2 * SPDX-License-Identifier: BSD-3-Clause
4 * Copyright (c) 1982, 1986, 1991, 1993
5 * The Regents of the University of California. All rights reserved.
6 * (c) UNIX System Laboratories, Inc.
7 * All or some portions of this file are derived from material licensed
8 * to the University of California by American Telephone and Telegraph
9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with
10 * the permission of UNIX System Laboratories, Inc.
12 * Redistribution and use in source and binary forms, with or without
13 * modification, are permitted provided that the following conditions
15 * 1. Redistributions of source code must retain the above copyright
16 * notice, this list of conditions and the following disclaimer.
17 * 2. Redistributions in binary form must reproduce the above copyright
18 * notice, this list of conditions and the following disclaimer in the
19 * documentation and/or other materials provided with the distribution.
20 * 3. Neither the name of the University nor the names of its contributors
21 * may be used to endorse or promote products derived from this software
22 * without specific prior written permission.
24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
36 * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94
39 #include <sys/cdefs.h>
40 __FBSDID("$FreeBSD$");
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/racct.h>
54 #include <sys/resourcevar.h>
55 #include <sys/rwlock.h>
56 #include <sys/sched.h>
58 #include <sys/syscallsubr.h>
59 #include <sys/sysctl.h>
60 #include <sys/sysent.h>
65 #include <vm/vm_param.h>
67 #include <vm/vm_map.h>
69 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
70 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
71 #define UIHASH(uid) (&uihashtbl[(uid) & uihash])
72 static struct rwlock uihashtbl_lock;
73 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
74 static u_long uihash; /* size of hash table - 1 */
76 static void calcru1(struct proc *p, struct rusage_ext *ruxp,
77 struct timeval *up, struct timeval *sp);
78 static int donice(struct thread *td, struct proc *chgp, int n);
79 static struct uidinfo *uilookup(uid_t uid);
80 static void ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td);
83 * Resource controls and accounting.
85 #ifndef _SYS_SYSPROTO_H_
86 struct getpriority_args {
92 sys_getpriority(struct thread *td, struct getpriority_args *uap)
95 return (kern_getpriority(td, uap->which, uap->who));
99 kern_getpriority(struct thread *td, int which, int who)
111 low = td->td_proc->p_nice;
116 if (p_cansee(td, p) == 0)
123 sx_slock(&proctree_lock);
125 pg = td->td_proc->p_pgrp;
130 sx_sunlock(&proctree_lock);
134 sx_sunlock(&proctree_lock);
135 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
137 if (p->p_state == PRS_NORMAL &&
138 p_cansee(td, p) == 0) {
149 who = td->td_ucred->cr_uid;
150 sx_slock(&allproc_lock);
151 FOREACH_PROC_IN_SYSTEM(p) {
153 if (p->p_state == PRS_NORMAL &&
154 p_cansee(td, p) == 0 &&
155 p->p_ucred->cr_uid == who) {
161 sx_sunlock(&allproc_lock);
168 if (low == PRIO_MAX + 1 && error == 0)
170 td->td_retval[0] = low;
174 #ifndef _SYS_SYSPROTO_H_
175 struct setpriority_args {
182 sys_setpriority(struct thread *td, struct setpriority_args *uap)
185 return (kern_setpriority(td, uap->which, uap->who, uap->prio));
189 kern_setpriority(struct thread *td, int which, int who, int prio)
191 struct proc *curp, *p;
193 int found = 0, error = 0;
200 error = donice(td, curp, prio);
206 error = p_cansee(td, p);
208 error = donice(td, p, prio);
215 sx_slock(&proctree_lock);
222 sx_sunlock(&proctree_lock);
226 sx_sunlock(&proctree_lock);
227 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
229 if (p->p_state == PRS_NORMAL &&
230 p_cansee(td, p) == 0) {
231 error = donice(td, p, prio);
241 who = td->td_ucred->cr_uid;
242 sx_slock(&allproc_lock);
243 FOREACH_PROC_IN_SYSTEM(p) {
245 if (p->p_state == PRS_NORMAL &&
246 p->p_ucred->cr_uid == who &&
247 p_cansee(td, p) == 0) {
248 error = donice(td, p, prio);
253 sx_sunlock(&allproc_lock);
260 if (found == 0 && error == 0)
266 * Set "nice" for a (whole) process.
