sys_aio.c revision 1.18
1 /* $NetBSD: sys_aio.c,v 1.18 2008/04/24 18:39:24 ad Exp $ */ 2 3 /* 4 * Copyright (c) 2007, Mindaugas Rasiukevicius <rmind at NetBSD org> 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS 17 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 18 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 19 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 20 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 21 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 22 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 23 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 24 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 25 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 26 * POSSIBILITY OF SUCH DAMAGE. 27 */ 28 29 /* 30 * TODO: 31 * 1. Additional work for VCHR and maybe VBLK devices. 32 * 2. Consider making the job-finding O(n) per one file descriptor. 33 */ 34 35 #include <sys/cdefs.h> 36 __KERNEL_RCSID(0, "$NetBSD: sys_aio.c,v 1.18 2008/04/24 18:39:24 ad Exp $"); 37 38 #include "opt_ddb.h" 39 40 #include <sys/param.h> 41 #include <sys/condvar.h> 42 #include <sys/file.h> 43 #include <sys/filedesc.h> 44 #include <sys/kernel.h> 45 #include <sys/kmem.h> 46 #include <sys/lwp.h> 47 #include <sys/mutex.h> 48 #include <sys/pool.h> 49 #include <sys/proc.h> 50 #include <sys/queue.h> 51 #include <sys/signal.h> 52 #include <sys/signalvar.h> 53 #include <sys/syscallargs.h> 54 #include <sys/sysctl.h> 55 #include <sys/systm.h> 56 #include <sys/types.h> 57 #include <sys/vnode.h> 58 #include <sys/atomic.h> 59 60 #include <uvm/uvm_extern.h> 61 62 /* 63 * System-wide limits and counter of AIO operations. 64 */ 65 static u_int aio_listio_max = AIO_LISTIO_MAX; 66 static u_int aio_max = AIO_MAX; 67 static u_int aio_jobs_count; 68 69 static struct pool aio_job_pool; 70 static struct pool aio_lio_pool; 71 72 /* Prototypes */ 73 void aio_worker(void *); 74 static void aio_process(struct aio_job *); 75 static void aio_sendsig(struct proc *, struct sigevent *); 76 static int aio_enqueue_job(int, void *, struct lio_req *); 77 78 /* 79 * Initialize the AIO system. 80 */ 81 void 82 aio_sysinit(void) 83 { 84 85 pool_init(&aio_job_pool, sizeof(struct aio_job), 0, 0, 0, 86 "aio_jobs_pool", &pool_allocator_nointr, IPL_NONE); 87 pool_init(&aio_lio_pool, sizeof(struct lio_req), 0, 0, 0, 88 "aio_lio_pool", &pool_allocator_nointr, IPL_NONE); 89 } 90 91 /* 92 * Initialize Asynchronous I/O data structures for the process. 93 */ 94 int 95 aio_init(struct proc *p) 96 { 97 struct aioproc *aio; 98 struct lwp *l; 99 int error; 100 bool inmem; 101 vaddr_t uaddr; 102 103 /* Allocate and initialize AIO structure */ 104 aio = kmem_zalloc(sizeof(struct aioproc), KM_SLEEP); 105 if (aio == NULL) 106 return EAGAIN; 107 108 /* Initialize queue and their synchronization structures */ 109 mutex_init(&aio->aio_mtx, MUTEX_DEFAULT, IPL_NONE); 110 cv_init(&aio->aio_worker_cv, "aiowork"); 111 cv_init(&aio->done_cv, "aiodone"); 112 TAILQ_INIT(&aio->jobs_queue); 113 114 /* 115 * Create an AIO worker thread. 116 * XXX: Currently, AIO thread is not protected against user's actions. 