1 /* $NetBSD: kern_time.c,v 1.72 2003/08/07 16:31:51 agc Exp $ */ 2 3 /*- 4 * Copyright (c) 2000 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 1.72 2003/08/07 16:31:51 agc 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> 101 102 static 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 */ 115 int 116 settime(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 */ 156 int 157 sys_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 */ 189 int 190 sys_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 } 207 208 209 int 210 clock_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 } 235 236 int 237 sys_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 */ 265 int 266 sys_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((caddr_t)SCARG(uap, rqtp), (caddr_t)&rqt, 279 sizeof(struct timespec)); 280 if (error) 281 return (error); 282 283 TIMESPEC_TO_TIMEVAL(&atv,&rqt) 284 if (itimerfix(&atv) || atv.tv_sec > 1000000000) 285 return (EINVAL); 286 287 s = splclock(); 288 timeradd(&atv,&time,&atv); 289 timo = hzto(&atv); 290 /* 291 * Avoid inadvertantly sleeping forever 292 */ 293 if (timo == 0) 294 timo = 1; 295 splx(s); 296 297 error = tsleep(&nanowait, PWAIT | PCATCH, "nanosleep", timo); 298 if (error == ERESTART) 299 error = EINTR; 300 if (error == EWOULDBLOCK) 301 error = 0; 302 303 if (SCARG(uap, rmtp)) { 304 int error; 305 306 s = splclock(); 307 utv = time; 308 splx(s); 309 310 timersub(&atv, &utv, &utv); 311 if (utv.tv_sec < 0) 312 timerclear(&utv); 313 314 TIMEVAL_TO_TIMESPEC(&utv,&rmt); 315 error = copyout((caddr_t)&rmt, (caddr_t)SCARG(uap,rmtp), 316 sizeof(rmt)); 317 if (error) 318 return (error); 319 } 320 321 return error; 322 } 323 324 /* ARGSUSED */ 325 int 326 sys_gettimeofday(struct lwp *l, void *v, register_t *retval) 327 { 328 struct sys_gettimeofday_args /* { 329 syscallarg(struct timeval *) tp; 330 syscallarg(struct timezone *) tzp; 331 } */ *uap = v; 332 struct timeval atv; 333 int error = 0; 334 struct timezone tzfake; 335 336 if (SCARG(uap, tp)) { 337 microtime(&atv); 338 error = copyout(&atv, SCARG(uap, tp), sizeof(atv)); 339 if (error) 340 return (error); 341 } 342 if (SCARG(uap, tzp)) { 343 /* 344 * NetBSD has no kernel notion of time zone, so we just 345 * fake up a timezone struct and return it if demanded. 346 */ 347 tzfake.tz_minuteswest = 0; 348 tzfake.tz_dsttime = 0; 349 error = copyout(&tzfake, SCARG(uap, tzp), sizeof(tzfake)); 350 } 351 return (error); 352 } 353 354 /* ARGSUSED */ 355 int 356 sys_settimeofday(struct lwp *l, void *v, register_t *retval) 357 { 358 struct sys_settimeofday_args /* { 359 syscallarg(const struct timeval *) tv; 360 syscallarg(const struct timezone *) tzp; 361 } */ *uap = v; 362 struct proc *p = l->l_proc; 363 int error; 364 365 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 366 return (error); 367 368 return settimeofday1(SCARG(uap, tv), SCARG(uap, tzp), p); 369 } 370 371 int 372 settimeofday1(utv, utzp, p) 373 const struct timeval *utv; 374 const struct timezone *utzp; 375 struct proc *p; 376 { 377 struct timeval atv; 378 struct timezone atz; 379 struct timeval *tv = NULL; 380 struct timezone *tzp = NULL; 381 int error; 382 383 /* Verify all parameters before changing time. */ 384 if (utv) { 385 if ((error = copyin(utv, &atv, sizeof(atv))) != 0) 386 return (error); 387 tv = &atv; 388 } 389 /* XXX since we don't use tz, probably no point in doing copyin. */ 390 if (utzp) { 391 if ((error = copyin(utzp, &atz, sizeof(atz))) != 0) 392 return (error); 393 tzp = &atz; 394 } 395 396 if (tv) 397 if ((error = settime(tv)) != 0) 398 return (error); 399 /* 400 * NetBSD has no kernel notion of time zone, and only an 401 * obsolete program would try to set it, so we log a warning. 