2 * Copyright (c) 2010 The FreeBSD Foundation
5 * This software was developed by Edward Tomasz Napierala under sponsorship
6 * from the FreeBSD Foundation.
8 * Redistribution and use in source and binary forms, with or without
9 * modification, are permitted provided that the following conditions
11 * 1. Redistributions of source code must retain the above copyright
12 * notice, this list of conditions and the following disclaimer.
13 * 2. Redistributions in binary form must reproduce the above copyright
14 * notice, this list of conditions and the following disclaimer in the
15 * documentation and/or other materials provided with the distribution.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
32 #include <sys/cdefs.h>
33 __FBSDID("$FreeBSD$");
35 #include <sys/param.h>
37 #include <sys/malloc.h>
38 #include <sys/queue.h>
39 #include <sys/refcount.h>
41 #include <sys/kernel.h>
42 #include <sys/limits.h>
43 #include <sys/loginclass.h>
46 #include <sys/racct.h>
48 #include <sys/resourcevar.h>
50 #include <sys/sysent.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, 0, "Resource Limits");
98 SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, maxbufsize, CTLFLAG_RWTUN,
99 &rctl_maxbufsize, 0, "Maximum output buffer size");
100 SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, log_rate_limit, CTLFLAG_RW,
101 &rctl_log_rate_limit, 0, "Maximum number of log messages per second");
102 SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, devctl_rate_limit, CTLFLAG_RWTUN,
103 &rctl_devctl_rate_limit, 0, "Maximum number of devctl messages per second");
104 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_min,
105 CTLTYPE_UINT | CTLFLAG_RWTUN, 0, 0, &rctl_throttle_min_sysctl, "IU",
106 "Shortest throttling duration, in hz");
107 TUNABLE_INT("kern.racct.rctl.throttle_min", &rctl_throttle_min);
108 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_max,
109 CTLTYPE_UINT | CTLFLAG_RWTUN, 0, 0, &rctl_throttle_max_sysctl, "IU",
110 "Longest throttling duration, in hz");
111 TUNABLE_INT("kern.racct.rctl.throttle_max", &rctl_throttle_max);
112 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_pct,
113 CTLTYPE_UINT | CTLFLAG_RWTUN, 0, 0, &rctl_throttle_pct_sysctl, "IU",
114 "Throttling penalty for process consumption, in percent");
115 TUNABLE_INT("kern.racct.rctl.throttle_pct", &rctl_throttle_pct);
116 SYSCTL_PROC(_kern_racct_rctl, OID_AUTO, throttle_pct2,
117 CTLTYPE_UINT | CTLFLAG_RWTUN, 0, 0, &rctl_throttle_pct2_sysctl, "IU",
118 "Throttling penalty for container consumption, in percent");
119 TUNABLE_INT("kern.racct.rctl.throttle_pct2", &rctl_throttle_pct2);
122 * 'rctl_rule_link' connects a rule with every racct it's related to.
123 * For example, rule 'user:X:openfiles:deny=N/process' is linked
124 * with uidinfo for user X, and to each process of that user.
126 struct rctl_rule_link {
127 LIST_ENTRY(rctl_rule_link) rrl_next;
128 struct rctl_rule *rrl_rule;
137 static struct dict subjectnames[] = {
138 { "process", RCTL_SUBJECT_TYPE_PROCESS },
139 { "user", RCTL_SUBJECT_TYPE_USER },
140 { "loginclass", RCTL_SUBJECT_TYPE_LOGINCLASS },
141 { "jail", RCTL_SUBJECT_TYPE_JAIL },
144 static struct dict resourcenames[] = {
145 { "cputime", RACCT_CPU },
146 { "datasize", RACCT_DATA },
147 { "stacksize", RACCT_STACK },
148 { "coredumpsize", RACCT_CORE },
149 { "memoryuse", RACCT_RSS },
150 { "memorylocked", RACCT_MEMLOCK },
151 { "maxproc", RACCT_NPROC },
152 { "openfiles", RACCT_NOFILE },
153 { "vmemoryuse", RACCT_VMEM },
154 { "pseudoterminals", RACCT_NPTS },
155 { "swapuse", RACCT_SWAP },
156 { "nthr", RACCT_NTHR },
157 { "msgqqueued", RACCT_MSGQQUEUED },
158 { "msgqsize", RACCT_MSGQSIZE },
159 { "nmsgq", RACCT_NMSGQ },
160 { "nsem", RACCT_NSEM },
161 { "nsemop", RACCT_NSEMOP },
162 { "nshm", RACCT_NSHM },
163 { "shmsize", RACCT_SHMSIZE },
164 { "wallclock", RACCT_WALLCLOCK },
165 { "pcpu", RACCT_PCTCPU },
166 { "readbps", RACCT_READBPS },
167 { "writebps", RACCT_WRITEBPS },
168 { "readiops", RACCT_READIOPS },
169 { "writeiops", RACCT_WRITEIOPS },
172 static struct dict actionnames[] = {
173 { "sighup", RCTL_ACTION_SIGHUP },
174 { "sigint", RCTL_ACTION_SIGINT },
175 { "sigquit", RCTL_ACTION_SIGQUIT },
176 { "sigill", RCTL_ACTION_SIGILL },
177 { "sigtrap", RCTL_ACTION_SIGTRAP },
178 { "sigabrt", RCTL_ACTION_SIGABRT },
179 { "sigemt", RCTL_ACTION_SIGEMT },
180 { "sigfpe", RCTL_ACTION_SIGFPE },
181 { "sigkill", RCTL_ACTION_SIGKILL },
182 { "sigbus", RCTL_ACTION_SIGBUS },
183 { "sigsegv", RCTL_ACTION_SIGSEGV },
184 { "sigsys", RCTL_ACTION_SIGSYS },
185 { "sigpipe", RCTL_ACTION_SIGPIPE },
186 { "sigalrm", RCTL_ACTION_SIGALRM },
187 { "sigterm", RCTL_ACTION_SIGTERM },
188 { "sigurg", RCTL_ACTION_SIGURG },
189 { "sigstop", RCTL_ACTION_SIGSTOP },
190 { "sigtstp", RCTL_ACTION_SIGTSTP },
191 { "sigchld", RCTL_ACTION_SIGCHLD },
192 { "sigttin", RCTL_ACTION_SIGTTIN },
193 { "sigttou", RCTL_ACTION_SIGTTOU },
194 { "sigio", RCTL_ACTION_SIGIO },
195 { "sigxcpu", RCTL_ACTION_SIGXCPU },
196 { "sigxfsz", RCTL_ACTION_SIGXFSZ },
197 { "sigvtalrm", RCTL_ACTION_SIGVTALRM },
198 { "sigprof", RCTL_ACTION_SIGPROF },
199 { "sigwinch", RCTL_ACTION_SIGWINCH },
200 { "siginfo", RCTL_ACTION_SIGINFO },
201 { "sigusr1", RCTL_ACTION_SIGUSR1 },
202 { "sigusr2", RCTL_ACTION_SIGUSR2 },
203 { "sigthr", RCTL_ACTION_SIGTHR },
204 { "deny", RCTL_ACTION_DENY },
205 { "log", RCTL_ACTION_LOG },
206 { "devctl", RCTL_ACTION_DEVCTL },
207 { "throttle", RCTL_ACTION_THROTTLE },
210 static void rctl_init(void);
211 SYSINIT(rctl, SI_SUB_RACCT, SI_ORDER_FIRST, rctl_init, NULL);
213 static uma_zone_t rctl_rule_zone;
214 static uma_zone_t rctl_rule_link_zone;
216 static int rctl_rule_fully_specified(const struct rctl_rule *rule);
217 static void rctl_rule_to_sbuf(struct sbuf *sb, const struct rctl_rule *rule);
219 static MALLOC_DEFINE(M_RCTL, "rctl", "Resource Limits");
221 static int rctl_throttle_min_sysctl(SYSCTL_HANDLER_ARGS)
223 int error, val = rctl_throttle_min;
225 error = sysctl_handle_int(oidp, &val, 0, req);
