2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
4 * Copyright (c) 2010 The FreeBSD Foundation
6 * This software was developed by Edward Tomasz Napierala under sponsorship
7 * from the FreeBSD Foundation.
9 * Redistribution and use in source and binary forms, with or without
10 * modification, are permitted provided that the following conditions
12 * 1. Redistributions of source code must retain the above copyright
13 * notice, this list of conditions and the following disclaimer.
14 * 2. Redistributions in binary form must reproduce the above copyright
15 * notice, this list of conditions and the following disclaimer in the
16 * documentation and/or other materials provided with the distribution.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
33 #include <sys/cdefs.h>
34 __FBSDID("$FreeBSD$");
36 #include <sys/param.h>
37 #include <sys/devctl.h>
38 #include <sys/malloc.h>
39 #include <sys/queue.h>
40 #include <sys/refcount.h>
42 #include <sys/kernel.h>
43 #include <sys/limits.h>
44 #include <sys/loginclass.h>
47 #include <sys/racct.h>
49 #include <sys/resourcevar.h>
51 #include <sys/sysproto.h>
52 #include <sys/systm.h>
53 #include <sys/types.h>
54 #include <sys/eventhandler.h>
56 #include <sys/mutex.h>
57 #include <sys/rwlock.h>
59 #include <sys/taskqueue.h>
65 #error "The RCTL option requires the RACCT option"
68 FEATURE(rctl, "Resource Limits");
71 #define HRF_DONT_INHERIT 1
72 #define HRF_DONT_ACCUMULATE 2
74 #define RCTL_MAX_INBUFSIZE 4 * 1024
75 #define RCTL_MAX_OUTBUFSIZE 16 * 1024 * 1024
76 #define RCTL_LOG_BUFSIZE 128
78 #define RCTL_PCPU_SHIFT (10 * 1000000)
80 static unsigned int rctl_maxbufsize = RCTL_MAX_OUTBUFSIZE;
81 static int rctl_log_rate_limit = 10;
82 static int rctl_devctl_rate_limit = 10;
85 * Values below are initialized in rctl_init().
87 static int rctl_throttle_min = -1;
88 static int rctl_throttle_max = -1;
89 static int rctl_throttle_pct = -1;
90 static int rctl_throttle_pct2 = -1;
92 static int rctl_throttle_min_sysctl(SYSCTL_HANDLER_ARGS);
93 static int rctl_throttle_max_sysctl(SYSCTL_HANDLER_ARGS);
94 static int rctl_throttle_pct_sysctl(SYSCTL_HANDLER_ARGS);
95 static int rctl_throttle_pct2_sysctl(SYSCTL_HANDLER_ARGS);
97 SYSCTL_NODE(_kern_racct, OID_AUTO, rctl, CTLFLAG_RW | CTLFLAG_MPSAFE, 0,
99 SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, maxbufsize, CTLFLAG_RWTUN,
100 &rctl_maxbufsize, 0, "Maximum output buffer size");
101 SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, log_rate_limit, CTLFLAG_RW,
102 &rctl_log_rate_limit, 0, "Maximum number of log messages per second");
103 SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, devctl_rate_limit, CTLFLAG_RWTUN,
104 &rctl_devctl_rate_limit, 0, "Maximum number of devctl messages per second");
105 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_min,
106 CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, 0,
107 &rctl_throttle_min_sysctl, "IU",
108 "Shortest throttling duration, in hz");
109 TUNABLE_INT("kern.racct.rctl.throttle_min", &rctl_throttle_min);
110 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_max,
111 CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, 0,
112 &rctl_throttle_max_sysctl, "IU",
113 "Longest throttling duration, in hz");
114 TUNABLE_INT("kern.racct.rctl.throttle_max", &rctl_throttle_max);
115 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_pct,
116 CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, 0,
117 &rctl_throttle_pct_sysctl, "IU",
118 "Throttling penalty for process consumption, in percent");
119 TUNABLE_INT("kern.racct.rctl.throttle_pct", &rctl_throttle_pct);
120 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_pct2,
121 CTLTYPE_UINT | CTLFLAG_RWTUN | CTLFLAG_MPSAFE, 0, 0,
122 &rctl_throttle_pct2_sysctl, "IU",
123 "Throttling penalty for container consumption, in percent");
124 TUNABLE_INT("kern.racct.rctl.throttle_pct2", &rctl_throttle_pct2);
127 * 'rctl_rule_link' connects a rule with every racct it's related to.
128 * For example, rule 'user:X:openfiles:deny=N/process' is linked
129 * with uidinfo for user X, and to each process of that user.
131 struct rctl_rule_link {
132 LIST_ENTRY(rctl_rule_link) rrl_next;
133 struct rctl_rule *rrl_rule;
142 static struct dict subjectnames[] = {
143 { "process", RCTL_SUBJECT_TYPE_PROCESS },
144 { "user", RCTL_SUBJECT_TYPE_USER },
145 { "loginclass", RCTL_SUBJECT_TYPE_LOGINCLASS },
146 { "jail", RCTL_SUBJECT_TYPE_JAIL },
149 static struct dict resourcenames[] = {
150 { "cputime", RACCT_CPU },
151 { "datasize", RACCT_DATA },
152 { "stacksize", RACCT_STACK },
153 { "coredumpsize", RACCT_CORE },
154 { "memoryuse", RACCT_RSS },
155 { "memorylocked", RACCT_MEMLOCK },
156 { "maxproc", RACCT_NPROC },
157 { "openfiles", RACCT_NOFILE },
158 { "vmemoryuse", RACCT_VMEM },
159 { "pseudoterminals", RACCT_NPTS },
160 { "swapuse", RACCT_SWAP },
161 { "nthr", RACCT_NTHR },
162 { "msgqqueued", RACCT_MSGQQUEUED },
163 { "msgqsize", RACCT_MSGQSIZE },
164 { "nmsgq", RACCT_NMSGQ },
165 { "nsem", RACCT_NSEM },
166 { "nsemop", RACCT_NSEMOP },
167 { "nshm", RACCT_NSHM },
168 { "shmsize", RACCT_SHMSIZE },
169 { "wallclock", RACCT_WALLCLOCK },
170 { "pcpu", RACCT_PCTCPU },
171 { "readbps", RACCT_READBPS },
172 { "writebps", RACCT_WRITEBPS },
173 { "readiops", RACCT_READIOPS },
174 { "writeiops", RACCT_WRITEIOPS },
177 static struct dict actionnames[] = {
178 { "sighup", RCTL_ACTION_SIGHUP },
179 { "sigint", RCTL_ACTION_SIGINT },
180 { "sigquit", RCTL_ACTION_SIGQUIT },
181 { "sigill", RCTL_ACTION_SIGILL },
182 { "sigtrap", RCTL_ACTION_SIGTRAP },
183 { "sigabrt", RCTL_ACTION_SIGABRT },
184 { "sigemt", RCTL_ACTION_SIGEMT },
185 { "sigfpe", RCTL_ACTION_SIGFPE },
186 { "sigkill", RCTL_ACTION_SIGKILL },
187 { "sigbus", RCTL_ACTION_SIGBUS },
188 { "sigsegv", RCTL_ACTION_SIGSEGV },
189 { "sigsys", RCTL_ACTION_SIGSYS },
190 { "sigpipe", RCTL_ACTION_SIGPIPE },
191 { "sigalrm", RCTL_ACTION_SIGALRM },
192 { "sigterm", RCTL_ACTION_SIGTERM },
193 { "sigurg", RCTL_ACTION_SIGURG },
194 { "sigstop", RCTL_ACTION_SIGSTOP },
195 { "sigtstp", RCTL_ACTION_SIGTSTP },
196 { "sigchld", RCTL_ACTION_SIGCHLD },
197 { "sigttin", RCTL_ACTION_SIGTTIN },
198 { "sigttou", RCTL_ACTION_SIGTTOU },
199 { "sigio", RCTL_ACTION_SIGIO },
200 { "sigxcpu", RCTL_ACTION_SIGXCPU },
201 { "sigxfsz", RCTL_ACTION_SIGXFSZ },
