kern_resource.c revision 228207
1145171Sdas/*- 2145171Sdas * Copyright (c) 1982, 1986, 1991, 1993 3145171Sdas * The Regents of the University of California. All rights reserved. 4145171Sdas * (c) UNIX System Laboratories, Inc. 5145171Sdas * All or some portions of this file are derived from material licensed 6145171Sdas * to the University of California by American Telephone and Telegraph 7145171Sdas * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8145171Sdas * the permission of UNIX System Laboratories, Inc. 9145171Sdas * 10145171Sdas * Redistribution and use in source and binary forms, with or without 11145171Sdas * modification, are permitted provided that the following conditions 12145171Sdas * are met: 13145171Sdas * 1. Redistributions of source code must retain the above copyright 14145171Sdas * notice, this list of conditions and the following disclaimer. 15145171Sdas * 2. Redistributions in binary form must reproduce the above copyright 16145171Sdas * notice, this list of conditions and the following disclaimer in the 17145171Sdas * documentation and/or other materials provided with the distribution. 18145171Sdas * 4. Neither the name of the University nor the names of its contributors 19145171Sdas * may be used to endorse or promote products derived from this software 20145171Sdas * without specific prior written permission. 21145171Sdas * 22145171Sdas * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23145171Sdas * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24145171Sdas * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25145171Sdas * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26145171Sdas * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27145171Sdas * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28145171Sdas * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29145171Sdas * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30145171Sdas * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31145171Sdas * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32145171Sdas * SUCH DAMAGE. 33145171Sdas * 34145171Sdas * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94 35145171Sdas */ 36145171Sdas 37145171Sdas#include <sys/cdefs.h> 38145171Sdas__FBSDID("$FreeBSD: head/sys/kern/kern_resource.c 228207 2011-12-02 19:59:46Z jhb $"); 39145171Sdas 40192760Sattilio#include "opt_compat.h" 41150067Sdas 42150067Sdas#include <sys/param.h> 43150067Sdas#include <sys/systm.h> 44217108Skib#include <sys/sysproto.h> 45217108Skib#include <sys/file.h> 46#include <sys/kernel.h> 47#include <sys/lock.h> 48#include <sys/malloc.h> 49#include <sys/mutex.h> 50#include <sys/priv.h> 51#include <sys/proc.h> 52#include <sys/refcount.h> 53#include <sys/racct.h> 54#include <sys/resourcevar.h> 55#include <sys/rwlock.h> 56#include <sys/sched.h> 57#include <sys/sx.h> 58#include <sys/syscallsubr.h> 59#include <sys/sysent.h> 60#include <sys/time.h> 61#include <sys/umtx.h> 62 63#include <vm/vm.h> 64#include <vm/vm_param.h> 65#include <vm/pmap.h> 66#include <vm/vm_map.h> 67 68 69static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures"); 70static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures"); 71#define UIHASH(uid) (&uihashtbl[(uid) & uihash]) 72static struct rwlock uihashtbl_lock; 73static LIST_HEAD(uihashhead, uidinfo) *uihashtbl; 74static u_long uihash; /* size of hash table - 1 */ 75 76static void calcru1(struct proc *p, struct rusage_ext *ruxp, 77 struct timeval *up, struct timeval *sp); 78static int donice(struct thread *td, struct proc *chgp, int n); 79static struct uidinfo *uilookup(uid_t uid); 80static void ruxagg_locked(struct rusage_ext *rux, struct thread *td); 81 82/* 83 * Resource controls and accounting. 84 */ 85#ifndef _SYS_SYSPROTO_H_ 86struct getpriority_args { 87 int which; 88 int who; 89}; 90#endif 91int 92sys_getpriority(td, uap) 93 struct thread *td; 94 register struct getpriority_args *uap; 95{ 96 struct proc *p; 97 struct pgrp *pg; 98 int error, low; 99 100 error = 0; 101 low = PRIO_MAX + 1; 102 switch (uap->which) { 103 104 case PRIO_PROCESS: 105 if (uap->who == 0) 106 low = td->td_proc->p_nice; 107 else { 108 p = pfind(uap->who); 109 if (p == NULL) 110 break; 111 if (p_cansee(td, p) == 0) 112 low = p->p_nice; 113 PROC_UNLOCK(p); 114 } 115 break; 116 117 case PRIO_PGRP: 118 sx_slock(&proctree_lock); 119 if (uap->who == 0) { 120 pg = td->td_proc->p_pgrp; 121 PGRP_LOCK(pg); 122 } else { 123 pg = pgfind(uap->who); 124 if (pg == NULL) { 125 sx_sunlock(&proctree_lock); 126 break; 127 } 128 } 129 sx_sunlock(&proctree_lock); 130 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 131 