269 donice(struct thread *td, struct proc *p, int n)
273 PROC_LOCK_ASSERT(p, MA_OWNED);
274 if ((error = p_cansched(td, p)))
280 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
286 static int unprivileged_idprio;
287 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW,
288 &unprivileged_idprio, 0, "Allow non-root users to set an idle priority");
291 * Set realtime priority for LWP.
293 #ifndef _SYS_SYSPROTO_H_
294 struct rtprio_thread_args {
301 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
308 /* Perform copyin before acquiring locks if needed. */
309 if (uap->function == RTP_SET)
310 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
314 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) {
319 /* Only look up thread in current process */
320 td1 = tdfind(uap->lwpid, curproc->p_pid);
326 switch (uap->function) {
328 if ((error = p_cansee(td, p)))
330 pri_to_rtp(td1, &rtp);
332 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
334 if ((error = p_cansched(td, p)) || (error = cierror))
337 /* Disallow setting rtprio in most cases if not superuser. */
340 * Realtime priority has to be restricted for reasons which
341 * should be obvious. However, for idleprio processes, there is
342 * a potential for system deadlock if an idleprio process gains
343 * a lock on a resource that other processes need (and the
344 * idleprio process can't run due to a CPU-bound normal
345 * process). Fix me! XXX
347 * This problem is not only related to idleprio process.
348 * A user level program can obtain a file lock and hold it
349 * indefinitely. Additionally, without idleprio processes it is
350 * still conceivable that a program with low priority will never
351 * get to run. In short, allowing this feature might make it
352 * easier to lock a resource indefinitely, but it is not the
353 * only thing that makes it possible.
355 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME ||
356 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
357 unprivileged_idprio == 0)) {
358 error = priv_check(td, PRIV_SCHED_RTPRIO);
362 error = rtp_to_pri(&rtp, td1);
373 * Set realtime priority.
375 #ifndef _SYS_SYSPROTO_H_
383 sys_rtprio(struct thread *td, struct rtprio_args *uap)
390 /* Perform copyin before acquiring locks if needed. */
391 if (uap->function == RTP_SET)
392 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
405 switch (uap->function) {
407 if ((error = p_cansee(td, p)))
410 * Return OUR priority if no pid specified,
411 * or if one is, report the highest priority
412 * in the process. There isn't much more you can do as
413 * there is only room to return a single priority.
414 * Note: specifying our own pid is not the same
415 * as leaving it zero.
418 pri_to_rtp(td, &rtp);
422 rtp.type = RTP_PRIO_IDLE;
423 rtp.prio = RTP_PRIO_MAX;
424 FOREACH_THREAD_IN_PROC(p, tdp) {
425 pri_to_rtp(tdp, &rtp2);
426 if (rtp2.type < rtp.type ||
427 (rtp2.type == rtp.type &&
428 rtp2.prio < rtp.prio)) {
429 rtp.type = rtp2.type;
430 rtp.prio = rtp2.prio;
435 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
437 if ((error = p_cansched(td, p)) || (error = cierror))
441 * Disallow setting rtprio in most cases if not superuser.
442 * See the comment in sys_rtprio_thread about idprio
443 * threads holding a lock.
445 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME ||
446 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
447 !unprivileged_idprio)) {
448 error = priv_check(td, PRIV_SCHED_RTPRIO);
454 * If we are setting our own priority, set just our
455 * thread but if we are doing another process,
456 * do all the threads on that process. If we
457 * specify our own pid we do the latter.