226 if (error || !req->newptr)
228 if (val < 1 || val > rctl_throttle_max)
232 rctl_throttle_min = val;
238 static int rctl_throttle_max_sysctl(SYSCTL_HANDLER_ARGS)
240 int error, val = rctl_throttle_max;
242 error = sysctl_handle_int(oidp, &val, 0, req);
243 if (error || !req->newptr)
245 if (val < rctl_throttle_min)
249 rctl_throttle_max = val;
255 static int rctl_throttle_pct_sysctl(SYSCTL_HANDLER_ARGS)
257 int error, val = rctl_throttle_pct;
259 error = sysctl_handle_int(oidp, &val, 0, req);
260 if (error || !req->newptr)
266 rctl_throttle_pct = val;
272 static int rctl_throttle_pct2_sysctl(SYSCTL_HANDLER_ARGS)
274 int error, val = rctl_throttle_pct2;
276 error = sysctl_handle_int(oidp, &val, 0, req);
277 if (error || !req->newptr)
283 rctl_throttle_pct2 = val;
290 rctl_subject_type_name(int subject)
294 for (i = 0; subjectnames[i].d_name != NULL; i++) {
295 if (subjectnames[i].d_value == subject)
296 return (subjectnames[i].d_name);
299 panic("rctl_subject_type_name: unknown subject type %d", subject);
303 rctl_action_name(int action)
307 for (i = 0; actionnames[i].d_name != NULL; i++) {
308 if (actionnames[i].d_value == action)
309 return (actionnames[i].d_name);
312 panic("rctl_action_name: unknown action %d", action);
316 rctl_resource_name(int resource)
320 for (i = 0; resourcenames[i].d_name != NULL; i++) {
321 if (resourcenames[i].d_value == resource)
322 return (resourcenames[i].d_name);
325 panic("rctl_resource_name: unknown resource %d", resource);
328 static struct racct *
329 rctl_proc_rule_to_racct(const struct proc *p, const struct rctl_rule *rule)
331 struct ucred *cred = p->p_ucred;
333 ASSERT_RACCT_ENABLED();
336 switch (rule->rr_per) {
337 case RCTL_SUBJECT_TYPE_PROCESS:
339 case RCTL_SUBJECT_TYPE_USER:
340 return (cred->cr_ruidinfo->ui_racct);
341 case RCTL_SUBJECT_TYPE_LOGINCLASS:
342 return (cred->cr_loginclass->lc_racct);
343 case RCTL_SUBJECT_TYPE_JAIL:
344 return (cred->cr_prison->pr_prison_racct->prr_racct);
346 panic("%s: unknown per %d", __func__, rule->rr_per);
351 * Return the amount of resource that can be allocated by 'p' before
355 rctl_available_resource(const struct proc *p, const struct rctl_rule *rule)
357 const struct racct *racct;
360 ASSERT_RACCT_ENABLED();
363 racct = rctl_proc_rule_to_racct(p, rule);
364 available = rule->rr_amount - racct->r_resources[rule->rr_resource];
370 * Called every second for proc, uidinfo, loginclass, and jail containers.
371 * If the limit isn't exceeded, it decreases the usage amount to zero.
372 * Otherwise, it decreases it by the value of the limit. This way
373 * resource consumption exceeding the limit "carries over" to the next
377 rctl_throttle_decay(struct racct *racct, int resource)
379 struct rctl_rule *rule;
380 struct rctl_rule_link *link;
381 int64_t minavailable;
383 ASSERT_RACCT_ENABLED();
386 minavailable = INT64_MAX;
388 LIST_FOREACH(link, &racct->r_rule_links, rrl_next) {
389 rule = link->rrl_rule;
391 if (rule->rr_resource != resource)
393 if (rule->rr_action != RCTL_ACTION_THROTTLE)
396 if (rule->rr_amount < minavailable)
397 minavailable = rule->rr_amount;
400 if (racct->r_resources[resource] < minavailable) {
401 racct->r_resources[resource] = 0;
404 * Cap utilization counter at ten times the limit. Otherwise,
405 * if we changed the rule lowering the allowed amount, it could
406 * take unreasonably long time for the accumulated resource
409 if (racct->r_resources[resource] > minavailable * 10)
410 racct->r_resources[resource] = minavailable * 10;
412 racct->r_resources[resource] -= minavailable;
417 * Special version of rctl_get_available() for the %CPU resource.
418 * We slightly cheat here and return less than we normally would.
421 rctl_pcpu_available(const struct proc *p) {
422 struct rctl_rule *rule;
423 struct rctl_rule_link *link;
424 int64_t available, minavailable, limit;
426 ASSERT_RACCT_ENABLED();
429 minavailable = INT64_MAX;
432 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
433 rule = link->rrl_rule;
434 if (rule->rr_resource != RACCT_PCTCPU)
436 if (rule->rr_action != RCTL_ACTION_DENY)
438 available = rctl_available_resource(p, rule);
439 if (available < minavailable) {
440 minavailable = available;
441 limit = rule->rr_amount;
446 * Return slightly less than actual value of the available
447 * %cpu resource. This makes %cpu throttling more aggressive
448 * and lets us act sooner than the limits are already exceeded.
451 if (limit > 2 * RCTL_PCPU_SHIFT)
452 minavailable -= RCTL_PCPU_SHIFT;
454 minavailable -= (limit / 2);
457 return (minavailable);
461 xadd(uint64_t a, uint64_t b)
477 xmul(uint64_t a, uint64_t b)
480 if (b != 0 && a > UINT64_MAX / b)
487 * Check whether the proc 'p' can allocate 'amount' of 'resource' in addition
488 * to what it keeps allocated now. Returns non-zero if the allocation should
489 * be denied, 0 otherwise.
492 rctl_enforce(struct proc *p, int resource, uint64_t amount)
494 static struct timeval log_lasttime, devctl_lasttime;
495 static int log_curtime = 0, devctl_curtime = 0;
496 struct rctl_rule *rule;
497 struct rctl_rule_link *link;
501 uint64_t sleep_ms, sleep_ratio;
504 ASSERT_RACCT_ENABLED();
508 * There may be more than one matching rule; go through all of them.
509 * Denial should be done last, after logging and sending signals.
511 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
512 rule = link->rrl_rule;
513 if (rule->rr_resource != resource)
516 available = rctl_available_resource(p, rule);
517 if (available >= (int64_t)amount) {
518 link->rrl_exceeded = 0;
522 switch (rule->rr_action) {
523 case RCTL_ACTION_DENY:
526 case RCTL_ACTION_LOG:
528 * If rrl_exceeded != 0, it means we've already
529 * logged a warning for this process.
531 if (link->rrl_exceeded != 0)
535 * If the process state is not fully initialized yet,
536 * we can't access most of the required fields, e.g.
537 * p->p_comm. This happens when called from fork1().
538 * Ignore this rule for now; it will be processed just
539 * after fork, when called from racct_proc_fork_done().