PROC_LOCK(p); 132 if (p->p_state == PRS_NORMAL && 133 p_cansee(td, p) == 0) { 134 if (p->p_nice < low) 135 low = p->p_nice; 136 } 137 PROC_UNLOCK(p); 138 } 139 PGRP_UNLOCK(pg); 140 break; 141 142 case PRIO_USER: 143 if (uap->who == 0) 144 uap->who = td->td_ucred->cr_uid; 145 sx_slock(&allproc_lock); 146 FOREACH_PROC_IN_SYSTEM(p) { 147 PROC_LOCK(p); 148 if (p->p_state == PRS_NORMAL && 149 p_cansee(td, p) == 0 && 150 p->p_ucred->cr_uid == uap->who) { 151 if (p->p_nice < low) 152 low = p->p_nice; 153 } 154 PROC_UNLOCK(p); 155 } 156 sx_sunlock(&allproc_lock); 157 break; 158 159 default: 160 error = EINVAL; 161 break; 162 } 163 if (low == PRIO_MAX + 1 && error == 0) 164 error = ESRCH; 165 td->td_retval[0] = low; 166 return (error); 167} 168 169#ifndef _SYS_SYSPROTO_H_ 170struct setpriority_args { 171 int which; 172 int who; 173 int prio; 174}; 175#endif 176int 177sys_setpriority(td, uap) 178 struct thread *td; 179 struct setpriority_args *uap; 180{ 181 struct proc *curp, *p; 182 struct pgrp *pg; 183 int found = 0, error = 0; 184 185 curp = td->td_proc; 186 switch (uap->which) { 187 case PRIO_PROCESS: 188 if (uap->who == 0) { 189 PROC_LOCK(curp); 190 error = donice(td, curp, uap->prio); 191 PROC_UNLOCK(curp); 192 } else { 193 p = pfind(uap->who); 194 if (p == NULL) 195 break; 196 error = p_cansee(td, p); 197 if (error == 0) 198 error = donice(td, p, uap->prio); 199 PROC_UNLOCK(p); 200 } 201 found++; 202 break; 203 204 case PRIO_PGRP: 205 sx_slock(&proctree_lock); 206 if (uap->who == 0) { 207 pg = curp->p_pgrp; 208 PGRP_LOCK(pg); 209 } else { 210 pg = pgfind(uap->who); 211 if (pg == NULL) { 212 sx_sunlock(&proctree_lock); 213 break; 214 } 215 } 216 sx_sunlock(&proctree_lock); 217 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 218 PROC_LOCK(p); 219 if (p->p_state == PRS_NORMAL && 220 p_cansee(td, p) == 0) { 221 error = donice(td, p, uap->prio); 222 found++; 223 } 224 PROC_UNLOCK(p); 225 } 226 PGRP_UNLOCK(pg); 227 break; 228 229 case PRIO_USER: 230 if (uap->who == 0) 231 uap->who = td->td_ucred->cr_uid; 232 sx_slock(&allproc_lock); 233 FOREACH_PROC_IN_SYSTEM(p) { 234 PROC_LOCK(p); 235 if (p->p_state == PRS_NORMAL && 236 p->p_ucred->cr_uid == uap->who && 237 p_cansee(td, p) == 0) { 238 error = donice(td, p, uap->prio); 239 found++; 240 } 241 PROC_UNLOCK(p); 242 } 243 sx_sunlock(&allproc_lock); 244 break; 245 246 default: 247 error = EINVAL; 248 break; 249 } 250 if (found == 0 && error == 0) 251 error = ESRCH; 252 return (error); 253} 254 255/* 256 * Set "nice" for a (whole) process. 257 */ 258static int 259donice(struct thread *td, struct proc *p, int n) 260{ 261 int error; 262 263 PROC_LOCK_ASSERT(p, MA_OWNED); 264 if ((error = p_cansched(td, p))) 265 return (error); 266 if (n > PRIO_MAX) 267 n = PRIO_MAX; 268 if (n < PRIO_MIN) 269 n = PRIO_MIN; 270 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0) 271 return (EACCES); 272 sched_nice(p, n); 273 return (0); 274} 275 276/* 277 * Set realtime priority for LWP. 278 */ 279#ifndef _SYS_SYSPROTO_H_ 280struct rtprio_thread_args { 281 int function; 282 lwpid_t lwpid; 283 struct rtprio *rtp; 284}; 285#endif 286int 287sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap) 288{ 289 struct proc *p; 290 struct rtprio rtp; 291 struct thread *td1; 292 int cierror, error; 293 294 /* Perform copyin before acquiring locks if needed. */ 295 if (uap->function == RTP_SET) 296 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 297 else 298 cierror = 0; 299 300 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) { 301 p = td->td_proc; 302 td1 = td; 303 PROC_LOCK(p); 304 } else { 305 /* Only look up thread in current process */ 306 td1 = tdfind(uap->lwpid, curproc->p_pid); 307 if (td1 == NULL) 308 return (ESRCH); 309 p = td1->td_proc; 310 } 311 312 switch (uap->function) { 313 case RTP_LOOKUP: 314 if ((error = p_cansee(td, p))) 315 break; 316 pri_to_rtp(td1, &rtp); 317 PROC_UNLOCK(p); 318 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 319 case RTP_SET: 320 if ((error = p_cansched(td, p)) || (error = cierror)) 321 break; 322 323 /* Disallow setting rtprio in most cases if not superuser. */ 324/* 325 * Realtime priority has to be restricted for reasons which should be 326 * obvious. However, for idle priority, there is a potential for 327 * system deadlock if an idleprio process gains a lock on a resource 328 * that other processes need (and the idleprio process can't run 329 * due to a CPU-bound normal process). Fix me! XXX 330 */ 331#if 0 332 if (RTP_PRIO_IS_REALTIME(rtp.type)) { 333#else 334 if (rtp.type != RTP_PRIO_NORMAL) { 335#endif 336 error = priv_check(td, PRIV_SCHED_RTPRIO); 337 if (error) 338 break; 339 } 340 error = rtp_to_pri(&rtp, td1); 341 break; 342 default: 343 error = EINVAL; 344 break; 345 } 346 PROC_UNLOCK(p); 347 return (error); 348} 349 350/* 351 * Set realtime priority. 