2 * Copyright (c) 1982, 1986, 1991, 1993
3 * The Regents of the University of California. All rights reserved.
4 * (c) UNIX System Laboratories, Inc.
5 * All or some portions of this file are derived from material licensed
6 * to the University of California by American Telephone and Telegraph
7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
8 * the permission of UNIX System Laboratories, Inc.
10 * Redistribution and use in source and binary forms, with or without
11 * modification, are permitted provided that the following conditions
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 * 4. Neither the name of the University nor the names of its contributors
19 * may be used to endorse or promote products derived from this software
20 * without specific prior written permission.
22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
34 * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94
37 #include <sys/cdefs.h>
38 __FBSDID("$FreeBSD$");
40 #include "opt_compat.h"
42 #include <sys/param.h>
43 #include <sys/systm.h>
44 #include <sys/sysproto.h>
46 #include <sys/kernel.h>
48 #include <sys/malloc.h>
49 #include <sys/mutex.h>
52 #include <sys/refcount.h>
53 #include <sys/resourcevar.h>
54 #include <sys/rwlock.h>
55 #include <sys/sched.h>
57 #include <sys/syscallsubr.h>
58 #include <sys/sysent.h>
63 #include <vm/vm_param.h>
65 #include <vm/vm_map.h>
68 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
69 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
70 #define UIHASH(uid) (&uihashtbl[(uid) & uihash])
71 static struct rwlock uihashtbl_lock;
72 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
73 static u_long uihash; /* size of hash table - 1 */
75 static void calcru1(struct proc *p, struct rusage_ext *ruxp,
76 struct timeval *up, struct timeval *sp);
77 static int donice(struct thread *td, struct proc *chgp, int n);
78 static struct uidinfo *uilookup(uid_t uid);
79 static void ruxagg_locked(struct rusage_ext *rux, struct thread *td);
82 * Resource controls and accounting.
84 #ifndef _SYS_SYSPROTO_H_
85 struct getpriority_args {
93 register struct getpriority_args *uap;
101 switch (uap->which) {
105 low = td->td_proc->p_nice;
110 if (p_cansee(td, p) == 0)
117 sx_slock(&proctree_lock);
119 pg = td->td_proc->p_pgrp;
122 pg = pgfind(uap->who);
124 sx_sunlock(&proctree_lock);
128 sx_sunlock(&proctree_lock);
129 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
131 if (p_cansee(td, p) == 0) {
142 uap->who = td->td_ucred->cr_uid;
143 sx_slock(&allproc_lock);
144 FOREACH_PROC_IN_SYSTEM(p) {
145 /* Do not bother to check PRS_NEW processes */
146 if (p->p_state == PRS_NEW)
149 if (p_cansee(td, p) == 0 &&
150 p->p_ucred->cr_uid == uap->who) {
156 sx_sunlock(&allproc_lock);
163 if (low == PRIO_MAX + 1 && error == 0)
165 td->td_retval[0] = low;
169 #ifndef _SYS_SYSPROTO_H_
170 struct setpriority_args {
179 struct setpriority_args *uap;
181 struct proc *curp, *p;
183 int found = 0, error = 0;
186 switch (uap->which) {
190 error = donice(td, curp, uap->prio);
196 error = p_cansee(td, p);
198 error = donice(td, p, uap->prio);
205 sx_slock(&proctree_lock);
210 pg = pgfind(uap->who);
212 sx_sunlock(&proctree_lock);
216 sx_sunlock(&proctree_lock);
217 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
219 if (p_cansee(td, p) == 0) {
220 error = donice(td, p, uap->prio);
230 uap->who = td->td_ucred->cr_uid;
231 sx_slock(&allproc_lock);
232 FOREACH_PROC_IN_SYSTEM(p) {
234 if (p->p_ucred->cr_uid == uap->who &&
235 p_cansee(td, p) == 0) {
236 error = donice(td, p, uap->prio);
241 sx_sunlock(&allproc_lock);
248 if (found == 0 && error == 0)
254 * Set "nice" for a (whole) process.
257 donice(struct thread *td, struct proc *p, int n)
261 PROC_LOCK_ASSERT(p, MA_OWNED);
262 if ((error = p_cansched(td, p)))
268 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
275 * Set realtime priority for LWP.