460 error = rtp_to_pri(&rtp, td);
462 FOREACH_THREAD_IN_PROC(p, td) {
463 if ((error = rtp_to_pri(&rtp, td)) != 0)
477 rtp_to_pri(struct rtprio *rtp, struct thread *td)
479 u_char newpri, oldclass, oldpri;
481 switch (RTP_PRIO_BASE(rtp->type)) {
482 case RTP_PRIO_REALTIME:
483 if (rtp->prio > RTP_PRIO_MAX)
485 newpri = PRI_MIN_REALTIME + rtp->prio;
487 case RTP_PRIO_NORMAL:
488 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE))
490 newpri = PRI_MIN_TIMESHARE + rtp->prio;
493 if (rtp->prio > RTP_PRIO_MAX)
495 newpri = PRI_MIN_IDLE + rtp->prio;
502 oldclass = td->td_pri_class;
503 sched_class(td, rtp->type); /* XXX fix */
504 oldpri = td->td_user_pri;
505 sched_user_prio(td, newpri);
506 if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL ||
507 td->td_pri_class != RTP_PRIO_NORMAL))
508 sched_prio(td, td->td_user_pri);
509 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
512 umtx_pi_adjust(td, oldpri);
520 pri_to_rtp(struct thread *td, struct rtprio *rtp)
524 switch (PRI_BASE(td->td_pri_class)) {
526 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
529 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
532 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
537 rtp->type = td->td_pri_class;
541 #if defined(COMPAT_43)
542 #ifndef _SYS_SYSPROTO_H_
543 struct osetrlimit_args {
549 osetrlimit(struct thread *td, struct osetrlimit_args *uap)
555 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
557 lim.rlim_cur = olim.rlim_cur;
558 lim.rlim_max = olim.rlim_max;
559 error = kern_setrlimit(td, uap->which, &lim);
563 #ifndef _SYS_SYSPROTO_H_
564 struct ogetrlimit_args {
570 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap)
576 if (uap->which >= RLIM_NLIMITS)
578 lim_rlimit(td, uap->which, &rl);
581 * XXX would be more correct to convert only RLIM_INFINITY to the
582 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
583 * values. Most 64->32 and 32->16 conversions, including not
584 * unimportant ones of uids are even more broken than what we
585 * do here (they blindly truncate). We don't do this correctly
586 * here since we have little experience with EOVERFLOW yet.
587 * Elsewhere, getuid() can't fail...
589 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
590 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
591 error = copyout(&olim, uap->rlp, sizeof(olim));
594 #endif /* COMPAT_43 */
596 #ifndef _SYS_SYSPROTO_H_
597 struct __setrlimit_args {
603 sys_setrlimit(struct thread *td, struct __setrlimit_args *uap)
608 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
610 error = kern_setrlimit(td, uap->which, &alim);
622 PROC_LOCK_ASSERT(p, MA_OWNED);
624 * Check if the process exceeds its cpu resource allocation. If
625 * it reaches the max, arrange to kill the process in ast().
627 if (p->p_cpulimit == RLIM_INFINITY)
630 FOREACH_THREAD_IN_PROC(p, td) {
634 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
635 lim_rlimit_proc(p, RLIMIT_CPU, &rlim);
636 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
637 killproc(p, "exceeded maximum CPU limit");
639 if (p->p_cpulimit < rlim.rlim_max)
641 kern_psignal(p, SIGXCPU);
644 if ((p->p_flag & P_WEXIT) == 0)
645 callout_reset_sbt(&p->p_limco, SBT_1S, 0,
646 lim_cb, p, C_PREL(1));
650 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp)
653 return (kern_proc_setrlimit(td, td->td_proc, which, limp));
657 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which,
660 struct plimit *newlim, *oldlim;
661 struct rlimit *alimp;
662 struct rlimit oldssiz;
665 if (which >= RLIM_NLIMITS)
669 * Preserve historical bugs by treating negative limits as unsigned.
