kern_time.c revision 1.89
1/* $NetBSD: kern_time.c,v 1.89 2005/05/29 22:24:15 christos Exp $ */ 2 3/*- 4 * Copyright (c) 2000, 2004, 2005 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Christopher G. Demetriou. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 3. All advertising materials mentioning features or use of this software 19 * must display the following acknowledgement: 20 * This product includes software developed by the NetBSD 21 * Foundation, Inc. and its contributors. 22 * 4. Neither the name of The NetBSD Foundation nor the names of its 23 * contributors may be used to endorse or promote products derived 24 * from this software without specific prior written permission. 25 * 26 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 27 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 28 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 29 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 30 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 31 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 32 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 33 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 34 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 35 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 36 * POSSIBILITY OF SUCH DAMAGE. 37 */ 38 39/* 40 * Copyright (c) 1982, 1986, 1989, 1993 41 * The Regents of the University of California. All rights reserved. 42 * 43 * Redistribution and use in source and binary forms, with or without 44 * modification, are permitted provided that the following conditions 45 * are met: 46 * 1. Redistributions of source code must retain the above copyright 47 * notice, this list of conditions and the following disclaimer. 48 * 2. Redistributions in binary form must reproduce the above copyright 49 * notice, this list of conditions and the following disclaimer in the 50 * documentation and/or other materials provided with the distribution. 51 * 3. Neither the name of the University nor the names of its contributors 52 * may be used to endorse or promote products derived from this software 53 * without specific prior written permission. 54 * 55 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 56 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 57 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 58 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 59 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 60 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 61 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 62 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 63 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 64 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 65 * SUCH DAMAGE. 66 * 67 * @(#)kern_time.c 8.4 (Berkeley) 5/26/95 68 */ 69 70#include <sys/cdefs.h> 71__KERNEL_RCSID(0, "$NetBSD: kern_time.c,v 1.89 2005/05/29 22:24:15 christos Exp $"); 72 73#include "fs_nfs.h" 74#include "opt_nfs.h" 75#include "opt_nfsserver.h" 76 77#include <sys/param.h> 78#include <sys/resourcevar.h> 79#include <sys/kernel.h> 80#include <sys/systm.h> 81#include <sys/malloc.h> 82#include <sys/proc.h> 83#include <sys/sa.h> 84#include <sys/savar.h> 85#include <sys/vnode.h> 86#include <sys/signalvar.h> 87#include <sys/syslog.h> 88 89#include <sys/mount.h> 90#include <sys/syscallargs.h> 91 92#include <uvm/uvm_extern.h> 93 94#if defined(NFS) || defined(NFSSERVER) 95#include <nfs/rpcv2.h> 96#include <nfs/nfsproto.h> 97#include <nfs/nfs_var.h> 98#endif 99 100#include <machine/cpu.h> 101 102static void timerupcall(struct lwp *, void *); 103 104 105/* Time of day and interval timer support. 106 * 107 * These routines provide the kernel entry points to get and set 108 * the time-of-day and per-process interval timers. Subroutines 109 * here provide support for adding and subtracting timeval structures 110 * and decrementing interval timers, optionally reloading the interval 111 * timers when they expire. 112 */ 113 114/* This function is used by clock_settime and settimeofday */ 115int 116settime(struct timeval *tv) 117{ 118 struct timeval delta; 119 struct cpu_info *ci; 120 int s; 121 122 /* WHAT DO WE DO ABOUT PENDING REAL-TIME TIMEOUTS??? */ 123 s = splclock(); 124 timersub(tv, &time, &delta); 125 if ((delta.tv_sec < 0 || delta.tv_usec < 0) && securelevel > 1) { 126 splx(s); 127 return (EPERM); 128 } 129#ifdef notyet 130 if ((delta.tv_sec < 86400) && securelevel > 0) { 131 splx(s); 132 return (EPERM); 133 } 134#endif 135 time = *tv; 136 (void) spllowersoftclock(); 137 timeradd(&boottime, &delta, &boottime); 138 /* 139 * XXXSMP 140 * This is wrong. We should traverse a list of all 141 * CPUs and add the delta to the runtime of those 142 * CPUs which have a