbe); 682 } 683 } 684 splx(s); 685 686 return (NULL); 687 } 688 689 /* 690 * The AIO processing activity. This is the code that does the I/O request for 691 * the non-physio version of the operations. The normal vn operations are used, 692 * and this code should work in all instances for every type of file, including 693 * pipes, sockets, fifos, and regular files. 694 */ 695 static void 696 aio_process(struct aiocblist *aiocbe) 697 { 698 struct ucred *td_savedcred; 699 struct thread *td; 700 struct proc *mycp; 701 struct aiocb *cb; 702 struct file *fp; 703 struct uio auio; 704 struct iovec aiov; 705 int cnt; 706 int error; 707 int oublock_st, oublock_end; 708 int inblock_st, inblock_end; 709 710 td = curthread; 711 td_savedcred = td->td_ucred; 712 td->td_ucred = aiocbe->cred; 713 mycp = td->td_proc; 714 cb = &aiocbe->uaiocb; 715 fp = aiocbe->fd_file; 716 717 aiov.iov_base = (void *)(uintptr_t)cb->aio_buf; 718 aiov.iov_len = cb->aio_nbytes; 719 720 auio.uio_iov = &aiov; 721 auio.uio_iovcnt = 1; 722 auio.uio_offset = cb->aio_offset; 723 auio.uio_resid = cb->aio_nbytes; 724 cnt = cb->aio_nbytes; 725 auio.uio_segflg = UIO_USERSPACE; 726 auio.uio_td = td; 727 728 inblock_st = mycp->p_stats->p_ru.ru_inblock; 729 oublock_st = mycp->p_stats->p_ru.ru_oublock; 730 /* 731 * _aio_aqueue() acquires a reference to the file that is 732 * released in aio_free_entry(). 733 */ 734 if (cb->aio_lio_opcode == LIO_READ) { 735 auio.uio_rw = UIO_READ; 736 error = fo_read(fp, &auio, fp->f_cred, FOF_OFFSET, td); 737 } else { 738 auio.uio_rw = UIO_WRITE; 739 error = fo_write(fp, &auio, fp->f_cred, FOF_OFFSET, td); 740 } 741 inblock_end = mycp->p_stats->p_ru.ru_inblock; 742 oublock_end = mycp->p_stats->p_ru.ru_oublock; 743 744 aiocbe->inputcharge = inblock_end - inblock_st; 745 aiocbe->outputcharge = oublock_end - oublock_st; 746 747 if ((error) && (auio.uio_resid != cnt)) { 748 if (error == ERESTART || error == EINTR || error == EWOULDBLOCK) 749 error = 0; 750 if ((error == EPIPE) && (cb->aio_lio_opcode == LIO_WRITE)) { 751 PROC_LOCK(aiocbe->userproc); 752 psignal(aiocbe->userproc, SIGPIPE); 753 PROC_UNLOCK(aiocbe->userproc); 754 } 755 } 756 757 cnt -= auio.uio_resid; 758 cb->_aiocb_private.error = error; 759 cb->_aiocb_private.status = cnt; 760 td->td_ucred = td_savedcred; 761 } 762 763 /* 764 * The AIO daemon, most of the actual work is done in aio_process, 765 * but the setup (and address space mgmt) is done in this routine. 766 */ 767 static void 768 aio_daemon(void *uproc) 769 { 770 int s; 771 struct aio_liojob *lj; 772 struct aiocb *cb; 773 struct aiocblist *aiocbe; 774 struct aiothreadlist *aiop; 775 struct kaioinfo *ki; 776 struct proc *curcp, *mycp, *userp; 777 struct vmspace *myvm, *tmpvm; 778 struct thread *td = curthread; 779 struct pgrp *newpgrp; 780 struct session *newsess; 781 782 mtx_lock(&Giant); 783 /* 784 * Local copies of curproc (cp) and vmspace (myvm) 785 */ 786 mycp = td->td_proc; 787 myvm = mycp->p_vmspace; 788 789 KASSERT(mycp->p_textvp == NULL, ("kthread has a textvp")); 790 791 /* 792 * Allocate and ready the aio control info. There is one aiop structure 793 * per daemon. 794 */ 795 aiop = uma_zalloc(aiop_zone, M_WAITOK); 796 aiop->aiothread = td; 797 aiop->aiothreadflags |= AIOP_FREE; 798 799 s = splnet(); 800 801 /* 802 * Place thread (lightweight process) onto the AIO free thread list. 803 */ 804 if (TAILQ_EMPTY(&aio_freeproc)) 805 wakeup(&aio_freeproc); 806 TAILQ_INSERT_HEAD(&aio_freeproc, aiop, list); 807 808 splx(s); 809 810 /* 811 * Get rid of our current filedescriptors. AIOD's don't need any 812 * filedescriptors, except as temporarily inherited from the client. 813 */ 814 fdfree(td); 815 816 mtx_unlock(&Giant); 817 /* The daemon resides in its own pgrp. */ 818 MALLOC(newpgrp, struct pgrp *, sizeof(struct pgrp), M_PGRP, 819 M_WAITOK | M_ZERO); 820 MALLOC(newsess, struct session *, sizeof(struct session), M_SESSION, 821 M_WAITOK | M_ZERO); 822 823 sx_xlock(&proctree_lock); 824 enterpgrp(mycp, mycp->p_pid, newpgrp, newsess); 825 sx_xunlock(&proctree_lock); 826 mtx_lock(&Giant); 827 828 /* 829 * Wakeup parent process. (Parent sleeps to keep from blasting away 830 * and creating too many daemons.) 831 */ 832 wakeup(mycp); 833 834 for (;;) { 835 /* 836 * curcp is the current daemon process context. 837 * userp is the current user process context. 838 */ 839 curcp = mycp; 840 841 /* 842 * Take daemon off of free queue 843 */ 844 if (aiop->aiothreadflags & AIOP_FREE) { 845 s = splnet(); 846 TAILQ_REMOVE(&aio_freeproc, aiop, list); 847 TAILQ_INSERT_TAIL(&aio_activeproc, aiop, list); 848 aiop->aiothreadflags &= ~AIOP_FREE; 849 splx(s); 850 } 851 aiop->aiothreadflags &= ~AIOP_SCHED; 852 853 /* 854 * Check for jobs. 855 */ 856 while ((aiocbe = aio_selectjob(aiop)) != NULL) { 857 cb = &aiocbe->uaiocb; 858 userp = aiocbe->userproc; 859 860 aiocbe->jobstate = JOBST_JOBRUNNING; 861 862 /* 863 * Connect to process address space for user program. 