277 #ifndef _SYS_SYSPROTO_H_
278 struct rtprio_thread_args {
285 rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
292 /* Perform copyin before acquiring locks if needed. */
293 if (uap->function == RTP_SET)
294 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
298 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) {
303 /* Only look up thread in current process */
304 td1 = tdfind(uap->lwpid, curproc->p_pid);
310 switch (uap->function) {
312 if ((error = p_cansee(td, p)))
314 pri_to_rtp(td1, &rtp);
316 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
318 if ((error = p_cansched(td, p)) || (error = cierror))
321 /* Disallow setting rtprio in most cases if not superuser. */
323 * Realtime priority has to be restricted for reasons which should be
324 * obvious. However, for idle priority, there is a potential for
325 * system deadlock if an idleprio process gains a lock on a resource
326 * that other processes need (and the idleprio process can't run
327 * due to a CPU-bound normal process). Fix me! XXX
330 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
332 if (rtp.type != RTP_PRIO_NORMAL) {
334 error = priv_check(td, PRIV_SCHED_RTPRIO);
338 error = rtp_to_pri(&rtp, td1);
349 * Set realtime priority.
351 #ifndef _SYS_SYSPROTO_H_
360 struct thread *td; /* curthread */
361 register struct rtprio_args *uap;
368 /* Perform copyin before acquiring locks if needed. */
369 if (uap->function == RTP_SET)
370 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
383 switch (uap->function) {
385 if ((error = p_cansee(td, p)))
388 * Return OUR priority if no pid specified,
389 * or if one is, report the highest priority
390 * in the process. There isn't much more you can do as
391 * there is only room to return a single priority.
392 * Note: specifying our own pid is not the same
393 * as leaving it zero.
396 pri_to_rtp(td, &rtp);
400 rtp.type = RTP_PRIO_IDLE;
401 rtp.prio = RTP_PRIO_MAX;
402 FOREACH_THREAD_IN_PROC(p, tdp) {
403 pri_to_rtp(tdp, &rtp2);
404 if (rtp2.type < rtp.type ||
405 (rtp2.type == rtp.type &&
406 rtp2.prio < rtp.prio)) {
407 rtp.type = rtp2.type;
408 rtp.prio = rtp2.prio;
413 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
415 if ((error = p_cansched(td, p)) || (error = cierror))
418 /* Disallow setting rtprio in most cases if not superuser. */
420 * Realtime priority has to be restricted for reasons which should be
421 * obvious. However, for idle priority, there is a potential for
422 * system deadlock if an idleprio process gains a lock on a resource
423 * that other processes need (and the idleprio process can't run
424 * due to a CPU-bound normal process). Fix me! XXX
427 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
429 if (rtp.type != RTP_PRIO_NORMAL) {
431 error = priv_check(td, PRIV_SCHED_RTPRIO);
437 * If we are setting our own priority, set just our
438 * thread but if we are doing another process,
439 * do all the threads on that process. If we
440 * specify our own pid we do the latter.
443 error = rtp_to_pri(&rtp, td);
445 FOREACH_THREAD_IN_PROC(p, td) {
446 if ((error = rtp_to_pri(&rtp, td)) != 0)
460 rtp_to_pri(struct rtprio *rtp, struct thread *td)
466 switch (RTP_PRIO_BASE(rtp->type)) {
467 case RTP_PRIO_REALTIME:
468 if (rtp->prio > RTP_PRIO_MAX) {
472 newpri = PRI_MIN_REALTIME + rtp->prio;
474 case RTP_PRIO_NORMAL:
475 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) {
479 newpri = PRI_MIN_TIMESHARE + rtp->prio;
482 newpri = PRI_MIN_IDLE + rtp->prio;
488 sched_class(td, rtp->type); /* XXX fix */
489 oldpri = td->td_user_pri;
490 sched_user_prio(td, newpri);
492 sched_prio(curthread, td->td_user_pri); /* XXX dubious */
493 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
495 umtx_pi_adjust(td, oldpri);
502 pri_to_rtp(struct thread *td, struct rtprio *rtp)
506 switch (PRI_BASE(td->td_pri_class)) {
508 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
511 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
514 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
519 rtp->type = td->td_pri_class;
523 #if defined(COMPAT_43)
524 #ifndef _SYS_SYSPROTO_H_
525 struct osetrlimit_args {
533 register struct osetrlimit_args *uap;
539 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
541 lim.rlim_cur = olim.rlim_cur;
542 lim.rlim_max = olim.rlim_max;
543 error = kern_setrlimit(td, uap->which, &lim);
547 #ifndef _SYS_SYSPROTO_H_
548 struct ogetrlimit_args {
556 register struct ogetrlimit_args *uap;
563 if (uap->which >= RLIM_NLIMITS)
567 lim_rlimit(p, uap->which, &rl);
571 * XXX would be more correct to convert only RLIM_INFINITY to the
572 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
573 * values. Most 64->32 and 32->16 conversions, including not
574 * unimportant ones of uids are even more broken than what we
575 * do here (they blindly truncate). We don't do this correctly
576 * here since we have little experience with EOVERFLOW yet.