process on them. 143 */ 144 ci = curcpu(); 145 timeradd(&ci->ci_schedstate.spc_runtime, &delta, 146 &ci->ci_schedstate.spc_runtime); 147# if (defined(NFS) && !defined (NFS_V2_ONLY)) || defined(NFSSERVER) 148 nqnfs_lease_updatetime(delta.tv_sec); 149# endif 150 splx(s); 151 resettodr(); 152 return (0); 153} 154 155/* ARGSUSED */ 156int 157sys_clock_gettime(struct lwp *l, void *v, register_t *retval) 158{ 159 struct sys_clock_gettime_args /* { 160 syscallarg(clockid_t) clock_id; 161 syscallarg(struct timespec *) tp; 162 } */ *uap = v; 163 clockid_t clock_id; 164 struct timeval atv; 165 struct timespec ats; 166 int s; 167 168 clock_id = SCARG(uap, clock_id); 169 switch (clock_id) { 170 case CLOCK_REALTIME: 171 microtime(&atv); 172 TIMEVAL_TO_TIMESPEC(&atv,&ats); 173 break; 174 case CLOCK_MONOTONIC: 175 /* XXX "hz" granularity */ 176 s = splclock(); 177 atv = mono_time; 178 splx(s); 179 TIMEVAL_TO_TIMESPEC(&atv,&ats); 180 break; 181 default: 182 return (EINVAL); 183 } 184 185 return copyout(&ats, SCARG(uap, tp), sizeof(ats)); 186} 187 188/* ARGSUSED */ 189int 190sys_clock_settime(l, v, retval) 191 struct lwp *l; 192 void *v; 193 register_t *retval; 194{ 195 struct sys_clock_settime_args /* { 196 syscallarg(clockid_t) clock_id; 197 syscallarg(const struct timespec *) tp; 198 } */ *uap = v; 199 struct proc *p = l->l_proc; 200 int error; 201 202 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 203 return (error); 204 205 return (clock_settime1(SCARG(uap, clock_id), SCARG(uap, tp))); 206} 207 208 209int 210clock_settime1(clock_id, tp) 211 clockid_t clock_id; 212 const struct timespec *tp; 213{ 214 struct timespec ats; 215 struct timeval atv; 216 int error; 217 218 if ((error = copyin(tp, &ats, sizeof(ats))) != 0) 219 return (error); 220 221 switch (clock_id) { 222 case CLOCK_REALTIME: 223 TIMESPEC_TO_TIMEVAL(&atv, &ats); 224 if ((error = settime(&atv)) != 0) 225 return (error); 226 break; 227 case CLOCK_MONOTONIC: 228 return (EINVAL); /* read-only clock */ 229 default: 230 return (EINVAL); 231 } 232 233 return 0; 234} 235 236int 237sys_clock_getres(struct lwp *l, void *v, register_t *retval) 238{ 239 struct sys_clock_getres_args /* { 240 syscallarg(clockid_t) clock_id; 241 syscallarg(struct timespec *) tp; 242 } */ *uap = v; 243 clockid_t clock_id; 244 struct timespec ts; 245 int error = 0; 246 247 clock_id = SCARG(uap, clock_id); 248 switch (clock_id) { 249 case CLOCK_REALTIME: 250 case CLOCK_MONOTONIC: 251 ts.tv_sec = 0; 252 ts.tv_nsec = 1000000000 / hz; 253 break; 254 default: 255 return (EINVAL); 256 } 257 258 if (SCARG(uap, tp)) 259 error = copyout(&ts, SCARG(uap, tp), sizeof(ts)); 260 261 return error; 262} 263 264/* ARGSUSED */ 265int 266sys_nanosleep(struct lwp *l, void *v, register_t *retval) 267{ 268 static int nanowait; 269 struct sys_nanosleep_args/* { 270 syscallarg(struct timespec *) rqtp; 271 syscallarg(struct timespec *) rmtp; 272 } */ *uap = v; 273 struct timespec rqt; 274 struct timespec rmt; 275 struct timeval atv, utv; 276 int error, s, timo; 277 278 error = copyin(SCARG(uap, rqtp), &rqt, sizeof(struct timespec)); 279 if (error) 280 return (error); 281 282 TIMESPEC_TO_TIMEVAL(&atv,&rqt); 283 if (itimerfix(&atv)) 284 return (EINVAL); 285 286 s = splclock(); 287 timeradd(&atv,&time,&atv); 288 timo = hzto(&atv); 289 /* 290 * Avoid inadvertantly sleeping forever 291 */ 292 if (timo == 0) 293 timo = 1; 294 splx(s); 295 296 error = tsleep(&nanowait, PWAIT | PCATCH, "nanosleep", timo); 297 if (error == ERESTART) 298 error = EINTR; 299 if (error == EWOULDBLOCK) 300 error = 0; 301 302 if (SCARG(uap, rmtp)) { 303 int error1; 304 305 s = splclock(); 306 utv = time; 307 splx(s); 308 309 timersub(&atv, &utv, &utv); 310 if (utv.tv_sec < 0) 311 timerclear(&utv); 312 313 TIMEVAL_TO_TIMESPEC(&utv,&rmt); 314 error1 = copyout((caddr_t)&rmt, (caddr_t)SCARG(uap,rmtp), 315 sizeof(rmt)); 316 if (error1) 317 return (error1); 318 } 319 320 return error; 321} 322 323/* ARGSUSED */ 324int 325sys_gettimeofday(struct lwp *l, void *v, register_t *retval) 326{ 327 struct sys_gettimeofday_args /* { 328 syscallarg(struct timeval *) tp; 329 syscallarg(void *) tzp; really "struct timezone *" 330 } */ *uap = v; 331 struct timeval atv; 332 int error = 0; 333 struct timezone tzfake; 334 335 if (SCARG(uap, tp)) { 336 microtime(&atv); 337 error = copyout(&atv, SCARG(uap, tp), sizeof(atv)); 338 if (error) 339 return (error); 340 } 341 if (SCARG(uap, tzp)) { 342 /* 343 * NetBSD has no kernel notion of time zone, so we just 344 * fake up a timezone struct and return it if demanded. 