864 */ 865 if (userp != curcp) { 866 /* 867 * Save the current address space that we are 868 * connected to. 869 */ 870 tmpvm = mycp->p_vmspace; 871 872 /* 873 * Point to the new user address space, and 874 * refer to it. 875 */ 876 mycp->p_vmspace = userp->p_vmspace; 877 atomic_add_int(&mycp->p_vmspace->vm_refcnt, 1); 878 879 /* Activate the new mapping. */ 880 pmap_activate(FIRST_THREAD_IN_PROC(mycp)); 881 882 /* 883 * If the old address space wasn't the daemons 884 * own address space, then we need to remove the 885 * daemon's reference from the other process 886 * that it was acting on behalf of. 887 */ 888 if (tmpvm != myvm) { 889 vmspace_free(tmpvm); 890 } 891 curcp = userp; 892 } 893 894 ki = userp->p_aioinfo; 895 lj = aiocbe->lio; 896 897 /* Account for currently active jobs. */ 898 ki->kaio_active_count++; 899 900 /* Do the I/O function. */ 901 aio_process(aiocbe); 902 903 /* Decrement the active job count. */ 904 ki->kaio_active_count--; 905 906 /* 907 * Increment the completion count for wakeup/signal 908 * comparisons. 909 */ 910 aiocbe->jobflags |= AIOCBLIST_DONE; 911 ki->kaio_queue_finished_count++; 912 if (lj) 913 lj->lioj_queue_finished_count++; 914 if ((ki->kaio_flags & KAIO_WAKEUP) || ((ki->kaio_flags 915 & KAIO_RUNDOWN) && (ki->kaio_active_count == 0))) { 916 ki->kaio_flags &= ~KAIO_WAKEUP; 917 wakeup(userp); 918 } 919 920 s = splbio(); 921 if (lj && (lj->lioj_flags & 922 (LIOJ_SIGNAL|LIOJ_SIGNAL_POSTED)) == LIOJ_SIGNAL) { 923 if ((lj->lioj_queue_finished_count == 924 lj->lioj_queue_count) && 925 (lj->lioj_buffer_finished_count == 926 lj->lioj_buffer_count)) { 927 PROC_LOCK(userp); 928 psignal(userp, 929 lj->lioj_signal.sigev_signo); 930 PROC_UNLOCK(userp); 931 lj->lioj_flags |= LIOJ_SIGNAL_POSTED; 932 } 933 } 934 splx(s); 935 936 aiocbe->jobstate = JOBST_JOBFINISHED; 937 938 s = splnet(); 939 TAILQ_REMOVE(&ki->kaio_jobqueue, aiocbe, plist); 940 TAILQ_INSERT_TAIL(&ki->kaio_jobdone, aiocbe, plist); 941 splx(s); 942 KNOTE_UNLOCKED(&aiocbe->klist, 0); 943 944 if (aiocbe->jobflags & AIOCBLIST_RUNDOWN) { 945 wakeup(aiocbe); 946 aiocbe->jobflags &= ~AIOCBLIST_RUNDOWN; 947 } 948 949 if (cb->aio_sigevent.sigev_notify == SIGEV_SIGNAL) { 950 PROC_LOCK(userp); 951 psignal(userp, cb->aio_sigevent.sigev_signo); 952 PROC_UNLOCK(userp); 953 } 954 } 955 956 /* 957 * Disconnect from user address space. 958 */ 959 if (curcp != mycp) { 960 /* Get the user address space to disconnect from. */ 961 tmpvm = mycp->p_vmspace; 962 963 /* Get original address space for daemon. */ 964 mycp->p_vmspace = myvm; 965 966 /* Activate the daemon's address space. */ 967 pmap_activate(FIRST_THREAD_IN_PROC(mycp)); 968 #ifdef DIAGNOSTIC 969 if (tmpvm == myvm) { 970 printf("AIOD: vmspace problem -- %d\n", 971 mycp->p_pid); 972 } 973 #endif 974 /* Remove our vmspace reference. */ 975 vmspace_free(tmpvm); 976 977 curcp = mycp; 978 } 979 980 /* 981 * If we are the first to be put onto the free queue, wakeup 982 * anyone waiting for a daemon. 983 */ 984 s = splnet(); 985 TAILQ_REMOVE(&aio_activeproc, aiop, list); 986 if (TAILQ_EMPTY(&aio_freeproc)) 987 wakeup(&aio_freeproc); 988 TAILQ_INSERT_HEAD(&aio_freeproc, aiop, list); 989 aiop->aiothreadflags |= AIOP_FREE; 990 splx(s); 991 992 /* 993 * If daemon is inactive for a long time, allow it to exit, 994 * thereby freeing resources. 995 */ 996 if ((aiop->aiothreadflags & AIOP_SCHED) == 0 && 997 tsleep(aiop->aiothread, PRIBIO, "aiordy", aiod_lifetime)) { 998 s = splnet(); 999 if (TAILQ_EMPTY(&aio_jobs)) { 1000 if ((aiop->aiothreadflags & AIOP_FREE) && 1001 (num_aio_procs > target_aio_procs)) { 1002 TAILQ_REMOVE(&aio_freeproc, aiop, list); 1003 splx(s); 1004 uma_zfree(aiop_zone, aiop); 1005 num_aio_procs--; 1006 #ifdef DIAGNOSTIC 1007 if (mycp->p_vmspace->vm_refcnt <= 1) { 1008 printf("AIOD: bad vm refcnt for" 1009 " exiting daemon: %d\n", 1010 mycp->p_vmspace->vm_refcnt); 1011 } 1012 #endif 1013 kthread_exit(0); 1014 } 1015 } 1016 splx(s); 1017 } 1018 } 1019 } 1020 1021 /* 1022 * Create a new AIO daemon. This is mostly a kernel-thread fork routine. The 1023 * AIO daemon modifies its environment itself. 1024 */ 1025 static int 1026 aio_newproc(void) 1027 { 1028 int error; 1029 struct proc *p; 1030 1031 error = kthread_create(aio_daemon, curproc, &p, RFNOWAIT, 0, "aiod%d", 1032 num_aio_procs); 1033 if (error) 1034 return (error); 1035 1036 /* 1037 * Wait until daemon is started, but continue on just in case to 1038 * handle error conditions. 1039 */ 1040 error = tsleep(p, PZERO, "aiosta", aiod_timeout); 1041 1042 num_aio_procs++; 1043 1044 return (error); 1045 } 1046 1047 /* 1048 * Try the high-performance, low-overhead physio method for eligible 1049 * VCHR devices. This method doesn't use an aio helper thread, and 1050 * thus has very low overhead. 1051 * 1052 * Assumes that the caller, _aio_aqueue(), has incremented the file 1053 * structure's reference count, preventing its deallocation for the 1054 * duration of this call. 1055 */ 1056 static int 1057 aio_qphysio(struct proc *p, struct aiocblist *aiocbe) 1058 { 1059 int error; 1060 struct aiocb *cb; 1061 struct file *fp; 1062 struct buf *bp; 1063 struct vnode *vp; 1064 struct kaioinfo *ki; 1065 struct aio_liojob *lj; 1066 int s; 1067 int notify; 1068 1069 cb = &aiocbe->uaiocb; 1070 fp = aiocbe->fd_file; 1071 1072 if (fp->f_type != DTYPE_VNODE) 1073 return (-1); 1074 1075 vp = fp->f_vnode; 1076 1077 /* 1078 * If its not a disk, we don't want to return a positive error. 1079 * It causes the aio code to not fall through to try the thread 1080 * way when you're talking to a regular file. 1081 */ 1082 if (!vn_isdisk(vp, &error)) { 1083 if (error == ENOTBLK) 1084 return (-1); 1085 else 1086 return (error); 1087 } 1088 1089 if (cb->aio_nbytes % vp->v_rdev->si_bsize_phys) 1090 return (-1); 1091 1092 if (cb->aio_nbytes > 1093 MAXPHYS - (((vm_offset_t) cb->aio_buf) & PAGE_MASK)) 1094 return (-1); 1095 1096 ki = p->p_aioinfo; 1097 if (ki->kaio_buffer_count >= ki->kaio_ballowed_count) 1098 return (-1); 1099 1100 ki->kaio_buffer_count++; 1101 1102 lj = aiocbe->lio; 1103 if (lj) 1104 lj->lioj_buffer_count++; 1105 1106 /* Create and build a buffer header for a transfer. */ 1107 bp = (struct buf *)getpbuf(NULL); 1108 BUF_KERNPROC(bp); 1109 1110 /* 1111 * Get a copy of the kva from the physical buffer. 