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>
62 #include <sys/umtxvar.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)
110 low = td->td_proc->p_nice;
115 if (p_cansee(td, p) == 0)
122 sx_slock(&proctree_lock);
124 pg = td->td_proc->p_pgrp;
129 sx_sunlock(&proctree_lock);
133 sx_sunlock(&proctree_lock);
134 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
136 if (p->p_state == PRS_NORMAL &&
137 p_cansee(td, p) == 0) {
148 who = td->td_ucred->cr_uid;
149 sx_slock(&allproc_lock);
150 FOREACH_PROC_IN_SYSTEM(p) {
152 if (p->p_state == PRS_NORMAL &&
153 p_cansee(td, p) == 0 &&
154 p->p_ucred->cr_uid == who) {
160 sx_sunlock(&allproc_lock);
167 if (low == PRIO_MAX + 1 && error == 0)
169 td->td_retval[0] = low;
173 #ifndef _SYS_SYSPROTO_H_
174 struct setpriority_args {
181 sys_setpriority(struct thread *td, struct setpriority_args *uap)
184 return (kern_setpriority(td, uap->which, uap->who, uap->prio));
188 kern_setpriority(struct thread *td, int which, int who, int prio)
190 struct proc *curp, *p;
192 int found = 0, error = 0;
199 error = donice(td, curp, prio);
205 error = p_cansee(td, p);
207 error = donice(td, p, prio);
214 sx_slock(&proctree_lock);
221 sx_sunlock(&proctree_lock);
225 sx_sunlock(&proctree_lock);
226 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
228 if (p->p_state == PRS_NORMAL &&
229 p_cansee(td, p) == 0) {
230 error = donice(td, p, prio);
240 who = td->td_ucred->cr_uid;
241 sx_slock(&allproc_lock);
242 FOREACH_PROC_IN_SYSTEM(p) {
244 if (p->p_state == PRS_NORMAL &&
245 p->p_ucred->cr_uid == who &&
246 p_cansee(td, p) == 0) {
247 error = donice(td, p, prio);
252 sx_sunlock(&allproc_lock);
259 if (found == 0 && error == 0)
265 * Set "nice" for a (whole) process.
268 donice(struct thread *td, struct proc *p, int n)
272 PROC_LOCK_ASSERT(p, MA_OWNED);
273 if ((error = p_cansched(td, p)))
279 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
285 static int unprivileged_idprio;
286 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW,
287 &unprivileged_idprio, 0,
288 "Allow non-root users to set an idle priority (deprecated)");
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 td1 = tdfind(uap->lwpid, -1);
325 switch (uap->function) {
327 if ((error = p_cansee(td, p)))
329 pri_to_rtp(td1, &rtp);
331 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
333 if ((error = p_cansched(td, p)) || (error = cierror))
336 /* Disallow setting rtprio in most cases if not superuser. */
339 * Realtime priority has to be restricted for reasons which
340 * should be obvious. However, for idleprio processes, there is
341 * a potential for system deadlock if an idleprio process gains
342 * a lock on a resource that other processes need (and the
343 * idleprio process can't run due to a CPU-bound normal
344 * process). Fix me! XXX
346 * This problem is not only related to idleprio process.
347 * A user level program can obtain a file lock and hold it
348 * indefinitely. Additionally, without idleprio processes it is
349 * still conceivable that a program with low priority will never
350 * get to run. In short, allowing this feature might make it
351 * easier to lock a resource indefinitely, but it is not the
352 * only thing that makes it possible.
354 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
355 (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
357 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
358 unprivileged_idprio == 0 &&
359 (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
361 error = rtp_to_pri(&rtp, td1);
372 * Set realtime priority.
374 #ifndef _SYS_SYSPROTO_H_
382 sys_rtprio(struct thread *td, struct rtprio_args *uap)
389 /* Perform copyin before acquiring locks if needed. */
390 if (uap->function == RTP_SET)
391 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
404 switch (uap->function) {
406 if ((error = p_cansee(td, p)))
409 * Return OUR priority if no pid specified,
410 * or if one is, report the highest priority
411 * in the process. There isn't much more you can do as
412 * there is only room to return a single priority.
413 * Note: specifying our own pid is not the same
414 * as leaving it zero.