402 */ 403 if (tzp) 404 log(LOG_WARNING, "pid %d attempted to set the " 405 "(obsolete) kernel time zone\n", p->p_pid); 406 return (0); 407 } 408 409 int tickdelta; /* current clock skew, us. per tick */ 410 long timedelta; /* unapplied time correction, us. */ 411 long bigadj = 1000000; /* use 10x skew above bigadj us. */ 412 int time_adjusted; /* set if an adjustment is made */ 413 414 /* ARGSUSED */ 415 int 416 sys_adjtime(struct lwp *l, void *v, register_t *retval) 417 { 418 struct sys_adjtime_args /* { 419 syscallarg(const struct timeval *) delta; 420 syscallarg(struct timeval *) olddelta; 421 } */ *uap = v; 422 struct proc *p = l->l_proc; 423 int error; 424 425 if ((error = suser(p->p_ucred, &p->p_acflag)) != 0) 426 return (error); 427 428 return adjtime1(SCARG(uap, delta), SCARG(uap, olddelta), p); 429 } 430 431 int 432 adjtime1(delta, olddelta, p) 433 const struct timeval *delta; 434 struct timeval *olddelta; 435 struct proc *p; 436 { 437 struct timeval atv; 438 long ndelta, ntickdelta, odelta; 439 int error; 440 int s; 441 442 error = copyin(delta, &atv, sizeof(struct timeval)); 443 if (error) 444 return (error); 445 446 if (olddelta != NULL) { 447 if (uvm_useracc((caddr_t)olddelta, 448 sizeof(struct timeval), B_WRITE) == FALSE) 449 return (EFAULT); 450 } 451 452 /* 453 * Compute the total correction and the rate at which to apply it. 454 * Round the adjustment down to a whole multiple of the per-tick 455 * delta, so that after some number of incremental changes in 456 * hardclock(), tickdelta will become zero, lest the correction 457 * overshoot and start taking us away from the desired final time. 458 */ 459 ndelta = atv.tv_sec * 1000000 + atv.tv_usec; 460 if (ndelta > bigadj || ndelta < -bigadj) 461 ntickdelta = 10 * tickadj; 462 else 463 ntickdelta = tickadj; 464 if (ndelta % ntickdelta) 465 ndelta = ndelta / ntickdelta * ntickdelta; 466 467 /* 468 * To make hardclock()'s job easier, make the per-tick delta negative 469 * if we want time to run slower; then hardclock can simply compute 470 * tick + tickdelta, and subtract tickdelta from timedelta. 471 */ 472 if (ndelta < 0) 473 ntickdelta = -ntickdelta; 474 if (ndelta != 0) 475 /* We need to save the system clock time during shutdown */ 476 time_adjusted |= 1; 477 s = splclock(); 478 odelta = timedelta; 479 timedelta = ndelta; 480 tickdelta = ntickdelta; 481 splx(s); 482 483 if (olddelta) { 484 atv.tv_sec = odelta / 1000000; 485 atv.tv_usec = odelta % 1000000; 486 (void) copyout(&atv, olddelta, sizeof(struct timeval)); 487 } 488 return (0); 489 } 490 491 /* 492 * Interval timer support. Both the BSD getitimer() family and the POSIX 493 * timer_*() family of routines are supported. 494 * 495 * All timers are kept in an array pointed to by p_timers, which is 496 * allocated on demand - many processes don't use timers at all. The 497 * first three elements in this array are reserved for the BSD timers: 498 * element 0 is ITIMER_REAL, element 1 is ITIMER_VIRTUAL, and element 499 * 2 is ITIMER_PROF. The rest may be allocated by the timer_create() 500 * syscall. 