117 */ 118 inmem = uvm_uarea_alloc(&uaddr); 119 if (uaddr == 0) { 120 aio_exit(p, aio); 121 return EAGAIN; 122 } 123 error = lwp_create(curlwp, p, uaddr, inmem, 0, NULL, 0, aio_worker, 124 NULL, &l, curlwp->l_class); 125 if (error != 0) { 126 uvm_uarea_free(uaddr, curcpu()); 127 aio_exit(p, aio); 128 return error; 129 } 130 131 /* Recheck if we are really first */ 132 mutex_enter(p->p_lock); 133 if (p->p_aio) { 134 mutex_exit(p->p_lock); 135 aio_exit(p, aio); 136 lwp_exit(l); 137 return 0; 138 } 139 p->p_aio = aio; 140 141 /* Complete the initialization of thread, and run it */ 142 aio->aio_worker = l; 143 p->p_nrlwps++; 144 lwp_lock(l); 145 l->l_stat = LSRUN; 146 l->l_priority = MAXPRI_USER; 147 sched_enqueue(l, false); 148 lwp_unlock(l); 149 mutex_exit(p->p_lock); 150 151 return 0; 152 } 153 154 /* 155 * Exit of Asynchronous I/O subsystem of process. 156 */ 157 void 158 aio_exit(struct proc *p, struct aioproc *aio) 159 { 160 struct aio_job *a_job; 161 162 if (aio == NULL) 163 return; 164 165 /* Free AIO queue */ 166 while (!TAILQ_EMPTY(&aio->jobs_queue)) { 167 a_job = TAILQ_FIRST(&aio->jobs_queue); 168 TAILQ_REMOVE(&aio->jobs_queue, a_job, list); 169 pool_put(&aio_job_pool, a_job); 170 atomic_dec_uint(&aio_jobs_count); 171 } 172 173 /* Destroy and free the entire AIO data structure */ 174 cv_destroy(&aio->aio_worker_cv); 175 cv_destroy(&aio->done_cv); 176 mutex_destroy(&aio->aio_mtx); 177 kmem_free(aio, sizeof(struct aioproc)); 178 } 179 180 /* 181 * AIO worker thread and processor. 182 */ 183 void 184 aio_worker(void *arg) 185 { 186 struct proc *p = curlwp->l_proc; 187 struct aioproc *aio = p->p_aio; 188 struct aio_job *a_job; 189 struct lio_req *lio; 190 sigset_t oss, nss; 191 int error, refcnt; 192 193 /* 194 * Make an empty signal mask, so it 195 * handles only SIGKILL and SIGSTOP. 196 */ 197 sigfillset(&nss); 198 mutex_enter(p->p_lock); 199 error = sigprocmask1(curlwp, SIG_SETMASK, &nss, &oss); 200 mutex_exit(p->p_lock); 201 KASSERT(error == 0); 202 203 for (;;) { 204 /* 205 * Loop for each job in the queue. If there 206 * are no jobs then sleep. 207 */ 208 mutex_enter(&aio->aio_mtx); 209 while ((a_job = TAILQ_FIRST(&aio->jobs_queue)) == NULL) { 210 if (cv_wait_sig(&aio->aio_worker_cv, &aio->aio_mtx)) { 211 /* 212 * Thread was interrupted - check for 213 * pending exit or suspend. 214 */ 215 mutex_exit(&aio->aio_mtx); 216 lwp_userret(curlwp); 217 mutex_enter(&aio->aio_mtx); 218 } 219 } 220 221 /* Take the job from the queue */ 222 aio->curjob = a_job; 223 TAILQ_REMOVE(&aio->jobs_queue, a_job, list); 224 225 atomic_dec_uint(&aio_jobs_count); 226 aio->jobs_count--; 227 228 mutex_exit(&aio->aio_mtx); 229 230 /* Process an AIO operation */ 231 aio_process(a_job); 232 233 /* Copy data structure back to the user-space */ 234 (void)copyout(&a_job->aiocbp, a_job->aiocb_uptr, 235 sizeof(struct aiocb)); 236 237 mutex_enter(&aio->aio_mtx); 238 aio->curjob = NULL; 239 240 /* Decrease a reference counter, if there is a LIO structure */ 241 lio = a_job->lio; 242 refcnt = (lio != NULL ? --lio->refcnt : -1); 243 244 /* Notify all suspenders */ 245 cv_broadcast(&aio->done_cv); 246 mutex_exit(&aio->aio_mtx); 247 248 /* Send a signal, if any */ 249 aio_sendsig(p, &a_job->aiocbp.aio_sigevent); 250 251 /* Destroy the LIO structure */ 252 if (refcnt == 0) { 253 aio_sendsig(p, &lio->sig); 254 pool_put(&aio_lio_pool, lio); 255 } 256 257 /* Destroy the the job */ 258 pool_put(&aio_job_pool, a_job); 259 } 260 261 /* NOTREACHED */ 262 } 263 264 static void 265 aio_process(struct aio_job *a_job) 266 { 267 struct proc *p = curlwp->l_proc; 268 struct aiocb *aiocbp = &a_job->aiocbp; 269 struct file *fp; 270 int fd = aiocbp->aio_fildes; 271 int error = 0; 272 273 KASSERT(a_job->aio_op != 0); 274 275 