352 */ 353#ifndef _SYS_SYSPROTO_H_ 354struct rtprio_args { 355 int function; 356 pid_t pid; 357 struct rtprio *rtp; 358}; 359#endif 360int 361sys_rtprio(td, uap) 362 struct thread *td; /* curthread */ 363 register struct rtprio_args *uap; 364{ 365 struct proc *p; 366 struct thread *tdp; 367 struct rtprio rtp; 368 int cierror, error; 369 370 /* Perform copyin before acquiring locks if needed. */ 371 if (uap->function == RTP_SET) 372 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 373 else 374 cierror = 0; 375 376 if (uap->pid == 0) { 377 p = td->td_proc; 378 PROC_LOCK(p); 379 } else { 380 p = pfind(uap->pid); 381 if (p == NULL) 382 return (ESRCH); 383 } 384 385 switch (uap->function) { 386 case RTP_LOOKUP: 387 if ((error = p_cansee(td, p))) 388 break; 389 /* 390 * Return OUR priority if no pid specified, 391 * or if one is, report the highest priority 392 * in the process. There isn't much more you can do as 393 * there is only room to return a single priority. 394 * Note: specifying our own pid is not the same 395 * as leaving it zero. 396 */ 397 if (uap->pid == 0) { 398 pri_to_rtp(td, &rtp); 399 } else { 400 struct rtprio rtp2; 401 402 rtp.type = RTP_PRIO_IDLE; 403 rtp.prio = RTP_PRIO_MAX; 404 FOREACH_THREAD_IN_PROC(p, tdp) { 405 pri_to_rtp(tdp, &rtp2); 406 if (rtp2.type < rtp.type || 407 (rtp2.type == rtp.type && 408 rtp2.prio < rtp.prio)) { 409 rtp.type = rtp2.type; 410 rtp.prio = rtp2.prio; 411 } 412 } 413 } 414 PROC_UNLOCK(p); 415 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 416 case RTP_SET: 417 if ((error = p_cansched(td, p)) || (error = cierror)) 418 break; 419 420 /* Disallow setting rtprio in most cases if not superuser. */ 421/* 422 * Realtime priority has to be restricted for reasons which should be 423 * obvious. However, for idle priority, there is a potential for 424 * system deadlock if an idleprio process gains a lock on a resource 425 * that other processes need (and the idleprio process can't run 426 * due to a CPU-bound normal process). Fix me! XXX 427 */ 428#if 0 429 if (RTP_PRIO_IS_REALTIME(rtp.type)) { 430#else 431 if (rtp.type != RTP_PRIO_NORMAL) { 432#endif 433 error = priv_check(td, PRIV_SCHED_RTPRIO); 434 if (error) 435 break; 436 } 437 438 /* 439 * If we are setting our own priority, set just our 440 * thread but if we are doing another process, 441 * do all the threads on that process. If we 442 * specify our own pid we do the latter. 443 */ 444 if (uap->pid == 0) { 445 error = rtp_to_pri(&rtp, td); 446 } else { 447 FOREACH_THREAD_IN_PROC(p, td) { 448 if ((error = rtp_to_pri(&rtp, td)) != 0) 449 break; 450 } 451 } 452 break; 453 default: 454 error = EINVAL; 455 break; 456 } 457 PROC_UNLOCK(p); 458 return (error); 459} 460 461int 462rtp_to_pri(struct rtprio *rtp, struct thread *td) 463{ 464 u_char newpri; 465 u_char oldpri; 466 467 switch (RTP_PRIO_BASE(rtp->type)) { 468 case RTP_PRIO_REALTIME: 469 if (rtp->prio > RTP_PRIO_MAX) 470 return (EINVAL); 471 newpri = PRI_MIN_REALTIME + rtp->prio; 472 break; 473 case RTP_PRIO_NORMAL: 474 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) 475 return (EINVAL); 476 newpri = PRI_MIN_TIMESHARE + rtp->prio; 477 break; 478 case RTP_PRIO_IDLE: 479 if (rtp->prio > RTP_PRIO_MAX) 480 return (EINVAL); 481 newpri = PRI_MIN_IDLE + rtp->prio; 482 break; 483 default: 484 return (EINVAL); 485 } 486 487 thread_lock(td); 488 sched_class(td, rtp->type); /* XXX fix */ 489 oldpri = td->td_user_pri; 490 sched_user_prio(td, newpri); 491 if (td->td_user_pri != oldpri && (td == curthread || 492 td->td_priority == oldpri || td->td_user_pri >= PRI_MAX_REALTIME)) 493 sched_prio(td, td->td_user_pri); 494 if (TD_ON_UPILOCK(td) && oldpri != newpri) { 495 critical_enter(); 496 thread_unlock(td); 497 umtx_pi_adjust(td, oldpri); 498 critical_exit(); 499 } else 500 thread_unlock(td); 501 return (0); 502} 503 504void 505pri_to_rtp(struct thread *td, struct rtprio *rtp) 506{ 507 508 thread_lock(td); 509 switch (PRI_BASE(td->td_pri_class)) { 510 case PRI_REALTIME: 511 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME; 512 break; 513 case PRI_TIMESHARE: 514 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE; 515 break; 516 case PRI_IDLE: 517 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE; 518 break; 519 default: 520 break; 521 } 522 rtp->type = td->td_pri_class; 523 thread_unlock(td); 524} 525 526#if defined(COMPAT_43) 527#ifndef _SYS_SYSPROTO_H_ 528struct osetrlimit_args { 