501 * 502 * Realtime timers are kept in the ptimer structure as an absolute 503 * time; virtual time timers are kept as a linked list of deltas. 504 * Virtual time timers are processed in the hardclock() routine of 505 * kern_clock.c. The real time timer is processed by a callout 506 * routine, called from the softclock() routine. Since a callout may 507 * be delayed in real time due to interrupt processing in the system, 508 * it is possible for the real time timeout routine (realtimeexpire, 509 * given below), to be delayed in real time past when it is supposed 510 * to occur. It does not suffice, therefore, to reload the real timer 511 * .it_value from the real time timers .it_interval. Rather, we 512 * compute the next time in absolute time the timer should go off. */ 513 514 /* Allocate a POSIX realtime timer. */ 515 int 516 sys_timer_create(struct lwp *l, void *v, register_t *retval) 517 { 518 struct sys_timer_create_args /* { 519 syscallarg(clockid_t) clock_id; 520 syscallarg(struct sigevent *) evp; 521 syscallarg(timer_t *) timerid; 522 } */ *uap = v; 523 struct proc *p = l->l_proc; 524 clockid_t id; 525 struct sigevent *evp; 526 struct ptimer *pt; 527 timer_t timerid; 528 int error; 529 530 id = SCARG(uap, clock_id); 531 if (id < CLOCK_REALTIME || 532 id > CLOCK_PROF) 533 return (EINVAL); 534 535 if (p->p_timers == NULL) 536 timers_alloc(p); 537 538 /* Find a free timer slot, skipping those reserved for setitimer(). */ 539 for (timerid = 3; timerid < TIMER_MAX; timerid++) 540 if (p->p_timers->pts_timers[timerid] == NULL) 541 break; 542 543 if (timerid == TIMER_MAX) 544 return EAGAIN; 545 546 pt = pool_get(&ptimer_pool, PR_WAITOK); 547 evp = SCARG(uap, evp); 548 if (evp) { 549 if (((error = 550 copyin(evp, &pt->pt_ev, sizeof (pt->pt_ev))) != 0) || 551 ((pt->pt_ev.sigev_notify < SIGEV_NONE) || 552 (pt->pt_ev.sigev_notify > SIGEV_SA))) { 553 pool_put(&ptimer_pool, pt); 554 return (error ? error : EINVAL); 555 } 556 } else { 557 pt->pt_ev.sigev_notify = SIGEV_SIGNAL; 558 switch (id) { 559 case CLOCK_REALTIME: 560 pt->pt_ev.sigev_signo = SIGALRM; 561 break; 562 case CLOCK_VIRTUAL: 563 pt->pt_ev.sigev_signo = SIGVTALRM; 564 break; 565 case CLOCK_PROF: 566 pt->pt_ev.sigev_signo = SIGPROF; 567 break; 568 } 569 pt->pt_ev.sigev_value.sival_int = timerid; 570 } 571 pt->pt_info.si_signo = pt->pt_ev.sigev_signo; 572 pt->pt_info.si_errno = 0; 573 pt->pt_info.si_code = 0; 574 pt->pt_info.si_pid = p->p_pid; 575 pt->pt_info.si_uid = p->p_cred->p_ruid; 576 pt->pt_info.si_sigval = pt->pt_ev.sigev_value; 577 578 pt->pt_type = id; 579 pt->pt_proc = p; 580 pt->pt_overruns = 0; 581 pt->pt_poverruns = 0; 582 pt->pt_entry = timerid; 583 timerclear(&pt->pt_time.it_value); 584 if (id == CLOCK_REALTIME) 585 callout_init(&pt->pt_ch); 586 else 587 pt->pt_active = 0; 588 589 p->p_timers->pts_timers[timerid] = pt; 590 591 return copyout(&timerid, SCARG(uap, timerid), sizeof(timerid)); 592 } 593 594 595 /* Delete a POSIX realtime timer */ 596 int 597 sys_timer_delete(struct lwp *l, void *v, register_t *retval) 598 { 599 struct sys_timer_delete_args /* { 600 syscallarg(timer_t) timerid; 601 } */ *uap = v; 602 struct proc *p = l->l_proc; 603 timer_t timerid; 604 struct ptimer *pt, *ptn; 605 int s; 606 607 timerid = SCARG(uap, timerid); 608 609 if ((p->p_timers == NULL) || 610 (timerid < 2) || (timerid >= TIMER_MAX) || 611 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 612 return (EINVAL); 613 614 if (pt->pt_type == CLOCK_REALTIME) 615 