202 { "sigvtalrm", RCTL_ACTION_SIGVTALRM },
203 { "sigprof", RCTL_ACTION_SIGPROF },
204 { "sigwinch", RCTL_ACTION_SIGWINCH },
205 { "siginfo", RCTL_ACTION_SIGINFO },
206 { "sigusr1", RCTL_ACTION_SIGUSR1 },
207 { "sigusr2", RCTL_ACTION_SIGUSR2 },
208 { "sigthr", RCTL_ACTION_SIGTHR },
209 { "deny", RCTL_ACTION_DENY },
210 { "log", RCTL_ACTION_LOG },
211 { "devctl", RCTL_ACTION_DEVCTL },
212 { "throttle", RCTL_ACTION_THROTTLE },
215 static void rctl_init(void);
216 SYSINIT(rctl, SI_SUB_RACCT, SI_ORDER_FIRST, rctl_init, NULL);
218 static uma_zone_t rctl_rule_zone;
219 static uma_zone_t rctl_rule_link_zone;
221 static int rctl_rule_fully_specified(const struct rctl_rule *rule);
222 static void rctl_rule_to_sbuf(struct sbuf *sb, const struct rctl_rule *rule);
224 static MALLOC_DEFINE(M_RCTL, "rctl", "Resource Limits");
226 static int rctl_throttle_min_sysctl(SYSCTL_HANDLER_ARGS)
228 int error, val = rctl_throttle_min;
230 error = sysctl_handle_int(oidp, &val, 0, req);
231 if (error || !req->newptr)
233 if (val < 1 || val > rctl_throttle_max)
237 rctl_throttle_min = val;
243 static int rctl_throttle_max_sysctl(SYSCTL_HANDLER_ARGS)
245 int error, val = rctl_throttle_max;
247 error = sysctl_handle_int(oidp, &val, 0, req);
248 if (error || !req->newptr)
250 if (val < rctl_throttle_min)
254 rctl_throttle_max = val;
260 static int rctl_throttle_pct_sysctl(SYSCTL_HANDLER_ARGS)
262 int error, val = rctl_throttle_pct;
264 error = sysctl_handle_int(oidp, &val, 0, req);
265 if (error || !req->newptr)
271 rctl_throttle_pct = val;
277 static int rctl_throttle_pct2_sysctl(SYSCTL_HANDLER_ARGS)
279 int error, val = rctl_throttle_pct2;
281 error = sysctl_handle_int(oidp, &val, 0, req);
282 if (error || !req->newptr)
288 rctl_throttle_pct2 = val;
295 rctl_subject_type_name(int subject)
299 for (i = 0; subjectnames[i].d_name != NULL; i++) {
300 if (subjectnames[i].d_value == subject)
301 return (subjectnames[i].d_name);
304 panic("rctl_subject_type_name: unknown subject type %d", subject);
308 rctl_action_name(int action)
312 for (i = 0; actionnames[i].d_name != NULL; i++) {
313 if (actionnames[i].d_value == action)
314 return (actionnames[i].d_name);
317 panic("rctl_action_name: unknown action %d", action);
321 rctl_resource_name(int resource)
325 for (i = 0; resourcenames[i].d_name != NULL; i++) {
326 if (resourcenames[i].d_value == resource)
327 return (resourcenames[i].d_name);
330 panic("rctl_resource_name: unknown resource %d", resource);
333 static struct racct *
334 rctl_proc_rule_to_racct(const struct proc *p, const struct rctl_rule *rule)
336 struct ucred *cred = p->p_ucred;
338 ASSERT_RACCT_ENABLED();
341 switch (rule->rr_per) {
342 case RCTL_SUBJECT_TYPE_PROCESS:
344 case RCTL_SUBJECT_TYPE_USER:
345 return (cred->cr_ruidinfo->ui_racct);
346 case RCTL_SUBJECT_TYPE_LOGINCLASS:
347 return (cred->cr_loginclass->lc_racct);
348 case RCTL_SUBJECT_TYPE_JAIL:
349 return (cred->cr_prison->pr_prison_racct->prr_racct);
351 panic("%s: unknown per %d", __func__, rule->rr_per);
356 * Return the amount of resource that can be allocated by 'p' before
360 rctl_available_resource(const struct proc *p, const struct rctl_rule *rule)
362 const struct racct *racct;
365 ASSERT_RACCT_ENABLED();
368 racct = rctl_proc_rule_to_racct(p, rule);
369 available = rule->rr_amount - racct->r_resources[rule->rr_resource];
375 * Called every second for proc, uidinfo, loginclass, and jail containers.
376 * If the limit isn't exceeded, it decreases the usage amount to zero.
377 * Otherwise, it decreases it by the value of the limit. This way
378 * resource consumption exceeding the limit "carries over" to the next
382 rctl_throttle_decay(struct racct *racct, int resource)
384 struct rctl_rule *rule;
385 struct rctl_rule_link *link;
386 int64_t minavailable;
388 ASSERT_RACCT_ENABLED();
391 minavailable = INT64_MAX;
393 LIST_FOREACH(link, &racct->r_rule_links, rrl_next) {
394 rule = link->rrl_rule;
396 if (rule->rr_resource != resource)
398 if (rule->rr_action != RCTL_ACTION_THROTTLE)
401 if (rule->rr_amount < minavailable)
402 minavailable = rule->rr_amount;
405 if (racct->r_resources[resource] < minavailable) {
406 racct->r_resources[resource] = 0;
409 * Cap utilization counter at ten times the limit. Otherwise,
410 * if we changed the rule lowering the allowed amount, it could
411 * take unreasonably long time for the accumulated resource
414 if (racct->r_resources[resource] > minavailable * 10)
415 racct->r_resources[resource] = minavailable * 10;
417 racct->r_resources[resource] -= minavailable;
422 * Special version of rctl_get_available() for the %CPU resource.
423 * We slightly cheat here and return less than we normally would.
426 rctl_pcpu_available(const struct proc *p) {
427 struct rctl_rule *rule;
428 struct rctl_rule_link *link;
429 int64_t available, minavailable, limit;
431 ASSERT_RACCT_ENABLED();
434 minavailable = INT64_MAX;
437 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
438 rule = link->rrl_rule;
439 if (rule->rr_resource != RACCT_PCTCPU)
441 if (rule->rr_action != RCTL_ACTION_DENY)
443 available = rctl_available_resource(p, rule);
444 if (available < minavailable) {
445 minavailable = available;
446 limit = rule->rr_amount;
451 * Return slightly less than actual value of the available
452 * %cpu resource. This makes %cpu throttling more aggressive
453 * and lets us act sooner than the limits are already exceeded.
456 if (limit > 2 * RCTL_PCPU_SHIFT)
457 minavailable -= RCTL_PCPU_SHIFT;
459 minavailable -= (limit / 2);
462 return (minavailable);
466 xadd(uint64_t a, uint64_t b)
482 xmul(uint64_t a, uint64_t b)
485 if (b != 0 && a > UINT64_MAX / b)
492 * Check whether the proc 'p' can allocate 'amount' of 'resource' in addition
493 * to what it keeps allocated now. Returns non-zero if the allocation should
494 * be denied, 0 otherwise.
497 rctl_enforce(struct proc *p, int resource, uint64_t amount)
499 static struct timeval log_lasttime, devctl_lasttime;
500 static int log_curtime = 0, devctl_curtime = 0;
501 struct rctl_rule *rule;
502 struct rctl_rule_link *link;
506 uint64_t sleep_ms, sleep_ratio;
509 ASSERT_RACCT_ENABLED();
513 * There may be more than one matching rule; go through all of them.
514 * Denial should be done last, after logging and sending signals.