529 u_int which; 530 struct orlimit *rlp; 531}; 532#endif 533int 534osetrlimit(td, uap) 535 struct thread *td; 536 register struct osetrlimit_args *uap; 537{ 538 struct orlimit olim; 539 struct rlimit lim; 540 int error; 541 542 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit)))) 543 return (error); 544 lim.rlim_cur = olim.rlim_cur; 545 lim.rlim_max = olim.rlim_max; 546 error = kern_setrlimit(td, uap->which, &lim); 547 return (error); 548} 549 550#ifndef _SYS_SYSPROTO_H_ 551struct ogetrlimit_args { 552 u_int which; 553 struct orlimit *rlp; 554}; 555#endif 556int 557ogetrlimit(td, uap) 558 struct thread *td; 559 register struct ogetrlimit_args *uap; 560{ 561 struct orlimit olim; 562 struct rlimit rl; 563 struct proc *p; 564 int error; 565 566 if (uap->which >= RLIM_NLIMITS) 567 return (EINVAL); 568 p = td->td_proc; 569 PROC_LOCK(p); 570 lim_rlimit(p, uap->which, &rl); 571 PROC_UNLOCK(p); 572 573 /* 574 * XXX would be more correct to convert only RLIM_INFINITY to the 575 * old RLIM_INFINITY and fail with EOVERFLOW for other larger 576 * values. Most 64->32 and 32->16 conversions, including not 577 * unimportant ones of uids are even more broken than what we 578 * do here (they blindly truncate). We don't do this correctly 579 * here since we have little experience with EOVERFLOW yet. 580 * Elsewhere, getuid() can't fail... 581 */ 582 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur; 583 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max; 584 error = copyout(&olim, uap->rlp, sizeof(olim)); 585 return (error); 586} 587#endif /* COMPAT_43 */ 588 589#ifndef _SYS_SYSPROTO_H_ 590struct __setrlimit_args { 591 u_int which; 592 struct rlimit *rlp; 593}; 594#endif 595int 596sys_setrlimit(td, uap) 597 struct thread *td; 598 register struct __setrlimit_args *uap; 599{ 600 struct rlimit alim; 601 int error; 602 603 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit)))) 604 return (error); 605 error = kern_setrlimit(td, uap->which, &alim); 606 return (error); 607} 608 609static void 610lim_cb(void *arg) 611{ 612 struct rlimit rlim; 613 struct thread *td; 614 struct proc *p; 615 616 p = arg; 617 PROC_LOCK_ASSERT(p, MA_OWNED); 618 /* 619 * Check if the process exceeds its cpu resource allocation. If 620 * it reaches the max, arrange to kill the process in ast(). 621 */ 622 if (p->p_cpulimit == RLIM_INFINITY) 623 return; 624 PROC_SLOCK(p); 625 FOREACH_THREAD_IN_PROC(p, td) { 626 ruxagg(p, td); 627 } 628 PROC_SUNLOCK(p); 629 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) { 630 lim_rlimit(p, RLIMIT_CPU, &rlim); 631 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) { 632 killproc(p, "exceeded maximum CPU limit"); 633 } else { 634 if (p->p_cpulimit < rlim.rlim_max) 635 p->p_cpulimit += 5; 636 kern_psignal(p, SIGXCPU); 637 } 638 } 639 if ((p->p_flag & P_WEXIT) == 0) 640 callout_reset(&p->p_limco, hz, lim_cb, p); 641} 642 643int 644kern_setrlimit(td, which, limp) 645 struct thread *td; 646 u_int which; 647 struct rlimit *limp; 648{ 649 struct plimit *newlim, *oldlim; 650 struct proc *p; 651 register struct rlimit *alimp; 652 struct rlimit oldssiz; 653 int error; 654 655 if (which >= RLIM_NLIMITS) 656 return (EINVAL); 657 658 /* 659 * Preserve historical bugs by treating negative limits as unsigned. 660 */ 661 if (limp->rlim_cur < 0) 662 limp->rlim_cur = RLIM_INFINITY; 663 if (limp->rlim_max < 0) 664 limp->rlim_max = RLIM_INFINITY; 665 666 oldssiz.rlim_cur = 0; 667 p = td->td_proc; 668 newlim = lim_alloc(); 669 PROC_LOCK(p); 670 oldlim = p->p_limit; 671 alimp = &oldlim->pl_rlimit[which]; 672 if (limp->rlim_cur > alimp->rlim_max || 673 limp->rlim_max > alimp->rlim_max) 674 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { 675 PROC_UNLOCK(p); 676 lim_free(newlim); 677 return (error); 678 } 679 if (limp->rlim_cur > limp->rlim_max) 680 limp->rlim_cur = limp->rlim_max; 681 lim_copy(newlim, oldlim); 682 alimp = &newlim->pl_rlimit[which]; 683 684 switch (which) { 685 686 case RLIMIT_CPU: 687 if (limp->rlim_cur != RLIM_INFINITY && 688 p->p_cpulimit == RLIM_INFINITY) 689 callout_reset(&p->p_limco, hz, lim_cb, p); 690 p->p_cpulimit = limp->rlim_cur; 691 break; 692 case RLIMIT_DATA: 693 if (limp->rlim_cur > maxdsiz) 694 limp->rlim_cur = maxdsiz; 695 if (limp->rlim_max > maxdsiz) 696 limp->rlim_max = maxdsiz; 697 break; 698 699 case RLIMIT_STACK: 700 if (limp->rlim_cur > maxssiz) 701 limp->rlim_cur = maxssiz; 702 if (limp->rlim_max > maxssiz) 703 limp->rlim_max = maxssiz; 704 oldssiz = *alimp; 705 if (p->p_sysent->sv_fixlimit != NULL) 706 p->p_sysent->sv_fixlimit(&oldssiz, 707 RLIMIT_STACK); 708 break; 709 710 case RLIMIT_NOFILE: 