671 if (limp->rlim_cur < 0)
672 limp->rlim_cur = RLIM_INFINITY;
673 if (limp->rlim_max < 0)
674 limp->rlim_max = RLIM_INFINITY;
676 oldssiz.rlim_cur = 0;
677 newlim = lim_alloc();
680 alimp = &oldlim->pl_rlimit[which];
681 if (limp->rlim_cur > alimp->rlim_max ||
682 limp->rlim_max > alimp->rlim_max)
683 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
688 if (limp->rlim_cur > limp->rlim_max)
689 limp->rlim_cur = limp->rlim_max;
690 lim_copy(newlim, oldlim);
691 alimp = &newlim->pl_rlimit[which];
696 if (limp->rlim_cur != RLIM_INFINITY &&
697 p->p_cpulimit == RLIM_INFINITY)
698 callout_reset_sbt(&p->p_limco, SBT_1S, 0,
699 lim_cb, p, C_PREL(1));
700 p->p_cpulimit = limp->rlim_cur;
703 if (limp->rlim_cur > maxdsiz)
704 limp->rlim_cur = maxdsiz;
705 if (limp->rlim_max > maxdsiz)
706 limp->rlim_max = maxdsiz;
710 if (limp->rlim_cur > maxssiz)
711 limp->rlim_cur = maxssiz;
712 if (limp->rlim_max > maxssiz)
713 limp->rlim_max = maxssiz;
715 if (p->p_sysent->sv_fixlimit != NULL)
716 p->p_sysent->sv_fixlimit(&oldssiz,
721 if (limp->rlim_cur > maxfilesperproc)
722 limp->rlim_cur = maxfilesperproc;
723 if (limp->rlim_max > maxfilesperproc)
724 limp->rlim_max = maxfilesperproc;
728 if (limp->rlim_cur > maxprocperuid)
729 limp->rlim_cur = maxprocperuid;
730 if (limp->rlim_max > maxprocperuid)
731 limp->rlim_max = maxprocperuid;
732 if (limp->rlim_cur < 1)
734 if (limp->rlim_max < 1)
738 if (p->p_sysent->sv_fixlimit != NULL)
739 p->p_sysent->sv_fixlimit(limp, which);
746 if (which == RLIMIT_STACK &&
748 * Skip calls from exec_new_vmspace(), done when stack is
751 (td != curthread || (p->p_flag & P_INEXEC) == 0)) {
753 * Stack is allocated to the max at exec time with only
754 * "rlim_cur" bytes accessible. If stack limit is going
755 * up make more accessible, if going down make inaccessible.
757 if (limp->rlim_cur != oldssiz.rlim_cur) {
762 if (limp->rlim_cur > oldssiz.rlim_cur) {
763 prot = p->p_sysent->sv_stackprot;
764 size = limp->rlim_cur - oldssiz.rlim_cur;
765 addr = p->p_sysent->sv_usrstack -
769 size = oldssiz.rlim_cur - limp->rlim_cur;
770 addr = p->p_sysent->sv_usrstack -
773 addr = trunc_page(addr);
774 size = round_page(size);
775 (void)vm_map_protect(&p->p_vmspace->vm_map,
776 addr, addr + size, prot, FALSE);
783 #ifndef _SYS_SYSPROTO_H_
784 struct __getrlimit_args {
791 sys_getrlimit(struct thread *td, struct __getrlimit_args *uap)
796 if (uap->which >= RLIM_NLIMITS)
798 lim_rlimit(td, uap->which, &rlim);
799 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
804 * Transform the running time and tick information for children of proc p
805 * into user and system time usage.
808 calccru(struct proc *p, struct timeval *up, struct timeval *sp)
811 PROC_LOCK_ASSERT(p, MA_OWNED);
812 calcru1(p, &p->p_crux, up, sp);
816 * Transform the running time and tick information in proc p into user
817 * and system time usage. If appropriate, include the current time slice
821 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
826 PROC_LOCK_ASSERT(p, MA_OWNED);
827 PROC_STATLOCK_ASSERT(p, MA_OWNED);
829 * If we are getting stats for the current process, then add in the
830 * stats that this thread has accumulated in its current time slice.