callout_stop(&pt->pt_ch); 616 else if (pt->pt_active) { 617 s = splclock(); 618 ptn = LIST_NEXT(pt, pt_list); 619 LIST_REMOVE(pt, pt_list); 620 for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list)) 621 timeradd(&pt->pt_time.it_value, &ptn->pt_time.it_value, 622 &ptn->pt_time.it_value); 623 splx(s); 624 } 625 626 p->p_timers->pts_timers[timerid] = NULL; 627 pool_put(&ptimer_pool, pt); 628 629 return (0); 630 } 631 632 /* 633 * Set up the given timer. The value in pt->pt_time.it_value is taken 634 * to be an absolute time for CLOCK_REALTIME timers and a relative 635 * time for virtual timers. 636 * Must be called at splclock(). 637 */ 638 void 639 timer_settime(struct ptimer *pt) 640 { 641 struct ptimer *ptn, *pptn; 642 struct ptlist *ptl; 643 644 if (pt->pt_type == CLOCK_REALTIME) { 645 callout_stop(&pt->pt_ch); 646 if (timerisset(&pt->pt_time.it_value)) { 647 /* 648 * Don't need to check hzto() return value, here. 649 * callout_reset() does it for us. 650 */ 651 callout_reset(&pt->pt_ch, hzto(&pt->pt_time.it_value), 652 realtimerexpire, pt); 653 } 654 } else { 655 if (pt->pt_active) { 656 ptn = LIST_NEXT(pt, pt_list); 657 LIST_REMOVE(pt, pt_list); 658 for ( ; ptn; ptn = LIST_NEXT(ptn, pt_list)) 659 timeradd(&pt->pt_time.it_value, 660 &ptn->pt_time.it_value, 661 &ptn->pt_time.it_value); 662 } 663 if (timerisset(&pt->pt_time.it_value)) { 664 if (pt->pt_type == CLOCK_VIRTUAL) 665 ptl = &pt->pt_proc->p_timers->pts_virtual; 666 else 667 ptl = &pt->pt_proc->p_timers->pts_prof; 668 669 for (ptn = LIST_FIRST(ptl), pptn = NULL; 670 ptn && timercmp(&pt->pt_time.it_value, 671 &ptn->pt_time.it_value, >); 672 pptn = ptn, ptn = LIST_NEXT(ptn, pt_list)) 673 timersub(&pt->pt_time.it_value, 674 &ptn->pt_time.it_value, 675 &pt->pt_time.it_value); 676 677 if (pptn) 678 LIST_INSERT_AFTER(pptn, pt, pt_list); 679 else 680 LIST_INSERT_HEAD(ptl, pt, pt_list); 681 682 for ( ; ptn ; ptn = LIST_NEXT(ptn, pt_list)) 683 timersub(&ptn->pt_time.it_value, 684 &pt->pt_time.it_value, 685 &ptn->pt_time.it_value); 686 687 pt->pt_active = 1; 688 } else 689 pt->pt_active = 0; 690 } 691 } 692 693 void 694 timer_gettime(struct ptimer *pt, struct itimerval *aitv) 695 { 696 struct ptimer *ptn; 697 698 *aitv = pt->pt_time; 699 if (pt->pt_type == CLOCK_REALTIME) { 700 /* 701 * Convert from absolute to relative time in .it_value 702 * part of real time timer. If time for real time 703 * timer has passed return 0, else return difference 704 * between current time and time for the timer to go 705 * off. 706 */ 707 if (timerisset(&aitv->it_value)) { 708 if (timercmp(&aitv->it_value, &time, <)) 709 timerclear(&aitv->it_value); 710 else 711 timersub(&aitv->it_value, &time, 712 &aitv->it_value); 713 } 714 } else if (pt->pt_active) { 715 if (pt->pt_type == CLOCK_VIRTUAL) 716 ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_virtual); 717 else 718 ptn = LIST_FIRST(&pt->pt_proc->p_timers->pts_prof); 719 for ( ; ptn && ptn != pt; ptn = LIST_NEXT(ptn, pt_list)) 720 timeradd(&aitv->it_value, 721 &ptn->pt_time.it_value, &aitv->it_value); 722 KASSERT(ptn != NULL); /* pt should be findable on the list */ 723 } else 724 timerclear(&aitv->it_value); 725 } 726 727 728 729 /* Set and arm a POSIX realtime timer */ 730 int 731 sys_timer_settime(struct lwp *l, void *v, register_t *retval) 732 { 733 struct sys_timer_settime_args /* { 734 syscallarg(timer_t) timerid; 735 syscallarg(int) flags; 736 syscallarg(const