711 if (limp->rlim_cur > maxfilesperproc) 712 limp->rlim_cur = maxfilesperproc; 713 if (limp->rlim_max > maxfilesperproc) 714 limp->rlim_max = maxfilesperproc; 715 break; 716 717 case RLIMIT_NPROC: 718 if (limp->rlim_cur > maxprocperuid) 719 limp->rlim_cur = maxprocperuid; 720 if (limp->rlim_max > maxprocperuid) 721 limp->rlim_max = maxprocperuid; 722 if (limp->rlim_cur < 1) 723 limp->rlim_cur = 1; 724 if (limp->rlim_max < 1) 725 limp->rlim_max = 1; 726 break; 727 } 728 if (p->p_sysent->sv_fixlimit != NULL) 729 p->p_sysent->sv_fixlimit(limp, which); 730 *alimp = *limp; 731 p->p_limit = newlim; 732 PROC_UNLOCK(p); 733 lim_free(oldlim); 734 735 if (which == RLIMIT_STACK) { 736 /* 737 * Stack is allocated to the max at exec time with only 738 * "rlim_cur" bytes accessible. If stack limit is going 739 * up make more accessible, if going down make inaccessible. 740 */ 741 if (limp->rlim_cur != oldssiz.rlim_cur) { 742 vm_offset_t addr; 743 vm_size_t size; 744 vm_prot_t prot; 745 746 if (limp->rlim_cur > oldssiz.rlim_cur) { 747 prot = p->p_sysent->sv_stackprot; 748 size = limp->rlim_cur - oldssiz.rlim_cur; 749 addr = p->p_sysent->sv_usrstack - 750 limp->rlim_cur; 751 } else { 752 prot = VM_PROT_NONE; 753 size = oldssiz.rlim_cur - limp->rlim_cur; 754 addr = p->p_sysent->sv_usrstack - 755 oldssiz.rlim_cur; 756 } 757 addr = trunc_page(addr); 758 size = round_page(size); 759 (void)vm_map_protect(&p->p_vmspace->vm_map, 760 addr, addr + size, prot, FALSE); 761 } 762 } 763 764 return (0); 765} 766 767#ifndef _SYS_SYSPROTO_H_ 768struct __getrlimit_args { 769 u_int which; 770 struct rlimit *rlp; 771}; 772#endif 773/* ARGSUSED */ 774int 775sys_getrlimit(td, uap) 776 struct thread *td; 777 register struct __getrlimit_args *uap; 778{ 779 struct rlimit rlim; 780 struct proc *p; 781 int error; 782 783 if (uap->which >= RLIM_NLIMITS) 784 return (EINVAL); 785 p = td->td_proc; 786 PROC_LOCK(p); 787 lim_rlimit(p, uap->which, &rlim); 788 PROC_UNLOCK(p); 789 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); 790 return (error); 791} 792 793/* 794 * Transform the running time and tick information for children of proc p 795 * into user and system time usage. 796 */ 797void 798calccru(p, up, sp) 799 struct proc *p; 800 struct timeval *up; 801 struct timeval *sp; 802{ 803 804 PROC_LOCK_ASSERT(p, MA_OWNED); 805 calcru1(p, &p->p_crux, up, sp); 806} 807 808/* 809 * Transform the running time and tick information in proc p into user 810 * and system time usage. If appropriate, include the current time slice 811 * on this CPU. 812 */ 813void 814calcru(struct proc *p, struct timeval *up, struct timeval *sp) 815{ 816 struct thread *td; 817 uint64_t runtime, u; 818 819 PROC_LOCK_ASSERT(p, MA_OWNED); 820 PROC_SLOCK_ASSERT(p, MA_OWNED); 821 /* 822 * If we are getting stats for the current process, then add in the 823 * stats that this thread has accumulated in its current time slice. 824 * We reset the thread and CPU state as if we had performed a context 825 * switch right here. 826 */ 827 td = curthread; 828 if (td->td_proc == p) { 829 u = cpu_ticks(); 830 runtime = u - PCPU_GET(switchtime); 831 td->td_runtime += runtime; 832 td->td_incruntime += runtime; 833 PCPU_SET(switchtime, u); 834 } 835 /* Make sure the per-thread stats are current. */ 836 FOREACH_THREAD_IN_PROC(p, td) { 837 if (td->td_incruntime == 0) 838 continue; 839 ruxagg(p, td); 840 } 841 calcru1(p, &p->p_rux, up, sp); 842} 843 844/* Collect resource usage for a single thread. */ 845void 846rufetchtd(struct thread *td, struct rusage *ru) 847{ 848 struct proc *p; 849 uint64_t runtime, u; 850 851 p = td->td_proc; 852 PROC_SLOCK_ASSERT(p, MA_OWNED); 853 THREAD_LOCK_ASSERT(td, MA_OWNED); 854 /* 855 * If we are getting stats for the current thread, then add in the 856 * stats that this thread has accumulated in its current time slice. 857 * We reset the thread and CPU state as if we had performed a context 858 * switch right here. 859 */ 860 if (td == curthread) { 861 u = cpu_ticks(); 862 runtime = u - PCPU_GET(switchtime); 863 td->td_runtime += runtime; 864 td->td_incruntime += runtime; 865 PCPU_SET(switchtime, u); 866 } 867 ruxagg(p, td); 868 *ru = td->td_ru; 869 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); 870} 871 872static void 873calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, 874 struct timeval *sp) 875{ 876 /* {user, system, interrupt, total} {ticks, usec}: */ 877 uint64_t ut, uu, st, su, it, tt, tu; 878 879 ut = ruxp->rux_uticks; 880 st = ruxp->rux_sticks; 881 it = ruxp->rux_iticks; 882 tt = ut + st + it; 883 if (tt == 0) { 884 /* Avoid divide by zero */ 885 st = 1; 886 tt = 1; 887 } 888 tu = cputick2usec(ruxp->rux_runtime); 889 if ((int64_t)tu < 0) { 890 /* XXX: this should be an assert /phk */ 891 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", 892 (intmax_t)tu, p->p_pid, p->p_comm); 893 tu = ruxp->rux_tu; 894 } 895 896 if (tu >= ruxp->rux_tu) { 897 /* 898 * The normal case, time increased. 