541 if (p->p_state != PRS_NORMAL)
544 if (!ppsratecheck(&log_lasttime, &log_curtime,
545 rctl_log_rate_limit))
548 buf = malloc(RCTL_LOG_BUFSIZE, M_RCTL, M_NOWAIT);
550 printf("rctl_enforce: out of memory\n");
553 sbuf_new(&sb, buf, RCTL_LOG_BUFSIZE, SBUF_FIXEDLEN);
554 rctl_rule_to_sbuf(&sb, rule);
556 printf("rctl: rule \"%s\" matched by pid %d "
557 "(%s), uid %d, jail %s\n", sbuf_data(&sb),
558 p->p_pid, p->p_comm, p->p_ucred->cr_uid,
559 p->p_ucred->cr_prison->pr_prison_racct->prr_name);
562 link->rrl_exceeded = 1;
564 case RCTL_ACTION_DEVCTL:
565 if (link->rrl_exceeded != 0)
568 if (p->p_state != PRS_NORMAL)
571 if (!ppsratecheck(&devctl_lasttime, &devctl_curtime,
572 rctl_devctl_rate_limit))
575 buf = malloc(RCTL_LOG_BUFSIZE, M_RCTL, M_NOWAIT);
577 printf("rctl_enforce: out of memory\n");
580 sbuf_new(&sb, buf, RCTL_LOG_BUFSIZE, SBUF_FIXEDLEN);
581 sbuf_printf(&sb, "rule=");
582 rctl_rule_to_sbuf(&sb, rule);
583 sbuf_printf(&sb, " pid=%d ruid=%d jail=%s",
584 p->p_pid, p->p_ucred->cr_ruid,
585 p->p_ucred->cr_prison->pr_prison_racct->prr_name);
587 devctl_notify_f("RCTL", "rule", "matched",
588 sbuf_data(&sb), M_NOWAIT);
591 link->rrl_exceeded = 1;
593 case RCTL_ACTION_THROTTLE:
594 if (p->p_state != PRS_NORMAL)
598 * Make the process sleep for a fraction of second
599 * proportional to the ratio of process' resource
600 * utilization compared to the limit. The point is
601 * to penalize resource hogs: processes that consume
602 * more of the available resources sleep for longer.
604 * We're trying to defer division until the very end,
605 * to minimize the rounding effects. The following
606 * calculation could have been written in a clearer
609 * sleep_ms = hz * p->p_racct->r_resources[resource] /
611 * sleep_ms *= rctl_throttle_pct / 100;
612 * if (sleep_ms < rctl_throttle_min)
613 * sleep_ms = rctl_throttle_min;
616 sleep_ms = xmul(hz, p->p_racct->r_resources[resource]);
617 sleep_ms = xmul(sleep_ms, rctl_throttle_pct) / 100;
618 if (sleep_ms < rctl_throttle_min * rule->rr_amount)
619 sleep_ms = rctl_throttle_min * rule->rr_amount;
622 * Multiply that by the ratio of the resource
623 * consumption for the container compared to the limit,
624 * squared. In other words, a process in a container
625 * that is two times over the limit will be throttled
626 * four times as much for hitting the same rule. The
627 * point is to penalize processes more if the container
628 * itself (eg certain UID or jail) is above the limit.
631 sleep_ratio = -available / rule->rr_amount;
634 sleep_ratio = xmul(sleep_ratio, sleep_ratio);
635 sleep_ratio = xmul(sleep_ratio, rctl_throttle_pct2) / 100;
636 sleep_ms = xadd(sleep_ms, xmul(sleep_ms, sleep_ratio));
639 * Finally the division.
641 sleep_ms /= rule->rr_amount;
643 if (sleep_ms > rctl_throttle_max)
644 sleep_ms = rctl_throttle_max;
646 printf("%s: pid %d (%s), %jd of %jd, will sleep for %ju ms (ratio %ju, available %jd)\n",
647 __func__, p->p_pid, p->p_comm,
648 p->p_racct->r_resources[resource],
649 rule->rr_amount, (uintmax_t)sleep_ms,
650 (uintmax_t)sleep_ratio, (intmax_t)available);
653 KASSERT(sleep_ms >= rctl_throttle_min, ("%s: %ju < %d\n",
654 __func__, (uintmax_t)sleep_ms, rctl_throttle_min));
655 racct_proc_throttle(p, sleep_ms);
658 if (link->rrl_exceeded != 0)
661 if (p->p_state != PRS_NORMAL)
664 KASSERT(rule->rr_action > 0 &&
665 rule->rr_action <= RCTL_ACTION_SIGNAL_MAX,
666 ("rctl_enforce: unknown action %d",
670 * We're using the fact that RCTL_ACTION_SIG* values
671 * are equal to their counterparts from sys/signal.h.
673 kern_psignal(p, rule->rr_action);
674 link->rrl_exceeded = 1;
681 * Return fake error code; the caller should change it
682 * into one proper for the situation - EFSIZ, ENOMEM etc.
691 rctl_get_limit(struct proc *p, int resource)
693 struct rctl_rule *rule;
694 struct rctl_rule_link *link;
695 uint64_t amount = UINT64_MAX;
697 ASSERT_RACCT_ENABLED();
701 * There may be more than one matching rule; go through all of them.
702 * Denial should be done last, after logging and sending signals.
704 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
705 rule = link->rrl_rule;
706 if (rule->rr_resource != resource)
708 if (rule->rr_action != RCTL_ACTION_DENY)
710 if (rule->rr_amount < amount)
711 amount = rule->rr_amount;
718 rctl_get_available(struct proc *p, int resource)
720 struct rctl_rule *rule;
721 struct rctl_rule_link *link;
722 int64_t available, minavailable, allocated;
724 minavailable = INT64_MAX;
726 ASSERT_RACCT_ENABLED();
730 * There may be more than one matching rule; go through all of them.
731 * Denial should be done last, after logging and sending signals.
733 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
734 rule = link->rrl_rule;
735 if (rule->rr_resource != resource)
737 if (rule->rr_action != RCTL_ACTION_DENY)
739 available = rctl_available_resource(p, rule);
740 if (available < minavailable)
741 minavailable = available;
745 * XXX: Think about this _hard_.
747 allocated = p->p_racct->r_resources[resource];
748 if (minavailable < INT64_MAX - allocated)
749 minavailable += allocated;
750 if (minavailable < 0)
753 return (minavailable);
757 rctl_rule_matches(const struct rctl_rule *rule, const struct rctl_rule *filter)
760 ASSERT_RACCT_ENABLED();
762 if (filter->rr_subject_type != RCTL_SUBJECT_TYPE_UNDEFINED) {
763 if (rule->rr_subject_type != filter->rr_subject_type)
766 switch (filter->rr_subject_type) {
767 case RCTL_SUBJECT_TYPE_PROCESS:
768 if (filter->rr_subject.rs_proc != NULL &&
769 rule->rr_subject.rs_proc !=
770 filter->rr_subject.rs_proc)
773 case RCTL_SUBJECT_TYPE_USER:
774 if (filter->rr_subject.rs_uip != NULL &&
775 rule->rr_subject.rs_uip !=
776 filter->rr_subject.rs_uip)
779 case RCTL_SUBJECT_TYPE_LOGINCLASS:
780 if (filter->rr_subject.rs_loginclass != NULL &&
781 rule->rr_subject.rs_loginclass !=
782 filter->rr_subject.rs_loginclass)
785 case RCTL_SUBJECT_TYPE_JAIL:
786 if (filter->rr_subject.rs_prison_racct != NULL &&
787 rule->rr_subject.rs_prison_racct !=
788 filter->rr_subject.rs_prison_racct)
792 panic("rctl_rule_matches: unknown subject type %d",
793 filter->rr_subject_type);
797 if (filter->rr_resource != RACCT_UNDEFINED) {
798 if (rule->rr_resource != filter->rr_resource)
802 if (filter->rr_action != RCTL_ACTION_UNDEFINED) {
803 if (rule->rr_action != filter->rr_action)
807 if (filter->rr_amount != RCTL_AMOUNT_UNDEFINED) {
808 if (rule->rr_amount != filter->rr_amount)
812 if (filter->rr_per != RCTL_SUBJECT_TYPE_UNDEFINED) {
813 if (rule->rr_per != filter->rr_per)
821 str2value(const char *str, int *value, struct dict *table)
828 for (i = 0; table[i].d_name != NULL; i++) {
829 if (strcasecmp(table[i].d_name, str) == 0) {
830 *value = table[i].d_value;
839 str2id(const char *str, id_t *value)
846 *value = strtoul(str, &end, 10);
847 if ((size_t)(end - str) != strlen(str))
854 str2int64(const char *str, int64_t *value)
861 *value = strtoul(str, &end, 10);
862 if ((size_t)(end - str) != strlen(str))
872 * Connect the rule to the racct, increasing refcount for the rule.