345 */ 346 tzfake.tz_minuteswest = 0; 347 tzfake.tz_dsttime = 0; 348 error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake)); 349 } 350 return (error); 351} 352 353/* ARGSUSED */ 354int 355sys_settimeofday(struct lwp *l, void *v, register_t *retval) 356{ 357 struct sys_settimeofday_args /* { 358 syscallarg(const struct timeval *) tv; 359 syscallarg(const void *) tzp; really "const struct timezone *" 360 } */ *uap = v; 361 struct proc *p = l->l_proc; 362 int error; 363 364 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 365 return (error); 366 367 return settimeofday1(SCARG(uap, tv), SCARG(uap, tzp), p); 368} 369 370int 371settimeofday1(utv, utzp, p) 372 const struct timeval *utv; 373 const struct timezone *utzp; 374 struct proc *p; 375{ 376 struct timeval atv; 377 struct timezone atz; 378 struct timeval *tv = NULL; 379 struct timezone *tzp = NULL; 380 int error; 381 382 /* Verify all parameters before changing time. */ 383 if (utv) { 384 if ((error = copyin(utv, &atv, sizeof(atv))) != 0) 385 return (error); 386 tv = &atv; 387 } 388 /* XXX since we don't use tz, probably no point in doing copyin. */ 389 if (utzp) { 390 if ((error = copyin(utzp, &atz, sizeof(atz))) != 0) 391 return (error); 392 tzp = &atz; 393 } 394 395 if (tv) 396 if ((error = settime(tv)) != 0) 397 return (error); 398 /* 399 * NetBSD has no kernel notion of time zone, and only an 400 * obsolete program would try to set it, so we log a warning. 401 */ 402 if (tzp) 403 log(LOG_WARNING, "pid %d attempted to set the " 404 "(obsolete) kernel time zone\n", p->p_pid); 405 return (0); 406} 407 408int tickdelta; /* current clock skew, us. per tick */ 409long timedelta; /* unapplied time correction, us. */ 410long bigadj = 1000000; /* use 10x skew above bigadj us. */ 411int time_adjusted; /* set if an adjustment is made */ 412 413/* ARGSUSED */ 414int 415sys_adjtime(struct lwp *l, void *v, register_t *retval) 416{ 417 struct sys_adjtime_args /* { 418 syscallarg(const struct timeval *) delta; 419 syscallarg(struct timeval *) olddelta; 420 } */ *uap = v; 421 struct proc *p = l->l_proc; 422 int error; 423 424 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 425 return (error); 426 427 return adjtime1(SCARG(uap, delta), SCARG(uap, olddelta), p); 428} 429 430int 431adjtime1(delta, olddelta, p) 432 const struct timeval *delta; 433 struct timeval *olddelta; 434 struct proc *p; 435{ 436 struct timeval atv; 437 long ndelta, ntickdelta, odelta; 438 int error; 439 int s; 440 441 error = copyin(delta, &atv, sizeof(struct timeval)); 442 if (error) 443 return (error); 444 445 /* 446 * Compute the total correction and the rate at which to apply it. 447 * Round the adjustment down to a whole multiple of the per-tick 448 * delta, so that after some number of incremental changes in 449 * hardclock(), tickdelta will become zero, lest the correction 450 * overshoot and start taking us away from the desired final time. 451 */ 452 ndelta = atv.tv_sec * 1000000 + atv.tv_usec; 453 if (ndelta > bigadj || ndelta < -bigadj) 454 ntickdelta = 10 * tickadj; 455 else 456 ntickdelta = tickadj; 457 if (ndelta % ntickdelta) 458 ndelta = ndelta / ntickdelta * ntickdelta; 459 460 /* 461 * To make hardclock()'s job easier, make the per-tick delta negative 462 * if we want time to run slower; then hardclock can simply compute 463 * tick + tickdelta, and subtract tickdelta from timedelta. 464 */ 465 if (ndelta < 0) 466 ntickdelta = -ntickdelta; 467 if (ndelta != 0) 468 /* We need to save the system clock time during shutdown */ 469 time_adjusted |= 1; 470 s = splclock(); 471 odelta = timedelta; 472 timedelta = ndelta; 473 tickdelta = ntickdelta; 474 splx(s); 475 476 if (olddelta) { 477 atv.tv_sec = odelta / 1000000; 478 atv.tv_usec = odelta % 1000000; 479 error = copyout(&atv, olddelta, sizeof(struct timeval)); 480 } 481 return error; 482} 483 484/* 485 * Interval timer support. Both the BSD getitimer() family and the POSIX 486 * timer_*() family of routines are supported. 487 * 488 * All timers are kept in an array pointed to by p_timers, which is 489 * allocated on demand - many processes don't use timers at all. The 490 * first three elements in this array are reserved for the BSD timers: 491 * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, and element 492 * 2 is ITIMER_PROF. The rest may be allocated by the timer_create() 493 * syscall. 494 * 495 * Realtime timers are kept in the ptimer structure as an absolute 496 * time; virtual time timers are kept as a linked list of deltas. 