if ((a_job->aio_op & (AIO_READ | AIO_WRITE)) != 0) { 276 struct iovec aiov; 277 struct uio auio; 278 279 if (aiocbp->aio_nbytes > SSIZE_MAX) { 280 error = EINVAL; 281 goto done; 282 } 283 284 fp = fd_getfile(fd); 285 if (fp == NULL) { 286 error = EBADF; 287 goto done; 288 } 289 290 aiov.iov_base = (void *)(uintptr_t)aiocbp->aio_buf; 291 aiov.iov_len = aiocbp->aio_nbytes; 292 auio.uio_iov = &aiov; 293 auio.uio_iovcnt = 1; 294 auio.uio_resid = aiocbp->aio_nbytes; 295 auio.uio_vmspace = p->p_vmspace; 296 297 if (a_job->aio_op & AIO_READ) { 298 /* 299 * Perform a Read operation 300 */ 301 KASSERT((a_job->aio_op & AIO_WRITE) == 0); 302 303 if ((fp->f_flag & FREAD) == 0) { 304 fd_putfile(fd); 305 error = EBADF; 306 goto done; 307 } 308 auio.uio_rw = UIO_READ; 309 error = (*fp->f_ops->fo_read)(fp, &aiocbp->aio_offset, 310 &auio, fp->f_cred, FOF_UPDATE_OFFSET); 311 } else { 312 /* 313 * Perform a Write operation 314 */ 315 KASSERT(a_job->aio_op & AIO_WRITE); 316 317 if ((fp->f_flag & FWRITE) == 0) { 318 fd_putfile(fd); 319 error = EBADF; 320 goto done; 321 } 322 auio.uio_rw = UIO_WRITE; 323 error = (*fp->f_ops->fo_write)(fp, &aiocbp->aio_offset, 324 &auio, fp->f_cred, FOF_UPDATE_OFFSET); 325 } 326 fd_putfile(fd); 327 328 /* Store the result value */ 329 a_job->aiocbp.aio_nbytes -= auio.uio_resid; 330 a_job->aiocbp._retval = (error == 0) ? 331 a_job->aiocbp.aio_nbytes : -1; 332 333 } else if ((a_job->aio_op & (AIO_SYNC | AIO_DSYNC)) != 0) { 334 /* 335 * Perform a file Sync operation 336 */ 337 struct vnode *vp; 338 339 if ((error = fd_getvnode(fd, &fp)) != 0) 340 goto done; 341 342 if ((fp->f_flag & FWRITE) == 0) { 343 fd_putfile(fd); 344 error = EBADF; 345 goto done; 346 } 347 348 vp = (struct vnode *)fp->f_data; 349 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 350 if (a_job->aio_op & AIO_DSYNC) { 351 error = VOP_FSYNC(vp, fp->f_cred, 352 FSYNC_WAIT | FSYNC_DATAONLY, 0, 0); 353 } else if (a_job->aio_op & AIO_SYNC) { 354 error = VOP_FSYNC(vp, fp->f_cred, 355 FSYNC_WAIT, 0, 0); 356 if (error == 0 && bioopsp != NULL && 357 vp->v_mount && 358 (vp->v_mount->mnt_flag & MNT_SOFTDEP)) 359 bioopsp->io_fsync(vp, 0); 360 } 361 VOP_UNLOCK(vp, 0); 362 fd_putfile(fd); 363 364 /* Store the result value */ 365 a_job->aiocbp._retval = (error == 0) ? 0 : -1; 366 367 } else 368 panic("aio_process: invalid operation code\n"); 369 370 done: 371 /* Job is done, set the error, if any */ 372 a_job->aiocbp._errno = error; 373 a_job->aiocbp._state = JOB_DONE; 374 } 375 376 /* 377 * Send AIO signal. 378 */ 379 static void 380 aio_sendsig(struct proc *p, struct sigevent *sig) 381 { 382 ksiginfo_t ksi; 383 384 if (sig->sigev_signo == 0 || sig->sigev_notify == SIGEV_NONE) 385 return; 386 387 KSI_INIT(&ksi); 388 ksi.ksi_signo = sig->sigev_signo; 389 ksi.ksi_code = SI_ASYNCIO; 390 ksi.ksi_value = sig->sigev_value; 391 mutex_enter(proc_lock); 392 kpsignal(p, &ksi, NULL); 393 mutex_exit(proc_lock); 394 } 395 396 /* 397 * Enqueue the job. 398 */ 399 static int 400 aio_enqueue_job(int op, void *aiocb_uptr, struct lio_req *lio) 401 { 402 struct proc *p = curlwp->l_proc; 403 struct aioproc *aio; 404 struct aio_job *a_job; 405 struct aiocb aiocbp; 406 struct sigevent *sig; 407 int error; 408 409 /* Non-accurate check for the limit */ 410 if (aio_jobs_count + 1 > aio_max) 411 return EAGAIN; 412 413 /* Get the data structure from user-space */ 414 error = copyin(aiocb_uptr, &aiocbp, sizeof(struct aiocb)); 415 if (error) 416 return error; 417 418 /* Check if signal is set, and validate it */ 419 sig = &aiocbp.aio_sigevent; 420 if (sig->sigev_signo < 0 || sig->sigev_signo >= NSIG || 421 sig->sigev_notify < SIGEV_NONE || sig->sigev_notify > SIGEV_SA) 422 return EINVAL; 423 424 /* Buffer and byte count */ 425 if (((AIO_SYNC | AIO_DSYNC) & op) == 0) 426 if (aiocbp.aio_buf == NULL || aiocbp.aio_nbytes > SSIZE_MAX) 427 return EINVAL; 428 429 /* Check the opcode, if LIO_NOP - simply ignore */ 430 if (op == AIO_LIO) { 431 KASSERT(lio != NULL); 432 if (aiocbp.aio_lio_opcode == LIO_WRITE) 433 op = AIO_WRITE; 434 else if (aiocbp.aio_lio_opcode == LIO_READ) 435 op = AIO_READ; 436 else 437 return (aiocbp.aio_lio_opcode == LIO_NOP) ? 0 : EINVAL; 438 } else { 439 KASSERT(lio == NULL); 440 } 441 442 /* 443 * Look for already existing job. If found - the job is in-progress. 444 * According to POSIX this is invalid, so return the error. 445 */ 446 aio = p->p_aio; 447 if (aio) { 448 mutex_enter(&aio->aio_mtx); 449 if (aio->curjob) { 450 a_job = aio->curjob; 451 if (a_job->aiocb_uptr == aiocb_uptr) { 452 mutex_exit(&aio->aio_mtx); 453 return EINVAL; 454 } 455 } 456 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) { 457 if (a_job->aiocb_uptr != aiocb_uptr) 458 continue; 459 mutex_exit(&aio->aio_mtx); 460 return EINVAL; 461 } 462 mutex_exit(&aio->aio_mtx); 463 } 464 465 /* 466 * Check if AIO structure is initialized, if not - initialize it. 467 * In LIO case, we did that already. We will recheck this with 468 * the lock in aio_init(). 469 */ 470 if (lio == NULL && p->p_aio == NULL) 471 if (aio_init(p)) 472 return EAGAIN; 473 aio = p->p_aio; 474 475 /* 476 * Set the state with errno, and copy data 477 * structure back to the user-space. 478 */ 479 aiocbp._state = JOB_WIP; 480 aiocbp._errno = EINPROGRESS; 481 aiocbp._retval = -1; 482 error = copyout(&aiocbp, aiocb_uptr, sizeof(struct aiocb)); 483 if (error) 484 return error; 485 486 /* Allocate and initialize a new AIO job */ 487 a_job = pool_get(&aio_job_pool, PR_WAITOK); 488 memset(a_job, 0, sizeof(struct aio_job)); 489 490 /* 491 * Set the data. 492 * Store the user-space pointer for searching. Since we 493 * are storing only per proc pointers - it is safe. 494 */ 495 memcpy(&a_job->aiocbp, &aiocbp, sizeof(struct aiocb)); 496 a_job->aiocb_uptr = aiocb_uptr; 497 a_job->aio_op |= op; 498 a_job->lio = lio; 499 500 /* 501 * Add the job to the queue, update the counters, and 502 * notify the AIO worker thread to handle the job. 503 */ 504 mutex_enter(&aio->aio_mtx); 505 506 /* Fail, if the limit was reached */ 507 if (atomic_inc_uint_nv(&aio_jobs_count) > aio_max || 508 aio->jobs_count >= aio_listio_max) { 509 atomic_dec_uint(&aio_jobs_count); 510 mutex_exit(&aio->aio_mtx); 511 pool_put(&aio_job_pool, a_job); 512 return EAGAIN; 513 } 514 515 TAILQ_INSERT_TAIL(&aio->jobs_queue, a_job, list); 516 aio->jobs_count++; 517 if (lio) 518 lio->refcnt++; 519 cv_signal(&aio->aio_worker_cv); 520 521 mutex_exit(&aio->aio_mtx); 522 523 /* 524 * One would handle the errors only with aio_error() function. 525 * This way is appropriate according to POSIX. 526 */ 527 return 0; 528 } 529 530 /* 531 * Syscall functions. 