875 rctl_racct_add_rule(struct racct *racct, struct rctl_rule *rule)
877 struct rctl_rule_link *link;
879 ASSERT_RACCT_ENABLED();
880 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified"));
882 rctl_rule_acquire(rule);
883 link = uma_zalloc(rctl_rule_link_zone, M_WAITOK);
884 link->rrl_rule = rule;
885 link->rrl_exceeded = 0;
888 LIST_INSERT_HEAD(&racct->r_rule_links, link, rrl_next);
893 rctl_racct_add_rule_locked(struct racct *racct, struct rctl_rule *rule)
895 struct rctl_rule_link *link;
897 ASSERT_RACCT_ENABLED();
898 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified"));
901 link = uma_zalloc(rctl_rule_link_zone, M_NOWAIT);
904 rctl_rule_acquire(rule);
905 link->rrl_rule = rule;
906 link->rrl_exceeded = 0;
908 LIST_INSERT_HEAD(&racct->r_rule_links, link, rrl_next);
914 * Remove limits for a rules matching the filter and release
915 * the refcounts for the rules, possibly freeing them. Returns
916 * the number of limit structures removed.
919 rctl_racct_remove_rules(struct racct *racct,
920 const struct rctl_rule *filter)
922 struct rctl_rule_link *link, *linktmp;
925 ASSERT_RACCT_ENABLED();
928 LIST_FOREACH_SAFE(link, &racct->r_rule_links, rrl_next, linktmp) {
929 if (!rctl_rule_matches(link->rrl_rule, filter))
932 LIST_REMOVE(link, rrl_next);
933 rctl_rule_release(link->rrl_rule);
934 uma_zfree(rctl_rule_link_zone, link);
941 rctl_rule_acquire_subject(struct rctl_rule *rule)
944 ASSERT_RACCT_ENABLED();
946 switch (rule->rr_subject_type) {
947 case RCTL_SUBJECT_TYPE_UNDEFINED:
948 case RCTL_SUBJECT_TYPE_PROCESS:
950 case RCTL_SUBJECT_TYPE_JAIL:
951 if (rule->rr_subject.rs_prison_racct != NULL)
952 prison_racct_hold(rule->rr_subject.rs_prison_racct);
954 case RCTL_SUBJECT_TYPE_USER:
955 if (rule->rr_subject.rs_uip != NULL)
956 uihold(rule->rr_subject.rs_uip);
958 case RCTL_SUBJECT_TYPE_LOGINCLASS:
959 if (rule->rr_subject.rs_loginclass != NULL)
960 loginclass_hold(rule->rr_subject.rs_loginclass);
963 panic("rctl_rule_acquire_subject: unknown subject type %d",
964 rule->rr_subject_type);
969 rctl_rule_release_subject(struct rctl_rule *rule)
972 ASSERT_RACCT_ENABLED();
974 switch (rule->rr_subject_type) {
975 case RCTL_SUBJECT_TYPE_UNDEFINED:
976 case RCTL_SUBJECT_TYPE_PROCESS:
978 case RCTL_SUBJECT_TYPE_JAIL:
979 if (rule->rr_subject.rs_prison_racct != NULL)
980 prison_racct_free(rule->rr_subject.rs_prison_racct);
982 case RCTL_SUBJECT_TYPE_USER:
983 if (rule->rr_subject.rs_uip != NULL)
984 uifree(rule->rr_subject.rs_uip);
986 case RCTL_SUBJECT_TYPE_LOGINCLASS:
987 if (rule->rr_subject.rs_loginclass != NULL)
988 loginclass_free(rule->rr_subject.rs_loginclass);
991 panic("rctl_rule_release_subject: unknown subject type %d",
992 rule->rr_subject_type);
997 rctl_rule_alloc(int flags)
999 struct rctl_rule *rule;
1001 ASSERT_RACCT_ENABLED();
1003 rule = uma_zalloc(rctl_rule_zone, flags);
1006 rule->rr_subject_type = RCTL_SUBJECT_TYPE_UNDEFINED;
1007 rule->rr_subject.rs_proc = NULL;
1008 rule->rr_subject.rs_uip = NULL;
1009 rule->rr_subject.rs_loginclass = NULL;
1010 rule->rr_subject.rs_prison_racct = NULL;
1011 rule->rr_per = RCTL_SUBJECT_TYPE_UNDEFINED;
1012 rule->rr_resource = RACCT_UNDEFINED;
1013 rule->rr_action = RCTL_ACTION_UNDEFINED;
1014 rule->rr_amount = RCTL_AMOUNT_UNDEFINED;
1015 refcount_init(&rule->rr_refcount, 1);
1021 rctl_rule_duplicate(const struct rctl_rule *rule, int flags)
1023 struct rctl_rule *copy;
1025 ASSERT_RACCT_ENABLED();
1027 copy = uma_zalloc(rctl_rule_zone, flags);
1030 copy->rr_subject_type = rule->rr_subject_type;
1031 copy->rr_subject.rs_proc = rule->rr_subject.rs_proc;
1032 copy->rr_subject.rs_uip = rule->rr_subject.rs_uip;
1033 copy->rr_subject.rs_loginclass = rule->rr_subject.rs_loginclass;
1034 copy->rr_subject.rs_prison_racct = rule->rr_subject.rs_prison_racct;
1035 copy->rr_per = rule->rr_per;
1036 copy->rr_resource = rule->rr_resource;
1037 copy->rr_action = rule->rr_action;
1038 copy->rr_amount = rule->rr_amount;
1039 refcount_init(©->rr_refcount, 1);
1040 rctl_rule_acquire_subject(copy);
1046 rctl_rule_acquire(struct rctl_rule *rule)
1049 ASSERT_RACCT_ENABLED();
1050 KASSERT(rule->rr_refcount > 0, ("rule->rr_refcount <= 0"));
1052 refcount_acquire(&rule->rr_refcount);
1056 rctl_rule_free(void *context, int pending)
1058 struct rctl_rule *rule;
1060 rule = (struct rctl_rule *)context;
1062 ASSERT_RACCT_ENABLED();
1063 KASSERT(rule->rr_refcount == 0, ("rule->rr_refcount != 0"));
1066 * We don't need locking here; rule is guaranteed to be inaccessible.
1069 rctl_rule_release_subject(rule);
1070 uma_zfree(rctl_rule_zone, rule);
1074 rctl_rule_release(struct rctl_rule *rule)
1077 ASSERT_RACCT_ENABLED();
1078 KASSERT(rule->rr_refcount > 0, ("rule->rr_refcount <= 0"));
1080 if (refcount_release(&rule->rr_refcount)) {
1082 * rctl_rule_release() is often called when iterating
1083 * over all the uidinfo structures in the system,
1084 * holding uihashtbl_lock. Since rctl_rule_free()
1085 * might end up calling uifree(), this would lead
1086 * to lock recursion. Use taskqueue to avoid this.