831 * We reset the thread and CPU state as if we had performed a context
835 if (td->td_proc == p) {
837 runtime = u - PCPU_GET(switchtime);
838 td->td_runtime += runtime;
839 td->td_incruntime += runtime;
840 PCPU_SET(switchtime, u);
842 /* Make sure the per-thread stats are current. */
843 FOREACH_THREAD_IN_PROC(p, td) {
844 if (td->td_incruntime == 0)
848 calcru1(p, &p->p_rux, up, sp);
851 /* Collect resource usage for a single thread. */
853 rufetchtd(struct thread *td, struct rusage *ru)
859 PROC_STATLOCK_ASSERT(p, MA_OWNED);
860 THREAD_LOCK_ASSERT(td, MA_OWNED);
862 * If we are getting stats for the current thread, then add in the
863 * stats that this thread has accumulated in its current time slice.
864 * We reset the thread and CPU state as if we had performed a context
867 if (td == curthread) {
869 runtime = u - PCPU_GET(switchtime);
870 td->td_runtime += runtime;
871 td->td_incruntime += runtime;
872 PCPU_SET(switchtime, u);
874 ruxagg_locked(p, td);
876 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime);
879 /* XXX: the MI version is too slow to use: */
880 #ifndef __HAVE_INLINE_FLSLL
881 #define flsll(x) (fls((x) >> 32) != 0 ? fls((x) >> 32) + 32 : fls(x))
885 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c)
887 uint64_t acc, bh, bl;
891 * Calculate (a * b) / c accurately enough without overflowing. c
892 * must be nonzero, and its top bit must be 0. a or b must be
893 * <= c, and the implementation is tuned for b <= c.
895 * The comments about times are for use in calcru1() with units of
896 * microseconds for 'a' and stathz ticks at 128 Hz for b and c.
898 * Let n be the number of top zero bits in c. Each iteration
899 * either returns, or reduces b by right shifting it by at least n.
900 * The number of iterations is at most 1 + 64 / n, and the error is
901 * at most the number of iterations.
903 * It is very unusual to need even 2 iterations. Previous
904 * implementations overflowed essentially by returning early in the
905 * first iteration, with n = 38 giving overflow at 105+ hours and
906 * n = 32 giving overlow at at 388+ days despite a more careful
907 * calculation. 388 days is a reasonable uptime, and the calculation
908 * needs to work for the uptime times the number of CPUs since 'a'
911 if (a >= (uint64_t)1 << 63)
912 return (0); /* Unsupported arg -- can't happen. */
914 for (i = 0; i < 128; i++) {
918 /* Up to 105 hours on first iteration. */
919 return (acc + (a * b) / c);
922 * This reduction is based on a = q * c + r, with the
923 * remainder r < c. 'a' may be large to start, and
924 * moving bits from b into 'a' at the end of the loop
925 * sets the top bit of 'a', so the reduction makes
926 * significant progress.
932 /* Up to 388 days on first iteration. */
933 return (acc + (a * b) / c);
937 * This step writes a * b as a * ((bh << s) + bl) =
938 * a * (bh << s) + a * bl = (a << s) * bh + a * bl. The 2
939 * additive terms are handled separately. Splitting in
940 * this way is linear except for rounding errors.
942 * s = 64 - sa is the maximum such that a << s fits in 64
943 * bits. Since a < c and c has at least 1 zero top bit,
944 * sa < 64 and s > 0. Thus this step makes progress by
945 * reducing b (it increases 'a', but taking remainders on
946 * the next iteration completes the reduction).
948 * Finally, the choice for s is just what is needed to keep
949 * a * bl from overflowing, so we don't need complications
950 * like a recursive call mul64_by_fraction(a, bl, c) to
951 * handle the second additive term.
960 return (0); /* Algorithm failure -- can't happen. */
964 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
967 /* {user, system, interrupt, total} {ticks, usec}: */
968 uint64_t ut, uu, st, su, it, tt, tu;
970 ut = ruxp->rux_uticks;
971 st = ruxp->rux_sticks;
972 it = ruxp->rux_iticks;
975 /* Avoid divide by zero */
979 tu = cputick2usec(ruxp->rux_runtime);
980 if ((int64_t)tu < 0) {
981 /* XXX: this should be an assert /phk */
982 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
983 (intmax_t)tu, p->p_pid, p->p_comm);
987 /* Subdivide tu. Avoid overflow in the multiplications. */
988 if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) {
989 /* Up to 76 hours when stathz is 128. */
993 uu = mul64_by_fraction(tu, ut, tt);
994 su = mul64_by_fraction(tu, st, tt);
997 if (tu >= ruxp->rux_tu) {
999 * The normal case, time increased.
