kern_rctl.c revision 290857
1/*- 2 * Copyright (c) 2010 The FreeBSD Foundation 3 * All rights reserved. 4 * 5 * This software was developed by Edward Tomasz Napierala under sponsorship 6 * from the FreeBSD Foundation. 7 * 8 * Redistribution and use in source and binary forms, with or without 9 * modification, are permitted provided that the following conditions 10 * are met: 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. 16 * 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 27 * SUCH DAMAGE. 28 * 29 * $FreeBSD: head/sys/kern/kern_rctl.c 290857 2015-11-15 12:10:51Z trasz $ 30 */ 31 32#include <sys/cdefs.h> 33__FBSDID("$FreeBSD: head/sys/kern/kern_rctl.c 290857 2015-11-15 12:10:51Z trasz $"); 34 35#include <sys/param.h> 36#include <sys/bus.h> 37#include <sys/malloc.h> 38#include <sys/queue.h> 39#include <sys/refcount.h> 40#include <sys/jail.h> 41#include <sys/kernel.h> 42#include <sys/limits.h> 43#include <sys/loginclass.h> 44#include <sys/priv.h> 45#include <sys/proc.h> 46#include <sys/racct.h> 47#include <sys/rctl.h> 48#include <sys/resourcevar.h> 49#include <sys/sx.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> 55#include <sys/lock.h> 56#include <sys/mutex.h> 57#include <sys/rwlock.h> 58#include <sys/sbuf.h> 59#include <sys/taskqueue.h> 60#include <sys/tree.h> 61#include <vm/uma.h> 62 63#ifdef RCTL 64#ifndef RACCT 65#error "The RCTL option requires the RACCT option" 66#endif 67 68FEATURE(rctl, "Resource Limits"); 69 70#define HRF_DEFAULT 0 71#define HRF_DONT_INHERIT 1 72#define HRF_DONT_ACCUMULATE 2 73 74#define RCTL_MAX_INBUFSIZE 4 * 1024 75#define RCTL_MAX_OUTBUFSIZE 16 * 1024 * 1024 76#define RCTL_LOG_BUFSIZE 128 77 78#define RCTL_PCPU_SHIFT (10 * 1000000) 79 80unsigned int rctl_maxbufsize = RCTL_MAX_OUTBUFSIZE; 81 82SYSCTL_NODE(_kern_racct, OID_AUTO, rctl, CTLFLAG_RW, 0, "Resource Limits"); 83SYSCTL_UINT(_kern_racct_rctl, OID_AUTO, maxbufsize, CTLFLAG_RWTUN, 84 &rctl_maxbufsize, 0, "Maximum output buffer size"); 85 86/* 87 * 'rctl_rule_link' connects a rule with every racct it's related to. 88 * For example, rule 'user:X:openfiles:deny=N/process' is linked 89 * with uidinfo for user X, and to each process of that user. 90 */ 91struct rctl_rule_link { 92 LIST_ENTRY(rctl_rule_link) rrl_next; 93 struct rctl_rule *rrl_rule; 94 int rrl_exceeded; 95}; 96 97struct dict { 98 const char *d_name; 99 int d_value; 100}; 101 102static struct dict subjectnames[] = { 103 { "process", RCTL_SUBJECT_TYPE_PROCESS }, 104 { "user", RCTL_SUBJECT_TYPE_USER }, 105 { "loginclass", RCTL_SUBJECT_TYPE_LOGINCLASS }, 106 { "jail", RCTL_SUBJECT_TYPE_JAIL }, 107 { NULL, -1 }}; 108 109static struct dict resourcenames[] = { 110 { "cputime", RACCT_CPU }, 111 { "datasize", RACCT_DATA }, 112 { "stacksize", RACCT_STACK }, 113 { "coredumpsize", RACCT_CORE }, 114 { "memoryuse", RACCT_RSS }, 115 { "memorylocked", RACCT_MEMLOCK }, 116 { "maxproc", RACCT_NPROC }, 117 { "openfiles", RACCT_NOFILE }, 118 { "vmemoryuse", RACCT_VMEM }, 119 { "pseudoterminals", RACCT_NPTS }, 120 { "swapuse", RACCT_SWAP }, 121 { "nthr", RACCT_NTHR }, 122 { "msgqqueued", RACCT_MSGQQUEUED }, 123 { "msgqsize", RACCT_MSGQSIZE }, 124 { "nmsgq", RACCT_NMSGQ }, 125 { "nsem", RACCT_NSEM }, 126 { "nsemop", RACCT_NSEMOP }, 127 { "nshm", RACCT_NSHM }, 128 { "shmsize", RACCT_SHMSIZE }, 129 { "wallclock", RACCT_WALLCLOCK }, 130 { "pcpu", RACCT_PCTCPU }, 131 { NULL, -1 }}; 132 133static struct dict actionnames[] = { 134 { "sighup", RCTL_ACTION_SIGHUP }, 135 { "sigint", RCTL_ACTION_SIGINT }, 136 { "sigquit", RCTL_ACTION_SIGQUIT }, 137 { "sigill", RCTL_ACTION_SIGILL }, 138 { "sigtrap", RCTL_ACTION_SIGTRAP }, 139 { "sigabrt", RCTL_ACTION_SIGABRT }, 140 { "sigemt", RCTL_ACTION_SIGEMT }, 141 { "sigfpe", RCTL_ACTION_SIGFPE }, 142 { "sigkill", RCTL_ACTION_SIGKILL }, 143 { "sigbus", RCTL_ACTION_SIGBUS }, 144 { "sigsegv", RCTL_ACTION_SIGSEGV }, 145 { "sigsys", RCTL_ACTION_SIGSYS }, 146 { "sigpipe", RCTL_ACTION_SIGPIPE }, 147 { "sigalrm", RCTL_ACTION_SIGALRM }, 148 { "sigterm", RCTL_ACTION_SIGTERM }, 149 { "sigurg", RCTL_ACTION_SIGURG }, 150 { "sigstop", RCTL_ACTION_SIGSTOP }, 151 { "sigtstp", RCTL_ACTION_SIGTSTP }, 152 { "sigchld", RCTL_ACTION_SIGCHLD }, 153 { "sigttin", RCTL_ACTION_SIGTTIN }, 154 { "sigttou", RCTL_ACTION_SIGTTOU }, 155 { "sigio", RCTL_ACTION_SIGIO }, 156 { "sigxcpu", RCTL_ACTION_SIGXCPU }, 157 { "sigxfsz", RCTL_ACTION_SIGXFSZ }, 158 { "sigvtalrm", RCTL_ACTION_SIGVTALRM }, 159 { "sigprof", RCTL_ACTION_SIGPROF }, 160 { "sigwinch", RCTL_ACTION_SIGWINCH }, 161 { "siginfo", RCTL_ACTION_SIGINFO }, 162 { "sigusr1", RCTL_ACTION_SIGUSR1 }, 163 { "sigusr2", RCTL_ACTION_SIGUSR2 }, 164 { "sigthr", RCTL_ACTION_SIGTHR }, 165 { "deny", RCTL_ACTION_DENY }, 166 { "log", RCTL_ACTION_LOG }, 167 { "devctl", RCTL_ACTION_DEVCTL }, 168 { NULL, -1 }}; 169 170static void rctl_init(void); 171SYSINIT(rctl, SI_SUB_RACCT, SI_ORDER_FIRST, rctl_init, NULL); 172 173static uma_zone_t rctl_rule_link_zone; 174static uma_zone_t rctl_rule_zone; 175static struct rwlock rctl_lock; 176RW_SYSINIT(rctl_lock, &rctl_lock, "RCTL lock"); 177 178static int rctl_rule_fully_specified(const struct rctl_rule *rule); 179static void rctl_rule_to_sbuf(struct sbuf *sb, const struct rctl_rule *rule); 180 181static MALLOC_DEFINE(M_RCTL, "rctl", "Resource Limits"); 182 183static const char * 184rctl_subject_type_name(int subject) 185{ 186 int i; 187 188 for (i = 0; subjectnames[i].d_name != NULL; i++) { 189 if (subjectnames[i].d_value == subject) 190 return (subjectnames[i].d_name); 191 } 192 193 panic("rctl_subject_type_name: unknown subject type %d", subject); 194} 195 196static const char * 197rctl_action_name(int action) 198{ 199 int i; 200 201 for (i = 0; actionnames[i].d_name != NULL; i++) { 202 if (actionnames[i].d_value == action) 203 return (actionnames[i].d_name); 204 } 205 206 panic("rctl_action_name: unknown action %d", action); 207} 208 209const char * 210rctl_resource_name(int resource) 211{ 212 int i; 213 214 for (i = 0; resourcenames[i].d_name != NULL; i++) { 215 if (resourcenames[i].d_value == resource) 216 return (resourcenames[i].d_name); 217 } 218 219 panic("rctl_resource_name: unknown resource %d", resource); 220} 221 222/* 223 * Return the amount of resource that can be allocated by 'p' before 224 * hitting 'rule'. 