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);
81 * Resource controls and accounting.
83 #ifndef _SYS_SYSPROTO_H_
84 struct getpriority_args {
92 register struct getpriority_args *uap;
100 switch (uap->which) {
104 low = td->td_proc->p_nice;
109 if (p_cansee(td, p) == 0)
116 sx_slock(&proctree_lock);
118 pg = td->td_proc->p_pgrp;
121 pg = pgfind(uap->who);
123 sx_sunlock(&proctree_lock);
127 sx_sunlock(&proctree_lock);
128 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
130 if (p_cansee(td, p) == 0) {
141 uap->who = td->td_ucred->cr_uid;
142 sx_slock(&allproc_lock);
143 FOREACH_PROC_IN_SYSTEM(p) {
144 /* Do not bother to check PRS_NEW processes */
145 if (p->p_state == PRS_NEW)
148 if (p_cansee(td, p) == 0 &&
149 p->p_ucred->cr_uid == uap->who) {
155 sx_sunlock(&allproc_lock);
162 if (low == PRIO_MAX + 1 && error == 0)
164 td->td_retval[0] = low;
168 #ifndef _SYS_SYSPROTO_H_
169 struct setpriority_args {
178 struct setpriority_args *uap;
180 struct proc *curp, *p;
182 int found = 0, error = 0;
185 switch (uap->which) {
189 error = donice(td, curp, uap->prio);
195 error = p_cansee(td, p);
197 error = donice(td, p, uap->prio);
204 sx_slock(&proctree_lock);
209 pg = pgfind(uap->who);
211 sx_sunlock(&proctree_lock);
215 sx_sunlock(&proctree_lock);
216 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
218 if (p_cansee(td, p) == 0) {
219 error = donice(td, p, uap->prio);
229 uap->who = td->td_ucred->cr_uid;
230 sx_slock(&allproc_lock);
231 FOREACH_PROC_IN_SYSTEM(p) {
233 if (p->p_ucred->cr_uid == uap->who &&
234 p_cansee(td, p) == 0) {
235 error = donice(td, p, uap->prio);
240 sx_sunlock(&allproc_lock);
247 if (found == 0 && error == 0)
253 * Set "nice" for a (whole) process.
256 donice(struct thread *td, struct proc *p, int n)
260 PROC_LOCK_ASSERT(p, MA_OWNED);
261 if ((error = p_cansched(td, p)))
267 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
274 * Set realtime priority for LWP.
276 #ifndef _SYS_SYSPROTO_H_
277 struct rtprio_thread_args {
284 rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
291 /* Perform copyin before acquiring locks if needed. */
292 if (uap->function == RTP_SET)
293 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
298 * Though lwpid is unique, only current process is supported
299 * since there is no efficient way to look up a LWP yet.
304 switch (uap->function) {
306 if ((error = p_cansee(td, p)))
308 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
311 td1 = thread_find(p, uap->lwpid);
313 pri_to_rtp(td1, &rtp);
317 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
319 if ((error = p_cansched(td, p)) || (error = cierror))
322 /* Disallow setting rtprio in most cases if not superuser. */
324 * Realtime priority has to be restricted for reasons which should be
325 * obvious. However, for idle priority, there is a potential for
326 * system deadlock if an idleprio process gains a lock on a resource
327 * that other processes need (and the idleprio process can't run
328 * due to a CPU-bound normal process). Fix me! XXX
331 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
333 if (rtp.type != RTP_PRIO_NORMAL) {
335 error = priv_check(td, PRIV_SCHED_RTPRIO);
340 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
343 td1 = thread_find(p, uap->lwpid);
345 error = rtp_to_pri(&rtp, td1);
358 * Set realtime priority.
360 #ifndef _SYS_SYSPROTO_H_
369 struct thread *td; /* curthread */
370 register struct rtprio_args *uap;
377 /* Perform copyin before acquiring locks if needed. */
378 if (uap->function == RTP_SET)
379 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
392 switch (uap->function) {
394 if ((error = p_cansee(td, p)))
397 * Return OUR priority if no pid specified,
398 * or if one is, report the highest priority
399 * in the process. There isn't much more you can do as
400 * there is only room to return a single priority.
401 * Note: specifying our own pid is not the same
402 * as leaving it zero.