417 pri_to_rtp(td, &rtp);
421 rtp.type = RTP_PRIO_IDLE;
422 rtp.prio = RTP_PRIO_MAX;
423 FOREACH_THREAD_IN_PROC(p, tdp) {
424 pri_to_rtp(tdp, &rtp2);
425 if (rtp2.type < rtp.type ||
426 (rtp2.type == rtp.type &&
427 rtp2.prio < rtp.prio)) {
428 rtp.type = rtp2.type;
429 rtp.prio = rtp2.prio;
434 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
436 if ((error = p_cansched(td, p)) || (error = cierror))
440 * Disallow setting rtprio in most cases if not superuser.
441 * See the comment in sys_rtprio_thread about idprio
442 * threads holding a lock.
444 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME &&
445 (error = priv_check(td, PRIV_SCHED_RTPRIO)) != 0)
447 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE &&
448 unprivileged_idprio == 0 &&
449 (error = priv_check(td, PRIV_SCHED_IDPRIO)) != 0)
453 * If we are setting our own priority, set just our
454 * thread but if we are doing another process,
455 * do all the threads on that process. If we
456 * specify our own pid we do the latter.
459 error = rtp_to_pri(&rtp, td);
461 FOREACH_THREAD_IN_PROC(p, td) {
462 if ((error = rtp_to_pri(&rtp, td)) != 0)
476 rtp_to_pri(struct rtprio *rtp, struct thread *td)
478 u_char newpri, oldclass, oldpri;
480 switch (RTP_PRIO_BASE(rtp->type)) {
481 case RTP_PRIO_REALTIME:
482 if (rtp->prio > RTP_PRIO_MAX)
484 newpri = PRI_MIN_REALTIME + rtp->prio;
486 case RTP_PRIO_NORMAL:
487 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE))
489 newpri = PRI_MIN_TIMESHARE + rtp->prio;
492 if (rtp->prio > RTP_PRIO_MAX)
494 newpri = PRI_MIN_IDLE + rtp->prio;
501 oldclass = td->td_pri_class;
502 sched_class(td, rtp->type); /* XXX fix */
503 oldpri = td->td_user_pri;
504 sched_user_prio(td, newpri);
505 if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL ||
506 td->td_pri_class != RTP_PRIO_NORMAL))
507 sched_prio(td, td->td_user_pri);
508 if (TD_ON_UPILOCK(td) && oldpri != newpri) {
511 umtx_pi_adjust(td, oldpri);
519 pri_to_rtp(struct thread *td, struct rtprio *rtp)
523 switch (PRI_BASE(td->td_pri_class)) {
525 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
528 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
531 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
536 rtp->type = td->td_pri_class;
540 #if defined(COMPAT_43)
541 #ifndef _SYS_SYSPROTO_H_
542 struct osetrlimit_args {
548 osetrlimit(struct thread *td, struct osetrlimit_args *uap)
554 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
556 lim.rlim_cur = olim.rlim_cur;
557 lim.rlim_max = olim.rlim_max;
558 error = kern_setrlimit(td, uap->which, &lim);
562 #ifndef _SYS_SYSPROTO_H_
563 struct ogetrlimit_args {
569 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap)
575 if (uap->which >= RLIM_NLIMITS)
577 lim_rlimit(td, uap->which, &rl);
580 * XXX would be more correct to convert only RLIM_INFINITY to the
581 * old RLIM_INFINITY and fail with EOVERFLOW for other larger
582 * values. Most 64->32 and 32->16 conversions, including not
583 * unimportant ones of uids are even more broken than what we
584 * do here (they blindly truncate). We don't do this correctly
585 * here since we have little experience with EOVERFLOW yet.
586 * Elsewhere, getuid() can't fail...
588 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
589 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
590 error = copyout(&olim, uap->rlp, sizeof(olim));
593 #endif /* COMPAT_43 */
595 #ifndef _SYS_SYSPROTO_H_
596 struct __setrlimit_args {
602 sys_setrlimit(struct thread *td, struct __setrlimit_args *uap)
607 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
609 error = kern_setrlimit(td, uap->which, &alim);
621 PROC_LOCK_ASSERT(p, MA_OWNED);
623 * Check if the process exceeds its cpu resource allocation. If
624 * it reaches the max, arrange to kill the process in ast().