struct itimerspec *) value; 737 syscallarg(struct itimerspec *) ovalue; 738 } */ *uap = v; 739 struct proc *p = l->l_proc; 740 int error, s, timerid; 741 struct itimerval val, oval; 742 struct itimerspec value, ovalue; 743 struct ptimer *pt; 744 745 timerid = SCARG(uap, timerid); 746 747 if ((p->p_timers == NULL) || 748 (timerid < 2) || (timerid >= TIMER_MAX) || 749 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 750 return (EINVAL); 751 752 if ((error = copyin(SCARG(uap, value), &value, 753 sizeof(struct itimerspec))) != 0) 754 return (error); 755 756 TIMESPEC_TO_TIMEVAL(&val.it_value, &value.it_value); 757 TIMESPEC_TO_TIMEVAL(&val.it_interval, &value.it_interval); 758 if (itimerfix(&val.it_value) || itimerfix(&val.it_interval)) 759 return (EINVAL); 760 761 oval = pt->pt_time; 762 pt->pt_time = val; 763 764 s = splclock(); 765 /* 766 * If we've been passed a relative time for a realtime timer, 767 * convert it to absolute; if an absolute time for a virtual 768 * timer, convert it to relative and make sure we don't set it 769 * to zero, which would cancel the timer, or let it go 770 * negative, which would confuse the comparison tests. 771 */ 772 if (timerisset(&pt->pt_time.it_value)) { 773 if (pt->pt_type == CLOCK_REALTIME) { 774 if ((SCARG(uap, flags) & TIMER_ABSTIME) == 0) 775 timeradd(&pt->pt_time.it_value, &time, 776 &pt->pt_time.it_value); 777 } else { 778 if ((SCARG(uap, flags) & TIMER_ABSTIME) != 0) { 779 timersub(&pt->pt_time.it_value, &time, 780 &pt->pt_time.it_value); 781 if (!timerisset(&pt->pt_time.it_value) || 782 pt->pt_time.it_value.tv_sec < 0) { 783 pt->pt_time.it_value.tv_sec = 0; 784 pt->pt_time.it_value.tv_usec = 1; 785 } 786 } 787 } 788 } 789 790 timer_settime(pt); 791 splx(s); 792 793 if (SCARG(uap, ovalue)) { 794 TIMEVAL_TO_TIMESPEC(&oval.it_value, &ovalue.it_value); 795 TIMEVAL_TO_TIMESPEC(&oval.it_interval, &ovalue.it_interval); 796 return copyout(&ovalue, SCARG(uap, ovalue), 797 sizeof(struct itimerspec)); 798 } 799 800 return (0); 801 } 802 803 /* Return the time remaining until a POSIX timer fires. */ 804 int 805 sys_timer_gettime(struct lwp *l, void *v, register_t *retval) 806 { 807 struct sys_timer_gettime_args /* { 808 syscallarg(timer_t) timerid; 809 syscallarg(struct itimerspec *) value; 810 } */ *uap = v; 811 struct itimerval aitv; 812 struct itimerspec its; 813 struct proc *p = l->l_proc; 814 int s, timerid; 815 struct ptimer *pt; 816 817 timerid = SCARG(uap, timerid); 818 819 if ((p->p_timers == NULL) || 820 (timerid < 2) || (timerid >= TIMER_MAX) || 821 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 822 return (EINVAL); 823 824 s = splclock(); 825 timer_gettime(pt, &aitv); 826 splx(s); 827 828 TIMEVAL_TO_TIMESPEC(&aitv.it_interval, &its.it_interval); 829 TIMEVAL_TO_TIMESPEC(&aitv.it_value, &its.it_value); 830 831 return copyout(&its, SCARG(uap, value), sizeof(its)); 832 } 833 834 /* 835 * Return the count of the number of times a periodic timer expired 836 * while a notification was already pending. The counter is reset when 837 * a timer expires and a notification can be posted. 