405 pri_to_rtp(td, &rtp);
409 rtp.type = RTP_PRIO_IDLE;
410 rtp.prio = RTP_PRIO_MAX;
411 FOREACH_THREAD_IN_PROC(p, tdp) {
412 pri_to_rtp(tdp, &rtp2);
413 if (rtp2.type < rtp.type ||
414 (rtp2.type == rtp.type &&
415 rtp2.prio < rtp.prio)) {
416 rtp.type = rtp2.type;
417 rtp.prio = rtp2.prio;
422 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
424 if ((error = p_cansched(td, p)) || (error = cierror))
427 /* Disallow setting rtprio in most cases if not superuser. */
429 * Realtime priority has to be restricted for reasons which should be
430 * obvious. However, for idle priority, there is a potential for
431 * system deadlock if an idleprio process gains a lock on a resource
432 * that other processes need (and the idleprio process can't run
433 * due to a CPU-bound normal process). Fix me! XXX
436 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
438 if (rtp.type != RTP_PRIO_NORMAL) {
440 error = priv_check(td, PRIV_SCHED_RTPRIO);
446 * If we are setting our own priority, set just our
447 * thread but if we are doing another process,
448 * do all the threads on that process. If we
449 * specify our own pid we do the latter.
452 error = rtp_to_pri(&rtp, td);
454 FOREACH_THREAD_IN_PROC(p, td) {
455 if ((error = rtp_to_pri(&rtp, td)) != 0)
469 rtp_to_pri(struct rtprio *rtp, struct thread *td)
474 if (rtp->prio > RTP_PRIO_MAX)
477 switch (RTP_PRIO_BASE(rtp->type)) {
478 case RTP_PRIO_REALTIME:
479 newpri = PRI_MIN_REALTIME + rtp->prio;
481 case RTP_PRIO_NORMAL:
482 newpri = PRI_MIN_TIMESHARE + rtp->prio;
485 newpri = PRI_MIN_IDLE + rtp->prio;
491 sched_class(td, rtp->type); /* XXX fix */
492 oldpri = td->td_user_pri;
493 sched_user_prio(td, newpri);
495 sched_prio(curthread, td->td_user_pri); /* XXX dubious */
496 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
498 umtx_pi_adjust(td, oldpri);
505 pri_to_rtp(struct thread *td, struct rtprio *rtp)
509 switch (PRI_BASE(td->td_pri_class)) {
511 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
514 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
517 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
522 rtp->type = td->td_pri_class;
526 #if defined(COMPAT_43)
527 #ifndef _SYS_SYSPROTO_H_
528 struct osetrlimit_args {
536 register struct osetrlimit_args *uap;
542 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
544 lim.rlim_cur = olim.rlim_cur;
545 lim.rlim_max = olim.rlim_max;
546 error = kern_setrlimit(td, uap->which, &lim);
550 #ifndef _SYS_SYSPROTO_H_
551 struct ogetrlimit_args {
559 register struct ogetrlimit_args *uap;
566 if (uap->which >= RLIM_NLIMITS)
570 lim_rlimit(p, uap->which, &rl);
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...
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));
587 #endif /* COMPAT_43 */
589 #ifndef _SYS_SYSPROTO_H_
590 struct __setrlimit_args {
598 register struct __setrlimit_args *uap;
603 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
605 error = kern_setrlimit(td, uap->which, &alim);
617 PROC_LOCK_ASSERT(p, MA_OWNED);
619 * Check if the process exceeds its cpu resource allocation. If
620 * it reaches the max, arrange to kill the process in ast().
622 if (p->p_cpulimit == RLIM_INFINITY)
625 FOREACH_THREAD_IN_PROC(p, td) {
627 ruxagg(&p->p_rux, td);
631 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
632 lim_rlimit(p, RLIMIT_CPU, &rlim);
633 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
634 killproc(p, "exceeded maximum CPU limit");
636 if (p->p_cpulimit < rlim.rlim_max)
641 if ((p->p_flag & P_WEXIT) == 0)
642 callout_reset(&p->p_limco, hz, lim_cb, p);
646 kern_setrlimit(td, which, limp)
651 struct plimit *newlim, *oldlim;
653 register struct rlimit *alimp;
654 struct rlimit oldssiz;
657 if (which >= RLIM_NLIMITS)
661 * Preserve historical bugs by treating negative limits as unsigned.