516 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
517 rule = link->rrl_rule;
518 if (rule->rr_resource != resource)
521 available = rctl_available_resource(p, rule);
522 if (available >= (int64_t)amount) {
523 link->rrl_exceeded = 0;
527 switch (rule->rr_action) {
528 case RCTL_ACTION_DENY:
531 case RCTL_ACTION_LOG:
533 * If rrl_exceeded != 0, it means we've already
534 * logged a warning for this process.
536 if (link->rrl_exceeded != 0)
540 * If the process state is not fully initialized yet,
541 * we can't access most of the required fields, e.g.
542 * p->p_comm. This happens when called from fork1().
543 * Ignore this rule for now; it will be processed just
544 * after fork, when called from racct_proc_fork_done().
546 if (p->p_state != PRS_NORMAL)
549 if (!ppsratecheck(&log_lasttime, &log_curtime,
550 rctl_log_rate_limit))
553 buf = malloc(RCTL_LOG_BUFSIZE, M_RCTL, M_NOWAIT);
555 printf("rctl_enforce: out of memory\n");
558 sbuf_new(&sb, buf, RCTL_LOG_BUFSIZE, SBUF_FIXEDLEN);
559 rctl_rule_to_sbuf(&sb, rule);
561 printf("rctl: rule \"%s\" matched by pid %d "
562 "(%s), uid %d, jail %s\n", sbuf_data(&sb),
563 p->p_pid, p->p_comm, p->p_ucred->cr_uid,
564 p->p_ucred->cr_prison->pr_prison_racct->prr_name);
567 link->rrl_exceeded = 1;
569 case RCTL_ACTION_DEVCTL:
570 if (link->rrl_exceeded != 0)
573 if (p->p_state != PRS_NORMAL)
576 if (!ppsratecheck(&devctl_lasttime, &devctl_curtime,
577 rctl_devctl_rate_limit))
580 buf = malloc(RCTL_LOG_BUFSIZE, M_RCTL, M_NOWAIT);
582 printf("rctl_enforce: out of memory\n");
585 sbuf_new(&sb, buf, RCTL_LOG_BUFSIZE, SBUF_FIXEDLEN);
586 sbuf_printf(&sb, "rule=");
587 rctl_rule_to_sbuf(&sb, rule);
588 sbuf_printf(&sb, " pid=%d ruid=%d jail=%s",
589 p->p_pid, p->p_ucred->cr_ruid,
590 p->p_ucred->cr_prison->pr_prison_racct->prr_name);
592 devctl_notify("RCTL", "rule", "matched",
596 link->rrl_exceeded = 1;
598 case RCTL_ACTION_THROTTLE:
599 if (p->p_state != PRS_NORMAL)
602 if (rule->rr_amount == 0) {
603 racct_proc_throttle(p, rctl_throttle_max);
608 * Make the process sleep for a fraction of second
609 * proportional to the ratio of process' resource
610 * utilization compared to the limit. The point is
611 * to penalize resource hogs: processes that consume
612 * more of the available resources sleep for longer.
614 * We're trying to defer division until the very end,
615 * to minimize the rounding effects. The following
616 * calculation could have been written in a clearer
619 * sleep_ms = hz * p->p_racct->r_resources[resource] /
621 * sleep_ms *= rctl_throttle_pct / 100;
622 * if (sleep_ms < rctl_throttle_min)
623 * sleep_ms = rctl_throttle_min;
626 sleep_ms = xmul(hz, p->p_racct->r_resources[resource]);
627 sleep_ms = xmul(sleep_ms, rctl_throttle_pct) / 100;
628 if (sleep_ms < rctl_throttle_min * rule->rr_amount)
629 sleep_ms = rctl_throttle_min * rule->rr_amount;
632 * Multiply that by the ratio of the resource
633 * consumption for the container compared to the limit,
634 * squared. In other words, a process in a container
635 * that is two times over the limit will be throttled
636 * four times as much for hitting the same rule. The
637 * point is to penalize processes more if the container
638 * itself (eg certain UID or jail) is above the limit.
641 sleep_ratio = -available / rule->rr_amount;
644 sleep_ratio = xmul(sleep_ratio, sleep_ratio);
645 sleep_ratio = xmul(sleep_ratio, rctl_throttle_pct2) / 100;
646 sleep_ms = xadd(sleep_ms, xmul(sleep_ms, sleep_ratio));
649 * Finally the division.
651 sleep_ms /= rule->rr_amount;
653 if (sleep_ms > rctl_throttle_max)
654 sleep_ms = rctl_throttle_max;
656 printf("%s: pid %d (%s), %jd of %jd, will sleep for %ju ms (ratio %ju, available %jd)\n",
657 __func__, p->p_pid, p->p_comm,
658 p->p_racct->r_resources[resource],
659 rule->rr_amount, (uintmax_t)sleep_ms,
660 (uintmax_t)sleep_ratio, (intmax_t)available);
663 KASSERT(sleep_ms >= rctl_throttle_min, ("%s: %ju < %d\n",
664 __func__, (uintmax_t)sleep_ms, rctl_throttle_min));
665 racct_proc_throttle(p, sleep_ms);
668 if (link->rrl_exceeded != 0)
671 if (p->p_state != PRS_NORMAL)
674 KASSERT(rule->rr_action > 0 &&
675 rule->rr_action <= RCTL_ACTION_SIGNAL_MAX,
676 ("rctl_enforce: unknown action %d",
680 * We're using the fact that RCTL_ACTION_SIG* values
681 * are equal to their counterparts from sys/signal.h.
683 kern_psignal(p, rule->rr_action);
684 link->rrl_exceeded = 1;
691 * Return fake error code; the caller should change it
692 * into one proper for the situation - EFSIZ, ENOMEM etc.
701 rctl_get_limit(struct proc *p, int resource)
703 struct rctl_rule *rule;
704 struct rctl_rule_link *link;
705 uint64_t amount = UINT64_MAX;
707 ASSERT_RACCT_ENABLED();
711 * There may be more than one matching rule; go through all of them.
712 * Denial should be done last, after logging and sending signals.
714 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
715 rule = link->rrl_rule;
716 if (rule->rr_resource != resource)
718 if (rule->rr_action != RCTL_ACTION_DENY)
720 if (rule->rr_amount < amount)
721 amount = rule->rr_amount;
728 rctl_get_available(struct proc *p, int resource)
730 struct rctl_rule *rule;
731 struct rctl_rule_link *link;
732 int64_t available, minavailable, allocated;
734 minavailable = INT64_MAX;
736 ASSERT_RACCT_ENABLED();
740 * There may be more than one matching rule; go through all of them.
741 * Denial should be done last, after logging and sending signals.
743 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
744 rule = link->rrl_rule;
745 if (rule->rr_resource != resource)
747 if (rule->rr_action != RCTL_ACTION_DENY)
749 available = rctl_available_resource(p, rule);
750 if (available < minavailable)
751 minavailable = available;
755 * XXX: Think about this _hard_.