225 */ 226static int64_t 227rctl_available_resource(const struct proc *p, const struct rctl_rule *rule) 228{ 229 int resource; 230 int64_t available = INT64_MAX; 231 struct ucred *cred = p->p_ucred; 232 233 ASSERT_RACCT_ENABLED(); 234 rw_assert(&rctl_lock, RA_LOCKED); 235 236 resource = rule->rr_resource; 237 switch (rule->rr_per) { 238 case RCTL_SUBJECT_TYPE_PROCESS: 239 available = rule->rr_amount - 240 p->p_racct->r_resources[resource]; 241 break; 242 case RCTL_SUBJECT_TYPE_USER: 243 available = rule->rr_amount - 244 cred->cr_ruidinfo->ui_racct->r_resources[resource]; 245 break; 246 case RCTL_SUBJECT_TYPE_LOGINCLASS: 247 available = rule->rr_amount - 248 cred->cr_loginclass->lc_racct->r_resources[resource]; 249 break; 250 case RCTL_SUBJECT_TYPE_JAIL: 251 available = rule->rr_amount - 252 cred->cr_prison->pr_prison_racct->prr_racct-> 253 r_resources[resource]; 254 break; 255 default: 256 panic("rctl_compute_available: unknown per %d", 257 rule->rr_per); 258 } 259 260 return (available); 261} 262 263/* 264 * Return non-zero if allocating 'amount' by proc 'p' would exceed 265 * resource limit specified by 'rule'. 266 */ 267static int 268rctl_would_exceed(const struct proc *p, const struct rctl_rule *rule, 269 int64_t amount) 270{ 271 int64_t available; 272 273 ASSERT_RACCT_ENABLED(); 274 275 rw_assert(&rctl_lock, RA_LOCKED); 276 277 available = rctl_available_resource(p, rule); 278 if (available >= amount) 279 return (0); 280 281 return (1); 282} 283 284/* 285 * Special version of rctl_available() function for the %cpu resource. 286 * We slightly cheat here and return less than we normally would. 287 */ 288int64_t 289rctl_pcpu_available(const struct proc *p) { 290 struct rctl_rule *rule; 291 struct rctl_rule_link *link; 292 int64_t available, minavailable, limit; 293 294 ASSERT_RACCT_ENABLED(); 295 296 minavailable = INT64_MAX; 297 limit = 0; 298 299 rw_rlock(&rctl_lock); 300 301 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 302 rule = link->rrl_rule; 303 if (rule->rr_resource != RACCT_PCTCPU) 304 continue; 305 if (rule->rr_action != RCTL_ACTION_DENY) 306 continue; 307 available = rctl_available_resource(p, rule); 308 if (available < minavailable) { 309 minavailable = available; 310 limit = rule->rr_amount; 311 } 312 } 313 314 rw_runlock(&rctl_lock); 315 316 /* 317 * Return slightly less than actual value of the available 318 * %cpu resource. This makes %cpu throttling more agressive 319 * and lets us act sooner than the limits are already exceeded. 320 */ 321 if (limit != 0) { 322 if (limit > 2 * RCTL_PCPU_SHIFT) 323 minavailable -= RCTL_PCPU_SHIFT; 324 else 325 minavailable -= (limit / 2); 326 } 327 328 return (minavailable); 329} 330 331/* 332 * Check whether the proc 'p' can allocate 'amount' of 'resource' in addition 333 * to what it keeps allocated now. Returns non-zero if the allocation should 334 * be denied, 0 otherwise. 335 */ 336int 337rctl_enforce(struct proc *p, int resource, uint64_t amount) 338{ 339 struct rctl_rule *rule; 340 struct rctl_rule_link *link; 341 struct sbuf sb; 342 int should_deny = 0; 343 char *buf; 344 static int curtime = 0; 345 static struct timeval lasttime; 346 347 ASSERT_RACCT_ENABLED(); 348 349 rw_rlock(&rctl_lock); 350 351 /* 352 * There may be more than one matching rule; go through all of them. 353 * Denial should be done last, after logging and sending signals. 354 */ 355 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 356 rule = link->rrl_rule; 357 if (rule->rr_resource != resource) 358 continue; 359 if (!rctl_would_exceed(p, rule, amount)) { 360 link->rrl_exceeded = 0; 361 continue; 362 } 363 364 switch (rule->rr_action) { 365 case RCTL_ACTION_DENY: 366 should_deny = 1; 367 continue; 368 case RCTL_ACTION_LOG: 369 /* 370 * If rrl_exceeded != 0, it means we've already 371 * logged a warning for this process. 372 */ 373 if (link->rrl_exceeded != 0) 374 continue; 375 376 /* 377 * If the process state is not fully initialized yet, 378 * we can't access most of the required fields, e.g. 379 * p->p_comm. This happens when called from fork1(). 380 * Ignore this rule for now; it will be processed just 381 * after fork, when called from racct_proc_fork_done(). 382 */ 383 if (p->p_state != PRS_NORMAL) 384 continue; 385 386 if (!ppsratecheck(&lasttime, &curtime, 10)) 387 continue; 388 389 buf = malloc(RCTL_LOG_BUFSIZE, M_RCTL, M_NOWAIT); 390 if (buf == NULL) { 391 printf("rctl_enforce: out of memory\n"); 392 continue; 393 } 394 sbuf_new(&sb, buf, RCTL_LOG_BUFSIZE, SBUF_FIXEDLEN); 395 rctl_rule_to_sbuf(&sb, rule); 396 sbuf_finish(&sb); 397 printf("rctl: rule \"%s\" matched by pid %d " 398 "(%s), uid %d, jail %s\n", sbuf_data(&sb), 399 p->p_pid, p->p_comm, p->p_ucred->cr_uid, 400 p->p_ucred->cr_prison->pr_prison_racct->prr_name); 401 sbuf_delete(&sb); 402 free(buf, M_RCTL); 403 link->rrl_exceeded = 1; 404 continue; 405 case RCTL_ACTION_DEVCTL: 406 if (link->rrl_exceeded != 0) 407 continue; 408 409 if (p->p_state != PRS_NORMAL) 410 continue; 411 412 buf = malloc(RCTL_LOG_BUFSIZE, M_RCTL, M_NOWAIT); 413 if (buf == NULL) { 414 printf("rctl_enforce: out of memory\n"); 415 continue; 416 } 417 sbuf_new(&sb, buf, RCTL_LOG_BUFSIZE, SBUF_FIXEDLEN); 418 sbuf_printf(&sb, "rule="); 419 rctl_rule_to_sbuf(&sb, rule); 420 sbuf_printf(&sb, " pid=%d ruid=%d jail=%s", 421 p->p_pid, p->p_ucred->cr_ruid, 422 p->p_ucred->cr_prison->pr_prison_racct->prr_name); 423 sbuf_finish(&sb); 424 devctl_notify_f("RCTL", "rule", "matched", 425 sbuf_data(&sb), M_NOWAIT); 426 sbuf_delete(&sb); 427 free(buf, M_RCTL); 428 link->rrl_exceeded = 1; 429 continue; 430 default: 431 if (link->rrl_exceeded != 0) 432 continue; 433 434 if (p->p_state != PRS_NORMAL) 435 continue; 436 437 KASSERT(rule->rr_action > 0 && 438 rule->rr_action <= RCTL_ACTION_SIGNAL_MAX, 439 ("rctl_enforce: unknown action %d", 440 rule->rr_action)); 441 442 /* 443 * We're using the fact that RCTL_ACTION_SIG* values 444 * are equal to their counterparts from sys/signal.h. 445 */ 446 kern_psignal(p, rule->rr_action); 447 link->rrl_exceeded = 1; 448 continue; 449 } 450 } 451 452 rw_runlock(&rctl_lock); 453 454 if (should_deny) { 455 /* 456 * Return fake error code; the caller should change it 457 * into one proper for the situation - EFSIZ, ENOMEM etc. 458 */ 459 return (EDOOFUS); 460 } 461 462 return (0); 463} 464 465uint64_t 466rctl_get_limit(struct proc *p, int resource) 467{ 468 struct rctl_rule *rule; 469 struct rctl_rule_link *link; 470 uint64_t amount = UINT64_MAX; 471 472 ASSERT_RACCT_ENABLED(); 473 474 rw_rlock(&rctl_lock); 475 476 /* 477 * There may be more than one matching rule; go through all of them. 478 * Denial should be done last, after logging and sending signals. 