899 * Enforce monotonicity of bucketed numbers. 900 */ 901 uu = (tu * ut) / tt; 902 if (uu < ruxp->rux_uu) 903 uu = ruxp->rux_uu; 904 su = (tu * st) / tt; 905 if (su < ruxp->rux_su) 906 su = ruxp->rux_su; 907 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { 908 /* 909 * When we calibrate the cputicker, it is not uncommon to 910 * see the presumably fixed frequency increase slightly over 911 * time as a result of thermal stabilization and NTP 912 * discipline (of the reference clock). We therefore ignore 913 * a bit of backwards slop because we expect to catch up 914 * shortly. We use a 3 microsecond limit to catch low 915 * counts and a 1% limit for high counts. 916 */ 917 uu = ruxp->rux_uu; 918 su = ruxp->rux_su; 919 tu = ruxp->rux_tu; 920 } else { /* tu < ruxp->rux_tu */ 921 /* 922 * What happened here was likely that a laptop, which ran at 923 * a reduced clock frequency at boot, kicked into high gear. 924 * The wisdom of spamming this message in that case is 925 * dubious, but it might also be indicative of something 926 * serious, so lets keep it and hope laptops can be made 927 * more truthful about their CPU speed via ACPI. 928 */ 929 printf("calcru: runtime went backwards from %ju usec " 930 "to %ju usec for pid %d (%s)\n", 931 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, 932 p->p_pid, p->p_comm); 933 uu = (tu * ut) / tt; 934 su = (tu * st) / tt; 935 } 936 937 ruxp->rux_uu = uu; 938 ruxp->rux_su = su; 939 ruxp->rux_tu = tu; 940 941 up->tv_sec = uu / 1000000; 942 up->tv_usec = uu % 1000000; 943 sp->tv_sec = su / 1000000; 944 sp->tv_usec = su % 1000000; 945} 946 947#ifndef _SYS_SYSPROTO_H_ 948struct getrusage_args { 949 int who; 950 struct rusage *rusage; 951}; 952#endif 953int 954sys_getrusage(td, uap) 955 register struct thread *td; 956 register struct getrusage_args *uap; 957{ 958 struct rusage ru; 959 int error; 960 961 error = kern_getrusage(td, uap->who, &ru); 962 if (error == 0) 963 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 964 return (error); 965} 966 967int 968kern_getrusage(struct thread *td, int who, struct rusage *rup) 969{ 970 struct proc *p; 971 int error; 972 973 error = 0; 974 p = td->td_proc; 975 PROC_LOCK(p); 976 switch (who) { 977 case RUSAGE_SELF: 978 rufetchcalc(p, rup, &rup->ru_utime, 979 &rup->ru_stime); 980 break; 981 982 case RUSAGE_CHILDREN: 983 *rup = p->p_stats->p_cru; 984 calccru(p, &rup->ru_utime, &rup->ru_stime); 985 break; 986 987 case RUSAGE_THREAD: 988 PROC_SLOCK(p); 989 thread_lock(td); 990 rufetchtd(td, rup); 991 thread_unlock(td); 992 PROC_SUNLOCK(p); 993 break; 994 995 default: 996 error = EINVAL; 997 } 998 PROC_UNLOCK(p); 999 return (error); 1000} 1001 1002void 1003rucollect(struct rusage *ru, struct rusage *ru2) 1004{ 1005 long *ip, *ip2; 1006 int i; 1007 1008 if (ru->ru_maxrss < ru2->ru_maxrss) 1009 ru->ru_maxrss = ru2->ru_maxrss; 1010 ip = &ru->ru_first; 1011 ip2 = &ru2->ru_first; 1012 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) 1013 *ip++ += *ip2++; 1014} 1015 1016void 1017ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, 1018 struct rusage_ext *rux2) 1019{ 1020 1021 rux->rux_runtime += rux2->rux_runtime; 1022 rux->rux_uticks += rux2->rux_uticks; 1023 rux->rux_sticks += rux2->rux_sticks; 1024 rux->rux_iticks += rux2->rux_iticks; 1025 rux->rux_uu += rux2->rux_uu; 1026 rux->rux_su += rux2->rux_su; 1027 rux->rux_tu += rux2->rux_tu; 1028 rucollect(ru, ru2); 1029} 1030 1031/* 1032 * Aggregate tick counts into the proc's rusage_ext. 1033 */ 1034static void 1035ruxagg_locked(struct rusage_ext *rux, struct thread *td) 1036{ 1037 1038 THREAD_LOCK_ASSERT(td, MA_OWNED); 1039 PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED); 1040 rux->rux_runtime += td->td_incruntime; 1041 rux->rux_uticks += td->td_uticks; 1042 rux->rux_sticks += td->td_sticks; 1043 rux->rux_iticks += td->td_iticks; 1044} 1045 1046void 1047ruxagg(struct proc *p, struct thread *td) 1048{ 1049 1050 thread_lock(td); 1051 ruxagg_locked(&p->p_rux, td); 1052 ruxagg_locked(&td->td_rux, td); 1053 td->td_incruntime = 0; 1054 td->td_uticks = 0; 1055 td->td_iticks = 0; 1056 td->td_sticks = 0; 1057 thread_unlock(td); 1058} 1059 1060/* 1061 * Update the rusage_ext structure and fetch a valid aggregate rusage 1062 * for proc p if storage for one is supplied. 