757 allocated = p->p_racct->r_resources[resource];
758 if (minavailable < INT64_MAX - allocated)
759 minavailable += allocated;
760 if (minavailable < 0)
763 return (minavailable);
767 rctl_rule_matches(const struct rctl_rule *rule, const struct rctl_rule *filter)
770 ASSERT_RACCT_ENABLED();
772 if (filter->rr_subject_type != RCTL_SUBJECT_TYPE_UNDEFINED) {
773 if (rule->rr_subject_type != filter->rr_subject_type)
776 switch (filter->rr_subject_type) {
777 case RCTL_SUBJECT_TYPE_PROCESS:
778 if (filter->rr_subject.rs_proc != NULL &&
779 rule->rr_subject.rs_proc !=
780 filter->rr_subject.rs_proc)
783 case RCTL_SUBJECT_TYPE_USER:
784 if (filter->rr_subject.rs_uip != NULL &&
785 rule->rr_subject.rs_uip !=
786 filter->rr_subject.rs_uip)
789 case RCTL_SUBJECT_TYPE_LOGINCLASS:
790 if (filter->rr_subject.rs_loginclass != NULL &&
791 rule->rr_subject.rs_loginclass !=
792 filter->rr_subject.rs_loginclass)
795 case RCTL_SUBJECT_TYPE_JAIL:
796 if (filter->rr_subject.rs_prison_racct != NULL &&
797 rule->rr_subject.rs_prison_racct !=
798 filter->rr_subject.rs_prison_racct)
802 panic("rctl_rule_matches: unknown subject type %d",
803 filter->rr_subject_type);
807 if (filter->rr_resource != RACCT_UNDEFINED) {
808 if (rule->rr_resource != filter->rr_resource)
812 if (filter->rr_action != RCTL_ACTION_UNDEFINED) {
813 if (rule->rr_action != filter->rr_action)
817 if (filter->rr_amount != RCTL_AMOUNT_UNDEFINED) {
818 if (rule->rr_amount != filter->rr_amount)
822 if (filter->rr_per != RCTL_SUBJECT_TYPE_UNDEFINED) {
823 if (rule->rr_per != filter->rr_per)
831 str2value(const char *str, int *value, struct dict *table)
838 for (i = 0; table[i].d_name != NULL; i++) {
839 if (strcasecmp(table[i].d_name, str) == 0) {
840 *value = table[i].d_value;
849 str2id(const char *str, id_t *value)
856 *value = strtoul(str, &end, 10);
857 if ((size_t)(end - str) != strlen(str))
864 str2int64(const char *str, int64_t *value)
871 *value = strtoul(str, &end, 10);
872 if ((size_t)(end - str) != strlen(str))
882 * Connect the rule to the racct, increasing refcount for the rule.
885 rctl_racct_add_rule(struct racct *racct, struct rctl_rule *rule)
887 struct rctl_rule_link *link;
889 ASSERT_RACCT_ENABLED();
890 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified"));
892 rctl_rule_acquire(rule);
893 link = uma_zalloc(rctl_rule_link_zone, M_WAITOK);
894 link->rrl_rule = rule;
895 link->rrl_exceeded = 0;
898 LIST_INSERT_HEAD(&racct->r_rule_links, link, rrl_next);
903 rctl_racct_add_rule_locked(struct racct *racct, struct rctl_rule *rule)
905 struct rctl_rule_link *link;
907 ASSERT_RACCT_ENABLED();
908 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified"));
911 link = uma_zalloc(rctl_rule_link_zone, M_NOWAIT);
914 rctl_rule_acquire(rule);
915 link->rrl_rule = rule;
916 link->rrl_exceeded = 0;
918 LIST_INSERT_HEAD(&racct->r_rule_links, link, rrl_next);
924 * Remove limits for a rules matching the filter and release
925 * the refcounts for the rules, possibly freeing them. Returns
926 * the number of limit structures removed.
929 rctl_racct_remove_rules(struct racct *racct,
930 const struct rctl_rule *filter)
932 struct rctl_rule_link *link, *linktmp;
935 ASSERT_RACCT_ENABLED();
938 LIST_FOREACH_SAFE(link, &racct->r_rule_links, rrl_next, linktmp) {
939 if (!rctl_rule_matches(link->rrl_rule, filter))
942 LIST_REMOVE(link, rrl_next);
943 rctl_rule_release(link->rrl_rule);
944 uma_zfree(rctl_rule_link_zone, link);
951 rctl_rule_acquire_subject(struct rctl_rule *rule)
954 ASSERT_RACCT_ENABLED();
956 switch (rule->rr_subject_type) {
957 case RCTL_SUBJECT_TYPE_UNDEFINED:
958 case RCTL_SUBJECT_TYPE_PROCESS:
960 case RCTL_SUBJECT_TYPE_JAIL:
961 if (rule->rr_subject.rs_prison_racct != NULL)
962 prison_racct_hold(rule->rr_subject.rs_prison_racct);
964 case RCTL_SUBJECT_TYPE_USER:
965 if (rule->rr_subject.rs_uip != NULL)
966 uihold(rule->rr_subject.rs_uip);
968 case RCTL_SUBJECT_TYPE_LOGINCLASS:
969 if (rule->rr_subject.rs_loginclass != NULL)
970 loginclass_hold(rule->rr_subject.rs_loginclass);
973 panic("rctl_rule_acquire_subject: unknown subject type %d",
974 rule->rr_subject_type);
979 rctl_rule_release_subject(struct rctl_rule *rule)
982 ASSERT_RACCT_ENABLED();
984 switch (rule->rr_subject_type) {
985 case RCTL_SUBJECT_TYPE_UNDEFINED:
986 case RCTL_SUBJECT_TYPE_PROCESS:
988 case RCTL_SUBJECT_TYPE_JAIL:
989 if (rule->rr_subject.rs_prison_racct != NULL)
990 prison_racct_free(rule->rr_subject.rs_prison_racct);
992 case RCTL_SUBJECT_TYPE_USER:
993 if (rule->rr_subject.rs_uip != NULL)
994 uifree(rule->rr_subject.rs_uip);
996 case RCTL_SUBJECT_TYPE_LOGINCLASS:
997 if (rule->rr_subject.rs_loginclass != NULL)
998 loginclass_free(rule->rr_subject.rs_loginclass);
1001 panic("rctl_rule_release_subject: unknown subject type %d",
1002 rule->rr_subject_type);
1007 rctl_rule_alloc(int flags)
1009 struct rctl_rule *rule;
1011 ASSERT_RACCT_ENABLED();
1013 rule = uma_zalloc(rctl_rule_zone, flags);
1016 rule->rr_subject_type = RCTL_SUBJECT_TYPE_UNDEFINED;
1017 rule->rr_subject.rs_proc = NULL;
1018 rule->rr_subject.rs_uip = NULL;
1019 rule->rr_subject.rs_loginclass = NULL;
1020 rule->rr_subject.rs_prison_racct = NULL;
1021 rule->rr_per = RCTL_SUBJECT_TYPE_UNDEFINED;
1022 rule->rr_resource = RACCT_UNDEFINED;
1023 rule->rr_action = RCTL_ACTION_UNDEFINED;
1024 rule->rr_amount = RCTL_AMOUNT_UNDEFINED;
1025 refcount_init(&rule->rr_refcount, 1);
1031 rctl_rule_duplicate(const struct rctl_rule *rule, int flags)
1033 struct rctl_rule *copy;
1035 ASSERT_RACCT_ENABLED();
1037 copy = uma_zalloc(rctl_rule_zone, flags);
1040 copy->rr_subject_type = rule->rr_subject_type;
1041 copy->rr_subject.rs_proc = rule->rr_subject.rs_proc;
1042 copy->rr_subject.rs_uip = rule->rr_subject.rs_uip;
1043 copy->rr_subject.rs_loginclass = rule->rr_subject.rs_loginclass;
1044 copy->rr_subject.rs_prison_racct = rule->rr_subject.rs_prison_racct;
1045 copy->rr_per = rule->rr_per;
1046 copy->rr_resource = rule->rr_resource;
1047 copy->rr_action = rule->rr_action;
1048 copy->rr_amount = rule->rr_amount;
1049 refcount_init(©->rr_refcount, 1);
1050 rctl_rule_acquire_subject(copy);
1056 rctl_rule_acquire(struct rctl_rule *rule)
1059 ASSERT_RACCT_ENABLED();
1060 KASSERT(rule->rr_refcount > 0, ("rule->rr_refcount <= 0"));
1062 refcount_acquire(&rule->rr_refcount);
1066 rctl_rule_free(void *context, int pending)
1068 struct rctl_rule *rule;
1070 rule = (struct rctl_rule *)context;
1072 ASSERT_RACCT_ENABLED();
1073 KASSERT(rule->rr_refcount == 0, ("rule->rr_refcount != 0"));
1076 * We don't need locking here; rule is guaranteed to be inaccessible.