1088 TASK_INIT(&rule->rr_task, 0, rctl_rule_free, rule);
1089 taskqueue_enqueue(taskqueue_thread, &rule->rr_task);
1094 rctl_rule_fully_specified(const struct rctl_rule *rule)
1097 ASSERT_RACCT_ENABLED();
1099 switch (rule->rr_subject_type) {
1100 case RCTL_SUBJECT_TYPE_UNDEFINED:
1102 case RCTL_SUBJECT_TYPE_PROCESS:
1103 if (rule->rr_subject.rs_proc == NULL)
1106 case RCTL_SUBJECT_TYPE_USER:
1107 if (rule->rr_subject.rs_uip == NULL)
1110 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1111 if (rule->rr_subject.rs_loginclass == NULL)
1114 case RCTL_SUBJECT_TYPE_JAIL:
1115 if (rule->rr_subject.rs_prison_racct == NULL)
1119 panic("rctl_rule_fully_specified: unknown subject type %d",
1120 rule->rr_subject_type);
1122 if (rule->rr_resource == RACCT_UNDEFINED)
1124 if (rule->rr_action == RCTL_ACTION_UNDEFINED)
1126 if (rule->rr_amount == RCTL_AMOUNT_UNDEFINED)
1128 if (rule->rr_per == RCTL_SUBJECT_TYPE_UNDEFINED)
1135 rctl_string_to_rule(char *rulestr, struct rctl_rule **rulep)
1137 struct rctl_rule *rule;
1138 char *subjectstr, *subject_idstr, *resourcestr, *actionstr,
1139 *amountstr, *perstr;
1143 ASSERT_RACCT_ENABLED();
1145 rule = rctl_rule_alloc(M_WAITOK);
1147 subjectstr = strsep(&rulestr, ":");
1148 subject_idstr = strsep(&rulestr, ":");
1149 resourcestr = strsep(&rulestr, ":");
1150 actionstr = strsep(&rulestr, "=/");
1151 amountstr = strsep(&rulestr, "/");
1154 if (subjectstr == NULL || subjectstr[0] == '\0')
1155 rule->rr_subject_type = RCTL_SUBJECT_TYPE_UNDEFINED;
1157 error = str2value(subjectstr, &rule->rr_subject_type, subjectnames);
1162 if (subject_idstr == NULL || subject_idstr[0] == '\0') {
1163 rule->rr_subject.rs_proc = NULL;
1164 rule->rr_subject.rs_uip = NULL;
1165 rule->rr_subject.rs_loginclass = NULL;
1166 rule->rr_subject.rs_prison_racct = NULL;
1168 switch (rule->rr_subject_type) {
1169 case RCTL_SUBJECT_TYPE_UNDEFINED:
1172 case RCTL_SUBJECT_TYPE_PROCESS:
1173 error = str2id(subject_idstr, &id);
1176 sx_assert(&allproc_lock, SA_LOCKED);
1177 rule->rr_subject.rs_proc = pfind(id);
1178 if (rule->rr_subject.rs_proc == NULL) {
1182 PROC_UNLOCK(rule->rr_subject.rs_proc);
1184 case RCTL_SUBJECT_TYPE_USER:
1185 error = str2id(subject_idstr, &id);
1188 rule->rr_subject.rs_uip = uifind(id);
1190 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1191 rule->rr_subject.rs_loginclass =
1192 loginclass_find(subject_idstr);
1193 if (rule->rr_subject.rs_loginclass == NULL) {
1194 error = ENAMETOOLONG;
1198 case RCTL_SUBJECT_TYPE_JAIL:
1199 rule->rr_subject.rs_prison_racct =
1200 prison_racct_find(subject_idstr);
1201 if (rule->rr_subject.rs_prison_racct == NULL) {
1202 error = ENAMETOOLONG;
1207 panic("rctl_string_to_rule: unknown subject type %d",
1208 rule->rr_subject_type);
1212 if (resourcestr == NULL || resourcestr[0] == '\0')
1213 rule->rr_resource = RACCT_UNDEFINED;
1215 error = str2value(resourcestr, &rule->rr_resource,
1221 if (actionstr == NULL || actionstr[0] == '\0')
1222 rule->rr_action = RCTL_ACTION_UNDEFINED;
1224 error = str2value(actionstr, &rule->rr_action, actionnames);
1229 if (amountstr == NULL || amountstr[0] == '\0')
1230 rule->rr_amount = RCTL_AMOUNT_UNDEFINED;
1232 error = str2int64(amountstr, &rule->rr_amount);
1235 if (RACCT_IS_IN_MILLIONS(rule->rr_resource)) {
1236 if (rule->rr_amount > INT64_MAX / 1000000) {
1240 rule->rr_amount *= 1000000;
1244 if (perstr == NULL || perstr[0] == '\0')
1245 rule->rr_per = RCTL_SUBJECT_TYPE_UNDEFINED;
1247 error = str2value(perstr, &rule->rr_per, subjectnames);
1256 rctl_rule_release(rule);
1262 * Link a rule with all the subjects it applies to.
1265 rctl_rule_add(struct rctl_rule *rule)
1269 struct uidinfo *uip;
1271 struct prison_racct *prr;
1272 struct loginclass *lc;
1273 struct rctl_rule *rule2;
1276 ASSERT_RACCT_ENABLED();
1277 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified"));
1280 * Some rules just don't make sense, like "deny" rule for an undeniable
1281 * resource. The exception are the RSS and %CPU resources - they are
1282 * not deniable in the racct sense, but the limit is enforced in
1285 if (rule->rr_action == RCTL_ACTION_DENY &&
1286 !RACCT_IS_DENIABLE(rule->rr_resource) &&
1287 rule->rr_resource != RACCT_RSS &&
1288 rule->rr_resource != RACCT_PCTCPU) {
1289 return (EOPNOTSUPP);
1292 if (rule->rr_action == RCTL_ACTION_THROTTLE &&
1293 !RACCT_IS_DECAYING(rule->rr_resource)) {
1294 return (EOPNOTSUPP);
1297 if (rule->rr_action == RCTL_ACTION_THROTTLE &&
1298 rule->rr_resource == RACCT_PCTCPU) {
1299 return (EOPNOTSUPP);
1302 if (rule->rr_per == RCTL_SUBJECT_TYPE_PROCESS &&
1303 RACCT_IS_SLOPPY(rule->rr_resource)) {
1304 return (EOPNOTSUPP);
1308 * Make sure there are no duplicated rules. Also, for the "deny"
1309 * rules, remove ones differing only by "amount".