577 * Elsewhere, getuid() can't fail...
579 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
580 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
581 error = copyout(&olim, uap->rlp, sizeof(olim));
584 #endif /* COMPAT_43 */
586 #ifndef _SYS_SYSPROTO_H_
587 struct __setrlimit_args {
595 register struct __setrlimit_args *uap;
600 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
602 error = kern_setrlimit(td, uap->which, &alim);
614 PROC_LOCK_ASSERT(p, MA_OWNED);
616 * Check if the process exceeds its cpu resource allocation. If
617 * it reaches the max, arrange to kill the process in ast().
619 if (p->p_cpulimit == RLIM_INFINITY)
622 FOREACH_THREAD_IN_PROC(p, td) {
626 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
627 lim_rlimit(p, RLIMIT_CPU, &rlim);
628 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
629 killproc(p, "exceeded maximum CPU limit");
631 if (p->p_cpulimit < rlim.rlim_max)
636 if ((p->p_flag & P_WEXIT) == 0)
637 callout_reset(&p->p_limco, hz, lim_cb, p);
641 kern_setrlimit(td, which, limp)
646 struct plimit *newlim, *oldlim;
648 register struct rlimit *alimp;
649 struct rlimit oldssiz;
652 if (which >= RLIM_NLIMITS)
656 * Preserve historical bugs by treating negative limits as unsigned.
658 if (limp->rlim_cur < 0)
659 limp->rlim_cur = RLIM_INFINITY;
660 if (limp->rlim_max < 0)
661 limp->rlim_max = RLIM_INFINITY;
663 oldssiz.rlim_cur = 0;
665 newlim = lim_alloc();
668 alimp = &oldlim->pl_rlimit[which];
669 if (limp->rlim_cur > alimp->rlim_max ||
670 limp->rlim_max > alimp->rlim_max)
671 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
676 if (limp->rlim_cur > limp->rlim_max)
677 limp->rlim_cur = limp->rlim_max;
678 lim_copy(newlim, oldlim);
679 alimp = &newlim->pl_rlimit[which];
684 if (limp->rlim_cur != RLIM_INFINITY &&
685 p->p_cpulimit == RLIM_INFINITY)
686 callout_reset(&p->p_limco, hz, lim_cb, p);
687 p->p_cpulimit = limp->rlim_cur;
690 if (limp->rlim_cur > maxdsiz)
691 limp->rlim_cur = maxdsiz;
692 if (limp->rlim_max > maxdsiz)
693 limp->rlim_max = maxdsiz;
697 if (limp->rlim_cur > maxssiz)
698 limp->rlim_cur = maxssiz;
699 if (limp->rlim_max > maxssiz)
700 limp->rlim_max = maxssiz;
702 if (p->p_sysent->sv_fixlimit != NULL)
703 p->p_sysent->sv_fixlimit(&oldssiz,
708 if (limp->rlim_cur > maxfilesperproc)
709 limp->rlim_cur = maxfilesperproc;
710 if (limp->rlim_max > maxfilesperproc)
711 limp->rlim_max = maxfilesperproc;
715 if (limp->rlim_cur > maxprocperuid)
716 limp->rlim_cur = maxprocperuid;
717 if (limp->rlim_max > maxprocperuid)
718 limp->rlim_max = maxprocperuid;
719 if (limp->rlim_cur < 1)
721 if (limp->rlim_max < 1)
725 if (p->p_sysent->sv_fixlimit != NULL)
726 p->p_sysent->sv_fixlimit(limp, which);
732 if (which == RLIMIT_STACK) {
734 * Stack is allocated to the max at exec time with only
735 * "rlim_cur" bytes accessible. If stack limit is going
736 * up make more accessible, if going down make inaccessible.