479 */ 480 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 481 rule = link->rrl_rule; 482 if (rule->rr_resource != resource) 483 continue; 484 if (rule->rr_action != RCTL_ACTION_DENY) 485 continue; 486 if (rule->rr_amount < amount) 487 amount = rule->rr_amount; 488 } 489 490 rw_runlock(&rctl_lock); 491 492 return (amount); 493} 494 495uint64_t 496rctl_get_available(struct proc *p, int resource) 497{ 498 struct rctl_rule *rule; 499 struct rctl_rule_link *link; 500 int64_t available, minavailable, allocated; 501 502 minavailable = INT64_MAX; 503 504 ASSERT_RACCT_ENABLED(); 505 506 rw_rlock(&rctl_lock); 507 508 /* 509 * There may be more than one matching rule; go through all of them. 510 * Denial should be done last, after logging and sending signals. 511 */ 512 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 513 rule = link->rrl_rule; 514 if (rule->rr_resource != resource) 515 continue; 516 if (rule->rr_action != RCTL_ACTION_DENY) 517 continue; 518 available = rctl_available_resource(p, rule); 519 if (available < minavailable) 520 minavailable = available; 521 } 522 523 rw_runlock(&rctl_lock); 524 525 /* 526 * XXX: Think about this _hard_. 527 */ 528 allocated = p->p_racct->r_resources[resource]; 529 if (minavailable < INT64_MAX - allocated) 530 minavailable += allocated; 531 if (minavailable < 0) 532 minavailable = 0; 533 return (minavailable); 534} 535 536static int 537rctl_rule_matches(const struct rctl_rule *rule, const struct rctl_rule *filter) 538{ 539 540 ASSERT_RACCT_ENABLED(); 541 542 if (filter->rr_subject_type != RCTL_SUBJECT_TYPE_UNDEFINED) { 543 if (rule->rr_subject_type != filter->rr_subject_type) 544 return (0); 545 546 switch (filter->rr_subject_type) { 547 case RCTL_SUBJECT_TYPE_PROCESS: 548 if (filter->rr_subject.rs_proc != NULL && 549 rule->rr_subject.rs_proc != 550 filter->rr_subject.rs_proc) 551 return (0); 552 break; 553 case RCTL_SUBJECT_TYPE_USER: 554 if (filter->rr_subject.rs_uip != NULL && 555 rule->rr_subject.rs_uip != 556 filter->rr_subject.rs_uip) 557 return (0); 558 break; 559 case RCTL_SUBJECT_TYPE_LOGINCLASS: 560 if (filter->rr_subject.rs_loginclass != NULL && 561 rule->rr_subject.rs_loginclass != 562 filter->rr_subject.rs_loginclass) 563 return (0); 564 break; 565 case RCTL_SUBJECT_TYPE_JAIL: 566 if (filter->rr_subject.rs_prison_racct != NULL && 567 rule->rr_subject.rs_prison_racct != 568 filter->rr_subject.rs_prison_racct) 569 return (0); 570 break; 571 default: 572 panic("rctl_rule_matches: unknown subject type %d", 573 filter->rr_subject_type); 574 } 575 } 576 577 if (filter->rr_resource != RACCT_UNDEFINED) { 578 if (rule->rr_resource != filter->rr_resource) 579 return (0); 580 } 581 582 if (filter->rr_action != RCTL_ACTION_UNDEFINED) { 583 if (rule->rr_action != filter->rr_action) 584 return (0); 585 } 586 587 if (filter->rr_amount != RCTL_AMOUNT_UNDEFINED) { 588 if (rule->rr_amount != filter->rr_amount) 589 return (0); 590 } 591 592 if (filter->rr_per != RCTL_SUBJECT_TYPE_UNDEFINED) { 593 if (rule->rr_per != filter->rr_per) 594 return (0); 595 } 596 597 return (1); 598} 599 600static int 601str2value(const char *str, int *value, struct dict *table) 602{ 603 int i; 604 605 if (value == NULL) 606 return (EINVAL); 607 608 for (i = 0; table[i].d_name != NULL; i++) { 609 if (strcasecmp(table[i].d_name, str) == 0) { 610 *value = table[i].d_value; 611 return (0); 612 } 613 } 614 615 return (EINVAL); 616} 617 618static int 619str2id(const char *str, id_t *value) 620{ 621 char *end; 622 623 if (str == NULL) 624 return (EINVAL); 625 626 *value = strtoul(str, &end, 10); 627 if ((size_t)(end - str) != strlen(str)) 628 return (EINVAL); 629 630 return (0); 631} 632 633static int 634str2int64(const char *str, int64_t *value) 635{ 636 char *end; 637 638 if (str == NULL) 639 return (EINVAL); 640 641 *value = strtoul(str, &end, 10); 642 if ((size_t)(end - str) != strlen(str)) 643 return (EINVAL); 644 645 return (0); 646} 647 648/* 649 * Connect the rule to the racct, increasing refcount for the rule. 650 */ 651static void 652rctl_racct_add_rule(struct racct *racct, struct rctl_rule *rule) 653{ 654 struct rctl_rule_link *link; 655 656 ASSERT_RACCT_ENABLED(); 657 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified")); 658 659 rctl_rule_acquire(rule); 660 link = uma_zalloc(rctl_rule_link_zone, M_WAITOK); 661 link->rrl_rule = rule; 662 link->rrl_exceeded = 0; 663 664 rw_wlock(&rctl_lock); 665 LIST_INSERT_HEAD(&racct->r_rule_links, link, rrl_next); 666 rw_wunlock(&rctl_lock); 667} 668 669static int 670rctl_racct_add_rule_locked(struct racct *racct, struct rctl_rule *rule) 671{ 672 struct rctl_rule_link *link; 673 674 ASSERT_RACCT_ENABLED(); 675 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified")); 676 rw_assert(&rctl_lock, RA_WLOCKED); 677 678 link = uma_zalloc(rctl_rule_link_zone, M_NOWAIT); 679 if (link == NULL) 680 return (ENOMEM); 681 rctl_rule_acquire(rule); 682 link->rrl_rule = rule; 683 link->rrl_exceeded = 0; 684 685 LIST_INSERT_HEAD(&racct->r_rule_links, link, rrl_next); 686 return (0); 687} 688 689/* 690 * Remove limits for a rules matching the filter and release 691 * the refcounts for the rules, possibly freeing them. Returns 692 * the number of limit structures removed. 693 */ 694static int 695rctl_racct_remove_rules(struct racct *racct, 696 const struct rctl_rule *filter) 697{ 698 int removed = 0; 699 struct rctl_rule_link *link, *linktmp; 700 701 ASSERT_RACCT_ENABLED(); 702 rw_assert(&rctl_lock, RA_WLOCKED); 703 704 LIST_FOREACH_SAFE(link, &racct->r_rule_links, rrl_next, linktmp) { 705 if (!rctl_rule_matches(link->rrl_rule, filter)) 706 continue; 707 708 LIST_REMOVE(link, rrl_next); 709 rctl_rule_release(link->rrl_rule); 710 uma_zfree(rctl_rule_link_zone, link); 711 removed++; 712 } 713 return (removed); 714} 715 716static void 717rctl_rule_acquire_subject(struct rctl_rule *rule) 718{ 719 720 ASSERT_RACCT_ENABLED(); 721 722 switch (rule->rr_subject_type) { 723 case RCTL_SUBJECT_TYPE_UNDEFINED: 724 case RCTL_SUBJECT_TYPE_PROCESS: 725 break; 726 case RCTL_SUBJECT_TYPE_JAIL: 727 if (rule->rr_subject.rs_prison_racct != NULL) 728 prison_racct_hold(rule->rr_subject.rs_prison_racct); 729 break; 730 case RCTL_SUBJECT_TYPE_USER: 731 if (rule->rr_subject.rs_uip != NULL) 732 uihold(rule->rr_subject.rs_uip); 733 break; 734 case RCTL_SUBJECT_TYPE_LOGINCLASS: 735 if (rule->rr_subject.rs_loginclass != NULL) 736 loginclass_hold(rule->rr_subject.rs_loginclass); 737 break; 738 default: 739 panic("rctl_rule_acquire_subject: unknown subject type %d", 740 rule->rr_subject_type); 741 } 742} 743 744static void 745rctl_rule_release_subject(struct rctl_rule *rule) 746{ 747 748 ASSERT_RACCT_ENABLED(); 749 750 switch (rule->rr_subject_type) { 751 case RCTL_SUBJECT_TYPE_UNDEFINED: 752 case RCTL_SUBJECT_TYPE_PROCESS: 753 break; 754 case RCTL_SUBJECT_TYPE_JAIL: 755 if (rule->rr_subject.rs_prison_racct != NULL) 756 prison_racct_free(rule->rr_subject.rs_prison_racct); 757 break; 758 case RCTL_SUBJECT_TYPE_USER: 759 if (rule->rr_subject.rs_uip != NULL) 760 uifree(rule->rr_subject.rs_uip); 761 break; 762 case RCTL_SUBJECT_TYPE_LOGINCLASS: 763 if (rule->rr_subject.rs_loginclass != NULL) 764 loginclass_free(rule->rr_subject.rs_loginclass); 765 break; 766 default: 767 panic("rctl_rule_release_subject: unknown subject type %d", 768 rule->rr_subject_type); 769 } 770} 771 772struct rctl_rule * 773rctl_rule_alloc(int flags) 774{ 775 struct rctl_rule *rule; 776 777 ASSERT_RACCT_ENABLED(); 778 779 rule = uma_zalloc(rctl_rule_zone, flags); 780 if (rule == NULL) 781 return (NULL); 782 rule->rr_subject_type = RCTL_SUBJECT_TYPE_UNDEFINED; 783 rule->rr_subject.rs_proc = NULL; 784 rule->rr_subject.rs_uip = NULL; 785 rule->rr_subject.rs_loginclass = NULL; 786 rule->rr_subject.rs_prison_racct = NULL; 787 rule->rr_per = RCTL_SUBJECT_TYPE_UNDEFINED; 788 rule->rr_resource = RACCT_UNDEFINED; 789 rule->rr_action = RCTL_ACTION_UNDEFINED; 790 rule->rr_amount = RCTL_AMOUNT_UNDEFINED; 791 refcount_init(&rule->rr_refcount, 1); 792 793 return (rule); 794} 795 796struct rctl_rule * 797rctl_rule_duplicate(const struct rctl_rule *rule, int flags) 798{ 799 struct rctl_rule *copy; 800 801 ASSERT_RACCT_ENABLED(); 802 803 copy = uma_zalloc(rctl_rule_zone, flags); 804 if (copy == NULL) 805 return (NULL); 806 copy->rr_subject_type = rule->rr_subject_type; 807 copy->rr_subject.rs_proc = rule->rr_subject.rs_proc; 808 copy->rr_subject.rs_uip = rule->rr_subject.rs_uip; 809 copy->rr_subject.rs_loginclass = rule->rr_subject.rs_loginclass; 810 copy->rr_subject.rs_prison_racct = rule->rr_subject.rs_prison_racct; 811 copy->rr_per = rule->rr_per; 812 copy->rr_resource = rule->rr_resource; 813 copy->rr_action = rule->rr_action; 814 copy->rr_amount = rule->rr_amount; 815 refcount_init(©->rr_refcount, 1); 816 rctl_rule_acquire_subject(copy); 817 818 return (copy); 819} 820 821void 822rctl_rule_acquire(struct rctl_rule *rule) 823{ 824 825 ASSERT_RACCT_ENABLED(); 826 KASSERT(rule->rr_refcount > 0, ("rule->rr_refcount <= 0")); 827 828 refcount_acquire(&rule->rr_refcount); 829} 830 831static void 832rctl_rule_free(void *context, int pending) 833{ 834 struct rctl_rule *rule; 835 836 rule = (struct rctl_rule *)context; 837 838 ASSERT_RACCT_ENABLED(); 839 KASSERT(rule->rr_refcount == 0, ("rule->rr_refcount != 0")); 840 841 /* 842 * We don't need locking here; rule is guaranteed to be inaccessible. 843 */ 844 845 rctl_rule_release_subject(rule); 846 uma_zfree(rctl_rule_zone, rule); 847} 848 849void 850rctl_rule_release(struct rctl_rule *rule) 851{ 852 853 ASSERT_RACCT_ENABLED(); 854 KASSERT(rule->rr_refcount > 0, ("rule->rr_refcount <= 0")); 855 856 if (refcount_release(&rule->rr_refcount)) { 857 /* 858 * rctl_rule_release() is often called when iterating 859 * over all the uidinfo structures in the system, 860 * holding uihashtbl_lock. Since rctl_rule_free() 861 * might end up calling uifree(), this would lead 862 * to lock recursion. Use taskqueue to avoid this. 863 */ 864 TASK_INIT(&rule->rr_task, 0, rctl_rule_free, rule); 865 taskqueue_enqueue(taskqueue_thread, &rule->rr_task); 866 } 867} 868 869static int 870rctl_rule_fully_specified(const struct rctl_rule *rule) 871{ 872 873 ASSERT_RACCT_ENABLED(); 874 875 switch (rule->rr_subject_type) { 876 case RCTL_SUBJECT_TYPE_UNDEFINED: 877 return (0); 878 case RCTL_SUBJECT_TYPE_PROCESS: 879 if (rule->rr_subject.rs_proc == NULL) 880 return (0); 881 break; 882 case RCTL_SUBJECT_TYPE_USER: 883 if (rule->rr_subject.rs_uip == NULL) 884 return (0); 885 break; 886 case RCTL_SUBJECT_TYPE_LOGINCLASS: 887 if (rule->rr_subject.rs_loginclass == NULL) 888 return (0); 889 break; 890 case RCTL_SUBJECT_TYPE_JAIL: 891 if (rule->rr_subject.rs_prison_racct == NULL) 892 return (0); 893 break; 894 default: 895 panic("rctl_rule_fully_specified: unknown subject type %d", 896 rule->rr_subject_type); 897 } 898 if (rule->rr_resource == RACCT_UNDEFINED) 899 return (0); 900 if (rule->rr_action == RCTL_ACTION_UNDEFINED) 901 return (0); 902 if (rule->rr_amount == RCTL_AMOUNT_UNDEFINED) 903 return (0); 904 if (rule->rr_per == RCTL_SUBJECT_TYPE_UNDEFINED) 905 return (0); 906 907 return (1); 908} 909 910static int 911rctl_string_to_rule(char *rulestr, struct rctl_rule **rulep) 912{ 913 int error = 0; 914 char *subjectstr, *subject_idstr, *resourcestr, *actionstr, 915 *amountstr, *perstr; 916 struct rctl_rule *rule; 917 id_t id; 918 919 ASSERT_RACCT_ENABLED(); 920 921 rule = rctl_rule_alloc(M_WAITOK); 922 923 subjectstr = strsep(&rulestr, ":"); 924 subject_idstr = strsep(&rulestr, ":"); 925 resourcestr = strsep(&rulestr, ":"); 926 actionstr = strsep(&rulestr, "=/"); 927 amountstr = strsep(&rulestr, "/"); 928 perstr = rulestr; 929 930 if (subjectstr == NULL || subjectstr[0] == '\0') 931 rule->rr_subject_type = RCTL_SUBJECT_TYPE_UNDEFINED; 932 else { 933 error = str2value(subjectstr, &rule->rr_subject_type, subjectnames); 934 if (error != 0) 935 goto out; 936 } 937 938 if (subject_idstr == NULL || subject_idstr[0] == '\0') { 939 rule->rr_subject.rs_proc = NULL; 940 rule->rr_subject.rs_uip = NULL; 941 rule->rr_subject.rs_loginclass = NULL; 942 rule->rr_subject.rs_prison_racct = NULL; 943 } else { 944 switch (rule->rr_subject_type) { 945 case RCTL_SUBJECT_TYPE_UNDEFINED: 946 error = EINVAL; 947 goto out; 948 case RCTL_SUBJECT_TYPE_PROCESS: 949 error = str2id(subject_idstr, &id); 950 if (error != 0) 951 goto out; 952 sx_assert(&allproc_lock, SA_LOCKED); 953 rule->rr_subject.rs_proc = pfind(id); 954 if (rule->rr_subject.rs_proc == NULL) { 955 error = ESRCH; 956 goto out; 957 } 958 PROC_UNLOCK(rule->rr_subject.rs_proc); 959 break; 960 case RCTL_SUBJECT_TYPE_USER: 961 error = str2id(subject_idstr, &id); 962 if (error != 0) 963 goto out; 964 rule->rr_subject.rs_uip = uifind(id); 965 break; 966 case RCTL_SUBJECT_TYPE_LOGINCLASS: 967 rule->rr_subject.rs_loginclass = 968 loginclass_find(subject_idstr); 969 if (rule->rr_subject.rs_loginclass == NULL) { 970 error = ENAMETOOLONG; 971 goto out; 972 } 973 break; 974 case