497 * Virtual time timers are processed in the hardclock() routine of 498 * kern_clock.c. The real time timer is processed by a callout 499 * routine, called from the softclock() routine. Since a callout may 500 * be delayed in real time due to interrupt processing in the system, 501 * it is possible for the real time timeout routine (realtimeexpire, 502 * given below), to be delayed in real time past when it is supposed 503 * to occur. It does not suffice, therefore, to reload the real timer 504 * .it_value from the real time timers .it_interval. Rather, we 505 * compute the next time in absolute time the timer should go off. */ 506 507/* Allocate a POSIX realtime timer. */ 508int 509sys_timer_create(struct lwp *l, void *v, register_t *retval) 510{ 511 struct sys_timer_create_args /* { 512 syscallarg(clockid_t) clock_id; 513 syscallarg(struct sigevent *) evp; 514 syscallarg(timer_t *) timerid; 515 } */ *uap = v; 516 struct proc *p = l->l_proc; 517 clockid_t id; 518 struct sigevent *evp; 519 struct ptimer *pt; 520 timer_t timerid; 521 int error; 522 523 id = SCARG(uap, clock_id); 524 if (id < CLOCK_REALTIME || 525 id > CLOCK_PROF) 526 return (EINVAL); 527 528 if (p->p_timers == NULL) 529 timers_alloc(p); 530 531 /* Find a free timer slot, skipping those reserved for setitimer(). */ 532 for (timerid = 3; timerid < TIMER_MAX; timerid++) 533 if (p->p_timers->pts_timers[timerid] == NULL) 534 break; 535 536 if (timerid == TIMER_MAX) 537 return EAGAIN; 538 539 pt = pool_get(&ptimer_pool, PR_WAITOK); 540 evp = SCARG(uap, evp); 541 if (evp) { 542 if (((error = 543 copyin(evp, &pt->pt_ev, sizeof (pt->pt_ev))) != 0) || 544 ((pt->pt_ev.sigev_notify < SIGEV_NONE) || 545 (pt->pt_ev.sigev_notify > SIGEV_SA))) { 546 pool_put(&ptimer_pool, pt); 547 return (error ? error : EINVAL); 548 } 549 } else { 550 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 551 switch (id) { 552 case CLOCK_REALTIME: 553 pt->pt_ev.sigev_signo = SIGALRM; 554 break; 555 case CLOCK_VIRTUAL: 556 pt->pt_ev.sigev_signo = SIGVTALRM; 557 break; 558 case CLOCK_PROF: 559 pt->pt_ev.sigev_signo = SIGPROF; 560 break; 561 } 562 pt->pt_ev.sigev_value.sival_int = timerid; 563 } 564 pt->pt_info.ksi_signo = pt->pt_ev.sigev_signo; 565 pt->pt_info.ksi_errno = 0; 566 pt->pt_info.ksi_code = 0; 567 pt->pt_info.ksi_pid = p->p_pid; 568 pt->pt_info.ksi_uid = p->p_cred->p_ruid; 569 pt->pt_info.ksi_sigval = pt->pt_ev.sigev_value; 570 571 pt->pt_type = id; 572 pt->pt_proc = p; 573 pt->pt_overruns = 0; 574 pt->pt_poverruns = 0; 575 pt->pt_entry = timerid; 576 timerclear(&pt->pt_time.it_value); 577 if (id == CLOCK_REALTIME) 578 callout_init(&pt->pt_ch); 579 else 580 pt->pt_active = 0; 581 582 p->p_timers->pts_timers[timerid] = pt; 583 584 return copyout(&timerid, SCARG(uap, timerid), sizeof(timerid)); 585} 586 587 588/* Delete a POSIX realtime timer */ 589int 590sys_timer_delete(struct lwp *l, void *v, register_t *retval) 591{ 592 struct sys_timer_delete_args /* { 593 syscallarg(timer_t) timerid; 594 } */ *uap = v; 595 struct proc *p = l->l_proc; 596 timer_t timerid; 597 struct ptimer *pt, *ptn; 598 int s; 599 600 timerid = SCARG(uap, timerid); 601 602 if ((p->p_timers == NULL) || 603 (timerid < 2) || (timerid >= TIMER_MAX) || 604 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 605 return (EINVAL); 606 607 if (pt->pt_type == CLOCK_REALTIME) 608 callout_stop(&pt->pt_ch); 609 else if (pt->pt_active) { 610 s = splclock(); 611 ptn = LIST_NEXT(pt, pt_list); 612 LIST_REMOVE(pt, pt_list); 613 for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list)) 614 timeradd(&pt->pt_time.it_value, &ptn->pt_time.it_value, 615 &ptn->pt_time.it_value); 616 splx(s); 617 } 618 619 p->p_timers->pts_timers[timerid] = NULL; 620 pool_put(&ptimer_pool, pt); 621 622 return (0); 623} 624 625/* 626 * Set up the given timer. The value in pt->pt_time.it_value is taken 627 * to be an absolute time for CLOCK_REALTIME timers and a relative 628 * time for virtual timers. 629 * Must be called at splclock(). 630 */ 631void 632timer_settime(struct ptimer *pt) 633{ 634 struct ptimer *ptn, *pptn; 635 struct ptlist *ptl; 636 637 if (pt->pt_type == CLOCK_REALTIME) { 638 callout_stop(&pt->pt_ch); 639 if (timerisset(&pt->pt_time.it_value)) { 640 /* 641 * Don't need to check hzto() return value, here. 642 * callout_reset() does it for us. 643 */ 644 callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value), 645 realtimerexpire, pt); 646 } 647 } else { 648 if (pt->pt_active) { 649 ptn = LIST_NEXT(pt, pt_list); 650 LIST_REMOVE(pt, pt_list); 651 for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list)) 652 timeradd(&pt->pt_time.it_value, 653 &ptn->pt_time.it_value, 654 &ptn->pt_time.it_value); 655 } 656 if (timerisset(&pt->pt_time.it_value)) { 657 if (pt->pt_type == CLOCK_VIRTUAL) 658 ptl = &pt->pt_proc->p_timers->pts_virtual; 659 else 660 ptl = &pt->pt_proc->p_timers->pts_prof; 661 662 for (ptn = LIST_FIRST(ptl), pptn = NULL; 663 ptn && timercmp(&pt->pt_time.it_value, 664 &ptn->pt_time.it_value, >); 665 pptn = ptn, ptn = LIST_NEXT(ptn, pt_list)) 666 timersub(&pt->pt_time.it_value, 667 &ptn->pt_time.it_value, 668 &pt->pt_time.it_value); 669 670 if (pptn) 671 