738 if (limp->rlim_cur != oldssiz.rlim_cur) {
743 if (limp->rlim_cur > oldssiz.rlim_cur) {
744 prot = p->p_sysent->sv_stackprot;
745 size = limp->rlim_cur - oldssiz.rlim_cur;
746 addr = p->p_sysent->sv_usrstack -
750 size = oldssiz.rlim_cur - limp->rlim_cur;
751 addr = p->p_sysent->sv_usrstack -
754 addr = trunc_page(addr);
755 size = round_page(size);
756 (void)vm_map_protect(&p->p_vmspace->vm_map,
757 addr, addr + size, prot, FALSE);
764 #ifndef _SYS_SYSPROTO_H_
765 struct __getrlimit_args {
774 register struct __getrlimit_args *uap;
780 if (uap->which >= RLIM_NLIMITS)
784 lim_rlimit(p, uap->which, &rlim);
786 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
791 * Transform the running time and tick information for children of proc p
792 * into user and system time usage.
801 PROC_LOCK_ASSERT(p, MA_OWNED);
802 calcru1(p, &p->p_crux, up, sp);
806 * Transform the running time and tick information in proc p into user
807 * and system time usage. If appropriate, include the current time slice
811 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
816 PROC_LOCK_ASSERT(p, MA_OWNED);
817 PROC_SLOCK_ASSERT(p, MA_OWNED);
819 * If we are getting stats for the current process, then add in the
820 * stats that this thread has accumulated in its current time slice.
821 * We reset the thread and CPU state as if we had performed a context
825 if (td->td_proc == p) {
827 p->p_rux.rux_runtime += u - PCPU_GET(switchtime);
828 PCPU_SET(switchtime, u);
830 /* Make sure the per-thread stats are current. */
831 FOREACH_THREAD_IN_PROC(p, td) {
832 if (td->td_incruntime == 0)
836 calcru1(p, &p->p_rux, up, sp);
840 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
843 /* {user, system, interrupt, total} {ticks, usec}: */
844 uint64_t ut, uu, st, su, it, tt, tu;
846 ut = ruxp->rux_uticks;
847 st = ruxp->rux_sticks;
848 it = ruxp->rux_iticks;
851 /* Avoid divide by zero */
855 tu = cputick2usec(ruxp->rux_runtime);
856 if ((int64_t)tu < 0) {
857 /* XXX: this should be an assert /phk */
858 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
859 (intmax_t)tu, p->p_pid, p->p_comm);
863 if (tu >= ruxp->rux_tu) {
865 * The normal case, time increased.
866 * Enforce monotonicity of bucketed numbers.
869 if (uu < ruxp->rux_uu)
872 if (su < ruxp->rux_su)
874 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
876 * When we calibrate the cputicker, it is not uncommon to
877 * see the presumably fixed frequency increase slightly over
878 * time as a result of thermal stabilization and NTP
879 * discipline (of the reference clock). We therefore ignore
880 * a bit of backwards slop because we expect to catch up
881 * shortly. We use a 3 microsecond limit to catch low
882 * counts and a 1% limit for high counts.
887 } else { /* tu < ruxp->rux_tu */
889 * What happened here was likely that a laptop, which ran at
890 * a reduced clock frequency at boot, kicked into high gear.
891 * The wisdom of spamming this message in that case is
892 * dubious, but it might also be indicative of something
893 * serious, so lets keep it and hope laptops can be made
894 * more truthful about their CPU speed via ACPI.
896 printf("calcru: runtime went backwards from %ju usec "
897 "to %ju usec for pid %d (%s)\n",
898 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
899 p->p_pid, p->p_comm);
908 up->tv_sec = uu / 1000000;
909 up->tv_usec = uu % 1000000;
910 sp->tv_sec = su / 1000000;
911 sp->tv_usec = su % 1000000;
914 #ifndef _SYS_SYSPROTO_H_
915 struct getrusage_args {
917 struct rusage *rusage;
922 register struct thread *td;
923 register struct getrusage_args *uap;
928 error = kern_getrusage(td, uap->who, &ru);
930 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
935 kern_getrusage(struct thread *td, int who, struct rusage *rup)
945 rufetchcalc(p, rup, &rup->ru_utime,
949 case RUSAGE_CHILDREN:
950 *rup = p->p_stats->p_cru;
951 calccru(p, &rup->ru_utime, &rup->ru_stime);
960 calcru1(p, &td->td_rux, &rup->ru_utime, &rup->ru_stime);
972 rucollect(struct rusage *ru, struct rusage *ru2)
977 if (ru->ru_maxrss < ru2->ru_maxrss)
978 ru->ru_maxrss = ru2->ru_maxrss;
980 ip2 = &ru2->ru_first;
981 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
986 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
987 struct rusage_ext *rux2)
990 rux->rux_runtime += rux2->rux_runtime;
991 rux->rux_uticks += rux2->rux_uticks;
992 rux->rux_sticks += rux2->rux_sticks;
993 rux->rux_iticks += rux2->rux_iticks;
994 rux->rux_uu += rux2->rux_uu;
995 rux->rux_su += rux2->rux_su;
996 rux->rux_tu += rux2->rux_tu;
1001 * Aggregate tick counts into the proc's rusage_ext.