RCTL_SUBJECT_TYPE_JAIL: 975 rule->rr_subject.rs_prison_racct = 976 prison_racct_find(subject_idstr); 977 if (rule->rr_subject.rs_prison_racct == NULL) { 978 error = ENAMETOOLONG; 979 goto out; 980 } 981 break; 982 default: 983 panic("rctl_string_to_rule: unknown subject type %d", 984 rule->rr_subject_type); 985 } 986 } 987 988 if (resourcestr == NULL || resourcestr[0] == '\0') 989 rule->rr_resource = RACCT_UNDEFINED; 990 else { 991 error = str2value(resourcestr, &rule->rr_resource, 992 resourcenames); 993 if (error != 0) 994 goto out; 995 } 996 997 if (actionstr == NULL || actionstr[0] == '\0') 998 rule->rr_action = RCTL_ACTION_UNDEFINED; 999 else { 1000 error = str2value(actionstr, &rule->rr_action, actionnames); 1001 if (error != 0) 1002 goto out; 1003 } 1004 1005 if (amountstr == NULL || amountstr[0] == '\0') 1006 rule->rr_amount = RCTL_AMOUNT_UNDEFINED; 1007 else { 1008 error = str2int64(amountstr, &rule->rr_amount); 1009 if (error != 0) 1010 goto out; 1011 if (RACCT_IS_IN_MILLIONS(rule->rr_resource)) 1012 rule->rr_amount *= 1000000; 1013 } 1014 1015 if (perstr == NULL || perstr[0] == '\0') 1016 rule->rr_per = RCTL_SUBJECT_TYPE_UNDEFINED; 1017 else { 1018 error = str2value(perstr, &rule->rr_per, subjectnames); 1019 if (error != 0) 1020 goto out; 1021 } 1022 1023out: 1024 if (error == 0) 1025 *rulep = rule; 1026 else 1027 rctl_rule_release(rule); 1028 1029 return (error); 1030} 1031 1032/* 1033 * Link a rule with all the subjects it applies to. 1034 */ 1035int 1036rctl_rule_add(struct rctl_rule *rule) 1037{ 1038 struct proc *p; 1039 struct ucred *cred; 1040 struct uidinfo *uip; 1041 struct prison *pr; 1042 struct prison_racct *prr; 1043 struct loginclass *lc; 1044 struct rctl_rule *rule2; 1045 int match; 1046 1047 ASSERT_RACCT_ENABLED(); 1048 KASSERT(rctl_rule_fully_specified(rule), ("rule not fully specified")); 1049 1050 /* 1051 * Some rules just don't make sense. Note that the one below 1052 * cannot be rewritten using RACCT_IS_DENIABLE(); the RACCT_PCTCPU, 1053 * for example, is not deniable in the racct sense, but the 1054 * limit is enforced in a different way, so "deny" rules for %CPU 1055 * do make sense. 1056 */ 1057 if (rule->rr_action == RCTL_ACTION_DENY && 1058 (rule->rr_resource == RACCT_CPU || 1059 rule->rr_resource == RACCT_WALLCLOCK)) 1060 return (EOPNOTSUPP); 1061 1062 if (rule->rr_per == RCTL_SUBJECT_TYPE_PROCESS && 1063 RACCT_IS_SLOPPY(rule->rr_resource)) 1064 return (EOPNOTSUPP); 1065 1066 /* 1067 * Make sure there are no duplicated rules. Also, for the "deny" 1068 * rules, remove ones differing only by "amount". 1069 */ 1070 if (rule->rr_action == RCTL_ACTION_DENY) { 1071 rule2 = rctl_rule_duplicate(rule, M_WAITOK); 1072 rule2->rr_amount = RCTL_AMOUNT_UNDEFINED; 1073 rctl_rule_remove(rule2); 1074 rctl_rule_release(rule2); 1075 } else 1076 rctl_rule_remove(rule); 1077 1078 switch (rule->rr_subject_type) { 1079 case RCTL_SUBJECT_TYPE_PROCESS: 1080 p = rule->rr_subject.rs_proc; 1081 KASSERT(p != NULL, ("rctl_rule_add: NULL proc")); 1082 1083 rctl_racct_add_rule(p->p_racct, rule); 1084 /* 1085 * In case of per-process rule, we don't have anything more 1086 * to do. 1087 */ 1088 return (0); 1089 1090 case RCTL_SUBJECT_TYPE_USER: 1091 uip = rule->rr_subject.rs_uip; 1092 KASSERT(uip != NULL, ("rctl_rule_add: NULL uip")); 1093 rctl_racct_add_rule(uip->ui_racct, rule); 1094 break; 1095 1096 case RCTL_SUBJECT_TYPE_LOGINCLASS: 1097 lc = rule->rr_subject.rs_loginclass; 1098 KASSERT(lc != NULL, ("rctl_rule_add: NULL loginclass")); 1099 rctl_racct_add_rule(lc->lc_racct, rule); 1100 break; 1101 1102 case RCTL_SUBJECT_TYPE_JAIL: 1103 prr = rule->rr_subject.rs_prison_racct; 1104 KASSERT(prr != NULL, ("rctl_rule_add: NULL pr")); 1105 rctl_racct_add_rule(prr->prr_racct, rule); 1106 break; 1107 1108 default: 1109 panic("rctl_rule_add: unknown subject type %d", 1110 rule->rr_subject_type); 1111 } 1112 1113 /* 1114 * Now go through all the processes and add the new rule to the ones 1115 * it applies to. 1116 */ 1117 sx_assert(&allproc_lock, SA_LOCKED); 1118 FOREACH_PROC_IN_SYSTEM(p) { 1119 cred = p->p_ucred; 1120 switch (rule->rr_subject_type) { 1121 case RCTL_SUBJECT_TYPE_USER: 1122 if (cred->cr_uidinfo == rule->rr_subject.rs_uip || 1123 cred->cr_ruidinfo == rule->rr_subject.rs_uip) 1124 break; 1125 continue; 1126 case RCTL_SUBJECT_TYPE_LOGINCLASS: 1127 if (cred->cr_loginclass == rule->rr_subject.rs_loginclass) 1128 break; 1129 continue; 1130 case RCTL_SUBJECT_TYPE_JAIL: 1131 match = 0; 1132 for (pr = cred->cr_prison; pr != NULL; pr = pr->pr_parent) { 1133 if (pr->pr_prison_racct == rule->rr_subject.rs_prison_racct) { 1134 match = 1; 1135 break; 1136 } 1137 } 1138 if (match) 1139 break; 1140 continue; 1141 default: 1142 panic("rctl_rule_add: unknown subject type %d", 1143 rule->rr_subject_type); 1144 } 1145 1146 rctl_racct_add_rule(p->p_racct, rule); 1147 } 1148 1149 return (0); 1150} 1151 1152static void 1153rctl_rule_pre_callback(void) 1154{ 1155 1156 rw_wlock(&rctl_lock); 1157} 1158 1159static void 1160rctl_rule_post_callback(void) 1161{ 1162 1163 rw_wunlock(&rctl_lock); 1164} 1165 1166static void 1167rctl_rule_remove_callback(struct racct *racct, void *arg2, void *arg3) 1168{ 1169 struct rctl_rule *filter = (struct rctl_rule *)arg2; 1170 int found = 0; 1171 1172 ASSERT_RACCT_ENABLED(); 1173 rw_assert(&rctl_lock, RA_WLOCKED); 1174 1175 found += rctl_racct_remove_rules(racct, filter); 1176 1177 *((int *)arg3) += found; 1178} 1179 1180/* 1181 * Remove all rules that match the filter. 1182 */ 1183int 1184rctl_rule_remove(struct rctl_rule *filter) 1185{ 1186 int found = 0; 1187 struct proc *p; 1188 1189 ASSERT_RACCT_ENABLED(); 1190 1191 if (filter->rr_subject_type == RCTL_SUBJECT_TYPE_PROCESS && 1192 filter->rr_subject.rs_proc != NULL) { 1193 p = filter->rr_subject.rs_proc; 1194 rw_wlock(&rctl_lock); 1195 found = rctl_racct_remove_rules(p->p_racct, filter); 1196 rw_wunlock(&rctl_lock); 1197 if (found) 1198 return (0); 1199 return (ESRCH); 1200 } 1201 1202 loginclass_racct_foreach(rctl_rule_remove_callback, 1203 rctl_rule_pre_callback, rctl_rule_post_callback, 1204 filter, (void *)&found); 1205 ui_racct_foreach(rctl_rule_remove_callback, 1206 rctl_rule_pre_callback, rctl_rule_post_callback, 1207 filter, (void *)&found); 1208 prison_racct_foreach(rctl_rule_remove_callback, 1209 rctl_rule_pre_callback, rctl_rule_post_callback, 1210 filter, (void *)&found); 1211 1212 sx_assert(&allproc_lock, SA_LOCKED); 1213 rw_wlock(&rctl_lock); 1214 FOREACH_PROC_IN_SYSTEM(p) { 