LIST_INSERT_AFTER(pptn, pt, pt_list); 672 else 673 LIST_INSERT_HEAD(ptl, pt, pt_list); 674 675 for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list)) 676 timersub(&ptn->pt_time.it_value, 677 &pt->pt_time.it_value, 678 &ptn->pt_time.it_value); 679 680 pt->pt_active = 1; 681 } else 682 pt->pt_active = 0; 683 } 684} 685 686void 687timer_gettime(struct ptimer *pt, struct itimerval *aitv) 688{ 689 struct ptimer *ptn; 690 691 *aitv = pt->pt_time; 692 if (pt->pt_type == CLOCK_REALTIME) { 693 /* 694 * Convert from absolute to relative time in .it_value 695 * part of real time timer. If time for real time 696 * timer has passed return 0, else return difference 697 * between current time and time for the timer to go 698 * off. 699 */ 700 if (timerisset(&aitv->it_value)) { 701 if (timercmp(&aitv->it_value, &time, <)) 702 timerclear(&aitv->it_value); 703 else 704 timersub(&aitv->it_value, &time, 705 &aitv->it_value); 706 } 707 } else if (pt->pt_active) { 708 if (pt->pt_type == CLOCK_VIRTUAL) 709 ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual); 710 else 711 ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof); 712 for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list)) 713 timeradd(&aitv->it_value, 714 &ptn->pt_time.it_value, &aitv->it_value); 715 KASSERT(ptn != NULL); /* pt should be findable on the list */ 716 } else 717 timerclear(&aitv->it_value); 718} 719 720 721 722/* Set and arm a POSIX realtime timer */ 723int 724sys_timer_settime(struct lwp *l, void *v, register_t *retval) 725{ 726 struct sys_timer_settime_args /* { 727 syscallarg(timer_t) timerid; 728 syscallarg(int) flags; 729 syscallarg(const struct itimerspec *) value; 730 syscallarg(struct itimerspec *) ovalue; 731 } */ *uap = v; 732 struct proc *p = l->l_proc; 733 int error, s, timerid; 734 struct itimerval val, oval; 735 struct itimerspec value, ovalue; 736 struct ptimer *pt; 737 738 timerid = SCARG(uap, timerid); 739 740 if ((p->p_timers == NULL) || 741 (timerid < 2) || (timerid >= TIMER_MAX) || 742 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 743 return (EINVAL); 744 745 if ((error = copyin(SCARG(uap, value), &value, 746 sizeof(struct itimerspec))) != 0) 747 return (error); 748 749 TIMESPEC_TO_TIMEVAL(&val.it_value, &value.it_value); 750 TIMESPEC_TO_TIMEVAL(&val.it_interval, &value.it_interval); 751 if (itimerfix(&val.it_value) || itimerfix(&val.it_interval)) 752 return (EINVAL); 753 754 oval = pt->pt_time; 755 pt->pt_time = val; 756 757 s = splclock(); 758 /* 759 * If we've been passed a relative time for a realtime timer, 760 * convert it to absolute; if an absolute time for a virtual 761 * timer, convert it to relative and make sure we don't set it 762 * to zero, which would cancel the timer, or let it go 763 * negative, which would confuse the comparison tests. 764 */ 765 if (timerisset(&pt->pt_time.it_value)) { 766 if (pt->pt_type == CLOCK_REALTIME) { 767 if ((SCARG(uap, flags) & TIMER_ABSTIME) == 0) 768 timeradd(&pt->pt_time.it_value, &time, 769 &pt->pt_time.it_value); 770 } else { 771 if ((SCARG(uap, flags) & TIMER_ABSTIME) != 0) { 772 timersub(&pt->pt_time.it_value, &time, 773 &pt->pt_time.it_value); 774 if (!timerisset(&pt->pt_time.it_value) || 775 pt->pt_time.it_value.tv_sec < 0) { 776 pt->pt_time.it_value.tv_sec = 0; 777 pt->pt_time.it_value.tv_usec = 1; 778 } 779 } 780 } 781 } 782 783 timer_settime(pt); 784 splx(s); 785 786 if (SCARG(uap, ovalue)) { 787 TIMEVAL_TO_TIMESPEC(&oval.it_value, &ovalue.it_value); 788 TIMEVAL_TO_TIMESPEC(&oval.it_interval, &ovalue.it_interval); 789 return copyout(&ovalue, SCARG(uap, ovalue), 790 sizeof(struct itimerspec)); 791 } 792 793 return (0); 794} 795 796/* Return the time remaining until a POSIX timer fires. */ 797int 798sys_timer_gettime(struct lwp *l, void *v, register_t *retval) 799{ 800 struct sys_timer_gettime_args /* { 801 syscallarg(timer_t) timerid; 802 syscallarg(struct itimerspec *) value; 803 } */ *uap = v; 804 struct itimerval aitv; 805 struct itimerspec its; 806 struct proc *p = l->l_proc; 807 int s, timerid; 808 struct ptimer *pt; 809 810 timerid = SCARG(uap, timerid); 811 812 if ((p->p_timers == NULL) || 813 (timerid < 2) || (timerid >= TIMER_MAX) || 814 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 815 return (EINVAL); 816 817 s = splclock(); 818 timer_gettime(pt, &aitv); 819 splx(s); 820 821 TIMEVAL_TO_TIMESPEC(&aitv.it_interval, &its.it_interval); 822 TIMEVAL_TO_TIMESPEC(&aitv.it_value, &its.it_value); 823 824 return copyout(&its, SCARG(uap, value), sizeof(its)); 825} 826 827/* 828 * Return the count of the number of times a periodic timer expired 829 * while a notification was already pending. The counter is reset when 830 * a timer expires and a notification can be posted. 