1112 */ 1113 bp->b_dev = vp->v_rdev; 1114 error = 0; 1115 1116 bp->b_bcount = cb->aio_nbytes; 1117 bp->b_bufsize = cb->aio_nbytes; 1118 bp->b_iodone = aio_physwakeup; 1119 bp->b_saveaddr = bp->b_data; 1120 bp->b_data = (void *)(uintptr_t)cb->aio_buf; 1121 bp->b_offset = cb->aio_offset; 1122 bp->b_iooffset = cb->aio_offset; 1123 bp->b_blkno = btodb(cb->aio_offset); 1124 bp->b_iocmd = cb->aio_lio_opcode == LIO_WRITE ? BIO_WRITE : BIO_READ; 1125 1126 /* 1127 * Bring buffer into kernel space. 1128 */ 1129 if (vmapbuf(bp) < 0) { 1130 error = EFAULT; 1131 goto doerror; 1132 } 1133 1134 s = splbio(); 1135 aiocbe->bp = bp; 1136 bp->b_caller1 = (void *)aiocbe; 1137 TAILQ_INSERT_TAIL(&aio_bufjobs, aiocbe, list); 1138 TAILQ_INSERT_TAIL(&ki->kaio_bufqueue, aiocbe, plist); 1139 aiocbe->jobstate = JOBST_JOBQBUF; 1140 cb->_aiocb_private.status = cb->aio_nbytes; 1141 num_buf_aio++; 1142 bp->b_error = 0; 1143 1144 splx(s); 1145 1146 /* Perform transfer. */ 1147 DEV_STRATEGY(bp); 1148 1149 notify = 0; 1150 s = splbio(); 1151 1152 /* 1153 * If we had an error invoking the request, or an error in processing 1154 * the request before we have returned, we process it as an error in 1155 * transfer. Note that such an I/O error is not indicated immediately, 1156 * but is returned using the aio_error mechanism. In this case, 1157 * aio_suspend will return immediately. 1158 */ 1159 if (bp->b_error || (bp->b_ioflags & BIO_ERROR)) { 1160 struct aiocb *job = aiocbe->uuaiocb; 1161 1162 aiocbe->uaiocb._aiocb_private.status = 0; 1163 suword(&job->_aiocb_private.status, 0); 1164 aiocbe->uaiocb._aiocb_private.error = bp->b_error; 1165 suword(&job->_aiocb_private.error, bp->b_error); 1166 1167 ki->kaio_buffer_finished_count++; 1168 1169 if (aiocbe->jobstate != JOBST_JOBBFINISHED) { 1170 aiocbe->jobstate = JOBST_JOBBFINISHED; 1171 aiocbe->jobflags |= AIOCBLIST_DONE; 1172 TAILQ_REMOVE(&aio_bufjobs, aiocbe, list); 1173 TAILQ_REMOVE(&ki->kaio_bufqueue, aiocbe, plist); 1174 TAILQ_INSERT_TAIL(&ki->kaio_bufdone, aiocbe, plist); 1175 notify = 1; 1176 } 1177 } 1178 splx(s); 1179 if (notify) 1180 KNOTE_UNLOCKED(&aiocbe->klist, 0); 1181 return (0); 1182 1183 doerror: 1184 ki->kaio_buffer_count--; 1185 if (lj) 1186 lj->lioj_buffer_count--; 1187 aiocbe->bp = NULL; 1188 relpbuf(bp, NULL); 1189 return (error); 1190 } 1191 1192 /* 1193 * This waits/tests physio completion. 1194 */ 1195 static int 1196 aio_fphysio(struct aiocblist *iocb) 1197 { 1198 int s; 1199 struct buf *bp; 1200 int error; 1201 1202 bp = iocb->bp; 1203 1204 s = splbio(); 1205 while ((bp->b_flags & B_DONE) == 0) { 1206 if (tsleep(bp, PRIBIO, "physstr", aiod_timeout)) { 1207 if ((bp->b_flags & B_DONE) == 0) { 1208 splx(s); 1209 return (EINPROGRESS); 1210 } else 1211 break; 1212 } 1213 } 1214 splx(s); 1215 1216 /* Release mapping into kernel space. */ 1217 vunmapbuf(bp); 1218 iocb->bp = 0; 1219 1220 error = 0; 1221 1222 /* Check for an error. */ 1223 if (bp->b_ioflags & BIO_ERROR) 1224 error = bp->b_error; 1225 1226 relpbuf(bp, NULL); 1227 return (error); 1228 } 1229 1230 /* 1231 * Wake up aio requests that may be serviceable now. 1232 */ 1233 static void 1234 aio_swake_cb(struct socket *so, struct sockbuf *sb) 1235 { 1236 struct aiocblist *cb,*cbn; 1237 struct proc *p; 1238 struct kaioinfo *ki = NULL; 1239 int opcode, wakecount = 0; 1240 struct aiothreadlist *aiop; 1241 1242 if (sb == &so->so_snd) { 1243 opcode = LIO_WRITE; 1244 SOCKBUF_LOCK(&so->so_snd); 1245 so->so_snd.sb_flags &= ~SB_AIO; 1246 SOCKBUF_UNLOCK(&so->so_snd); 1247 } else { 1248 opcode = LIO_READ; 1249 SOCKBUF_LOCK(&so->so_rcv); 1250 so->so_rcv.sb_flags &= ~SB_AIO; 1251 SOCKBUF_UNLOCK(&so->so_rcv); 1252 } 1253 1254 for (cb = TAILQ_FIRST(&so->so_aiojobq); cb; cb = cbn) { 1255 cbn = TAILQ_NEXT(cb, list); 1256 if (opcode == cb->uaiocb.aio_lio_opcode) { 1257 p = cb->userproc; 1258 ki = p->p_aioinfo; 1259 TAILQ_REMOVE(&so->so_aiojobq, cb, list); 1260 TAILQ_REMOVE(&ki->kaio_sockqueue, cb, plist); 1261 TAILQ_INSERT_TAIL(&aio_jobs, cb, list); 1262 TAILQ_INSERT_TAIL(&ki->kaio_jobqueue, cb, plist); 1263 wakecount++; 1264 if (cb->jobstate != JOBST_JOBQGLOBAL) 1265 panic("invalid queue value"); 1266 } 1267 } 1268 1269 while (wakecount--) { 1270 if ((aiop = TAILQ_FIRST(&aio_freeproc)) != 0) { 1271 TAILQ_REMOVE(&aio_freeproc, aiop, list); 1272 TAILQ_INSERT_TAIL(&aio_activeproc, aiop, list); 1273 aiop->aiothreadflags &= ~AIOP_FREE; 1274 wakeup(aiop->aiothread); 1275 } 1276 } 1277 } 1278 1279 /* 1280 * Queue a new AIO request. Choosing either the threaded or direct physio VCHR 1281 * technique is done in this code. 1282 */ 1283 static int 1284 _aio_aqueue(struct thread *td, struct aiocb *job, struct aio_liojob *lj, int type) 1285 { 1286 struct proc *p = td->td_proc; 1287 struct filedesc *fdp; 1288 struct file *fp; 1289 unsigned int fd; 1290 struct socket *so; 1291 int s; 1292 int error; 1293 int opcode, user_opcode; 1294 struct aiocblist *aiocbe; 1295 struct aiothreadlist *aiop; 1296 struct kaioinfo *ki; 1297 struct kevent kev; 1298 struct kqueue *kq; 1299 struct file *kq_fp; 1300 1301 aiocbe = uma_zalloc(aiocb_zone, M_WAITOK); 1302 aiocbe->inputcharge = 0; 1303 aiocbe->outputcharge = 0; 1304 callout_handle_init(&aiocbe->timeouthandle); 1305 /* XXX - need a lock */ 1306 knlist_init(&aiocbe->klist, NULL); 1307 1308 suword(&job->_aiocb_private.status, -1); 1309 suword(&job->_aiocb_private.error, 0); 1310 suword(&job->_aiocb_private.kernelinfo, -1); 1311 1312 error = copyin(job, &aiocbe->uaiocb, sizeof(aiocbe->uaiocb)); 1313 if (error) { 1314 suword(&job->_aiocb_private.error, error); 1315 uma_zfree(aiocb_zone, aiocbe); 1316 return (error); 1317 } 1318 if (aiocbe->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL && 1319 !