1000 * Enforce monotonicity of bucketed numbers.
1002 if (uu < ruxp->rux_uu)
1004 if (su < ruxp->rux_su)
1006 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
1008 * When we calibrate the cputicker, it is not uncommon to
1009 * see the presumably fixed frequency increase slightly over
1010 * time as a result of thermal stabilization and NTP
1011 * discipline (of the reference clock). We therefore ignore
1012 * a bit of backwards slop because we expect to catch up
1013 * shortly. We use a 3 microsecond limit to catch low
1014 * counts and a 1% limit for high counts.
1019 } else { /* tu < ruxp->rux_tu */
1021 * What happened here was likely that a laptop, which ran at
1022 * a reduced clock frequency at boot, kicked into high gear.
1023 * The wisdom of spamming this message in that case is
1024 * dubious, but it might also be indicative of something
1025 * serious, so lets keep it and hope laptops can be made
1026 * more truthful about their CPU speed via ACPI.
1028 printf("calcru: runtime went backwards from %ju usec "
1029 "to %ju usec for pid %d (%s)\n",
1030 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
1031 p->p_pid, p->p_comm);
1038 up->tv_sec = uu / 1000000;
1039 up->tv_usec = uu % 1000000;
1040 sp->tv_sec = su / 1000000;
1041 sp->tv_usec = su % 1000000;
1044 #ifndef _SYS_SYSPROTO_H_
1045 struct getrusage_args {
1047 struct rusage *rusage;
1051 sys_getrusage(struct thread *td, struct getrusage_args *uap)
1056 error = kern_getrusage(td, uap->who, &ru);
1058 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
1063 kern_getrusage(struct thread *td, int who, struct rusage *rup)
1073 rufetchcalc(p, rup, &rup->ru_utime,
1077 case RUSAGE_CHILDREN:
1078 *rup = p->p_stats->p_cru;
1079 calccru(p, &rup->ru_utime, &rup->ru_stime);
1098 rucollect(struct rusage *ru, struct rusage *ru2)
1103 if (ru->ru_maxrss < ru2->ru_maxrss)
1104 ru->ru_maxrss = ru2->ru_maxrss;
1106 ip2 = &ru2->ru_first;
1107 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
1112 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1113 struct rusage_ext *rux2)
1116 rux->rux_runtime += rux2->rux_runtime;
1117 rux->rux_uticks += rux2->rux_uticks;
1118 rux->rux_sticks += rux2->rux_sticks;
1119 rux->rux_iticks += rux2->rux_iticks;
1120 rux->rux_uu += rux2->rux_uu;
1121 rux->rux_su += rux2->rux_su;
1122 rux->rux_tu += rux2->rux_tu;
1127 * Aggregate tick counts into the proc's rusage_ext.
1130 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td)
1133 rux->rux_runtime += td->td_incruntime;
1134 rux->rux_uticks += td->td_uticks;
1135 rux->rux_sticks += td->td_sticks;
1136 rux->rux_iticks += td->td_iticks;
1140 ruxagg_locked(struct proc *p, struct thread *td)
1142 THREAD_LOCK_ASSERT(td, MA_OWNED);
1143 PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED);
1145 ruxagg_ext_locked(&p->p_rux, td);
1146 ruxagg_ext_locked(&td->td_rux, td);
1147 td->td_incruntime = 0;
1154 ruxagg(struct proc *p, struct thread *td)
1158 ruxagg_locked(p, td);
1163 * Update the rusage_ext structure and fetch a valid aggregate rusage
1164 * for proc p if storage for one is supplied.