626 if (p->p_cpulimit == RLIM_INFINITY)
629 FOREACH_THREAD_IN_PROC(p, td) {
633 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
634 lim_rlimit_proc(p, RLIMIT_CPU, &rlim);
635 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
636 killproc(p, "exceeded maximum CPU limit");
638 if (p->p_cpulimit < rlim.rlim_max)
640 kern_psignal(p, SIGXCPU);
643 if ((p->p_flag & P_WEXIT) == 0)
644 callout_reset_sbt(&p->p_limco, SBT_1S, 0,
645 lim_cb, p, C_PREL(1));
649 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp)
652 return (kern_proc_setrlimit(td, td->td_proc, which, limp));
656 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which,
659 struct plimit *newlim, *oldlim;
660 struct rlimit *alimp;
661 struct rlimit oldssiz;
664 if (which >= RLIM_NLIMITS)
668 * Preserve historical bugs by treating negative limits as unsigned.
670 if (limp->rlim_cur < 0)
671 limp->rlim_cur = RLIM_INFINITY;
672 if (limp->rlim_max < 0)
673 limp->rlim_max = RLIM_INFINITY;
675 oldssiz.rlim_cur = 0;
676 newlim = lim_alloc();
679 alimp = &oldlim->pl_rlimit[which];
680 if (limp->rlim_cur > alimp->rlim_max ||
681 limp->rlim_max > alimp->rlim_max)
682 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
687 if (limp->rlim_cur > limp->rlim_max)
688 limp->rlim_cur = limp->rlim_max;
689 lim_copy(newlim, oldlim);
690 alimp = &newlim->pl_rlimit[which];
694 if (limp->rlim_cur != RLIM_INFINITY &&
695 p->p_cpulimit == RLIM_INFINITY)
696 callout_reset_sbt(&p->p_limco, SBT_1S, 0,
697 lim_cb, p, C_PREL(1));
698 p->p_cpulimit = limp->rlim_cur;
701 if (limp->rlim_cur > maxdsiz)
702 limp->rlim_cur = maxdsiz;
703 if (limp->rlim_max > maxdsiz)
704 limp->rlim_max = maxdsiz;
708 if (limp->rlim_cur > maxssiz)
709 limp->rlim_cur = maxssiz;
710 if (limp->rlim_max > maxssiz)
711 limp->rlim_max = maxssiz;
713 if (p->p_sysent->sv_fixlimit != NULL)
714 p->p_sysent->sv_fixlimit(&oldssiz,
719 if (limp->rlim_cur > maxfilesperproc)
720 limp->rlim_cur = maxfilesperproc;
721 if (limp->rlim_max > maxfilesperproc)
722 limp->rlim_max = maxfilesperproc;
726 if (limp->rlim_cur > maxprocperuid)
727 limp->rlim_cur = maxprocperuid;
728 if (limp->rlim_max > maxprocperuid)
729 limp->rlim_max = maxprocperuid;
730 if (limp->rlim_cur < 1)
732 if (limp->rlim_max < 1)
736 if (p->p_sysent->sv_fixlimit != NULL)
737 p->p_sysent->sv_fixlimit(limp, which);
744 if (which == RLIMIT_STACK &&
746 * Skip calls from exec_new_vmspace(), done when stack is
749 (td != curthread || (p->p_flag & P_INEXEC) == 0)) {
751 * Stack is allocated to the max at exec time with only
752 * "rlim_cur" bytes accessible. If stack limit is going
753 * up make more accessible, if going down make inaccessible.
755 if (limp->rlim_cur != oldssiz.rlim_cur) {
760 if (limp->rlim_cur > oldssiz.rlim_cur) {
761 prot = p->p_sysent->sv_stackprot;
762 size = limp->rlim_cur - oldssiz.rlim_cur;
763 addr = round_page(p->p_vmspace->vm_stacktop) -
767 size = oldssiz.rlim_cur - limp->rlim_cur;
768 addr = round_page(p->p_vmspace->vm_stacktop) -
771 addr = trunc_page(addr);
772 size = round_page(size);
773 (void)vm_map_protect(&p->p_vmspace->vm_map,
774 addr, addr + size, prot, 0,
775 VM_MAP_PROTECT_SET_PROT);
782 #ifndef _SYS_SYSPROTO_H_
783 struct __getrlimit_args {
790 sys_getrlimit(struct thread *td, struct __getrlimit_args *uap)
795 if (uap->which >= RLIM_NLIMITS)
797 lim_rlimit(td, uap->which, &rlim);
798 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
803 * Transform the running time and tick information for children of proc p
804 * into user and system time usage.