663 if (limp->rlim_cur < 0)
664 limp->rlim_cur = RLIM_INFINITY;
665 if (limp->rlim_max < 0)
666 limp->rlim_max = RLIM_INFINITY;
668 oldssiz.rlim_cur = 0;
670 newlim = lim_alloc();
673 alimp = &oldlim->pl_rlimit[which];
674 if (limp->rlim_cur > alimp->rlim_max ||
675 limp->rlim_max > alimp->rlim_max)
676 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
681 if (limp->rlim_cur > limp->rlim_max)
682 limp->rlim_cur = limp->rlim_max;
683 lim_copy(newlim, oldlim);
684 alimp = &newlim->pl_rlimit[which];
689 if (limp->rlim_cur != RLIM_INFINITY &&
690 p->p_cpulimit == RLIM_INFINITY)
691 callout_reset(&p->p_limco, hz, lim_cb, p);
692 p->p_cpulimit = limp->rlim_cur;
695 if (limp->rlim_cur > maxdsiz)
696 limp->rlim_cur = maxdsiz;
697 if (limp->rlim_max > maxdsiz)
698 limp->rlim_max = maxdsiz;
702 if (limp->rlim_cur > maxssiz)
703 limp->rlim_cur = maxssiz;
704 if (limp->rlim_max > maxssiz)
705 limp->rlim_max = maxssiz;
707 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
708 td->td_proc->p_sysent->sv_fixlimit(&oldssiz,
713 if (limp->rlim_cur > maxfilesperproc)
714 limp->rlim_cur = maxfilesperproc;
715 if (limp->rlim_max > maxfilesperproc)
716 limp->rlim_max = maxfilesperproc;
720 if (limp->rlim_cur > maxprocperuid)
721 limp->rlim_cur = maxprocperuid;
722 if (limp->rlim_max > maxprocperuid)
723 limp->rlim_max = maxprocperuid;
724 if (limp->rlim_cur < 1)
726 if (limp->rlim_max < 1)
730 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
731 td->td_proc->p_sysent->sv_fixlimit(limp, which);
737 if (which == RLIMIT_STACK) {
739 * Stack is allocated to the max at exec time with only
740 * "rlim_cur" bytes accessible. If stack limit is going
741 * up make more accessible, if going down make inaccessible.
743 if (limp->rlim_cur != oldssiz.rlim_cur) {
748 if (limp->rlim_cur > oldssiz.rlim_cur) {
749 prot = p->p_sysent->sv_stackprot;
750 size = limp->rlim_cur - oldssiz.rlim_cur;
751 addr = p->p_sysent->sv_usrstack -
755 size = oldssiz.rlim_cur - limp->rlim_cur;
756 addr = p->p_sysent->sv_usrstack -
759 addr = trunc_page(addr);
760 size = round_page(size);
761 (void)vm_map_protect(&p->p_vmspace->vm_map,
762 addr, addr + size, prot, FALSE);
769 #ifndef _SYS_SYSPROTO_H_
770 struct __getrlimit_args {
779 register struct __getrlimit_args *uap;
785 if (uap->which >= RLIM_NLIMITS)
789 lim_rlimit(p, uap->which, &rlim);
791 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
796 * Transform the running time and tick information for children of proc p
797 * into user and system time usage.
806 PROC_LOCK_ASSERT(p, MA_OWNED);
807 calcru1(p, &p->p_crux, up, sp);
811 * Transform the running time and tick information in proc p into user
812 * and system time usage. If appropriate, include the current time slice
816 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
821 PROC_LOCK_ASSERT(p, MA_OWNED);
822 PROC_SLOCK_ASSERT(p, MA_OWNED);
824 * If we are getting stats for the current process, then add in the
825 * stats that this thread has accumulated in its current time slice.
826 * We reset the thread and CPU state as if we had performed a context
830 if (td->td_proc == p) {
832 p->p_rux.rux_runtime += u - PCPU_GET(switchtime);
833 PCPU_SET(switchtime, u);
835 /* Make sure the per-thread stats are current. */
836 FOREACH_THREAD_IN_PROC(p, td) {
837 if (td->td_incruntime == 0)
840 ruxagg(&p->p_rux, td);
843 calcru1(p, &p->p_rux, up, sp);
847 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
850 /* {user, system, interrupt, total} {ticks, usec}: */
851 u_int64_t ut, uu, st, su, it, tt, tu;
853 ut = ruxp->rux_uticks;
854 st = ruxp->rux_sticks;
855 it = ruxp->rux_iticks;
858 /* Avoid divide by zero */
862 tu = cputick2usec(ruxp->rux_runtime);
863 if ((int64_t)tu < 0) {
864 /* XXX: this should be an assert /phk */
865 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
866 (intmax_t)tu, p->p_pid, p->p_comm);
870 if (tu >= ruxp->rux_tu) {
872 * The normal case, time increased.