532 */ 533 534 int 535 sys_aio_cancel(struct lwp *l, const struct sys_aio_cancel_args *uap, register_t *retval) 536 { 537 /* { 538 syscallarg(int) fildes; 539 syscallarg(struct aiocb *) aiocbp; 540 } */ 541 struct proc *p = l->l_proc; 542 struct aioproc *aio; 543 struct aio_job *a_job; 544 struct aiocb *aiocbp_ptr; 545 struct lio_req *lio; 546 struct filedesc *fdp = p->p_fd; 547 unsigned int cn, errcnt, fildes; 548 549 TAILQ_HEAD(, aio_job) tmp_jobs_list; 550 551 /* Check for invalid file descriptor */ 552 fildes = (unsigned int)SCARG(uap, fildes); 553 if (fildes >= fdp->fd_nfiles) 554 return EBADF; 555 membar_consumer(); 556 if (fdp->fd_ofiles[fildes] == NULL || fdp->fd_ofiles[fildes]->ff_file == NULL) 557 return EBADF; 558 559 /* Check if AIO structure is initialized */ 560 if (p->p_aio == NULL) { 561 *retval = AIO_NOTCANCELED; 562 return 0; 563 } 564 565 aio = p->p_aio; 566 aiocbp_ptr = (struct aiocb *)SCARG(uap, aiocbp); 567 568 mutex_enter(&aio->aio_mtx); 569 570 /* Cancel the jobs, and remove them from the queue */ 571 cn = 0; 572 TAILQ_INIT(&tmp_jobs_list); 573 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) { 574 if (aiocbp_ptr) { 575 if (aiocbp_ptr != a_job->aiocb_uptr) 576 continue; 577 if (fildes != a_job->aiocbp.aio_fildes) { 578 mutex_exit(&aio->aio_mtx); 579 return EBADF; 580 } 581 } else if (a_job->aiocbp.aio_fildes != fildes) 582 continue; 583 584 TAILQ_REMOVE(&aio->jobs_queue, a_job, list); 585 TAILQ_INSERT_TAIL(&tmp_jobs_list, a_job, list); 586 587 /* Decrease the counters */ 588 atomic_dec_uint(&aio_jobs_count); 589 aio->jobs_count--; 590 lio = a_job->lio; 591 if (lio != NULL && --lio->refcnt != 0) 592 a_job->lio = NULL; 593 594 cn++; 595 if (aiocbp_ptr) 596 break; 597 } 598 599 /* There are canceled jobs */ 600 if (cn) 601 *retval = AIO_CANCELED; 602 603 /* We cannot cancel current job */ 604 a_job = aio->curjob; 605 if (a_job && ((a_job->aiocbp.aio_fildes == fildes) || 606 (a_job->aiocb_uptr == aiocbp_ptr))) 607 *retval = AIO_NOTCANCELED; 608 609 mutex_exit(&aio->aio_mtx); 610 611 /* Free the jobs after the lock */ 612 errcnt = 0; 613 while (!TAILQ_EMPTY(&tmp_jobs_list)) { 614 a_job = TAILQ_FIRST(&tmp_jobs_list); 615 TAILQ_REMOVE(&tmp_jobs_list, a_job, list); 616 /* Set the errno and copy structures back to the user-space */ 617 a_job->aiocbp._errno = ECANCELED; 618 a_job->aiocbp._state = JOB_DONE; 619 if (copyout(&a_job->aiocbp, a_job->aiocb_uptr, 620 sizeof(struct aiocb))) 621 errcnt++; 622 /* Send a signal if any */ 623 aio_sendsig(p, &a_job->aiocbp.aio_sigevent); 624 if (a_job->lio) { 625 lio = a_job->lio; 626 aio_sendsig(p, &lio->sig); 627 pool_put(&aio_lio_pool, lio); 628 } 629 pool_put(&aio_job_pool, a_job); 630 } 631 632 if (errcnt) 633 return EFAULT; 634 635 /* Set a correct return value */ 636 if (*retval == 0) 637 *retval = AIO_ALLDONE; 638 639 return 0; 640 } 641 642 int 643 sys_aio_error(struct lwp *l, const struct sys_aio_error_args *uap, register_t *retval) 644 { 645 /* { 646 syscallarg(const struct aiocb *) aiocbp; 647 } */ 648 struct proc *p = l->l_proc; 649 struct aioproc *aio = p->p_aio; 650 struct aiocb aiocbp; 651 int error; 652 653 if (aio == NULL) 654 return EINVAL; 655 656 error = copyin(SCARG(uap, aiocbp), &aiocbp, sizeof(struct aiocb)); 657 if (error) 658 return error; 659 660 if (aiocbp._state == JOB_NONE) 661 return EINVAL; 662 663 *retval = aiocbp._errno; 664 665 return 0; 666 } 667 668 int 669 sys_aio_fsync(struct lwp *l, const struct sys_aio_fsync_args *uap, register_t *retval) 670 { 671 /* { 672 syscallarg(int) op; 673 syscallarg(struct aiocb *) aiocbp; 674 } */ 675 int op = SCARG(uap, op); 676 677 if ((op != O_DSYNC) && (op != O_SYNC)) 678 return EINVAL; 679 680 op = O_DSYNC ? AIO_DSYNC : AIO_SYNC; 681 682 return aio_enqueue_job(op, SCARG(uap, aiocbp), NULL); 683 } 684 685 int 686 sys_aio_read(struct lwp *l, const struct sys_aio_read_args *uap, register_t *retval) 687 { 688 /* { 689 syscallarg(struct aiocb *) aiocbp; 690 } */ 691 692 return aio_enqueue_job(AIO_READ, SCARG(uap, aiocbp), NULL); 693 } 694 695 int 696 sys_aio_return(struct lwp *l, const struct sys_aio_return_args *uap, register_t *retval) 697 { 698 /* { 699 syscallarg(struct aiocb *) aiocbp; 700 } */ 701 struct proc *p = l->l_proc; 702 struct aioproc *aio = p->p_aio; 703 struct aiocb aiocbp; 704 int error; 705 706 if (aio == NULL) 707 return EINVAL; 708 709 error = copyin(SCARG(uap, aiocbp), &aiocbp, sizeof(struct aiocb)); 710 if (error) 711 return error; 712 713 if (aiocbp._errno == EINPROGRESS || aiocbp._state != JOB_DONE) 714 return EINVAL; 715 716 *retval = aiocbp._retval; 717 718 /* Reset the internal variables */ 719 aiocbp._errno = 0; 720 aiocbp._retval = -1; 721 aiocbp._state = JOB_NONE; 722 error = copyout(&aiocbp, SCARG(uap, aiocbp), sizeof(struct aiocb)); 723 724 return error; 725 } 726 727 int 728 sys_aio_suspend(struct lwp *l, const struct sys_aio_suspend_args *uap, register_t *retval) 729 { 730 /* { 731 syscallarg(const struct aiocb *const[]) list; 732 syscallarg(int) nent; 733 syscallarg(const struct timespec *) timeout; 734 } */ 735 struct proc *p = l->l_proc; 736 struct aioproc *aio; 737 struct aio_job *a_job; 738 struct aiocb **aiocbp_list; 739 struct timespec ts; 740 int i, error, nent, timo; 741 742 if (p->p_aio == NULL) 743 return EAGAIN; 744 aio = p->p_aio; 745 746 nent = SCARG(uap, nent); 747 if (nent <= 0 || nent > aio_listio_max) 748 return EAGAIN; 749 750 if (SCARG(uap, timeout)) { 751 /* Convert timespec to ticks */ 752 error = copyin(SCARG(uap, timeout), &ts, 753 sizeof(struct timespec)); 754 if (error) 755 return error; 756 timo = mstohz((ts.tv_sec * 1000) + (ts.tv_nsec / 1000000)); 757 if (timo == 0 && ts.tv_sec == 0 && ts.tv_nsec > 0) 758 timo = 1; 759 if (timo <= 0) 760 return EAGAIN; 761 } else 762 timo = 0; 763 764 /* Get the list from user-space */ 765 aiocbp_list = kmem_zalloc(nent * sizeof(struct aio_job), KM_SLEEP); 766 error = copyin(SCARG(uap, list), aiocbp_list, 767 nent * sizeof(struct aiocb)); 768 if (error) { 769 kmem_free(aiocbp_list, nent * sizeof(struct aio_job)); 770 return error; 771 } 772 773 mutex_enter(&aio->aio_mtx); 774 for (;;) { 775 776 for (i = 0; i < nent; i++) { 777 778 /* Skip NULL entries */ 779 if (aiocbp_list[i] == NULL) 780 continue; 781 782 /* Skip current job */ 783 if (aio->curjob) { 784 a_job = aio->curjob; 785 if (a_job->aiocb_uptr == aiocbp_list[i]) 786 continue; 787 } 788 789 /* Look for a job in the queue */ 790 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) 791 if (a_job->aiocb_uptr == aiocbp_list[i]) 792 break; 793 794 if (a_job == NULL) { 795 struct aiocb aiocbp; 796 797 mutex_exit(&aio->aio_mtx); 798 799 error = copyin(aiocbp_list[i], &aiocbp, 800 sizeof(struct aiocb)); 801 if (error == 0 && aiocbp._state != JOB_DONE) { 802 mutex_enter(&aio->aio_mtx); 803 continue; 804 } 805 806 kmem_free(aiocbp_list, 807 nent * sizeof(struct aio_job)); 808 return error; 809 } 810 } 811 812 /* Wait for a signal or when timeout occurs */ 813 error = cv_timedwait_sig(&aio->done_cv, &aio->aio_mtx, timo); 814 if (error) { 815 if (error == EWOULDBLOCK) 816 error = EAGAIN; 817 break; 818 } 819 } 820 mutex_exit(&aio->aio_mtx); 