1079 rctl_rule_release_subject(rule);
1080 uma_zfree(rctl_rule_zone, rule);
1084 rctl_rule_release(struct rctl_rule *rule)
1087 ASSERT_RACCT_ENABLED();
1088 KASSERT(rule->rr_refcount > 0, ("rule->rr_refcount <= 0"));
1090 if (refcount_release(&rule->rr_refcount)) {
1092 * rctl_rule_release() is often called when iterating
1093 * over all the uidinfo structures in the system,
1094 * holding uihashtbl_lock. Since rctl_rule_free()
1095 * might end up calling uifree(), this would lead
1096 * to lock recursion. Use taskqueue to avoid this.
1098 TASK_INIT(&rule->rr_task, 0, rctl_rule_free, rule);
1099 taskqueue_enqueue(taskqueue_thread, &rule->rr_task);
1104 rctl_rule_fully_specified(const struct rctl_rule *rule)
1107 ASSERT_RACCT_ENABLED();
1109 switch (rule->rr_subject_type) {
1110 case RCTL_SUBJECT_TYPE_UNDEFINED:
1112 case RCTL_SUBJECT_TYPE_PROCESS:
1113 if (rule->rr_subject.rs_proc == NULL)
1116 case RCTL_SUBJECT_TYPE_USER:
1117 if (rule->rr_subject.rs_uip == NULL)
1120 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1121 if (rule->rr_subject.rs_loginclass == NULL)
1124 case RCTL_SUBJECT_TYPE_JAIL:
1125 if (rule->rr_subject.rs_prison_racct == NULL)
1129 panic("rctl_rule_fully_specified: unknown subject type %d",
1130 rule->rr_subject_type);
1132 if (rule->rr_resource == RACCT_UNDEFINED)
1134 if (rule->rr_action == RCTL_ACTION_UNDEFINED)
1136 if (rule->rr_amount == RCTL_AMOUNT_UNDEFINED)
1138 if (rule->rr_per == RCTL_SUBJECT_TYPE_UNDEFINED)
1145 rctl_string_to_rule(char *rulestr, struct rctl_rule **rulep)
1147 struct rctl_rule *rule;
1148 char *subjectstr, *subject_idstr, *resourcestr, *actionstr,
1149 *amountstr, *perstr;
1153 ASSERT_RACCT_ENABLED();
1155 rule = rctl_rule_alloc(M_WAITOK);
1157 subjectstr = strsep(&rulestr, ":");
1158 subject_idstr = strsep(&rulestr, ":");
1159 resourcestr = strsep(&rulestr, ":");
1160 actionstr = strsep(&rulestr, "=/");
1161 amountstr = strsep(&rulestr, "/");
1164 if (subjectstr == NULL || subjectstr[0] == '\0')
1165 rule->rr_subject_type = RCTL_SUBJECT_TYPE_UNDEFINED;
1167 error = str2value(subjectstr, &rule->rr_subject_type, subjectnames);
1172 if (subject_idstr == NULL || subject_idstr[0] == '\0') {
1173 rule->rr_subject.rs_proc = NULL;
1174 rule->rr_subject.rs_uip = NULL;
1175 rule->rr_subject.rs_loginclass = NULL;
1176 rule->rr_subject.rs_prison_racct = NULL;
1178 switch (rule->rr_subject_type) {
1179 case RCTL_SUBJECT_TYPE_UNDEFINED:
1182 case RCTL_SUBJECT_TYPE_PROCESS:
1183 error = str2id(subject_idstr, &id);
1186 sx_assert(&allproc_lock, SA_LOCKED);
1187 rule->rr_subject.rs_proc = pfind(id);
1188 if (rule->rr_subject.rs_proc == NULL) {
1192 PROC_UNLOCK(rule->rr_subject.rs_proc);
1194 case RCTL_SUBJECT_TYPE_USER:
1195 error = str2id(subject_idstr, &id);
1198 rule->rr_subject.rs_uip = uifind(id);
1200 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1201 rule->rr_subject.rs_loginclass =
1202 loginclass_find(subject_idstr);
1203 if (rule->rr_subject.rs_loginclass == NULL) {
1204 error = ENAMETOOLONG;
1208 case RCTL_SUBJECT_TYPE_JAIL:
1209 rule->rr_subject.rs_prison_racct =
1210 prison_racct_find(subject_idstr);
1211 if (rule->rr_subject.rs_prison_racct == NULL) {
1212 error = ENAMETOOLONG;
1217 panic("rctl_string_to_rule: unknown subject type %d",
1218 rule->rr_subject_type);
1222 if (resourcestr == NULL || resourcestr[0] == '\0')
1223 rule->rr_resource = RACCT_UNDEFINED;
1225 error = str2value(resourcestr, &rule->rr_resource,
1231 if (actionstr == NULL || actionstr[0] == '\0')
1232 rule->rr_action = RCTL_ACTION_UNDEFINED;
1234 error = str2value(actionstr, &rule->rr_action, actionnames);
1239 if (amountstr == NULL || amountstr[0] == '\0')
1240 rule->rr_amount = RCTL_AMOUNT_UNDEFINED;
1242 error = str2int64(amountstr, &rule->rr_amount);
1245 if (RACCT_IS_IN_MILLIONS(rule->rr_resource)) {
1246 if (rule->rr_amount > INT64_MAX / 1000000) {
1250 rule->rr_amount *= 1000000;
1254 if (perstr == NULL || perstr[0] == '\0')
1255 rule->rr_per = RCTL_SUBJECT_TYPE_UNDEFINED;
1257 error = str2value(perstr, &rule->rr_per, subjectnames);
1266 rctl_rule_release(rule);
1272 * Link a rule with all the subjects it applies to.
1275 rctl_rule_add(struct rctl_rule *rule)
1279 struct uidinfo *uip;
1281 struct prison_racct *prr;
1282 struct loginclass *lc;
1283 struct rctl_rule *rule2;
1286 ASSERT_RACCT_ENABLED();
1287 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified"));
1290 * Some rules just don't make sense, like "deny" rule for an undeniable
1291 * resource. The exception are the RSS and %CPU resources - they are
1292 * not deniable in the racct sense, but the limit is enforced in
1295 if (rule->rr_action == RCTL_ACTION_DENY &&
1296 !RACCT_IS_DENIABLE(rule->rr_resource) &&
1297 rule->rr_resource != RACCT_RSS &&
1298 rule->rr_resource != RACCT_PCTCPU) {
1299 return (EOPNOTSUPP);
1302 if (rule->rr_action == RCTL_ACTION_THROTTLE &&
1303 !RACCT_IS_DECAYING(rule->rr_resource)) {
1304 return (EOPNOTSUPP);
1307 if (rule->rr_action == RCTL_ACTION_THROTTLE &&
1308 rule->rr_resource == RACCT_PCTCPU) {
1309 return (EOPNOTSUPP);
1312 if (rule->rr_per == RCTL_SUBJECT_TYPE_PROCESS &&
1313 RACCT_IS_SLOPPY(rule->rr_resource)) {
1314 return (EOPNOTSUPP);
1318 * Make sure there are no duplicated rules. Also, for the "deny"
1319 * rules, remove ones differing only by "amount".