807 calccru(struct proc *p, struct timeval *up, struct timeval *sp)
810 PROC_LOCK_ASSERT(p, MA_OWNED);
811 calcru1(p, &p->p_crux, up, sp);
815 * Transform the running time and tick information in proc p into user
816 * and system time usage. If appropriate, include the current time slice
820 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
825 PROC_LOCK_ASSERT(p, MA_OWNED);
826 PROC_STATLOCK_ASSERT(p, MA_OWNED);
828 * If we are getting stats for the current process, then add in the
829 * stats that this thread has accumulated in its current time slice.
830 * We reset the thread and CPU state as if we had performed a context
834 if (td->td_proc == p) {
836 runtime = u - PCPU_GET(switchtime);
837 td->td_runtime += runtime;
838 td->td_incruntime += runtime;
839 PCPU_SET(switchtime, u);
841 /* Make sure the per-thread stats are current. */
842 FOREACH_THREAD_IN_PROC(p, td) {
843 if (td->td_incruntime == 0)
847 calcru1(p, &p->p_rux, up, sp);
850 /* Collect resource usage for a single thread. */
852 rufetchtd(struct thread *td, struct rusage *ru)
858 PROC_STATLOCK_ASSERT(p, MA_OWNED);
859 THREAD_LOCK_ASSERT(td, MA_OWNED);
861 * If we are getting stats for the current thread, then add in the
862 * stats that this thread has accumulated in its current time slice.
863 * We reset the thread and CPU state as if we had performed a context
866 if (td == curthread) {
868 runtime = u - PCPU_GET(switchtime);
869 td->td_runtime += runtime;
870 td->td_incruntime += runtime;
871 PCPU_SET(switchtime, u);
873 ruxagg_locked(p, td);
875 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime);
878 /* XXX: the MI version is too slow to use: */
879 #ifndef __HAVE_INLINE_FLSLL
880 #define flsll(x) (fls((x) >> 32) != 0 ? fls((x) >> 32) + 32 : fls(x))
884 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c)
886 uint64_t acc, bh, bl;
890 * Calculate (a * b) / c accurately enough without overflowing. c
891 * must be nonzero, and its top bit must be 0. a or b must be
892 * <= c, and the implementation is tuned for b <= c.
894 * The comments about times are for use in calcru1() with units of
895 * microseconds for 'a' and stathz ticks at 128 Hz for b and c.
897 * Let n be the number of top zero bits in c. Each iteration
898 * either returns, or reduces b by right shifting it by at least n.
899 * The number of iterations is at most 1 + 64 / n, and the error is
900 * at most the number of iterations.
902 * It is very unusual to need even 2 iterations. Previous
903 * implementations overflowed essentially by returning early in the
904 * first iteration, with n = 38 giving overflow at 105+ hours and
905 * n = 32 giving overlow at at 388+ days despite a more careful
906 * calculation. 388 days is a reasonable uptime, and the calculation
907 * needs to work for the uptime times the number of CPUs since 'a'
910 if (a >= (uint64_t)1 << 63)
911 return (0); /* Unsupported arg -- can't happen. */
913 for (i = 0; i < 128; i++) {
917 /* Up to 105 hours on first iteration. */
918 return (acc + (a * b) / c);
921 * This reduction is based on a = q * c + r, with the
922 * remainder r < c. 'a' may be large to start, and
923 * moving bits from b into 'a' at the end of the loop
924 * sets the top bit of 'a', so the reduction makes
925 * significant progress.
931 /* Up to 388 days on first iteration. */
932 return (acc + (a * b) / c);
936 * This step writes a * b as a * ((bh << s) + bl) =
937 * a * (bh << s) + a * bl = (a << s) * bh + a * bl. The 2
938 * additive terms are handled separately. Splitting in
939 * this way is linear except for rounding errors.
941 * s = 64 - sa is the maximum such that a << s fits in 64
942 * bits. Since a < c and c has at least 1 zero top bit,
943 * sa < 64 and s > 0. Thus this step makes progress by
944 * reducing b (it increases 'a', but taking remainders on
945 * the next iteration completes the reduction).
947 * Finally, the choice for s is just what is needed to keep
948 * a * bl from overflowing, so we don't need complications
949 * like a recursive call mul64_by_fraction(a, bl, c) to
950 * handle the second additive term.