1004 ruxagg_locked(struct rusage_ext *rux, struct thread *td)
1007 THREAD_LOCK_ASSERT(td, MA_OWNED);
1008 PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED);
1009 rux->rux_runtime += td->td_incruntime;
1010 rux->rux_uticks += td->td_uticks;
1011 rux->rux_sticks += td->td_sticks;
1012 rux->rux_iticks += td->td_iticks;
1016 ruxagg(struct proc *p, struct thread *td)
1020 ruxagg_locked(&p->p_rux, td);
1021 ruxagg_locked(&td->td_rux, td);
1022 td->td_incruntime = 0;
1030 * Update the rusage_ext structure and fetch a valid aggregate rusage
1031 * for proc p if storage for one is supplied.
1034 rufetch(struct proc *p, struct rusage *ru)
1038 PROC_SLOCK_ASSERT(p, MA_OWNED);
1041 if (p->p_numthreads > 0) {
1042 FOREACH_THREAD_IN_PROC(p, td) {
1044 rucollect(ru, &td->td_ru);
1050 * Atomically perform a rufetch and a calcru together.
1051 * Consumers, can safely assume the calcru is executed only once
1052 * rufetch is completed.
1055 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1066 * Allocate a new resource limits structure and initialize its
1067 * reference count and mutex pointer.
1072 struct plimit *limp;
1074 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1075 refcount_init(&limp->pl_refcnt, 1);
1081 struct plimit *limp;
1084 refcount_acquire(&limp->pl_refcnt);
1089 lim_fork(struct proc *p1, struct proc *p2)
1091 p2->p_limit = lim_hold(p1->p_limit);
1092 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1093 if (p1->p_cpulimit != RLIM_INFINITY)
1094 callout_reset(&p2->p_limco, hz, lim_cb, p2);
1099 struct plimit *limp;
1102 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
1103 if (refcount_release(&limp->pl_refcnt))
1104 free((void *)limp, M_PLIMIT);
1108 * Make a copy of the plimit structure.
1109 * We share these structures copy-on-write after fork.
1113 struct plimit *dst, *src;
1116 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
1117 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1121 * Return the hard limit for a particular system resource. The
1122 * which parameter specifies the index into the rlimit array.
1125 lim_max(struct proc *p, int which)
1129 lim_rlimit(p, which, &rl);
1130 return (rl.rlim_max);
1134 * Return the current (soft) limit for a particular system resource.
1135 * The which parameter which specifies the index into the rlimit array
1138 lim_cur(struct proc *p, int which)
1142 lim_rlimit(p, which, &rl);
1143 return (rl.rlim_cur);
1147 * Return a copy of the entire rlimit structure for the system limit
1148 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1151 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
1154 PROC_LOCK_ASSERT(p, MA_OWNED);
1155 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1156 ("request for invalid resource limit"));
1157 *rlp = p->p_limit->pl_rlimit[which];
1158 if (p->p_sysent->sv_fixlimit != NULL)
1159 p->p_sysent->sv_fixlimit(rlp, which);
1166 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1167 rw_init(&uihashtbl_lock, "uidinfo hash");
1171 * Look up a uidinfo struct for the parameter uid.
1172 * uihashtbl_lock must be locked.
1174 static struct uidinfo *
1178 struct uihashhead *uipp;
1179 struct uidinfo *uip;
1181 rw_assert(&uihashtbl_lock, RA_LOCKED);
1183 LIST_FOREACH(uip, uipp, ui_hash)
1184 if (uip->ui_uid == uid)
1191 * Find or allocate a struct uidinfo for a particular uid.