838 */ 839 int 840 sys_timer_getoverrun(struct lwp *l, void *v, register_t *retval) 841 { 842 struct sys_timer_getoverrun_args /* { 843 syscallarg(timer_t) timerid; 844 } */ *uap = v; 845 struct proc *p = l->l_proc; 846 int timerid; 847 struct ptimer *pt; 848 849 timerid = SCARG(uap, timerid); 850 851 if ((p->p_timers == NULL) || 852 (timerid < 2) || (timerid >= TIMER_MAX) || 853 ((pt = p->p_timers->pts_timers[timerid]) == NULL)) 854 return (EINVAL); 855 856 *retval = pt->pt_poverruns; 857 858 return (0); 859 } 860 861 /* Glue function that triggers an upcall; called from userret(). */ 862 static void 863 timerupcall(struct lwp *l, void *arg) 864 { 865 struct ptimers *pt = (struct ptimers *)arg; 866 unsigned int i, fired, done; 867 KERNEL_PROC_LOCK(l); 868 869 { 870 struct proc *p = l->l_proc; 871 struct sadata *sa = p->p_sa; 872 873 /* Bail out if we do not own the virtual processor */ 874 if (sa->sa_vp != l) { 875 KERNEL_PROC_UNLOCK(l); 876 return ; 877 } 878 } 879 880 fired = pt->pts_fired; 881 done = 0; 882 while ((i = ffs(fired)) != 0) { 883 i--; 884 if (sa_upcall(l, SA_UPCALL_SIGEV | SA_UPCALL_DEFER, NULL, l, 885 sizeof(siginfo_t), &pt->pts_timers[i]->pt_info) == 0) 886 done |= 1 << i; 887 fired &= ~(1 << i); 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 */ 905 void 906 realtimerexpire(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 */ 940 int 941 sys_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 */ 971 int 972 sys_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_overruns = 0; 1018 pt->pt_proc = p; 1019 pt->pt_type = which; 1020 pt->pt_entry = which; 1021 switch (which) { 1022 case ITIMER_REAL: 1023 callout_init(&pt->pt_ch); 1024 pt->pt_ev.sigev_signo = SIGALRM; 1025 break; 1026 case ITIMER_VIRTUAL: 1027 pt->pt_active = 0; 1028 pt->pt_ev.sigev_signo = SIGVTALRM; 1029 break; 1030 case ITIMER_PROF: 1031 pt->pt_active = 0; 1032 pt->pt_ev.sigev_signo = SIGPROF; 1033 break; 1034 } 1035 } else 1036 pt = p->p_timers->pts_timers[which]; 1037 1038 pt->pt_time = aitv; 1039 p->p_timers->pts_timers[which] = pt; 1040 1041 s = splclock(); 1042 if ((which == ITIMER_REAL) && timerisset(&pt->pt_time.it_value)) { 1043 /* Convert to absolute time */ 1044 timeradd(&pt->pt_time.it_value, &time, &pt->pt_time.it_value); 1045 } 1046 timer_settime(pt); 1047 splx(s); 1048 1049 return (0); 1050 } 1051 1052 /* Utility routines to manage the array of pointers to timers. */ 1053 void 1054 timers_alloc(struct proc *p) 1055 { 1056 int i; 1057 struct ptimers *pts; 1058 1059 pts = malloc(sizeof (struct ptimers), M_SUBPROC, 0); 1060 LIST_INIT(&pts->pts_virtual); 1061 LIST_INIT(&pts->pts_prof); 1062 for (i = 0; i < TIMER_MAX; i++) 1063 pts->pts_timers[i] = NULL; 1064 pts->pts_fired = 0; 1065 p->p_timers = pts; 1066 } 1067 1068 /* 1069 * Clean up the per-process timers. If "which" is set to TIMERS_ALL, 1070 * then clean up all timers and free all the data structures. If 1071 * "which" is set to TIMERS_POSIX, only clean up the timers allocated 1072 * by timer_create(), not the BSD setitimer() timers, and only free the 1073 * structure if none of those remain. 1074 */ 1075 void 1076 timers_free(struct proc *p, int which) 1077 { 1078 int i, s; 1079 struct ptimers *pts; 1080 struct ptimer *pt, *ptn; 1081 struct timeval tv; 1082 1083 if (p->p_timers) { 1084 pts = p->p_timers; 1085 if (which == TIMERS_ALL) 1086 i = 0; 1087 else { 1088 s = splclock(); 1089 timerclear(&tv); 1090 for (ptn = LIST_FIRST(&p->p_timers->pts_virtual); 1091 ptn && ptn != pts->pts_timers[ITIMER_VIRTUAL]; 1092 ptn = LIST_NEXT(ptn, pt_list)) 1093 timeradd(&tv, &ptn->pt_time.it_value, &tv); 1094 LIST_FIRST(&p->p_timers->pts_virtual) = NULL; 1095 if (ptn) { 1096 timeradd(&tv, &ptn->pt_time.it_value, 1097 &ptn->pt_time.it_value); 1098 