959 return (0); /* Algorithm failure -- can't happen. */
963 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
966 /* {user, system, interrupt, total} {ticks, usec}: */
967 uint64_t ut, uu, st, su, it, tt, tu;
969 ut = ruxp->rux_uticks;
970 st = ruxp->rux_sticks;
971 it = ruxp->rux_iticks;
974 /* Avoid divide by zero */
978 tu = cputick2usec(ruxp->rux_runtime);
979 if ((int64_t)tu < 0) {
980 /* XXX: this should be an assert /phk */
981 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
982 (intmax_t)tu, p->p_pid, p->p_comm);
986 /* Subdivide tu. Avoid overflow in the multiplications. */
987 if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) {
988 /* Up to 76 hours when stathz is 128. */
992 uu = mul64_by_fraction(tu, ut, tt);
993 su = mul64_by_fraction(tu, st, tt);
996 if (tu >= ruxp->rux_tu) {
998 * The normal case, time increased.
999 * Enforce monotonicity of bucketed numbers.
1001 if (uu < ruxp->rux_uu)
1003 if (su < ruxp->rux_su)
1005 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
1007 * When we calibrate the cputicker, it is not uncommon to
1008 * see the presumably fixed frequency increase slightly over
1009 * time as a result of thermal stabilization and NTP
1010 * discipline (of the reference clock). We therefore ignore
1011 * a bit of backwards slop because we expect to catch up
1012 * shortly. We use a 3 microsecond limit to catch low
1013 * counts and a 1% limit for high counts.
1018 } else { /* tu < ruxp->rux_tu */
1020 * What happened here was likely that a laptop, which ran at
1021 * a reduced clock frequency at boot, kicked into high gear.
1022 * The wisdom of spamming this message in that case is
1023 * dubious, but it might also be indicative of something
1024 * serious, so lets keep it and hope laptops can be made
1025 * more truthful about their CPU speed via ACPI.
1027 printf("calcru: runtime went backwards from %ju usec "
1028 "to %ju usec for pid %d (%s)\n",
1029 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
1030 p->p_pid, p->p_comm);
1037 up->tv_sec = uu / 1000000;
1038 up->tv_usec = uu % 1000000;
1039 sp->tv_sec = su / 1000000;
1040 sp->tv_usec = su % 1000000;
1043 #ifndef _SYS_SYSPROTO_H_
1044 struct getrusage_args {
1046 struct rusage *rusage;
1050 sys_getrusage(struct thread *td, struct getrusage_args *uap)
1055 error = kern_getrusage(td, uap->who, &ru);
1057 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
1062 kern_getrusage(struct thread *td, int who, struct rusage *rup)
1072 rufetchcalc(p, rup, &rup->ru_utime,
1076 case RUSAGE_CHILDREN:
1077 *rup = p->p_stats->p_cru;
1078 calccru(p, &rup->ru_utime, &rup->ru_stime);
1097 rucollect(struct rusage *ru, struct rusage *ru2)
1102 if (ru->ru_maxrss < ru2->ru_maxrss)
1103 ru->ru_maxrss = ru2->ru_maxrss;
1105 ip2 = &ru2->ru_first;
1106 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
1111 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
1112 struct rusage_ext *rux2)
1115 rux->rux_runtime += rux2->rux_runtime;
1116 rux->rux_uticks += rux2->rux_uticks;
1117 rux->rux_sticks += rux2->rux_sticks;
1118 rux->rux_iticks += rux2->rux_iticks;
1119 rux->rux_uu += rux2->rux_uu;
1120 rux->rux_su += rux2->rux_su;
1121 rux->rux_tu += rux2->rux_tu;
1126 * Aggregate tick counts into the proc's rusage_ext.
1129 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td)
1132 rux->rux_runtime += td->td_incruntime;
1133 rux->rux_uticks += td->td_uticks;
1134 rux->rux_sticks += td->td_sticks;
1135 rux->rux_iticks += td->td_iticks;
1139 ruxagg_locked(struct proc *p, struct thread *td)
1141 THREAD_LOCK_ASSERT(td, MA_OWNED);
1142 PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED);
1144 ruxagg_ext_locked(&p->p_rux, td);
1145 ruxagg_ext_locked(&td->td_rux, td);
1146 td->td_incruntime = 0;
1153 ruxagg(struct proc *p, struct thread *td)
1157 ruxagg_locked(p, td);
1162 * Update the rusage_ext structure and fetch a valid aggregate rusage
1163 * for proc p if storage for one is supplied.