1311 if (rule->rr_action == RCTL_ACTION_DENY) {
1312 rule2 = rctl_rule_duplicate(rule, M_WAITOK);
1313 rule2->rr_amount = RCTL_AMOUNT_UNDEFINED;
1314 rctl_rule_remove(rule2);
1315 rctl_rule_release(rule2);
1317 rctl_rule_remove(rule);
1319 switch (rule->rr_subject_type) {
1320 case RCTL_SUBJECT_TYPE_PROCESS:
1321 p = rule->rr_subject.rs_proc;
1322 KASSERT(p != NULL, ("rctl_rule_add: NULL proc"));
1324 rctl_racct_add_rule(p->p_racct, rule);
1326 * In case of per-process rule, we don't have anything more
1331 case RCTL_SUBJECT_TYPE_USER:
1332 uip = rule->rr_subject.rs_uip;
1333 KASSERT(uip != NULL, ("rctl_rule_add: NULL uip"));
1334 rctl_racct_add_rule(uip->ui_racct, rule);
1337 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1338 lc = rule->rr_subject.rs_loginclass;
1339 KASSERT(lc != NULL, ("rctl_rule_add: NULL loginclass"));
1340 rctl_racct_add_rule(lc->lc_racct, rule);
1343 case RCTL_SUBJECT_TYPE_JAIL:
1344 prr = rule->rr_subject.rs_prison_racct;
1345 KASSERT(prr != NULL, ("rctl_rule_add: NULL pr"));
1346 rctl_racct_add_rule(prr->prr_racct, rule);
1350 panic("rctl_rule_add: unknown subject type %d",
1351 rule->rr_subject_type);
1355 * Now go through all the processes and add the new rule to the ones
1358 sx_assert(&allproc_lock, SA_LOCKED);
1359 FOREACH_PROC_IN_SYSTEM(p) {
1361 switch (rule->rr_subject_type) {
1362 case RCTL_SUBJECT_TYPE_USER:
1363 if (cred->cr_uidinfo == rule->rr_subject.rs_uip ||
1364 cred->cr_ruidinfo == rule->rr_subject.rs_uip)
1367 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1368 if (cred->cr_loginclass == rule->rr_subject.rs_loginclass)
1371 case RCTL_SUBJECT_TYPE_JAIL:
1373 for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) {
1374 if (pr->pr_prison_racct == rule->rr_subject.rs_prison_racct) {
1383 panic("rctl_rule_add: unknown subject type %d",
1384 rule->rr_subject_type);
1387 rctl_racct_add_rule(p->p_racct, rule);
1394 rctl_rule_pre_callback(void)
1401 rctl_rule_post_callback(void)
1408 rctl_rule_remove_callback(struct racct *racct, void *arg2, void *arg3)
1410 struct rctl_rule *filter = (struct rctl_rule *)arg2;
1413 ASSERT_RACCT_ENABLED();
1414 RACCT_LOCK_ASSERT();
1416 found += rctl_racct_remove_rules(racct, filter);
1418 *((int *)arg3) += found;
1422 * Remove all rules that match the filter.
1425 rctl_rule_remove(struct rctl_rule *filter)
1430 ASSERT_RACCT_ENABLED();
1432 if (filter->rr_subject_type == RCTL_SUBJECT_TYPE_PROCESS &&
1433 filter->rr_subject.rs_proc != NULL) {
1434 p = filter->rr_subject.rs_proc;
1436 found = rctl_racct_remove_rules(p->p_racct, filter);
1443 loginclass_racct_foreach(rctl_rule_remove_callback,
1444 rctl_rule_pre_callback, rctl_rule_post_callback,
1445 filter, (void *)&found);
1446 ui_racct_foreach(rctl_rule_remove_callback,
1447 rctl_rule_pre_callback, rctl_rule_post_callback,
1448 filter, (void *)&found);
1449 prison_racct_foreach(rctl_rule_remove_callback,
1450 rctl_rule_pre_callback, rctl_rule_post_callback,
1451 filter, (void *)&found);
1453 sx_assert(&allproc_lock, SA_LOCKED);
1455 FOREACH_PROC_IN_SYSTEM(p) {
1456 found += rctl_racct_remove_rules(p->p_racct, filter);
1466 * Appends a rule to the sbuf.
1469 rctl_rule_to_sbuf(struct sbuf *sb, const struct rctl_rule *rule)
1473 ASSERT_RACCT_ENABLED();
1475 sbuf_printf(sb, "%s:", rctl_subject_type_name(rule->rr_subject_type));
1477 switch (rule->rr_subject_type) {
1478 case RCTL_SUBJECT_TYPE_PROCESS:
1479 if (rule->rr_subject.rs_proc == NULL)
1480 sbuf_printf(sb, ":");
1482 sbuf_printf(sb, "%d:",
1483 rule->rr_subject.rs_proc->p_pid);
1485 case RCTL_SUBJECT_TYPE_USER:
1486 if (rule->rr_subject.rs_uip == NULL)
1487 sbuf_printf(sb, ":");
1489 sbuf_printf(sb, "%d:",
1490 rule->rr_subject.rs_uip->ui_uid);
1492 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1493 if (rule->rr_subject.rs_loginclass == NULL)
1494 sbuf_printf(sb, ":");
1496 sbuf_printf(sb, "%s:",
1497 rule->rr_subject.rs_loginclass->lc_name);
1499 case RCTL_SUBJECT_TYPE_JAIL:
1500 if (rule->rr_subject.rs_prison_racct == NULL)
1501 sbuf_printf(sb, ":");
1503 sbuf_printf(sb, "%s:",
1504 rule->rr_subject.rs_prison_racct->prr_name);
1507 panic("rctl_rule_to_sbuf: unknown subject type %d",
1508 rule->rr_subject_type);
1511 amount = rule->rr_amount;
1512 if (amount != RCTL_AMOUNT_UNDEFINED &&
1513 RACCT_IS_IN_MILLIONS(rule->rr_resource))
1516 sbuf_printf(sb, "%s:%s=%jd",
1517 rctl_resource_name(rule->rr_resource),
1518 rctl_action_name(rule->rr_action),
1521 if (rule->rr_per != rule->rr_subject_type)
1522 sbuf_printf(sb, "/%s", rctl_subject_type_name(rule->rr_per));
1526 * Routine used by RCTL syscalls to read in input string.
1529 rctl_read_inbuf(char **inputstr, const char *inbufp, size_t inbuflen)
1534 ASSERT_RACCT_ENABLED();
1538 if (inbuflen > RCTL_MAX_INBUFSIZE)
1541 str = malloc(inbuflen + 1, M_RCTL, M_WAITOK);
1542 error = copyinstr(inbufp, str, inbuflen, NULL);
1554 * Routine used by RCTL syscalls to write out output string.