_SIG_VALID(aiocbe->uaiocb.aio_sigevent.sigev_signo)) { 1320 uma_zfree(aiocb_zone, aiocbe); 1321 return (EINVAL); 1322 } 1323 1324 /* Save userspace address of the job info. */ 1325 aiocbe->uuaiocb = job; 1326 1327 /* Get the opcode. */ 1328 user_opcode = aiocbe->uaiocb.aio_lio_opcode; 1329 if (type != LIO_NOP) 1330 aiocbe->uaiocb.aio_lio_opcode = type; 1331 opcode = aiocbe->uaiocb.aio_lio_opcode; 1332 1333 /* Get the fd info for process. */ 1334 fdp = p->p_fd; 1335 1336 /* 1337 * Range check file descriptor. 1338 */ 1339 FILEDESC_LOCK(fdp); 1340 fd = aiocbe->uaiocb.aio_fildes; 1341 if (fd >= fdp->fd_nfiles) { 1342 FILEDESC_UNLOCK(fdp); 1343 uma_zfree(aiocb_zone, aiocbe); 1344 if (type == 0) 1345 suword(&job->_aiocb_private.error, EBADF); 1346 return (EBADF); 1347 } 1348 1349 fp = aiocbe->fd_file = fdp->fd_ofiles[fd]; 1350 if ((fp == NULL) || 1351 ((opcode == LIO_WRITE) && ((fp->f_flag & FWRITE) == 0)) || 1352 ((opcode == LIO_READ) && ((fp->f_flag & FREAD) == 0))) { 1353 FILEDESC_UNLOCK(fdp); 1354 uma_zfree(aiocb_zone, aiocbe); 1355 if (type == 0) 1356 suword(&job->_aiocb_private.error, EBADF); 1357 return (EBADF); 1358 } 1359 fhold(fp); 1360 FILEDESC_UNLOCK(fdp); 1361 1362 if (aiocbe->uaiocb.aio_offset == -1LL) { 1363 error = EINVAL; 1364 goto aqueue_fail; 1365 } 1366 error = suword(&job->_aiocb_private.kernelinfo, jobrefid); 1367 if (error) { 1368 error = EINVAL; 1369 goto aqueue_fail; 1370 } 1371 aiocbe->uaiocb._aiocb_private.kernelinfo = (void *)(intptr_t)jobrefid; 1372 if (jobrefid == LONG_MAX) 1373 jobrefid = 1; 1374 else 1375 jobrefid++; 1376 1377 if (opcode == LIO_NOP) { 1378 fdrop(fp, td); 1379 uma_zfree(aiocb_zone, aiocbe); 1380 if (type == 0) { 1381 suword(&job->_aiocb_private.error, 0); 1382 suword(&job->_aiocb_private.status, 0); 1383 suword(&job->_aiocb_private.kernelinfo, 0); 1384 } 1385 return (0); 1386 } 1387 if ((opcode != LIO_READ) && (opcode != LIO_WRITE)) { 1388 if (type == 0) 1389 suword(&job->_aiocb_private.status, 0); 1390 error = EINVAL; 1391 goto aqueue_fail; 1392 } 1393 1394 if (aiocbe->uaiocb.aio_sigevent.sigev_notify == SIGEV_KEVENT) { 1395 kev.ident = aiocbe->uaiocb.aio_sigevent.sigev_notify_kqueue; 1396 kev.udata = aiocbe->uaiocb.aio_sigevent.sigev_value.sigval_ptr; 1397 } 1398 else { 1399 /* 1400 * This method for requesting kevent-based notification won't 1401 * work on the alpha, since we're passing in a pointer 1402 * via aio_lio_opcode, which is an int. Use the SIGEV_KEVENT- 1403 * based method instead. 1404 */ 1405 if (user_opcode == LIO_NOP || user_opcode == LIO_READ || 1406 user_opcode == LIO_WRITE) 1407 goto no_kqueue; 1408 1409 error = copyin((struct kevent *)(uintptr_t)user_opcode, 1410 &kev, sizeof(kev)); 1411 if (error) 1412 goto aqueue_fail; 1413 } 1414 if ((u_int)kev.ident >= fdp->fd_nfiles || 1415 (kq_fp = fdp->fd_ofiles[kev.ident]) == NULL || 1416 (kq_fp->f_type != DTYPE_KQUEUE)) { 1417 error = EBADF; 1418 goto aqueue_fail; 1419 } 1420 kq = kq_fp->f_data; 1421 kev.ident = (uintptr_t)aiocbe->uuaiocb; 1422 kev.filter = EVFILT_AIO; 1423 kev.flags = EV_ADD | EV_ENABLE | EV_FLAG1; 1424 kev.data = (intptr_t)aiocbe; 1425 error = kqueue_register(kq, &kev, td, 1); 1426 aqueue_fail: 1427 if (error) { 1428 fdrop(fp, td); 1429 uma_zfree(aiocb_zone, aiocbe); 1430 if (type == 0) 1431 suword(&job->_aiocb_private.error, error); 1432 goto done; 1433 } 1434 no_kqueue: 1435 1436 suword(&job->_aiocb_private.error, EINPROGRESS); 1437 aiocbe->uaiocb._aiocb_private.error = EINPROGRESS; 1438 aiocbe->userproc = p; 1439 aiocbe->cred = crhold(td->td_ucred); 1440 aiocbe->jobflags = 0; 1441 aiocbe->lio = lj; 1442 ki = p->p_aioinfo; 1443 1444 if (fp->f_type == DTYPE_SOCKET) { 1445 /* 1446 * Alternate queueing for socket ops: Reach down into the 1447 * descriptor to get the socket data. Then check to see if the 1448 * socket is ready to be read or written (based on the requested 1449 * operation). 1450 * 1451 * If it is not ready for io, then queue the aiocbe on the 1452 * socket, and set the flags so we get a call when sbnotify() 1453 * happens. 1454 */ 1455 so = fp->f_data; 1456 s = splnet(); 1457 if (((opcode == LIO_READ) && (!soreadable(so))) || ((opcode == 1458 LIO_WRITE) && (!sowriteable(so)))) { 1459 TAILQ_INSERT_TAIL(&so->so_aiojobq, aiocbe, list); 1460 TAILQ_INSERT_TAIL(&ki->kaio_sockqueue, aiocbe, plist); 1461 if (opcode == LIO_READ) { 1462 SOCKBUF_LOCK(&so->so_rcv); 1463 so->so_rcv.sb_flags |= SB_AIO; 1464 SOCKBUF_UNLOCK(&so->so_rcv); 1465 } else { 1466 SOCKBUF_LOCK(&so->so_snd); 1467 so->so_snd.sb_flags |= SB_AIO; 1468 SOCKBUF_UNLOCK(&so->so_snd); 1469 } 1470 aiocbe->jobstate = JOBST_JOBQGLOBAL; /* XXX */ 1471 ki->kaio_queue_count++; 1472 num_queue_count++; 1473 splx(s); 1474 error = 0; 1475 goto done; 1476 } 1477 splx(s); 1478 } 1479 1480 if ((error = aio_qphysio(p, aiocbe)) == 0) 1481 goto done; 1482 if (error > 0) { 1483 suword(&job->_aiocb_private.status, 0); 1484 