1166 rufetch(struct proc *p, struct rusage *ru)
1170 PROC_STATLOCK_ASSERT(p, MA_OWNED);
1173 if (p->p_numthreads > 0) {
1174 FOREACH_THREAD_IN_PROC(p, td) {
1176 rucollect(ru, &td->td_ru);
1182 * Atomically perform a rufetch and a calcru together.
1183 * Consumers, can safely assume the calcru is executed only once
1184 * rufetch is completed.
1187 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1198 * Allocate a new resource limits structure and initialize its
1199 * reference count and mutex pointer.
1204 struct plimit *limp;
1206 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1207 refcount_init(&limp->pl_refcnt, 1);
1212 lim_hold(struct plimit *limp)
1215 refcount_acquire(&limp->pl_refcnt);
1220 lim_fork(struct proc *p1, struct proc *p2)
1223 PROC_LOCK_ASSERT(p1, MA_OWNED);
1224 PROC_LOCK_ASSERT(p2, MA_OWNED);
1226 p2->p_limit = lim_hold(p1->p_limit);
1227 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1228 if (p1->p_cpulimit != RLIM_INFINITY)
1229 callout_reset_sbt(&p2->p_limco, SBT_1S, 0,
1230 lim_cb, p2, C_PREL(1));
1234 lim_free(struct plimit *limp)
1237 if (refcount_release(&limp->pl_refcnt))
1238 free((void *)limp, M_PLIMIT);
1242 lim_freen(struct plimit *limp, int n)
1245 if (refcount_releasen(&limp->pl_refcnt, n))
1246 free((void *)limp, M_PLIMIT);
1250 * Make a copy of the plimit structure.
1251 * We share these structures copy-on-write after fork.
1254 lim_copy(struct plimit *dst, struct plimit *src)
1257 KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit"));
1258 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1262 * Return the hard limit for a particular system resource. The
1263 * which parameter specifies the index into the rlimit array.
1266 lim_max(struct thread *td, int which)
1270 lim_rlimit(td, which, &rl);
1271 return (rl.rlim_max);
1275 lim_max_proc(struct proc *p, int which)
1279 lim_rlimit_proc(p, which, &rl);
1280 return (rl.rlim_max);
1284 * Return the current (soft) limit for a particular system resource.
1285 * The which parameter which specifies the index into the rlimit array
1288 (lim_cur)(struct thread *td, int which)
1292 lim_rlimit(td, which, &rl);
1293 return (rl.rlim_cur);
1297 lim_cur_proc(struct proc *p, int which)
1301 lim_rlimit_proc(p, which, &rl);
1302 return (rl.rlim_cur);
1306 * Return a copy of the entire rlimit structure for the system limit
1307 * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1310 lim_rlimit(struct thread *td, int which, struct rlimit *rlp)
1312 struct proc *p = td->td_proc;
1314 MPASS(td == curthread);
1315 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1316 ("request for invalid resource limit"));
1317 *rlp = td->td_limit->pl_rlimit[which];
1318 if (p->p_sysent->sv_fixlimit != NULL)
1319 p->p_sysent->sv_fixlimit(rlp, which);
1323 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp)
1326 PROC_LOCK_ASSERT(p, MA_OWNED);
1327 KASSERT(which >= 0 && which < RLIM_NLIMITS,
1328 ("request for invalid resource limit"));
1329 *rlp = p->p_limit->pl_rlimit[which];
1330 if (p->p_sysent->sv_fixlimit != NULL)
1331 p->p_sysent->sv_fixlimit(rlp, which);
1338 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1339 rw_init(&uihashtbl_lock, "uidinfo hash");
1343 * Look up a uidinfo struct for the parameter uid.
1344 * uihashtbl_lock must be locked.
1345 * Increase refcount on uidinfo struct returned.
1347 static struct uidinfo *
1350 struct uihashhead *uipp;
1351 struct uidinfo *uip;
1353 rw_assert(&uihashtbl_lock, RA_LOCKED);
1355 LIST_FOREACH(uip, uipp, ui_hash)
1356 if (uip->ui_uid == uid) {
1365 * Find or allocate a struct uidinfo for a particular uid.
1366 * Returns with uidinfo struct referenced.
1367 * uifree() should be called on a struct uidinfo when released.