1063 */ 1064void 1065rufetch(struct proc *p, struct rusage *ru) 1066{ 1067 struct thread *td; 1068 1069 PROC_SLOCK_ASSERT(p, MA_OWNED); 1070 1071 *ru = p->p_ru; 1072 if (p->p_numthreads > 0) { 1073 FOREACH_THREAD_IN_PROC(p, td) { 1074 ruxagg(p, td); 1075 rucollect(ru, &td->td_ru); 1076 } 1077 } 1078} 1079 1080/* 1081 * Atomically perform a rufetch and a calcru together. 1082 * Consumers, can safely assume the calcru is executed only once 1083 * rufetch is completed. 1084 */ 1085void 1086rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, 1087 struct timeval *sp) 1088{ 1089 1090 PROC_SLOCK(p); 1091 rufetch(p, ru); 1092 calcru(p, up, sp); 1093 PROC_SUNLOCK(p); 1094} 1095 1096/* 1097 * Allocate a new resource limits structure and initialize its 1098 * reference count and mutex pointer. 1099 */ 1100struct plimit * 1101lim_alloc() 1102{ 1103 struct plimit *limp; 1104 1105 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); 1106 refcount_init(&limp->pl_refcnt, 1); 1107 return (limp); 1108} 1109 1110struct plimit * 1111lim_hold(limp) 1112 struct plimit *limp; 1113{ 1114 1115 refcount_acquire(&limp->pl_refcnt); 1116 return (limp); 1117} 1118 1119void 1120lim_fork(struct proc *p1, struct proc *p2) 1121{ 1122 1123 PROC_LOCK_ASSERT(p1, MA_OWNED); 1124 PROC_LOCK_ASSERT(p2, MA_OWNED); 1125 1126 p2->p_limit = lim_hold(p1->p_limit); 1127 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); 1128 if (p1->p_cpulimit != RLIM_INFINITY) 1129 callout_reset(&p2->p_limco, hz, lim_cb, p2); 1130} 1131 1132void 1133lim_free(limp) 1134 struct plimit *limp; 1135{ 1136 1137 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow")); 1138 if (refcount_release(&limp->pl_refcnt)) 1139 free((void *)limp, M_PLIMIT); 1140} 1141 1142/* 1143 * Make a copy of the plimit structure. 1144 * We share these structures copy-on-write after fork. 1145 */ 1146void 1147lim_copy(dst, src) 1148 struct plimit *dst, *src; 1149{ 1150 1151 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit")); 1152 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); 1153} 1154 1155/* 1156 * Return the hard limit for a particular system resource. The 1157 * which parameter specifies the index into the rlimit array. 1158 */ 1159rlim_t 1160lim_max(struct proc *p, int which) 1161{ 1162 struct rlimit rl; 1163 1164 lim_rlimit(p, which, &rl); 1165 return (rl.rlim_max); 1166} 1167 1168/* 1169 * Return the current (soft) limit for a particular system resource. 1170 * The which parameter which specifies the index into the rlimit array 1171 */ 1172rlim_t 1173lim_cur(struct proc *p, int which) 1174{ 1175 struct rlimit rl; 1176 1177 lim_rlimit(p, which, &rl); 1178 return (rl.rlim_cur); 1179} 1180 1181/* 1182 * Return a copy of the entire rlimit structure for the system limit 1183 * specified by 'which' in the rlimit structure pointed to by 'rlp'. 1184 */ 1185void 1186lim_rlimit(struct proc *p, int which, struct rlimit *rlp) 1187{ 1188 1189 PROC_LOCK_ASSERT(p, MA_OWNED); 1190 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1191 ("request for invalid resource limit")); 1192 *rlp = p->p_limit->pl_rlimit[which]; 1193 if (p->p_sysent->sv_fixlimit != NULL) 1194 p->p_sysent->sv_fixlimit(rlp, which); 1195} 1196 1197void 1198uihashinit() 1199{ 1200 1201 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); 1202 rw_init(&uihashtbl_lock, "uidinfo hash"); 1203} 1204 1205/* 1206 * Look up a uidinfo struct for the parameter uid. 1207 * uihashtbl_lock must be locked. 1208 */ 1209static struct uidinfo * 1210uilookup(uid) 1211 uid_t uid; 1212{ 1213 struct uihashhead *uipp; 1214 struct uidinfo *uip; 1215 1216 rw_assert(&uihashtbl_lock, RA_LOCKED); 1217 uipp = UIHASH(uid); 1218 LIST_FOREACH(uip, uipp, ui_hash) 1219 if (uip->ui_uid == uid) 1220 break; 1221 1222 return (uip); 1223} 1224 1225/* 1226 * Find or allocate a struct uidinfo for a particular uid. 1227 * Increase refcount on uidinfo struct returned. 1228 * uifree() should be called on a struct uidinfo when released. 1229 */ 1230struct uidinfo * 1231uifind(uid) 1232 uid_t uid; 1233{ 1234 struct uidinfo *old_uip, *uip; 1235 1236 rw_rlock(&uihashtbl_lock); 1237 uip = uilookup(uid); 1238 if (uip == NULL) { 1239 rw_runlock(&uihashtbl_lock); 1240 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO); 1241 racct_create(&uip->ui_racct); 1242 rw_wlock(&uihashtbl_lock); 1243 /* 1244 * There's a chance someone created our uidinfo while we 1245 * were in malloc and not holding the lock, so we have to 1246 * make sure we don't insert a duplicate uidinfo. 