aiocbe->uaiocb._aiocb_private.error = error; 1485 suword(&job->_aiocb_private.error, error); 1486 goto done; 1487 } 1488 1489 /* No buffer for daemon I/O. */ 1490 aiocbe->bp = NULL; 1491 1492 ki->kaio_queue_count++; 1493 if (lj) 1494 lj->lioj_queue_count++; 1495 s = splnet(); 1496 TAILQ_INSERT_TAIL(&ki->kaio_jobqueue, aiocbe, plist); 1497 TAILQ_INSERT_TAIL(&aio_jobs, aiocbe, list); 1498 splx(s); 1499 aiocbe->jobstate = JOBST_JOBQGLOBAL; 1500 1501 num_queue_count++; 1502 error = 0; 1503 1504 /* 1505 * If we don't have a free AIO process, and we are below our quota, then 1506 * start one. Otherwise, depend on the subsequent I/O completions to 1507 * pick-up this job. If we don't sucessfully create the new process 1508 * (thread) due to resource issues, we return an error for now (EAGAIN), 1509 * which is likely not the correct thing to do. 1510 */ 1511 s = splnet(); 1512 retryproc: 1513 if ((aiop = TAILQ_FIRST(&aio_freeproc)) != NULL) { 1514 TAILQ_REMOVE(&aio_freeproc, aiop, list); 1515 TAILQ_INSERT_TAIL(&aio_activeproc, aiop, list); 1516 aiop->aiothreadflags &= ~AIOP_FREE; 1517 wakeup(aiop->aiothread); 1518 } else if (((num_aio_resv_start + num_aio_procs) < max_aio_procs) && 1519 ((ki->kaio_active_count + num_aio_resv_start) < 1520 ki->kaio_maxactive_count)) { 1521 num_aio_resv_start++; 1522 if ((error = aio_newproc()) == 0) { 1523 num_aio_resv_start--; 1524 goto retryproc; 1525 } 1526 num_aio_resv_start--; 1527 } 1528 splx(s); 1529 done: 1530 return (error); 1531 } 1532 1533 /* 1534 * This routine queues an AIO request, checking for quotas. 1535 */ 1536 static int 1537 aio_aqueue(struct thread *td, struct aiocb *job, int type) 1538 { 1539 struct proc *p = td->td_proc; 1540 struct kaioinfo *ki; 1541 1542 if (p->p_aioinfo == NULL) 1543 aio_init_aioinfo(p); 1544 1545 if (num_queue_count >= max_queue_count) 1546 return (EAGAIN); 1547 1548 ki = p->p_aioinfo; 1549 if (ki->kaio_queue_count >= ki->kaio_qallowed_count) 1550 return (EAGAIN); 1551 1552 return _aio_aqueue(td, job, NULL, type); 1553 } 1554 1555 /* 1556 * Support the aio_return system call, as a side-effect, kernel resources are 1557 * released. 1558 */ 1559 int 1560 aio_return(struct thread *td, struct aio_return_args *uap) 1561 { 1562 struct proc *p = td->td_proc; 1563 int s; 1564 long jobref; 1565 struct aiocblist *cb, *ncb; 1566 struct aiocb *ujob; 1567 struct kaioinfo *ki; 1568 1569 ujob = uap->aiocbp; 1570 jobref = fuword(&ujob->_aiocb_private.kernelinfo); 1571 if (jobref == -1 || jobref == 0) 1572 return (EINVAL); 1573 1574 ki = p->p_aioinfo; 1575 if (ki == NULL) 1576 return (EINVAL); 1577 TAILQ_FOREACH(cb, &ki->kaio_jobdone, plist) { 1578 if (((intptr_t) cb->uaiocb._aiocb_private.kernelinfo) == 1579 jobref) { 1580 if (cb->uaiocb.aio_lio_opcode == LIO_WRITE) { 1581 p->p_stats->p_ru.ru_oublock += 1582 cb->outputcharge; 1583 cb->outputcharge = 0; 1584 } else if (cb->uaiocb.aio_lio_opcode == LIO_READ) { 1585 p->p_stats->p_ru.ru_inblock += cb->inputcharge; 1586 cb->inputcharge = 0; 1587 } 1588 goto done; 1589 } 1590 } 1591 s = splbio(); 1592 for (cb = TAILQ_FIRST(&ki->kaio_bufdone); cb; cb = ncb) { 1593 ncb = TAILQ_NEXT(cb, plist); 1594 if (((intptr_t) cb->uaiocb._aiocb_private.kernelinfo) 1595 == jobref) { 1596 break; 1597 } 1598 } 1599 splx(s); 1600 done: 1601 if (cb != NULL) { 1602 if (ujob == cb->uuaiocb) { 1603 td->td_retval[0] = 1604 cb->uaiocb._aiocb_private.status; 1605 } else 1606 td->td_retval[0] = EFAULT; 1607 aio_free_entry(cb); 1608 return (0); 1609 } 1610 return (EINVAL); 1611 } 1612 1613 /* 1614 * Allow a process to wakeup when any of the I/O requests are completed. 1615 */ 1616 int 1617 aio_suspend(struct thread *td, struct aio_suspend_args *uap) 1618 { 1619 struct proc *p = td->td_proc; 1620 struct timeval atv; 1621 struct timespec ts; 1622 struct aiocb *const *cbptr, *cbp; 1623 struct kaioinfo *ki; 1624 struct aiocblist *cb; 1625 int i; 1626 int njoblist; 1627 int error, s, timo; 1628 long *ijoblist; 1629 struct aiocb **ujoblist; 1630 1631 if (uap->nent < 0 || uap->nent > AIO_LISTIO_MAX) 1632 return (EINVAL); 1633 1634 timo = 0; 1635 if (uap->timeout) { 1636 /* Get timespec struct. */ 1637 if ((error = copyin(uap->timeout, &ts, sizeof(ts))) != 0) 1638 return (error); 1639 1640 if (ts.tv_nsec < 0 || ts.tv_nsec >= 1000000000) 1641 return (EINVAL); 1642 1643 TIMESPEC_TO_TIMEVAL(&atv, &ts); 1644 if (itimerfix(&atv)) 1645 return (EINVAL); 1646 timo = tvtohz(&atv); 1647 } 1648 1649 ki = p->p_aioinfo; 1650 if (ki == NULL) 1651 return (EAGAIN); 1652 1653 njoblist = 0; 1654 ijoblist = uma_zalloc(aiol_zone, M_WAITOK); 1655 ujoblist = uma_zalloc(aiol_zone, M_WAITOK); 1656 cbptr = uap->aiocbp; 1657 1658 for (i = 0; i < uap->nent; i++) { 1659 cbp = (struct aiocb *)(intptr_t)fuword(&cbptr[i]); 1660 if (cbp == 0) 1661 continue; 1662 ujoblist[njoblist] = cbp; 1663 ijoblist[njoblist] = fuword(&cbp->_aiocb_private.kernelinfo); 1664 njoblist++; 1665 } 1666 1667 if (njoblist == 0) { 1668 uma_zfree(aiol_zone, ijoblist); 1669 uma_zfree(aiol_zone, ujoblist); 1670 return (0); 1671 } 1672 1673 error = 0; 1674 for (;;) { 1675 TAILQ_FOREACH(cb, &ki->kaio_jobdone, plist) { 1676 for (i = 0; i < njoblist; i++) { 1677 if (((intptr_t) 1678 cb->uaiocb._aiocb_private.kernelinfo) == 1679 ijoblist[i]) { 1680 if (ujoblist[i] != cb->uuaiocb) 1681 error = EINVAL; 1682 uma_zfree(aiol_zone, ijoblist); 1683 uma_zfree(aiol_zone, ujoblist); 1684 return (error); 1685 } 1686 } 1687 } 1688 1689 s = splbio(); 1690 for (cb = TAILQ_FIRST(&ki->kaio_bufdone); cb; cb = 1691 TAILQ_NEXT(cb, plist)) { 1692 for (i = 0; i < njoblist; i++) { 1693 if (((intptr_t) 1694 cb->uaiocb._aiocb_private.kernelinfo) == 1695 ijoblist[i]) { 1696 splx(s); 1697 if (ujoblist[i] != cb->uuaiocb) 1698 error = EINVAL; 1699 uma_zfree(aiol_zone, ijoblist); 1700 uma_zfree(aiol_zone, ujoblist); 1701 return (error); 1702 } 1703 } 1704 } 1705 1706 ki->kaio_flags |= KAIO_WAKEUP; 1707 error = tsleep(p, PRIBIO | PCATCH, "aiospn", timo); 1708 splx(s); 1709 1710 if (error == ERESTART || error == EINTR) { 1711 uma_zfree(aiol_zone, ijoblist); 1712 uma_zfree(aiol_zone, ujoblist); 1713 return (EINTR); 1714 } else if (error == EWOULDBLOCK) { 1715 uma_zfree(aiol_zone, ijoblist); 1716 uma_zfree(aiol_zone, ujoblist); 1717 return (EAGAIN); 1718 } 1719 } 1720 1721 /* NOTREACHED */ 1722 return (EINVAL); 1723 } 1724 1725 /* 1726 * aio_cancel cancels any non-physio aio operations not currently in 1727 * progress. 1728 */ 1729 int 1730 aio_cancel(struct thread *td, struct aio_cancel_args *uap) 1731 { 1732 struct proc *p = td->td_proc; 1733 struct kaioinfo *ki; 1734 struct aiocblist *cbe, *cbn; 1735 struct file *fp; 1736 struct filedesc *fdp; 1737 struct socket *so; 1738 struct proc *po; 1739 int s,error; 1740 int cancelled=0; 1741 int notcancelled=0; 1742 struct vnode *vp; 1743 1744 fdp = p->p_fd; 1745 if ((u_int)uap->fd >= fdp->fd_nfiles || 1746 (fp = fdp->fd_ofiles[uap->fd]) == NULL) 1747 return (EBADF); 1748 1749 if (fp->f_type == DTYPE_VNODE) { 1750 vp = fp->f_vnode; 1751 1752 if (vn_isdisk(vp,&error)) { 1753 td->td_retval[0] = AIO_NOTCANCELED; 1754 return (0); 1755 } 1756 } else if (fp->f_type == DTYPE_SOCKET) { 1757 so = fp->f_data; 1758 1759 s = splnet(); 1760 1761 for (cbe = TAILQ_FIRST(&so->so_aiojobq); cbe; cbe = cbn) { 1762 cbn = TAILQ_NEXT(cbe, list); 1763 if ((uap->aiocbp == NULL) || 1764 (uap->aiocbp == cbe->uuaiocb) ) { 1765 po = cbe->userproc; 1766 ki = po->p_aioinfo; 1767 TAILQ_REMOVE(&so->so_aiojobq, cbe, list); 1768 TAILQ_REMOVE(&ki->kaio_sockqueue, cbe, plist); 1769 TAILQ_INSERT_TAIL(&ki->kaio_jobdone, cbe, plist); 1770 if (ki->kaio_flags & KAIO_WAKEUP) { 1771 wakeup(po); 1772 } 1773 cbe->jobstate = JOBST_JOBFINISHED; 1774 cbe->uaiocb._aiocb_private.status=-1; 1775 cbe->uaiocb._aiocb_private.error=ECANCELED; 1776 cancelled++; 1777 /* XXX cancelled, knote? */ 1778 if (cbe->uaiocb.aio_sigevent.sigev_notify == 1779 SIGEV_SIGNAL) { 1780 PROC_LOCK(cbe->userproc); 1781 psignal(cbe->userproc, cbe->uaiocb.aio_sigevent.sigev_signo); 1782 PROC_UNLOCK(cbe->userproc); 1783 } 1784 if (uap->aiocbp) 1785 break; 1786 } 1787 } 1788 splx(s); 1789 1790 if ((cancelled) && (uap->aiocbp)) { 1791 td->td_retval[0] = AIO_CANCELED; 1792 return (0); 1793 } 1794 } 1795 ki=p->p_aioinfo; 1796 if (ki == NULL) 1797 goto done; 1798 s = splnet(); 1799 1800 for (cbe = TAILQ_FIRST(&ki->kaio_jobqueue); cbe; cbe = cbn) { 1801 cbn = TAILQ_NEXT(cbe, plist); 1802 1803 if ((uap->fd == cbe->uaiocb.aio_fildes) && 1804 ((uap->aiocbp == NULL ) || 1805 (uap->aiocbp == cbe->uuaiocb))) { 1806 1807 if (cbe->jobstate == JOBST_JOBQGLOBAL) { 1808 TAILQ_REMOVE(&aio_jobs, cbe, list); 1809 TAILQ_REMOVE(&ki->kaio_jobqueue, cbe, plist); 1810 TAILQ_INSERT_TAIL(&ki->kaio_jobdone, cbe, 1811 plist); 1812 cancelled++; 1813 ki->kaio_queue_finished_count++; 1814 cbe->jobstate = JOBST_JOBFINISHED; 1815 cbe->uaiocb._aiocb_private.status = -1; 1816 cbe->uaiocb._aiocb_private.error = ECANCELED; 1817 /* XXX cancelled, knote? */ 1818 if (cbe->uaiocb.aio_sigevent.sigev_notify == 1819 SIGEV_SIGNAL) { 1820 PROC_LOCK(cbe->userproc); 1821 psignal(cbe->userproc, cbe->uaiocb.aio_sigevent.sigev_signo); 1822 PROC_UNLOCK(cbe->userproc); 1823 } 1824 } else { 1825 notcancelled++; 1826 } 1827 } 1828 } 1829 splx(s); 1830 done: 1831 if (notcancelled) { 1832 td->td_retval[0] = AIO_NOTCANCELED; 1833 return (0); 1834 } 1835 if (cancelled) { 1836 td->td_retval[0] = AIO_CANCELED; 1837 return (0); 1838 } 1839 td->td_retval[0] = AIO_ALLDONE; 1840 1841 return (0); 1842 } 1843 1844 /* 1845 * aio_error is implemented in the kernel level for compatibility purposes only. 1846 * For a user mode async implementation, it would be best to do it in a userland 1847 * subroutine. 1848 */ 1849 int 1850 aio_error(struct thread *td, struct aio_error_args *uap) 1851 { 1852 struct proc *p = td->td_proc; 1853 int s; 1854 struct aiocblist *cb; 1855 struct kaioinfo *ki; 1856 long jobref; 1857 1858 ki = p->p_aioinfo; 1859 if (ki == NULL) 1860 return (EINVAL); 1861 1862 jobref = fuword(&uap->aiocbp->_aiocb_private.kernelinfo); 1863 if ((jobref == -1) || (jobref == 0)) 1864 return (EINVAL); 1865 1866 TAILQ_FOREACH(cb, &ki->kaio_jobdone, plist) { 1867 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) == 1868 jobref) { 1869 td->td_retval[0] = cb->uaiocb._aiocb_private.error; 1870 return (0); 1871 } 1872 } 1873 1874 s = splnet(); 1875 1876 for (cb = TAILQ_FIRST(&ki->kaio_jobqueue); cb; cb = TAILQ_NEXT(cb, 1877 plist)) { 1878 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) == 1879 jobref) { 1880 td->td_retval[0] = EINPROGRESS; 1881 splx(s); 1882 return (0); 1883 } 1884 } 1885 1886 for (cb = TAILQ_FIRST(&ki->kaio_sockqueue); cb; cb = TAILQ_NEXT(cb, 1887 plist)) { 1888 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) == 1889 jobref) { 1890 td->td_retval[0] = EINPROGRESS; 1891 splx(s); 1892 return (0); 1893 } 1894 } 1895 splx(s); 1896 1897 s = splbio(); 1898 for (cb = TAILQ_FIRST(&ki->kaio_bufdone); cb; cb = TAILQ_NEXT(cb, 1899 plist)) { 1900 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) == 1901 jobref) { 1902 td->td_retval[0] = cb->uaiocb._aiocb_private.error; 1903 splx(s); 1904 return (0); 1905 } 1906 } 1907 1908 for (cb = TAILQ_FIRST(&ki->kaio_bufqueue); cb; cb = TAILQ_NEXT(cb, 1909 plist)) { 1910 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) == 1911 jobref) { 1912 td->td_retval[0] = EINPROGRESS; 1913 splx(s); 1914 return (0); 1915 } 1916 } 1917 splx(s); 1918 1919 #if (0) 1920 /* 1921 * Hack for lio. 1922 */ 1923 status = fuword(&uap->aiocbp->_aiocb_private.status); 1924 if (status == -1) 1925 return fuword(&uap->aiocbp->_aiocb_private.error); 1926 #endif 1927 return (EINVAL); 1928 } 1929 1930 /* syscall - asynchronous read from a file (REALTIME) */ 1931 int 1932 aio_read(struct thread *td, struct aio_read_args *uap) 1933 { 1934 1935 return aio_aqueue(td, uap->aiocbp, LIO_READ); 1936 } 1937 1938 /* syscall - asynchronous write to a file (REALTIME) */ 1939 int 1940 aio_write(struct thread *td, struct aio_write_args *uap) 1941 { 1942 1943 return aio_aqueue(td, uap->aiocbp, LIO_WRITE); 1944 } 1945 1946 /* syscall - list directed I/O (REALTIME) */ 1947 int 1948 lio_listio(struct thread *td, struct lio_listio_args *uap) 1949 { 1950 struct proc *p = td->td_proc; 1951 int nent, nentqueued; 1952 struct aiocb *iocb, * const *cbptr; 1953 struct aiocblist *cb; 1954 struct kaioinfo *ki; 1955 struct aio_liojob *lj; 1956 int error, runningcode; 1957 int nerror; 1958 int i; 1959 int s; 1960 1961 if ((uap->mode != LIO_NOWAIT) && (uap->mode != LIO_WAIT)) 1962 return (EINVAL); 1963 1964 nent = uap->nent; 1965 if (nent < 0 || nent > AIO_LISTIO_MAX) 1966 return (EINVAL); 1967 1968 if (p->p_aioinfo == NULL) 1969 aio_init_aioinfo(p); 1970 1971 if ((nent + num_queue_count) > max_queue_count) 1972 return (EAGAIN); 1973 1974 ki = p->p_aioinfo; 1975 if ((nent + ki->kaio_queue_count) > ki->kaio_qallowed_count) 1976 return (EAGAIN); 1977 1978 lj = uma_zalloc(aiolio_zone, M_WAITOK); 1979 if (!lj) 1980 return (EAGAIN); 1981 1982 lj->lioj_flags = 0; 1983 lj->lioj_buffer_count = 0; 1984 lj->lioj_buffer_finished_count = 0; 1985 lj->lioj_queue_count = 0; 1986 lj->lioj_queue_finished_count = 0; 1987 lj->lioj_ki = ki; 1988 1989 /* 1990 * Setup signal. 1991 */ 1992 if (uap->sig && (uap->mode == LIO_NOWAIT)) { 1993 error = copyin(uap->sig, &lj->lioj_signal, 1994 sizeof(lj->lioj_signal)); 1995 if (error) { 1996 uma_zfree(aiolio_zone, lj); 1997 return (error); 1998 } 1999 if (!_SIG_VALID(lj->lioj_signal.sigev_signo)) { 2000 uma_zfree(aiolio_zone, lj); 2001 return (EINVAL); 2002 } 2003 lj->lioj_flags |= LIOJ_SIGNAL; 2004 } 2005 TAILQ_INSERT_TAIL(&ki->kaio_liojoblist, lj, lioj_list); 2006 /* 2007 * Get pointers to the list of I/O requests. 2008 */ 2009 nerror = 0; 2010 nentqueued = 0; 2011 cbptr = uap->acb_list; 2012 for (i = 0; i < uap->nent; i++) { 2013 iocb = (struct aiocb *)(intptr_t)fuword(&cbptr[i]); 2014 if (((intptr_t)iocb != -1) && ((intptr_t)iocb != 0)) { 2015 error = _aio_aqueue(td, iocb, lj, 0); 2016 if (error == 0) 2017 nentqueued++; 2018 else 2019 nerror++; 2020 } 2021 } 2022 2023 /* 2024 * If we haven't queued any, then just return error. 2025 */ 2026 if (nentqueued == 0) 2027 return (0); 2028 2029 /* 2030 * Calculate the appropriate error return. 2031 */ 2032 runningcode = 0; 2033 if (nerror) 2034 runningcode = EIO; 2035 2036 if (uap->mode == LIO_WAIT) { 2037 int command, found, jobref; 2038 2039 for (;;) { 2040 found = 0; 2041 for (i = 0; i < uap->nent; i++) { 2042 /* 2043 * Fetch address of the control buf pointer in 2044 * user space. 2045 */ 2046 iocb = (struct aiocb *) 2047 (intptr_t)fuword(&cbptr[i]); 2048 if (((intptr_t)iocb == -1) || ((intptr_t)iocb 2049 == 0)) 2050 continue; 2051 2052 /* 2053 * Fetch the associated command from user space. 2054 */ 2055 command = fuword(&iocb->aio_lio_opcode); 2056 if (command == LIO_NOP) { 2057 found++; 2058 continue; 2059 } 2060 2061 jobref = 2062 fuword(&iocb->_aiocb_private.kernelinfo); 2063 2064 TAILQ_FOREACH(cb, &ki->kaio_jobdone, plist) { 2065 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) 2066 == jobref) { 2067 if (cb->uaiocb.aio_lio_opcode 2068 == LIO_WRITE) { 2069 p->p_stats->p_ru.ru_oublock 2070 += 2071 cb->outputcharge; 2072 cb->outputcharge = 0; 2073 } else if (cb->uaiocb.aio_lio_opcode 2074 == LIO_READ) { 2075 p->p_stats->p_ru.ru_inblock 2076 += cb->inputcharge; 2077 cb->inputcharge = 0; 2078 } 2079 found++; 2080 break; 2081 } 2082 } 2083 2084 s = splbio(); 2085 TAILQ_FOREACH(cb, &ki->kaio_bufdone, plist) { 2086 if (((intptr_t)cb->uaiocb._aiocb_private.kernelinfo) 2087 == jobref) { 2088 found++; 2089 break; 2090 } 2091 } 2092 splx(s); 2093 } 2094 2095 /* 2096 * If all I/Os have been disposed of, then we can 2097 * return. 2098 */ 2099 if (found == nentqueued) 2100 return (runningcode); 2101 2102 ki->kaio_flags |= KAIO_WAKEUP; 2103 error = tsleep(p, PRIBIO | PCATCH, "aiospn", 0); 2104 2105 if (error == EINTR) 2106 return (EINTR); 2107 else if (error == EWOULDBLOCK) 2108 return (EAGAIN); 2109 } 2110 } 2111 2112 return (runningcode); 2113 } 2114 2115 /* 2116 * This is a weird hack so that we can post a signal. It is safe to do so from 2117 * a timeout routine, but *not* from an interrupt routine. 