1557 rctl_write_outbuf(struct sbuf *outputsbuf, char *outbufp, size_t outbuflen)
1561 ASSERT_RACCT_ENABLED();
1563 if (outputsbuf == NULL)
1566 sbuf_finish(outputsbuf);
1567 if (outbuflen < sbuf_len(outputsbuf) + 1) {
1568 sbuf_delete(outputsbuf);
1571 error = copyout(sbuf_data(outputsbuf), outbufp,
1572 sbuf_len(outputsbuf) + 1);
1573 sbuf_delete(outputsbuf);
1577 static struct sbuf *
1578 rctl_racct_to_sbuf(struct racct *racct, int sloppy)
1584 ASSERT_RACCT_ENABLED();
1586 sb = sbuf_new_auto();
1587 for (i = 0; i <= RACCT_MAX; i++) {
1588 if (sloppy == 0 && RACCT_IS_SLOPPY(i))
1591 amount = racct->r_resources[i];
1593 if (RACCT_IS_IN_MILLIONS(i))
1595 sbuf_printf(sb, "%s=%jd,", rctl_resource_name(i), amount);
1597 sbuf_setpos(sb, sbuf_len(sb) - 1);
1602 sys_rctl_get_racct(struct thread *td, struct rctl_get_racct_args *uap)
1604 struct rctl_rule *filter;
1605 struct sbuf *outputsbuf = NULL;
1607 struct uidinfo *uip;
1608 struct loginclass *lc;
1609 struct prison_racct *prr;
1616 error = priv_check(td, PRIV_RCTL_GET_RACCT);
1620 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1624 sx_slock(&allproc_lock);
1625 error = rctl_string_to_rule(inputstr, &filter);
1626 free(inputstr, M_RCTL);
1628 sx_sunlock(&allproc_lock);
1632 switch (filter->rr_subject_type) {
1633 case RCTL_SUBJECT_TYPE_PROCESS:
1634 p = filter->rr_subject.rs_proc;
1639 outputsbuf = rctl_racct_to_sbuf(p->p_racct, 0);
1641 case RCTL_SUBJECT_TYPE_USER:
1642 uip = filter->rr_subject.rs_uip;
1647 outputsbuf = rctl_racct_to_sbuf(uip->ui_racct, 1);
1649 case RCTL_SUBJECT_TYPE_LOGINCLASS:
1650 lc = filter->rr_subject.rs_loginclass;
1655 outputsbuf = rctl_racct_to_sbuf(lc->lc_racct, 1);
1657 case RCTL_SUBJECT_TYPE_JAIL:
1658 prr = filter->rr_subject.rs_prison_racct;
1663 outputsbuf = rctl_racct_to_sbuf(prr->prr_racct, 1);
1669 rctl_rule_release(filter);
1670 sx_sunlock(&allproc_lock);
1674 error = rctl_write_outbuf(outputsbuf, uap->outbufp, uap->outbuflen);
1680 rctl_get_rules_callback(struct racct *racct, void *arg2, void *arg3)
1682 struct rctl_rule *filter = (struct rctl_rule *)arg2;
1683 struct rctl_rule_link *link;
1684 struct sbuf *sb = (struct sbuf *)arg3;
1686 ASSERT_RACCT_ENABLED();
1687 RACCT_LOCK_ASSERT();
1689 LIST_FOREACH(link, &racct->r_rule_links, rrl_next) {
1690 if (!rctl_rule_matches(link->rrl_rule, filter))
1692 rctl_rule_to_sbuf(sb, link->rrl_rule);
1693 sbuf_printf(sb, ",");
1698 sys_rctl_get_rules(struct thread *td, struct rctl_get_rules_args *uap)
1701 struct rctl_rule *filter;
1702 struct rctl_rule_link *link;
1704 char *inputstr, *buf;
1711 error = priv_check(td, PRIV_RCTL_GET_RULES);
1715 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1719 sx_slock(&allproc_lock);
1720 error = rctl_string_to_rule(inputstr, &filter);
1721 free(inputstr, M_RCTL);
1723 sx_sunlock(&allproc_lock);
1727 bufsize = uap->outbuflen;
1728 if (bufsize > rctl_maxbufsize) {
1729 sx_sunlock(&allproc_lock);
1733 buf = malloc(bufsize, M_RCTL, M_WAITOK);
1734 sb = sbuf_new(NULL, buf, bufsize, SBUF_FIXEDLEN);
1735 KASSERT(sb != NULL, ("sbuf_new failed"));
1737 FOREACH_PROC_IN_SYSTEM(p) {
1739 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
1741 * Non-process rules will be added to the buffer later.
1742 * Adding them here would result in duplicated output.
1744 if (link->rrl_rule->rr_subject_type !=
1745 RCTL_SUBJECT_TYPE_PROCESS)
1747 if (!rctl_rule_matches(link->rrl_rule, filter))
1749 rctl_rule_to_sbuf(sb, link->rrl_rule);
1750 sbuf_printf(sb, ",");
1755 loginclass_racct_foreach(rctl_get_rules_callback,
1756 rctl_rule_pre_callback, rctl_rule_post_callback,
1758 ui_racct_foreach(rctl_get_rules_callback,
1759 rctl_rule_pre_callback, rctl_rule_post_callback,
1761 prison_racct_foreach(rctl_get_rules_callback,
1762 rctl_rule_pre_callback, rctl_rule_post_callback,
1764 if (sbuf_error(sb) == ENOMEM) {
1770 * Remove trailing ",".
1772 if (sbuf_len(sb) > 0)
1773 sbuf_setpos(sb, sbuf_len(sb) - 1);
1775 error = rctl_write_outbuf(sb, uap->outbufp, uap->outbuflen);
1777 rctl_rule_release(filter);
1778 sx_sunlock(&allproc_lock);
1784 sys_rctl_get_limits(struct thread *td, struct rctl_get_limits_args *uap)
1787 struct rctl_rule *filter;
1788 struct rctl_rule_link *link;
1789 char *inputstr, *buf;
1796 error = priv_check(td, PRIV_RCTL_GET_LIMITS);
1800 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1804 sx_slock(&allproc_lock);
1805 error = rctl_string_to_rule(inputstr, &filter);
1806 free(inputstr, M_RCTL);
1808 sx_sunlock(&allproc_lock);
1812 if (filter->rr_subject_type == RCTL_SUBJECT_TYPE_UNDEFINED) {
1813 rctl_rule_release(filter);
1814 sx_sunlock(&allproc_lock);
1817 if (filter->rr_subject_type != RCTL_SUBJECT_TYPE_PROCESS) {
1818 rctl_rule_release(filter);
1819 sx_sunlock(&allproc_lock);
1820 return (EOPNOTSUPP);
1822 if (filter->rr_subject.rs_proc == NULL) {
1823 rctl_rule_release(filter);
1824 sx_sunlock(&allproc_lock);
1828 bufsize = uap->outbuflen;
1829 if (bufsize > rctl_maxbufsize) {
1830 rctl_rule_release(filter);
1831 sx_sunlock(&allproc_lock);
1835 buf = malloc(bufsize, M_RCTL, M_WAITOK);
1836 sb = sbuf_new(NULL, buf, bufsize, SBUF_FIXEDLEN);
1837 KASSERT(sb != NULL, ("sbuf_new failed"));
1840 LIST_FOREACH(link, &filter->rr_subject.rs_proc->p_racct->r_rule_links,
1842 rctl_rule_to_sbuf(sb, link->rrl_rule);
1843 sbuf_printf(sb, ",");
1846 if (sbuf_error(sb) == ENOMEM) {
1853 * Remove trailing ",".
1855 if (sbuf_len(sb) > 0)
1856 sbuf_setpos(sb, sbuf_len(sb) - 1);
1858 error = rctl_write_outbuf(sb, uap->outbufp, uap->outbuflen);
1860 rctl_rule_release(filter);
1861 sx_sunlock(&allproc_lock);
1867 sys_rctl_add_rule(struct thread *td, struct rctl_add_rule_args *uap)
1869 struct rctl_rule *rule;
1876 error = priv_check(td, PRIV_RCTL_ADD_RULE);
1880 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1884 sx_slock(&allproc_lock);
1885 error = rctl_string_to_rule(inputstr, &rule);
1886 free(inputstr, M_RCTL);
1888 sx_sunlock(&allproc_lock);
1892 * The 'per' part of a rule is optional.
1894 if (rule->rr_per == RCTL_SUBJECT_TYPE_UNDEFINED &&
1895 rule->rr_subject_type != RCTL_SUBJECT_TYPE_UNDEFINED)
1896 rule->rr_per = rule->rr_subject_type;
1898 if (!rctl_rule_fully_specified(rule)) {
1903 error = rctl_rule_add(rule);
1906 rctl_rule_release(rule);
1907 sx_sunlock(&allproc_lock);
1912 sys_rctl_remove_rule(struct thread *td, struct rctl_remove_rule_args *uap)
1914 struct rctl_rule *filter;
1921 error = priv_check(td, PRIV_RCTL_REMOVE_RULE);
1925 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen);
1929 sx_slock(&allproc_lock);
1930 error = rctl_string_to_rule(inputstr, &filter);
1931 free(inputstr, M_RCTL);
1933 sx_sunlock(&allproc_lock);
1937 error = rctl_rule_remove(filter);
1938 rctl_rule_release(filter);
1939 sx_sunlock(&allproc_lock);
1945 * Update RCTL rule list after credential change.