873 * Enforce monotonicity of bucketed numbers.
876 if (uu < ruxp->rux_uu)
879 if (su < ruxp->rux_su)
881 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
883 * When we calibrate the cputicker, it is not uncommon to
884 * see the presumably fixed frequency increase slightly over
885 * time as a result of thermal stabilization and NTP
886 * discipline (of the reference clock). We therefore ignore
887 * a bit of backwards slop because we expect to catch up
888 * shortly. We use a 3 microsecond limit to catch low
889 * counts and a 1% limit for high counts.
894 } else { /* tu < ruxp->rux_tu */
896 * What happened here was likely that a laptop, which ran at
897 * a reduced clock frequency at boot, kicked into high gear.
898 * The wisdom of spamming this message in that case is
899 * dubious, but it might also be indicative of something
900 * serious, so lets keep it and hope laptops can be made
901 * more truthful about their CPU speed via ACPI.
903 printf("calcru: runtime went backwards from %ju usec "
904 "to %ju usec for pid %d (%s)\n",
905 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
906 p->p_pid, p->p_comm);
915 up->tv_sec = uu / 1000000;
916 up->tv_usec = uu % 1000000;
917 sp->tv_sec = su / 1000000;
918 sp->tv_usec = su % 1000000;
921 #ifndef _SYS_SYSPROTO_H_
922 struct getrusage_args {
924 struct rusage *rusage;
929 register struct thread *td;
930 register struct getrusage_args *uap;
935 error = kern_getrusage(td, uap->who, &ru);
937 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
942 kern_getrusage(td, who, rup)
955 rufetchcalc(p, rup, &rup->ru_utime,
959 case RUSAGE_CHILDREN:
960 *rup = p->p_stats->p_cru;
961 calccru(p, &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(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;
1013 td->td_incruntime = 0;
1020 * Update the rusage_ext structure and fetch a valid aggregate rusage
1021 * for proc p if storage for one is supplied.
1024 rufetch(struct proc *p, struct rusage *ru)
1028 PROC_SLOCK_ASSERT(p, MA_OWNED);
1031 if (p->p_numthreads > 0) {
1032 FOREACH_THREAD_IN_PROC(p, td) {
1034 ruxagg(&p->p_rux, td);
1036 rucollect(ru, &td->td_ru);
1042 * Atomically perform a rufetch and a calcru together.
1043 * Consumers, can safely assume the calcru is executed only once
1044 * rufetch is completed.
1047 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1058 * Allocate a new resource limits structure and initialize its
1059 * reference count and mutex pointer.
1064 struct plimit *limp;
1066 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1067 refcount_init(&limp->pl_refcnt, 1);
1073 struct plimit *limp;
1076 refcount_acquire(&limp->pl_refcnt);
1081 lim_fork(struct proc *p1, struct proc *p2)
1083 p2->p_limit = lim_hold(p1->p_limit);
1084 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1085 if (p1->p_cpulimit != RLIM_INFINITY)
1086 callout_reset(&p2->p_limco, hz, lim_cb, p2);
1091 struct plimit *limp;
1094 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
1095 if (refcount_release(&limp->pl_refcnt))
1096 free((void *)limp, M_PLIMIT);
1100 * Make a copy of the plimit structure.
1101 * We share these structures copy-on-write after fork.
1105 struct plimit *dst, *src;
1108 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
1109 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1113 * Return the hard limit for a particular system resource. The
1114 * which parameter specifies the index into the rlimit array.
1117 lim_max(struct proc *p, int which)
1121 lim_rlimit(p, which, &rl);
1122 return (rl.rlim_max);
1126 * Return the current (soft) limit for a particular system resource.
1127 * The which parameter which specifies the index into the rlimit array
1130 lim_cur(struct proc *p, int which)
1134 lim_rlimit(p, which, &rl);
1135 return (rl.rlim_cur);
1139 * Return a copy of the entire rlimit structure for the system limit
1140 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1143 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
1146 PROC_LOCK_ASSERT(p, MA_OWNED);
1147 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1148 ("request for invalid resource limit"));
1149 *rlp = p->p_limit->pl_rlimit[which];
1150 if (p->p_sysent->sv_fixlimit != NULL)
1151 p->p_sysent->sv_fixlimit(rlp, which);
1155 * Find the uidinfo structure for a uid. This structure is used to
1156 * track the total resource consumption (process count, socket buffer
1157 * size, etc.) for the uid and impose limits.