1192 * Increase refcount on uidinfo struct returned.
1193 * uifree() should be called on a struct uidinfo when released.
1199 struct uidinfo *old_uip, *uip;
1201 rw_rlock(&uihashtbl_lock);
1202 uip = uilookup(uid);
1204 rw_runlock(&uihashtbl_lock);
1205 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1206 rw_wlock(&uihashtbl_lock);
1208 * There's a chance someone created our uidinfo while we
1209 * were in malloc and not holding the lock, so we have to
1210 * make sure we don't insert a duplicate uidinfo.
1212 if ((old_uip = uilookup(uid)) != NULL) {
1213 /* Someone else beat us to it. */
1214 free(uip, M_UIDINFO);
1217 refcount_init(&uip->ui_ref, 0);
1219 mtx_init(&uip->ui_vmsize_mtx, "ui_vmsize", NULL,
1221 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1225 rw_unlock(&uihashtbl_lock);
1230 * Place another refcount on a uidinfo struct.
1234 struct uidinfo *uip;
1237 refcount_acquire(&uip->ui_ref);
1241 * Since uidinfo structs have a long lifetime, we use an
1242 * opportunistic refcounting scheme to avoid locking the lookup hash
1245 * If the refcount hits 0, we need to free the structure,
1246 * which means we need to lock the hash.
1248 * After locking the struct and lowering the refcount, if we find
1249 * that we don't need to free, simply unlock and return.
1251 * If refcount lowering results in need to free, bump the count
1252 * back up, lose the lock and acquire the locks in the proper
1253 * order to try again.
1257 struct uidinfo *uip;
1261 /* Prepare for optimal case. */
1263 if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1))
1266 /* Prepare for suboptimal case. */
1267 rw_wlock(&uihashtbl_lock);
1268 if (refcount_release(&uip->ui_ref)) {
1269 LIST_REMOVE(uip, ui_hash);
1270 rw_wunlock(&uihashtbl_lock);
1271 if (uip->ui_sbsize != 0)
1272 printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1273 uip->ui_uid, uip->ui_sbsize);
1274 if (uip->ui_proccnt != 0)
1275 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1276 uip->ui_uid, uip->ui_proccnt);
1277 if (uip->ui_vmsize != 0)
1278 printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1279 uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1280 mtx_destroy(&uip->ui_vmsize_mtx);
1281 free(uip, M_UIDINFO);
1285 * Someone added a reference between atomic_cmpset_int() and
1286 * rw_wlock(&uihashtbl_lock).
1288 rw_wunlock(&uihashtbl_lock);
1292 * Change the count associated with number of processes
1293 * a given user is using. When 'max' is 0, don't enforce a limit
1296 chgproccnt(uip, diff, max)
1297 struct uidinfo *uip;
1302 /* Don't allow them to exceed max, but allow subtraction. */
1303 if (diff > 0 && max != 0) {
1304 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) {
1305 atomic_subtract_long(&uip->ui_proccnt, (long)diff);
1309 atomic_add_long(&uip->ui_proccnt, (long)diff);
1310 if (uip->ui_proccnt < 0)
1311 printf("negative proccnt for uid = %d\n", uip->ui_uid);
1317 * Change the total socket buffer size a user has used.
1320 chgsbsize(uip, hiwat, to, max)
1321 struct uidinfo *uip;
1330 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) {
1331 atomic_subtract_long(&uip->ui_sbsize, (long)diff);
1335 atomic_add_long(&uip->ui_sbsize, (long)diff);
1336 if (uip->ui_sbsize < 0)
1337 printf("negative sbsize for uid = %d\n", uip->ui_uid);
1344 * Change the count associated with number of pseudo-terminals
1345 * a given user is using. When 'max' is 0, don't enforce a limit
1348 chgptscnt(uip, diff, max)
1349 struct uidinfo *uip;
1354 /* Don't allow them to exceed max, but allow subtraction. */
1355 if (diff > 0 && max != 0) {
1356 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) {
1357 atomic_subtract_long(&uip->ui_ptscnt, (long)diff);
1361 atomic_add_long(&uip->ui_ptscnt, (long)diff);
1362 if (uip->ui_ptscnt < 0)
1363 printf("negative ptscnt for uid = %d\n", uip->ui_uid);