1215 found += rctl_racct_remove_rules(p->p_racct, filter); 1216 } 1217 rw_wunlock(&rctl_lock); 1218 1219 if (found) 1220 return (0); 1221 return (ESRCH); 1222} 1223 1224/* 1225 * Appends a rule to the sbuf. 1226 */ 1227static void 1228rctl_rule_to_sbuf(struct sbuf *sb, const struct rctl_rule *rule) 1229{ 1230 int64_t amount; 1231 1232 ASSERT_RACCT_ENABLED(); 1233 1234 sbuf_printf(sb, "%s:", rctl_subject_type_name(rule->rr_subject_type)); 1235 1236 switch (rule->rr_subject_type) { 1237 case RCTL_SUBJECT_TYPE_PROCESS: 1238 if (rule->rr_subject.rs_proc == NULL) 1239 sbuf_printf(sb, ":"); 1240 else 1241 sbuf_printf(sb, "%d:", 1242 rule->rr_subject.rs_proc->p_pid); 1243 break; 1244 case RCTL_SUBJECT_TYPE_USER: 1245 if (rule->rr_subject.rs_uip == NULL) 1246 sbuf_printf(sb, ":"); 1247 else 1248 sbuf_printf(sb, "%d:", 1249 rule->rr_subject.rs_uip->ui_uid); 1250 break; 1251 case RCTL_SUBJECT_TYPE_LOGINCLASS: 1252 if (rule->rr_subject.rs_loginclass == NULL) 1253 sbuf_printf(sb, ":"); 1254 else 1255 sbuf_printf(sb, "%s:", 1256 rule->rr_subject.rs_loginclass->lc_name); 1257 break; 1258 case RCTL_SUBJECT_TYPE_JAIL: 1259 if (rule->rr_subject.rs_prison_racct == NULL) 1260 sbuf_printf(sb, ":"); 1261 else 1262 sbuf_printf(sb, "%s:", 1263 rule->rr_subject.rs_prison_racct->prr_name); 1264 break; 1265 default: 1266 panic("rctl_rule_to_sbuf: unknown subject type %d", 1267 rule->rr_subject_type); 1268 } 1269 1270 amount = rule->rr_amount; 1271 if (amount != RCTL_AMOUNT_UNDEFINED && 1272 RACCT_IS_IN_MILLIONS(rule->rr_resource)) 1273 amount /= 1000000; 1274 1275 sbuf_printf(sb, "%s:%s=%jd", 1276 rctl_resource_name(rule->rr_resource), 1277 rctl_action_name(rule->rr_action), 1278 amount); 1279 1280 if (rule->rr_per != rule->rr_subject_type) 1281 sbuf_printf(sb, "/%s", rctl_subject_type_name(rule->rr_per)); 1282} 1283 1284/* 1285 * Routine used by RCTL syscalls to read in input string. 1286 */ 1287static int 1288rctl_read_inbuf(char **inputstr, const char *inbufp, size_t inbuflen) 1289{ 1290 int error; 1291 char *str; 1292 1293 ASSERT_RACCT_ENABLED(); 1294 1295 if (inbuflen <= 0) 1296 return (EINVAL); 1297 if (inbuflen > RCTL_MAX_INBUFSIZE) 1298 return (E2BIG); 1299 1300 str = malloc(inbuflen + 1, M_RCTL, M_WAITOK); 1301 error = copyinstr(inbufp, str, inbuflen, NULL); 1302 if (error != 0) { 1303 free(str, M_RCTL); 1304 return (error); 1305 } 1306 1307 *inputstr = str; 1308 1309 return (0); 1310} 1311 1312/* 1313 * Routine used by RCTL syscalls to write out output string. 1314 */ 1315static int 1316rctl_write_outbuf(struct sbuf *outputsbuf, char *outbufp, size_t outbuflen) 1317{ 1318 int error; 1319 1320 ASSERT_RACCT_ENABLED(); 1321 1322 if (outputsbuf == NULL) 1323 return (0); 1324 1325 sbuf_finish(outputsbuf); 1326 if (outbuflen < sbuf_len(outputsbuf) + 1) { 1327 sbuf_delete(outputsbuf); 1328 return (ERANGE); 1329 } 1330 error = copyout(sbuf_data(outputsbuf), outbufp, 1331 sbuf_len(outputsbuf) + 1); 1332 sbuf_delete(outputsbuf); 1333 return (error); 1334} 1335 1336static struct sbuf * 1337rctl_racct_to_sbuf(struct racct *racct, int sloppy) 1338{ 1339 int i; 1340 int64_t amount; 1341 struct sbuf *sb; 1342 1343 ASSERT_RACCT_ENABLED(); 1344 1345 sb = sbuf_new_auto(); 1346 for (i = 0; i <= RACCT_MAX; i++) { 1347 if (sloppy == 0 && RACCT_IS_SLOPPY(i)) 1348 continue; 1349 amount = racct->r_resources[i]; 1350 if (RACCT_IS_IN_MILLIONS(i)) 1351 amount /= 1000000; 1352 sbuf_printf(sb, "%s=%jd,", rctl_resource_name(i), amount); 1353 } 1354 sbuf_setpos(sb, sbuf_len(sb) - 1); 1355 return (sb); 1356} 1357 1358int 1359sys_rctl_get_racct(struct thread *td, struct rctl_get_racct_args *uap) 1360{ 1361 int error; 1362 char *inputstr; 1363 struct rctl_rule *filter; 1364 struct sbuf *outputsbuf = NULL; 1365 struct proc *p; 1366 struct uidinfo *uip; 1367 struct loginclass *lc; 1368 struct prison_racct *prr; 1369 1370 if (!racct_enable) 1371 return (ENOSYS); 1372 1373 error = priv_check(td, PRIV_RCTL_GET_RACCT); 1374 if (error != 0) 1375 return (error); 1376 1377 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen); 1378 if (error != 0) 1379 return (error); 1380 1381 sx_slock(&allproc_lock); 1382 error = rctl_string_to_rule(inputstr, &filter); 1383 free(inputstr, M_RCTL); 1384 if (error != 0) { 1385 sx_sunlock(&allproc_lock); 1386 return (error); 1387 } 1388 1389 switch (filter->rr_subject_type) { 1390 case RCTL_SUBJECT_TYPE_PROCESS: 1391 p = filter->rr_subject.rs_proc; 1392 if (p == NULL) { 1393 error = EINVAL; 1394 goto out; 1395 } 1396 outputsbuf = rctl_racct_to_sbuf(p->p_racct, 0); 1397 break; 1398 case RCTL_SUBJECT_TYPE_USER: 1399 uip = filter->rr_subject.rs_uip; 1400 if (uip == NULL) { 1401 error = EINVAL; 1402 goto out; 1403 } 1404 outputsbuf = rctl_racct_to_sbuf(uip->ui_racct, 1); 1405 break; 1406 case RCTL_SUBJECT_TYPE_LOGINCLASS: 1407 lc = filter->rr_subject.rs_loginclass; 1408 if (lc == NULL) { 1409 error = EINVAL; 1410 goto out; 1411 } 1412 outputsbuf = rctl_racct_to_sbuf(lc->lc_racct, 1); 1413 break; 1414 case RCTL_SUBJECT_TYPE_JAIL: 1415 prr = filter->rr_subject.rs_prison_racct; 1416 if (prr == NULL) { 1417 error = EINVAL; 1418 goto out; 1419 } 1420 outputsbuf = rctl_racct_to_sbuf(prr->prr_racct, 1); 1421 break; 1422 default: 1423 error = EINVAL; 1424 } 1425out: 1426 rctl_rule_release(filter); 1427 sx_sunlock(&allproc_lock); 1428 if (error != 0) 1429 return (error); 1430 1431 error = rctl_write_outbuf(outputsbuf, uap->outbufp, uap->outbuflen); 1432 1433 return (error); 1434} 1435 1436static void 1437rctl_get_rules_callback(struct racct *racct, void *arg2, void *arg3) 1438{ 1439 struct rctl_rule *filter = (struct rctl_rule *)arg2; 1440 struct rctl_rule_link *link; 1441 struct sbuf *sb = (struct sbuf *)arg3; 1442 1443 ASSERT_RACCT_ENABLED(); 1444 rw_assert(&rctl_lock, RA_LOCKED); 1445 1446 LIST_FOREACH(link, &racct->r_rule_links, rrl_next) { 1447 if (!rctl_rule_matches(link->rrl_rule, filter)) 1448 continue; 1449 rctl_rule_to_sbuf(sb, link->rrl_rule); 1450 sbuf_printf(sb, ","); 1451 } 1452} 1453 1454int 1455sys_rctl_get_rules(struct thread *td, struct rctl_get_rules_args *uap) 1456{ 1457 int error; 1458 size_t bufsize; 1459 char *inputstr, *buf; 1460 struct sbuf *sb; 1461 struct rctl_rule *filter; 1462 struct rctl_rule_link *link; 1463 struct proc *p; 1464 1465 if (!racct_enable) 1466 return (ENOSYS); 1467 1468 error = priv_check(td, PRIV_RCTL_GET_RULES); 1469 if (error != 0) 1470 return (error); 1471 1472 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen); 1473 if (error != 0) 1474 return (error); 1475 1476 sx_slock(&allproc_lock); 1477 error = rctl_string_to_rule(inputstr, &filter); 1478 free(inputstr, M_RCTL); 1479 if (error != 0) { 1480 sx_sunlock(&allproc_lock); 1481 return (error); 1482 } 1483 1484 bufsize = uap->outbuflen; 1485 if (bufsize > rctl_maxbufsize) { 1486 sx_sunlock(&allproc_lock); 1487 return (E2BIG); 1488 } 1489 1490 buf = malloc(bufsize, M_RCTL, M_WAITOK); 1491 sb = sbuf_new(NULL, buf, bufsize, SBUF_FIXEDLEN); 1492 KASSERT(sb != NULL, ("sbuf_new failed")); 1493 1494 FOREACH_PROC_IN_SYSTEM(p) { 1495 rw_rlock(&rctl_lock); 1496 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 1497 /* 1498 * Non-process rules will be added to the buffer later. 