1247 */ 1248 if ((old_uip = uilookup(uid)) != NULL) { 1249 /* Someone else beat us to it. */ 1250 racct_destroy(&uip->ui_racct); 1251 free(uip, M_UIDINFO); 1252 uip = old_uip; 1253 } else { 1254 refcount_init(&uip->ui_ref, 0); 1255 uip->ui_uid = uid; 1256 mtx_init(&uip->ui_vmsize_mtx, "ui_vmsize", NULL, 1257 MTX_DEF); 1258 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash); 1259 } 1260 } 1261 uihold(uip); 1262 rw_unlock(&uihashtbl_lock); 1263 return (uip); 1264} 1265 1266/* 1267 * Place another refcount on a uidinfo struct. 1268 */ 1269void 1270uihold(uip) 1271 struct uidinfo *uip; 1272{ 1273 1274 refcount_acquire(&uip->ui_ref); 1275} 1276 1277/*- 1278 * Since uidinfo structs have a long lifetime, we use an 1279 * opportunistic refcounting scheme to avoid locking the lookup hash 1280 * for each release. 1281 * 1282 * If the refcount hits 0, we need to free the structure, 1283 * which means we need to lock the hash. 1284 * Optimal case: 1285 * After locking the struct and lowering the refcount, if we find 1286 * that we don't need to free, simply unlock and return. 1287 * Suboptimal case: 1288 * If refcount lowering results in need to free, bump the count 1289 * back up, lose the lock and acquire the locks in the proper 1290 * order to try again. 1291 */ 1292void 1293uifree(uip) 1294 struct uidinfo *uip; 1295{ 1296 int old; 1297 1298 /* Prepare for optimal case. */ 1299 old = uip->ui_ref; 1300 if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1)) 1301 return; 1302 1303 /* Prepare for suboptimal case. */ 1304 rw_wlock(&uihashtbl_lock); 1305 if (refcount_release(&uip->ui_ref)) { 1306 racct_destroy(&uip->ui_racct); 1307 LIST_REMOVE(uip, ui_hash); 1308 rw_wunlock(&uihashtbl_lock); 1309 if (uip->ui_sbsize != 0) 1310 printf("freeing uidinfo: uid = %d, sbsize = %ld\n", 1311 uip->ui_uid, uip->ui_sbsize); 1312 if (uip->ui_proccnt != 0) 1313 printf("freeing uidinfo: uid = %d, proccnt = %ld\n", 1314 uip->ui_uid, uip->ui_proccnt); 1315 if (uip->ui_vmsize != 0) 1316 printf("freeing uidinfo: uid = %d, swapuse = %lld\n", 1317 uip->ui_uid, (unsigned long long)uip->ui_vmsize); 1318 mtx_destroy(&uip->ui_vmsize_mtx); 1319 free(uip, M_UIDINFO); 1320 return; 1321 } 1322 /* 1323 * Someone added a reference between atomic_cmpset_int() and 1324 * rw_wlock(&uihashtbl_lock). 1325 */ 1326 rw_wunlock(&uihashtbl_lock); 1327} 1328 1329void 1330ui_racct_foreach(void (*callback)(struct racct *racct, 1331 void *arg2, void *arg3), void *arg2, void *arg3) 1332{ 1333 struct uidinfo *uip; 1334 struct uihashhead *uih; 1335 1336 rw_rlock(&uihashtbl_lock); 1337 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { 1338 LIST_FOREACH(uip, uih, ui_hash) { 1339 (callback)(uip->ui_racct, arg2, arg3); 1340 } 1341 } 1342 rw_runlock(&uihashtbl_lock); 1343} 1344 1345/* 1346 * Change the count associated with number of processes 1347 * a given user is using. When 'max' is 0, don't enforce a limit 1348 */ 1349int 1350chgproccnt(uip, diff, max) 1351 struct uidinfo *uip; 1352 int diff; 1353 rlim_t max; 1354{ 1355 1356 /* Don't allow them to exceed max, but allow subtraction. */ 1357 if (diff > 0 && max != 0) { 1358 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) { 1359 atomic_subtract_long(&uip->ui_proccnt, (long)diff); 1360 return (0); 1361 } 1362 } else { 1363 atomic_add_long(&uip->ui_proccnt, (long)diff); 1364 if (uip->ui_proccnt < 0) 1365 printf("negative proccnt for uid = %d\n", uip->ui_uid); 1366 } 1367 return (1); 1368} 1369 1370/* 1371 * Change the total socket buffer size a user has used. 1372 */ 1373int 1374chgsbsize(uip, hiwat, to, max) 1375 struct uidinfo *uip; 1376 u_int *hiwat; 1377 u_int to; 1378 rlim_t max; 1379{ 1380 int diff; 1381 1382 diff = to - *hiwat; 1383 if (diff > 0) { 1384 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) { 1385 atomic_subtract_long(&uip->ui_sbsize, (long)diff); 1386 return (0); 1387 } 1388 } else { 1389 atomic_add_long(&uip->ui_sbsize, (long)diff); 1390 if (uip->ui_sbsize < 0) 1391 printf("negative sbsize for uid = %d\n", uip->ui_uid); 1392 } 1393 *hiwat = to; 1394 return (1); 1395} 1396 1397/* 1398 * Change the count associated with number of pseudo-terminals 1399 * a given user is using. When 'max' is 0, don't enforce a limit 1400 */ 1401int 1402chgptscnt(uip, diff, max) 1403 struct uidinfo *uip; 1404 int diff; 1405 rlim_t max; 1406{ 1407 1408 /* Don't allow them to exceed max, but allow subtraction. */ 1409 if (diff > 0 && max != 0) { 1410 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) { 1411 atomic_subtract_long(&uip->ui_ptscnt, (long)diff); 1412 return (0); 1413 } 1414 } else { 1415 atomic_add_long(&uip->ui_ptscnt, (long)diff); 1416 if (uip->ui_ptscnt < 0) 1417 printf("negative ptscnt for uid = %d\n", uip->ui_uid); 1418 } 1419 return (1); 1420} 1421