821 822 kmem_free(aiocbp_list, nent * sizeof(struct aio_job)); 823 return error; 824 } 825 826 int 827 sys_aio_write(struct lwp *l, const struct sys_aio_write_args *uap, register_t *retval) 828 { 829 /* { 830 syscallarg(struct aiocb *) aiocbp; 831 } */ 832 833 return aio_enqueue_job(AIO_WRITE, SCARG(uap, aiocbp), NULL); 834 } 835 836 int 837 sys_lio_listio(struct lwp *l, const struct sys_lio_listio_args *uap, register_t *retval) 838 { 839 /* { 840 syscallarg(int) mode; 841 syscallarg(struct aiocb *const[]) list; 842 syscallarg(int) nent; 843 syscallarg(struct sigevent *) sig; 844 } */ 845 struct proc *p = l->l_proc; 846 struct aioproc *aio; 847 struct aiocb **aiocbp_list; 848 struct lio_req *lio; 849 int i, error, errcnt, mode, nent; 850 851 mode = SCARG(uap, mode); 852 nent = SCARG(uap, nent); 853 854 /* Non-accurate checks for the limit and invalid values */ 855 if (nent < 1 || nent > aio_listio_max) 856 return EINVAL; 857 if (aio_jobs_count + nent > aio_max) 858 return EAGAIN; 859 860 /* Check if AIO structure is initialized, if not - initialize it */ 861 if (p->p_aio == NULL) 862 if (aio_init(p)) 863 return EAGAIN; 864 aio = p->p_aio; 865 866 /* Create a LIO structure */ 867 lio = pool_get(&aio_lio_pool, PR_WAITOK); 868 lio->refcnt = 1; 869 error = 0; 870 871 switch (mode) { 872 case LIO_WAIT: 873 memset(&lio->sig, 0, sizeof(struct sigevent)); 874 break; 875 case LIO_NOWAIT: 876 /* Check for signal, validate it */ 877 if (SCARG(uap, sig)) { 878 struct sigevent *sig = &lio->sig; 879 880 error = copyin(SCARG(uap, sig), &lio->sig, 881 sizeof(struct sigevent)); 882 if (error == 0 && 883 (sig->sigev_signo < 0 || 884 sig->sigev_signo >= NSIG || 885 sig->sigev_notify < SIGEV_NONE || 886 sig->sigev_notify > SIGEV_SA)) 887 error = EINVAL; 888 } else 889 memset(&lio->sig, 0, sizeof(struct sigevent)); 890 break; 891 default: 892 error = EINVAL; 893 break; 894 } 895 896 if (error != 0) { 897 pool_put(&aio_lio_pool, lio); 898 return error; 899 } 900 901 /* Get the list from user-space */ 902 aiocbp_list = kmem_zalloc(nent * sizeof(struct aio_job), KM_SLEEP); 903 error = copyin(SCARG(uap, list), aiocbp_list, 904 nent * sizeof(struct aiocb)); 905 if (error) { 906 mutex_enter(&aio->aio_mtx); 907 goto err; 908 } 909 910 /* Enqueue all jobs */ 911 errcnt = 0; 912 for (i = 0; i < nent; i++) { 913 error = aio_enqueue_job(AIO_LIO, aiocbp_list[i], lio); 914 /* 915 * According to POSIX, in such error case it may 916 * fail with other I/O operations initiated. 917 */ 918 if (error) 919 errcnt++; 920 } 921 922 mutex_enter(&aio->aio_mtx); 923 924 /* Return an error, if any */ 925 if (errcnt) { 926 error = EIO; 927 goto err; 928 } 929 930 if (mode == LIO_WAIT) { 931 /* 932 * Wait for AIO completion. In such case, 933 * the LIO structure will be freed here. 934 */ 935 while (lio->refcnt > 1 && error == 0) 936 error = cv_wait_sig(&aio->done_cv, &aio->aio_mtx); 937 if (error) 938 error = EINTR; 939 } 940 941 err: 942 if (--lio->refcnt != 0) 943 lio = NULL; 944 mutex_exit(&aio->aio_mtx); 945 if (lio != NULL) { 946 aio_sendsig(p, &lio->sig); 947 pool_put(&aio_lio_pool, lio); 948 } 949 kmem_free(aiocbp_list, nent * sizeof(struct aio_job)); 950 return error; 951 } 952 953 /* 954 * SysCtl 955 */ 956 957 static int 958 sysctl_aio_listio_max(SYSCTLFN_ARGS) 959 { 960 struct sysctlnode node; 961 int error, newsize; 962 963 node = *rnode; 964 node.sysctl_data = &newsize; 965 966 newsize = aio_listio_max; 