831 */ 832int 833sys_timer_getoverrun(struct lwp *l, void *v, register_t *retval) 834{ 835 struct sys_timer_getoverrun_args /* { 836 syscallarg(timer_t) timerid; 837 } */ *uap = v; 838 struct proc *p = l->l_proc; 839 int timerid; 840 struct ptimer *pt; 841 842 timerid = SCARG(uap, timerid); 843 844 if ((p->p_timers == NULL) || 845 (timerid < 2) || (timerid >= TIMER_MAX) || 846 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 847 return (EINVAL); 848 849 *retval = pt->pt_poverruns; 850 851 return (0); 852} 853 854/* Glue function that triggers an upcall; called from userret(). */ 855static void 856timerupcall(struct lwp *l, void *arg) 857{ 858 struct ptimers *pt = (struct ptimers *)arg; 859 unsigned int i, fired, done; 860 extern struct pool siginfo_pool; /* XXX Ew. */ 861 862 KDASSERT(l->l_proc->p_sa); 863 /* Bail out if we do not own the virtual processor */ 864 if (l->l_savp->savp_lwp != l) 865 return ; 866 867 KERNEL_PROC_LOCK(l); 868 869 fired = pt->pts_fired; 870 done = 0; 871 while ((i = ffs(fired)) != 0) { 872 siginfo_t *si; 873 int mask = 1 << --i; 874 int f; 875 876 f = l->l_flag & L_SA; 877 l->l_flag &= ~L_SA; 878 si = pool_get(&siginfo_pool, PR_WAITOK); 879 si->_info = pt->pts_timers[i]->pt_info.ksi_info; 880 if (sa_upcall(l, SA_UPCALL_SIGEV | SA_UPCALL_DEFER, NULL, l, 881 sizeof(*si), si) != 0) { 882 pool_put(&siginfo_pool, si); 883 /* XXX What do we do here?? */ 884 } else 885 done |= mask; 886 fired &= ~mask; 887 l->l_flag |= f; 888 } 889 pt->pts_fired &= ~done; 890 if (pt->pts_fired == 0) 891 l->l_proc->p_userret = NULL; 892 893 KERNEL_PROC_UNLOCK(l); 894} 895 896 897/* 898 * Real interval timer expired: 899 * send process whose timer expired an alarm signal. 900 * If time is not set up to reload, then just return. 901 * Else compute next time timer should go off which is > current time. 902 * This is where delay in processing this timeout causes multiple 903 * SIGALRM calls to be compressed into one. 904 */ 905void 906realtimerexpire(void *arg) 907{ 908 struct ptimer *pt; 909 int s; 910 911 pt = (struct ptimer *)arg; 912 913 itimerfire(pt); 914 915 if (!timerisset(&pt->pt_time.it_interval)) { 916 timerclear(&pt->pt_time.it_value); 917 return; 918 } 919 for (;;) { 920 s = splclock(); 921 timeradd(&pt->pt_time.it_value, 922 &pt->pt_time.it_interval, &pt->pt_time.it_value); 923 if (timercmp(&pt->pt_time.it_value, &time, >)) { 924 /* 925 * Don't need to check hzto() return value, here. 926 * callout_reset() does it for us. 927 */ 928 callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value), 929 realtimerexpire, pt); 930 splx(s); 931 return; 932 } 933 splx(s); 934 pt->pt_overruns++; 935 } 936} 937 938/* BSD routine to get the value of an interval timer. */ 939/* ARGSUSED */ 940int 941sys_getitimer(struct lwp *l, void *v, register_t *retval) 942{ 943 struct sys_getitimer_args /* { 944 syscallarg(int) which; 945 syscallarg(struct itimerval *) itv; 946 } */ *uap = v; 947 struct proc *p = l->l_proc; 948 struct itimerval aitv; 949 int s, which; 950 951 which = SCARG(uap, which); 952 953 if ((u_int)which > ITIMER_PROF) 954 return (EINVAL); 955 956 if ((p->p_timers == NULL) || (p->p_timers->pts_timers[which] == NULL)){ 957 timerclear(&aitv.it_value); 958 timerclear(&aitv.it_interval); 959 } else { 960 s = splclock(); 961 timer_gettime(p->p_timers->pts_timers[which], &aitv); 962 splx(s); 963 } 964 965 return (copyout(&aitv, SCARG(uap, itv), sizeof(struct itimerval))); 966 967} 968 969/* BSD routine to set/arm an interval timer. */ 970/* ARGSUSED */ 971int 972sys_setitimer(struct lwp *l, void *v, register_t *retval) 973{ 974 struct sys_setitimer_args /* { 975 syscallarg(int) which; 976 syscallarg(const struct itimerval *) itv; 977 syscallarg(struct itimerval *) oitv; 978 } */ *uap = v; 979 struct proc *p = l->l_proc; 980 int which = SCARG(uap, which); 981 struct sys_getitimer_args getargs; 982 struct itimerval aitv; 983 const struct itimerval *itvp; 984 struct ptimer *pt; 985 int s, error; 986 987 if ((u_int)which > ITIMER_PROF) 988 return (EINVAL); 989 itvp = SCARG(uap, itv); 990 if (itvp && 991 (error = copyin(itvp, &aitv, sizeof(struct itimerval)) != 0)) 992 return (error); 993 if (SCARG(uap, oitv) != NULL) { 994 SCARG(&getargs, which) = which; 995 SCARG(&getargs, itv) = SCARG(uap, oitv); 996 if ((error = sys_getitimer(l, &getargs, retval)) != 0) 997 return (error); 998 } 999 if (itvp == 0) 1000 return (0); 1001 if (itimerfix(&aitv.it_value) || itimerfix(&aitv.it_interval)) 1002 return (EINVAL); 1003 1004 /* 1005 * Don't bother allocating data structures if the process just 1006 * wants to clear the timer. 