1167 rufetch(struct proc *p, struct rusage *ru)
1171 PROC_STATLOCK_ASSERT(p, MA_OWNED);
1174 if (p->p_numthreads > 0) {
1175 FOREACH_THREAD_IN_PROC(p, td) {
1177 rucollect(ru, &td->td_ru);
1183 * Atomically perform a rufetch and a calcru together.
1184 * Consumers, can safely assume the calcru is executed only once
1185 * rufetch is completed.
1188 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1199 * Allocate a new resource limits structure and initialize its
1200 * reference count and mutex pointer.
1205 struct plimit *limp;
1207 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1208 refcount_init(&limp->pl_refcnt, 1);
1213 lim_hold(struct plimit *limp)
1216 refcount_acquire(&limp->pl_refcnt);
1221 lim_fork(struct proc *p1, struct proc *p2)
1224 PROC_LOCK_ASSERT(p1, MA_OWNED);
1225 PROC_LOCK_ASSERT(p2, MA_OWNED);
1227 p2->p_limit = lim_hold(p1->p_limit);
1228 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1229 if (p1->p_cpulimit != RLIM_INFINITY)
1230 callout_reset_sbt(&p2->p_limco, SBT_1S, 0,
1231 lim_cb, p2, C_PREL(1));
1235 lim_free(struct plimit *limp)
1238 if (refcount_release(&limp->pl_refcnt))
1239 free((void *)limp, M_PLIMIT);
1243 * Make a copy of the plimit structure.
1244 * We share these structures copy-on-write after fork.
1247 lim_copy(struct plimit *dst, struct plimit *src)
1250 KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit"));
1251 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1255 * Return the hard limit for a particular system resource. The
1256 * which parameter specifies the index into the rlimit array.
1259 lim_max(struct thread *td, int which)
1263 lim_rlimit(td, which, &rl);
1264 return (rl.rlim_max);
1268 lim_max_proc(struct proc *p, int which)
1272 lim_rlimit_proc(p, which, &rl);
1273 return (rl.rlim_max);
1277 * Return the current (soft) limit for a particular system resource.
1278 * The which parameter which specifies the index into the rlimit array
1281 (lim_cur)(struct thread *td, int which)
1285 lim_rlimit(td, which, &rl);
1286 return (rl.rlim_cur);
1290 lim_cur_proc(struct proc *p, int which)
1294 lim_rlimit_proc(p, which, &rl);
1295 return (rl.rlim_cur);
1299 * Return a copy of the entire rlimit structure for the system limit
1300 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1303 lim_rlimit(struct thread *td, int which, struct rlimit *rlp)
1305 struct proc *p = td->td_proc;
1307 MPASS(td == curthread);
1308 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1309 ("request for invalid resource limit"));
1310 *rlp = td->td_limit->pl_rlimit[which];
1311 if (p->p_sysent->sv_fixlimit != NULL)
1312 p->p_sysent->sv_fixlimit(rlp, which);
1316 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp)
1319 PROC_LOCK_ASSERT(p, MA_OWNED);
1320 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1321 ("request for invalid resource limit"));
1322 *rlp = p->p_limit->pl_rlimit[which];
1323 if (p->p_sysent->sv_fixlimit != NULL)
1324 p->p_sysent->sv_fixlimit(rlp, which);
1331 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1332 rw_init(&uihashtbl_lock, "uidinfo hash");
1336 * Look up a uidinfo struct for the parameter uid.
1337 * uihashtbl_lock must be locked.
1338 * Increase refcount on uidinfo struct returned.
1340 static struct uidinfo *
1343 struct uihashhead *uipp;
1344 struct uidinfo *uip;
1346 rw_assert(&uihashtbl_lock, RA_LOCKED);
1348 LIST_FOREACH(uip, uipp, ui_hash)
1349 if (uip->ui_uid == uid) {
1358 * Find or allocate a struct uidinfo for a particular uid.
1359 * Returns with uidinfo struct referenced.
1360 * uifree() should be called on a struct uidinfo when released.