1163 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1164 rw_init(&uihashtbl_lock, "uidinfo hash");
1168 * Look up a uidinfo struct for the parameter uid.
1169 * uihashtbl_lock must be locked.
1171 static struct uidinfo *
1175 struct uihashhead *uipp;
1176 struct uidinfo *uip;
1178 rw_assert(&uihashtbl_lock, RA_LOCKED);
1180 LIST_FOREACH(uip, uipp, ui_hash)
1181 if (uip->ui_uid == uid)
1188 * Find or allocate a struct uidinfo for a particular uid.
1189 * Increase refcount on uidinfo struct returned.
1190 * uifree() should be called on a struct uidinfo when released.
1196 struct uidinfo *old_uip, *uip;
1198 rw_rlock(&uihashtbl_lock);
1199 uip = uilookup(uid);
1201 rw_runlock(&uihashtbl_lock);
1202 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1203 rw_wlock(&uihashtbl_lock);
1205 * There's a chance someone created our uidinfo while we
1206 * were in malloc and not holding the lock, so we have to
1207 * make sure we don't insert a duplicate uidinfo.
1209 if ((old_uip = uilookup(uid)) != NULL) {
1210 /* Someone else beat us to it. */
1211 free(uip, M_UIDINFO);
1214 refcount_init(&uip->ui_ref, 0);
1216 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1220 rw_unlock(&uihashtbl_lock);
1225 * Place another refcount on a uidinfo struct.
1229 struct uidinfo *uip;
1232 refcount_acquire(&uip->ui_ref);
1236 * Since uidinfo structs have a long lifetime, we use an
1237 * opportunistic refcounting scheme to avoid locking the lookup hash
1240 * If the refcount hits 0, we need to free the structure,
1241 * which means we need to lock the hash.
1243 * After locking the struct and lowering the refcount, if we find
1244 * that we don't need to free, simply unlock and return.
1246 * If refcount lowering results in need to free, bump the count
1247 * back up, lose the lock and acquire the locks in the proper
1248 * order to try again.
1252 struct uidinfo *uip;
1256 /* Prepare for optimal case. */
1258 if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1))
1261 /* Prepare for suboptimal case. */
1262 rw_wlock(&uihashtbl_lock);
1263 if (refcount_release(&uip->ui_ref)) {
1264 LIST_REMOVE(uip, ui_hash);
1265 rw_wunlock(&uihashtbl_lock);
1266 if (uip->ui_sbsize != 0)
1267 printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1268 uip->ui_uid, uip->ui_sbsize);
1269 if (uip->ui_proccnt != 0)
1270 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1271 uip->ui_uid, uip->ui_proccnt);
1272 free(uip, M_UIDINFO);
1276 * Someone added a reference between atomic_cmpset_int() and
1277 * rw_wlock(&uihashtbl_lock).
1279 rw_wunlock(&uihashtbl_lock);
1283 * Change the count associated with number of processes
1284 * a given user is using. When 'max' is 0, don't enforce a limit
1287 chgproccnt(uip, diff, max)
1288 struct uidinfo *uip;
1293 /* Don't allow them to exceed max, but allow subtraction. */
1294 if (diff > 0 && max != 0) {
1295 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) {
1296 atomic_subtract_long(&uip->ui_proccnt, (long)diff);
1300 atomic_add_long(&uip->ui_proccnt, (long)diff);
1301 if (uip->ui_proccnt < 0)
1302 printf("negative proccnt for uid = %d\n", uip->ui_uid);
1308 * Change the total socket buffer size a user has used.
1311 chgsbsize(uip, hiwat, to, max)
1312 struct uidinfo *uip;
1321 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) {
1322 atomic_subtract_long(&uip->ui_sbsize, (long)diff);
1326 atomic_add_long(&uip->ui_sbsize, (long)diff);
1327 if (uip->ui_sbsize < 0)
1328 printf("negative sbsize for uid = %d\n", uip->ui_uid);
1335 * Change the count associated with number of pseudo-terminals
1336 * a given user is using. When 'max' is 0, don't enforce a limit
1339 chgptscnt(uip, diff, max)
1340 struct uidinfo *uip;
1345 /* Don't allow them to exceed max, but allow subtraction. */
1346 if (diff > 0 && max != 0) {
1347 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) {
1348 atomic_subtract_long(&uip->ui_ptscnt, (long)diff);
1352 atomic_add_long(&uip->ui_ptscnt, (long)diff);
1353 if (uip->ui_ptscnt < 0)
1354 printf("negative ptscnt for uid = %d\n", uip->ui_uid);