1499 * Adding them here would result in duplicated output. 1500 */ 1501 if (link->rrl_rule->rr_subject_type != 1502 RCTL_SUBJECT_TYPE_PROCESS) 1503 continue; 1504 if (!rctl_rule_matches(link->rrl_rule, filter)) 1505 continue; 1506 rctl_rule_to_sbuf(sb, link->rrl_rule); 1507 sbuf_printf(sb, ","); 1508 } 1509 rw_runlock(&rctl_lock); 1510 } 1511 1512 loginclass_racct_foreach(rctl_get_rules_callback, 1513 rctl_rule_pre_callback, rctl_rule_post_callback, 1514 filter, sb); 1515 ui_racct_foreach(rctl_get_rules_callback, 1516 rctl_rule_pre_callback, rctl_rule_post_callback, 1517 filter, sb); 1518 prison_racct_foreach(rctl_get_rules_callback, 1519 rctl_rule_pre_callback, rctl_rule_post_callback, 1520 filter, sb); 1521 if (sbuf_error(sb) == ENOMEM) { 1522 error = ERANGE; 1523 goto out; 1524 } 1525 1526 /* 1527 * Remove trailing ",". 1528 */ 1529 if (sbuf_len(sb) > 0) 1530 sbuf_setpos(sb, sbuf_len(sb) - 1); 1531 1532 error = rctl_write_outbuf(sb, uap->outbufp, uap->outbuflen); 1533out: 1534 rctl_rule_release(filter); 1535 sx_sunlock(&allproc_lock); 1536 free(buf, M_RCTL); 1537 return (error); 1538} 1539 1540int 1541sys_rctl_get_limits(struct thread *td, struct rctl_get_limits_args *uap) 1542{ 1543 int error; 1544 size_t bufsize; 1545 char *inputstr, *buf; 1546 struct sbuf *sb; 1547 struct rctl_rule *filter; 1548 struct rctl_rule_link *link; 1549 1550 if (!racct_enable) 1551 return (ENOSYS); 1552 1553 error = priv_check(td, PRIV_RCTL_GET_LIMITS); 1554 if (error != 0) 1555 return (error); 1556 1557 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen); 1558 if (error != 0) 1559 return (error); 1560 1561 sx_slock(&allproc_lock); 1562 error = rctl_string_to_rule(inputstr, &filter); 1563 free(inputstr, M_RCTL); 1564 if (error != 0) { 1565 sx_sunlock(&allproc_lock); 1566 return (error); 1567 } 1568 1569 if (filter->rr_subject_type == RCTL_SUBJECT_TYPE_UNDEFINED) { 1570 rctl_rule_release(filter); 1571 sx_sunlock(&allproc_lock); 1572 return (EINVAL); 1573 } 1574 if (filter->rr_subject_type != RCTL_SUBJECT_TYPE_PROCESS) { 1575 rctl_rule_release(filter); 1576 sx_sunlock(&allproc_lock); 1577 return (EOPNOTSUPP); 1578 } 1579 if (filter->rr_subject.rs_proc == NULL) { 1580 rctl_rule_release(filter); 1581 sx_sunlock(&allproc_lock); 1582 return (EINVAL); 1583 } 1584 1585 bufsize = uap->outbuflen; 1586 if (bufsize > rctl_maxbufsize) { 1587 rctl_rule_release(filter); 1588 sx_sunlock(&allproc_lock); 1589 return (E2BIG); 1590 } 1591 1592 buf = malloc(bufsize, M_RCTL, M_WAITOK); 1593 sb = sbuf_new(NULL, buf, bufsize, SBUF_FIXEDLEN); 1594 KASSERT(sb != NULL, ("sbuf_new failed")); 1595 1596 rw_rlock(&rctl_lock); 1597 LIST_FOREACH(link, &filter->rr_subject.rs_proc->p_racct->r_rule_links, 1598 rrl_next) { 1599 rctl_rule_to_sbuf(sb, link->rrl_rule); 1600 sbuf_printf(sb, ","); 1601 } 1602 rw_runlock(&rctl_lock); 1603 if (sbuf_error(sb) == ENOMEM) { 1604 error = ERANGE; 1605 goto out; 1606 } 1607 1608 /* 1609 * Remove trailing ",". 1610 */ 1611 if (sbuf_len(sb) > 0) 1612 sbuf_setpos(sb, sbuf_len(sb) - 1); 1613 1614 error = rctl_write_outbuf(sb, uap->outbufp, uap->outbuflen); 1615out: 1616 rctl_rule_release(filter); 1617 sx_sunlock(&allproc_lock); 1618 free(buf, M_RCTL); 1619 return (error); 1620} 1621 1622int 1623sys_rctl_add_rule(struct thread *td, struct rctl_add_rule_args *uap) 1624{ 1625 int error; 1626 struct rctl_rule *rule; 1627 char *inputstr; 1628 1629 if (!racct_enable) 1630 return (ENOSYS); 1631 1632 error = priv_check(td, PRIV_RCTL_ADD_RULE); 1633 if (error != 0) 1634 return (error); 1635 1636 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen); 1637 if (error != 0) 1638 return (error); 1639 1640 sx_slock(&allproc_lock); 1641 error = rctl_string_to_rule(inputstr, &rule); 1642 free(inputstr, M_RCTL); 1643 if (error != 0) { 1644 sx_sunlock(&allproc_lock); 1645 return (error); 1646 } 1647 /* 1648 * The 'per' part of a rule is optional. 1649 */ 1650 if (rule->rr_per == RCTL_SUBJECT_TYPE_UNDEFINED && 1651 rule->rr_subject_type != RCTL_SUBJECT_TYPE_UNDEFINED) 1652 rule->rr_per = rule->rr_subject_type; 1653 1654 if (!rctl_rule_fully_specified(rule)) { 1655 error = EINVAL; 1656 goto out; 1657 } 1658 1659 error = rctl_rule_add(rule); 1660 1661out: 1662 rctl_rule_release(rule); 1663 sx_sunlock(&allproc_lock); 1664 return (error); 1665} 1666 1667int 1668sys_rctl_remove_rule(struct thread *td, struct rctl_remove_rule_args *uap) 1669{ 1670 int error; 1671 struct rctl_rule *filter; 1672 char *inputstr; 1673 1674 if (!racct_enable) 1675 return (ENOSYS); 1676 1677 error = priv_check(td, PRIV_RCTL_REMOVE_RULE); 1678 if (error != 0) 1679 return (error); 1680 1681 error = rctl_read_inbuf(&inputstr, uap->inbufp, uap->inbuflen); 1682 if (error != 0) 1683 return (error); 1684 1685 sx_slock(&allproc_lock); 1686 error = rctl_string_to_rule(inputstr, &filter); 1687 free(inputstr, M_RCTL); 1688 if (error != 0) { 1689 sx_sunlock(&allproc_lock); 1690 return (error); 1691 } 1692 1693 error = rctl_rule_remove(filter); 1694 rctl_rule_release(filter); 1695 sx_sunlock(&allproc_lock); 1696 1697 return (error); 1698} 1699 1700/* 1701 * Update RCTL rule list after credential change. 1702 */ 1703void 1704rctl_proc_ucred_changed(struct proc *p, struct ucred *newcred) 1705{ 1706 int rulecnt, i; 1707 struct rctl_rule_link *link, *newlink; 1708 struct uidinfo *newuip; 1709 struct loginclass *newlc; 1710 struct prison_racct *newprr; 1711 LIST_HEAD(, rctl_rule_link) newrules; 1712 1713 ASSERT_RACCT_ENABLED(); 1714 1715 newuip = newcred->cr_ruidinfo; 1716 newlc = newcred->cr_loginclass; 1717 newprr = newcred->cr_prison->pr_prison_racct; 1718 1719 LIST_INIT(&newrules); 1720 1721again: 1722 /* 1723 * First, count the rules that apply to the process with new 1724 * credentials. 