1007 */ 1008 if (!timerisset(&aitv.it_value) && 1009 ((p->p_timers == NULL) ||(p->p_timers->pts_timers[which] == NULL))) 1010 return (0); 1011 1012 if (p->p_timers == NULL) 1013 timers_alloc(p); 1014 if (p->p_timers->pts_timers[which] == NULL) { 1015 pt = pool_get(&ptimer_pool, PR_WAITOK); 1016 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 1017 pt->pt_ev.sigev_value.sival_int = which; 1018 pt->pt_overruns = 0; 1019 pt->pt_proc = p; 1020 pt->pt_type = which; 1021 pt->pt_entry = which; 1022 switch (which) { 1023 case ITIMER_REAL: 1024 callout_init(&pt->pt_ch); 1025 pt->pt_ev.sigev_signo = SIGALRM; 1026 break; 1027 case ITIMER_VIRTUAL: 1028 pt->pt_active = 0; 1029 pt->pt_ev.sigev_signo = SIGVTALRM; 1030 break; 1031 case ITIMER_PROF: 1032 pt->pt_active = 0; 1033 pt->pt_ev.sigev_signo = SIGPROF; 1034 break; 1035 } 1036 } else 1037 pt = p->p_timers->pts_timers[which]; 1038 1039 pt->pt_time = aitv; 1040 p->p_timers->pts_timers[which] = pt; 1041 1042 s = splclock(); 1043 if ((which == ITIMER_REAL) && timerisset(&pt->pt_time.it_value)) { 1044 /* Convert to absolute time */ 1045 timeradd(&pt->pt_time.it_value, &time, &pt->pt_time.it_value); 1046 } 1047 timer_settime(pt); 1048 splx(s); 1049 1050 return (0); 1051} 1052 1053/* Utility routines to manage the array of pointers to timers. */ 1054void 1055timers_alloc(struct proc *p) 1056{ 1057 int i; 1058 struct ptimers *pts; 1059 1060 pts = malloc(sizeof (struct ptimers), M_SUBPROC, 0); 1061 LIST_INIT(&pts->pts_virtual); 1062 LIST_INIT(&pts->pts_prof); 1063 for (i = 0; i < TIMER_MAX; i++) 1064 pts->pts_timers[i] = NULL; 1065 pts->pts_fired = 0; 1066 p->p_timers = pts; 1067} 1068 1069/* 1070 * Clean up the per-process timers. If "which" is set to TIMERS_ALL, 1071 * then clean up all timers and free all the data structures. If 1072 * "which" is set to TIMERS_POSIX, only clean up the timers allocated 1073 * by timer_create(), not the BSD setitimer() timers, and only free the 1074 * structure if none of those remain. 1075 */ 1076void 1077timers_free(struct proc *p, int which) 1078{ 1079 int i, s; 1080 struct ptimers *pts; 1081 struct ptimer *pt, *ptn; 1082 struct timeval tv; 1083 1084 if (p->p_timers) { 1085 pts = p->p_timers; 1086 if (which == TIMERS_ALL) 1087 i = 0; 1088 else { 1089 s = splclock(); 1090 timerclear(&tv); 1091 for (ptn = LIST_FIRST(&p->p_timers->pts_virtual); 1092 ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL]; 1093 ptn = LIST_NEXT(ptn, pt_list)) 1094 timeradd(&tv, &ptn->pt_time.it_value, &tv); 1095 LIST_FIRST(&p->p_timers->pts_virtual) = NULL; 1096 if (ptn) { 1097 timeradd(&tv, &ptn->pt_time.it_value, 1098 &ptn->pt_time.it_value); 1099 LIST_INSERT_HEAD(&p->p_timers->pts_virtual, 1100 ptn, pt_list); 1101 } 1102 1103 timerclear(&tv); 1104 for (ptn = LIST_FIRST(&p->p_timers->pts_prof); 1105 ptn && ptn != pts->pts_timers[ITIMER_PROF]; 1106 ptn = LIST_NEXT(ptn, pt_list)) 1107 timeradd(&tv, &ptn->pt_time.it_value, &tv); 1108 LIST_FIRST(&p->p_timers->pts_prof) = NULL; 1109 if (ptn) { 1110 timeradd(&tv, &ptn->pt_time.it_value, 1111 &ptn->pt_time.it_value); 1112 LIST_INSERT_HEAD(&p->p_timers->pts_prof, ptn, 1113 pt_list); 1114 } 1115 splx(s); 1116 i = 3; 1117 } 1118 for ( ; i < TIMER_MAX; i++) 1119 if ((pt = pts->pts_timers[i]) != NULL) { 1120 if (pt->pt_type == CLOCK_REALTIME) 1121 callout_stop(&pt->pt_ch); 1122 pts->pts_timers[i] = NULL; 1123 pool_put(&ptimer_pool, pt); 1124 } 1125 if ((pts->pts_timers[0] == NULL) && 1126 (pts->pts_timers[1] == NULL) && 1127 (pts->pts_timers[2] == NULL)) { 1128 p->p_timers = NULL; 1129 free(pts, M_SUBPROC); 1130 } 1131 } 1132} 1133 1134/* 1135 * Check that a proposed value to load into the .it_value or 1136 * .it_interval part of an interval timer is acceptable, and 1137 * fix it to have at least minimal value (i.e. if it is less 1138 * than the resolution of the clock, round it up.) 1139 */ 1140int 1141itimerfix(struct timeval *tv) 1142{ 1143 1144 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 1145 return (EINVAL); 1146 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick) 1147 tv->tv_usec = tick; 1148 return (0); 1149} 1150 1151/* 1152 * Decrement an interval timer by a specified number 1153 * of microseconds, which must be less than a second, 1154 * i.e. < 1000000. If the timer expires, then reload 1155 * it. In this case, carry over (usec - old value) to 1156 * reduce the value reloaded into the timer so that 1157 * the timer does not drift. This routine assumes 1158 * that it is called in a context where the timers 1159 * on which it is operating cannot change in value. 