LIST_INSERT_HEAD(&p->p_timers->pts_virtual, 1099 ptn, pt_list); 1100 } 1101 1102 timerclear(&tv); 1103 for (ptn = LIST_FIRST(&p->p_timers->pts_prof); 1104 ptn && ptn != pts->pts_timers[ITIMER_PROF]; 1105 ptn = LIST_NEXT(ptn, pt_list)) 1106 timeradd(&tv, &ptn->pt_time.it_value, &tv); 1107 LIST_FIRST(&p->p_timers->pts_prof) = NULL; 1108 if (ptn) { 1109 timeradd(&tv, &ptn->pt_time.it_value, 1110 &ptn->pt_time.it_value); 1111 LIST_INSERT_HEAD(&p->p_timers->pts_prof, ptn, 1112 pt_list); 1113 } 1114 splx(s); 1115 i = 3; 1116 } 1117 for ( ; i < TIMER_MAX; i++) 1118 if ((pt = pts->pts_timers[i]) != NULL) { 1119 if (pt->pt_type == CLOCK_REALTIME) 1120 callout_stop(&pt->pt_ch); 1121 pts->pts_timers[i] = NULL; 1122 pool_put(&ptimer_pool, pt); 1123 } 1124 if ((pts->pts_timers[0] == NULL) && 1125 (pts->pts_timers[1] == NULL) && 1126 (pts->pts_timers[2] == NULL)) { 1127 p->p_timers = NULL; 1128 free(pts, M_SUBPROC); 1129 } 1130 } 1131 } 1132 1133 /* 1134 * Check that a proposed value to load into the .it_value or 1135 * .it_interval part of an interval timer is acceptable, and 1136 * fix it to have at least minimal value (i.e. if it is less 1137 * than the resolution of the clock, round it up.) 1138 */ 1139 int 1140 itimerfix(struct timeval *tv) 1141 { 1142 1143 if (tv->tv_sec < 0 || tv->tv_usec < 0 || tv->tv_usec >= 1000000) 1144 return (EINVAL); 1145 if (tv->tv_sec == 0 && tv->tv_usec != 0 && tv->tv_usec < tick) 1146 tv->tv_usec = tick; 1147 return (0); 1148 } 1149 1150 /* 1151 * Decrement an interval timer by a specified number 1152 * of microseconds, which must be less than a second, 1153 * i.e. < 1000000. If the timer expires, then reload 1154 * it. In this case, carry over (usec - old value) to 1155 * reduce the value reloaded into the timer so that 1156 * the timer does not drift. This routine assumes 1157 * that it is called in a context where the timers 1158 * on which it is operating cannot change in value. 1159 */ 1160 int 1161 itimerdecr(struct ptimer *pt, int usec) 1162 { 1163 struct itimerval *itp; 1164 1165 itp = &pt->pt_time; 1166 if (itp->it_value.tv_usec < usec) { 1167 if (itp->it_value.tv_sec == 0) { 1168 /* expired, and already in next interval */ 1169 usec -= itp->it_value.tv_usec; 1170 goto expire; 1171 } 1172 itp->it_value.tv_usec += 1000000; 1173 itp->it_value.tv_sec--; 1174 } 1175 itp->it_value.tv_usec -= usec; 1176 usec = 0; 1177 if (timerisset(&itp->it_value)) 1178 return (1); 1179 /* expired, exactly at end of interval */ 1180 expire: 1181 if (timerisset(&itp->it_interval)) { 1182 itp->it_value = itp->it_interval; 1183 itp->it_value.tv_usec -= usec; 1184 if (itp->it_value.tv_usec < 0) { 1185 itp->it_value.tv_usec += 1000000; 1186 itp->it_value.tv_sec--; 1187 } 1188 timer_settime(pt); 1189 } else 1190 itp->it_value.tv_usec = 0; /* sec is already 0 */ 1191 return (0); 1192 } 1193 1194 void 1195 itimerfire(struct ptimer *pt) 1196 { 1197 struct proc *p = pt->pt_proc; 1198 #if 0 1199 int s; 1200 #endif 1201 if (pt->pt_ev.sigev_notify == SIGEV_SIGNAL) { 1202 /* 1203 * No RT signal infrastructure exists at this time; 1204 * just post the signal number and throw away the 1205 * value. 