1725 */ 1726 rulecnt = 0; 1727 rw_rlock(&rctl_lock); 1728 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 1729 if (link->rrl_rule->rr_subject_type == 1730 RCTL_SUBJECT_TYPE_PROCESS) 1731 rulecnt++; 1732 } 1733 LIST_FOREACH(link, &newuip->ui_racct->r_rule_links, rrl_next) 1734 rulecnt++; 1735 LIST_FOREACH(link, &newlc->lc_racct->r_rule_links, rrl_next) 1736 rulecnt++; 1737 LIST_FOREACH(link, &newprr->prr_racct->r_rule_links, rrl_next) 1738 rulecnt++; 1739 rw_runlock(&rctl_lock); 1740 1741 /* 1742 * Create temporary list. We've dropped the rctl_lock in order 1743 * to use M_WAITOK. 1744 */ 1745 for (i = 0; i < rulecnt; i++) { 1746 newlink = uma_zalloc(rctl_rule_link_zone, M_WAITOK); 1747 newlink->rrl_rule = NULL; 1748 LIST_INSERT_HEAD(&newrules, newlink, rrl_next); 1749 } 1750 1751 newlink = LIST_FIRST(&newrules); 1752 1753 /* 1754 * Assign rules to the newly allocated list entries. 1755 */ 1756 rw_wlock(&rctl_lock); 1757 LIST_FOREACH(link, &p->p_racct->r_rule_links, rrl_next) { 1758 if (link->rrl_rule->rr_subject_type == 1759 RCTL_SUBJECT_TYPE_PROCESS) { 1760 if (newlink == NULL) 1761 goto goaround; 1762 rctl_rule_acquire(link->rrl_rule); 1763 newlink->rrl_rule = link->rrl_rule; 1764 newlink = LIST_NEXT(newlink, rrl_next); 1765 rulecnt--; 1766 } 1767 } 1768 1769 LIST_FOREACH(link, &newuip->ui_racct->r_rule_links, rrl_next) { 1770 if (newlink == NULL) 1771 goto goaround; 1772 rctl_rule_acquire(link->rrl_rule); 1773 newlink->rrl_rule = link->rrl_rule; 1774 newlink = LIST_NEXT(newlink, rrl_next); 1775 rulecnt--; 1776 } 1777 1778 LIST_FOREACH(link, &newlc->lc_racct->r_rule_links, rrl_next) { 1779 if (newlink == NULL) 1780 goto goaround; 1781 rctl_rule_acquire(link->rrl_rule); 1782 newlink->rrl_rule = link->rrl_rule; 1783 newlink = LIST_NEXT(newlink, rrl_next); 1784 rulecnt--; 1785 } 1786 1787 LIST_FOREACH(link, &newprr->prr_racct->r_rule_links, rrl_next) { 1788 if (newlink == NULL) 1789 goto goaround; 1790 rctl_rule_acquire(link->rrl_rule); 1791 newlink->rrl_rule = link->rrl_rule; 1792 newlink = LIST_NEXT(newlink, rrl_next); 1793 rulecnt--; 1794 } 1795 1796 if (rulecnt == 0) { 1797 /* 1798 * Free the old rule list. 1799 */ 1800 while (!LIST_EMPTY(&p->p_racct->r_rule_links)) { 1801 link = LIST_FIRST(&p->p_racct->r_rule_links); 1802 LIST_REMOVE(link, rrl_next); 1803 rctl_rule_release(link->rrl_rule); 1804 uma_zfree(rctl_rule_link_zone, link); 1805 } 1806 1807 /* 1808 * Replace lists and we're done. 1809 * 1810 * XXX: Is there any way to switch list heads instead 1811 * of iterating here? 1812 */ 1813 while (!LIST_EMPTY(&newrules)) { 1814 newlink = LIST_FIRST(&newrules); 1815 LIST_REMOVE(newlink, rrl_next); 1816 LIST_INSERT_HEAD(&p->p_racct->r_rule_links, 1817 newlink, rrl_next); 1818 } 1819 1820 rw_wunlock(&rctl_lock); 1821 1822 return; 1823 } 1824 1825goaround: 1826 rw_wunlock(&rctl_lock); 1827 1828 /* 1829 * Rule list changed while we were not holding the rctl_lock. 1830 * Free the new list and try again. 1831 */ 1832 while (!LIST_EMPTY(&newrules)) { 1833 newlink = LIST_FIRST(&newrules); 1834 LIST_REMOVE(newlink, rrl_next); 1835 if (newlink->rrl_rule != NULL) 1836 rctl_rule_release(newlink->rrl_rule); 1837 uma_zfree(rctl_rule_link_zone, newlink); 1838 } 1839 1840 goto again; 1841} 1842 1843/* 1844 * Assign RCTL rules to the newly created process. 1845 */ 1846int 1847rctl_proc_fork(struct proc *parent, struct proc *child) 1848{ 1849 int error; 1850 struct rctl_rule_link *link; 1851 struct rctl_rule *rule; 1852 1853 LIST_INIT(&child->p_racct->r_rule_links); 1854 1855 ASSERT_RACCT_ENABLED(); 1856 KASSERT(parent->p_racct != NULL, ("process without racct; p = %p", parent)); 1857 1858 rw_wlock(&rctl_lock); 1859 1860 /* 1861 * Go through limits applicable to the parent and assign them 1862 * to the child. Rules with 'process' subject have to be duplicated 1863 * in order to make their rr_subject point to the new process. 1864 */ 1865 LIST_FOREACH(link, &parent->p_racct->r_rule_links, rrl_next) { 1866 if (link->rrl_rule->rr_subject_type == 1867 RCTL_SUBJECT_TYPE_PROCESS) { 1868 rule = rctl_rule_duplicate(link->rrl_rule, M_NOWAIT); 1869 if (rule == NULL) 1870 goto fail; 1871 KASSERT(rule->rr_subject.rs_proc == parent, 1872 ("rule->rr_subject.rs_proc != parent")); 1873 rule->rr_subject.rs_proc = child; 1874 error = rctl_racct_add_rule_locked(child->p_racct, 1875 rule); 1876 rctl_rule_release(rule); 1877 if (error != 0) 1878 goto fail; 1879 } else { 1880 error = rctl_racct_add_rule_locked(child->p_racct, 1881 link->rrl_rule); 1882 if (error != 0) 1883 goto fail; 1884 } 1885 } 1886 1887 rw_wunlock(&rctl_lock); 1888 return (0); 1889 1890fail: 1891 while (!LIST_EMPTY(&child->p_racct->r_rule_links)) { 1892 link = LIST_FIRST(&child->p_racct->r_rule_links); 1893 LIST_REMOVE(link, rrl_next); 1894 rctl_rule_release(link->rrl_rule); 1895 uma_zfree(rctl_rule_link_zone, link); 1896 } 1897 rw_wunlock(&rctl_lock); 1898 return (EAGAIN); 1899} 1900 1901/* 1902 * Release rules attached to the racct. 1903 */ 1904void 1905rctl_racct_release(struct racct *racct) 1906{ 1907 struct rctl_rule_link *link; 1908 1909 ASSERT_RACCT_ENABLED(); 1910 1911 rw_wlock(&rctl_lock); 1912 while (!LIST_EMPTY(&racct->r_rule_links)) { 1913 link = LIST_FIRST(&racct->r_rule_links); 1914 LIST_REMOVE(link, rrl_next); 1915 rctl_rule_release(link->rrl_rule); 1916 uma_zfree(rctl_rule_link_zone, link); 1917 } 1918 rw_wunlock(&rctl_lock); 1919} 1920 1921static void 1922rctl_init(void) 1923{ 1924 1925 if (!racct_enable) 1926 return; 1927 1928 rctl_rule_link_zone = uma_zcreate("rctl_rule_link", 1929 sizeof(struct rctl_rule_link), NULL, NULL, NULL, NULL, 1930 UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1931 rctl_rule_zone = uma_zcreate("rctl_rule", sizeof(struct rctl_rule), 1932 NULL, NULL, NULL, NULL, UMA_ALIGN_PTR, UMA_ZONE_NOFREE); 1933} 1934 1935#else /* !RCTL */ 1936 1937int 1938sys_rctl_get_racct(struct thread *td, struct rctl_get_racct_args *uap) 1939{ 1940 1941 return (ENOSYS); 1942} 1943 1944int 1945sys_rctl_get_rules(struct thread *td, struct rctl_get_rules_args *uap) 1946{ 1947 1948 return (ENOSYS); 1949} 1950 1951int 1952sys_rctl_get_limits(struct thread *td, struct rctl_get_limits_args *uap) 1953{ 1954 1955 return (ENOSYS); 1956} 1957 1958int 1959sys_rctl_add_rule(struct thread *td, struct rctl_add_rule_args *uap) 1960{ 1961 1962 return (ENOSYS); 1963} 1964 1965int 1966sys_rctl_remove_rule(struct thread *td, struct rctl_remove_rule_args *uap) 1967{ 1968 1969 return (ENOSYS); 1970} 1971 1972#endif /* !RCTL */ 1973