2118 */ 2119 static void 2120 process_signal(void *aioj) 2121 { 2122 struct aiocblist *aiocbe = aioj; 2123 struct aio_liojob *lj = aiocbe->lio; 2124 struct aiocb *cb = &aiocbe->uaiocb; 2125 2126 if ((lj) && (lj->lioj_signal.sigev_notify == SIGEV_SIGNAL) && 2127 (lj->lioj_queue_count == lj->lioj_queue_finished_count)) { 2128 PROC_LOCK(lj->lioj_ki->kaio_p); 2129 psignal(lj->lioj_ki->kaio_p, lj->lioj_signal.sigev_signo); 2130 PROC_UNLOCK(lj->lioj_ki->kaio_p); 2131 lj->lioj_flags |= LIOJ_SIGNAL_POSTED; 2132 } 2133 2134 if (cb->aio_sigevent.sigev_notify == SIGEV_SIGNAL) { 2135 PROC_LOCK(aiocbe->userproc); 2136 psignal(aiocbe->userproc, cb->aio_sigevent.sigev_signo); 2137 PROC_UNLOCK(aiocbe->userproc); 2138 } 2139 } 2140 2141 /* 2142 * Interrupt handler for physio, performs the necessary process wakeups, and 2143 * signals. 2144 */ 2145 static void 2146 aio_physwakeup(struct buf *bp) 2147 { 2148 struct aiocblist *aiocbe; 2149 struct proc *p; 2150 struct kaioinfo *ki; 2151 struct aio_liojob *lj; 2152 2153 wakeup(bp); 2154 2155 aiocbe = (struct aiocblist *)bp->b_caller1; 2156 if (aiocbe) { 2157 p = aiocbe->userproc; 2158 2159 aiocbe->jobstate = JOBST_JOBBFINISHED; 2160 aiocbe->uaiocb._aiocb_private.status -= bp->b_resid; 2161 aiocbe->uaiocb._aiocb_private.error = 0; 2162 aiocbe->jobflags |= AIOCBLIST_DONE; 2163 2164 if (bp->b_ioflags & BIO_ERROR) 2165 aiocbe->uaiocb._aiocb_private.error = bp->b_error; 2166 2167 lj = aiocbe->lio; 2168 if (lj) { 2169 lj->lioj_buffer_finished_count++; 2170 2171 /* 2172 * wakeup/signal if all of the interrupt jobs are done. 2173 */ 2174 if (lj->lioj_buffer_finished_count == 2175 lj->lioj_buffer_count) { 2176 /* 2177 * Post a signal if it is called for. 2178 */ 2179 if ((lj->lioj_flags & 2180 (LIOJ_SIGNAL|LIOJ_SIGNAL_POSTED)) == 2181 LIOJ_SIGNAL) { 2182 lj->lioj_flags |= LIOJ_SIGNAL_POSTED; 2183 aiocbe->timeouthandle = 2184 timeout(process_signal, 2185 aiocbe, 0); 2186 } 2187 } 2188 } 2189 2190 ki = p->p_aioinfo; 2191 if (ki) { 2192 ki->kaio_buffer_finished_count++; 2193 TAILQ_REMOVE(&aio_bufjobs, aiocbe, list); 2194 TAILQ_REMOVE(&ki->kaio_bufqueue, aiocbe, plist); 2195 TAILQ_INSERT_TAIL(&ki->kaio_bufdone, aiocbe, plist); 2196 2197 KNOTE_UNLOCKED(&aiocbe->klist, 0); 2198 /* Do the wakeup. */ 2199 if (ki->kaio_flags & (KAIO_RUNDOWN|KAIO_WAKEUP)) { 2200 ki->kaio_flags &= ~KAIO_WAKEUP; 2201 wakeup(p); 2202 } 2203 } 2204 2205 if (aiocbe->uaiocb.aio_sigevent.sigev_notify == SIGEV_SIGNAL) 2206 aiocbe->timeouthandle = 2207 timeout(process_signal, aiocbe, 0); 2208 } 2209 } 2210 2211 /* syscall - wait for the next completion of an aio request */ 2212 int 2213 aio_waitcomplete(struct thread *td, struct aio_waitcomplete_args *uap) 2214 { 2215 struct proc *p = td->td_proc; 2216 struct timeval atv; 2217 struct timespec ts; 2218 struct kaioinfo *ki; 2219 struct aiocblist *cb = NULL; 2220 int error, s, timo; 2221 2222 suword(uap->aiocbp, (int)NULL); 2223 2224 timo = 0; 2225 if (uap->timeout) { 2226 /* Get timespec struct. */ 2227 error = copyin(uap->timeout, &ts, sizeof(ts)); 2228 if (error) 2229 return (error); 2230 2231 if ((ts.tv_nsec < 0) || (ts.tv_nsec >= 1000000000)) 2232 return (EINVAL); 2233 2234 TIMESPEC_TO_TIMEVAL(&atv, &ts); 2235 if (itimerfix(&atv)) 2236 return (EINVAL); 2237 timo = tvtohz(&atv); 2238 } 2239 2240 ki = p->p_aioinfo; 2241 if (ki == NULL) 2242 return (EAGAIN); 2243 2244 for (;;) { 2245 if ((cb = TAILQ_FIRST(&ki->kaio_jobdone)) != 0) { 2246 suword(uap->aiocbp, (uintptr_t)cb->uuaiocb); 2247 td->td_retval[0] = cb->uaiocb._aiocb_private.status; 2248 if (cb->uaiocb.aio_lio_opcode == LIO_WRITE) { 2249 p->p_stats->p_ru.ru_oublock += 2250 cb->outputcharge; 2251 cb->outputcharge = 0; 2252 } else if (cb->uaiocb.aio_lio_opcode == LIO_READ) { 2253 p->p_stats->p_ru.ru_inblock += cb->inputcharge; 2254 cb->inputcharge = 0; 2255 } 2256 aio_free_entry(cb); 2257 return (cb->uaiocb._aiocb_private.error); 2258 } 2259 2260 s = splbio(); 2261 if ((cb = TAILQ_FIRST(&ki->kaio_bufdone)) != 0 ) { 2262 splx(s); 2263 suword(uap->aiocbp, (uintptr_t)cb->uuaiocb); 2264 td->td_retval[0] = cb->uaiocb._aiocb_private.status; 2265 aio_free_entry(cb); 2266 return (cb->uaiocb._aiocb_private.error); 2267 } 2268 2269 ki->kaio_flags |= KAIO_WAKEUP; 2270 error = tsleep(p, PRIBIO | PCATCH, "aiowc", timo); 2271 splx(s); 2272 2273 if (error == ERESTART) 2274 return (EINTR); 2275 else if (error < 0) 2276 return (error); 2277 else if (error == EINTR) 2278 return (EINTR); 2279 else if (error == EWOULDBLOCK) 2280 return (EAGAIN); 2281 } 2282 } 2283 2284 /* kqueue attach function */ 2285 static int 2286 filt_aioattach(struct knote *kn) 2287 { 2288 struct aiocblist *aiocbe = (struct aiocblist *)kn->kn_sdata; 2289 2290 /* 2291 * The aiocbe pointer must be validated before using it, so 2292 * registration is restricted to the kernel; the user cannot 2293 * set EV_FLAG1. 2294 */ 2295 if ((kn->kn_flags & EV_FLAG1) == 0) 2296 return (EPERM); 2297 kn->kn_flags &= ~EV_FLAG1; 2298 2299 knlist_add(&aiocbe->klist, kn, 0); 2300 2301 return (0); 2302 } 2303 2304 /* kqueue detach function */ 2305 static void 2306 filt_aiodetach(struct knote *kn) 2307 { 2308 struct aiocblist *aiocbe = (struct aiocblist *)kn->kn_sdata; 2309 2310 knlist_remove(&aiocbe->klist, kn, 0); 2311 } 2312 2313 /* kqueue filter function */ 2314 /*ARGSUSED*/ 2315 static int 2316 filt_aio(struct knote *kn, long hint) 2317 { 2318 struct aiocblist *aiocbe = (struct aiocblist *)kn->kn_sdata; 2319 2320 kn->kn_data = aiocbe->uaiocb._aiocb_private.error; 2321 if (aiocbe->jobstate != JOBST_JOBFINISHED && 2322 aiocbe->jobstate != JOBST_JOBBFINISHED) 2323 return (0); 2324 kn->kn_flags |= EV_EOF; 2325 return (1); 2326 } 2327