1365 struct uidinfo *new_uip, *uip;
1368 cred = curthread->td_ucred;
1369 if (cred->cr_uidinfo->ui_uid == uid) {
1370 uip = cred->cr_uidinfo;
1373 } else if (cred->cr_ruidinfo->ui_uid == uid) {
1374 uip = cred->cr_ruidinfo;
1379 rw_rlock(&uihashtbl_lock);
1380 uip = uilookup(uid);
1381 rw_runlock(&uihashtbl_lock);
1385 new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO);
1386 racct_create(&new_uip->ui_racct);
1387 refcount_init(&new_uip->ui_ref, 1);
1388 new_uip->ui_uid = uid;
1390 rw_wlock(&uihashtbl_lock);
1392 * There's a chance someone created our uidinfo while we
1393 * were in malloc and not holding the lock, so we have to
1394 * make sure we don't insert a duplicate uidinfo.
1396 if ((uip = uilookup(uid)) == NULL) {
1397 LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash);
1398 rw_wunlock(&uihashtbl_lock);
1401 rw_wunlock(&uihashtbl_lock);
1402 racct_destroy(&new_uip->ui_racct);
1403 free(new_uip, M_UIDINFO);
1409 * Place another refcount on a uidinfo struct.
1412 uihold(struct uidinfo *uip)
1415 refcount_acquire(&uip->ui_ref);
1419 * Since uidinfo structs have a long lifetime, we use an
1420 * opportunistic refcounting scheme to avoid locking the lookup hash
1423 * If the refcount hits 0, we need to free the structure,
1424 * which means we need to lock the hash.
1426 * After locking the struct and lowering the refcount, if we find
1427 * that we don't need to free, simply unlock and return.
1429 * If refcount lowering results in need to free, bump the count
1430 * back up, lose the lock and acquire the locks in the proper
1431 * order to try again.
1434 uifree(struct uidinfo *uip)
1437 if (refcount_release_if_not_last(&uip->ui_ref))
1440 rw_wlock(&uihashtbl_lock);
1441 if (refcount_release(&uip->ui_ref) == 0) {
1442 rw_wunlock(&uihashtbl_lock);
1446 racct_destroy(&uip->ui_racct);
1447 LIST_REMOVE(uip, ui_hash);
1448 rw_wunlock(&uihashtbl_lock);
1450 if (uip->ui_sbsize != 0)
1451 printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1452 uip->ui_uid, uip->ui_sbsize);
1453 if (uip->ui_proccnt != 0)
1454 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1455 uip->ui_uid, uip->ui_proccnt);
1456 if (uip->ui_vmsize != 0)
1457 printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1458 uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1459 free(uip, M_UIDINFO);
1464 ui_racct_foreach(void (*callback)(struct racct *racct,
1465 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
1466 void *arg2, void *arg3)
1468 struct uidinfo *uip;
1469 struct uihashhead *uih;
1471 rw_rlock(&uihashtbl_lock);
1474 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) {
1475 LIST_FOREACH(uip, uih, ui_hash) {
1476 (callback)(uip->ui_racct, arg2, arg3);
1481 rw_runlock(&uihashtbl_lock);
1486 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name)
1490 /* Don't allow them to exceed max, but allow subtraction. */
1491 new = atomic_fetchadd_long(limit, (long)diff) + diff;
1492 if (diff > 0 && max != 0) {
1493 if (new < 0 || new > max) {
1494 atomic_subtract_long(limit, (long)diff);
1498 printf("negative %s for uid = %d\n", name, uip->ui_uid);
1503 * Change the count associated with number of processes
1504 * a given user is using. When 'max' is 0, don't enforce a limit
1507 chgproccnt(struct uidinfo *uip, int diff, rlim_t max)
1510 return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt"));
1514 * Change the total socket buffer size a user has used.
1517 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max)
1522 if (diff > 0 && max == 0) {
1525 rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize");
1533 * Change the count associated with number of pseudo-terminals
1534 * a given user is using. When 'max' is 0, don't enforce a limit
1537 chgptscnt(struct uidinfo *uip, int diff, rlim_t max)
1540 return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt"));
1544 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max)
1547 return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt"));
1551 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max)
1554 return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt"));