1321 if (rule->rr_action == RCTL_ACTION_DENY) {
1322 rule2 = rctl_rule_duplicate(rule, M_WAITOK);
1323 rule2->rr_amount = RCTL_AMOUNT_UNDEFINED;
1324 rctl_rule_remove(rule2);
1325 rctl_rule_release(rule2);
1327 rctl_rule_remove(rule);
1329 switch (rule->rr_subject_type) {
1330 case RCTL_SUBJECT_TYPE_PROCESS:
1331 p = rule->rr_subject.rs_proc;
1332 KASSERT(p != NULL, ("rctl_rule_add: NULL proc"));
1334 rctl_racct_add_rule(p->p_racct, rule);
1336 * In case of per-process rule, we don't have anything more
1341 case RCTL_SUBJECT_TYPE_USER:
1342 uip = rule->rr_subject.rs_uip;
1343 KASSERT(uip != NULL, ("rctl_rule_add: NULL uip"));
1344 rctl_racct_add_rule(uip->ui_racct, rule);
1347 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1348 lc = rule->rr_subject.rs_loginclass;
1349 KASSERT(lc != NULL, ("rctl_rule_add: NULL loginclass"));
1350 rctl_racct_add_rule(lc->lc_racct, rule);
1353 case RCTL_SUBJECT_TYPE_JAIL:
1354 prr = rule->rr_subject.rs_prison_racct;
1355 KASSERT(prr != NULL, ("rctl_rule_add: NULL pr"));
1356 rctl_racct_add_rule(prr->prr_racct, rule);
1360 panic("rctl_rule_add: unknown subject type %d",
1361 rule->rr_subject_type);
1365 * Now go through all the processes and add the new rule to the ones
1368 sx_assert(&allproc_lock, SA_LOCKED);
1369 FOREACH_PROC_IN_SYSTEM(p) {
1371 switch (rule->rr_subject_type) {
1372 case RCTL_SUBJECT_TYPE_USER:
1373 if (cred->cr_uidinfo == rule->rr_subject.rs_uip ||
1374 cred->cr_ruidinfo == rule->rr_subject.rs_uip)
1377 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1378 if (cred->cr_loginclass == rule->rr_subject.rs_loginclass)
1381 case RCTL_SUBJECT_TYPE_JAIL:
1383 for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) {
1384 if (pr->pr_prison_racct == rule->rr_subject.rs_prison_racct) {
1393 panic("rctl_rule_add: unknown subject type %d",
1394 rule->rr_subject_type);
1397 rctl_racct_add_rule(p->p_racct, rule);
1404 rctl_rule_pre_callback(void)
1411 rctl_rule_post_callback(void)
1418 rctl_rule_remove_callback(struct racct *racct, void *arg2, void *arg3)
1420 struct rctl_rule *filter = (struct rctl_rule *)arg2;
1423 ASSERT_RACCT_ENABLED();
1424 RACCT_LOCK_ASSERT();
1426 found += rctl_racct_remove_rules(racct, filter);
1428 *((int *)arg3) += found;
1432 * Remove all rules that match the filter.
1435 rctl_rule_remove(struct rctl_rule *filter)
1440 ASSERT_RACCT_ENABLED();
1442 if (filter->rr_subject_type == RCTL_SUBJECT_TYPE_PROCESS &&
1443 filter->rr_subject.rs_proc != NULL) {
1444 p = filter->rr_subject.rs_proc;
1446 found = rctl_racct_remove_rules(p->p_racct, filter);
1453 loginclass_racct_foreach(rctl_rule_remove_callback,
1454 rctl_rule_pre_callback, rctl_rule_post_callback,
1455 filter, (void *)&found);
1456 ui_racct_foreach(rctl_rule_remove_callback,
1457 rctl_rule_pre_callback, rctl_rule_post_callback,
1458 filter, (void *)&found);
1459 prison_racct_foreach(rctl_rule_remove_callback,
1460 rctl_rule_pre_callback, rctl_rule_post_callback,
1461 filter, (void *)&found);
1463 sx_assert(&allproc_lock, SA_LOCKED);
1465 FOREACH_PROC_IN_SYSTEM(p) {
1466 found += rctl_racct_remove_rules(p->p_racct, filter);
1476 * Appends a rule to the sbuf.
1479 rctl_rule_to_sbuf(struct sbuf *sb, const struct rctl_rule *rule)
1483 ASSERT_RACCT_ENABLED();
1485 sbuf_printf(sb, "%s:", rctl_subject_type_name(rule->rr_subject_type));
1487 switch (rule->rr_subject_type) {
1488 case RCTL_SUBJECT_TYPE_PROCESS:
1489 if (rule->rr_subject.rs_proc == NULL)
1490 sbuf_printf(sb, ":");
1492 sbuf_printf(sb, "%d:",
1493 rule->rr_subject.rs_proc->p_pid);
1495 case RCTL_SUBJECT_TYPE_USER:
1496 if (rule->rr_subject.rs_uip == NULL)
1497 sbuf_printf(sb, ":");
1499 sbuf_printf(sb, "%d:",
1500 rule->rr_subject.rs_uip->ui_uid);
1502 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1503 if (rule->rr_subject.rs_loginclass == NULL)
1504 sbuf_printf(sb, ":");
1506 sbuf_printf(sb, "%s:",
1507 rule->rr_subject.rs_loginclass->lc_name);
1509 case RCTL_SUBJECT_TYPE_JAIL:
1510 if (rule->rr_subject.rs_prison_racct == NULL)
1511 sbuf_printf(sb, ":");
1513 sbuf_printf(sb, "%s:",
1514 rule->rr_subject.rs_prison_racct->prr_name);
1517 panic("rctl_rule_to_sbuf: unknown subject type %d",
1518 rule->rr_subject_type);
1521 amount = rule->rr_amount;
1522 if (amount != RCTL_AMOUNT_UNDEFINED &&
1523 RACCT_IS_IN_MILLIONS(rule->rr_resource))
1526 sbuf_printf(sb, "%s:%s=%jd",
1527 rctl_resource_name(rule->rr_resource),
1528 rctl_action_name(rule->rr_action),
1531 if (rule->rr_per != rule->rr_subject_type)
1532 sbuf_printf(sb, "/%s", rctl_subject_type_name(rule->rr_per));
1536 * Routine used by RCTL syscalls to read in input string.
1539 rctl_read_inbuf(char **inputstr, const char *inbufp, size_t inbuflen)
1544 ASSERT_RACCT_ENABLED();
1548 if (inbuflen > RCTL_MAX_INBUFSIZE)
1551 str = malloc(inbuflen + 1, M_RCTL, M_WAITOK);
1552 error = copyinstr(inbufp, str, inbuflen, NULL);
1564 * Routine used by RCTL syscalls to write out output string.