1206 */ 1207 if (sigismember(&p->p_sigctx.ps_siglist, pt->pt_ev.sigev_signo)) 1208 pt->pt_overruns++; 1209 else { 1210 pt->pt_poverruns = pt->pt_overruns; 1211 pt->pt_overruns = 0; 1212 psignal(p, pt->pt_ev.sigev_signo); 1213 } 1214 } else if (pt->pt_ev.sigev_notify == SIGEV_SA && (p->p_flag & P_SA)) { 1215 /* Cause the process to generate an upcall when it returns. */ 1216 struct sadata *sa = p->p_sa; 1217 unsigned int i; 1218 1219 if (p->p_userret == NULL) { 1220 /* 1221 * XXX stop signals can be processed inside tsleep, 1222 * which can be inside sa_yield's inner loop, which 1223 * makes testing for sa_idle alone insuffucent to 1224 * determine if we really should call setrunnable. 1225 */ 1226 #if 0 1227 1228 if ((sa->sa_idle) && (p->p_stat != SSTOP)) { 1229 SCHED_LOCK(s); 1230 setrunnable(sa->sa_idle); 1231 SCHED_UNLOCK(s); 1232 } 1233 #endif 1234 pt->pt_poverruns = pt->pt_overruns; 1235 pt->pt_overruns = 0; 1236 i = 1 << pt->pt_entry; 1237 p->p_timers->pts_fired = i; 1238 p->p_userret = timerupcall; 1239 p->p_userret_arg = p->p_timers; 1240 1241 if (sa->sa_idle) 1242 wakeup(sa->sa_idle); 1243 1244 } else if (p->p_userret == timerupcall) { 1245 i = 1 << pt->pt_entry; 1246 if ((p->p_timers->pts_fired & i) == 0) { 1247 pt->pt_poverruns = pt->pt_overruns; 1248 pt->pt_overruns = 0; 1249 p->p_timers->pts_fired |= i; 1250 } else 1251 pt->pt_overruns++; 1252 } else { 1253 pt->pt_overruns++; 1254 printf("itimerfire(%d): overrun %d on timer %x (userret is %p)\n", 1255 p->p_pid, pt->pt_overruns, 1256 pt->pt_ev.sigev_value.sival_int, 1257 p->p_userret); 1258 } 1259 } 1260 1261 } 1262 1263 /* 1264 * ratecheck(): simple time-based rate-limit checking. see ratecheck(9) 1265 * for usage and rationale. 1266 */ 1267 int 1268 ratecheck(struct timeval *lasttime, const struct timeval *mininterval) 1269 { 1270 struct timeval tv, delta; 1271 int s, rv = 0; 1272 1273 s = splclock(); 1274 tv = mono_time; 1275 splx(s); 1276 1277 timersub(&tv, lasttime, &delta); 1278 1279 /* 1280 * check for 0,0 is so that the message will be seen at least once, 1281 * even if interval is huge. 1282 */ 1283 if (timercmp(&delta, mininterval, >=) || 1284 (lasttime->tv_sec == 0 && lasttime->tv_usec == 0)) { 1285 *lasttime = tv; 1286 rv = 1; 1287 } 1288 1289 return (rv); 1290 } 1291 1292 /* 1293 * ppsratecheck(): packets (or events) per second limitation. 1294 */ 1295 int 1296 ppsratecheck(struct timeval *lasttime, int *curpps, int maxpps) 1297 { 1298 struct timeval tv, delta; 1299 int s, rv; 1300 1301 s = splclock(); 1302 tv = mono_time; 1303 splx(s); 1304 1305 timersub(&tv, lasttime, &delta); 1306 1307 /* 1308 * check for 0,0 is so that the message will be seen at least once. 1309 * if more than one second have passed since the last update of 1310 * lasttime, reset the counter. 1311 * 1312 * we do increment *curpps even in *curpps < maxpps case, as some may 1313 * try to use *curpps for stat purposes as well. 1314 */ 1315 if ((lasttime->tv_sec == 0 && lasttime->tv_usec == 0) || 1316 delta.tv_sec >= 1) { 1317 *lasttime = tv; 1318 *curpps = 0; 1319 } 1320 if (maxpps < 0) 1321 rv = 1; 1322 else if (*curpps < maxpps) 1323 rv = 1; 1324 else 1325 rv = 0; 1326 1327 #if 1 /*DIAGNOSTIC?*/ 1328 /* be careful about wrap-around */ 1329 if (*curpps + 1 > *curpps) 1330 *curpps = *curpps + 1; 1331 #else 1332 /* 1333 * assume that there's not too many calls to this function. 1334 * not sure if the assumption holds, as it depends on *caller's* 1335 * behavior, not the behavior of this function. 1336 * IMHO it is wrong to make assumption on the caller's behavior, 1337 * so the above #if is #if 1, not #ifdef DIAGNOSTIC. 1338 */ 1339 *curpps = *curpps + 1; 1340 #endif 1341 1342 return (rv); 1343 } 1344