1372 struct uidinfo *new_uip, *uip;
1375 cred = curthread->td_ucred;
1376 if (cred->cr_uidinfo->ui_uid == uid) {
1377 uip = cred->cr_uidinfo;
1380 } else if (cred->cr_ruidinfo->ui_uid == uid) {
1381 uip = cred->cr_ruidinfo;
1386 rw_rlock(&uihashtbl_lock);
1387 uip = uilookup(uid);
1388 rw_runlock(&uihashtbl_lock);
1392 new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO);
1393 racct_create(&new_uip->ui_racct);
1394 refcount_init(&new_uip->ui_ref, 1);
1395 new_uip->ui_uid = uid;
1397 rw_wlock(&uihashtbl_lock);
1399 * There's a chance someone created our uidinfo while we
1400 * were in malloc and not holding the lock, so we have to
1401 * make sure we don't insert a duplicate uidinfo.
1403 if ((uip = uilookup(uid)) == NULL) {
1404 LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash);
1405 rw_wunlock(&uihashtbl_lock);
1408 rw_wunlock(&uihashtbl_lock);
1409 racct_destroy(&new_uip->ui_racct);
1410 free(new_uip, M_UIDINFO);
1416 * Place another refcount on a uidinfo struct.
1419 uihold(struct uidinfo *uip)
1422 refcount_acquire(&uip->ui_ref);
1426 * Since uidinfo structs have a long lifetime, we use an
1427 * opportunistic refcounting scheme to avoid locking the lookup hash
1430 * If the refcount hits 0, we need to free the structure,
1431 * which means we need to lock the hash.
1433 * After locking the struct and lowering the refcount, if we find
1434 * that we don't need to free, simply unlock and return.
1436 * If refcount lowering results in need to free, bump the count
1437 * back up, lose the lock and acquire the locks in the proper
1438 * order to try again.
1441 uifree(struct uidinfo *uip)
1444 if (refcount_release_if_not_last(&uip->ui_ref))
1447 rw_wlock(&uihashtbl_lock);
1448 if (refcount_release(&uip->ui_ref) == 0) {
1449 rw_wunlock(&uihashtbl_lock);
1453 racct_destroy(&uip->ui_racct);
1454 LIST_REMOVE(uip, ui_hash);
1455 rw_wunlock(&uihashtbl_lock);
1457 if (uip->ui_sbsize != 0)
1458 printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1459 uip->ui_uid, uip->ui_sbsize);
1460 if (uip->ui_proccnt != 0)
1461 printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1462 uip->ui_uid, uip->ui_proccnt);
1463 if (uip->ui_vmsize != 0)
1464 printf("freeing uidinfo: uid = %d, swapuse = %lld\n",
1465 uip->ui_uid, (unsigned long long)uip->ui_vmsize);
1466 free(uip, M_UIDINFO);
1471 ui_racct_foreach(void (*callback)(struct racct *racct,
1472 void *arg2, void *arg3), void (*pre)(void), void (*post)(void),
1473 void *arg2, void *arg3)
1475 struct uidinfo *uip;
1476 struct uihashhead *uih;
1478 rw_rlock(&uihashtbl_lock);
1481 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) {
1482 LIST_FOREACH(uip, uih, ui_hash) {
1483 (callback)(uip->ui_racct, arg2, arg3);
1488 rw_runlock(&uihashtbl_lock);
1493 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name)
1497 /* Don't allow them to exceed max, but allow subtraction. */
1498 new = atomic_fetchadd_long(limit, (long)diff) + diff;
1499 if (diff > 0 && max != 0) {
1500 if (new < 0 || new > max) {
1501 atomic_subtract_long(limit, (long)diff);
1505 printf("negative %s for uid = %d\n", name, uip->ui_uid);
1510 * Change the count associated with number of processes
1511 * a given user is using. When 'max' is 0, don't enforce a limit
1514 chgproccnt(struct uidinfo *uip, int diff, rlim_t max)
1517 return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt"));
1521 * Change the total socket buffer size a user has used.
1524 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max)
1529 if (diff > 0 && max == 0) {
1532 rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize");
1540 * Change the count associated with number of pseudo-terminals
1541 * a given user is using. When 'max' is 0, don't enforce a limit
1544 chgptscnt(struct uidinfo *uip, int diff, rlim_t max)
1547 return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt"));
1551 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max)
1554 return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt"));
1558 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max)
1561 return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt"));