1567 rctl_write_outbuf(struct sbuf *outputsbuf, char *outbufp, size_t outbuflen)
1571 ASSERT_RACCT_ENABLED();
1573 if (outputsbuf == NULL)
1576 sbuf_finish(outputsbuf);
1577 if (outbuflen < sbuf_len(outputsbuf) + 1) {
1578 sbuf_delete(outputsbuf);
1581 error = copyout(sbuf_data(outputsbuf), outbufp,
1582 sbuf_len(outputsbuf) + 1);
1583 sbuf_delete(outputsbuf);
1587 static struct sbuf *
1588 rctl_racct_to_sbuf(struct racct *racct, int sloppy)
1594 ASSERT_RACCT_ENABLED();
1596 sb = sbuf_new_auto();
1597 for (i = 0; i <= RACCT_MAX; i++) {
1598 if (sloppy == 0 && RACCT_IS_SLOPPY(i))
1601 amount = racct->r_resources[i];
1603 if (RACCT_IS_IN_MILLIONS(i))
1605 sbuf_printf(sb, "%s=%jd,", rctl_resource_name(i), amount);
1607 sbuf_setpos(sb, sbuf_len(sb) - 1);
1612 sys_rctl_get_racct(struct thread *td, struct rctl_get_racct_args *uap)
1614 struct rctl_rule *filter;
1615 struct sbuf *outputsbuf = NULL;
1617 struct uidinfo *uip;
1618 struct loginclass *lc;
1619 struct prison_racct *prr;
1626 error = priv_check(td, PRIV_RCTL_GET_RACCT);
1630 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1634 sx_slock(&allproc_lock);
1635 error = rctl_string_to_rule(inputstr, &filter);
1636 free(inputstr, M_RCTL);
1638 sx_sunlock(&allproc_lock);
1642 switch (filter->rr_subject_type) {
1643 case RCTL_SUBJECT_TYPE_PROCESS:
1644 p = filter->rr_subject.rs_proc;
1649 outputsbuf = rctl_racct_to_sbuf(p->p_racct, 0);
1651 case RCTL_SUBJECT_TYPE_USER:
1652 uip = filter->rr_subject.rs_uip;
1657 outputsbuf = rctl_racct_to_sbuf(uip->ui_racct, 1);
1659 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1660 lc = filter->rr_subject.rs_loginclass;
1665 outputsbuf = rctl_racct_to_sbuf(lc->lc_racct, 1);
1667 case RCTL_SUBJECT_TYPE_JAIL:
1668 prr = filter->rr_subject.rs_prison_racct;
1673 outputsbuf = rctl_racct_to_sbuf(prr->prr_racct, 1);
1679 rctl_rule_release(filter);
1680 sx_sunlock(&allproc_lock);
1684 error = rctl_write_outbuf(outputsbuf, uap->outbufp, uap->outbuflen);
1690 rctl_get_rules_callback(struct racct *racct, void *arg2, void *arg3)
1692 struct rctl_rule *filter = (struct rctl_rule *)arg2;
1693 struct rctl_rule_link *link;
1694 struct sbuf *sb = (struct sbuf *)arg3;
1696 ASSERT_RACCT_ENABLED();
1697 RACCT_LOCK_ASSERT();
1699 LIST_FOREACH(link, &racct->r_rule_links, rrl_next) {
1700 if (!rctl_rule_matches(link->rrl_rule, filter))
1702 rctl_rule_to_sbuf(sb, link->rrl_rule);
1703 sbuf_printf(sb, ",");
1708 sys_rctl_get_rules(struct thread *td, struct rctl_get_rules_args *uap)
1711 struct rctl_rule *filter;
1712 struct rctl_rule_link *link;
1714 char *inputstr, *buf;
1721 error = priv_check(td, PRIV_RCTL_GET_RULES);
1725 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1729 sx_slock(&allproc_lock);
1730 error = rctl_string_to_rule(inputstr, &filter);
1731 free(inputstr, M_RCTL);
1733 sx_sunlock(&allproc_lock);
1737 bufsize = uap->outbuflen;
1738 if (bufsize > rctl_maxbufsize) {
1739 sx_sunlock(&allproc_lock);
1743 buf = malloc(bufsize, M_RCTL, M_WAITOK);
1744 sb = sbuf_new(NULL, buf, bufsize, SBUF_FIXEDLEN);
1745 KASSERT(sb != NULL, ("sbuf_new failed"));
1747 FOREACH_PROC_IN_SYSTEM(p) {
1749 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
1751 * Non-process rules will be added to the buffer later.
1752 * Adding them here would result in duplicated output.
1754 if (link->rrl_rule->rr_subject_type !=
1755 RCTL_SUBJECT_TYPE_PROCESS)
1757 if (!rctl_rule_matches(link->rrl_rule, filter))
1759 rctl_rule_to_sbuf(sb, link->rrl_rule);
1760 sbuf_printf(sb, ",");
1765 loginclass_racct_foreach(rctl_get_rules_callback,
1766 rctl_rule_pre_callback, rctl_rule_post_callback,
1768 ui_racct_foreach(rctl_get_rules_callback,
1769 rctl_rule_pre_callback, rctl_rule_post_callback,
1771 prison_racct_foreach(rctl_get_rules_callback,
1772 rctl_rule_pre_callback, rctl_rule_post_callback,
1774 if (sbuf_error(sb) == ENOMEM) {
1780 * Remove trailing ",".
1782 if (sbuf_len(sb) > 0)
1783 sbuf_setpos(sb, sbuf_len(sb) - 1);
1785 error = rctl_write_outbuf(sb, uap->outbufp, uap->outbuflen);
1787 rctl_rule_release(filter);
1788 sx_sunlock(&allproc_lock);
1794 sys_rctl_get_limits(struct thread *td, struct rctl_get_limits_args *uap)
1797 struct rctl_rule *filter;
1798 struct rctl_rule_link *link;
1799 char *inputstr, *buf;
1806 error = priv_check(td, PRIV_RCTL_GET_LIMITS);
1810 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1814 sx_slock(&allproc_lock);
1815 error = rctl_string_to_rule(inputstr, &filter);
1816 free(inputstr, M_RCTL);
1818 sx_sunlock(&allproc_lock);
1822 if (filter->rr_subject_type == RCTL_SUBJECT_TYPE_UNDEFINED) {
1823 rctl_rule_release(filter);
1824 sx_sunlock(&allproc_lock);
1827 if (filter->rr_subject_type != RCTL_SUBJECT_TYPE_PROCESS) {
1828 rctl_rule_release(filter);
1829 sx_sunlock(&allproc_lock);
1830 return (EOPNOTSUPP);
1832 if (filter->rr_subject.rs_proc == NULL) {
1833 rctl_rule_release(filter);
1834 sx_sunlock(&allproc_lock);
1838 bufsize = uap->outbuflen;
1839 if (bufsize > rctl_maxbufsize) {
1840 rctl_rule_release(filter);
1841 sx_sunlock(&allproc_lock);
1845 buf = malloc(bufsize, M_RCTL, M_WAITOK);
1846 sb = sbuf_new(NULL, buf, bufsize, SBUF_FIXEDLEN);
1847 KASSERT(sb != NULL, ("sbuf_new failed"));
1850 LIST_FOREACH(link, &filter->rr_subject.rs_proc->p_racct->r_rule_links,
1852 rctl_rule_to_sbuf(sb, link->rrl_rule);
1853 sbuf_printf(sb, ",");
1856 if (sbuf_error(sb) == ENOMEM) {
1863 * Remove trailing ",".
1865 if (sbuf_len(sb) > 0)
1866 sbuf_setpos(sb, sbuf_len(sb) - 1);
1868 error = rctl_write_outbuf(sb, uap->outbufp, uap->outbuflen);
1870 rctl_rule_release(filter);
1871 sx_sunlock(&allproc_lock);
1877 sys_rctl_add_rule(struct thread *td, struct rctl_add_rule_args *uap)
1879 struct rctl_rule *rule;
1886 error = priv_check(td, PRIV_RCTL_ADD_RULE);
1890 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1894 sx_slock(&allproc_lock);
1895 error = rctl_string_to_rule(inputstr, &rule);
1896 free(inputstr, M_RCTL);
1898 sx_sunlock(&allproc_lock);
1902 * The 'per' part of a rule is optional.
1904 if (rule->rr_per == RCTL_SUBJECT_TYPE_UNDEFINED &&
1905 rule->rr_subject_type != RCTL_SUBJECT_TYPE_UNDEFINED)
1906 rule->rr_per = rule->rr_subject_type;
1908 if (!rctl_rule_fully_specified(rule)) {
1913 error = rctl_rule_add(rule);
1916 rctl_rule_release(rule);
1917 sx_sunlock(&allproc_lock);
1922 sys_rctl_remove_rule(struct thread *td, struct rctl_remove_rule_args *uap)
1924 struct rctl_rule *filter;
1931 error = priv_check(td, PRIV_RCTL_REMOVE_RULE);
1935 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1939 sx_slock(&allproc_lock);
1940 error = rctl_string_to_rule(inputstr, &filter);
1941 free(inputstr, M_RCTL);
1943 sx_sunlock(&allproc_lock);
1947 error = rctl_rule_remove(filter);
1948 rctl_rule_release(filter);
1949 sx_sunlock(&allproc_lock);
1955 * Update RCTL rule list after credential change.