1160 */ 1161int 1162itimerdecr(struct ptimer *pt, int usec) 1163{ 1164 struct itimerval *itp; 1165 1166 itp = &pt->pt_time; 1167 if (itp->it_value.tv_usec < usec) { 1168 if (itp->it_value.tv_sec == 0) { 1169 /* expired, and already in next interval */ 1170 usec -= itp->it_value.tv_usec; 1171 goto expire; 1172 } 1173 itp->it_value.tv_usec += 1000000; 1174 itp->it_value.tv_sec--; 1175 } 1176 itp->it_value.tv_usec -= usec; 1177 usec = 0; 1178 if (timerisset(&itp->it_value)) 1179 return (1); 1180 /* expired, exactly at end of interval */ 1181expire: 1182 if (timerisset(&itp->it_interval)) { 1183 itp->it_value = itp->it_interval; 1184 itp->it_value.tv_usec -= usec; 1185 if (itp->it_value.tv_usec < 0) { 1186 itp->it_value.tv_usec += 1000000; 1187 itp->it_value.tv_sec--; 1188 } 1189 timer_settime(pt); 1190 } else 1191 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1192 return (0); 1193} 1194 1195void 1196itimerfire(struct ptimer *pt) 1197{ 1198 struct proc *p = pt->pt_proc; 1199 struct sadata_vp *vp; 1200 int s; 1201 unsigned int i; 1202 1203 if (pt->pt_ev.sigev_notify == SIGEV_SIGNAL) { 1204 /* 1205 * No RT signal infrastructure exists at this time; 1206 * just post the signal number and throw away the 1207 * value. 1208 */ 1209 if (sigismember(&p->p_sigctx.ps_siglist, pt->pt_ev.sigev_signo)) 1210 pt->pt_overruns++; 1211 else { 1212 ksiginfo_t ksi; 1213 (void)memset(&ksi, 0, sizeof(ksi)); 1214 ksi.ksi_signo = pt->pt_ev.sigev_signo; 1215 ksi.ksi_code = SI_TIMER; 1216 ksi.ksi_sigval = pt->pt_ev.sigev_value; 1217 pt->pt_poverruns = pt->pt_overruns; 1218 pt->pt_overruns = 0; 1219 kpsignal(p, &ksi, NULL); 1220 } 1221 } else if (pt->pt_ev.sigev_notify == SIGEV_SA && (p->p_flag & P_SA)) { 1222 /* Cause the process to generate an upcall when it returns. */ 1223 1224 if (p->p_userret == NULL) { 1225 /* 1226 * XXX stop signals can be processed inside tsleep, 1227 * which can be inside sa_yield's inner loop, which 1228 * makes testing for sa_idle alone insuffucent to 1229 * determine if we really should call setrunnable. 1230 */ 1231 pt->pt_poverruns = pt->pt_overruns; 1232 pt->pt_overruns = 0; 1233 i = 1 << pt->pt_entry; 1234 p->p_timers->pts_fired = i; 1235 p->p_userret = timerupcall; 1236 p->p_userret_arg = p->p_timers; 1237 1238 SCHED_LOCK(s); 1239 SLIST_FOREACH(vp, &p->p_sa->sa_vps, savp_next) { 1240 if (vp->savp_lwp->l_flag & L_SA_IDLE) { 1241 vp->savp_lwp->l_flag &= ~L_SA_IDLE; 1242 sched_wakeup(vp->savp_lwp); 1243 break; 1244 } 1245 } 1246 SCHED_UNLOCK(s); 1247 } else if (p->p_userret == timerupcall) { 1248 i = 1 << pt->pt_entry; 1249 if ((p->p_timers->pts_fired & i) == 0) { 1250 pt->pt_poverruns = pt->pt_overruns; 1251 pt->pt_overruns = 0; 1252 p->p_timers->pts_fired |= i; 1253 } else 1254 pt->pt_overruns++; 1255 } else { 1256 pt->pt_overruns++; 1257 if ((p->p_flag & P_WEXIT) == 0) 1258 printf("itimerfire(%d): overrun %d on timer %x (userret is %p)\n", 1259 p->p_pid, pt->pt_overruns, 1260 pt->pt_ev.sigev_value.sival_int, 1261 p->p_userret); 1262 } 1263 } 1264 1265} 1266 1267/* 1268 * ratecheck(): simple time-based rate-limit checking. see ratecheck(9) 1269 * for usage and rationale. 1270 */ 1271int 1272ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1273{ 1274 struct timeval tv, delta; 1275 int s, rv = 0; 1276 1277 s = splclock(); 1278 tv = mono_time; 1279 splx(s); 1280 1281 timersub(&tv, lasttime, &delta); 1282 1283 /* 1284 * check for 0,0 is so that the message will be seen at least once, 1285 * even if interval is huge. 1286 */ 1287 if (timercmp(&delta, mininterval, >=) || 1288 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1289 *lasttime = tv; 1290 rv = 1; 1291 } 1292 1293 return (rv); 1294} 1295 1296/* 1297 * ppsratecheck(): packets (or events) per second limitation. 1298 */ 1299int 1300ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1301{ 1302 struct timeval tv, delta; 1303 int s, rv; 1304 1305 s = splclock(); 1306 tv = mono_time; 1307 splx(s); 1308 1309 timersub(&tv, lasttime, &delta); 1310 1311 /* 1312 * check for 0,0 is so that the message will be seen at least once. 1313 * if more than one second have passed since the last update of 1314 * lasttime, reset the counter. 1315 * 1316 * we do increment *curpps even in *curpps < maxpps case, as some may 1317 * try to use *curpps for stat purposes as well. 1318 */ 1319 if ((lasttime->tv_sec == 0 && lasttime->tv_usec == 0) || 1320 delta.tv_sec >= 1) { 1321 *lasttime = tv; 1322 *curpps = 0; 1323 } 1324 if (maxpps < 0) 1325 rv = 1; 1326 else if (*curpps < maxpps) 1327 rv = 1; 1328 else 1329 rv = 0; 1330 1331#if 1 /*DIAGNOSTIC?*/ 1332 /* be careful about wrap-around */ 1333 if (*curpps + 1 > *curpps) 1334 *curpps = *curpps + 1; 1335#else 1336 /* 1337 * assume that there's not too many calls to this function. 1338 * not sure if the assumption holds, as it depends on *caller's* 1339 * behavior, not the behavior of this function. 1340 * IMHO it is wrong to make assumption on the caller's behavior, 1341 * so the above #if is #if 1, not #ifdef DIAGNOSTIC. 1342 */ 1343 *curpps = *curpps + 1; 1344#endif 1345 1346 return (rv); 1347} 1348