967 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 968 if (error || newp == NULL) 969 return error; 970 971 if (newsize < 1 || newsize > aio_max) 972 return EINVAL; 973 aio_listio_max = newsize; 974 975 return 0; 976 } 977 978 static int 979 sysctl_aio_max(SYSCTLFN_ARGS) 980 { 981 struct sysctlnode node; 982 int error, newsize; 983 984 node = *rnode; 985 node.sysctl_data = &newsize; 986 987 newsize = aio_max; 988 error = sysctl_lookup(SYSCTLFN_CALL(&node)); 989 if (error || newp == NULL) 990 return error; 991 992 if (newsize < 1 || newsize < aio_listio_max) 993 return EINVAL; 994 aio_max = newsize; 995 996 return 0; 997 } 998 999 SYSCTL_SETUP(sysctl_aio_setup, "sysctl aio setup") 1000 { 1001 1002 sysctl_createv(clog, 0, NULL, NULL, 1003 CTLFLAG_PERMANENT, 1004 CTLTYPE_NODE, "kern", NULL, 1005 NULL, 0, NULL, 0, 1006 CTL_KERN, CTL_EOL); 1007 sysctl_createv(clog, 0, NULL, NULL, 1008 CTLFLAG_PERMANENT | CTLFLAG_IMMEDIATE, 1009 CTLTYPE_INT, "posix_aio", 1010 SYSCTL_DESCR("Version of IEEE Std 1003.1 and its " 1011 "Asynchronous I/O option to which the " 1012 "system attempts to conform"), 1013 NULL, _POSIX_ASYNCHRONOUS_IO, NULL, 0, 1014 CTL_KERN, CTL_CREATE, CTL_EOL); 1015 sysctl_createv(clog, 0, NULL, NULL, 1016 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1017 CTLTYPE_INT, "aio_listio_max", 1018 SYSCTL_DESCR("Maximum number of asynchronous I/O " 1019 "operations in a single list I/O call"), 1020 sysctl_aio_listio_max, 0, &aio_listio_max, 0, 1021 CTL_KERN, CTL_CREATE, CTL_EOL); 1022 sysctl_createv(clog, 0, NULL, NULL, 1023 CTLFLAG_PERMANENT | CTLFLAG_READWRITE, 1024 CTLTYPE_INT, "aio_max", 1025 SYSCTL_DESCR("Maximum number of asynchronous I/O " 1026 "operations"), 1027 sysctl_aio_max, 0, &aio_max, 0, 1028 CTL_KERN, CTL_CREATE, CTL_EOL); 1029 } 1030 1031 /* 1032 * Debugging 1033 */ 1034 #if defined(DDB) 1035 void 1036 aio_print_jobs(void (*pr)(const char *, ...)) 1037 { 1038 struct proc *p = (curlwp == NULL ? NULL : curlwp->l_proc); 1039 struct aioproc *aio; 1040 struct aio_job *a_job; 1041 struct aiocb *aiocbp; 1042 1043 if (p == NULL) { 1044 (*pr)("AIO: We are not in the processes right now.\n"); 1045 return; 1046 } 1047 1048 aio = p->p_aio; 1049 if (aio == NULL) { 1050 (*pr)("AIO data is not initialized (PID = %d).\n", p->p_pid); 1051 return; 1052 } 1053 1054 (*pr)("AIO: PID = %d\n", p->p_pid); 1055 (*pr)("AIO: Global count of the jobs = %u\n", aio_jobs_count); 1056 (*pr)("AIO: Count of the jobs = %u\n", aio->jobs_count); 1057 1058 if (aio->curjob) { 1059 a_job = aio->curjob; 1060 (*pr)("\nAIO current job:\n"); 1061 (*pr)(" opcode = %d, errno = %d, state = %d, aiocb_ptr = %p\n", 1062 a_job->aio_op, a_job->aiocbp._errno, 1063 a_job->aiocbp._state, a_job->aiocb_uptr); 1064 aiocbp = &a_job->aiocbp; 1065 (*pr)(" fd = %d, offset = %u, buf = %p, nbytes = %u\n", 1066 aiocbp->aio_fildes, aiocbp->aio_offset, 1067 aiocbp->aio_buf, aiocbp->aio_nbytes); 1068 } 1069 1070 (*pr)("\nAIO queue:\n"); 1071 TAILQ_FOREACH(a_job, &aio->jobs_queue, list) { 1072 (*pr)(" opcode = %d, errno = %d, state = %d, aiocb_ptr = %p\n", 1073 a_job->aio_op, a_job->aiocbp._errno, 1074 a_job->aiocbp._state, a_job->aiocb_uptr); 1075 aiocbp = &a_job->aiocbp; 1076 (*pr)(" fd = %d, offset = %u, buf = %p, nbytes = %u\n", 1077 aiocbp->aio_fildes, aiocbp->aio_offset, 1078 aiocbp->aio_buf, aiocbp->aio_nbytes); 1079 } 1080 } 1081 #endif /* defined(DDB) */ 1082
Indexes created Mon Sep 22 13:09:51 GMT 2025