1958 rctl_proc_ucred_changed(struct proc *p, struct ucred *newcred)
1960 LIST_HEAD(, rctl_rule_link) newrules;
1961 struct rctl_rule_link *link, *newlink;
1962 struct uidinfo *newuip;
1963 struct loginclass *newlc;
1964 struct prison_racct *newprr;
1970 PROC_LOCK_ASSERT(p, MA_NOTOWNED);
1972 newuip = newcred->cr_ruidinfo;
1973 newlc = newcred->cr_loginclass;
1974 newprr = newcred->cr_prison->pr_prison_racct;
1976 LIST_INIT(&newrules);
1980 * First, count the rules that apply to the process with new
1985 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
1986 if (link->rrl_rule->rr_subject_type ==
1987 RCTL_SUBJECT_TYPE_PROCESS)
1990 LIST_FOREACH(link, &newuip->ui_racct->r_rule_links, rrl_next)
1992 LIST_FOREACH(link, &newlc->lc_racct->r_rule_links, rrl_next)
1994 LIST_FOREACH(link, &newprr->prr_racct->r_rule_links, rrl_next)
1999 * Create temporary list. We've dropped the rctl_lock in order
2002 for (i = 0; i < rulecnt; i++) {
2003 newlink = uma_zalloc(rctl_rule_link_zone, M_WAITOK);
2004 newlink->rrl_rule = NULL;
2005 newlink->rrl_exceeded = 0;
2006 LIST_INSERT_HEAD(&newrules, newlink, rrl_next);
2009 newlink = LIST_FIRST(&newrules);
2012 * Assign rules to the newly allocated list entries.
2015 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
2016 if (link->rrl_rule->rr_subject_type ==
2017 RCTL_SUBJECT_TYPE_PROCESS) {
2018 if (newlink == NULL)
2020 rctl_rule_acquire(link->rrl_rule);
2021 newlink->rrl_rule = link->rrl_rule;
2022 newlink->rrl_exceeded = link->rrl_exceeded;
2023 newlink = LIST_NEXT(newlink, rrl_next);
2028 LIST_FOREACH(link, &newuip->ui_racct->r_rule_links, rrl_next) {
2029 if (newlink == NULL)
2031 rctl_rule_acquire(link->rrl_rule);
2032 newlink->rrl_rule = link->rrl_rule;
2033 newlink->rrl_exceeded = link->rrl_exceeded;
2034 newlink = LIST_NEXT(newlink, rrl_next);
2038 LIST_FOREACH(link, &newlc->lc_racct->r_rule_links, rrl_next) {
2039 if (newlink == NULL)
2041 rctl_rule_acquire(link->rrl_rule);
2042 newlink->rrl_rule = link->rrl_rule;
2043 newlink->rrl_exceeded = link->rrl_exceeded;
2044 newlink = LIST_NEXT(newlink, rrl_next);
2048 LIST_FOREACH(link, &newprr->prr_racct->r_rule_links, rrl_next) {
2049 if (newlink == NULL)
2051 rctl_rule_acquire(link->rrl_rule);
2052 newlink->rrl_rule = link->rrl_rule;
2053 newlink->rrl_exceeded = link->rrl_exceeded;
2054 newlink = LIST_NEXT(newlink, rrl_next);
2060 * Free the old rule list.
2062 while (!LIST_EMPTY(&p->p_racct->r_rule_links)) {
2063 link = LIST_FIRST(&p->p_racct->r_rule_links);
2064 LIST_REMOVE(link, rrl_next);
2065 rctl_rule_release(link->rrl_rule);
2066 uma_zfree(rctl_rule_link_zone, link);
2070 * Replace lists and we're done.
2072 * XXX: Is there any way to switch list heads instead
2073 * of iterating here?
2075 while (!LIST_EMPTY(&newrules)) {
2076 newlink = LIST_FIRST(&newrules);
2077 LIST_REMOVE(newlink, rrl_next);
2078 LIST_INSERT_HEAD(&p->p_racct->r_rule_links,
2091 * Rule list changed while we were not holding the rctl_lock.
2092 * Free the new list and try again.
2094 while (!LIST_EMPTY(&newrules)) {
2095 newlink = LIST_FIRST(&newrules);
2096 LIST_REMOVE(newlink, rrl_next);
2097 if (newlink->rrl_rule != NULL)
2098 rctl_rule_release(newlink->rrl_rule);
2099 uma_zfree(rctl_rule_link_zone, newlink);
2106 * Assign RCTL rules to the newly created process.
2109 rctl_proc_fork(struct proc *parent, struct proc *child)
2111 struct rctl_rule *rule;
2112 struct rctl_rule_link *link;
2115 ASSERT_RACCT_ENABLED();
2116 RACCT_LOCK_ASSERT();
2117 KASSERT(parent->p_racct != NULL, ("process without racct; p = %p", parent));
2119 LIST_INIT(&child->p_racct->r_rule_links);
2122 * Go through limits applicable to the parent and assign them
2123 * to the child. Rules with 'process' subject have to be duplicated
2124 * in order to make their rr_subject point to the new process.
2126 LIST_FOREACH(link, &parent->p_racct->r_rule_links, rrl_next) {
2127 if (link->rrl_rule->rr_subject_type ==
2128 RCTL_SUBJECT_TYPE_PROCESS) {
2129 rule = rctl_rule_duplicate(link->rrl_rule, M_NOWAIT);
2132 KASSERT(rule->rr_subject.rs_proc == parent,
2133 ("rule->rr_subject.rs_proc != parent"));
2134 rule->rr_subject.rs_proc = child;
2135 error = rctl_racct_add_rule_locked(child->p_racct,
2137 rctl_rule_release(rule);
2141 error = rctl_racct_add_rule_locked(child->p_racct,
2151 while (!LIST_EMPTY(&child->p_racct->r_rule_links)) {
2152 link = LIST_FIRST(&child->p_racct->r_rule_links);
2153 LIST_REMOVE(link, rrl_next);
2154 rctl_rule_release(link->rrl_rule);
2155 uma_zfree(rctl_rule_link_zone, link);
2162 * Release rules attached to the racct.
2165 rctl_racct_release(struct racct *racct)
2167 struct rctl_rule_link *link;
2169 ASSERT_RACCT_ENABLED();
2170 RACCT_LOCK_ASSERT();
2172 while (!LIST_EMPTY(&racct->r_rule_links)) {
2173 link = LIST_FIRST(&racct->r_rule_links);
2174 LIST_REMOVE(link, rrl_next);
2175 rctl_rule_release(link->rrl_rule);
2176 uma_zfree(rctl_rule_link_zone, link);
2187 rctl_rule_zone = uma_zcreate("rctl_rule", sizeof(struct rctl_rule),
2188 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2189 rctl_rule_link_zone = uma_zcreate("rctl_rule_link",
2190 sizeof(struct rctl_rule_link), NULL, NULL, NULL, NULL,
2194 * Set default values, making sure not to overwrite the ones
2195 * fetched from tunables. Most of those could be set at the
2196 * declaration, except for the rctl_throttle_max - we cannot
2197 * set it there due to hz not being compile time constant.
2199 if (rctl_throttle_min < 1)
2200 rctl_throttle_min = 1;
2201 if (rctl_throttle_max < rctl_throttle_min)
2202 rctl_throttle_max = 2 * hz;
2203 if (rctl_throttle_pct < 0)
2204 rctl_throttle_pct = 100;
2205 if (rctl_throttle_pct2 < 0)
2206 rctl_throttle_pct2 = 100;
2212 sys_rctl_get_racct(struct thread *td, struct rctl_get_racct_args *uap)
2219 sys_rctl_get_rules(struct thread *td, struct rctl_get_rules_args *uap)
2226 sys_rctl_get_limits(struct thread *td, struct rctl_get_limits_args *uap)
2233 sys_rctl_add_rule(struct thread *td, struct rctl_add_rule_args *uap)
2240 sys_rctl_remove_rule(struct thread *td, struct rctl_remove_rule_args *uap)