1948 rctl_proc_ucred_changed(struct proc *p, struct ucred *newcred)
1950 LIST_HEAD(, rctl_rule_link) newrules;
1951 struct rctl_rule_link *link, *newlink;
1952 struct uidinfo *newuip;
1953 struct loginclass *newlc;
1954 struct prison_racct *newprr;
1957 ASSERT_RACCT_ENABLED();
1959 newuip = newcred->cr_ruidinfo;
1960 newlc = newcred->cr_loginclass;
1961 newprr = newcred->cr_prison->pr_prison_racct;
1963 LIST_INIT(&newrules);
1967 * First, count the rules that apply to the process with new
1972 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
1973 if (link->rrl_rule->rr_subject_type ==
1974 RCTL_SUBJECT_TYPE_PROCESS)
1977 LIST_FOREACH(link, &newuip->ui_racct->r_rule_links, rrl_next)
1979 LIST_FOREACH(link, &newlc->lc_racct->r_rule_links, rrl_next)
1981 LIST_FOREACH(link, &newprr->prr_racct->r_rule_links, rrl_next)
1986 * Create temporary list. We've dropped the rctl_lock in order
1989 for (i = 0; i < rulecnt; i++) {
1990 newlink = uma_zalloc(rctl_rule_link_zone, M_WAITOK);
1991 newlink->rrl_rule = NULL;
1992 newlink->rrl_exceeded = 0;
1993 LIST_INSERT_HEAD(&newrules, newlink, rrl_next);
1996 newlink = LIST_FIRST(&newrules);
1999 * Assign rules to the newly allocated list entries.
2002 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) {
2003 if (link->rrl_rule->rr_subject_type ==
2004 RCTL_SUBJECT_TYPE_PROCESS) {
2005 if (newlink == NULL)
2007 rctl_rule_acquire(link->rrl_rule);
2008 newlink->rrl_rule = link->rrl_rule;
2009 newlink->rrl_exceeded = link->rrl_exceeded;
2010 newlink = LIST_NEXT(newlink, rrl_next);
2015 LIST_FOREACH(link, &newuip->ui_racct->r_rule_links, rrl_next) {
2016 if (newlink == NULL)
2018 rctl_rule_acquire(link->rrl_rule);
2019 newlink->rrl_rule = link->rrl_rule;
2020 newlink->rrl_exceeded = link->rrl_exceeded;
2021 newlink = LIST_NEXT(newlink, rrl_next);
2025 LIST_FOREACH(link, &newlc->lc_racct->r_rule_links, rrl_next) {
2026 if (newlink == NULL)
2028 rctl_rule_acquire(link->rrl_rule);
2029 newlink->rrl_rule = link->rrl_rule;
2030 newlink->rrl_exceeded = link->rrl_exceeded;
2031 newlink = LIST_NEXT(newlink, rrl_next);
2035 LIST_FOREACH(link, &newprr->prr_racct->r_rule_links, rrl_next) {
2036 if (newlink == NULL)
2038 rctl_rule_acquire(link->rrl_rule);
2039 newlink->rrl_rule = link->rrl_rule;
2040 newlink->rrl_exceeded = link->rrl_exceeded;
2041 newlink = LIST_NEXT(newlink, rrl_next);
2047 * Free the old rule list.
2049 while (!LIST_EMPTY(&p->p_racct->r_rule_links)) {
2050 link = LIST_FIRST(&p->p_racct->r_rule_links);
2051 LIST_REMOVE(link, rrl_next);
2052 rctl_rule_release(link->rrl_rule);
2053 uma_zfree(rctl_rule_link_zone, link);
2057 * Replace lists and we're done.
2059 * XXX: Is there any way to switch list heads instead
2060 * of iterating here?
2062 while (!LIST_EMPTY(&newrules)) {
2063 newlink = LIST_FIRST(&newrules);
2064 LIST_REMOVE(newlink, rrl_next);
2065 LIST_INSERT_HEAD(&p->p_racct->r_rule_links,
2078 * Rule list changed while we were not holding the rctl_lock.
2079 * Free the new list and try again.
2081 while (!LIST_EMPTY(&newrules)) {
2082 newlink = LIST_FIRST(&newrules);
2083 LIST_REMOVE(newlink, rrl_next);
2084 if (newlink->rrl_rule != NULL)
2085 rctl_rule_release(newlink->rrl_rule);
2086 uma_zfree(rctl_rule_link_zone, newlink);
2093 * Assign RCTL rules to the newly created process.
2096 rctl_proc_fork(struct proc *parent, struct proc *child)
2098 struct rctl_rule *rule;
2099 struct rctl_rule_link *link;
2102 ASSERT_RACCT_ENABLED();
2103 RACCT_LOCK_ASSERT();
2104 KASSERT(parent->p_racct != NULL, ("process without racct; p = %p", parent));
2106 LIST_INIT(&child->p_racct->r_rule_links);
2109 * Go through limits applicable to the parent and assign them
2110 * to the child. Rules with 'process' subject have to be duplicated
2111 * in order to make their rr_subject point to the new process.
2113 LIST_FOREACH(link, &parent->p_racct->r_rule_links, rrl_next) {
2114 if (link->rrl_rule->rr_subject_type ==
2115 RCTL_SUBJECT_TYPE_PROCESS) {
2116 rule = rctl_rule_duplicate(link->rrl_rule, M_NOWAIT);
2119 KASSERT(rule->rr_subject.rs_proc == parent,
2120 ("rule->rr_subject.rs_proc != parent"));
2121 rule->rr_subject.rs_proc = child;
2122 error = rctl_racct_add_rule_locked(child->p_racct,
2124 rctl_rule_release(rule);
2128 error = rctl_racct_add_rule_locked(child->p_racct,
2138 while (!LIST_EMPTY(&child->p_racct->r_rule_links)) {
2139 link = LIST_FIRST(&child->p_racct->r_rule_links);
2140 LIST_REMOVE(link, rrl_next);
2141 rctl_rule_release(link->rrl_rule);
2142 uma_zfree(rctl_rule_link_zone, link);
2149 * Release rules attached to the racct.
2152 rctl_racct_release(struct racct *racct)
2154 struct rctl_rule_link *link;
2156 ASSERT_RACCT_ENABLED();
2157 RACCT_LOCK_ASSERT();
2159 while (!LIST_EMPTY(&racct->r_rule_links)) {
2160 link = LIST_FIRST(&racct->r_rule_links);
2161 LIST_REMOVE(link, rrl_next);
2162 rctl_rule_release(link->rrl_rule);
2163 uma_zfree(rctl_rule_link_zone, link);
2174 rctl_rule_zone = uma_zcreate("rctl_rule", sizeof(struct rctl_rule),
2175 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, 0);
2176 rctl_rule_link_zone = uma_zcreate("rctl_rule_link",
2177 sizeof(struct rctl_rule_link), NULL, NULL, NULL, NULL,
2181 * Set default values, making sure not to overwrite the ones
2182 * fetched from tunables. Most of those could be set at the
2183 * declaration, except for the rctl_throttle_max - we cannot
2184 * set it there due to hz not being compile time constant.
2186 if (rctl_throttle_min < 1)
2187 rctl_throttle_min = 1;
2188 if (rctl_throttle_max < rctl_throttle_min)
2189 rctl_throttle_max = 2 * hz;
2190 if (rctl_throttle_pct < 0)
2191 rctl_throttle_pct = 100;
2192 if (rctl_throttle_pct2 < 0)
2193 rctl_throttle_pct2 = 100;
2199 sys_rctl_get_racct(struct thread *td, struct rctl_get_racct_args *uap)
2206 sys_rctl_get_rules(struct thread *td, struct rctl_get_rules_args *uap)
2213 sys_rctl_get_limits(struct thread *td, struct rctl_get_limits_args *uap)
2220 sys_rctl_add_rule(struct thread *td, struct rctl_add_rule_args *uap)
2227 sys_rctl_remove_rule(struct thread *td, struct rctl_remove_rule_args *uap)