1 /*- 2 * Copyright (c) 1982, 1986, 1991, 1993 3 * The Regents of the University of California. All rights reserved. 4 * (c) UNIX System Laboratories, Inc. 5 * All or some portions of this file are derived from material licensed 6 * to the University of California by American Telephone and Telegraph 7 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 8 * the permission of UNIX System Laboratories, Inc. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 4. Neither the name of the University nor the names of its contributors 19 * may be used to endorse or promote products derived from this software 20 * without specific prior written permission. 21 * 22 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 23 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 24 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 25 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 26 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 27 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 28 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 29 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 30 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 31 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 32 * SUCH DAMAGE. 33 * 34 * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94 35 */ 36 37 #include <sys/cdefs.h> 38 __FBSDID("$FreeBSD$"); 39 40 #include "opt_compat.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/sysproto.h> 45 #include <sys/file.h> 46 #include <sys/kernel.h> 47 #include <sys/lock.h> 48 #include <sys/malloc.h> 49 #include <sys/mutex.h> 50 #include <sys/priv.h> 51 #include <sys/proc.h> 52 #include <sys/refcount.h> 53 #include <sys/racct.h> 54 #include <sys/resourcevar.h> 55 #include <sys/rwlock.h> 56 #include <sys/sched.h> 57 #include <sys/sx.h> 58 #include <sys/syscallsubr.h> 59 #include <sys/sysctl.h> 60 #include <sys/sysent.h> 61 #include <sys/time.h> 62 #include <sys/umtx.h> 63 64 #include <vm/vm.h> 65 #include <vm/vm_param.h> 66 #include <vm/pmap.h> 67 #include <vm/vm_map.h> 68 69 70 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures"); 71 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures"); 72 #define UIHASH(uid) (&uihashtbl[(uid) & uihash]) 73 static struct rwlock uihashtbl_lock; 74 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl; 75 static u_long uihash; /* size of hash table - 1 */ 76 77 static void calcru1(struct proc *p, struct rusage_ext *ruxp, 78 struct timeval *up, struct timeval *sp); 79 static int donice(struct thread *td, struct proc *chgp, int n); 80 static struct uidinfo *uilookup(uid_t uid); 81 static void ruxagg_locked(struct rusage_ext *rux, struct thread *td); 82 83 /* 84 * Resource controls and accounting. 85 */ 86 #ifndef _SYS_SYSPROTO_H_ 87 struct getpriority_args { 88 int which; 89 int who; 90 }; 91 #endif 92 int 93 sys_getpriority(td, uap) 94 struct thread *td; 95 register struct getpriority_args *uap; 96 { 97 struct proc *p; 98 struct pgrp *pg; 99 int error, low; 100 101 error = 0; 102 low = PRIO_MAX + 1; 103 switch (uap->which) { 104 105 case PRIO_PROCESS: 106 if (uap->who == 0) 107 low = td->td_proc->p_nice; 108 else { 109 p = pfind(uap->who); 110 if (p == NULL) 111 break; 112 if (p_cansee(td, p) == 0) 113 low = p->p_nice; 114 PROC_UNLOCK(p); 115 } 116 break; 117 118 case PRIO_PGRP: 119 sx_slock(&proctree_lock); 120 if (uap->who == 0) { 121 pg = td->td_proc->p_pgrp; 122 PGRP_LOCK(pg); 123 } else { 124 pg = pgfind(uap->who); 125 if (pg == NULL) { 126 sx_sunlock(&proctree_lock); 127 break; 128 } 129 } 130 sx_sunlock(&proctree_lock); 131 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 132 PROC_LOCK(p); 133 if (p->p_state == PRS_NORMAL && 134 p_cansee(td, p) == 0) { 135 if (p->p_nice < low) 136 low = p->p_nice; 137 } 138 PROC_UNLOCK(p); 139 } 140 PGRP_UNLOCK(pg); 141 break; 142 143 case PRIO_USER: 144 if (uap->who == 0) 145 uap->who = td->td_ucred->cr_uid; 146 sx_slock(&allproc_lock); 147 FOREACH_PROC_IN_SYSTEM(p) { 148 PROC_LOCK(p); 149 if (p->p_state == PRS_NORMAL && 150 p_cansee(td, p) == 0 && 151 p->p_ucred->cr_uid == uap->who) { 152 if (p->p_nice < low) 153 low = p->p_nice; 154 } 155 PROC_UNLOCK(p); 156 } 157 sx_sunlock(&allproc_lock); 158 break; 159 160 default: 161 error = EINVAL; 162 break; 163 } 164 if (low == PRIO_MAX + 1 && error == 0) 165 error = ESRCH; 166 td->td_retval[0] = low; 167 return (error); 168 } 169 170 #ifndef _SYS_SYSPROTO_H_ 171 struct setpriority_args { 172 int which; 173 int who; 174 int prio; 175 }; 176 #endif 177 int 178 sys_setpriority(td, uap) 179 struct thread *td; 180 struct setpriority_args *uap; 181 { 182 struct proc *curp, *p; 183 struct pgrp *pg; 184 int found = 0, error = 0; 185 186 curp = td->td_proc; 187 switch (uap->which) { 188 case PRIO_PROCESS: 189 if (uap->who == 0) { 190 PROC_LOCK(curp); 191 error = donice(td, curp, uap->prio); 192 PROC_UNLOCK(curp); 193 } else { 194 p = pfind(uap->who); 195 if (p == NULL) 196 break; 197 error = p_cansee(td, p); 198 if (error == 0) 199 error = donice(td, p, uap->prio); 200 PROC_UNLOCK(p); 201 } 202 found++; 203 break; 204 205 case PRIO_PGRP: 206 sx_slock(&proctree_lock); 207 if (uap->who == 0) { 208 pg = curp->p_pgrp; 209 PGRP_LOCK(pg); 210 } else { 211 pg = pgfind(uap->who); 212 if (pg == NULL) { 213 sx_sunlock(&proctree_lock); 214 break; 215 } 216 } 217 sx_sunlock(&proctree_lock); 218 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 219 PROC_LOCK(p); 220 if (p->p_state == PRS_NORMAL && 221 p_cansee(td, p) == 0) { 222 error = donice(td, p, uap->prio); 223 found++; 224 } 225 PROC_UNLOCK(p); 226 } 227 PGRP_UNLOCK(pg); 228 break; 229 230 case PRIO_USER: 231 if (uap->who == 0) 232 uap->who = td->td_ucred->cr_uid; 233 sx_slock(&allproc_lock); 234 FOREACH_PROC_IN_SYSTEM(p) { 235 PROC_LOCK(p); 236 if (p->p_state == PRS_NORMAL && 237 p->p_ucred->cr_uid == uap->who && 238 p_cansee(td, p) == 0) { 239 error = donice(td, p, uap->prio); 240 found++; 241 } 242 PROC_UNLOCK(p); 243 } 244 sx_sunlock(&allproc_lock); 245 break; 246 247 default: 248 error = EINVAL; 249 break; 250 } 251 if (found == 0 && error == 0) 252 error = ESRCH; 253 return (error); 254 } 255 256 /* 257 * Set "nice" for a (whole) process. 258 */ 259 static int 260 donice(struct thread *td, struct proc *p, int n) 261 { 262 int error; 263 264 PROC_LOCK_ASSERT(p, MA_OWNED); 265 if ((error = p_cansched(td, p))) 266 return (error); 267 if (n > PRIO_MAX) 268 n = PRIO_MAX; 269 if (n < PRIO_MIN) 270 n = PRIO_MIN; 271 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0) 272 return (EACCES); 273 sched_nice(p, n); 274 return (0); 275 } 276 277 static int unprivileged_idprio; 278 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW, 279 &unprivileged_idprio, 0, "Allow non-root users to set an idle priority"); 280 281 /* 282 * Set realtime priority for LWP. 283 */ 284 #ifndef _SYS_SYSPROTO_H_ 285 struct rtprio_thread_args { 286 int function; 287 lwpid_t lwpid; 288 struct rtprio *rtp; 289 }; 290 #endif 291 int 292 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap) 293 { 294 struct proc *p; 295 struct rtprio rtp; 296 struct thread *td1; 297 int cierror, error; 298 299 /* Perform copyin before acquiring locks if needed. */ 300 if (uap->function == RTP_SET) 301 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 302 else 303 cierror = 0; 304 305 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) { 306 p = td->td_proc; 307 td1 = td; 308 PROC_LOCK(p); 309 } else { 310 /* Only look up thread in current process */ 311 td1 = tdfind(uap->lwpid, curproc->p_pid); 312 if (td1 == NULL) 313 return (ESRCH); 314 p = td1->td_proc; 315 } 316 317 switch (uap->function) { 318 case RTP_LOOKUP: 319 if ((error = p_cansee(td, p))) 320 break; 321 pri_to_rtp(td1, &rtp); 322 PROC_UNLOCK(p); 323 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 324 case RTP_SET: 325 if ((error = p_cansched(td, p)) || (error = cierror)) 326 break; 327 328 /* Disallow setting rtprio in most cases if not superuser. */ 329 330 /* 331 * Realtime priority has to be restricted for reasons which 332 * should be obvious. However, for idleprio processes, there is 333 * a potential for system deadlock if an idleprio process gains 334 * a lock on a resource that other processes need (and the 335 * idleprio process can't run due to a CPU-bound normal 336 * process). Fix me! XXX 337 * 338 * This problem is not only related to idleprio process. 339 * A user level program can obtain a file lock and hold it 340 * indefinitely. Additionally, without idleprio processes it is 341 * still conceivable that a program with low priority will never 342 * get to run. In short, allowing this feature might make it 343 * easier to lock a resource indefinitely, but it is not the 344 * only thing that makes it possible. 345 */ 346 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || 347 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && 348 unprivileged_idprio == 0)) { 349 error = priv_check(td, PRIV_SCHED_RTPRIO); 350 if (error) 351 break; 352 } 353 error = rtp_to_pri(&rtp, td1); 354 break; 355 default: 356 error = EINVAL; 357 break; 358 } 359 PROC_UNLOCK(p); 360 return (error); 361 } 362 363 /* 364 * Set realtime priority. 365 */ 366 #ifndef _SYS_SYSPROTO_H_ 367 struct rtprio_args { 368 int function; 369 pid_t pid; 370 struct rtprio *rtp; 371 }; 372 #endif 373 int 374 sys_rtprio(td, uap) 375 struct thread *td; /* curthread */ 376 register struct rtprio_args *uap; 377 { 378 struct proc *p; 379 struct thread *tdp; 380 struct rtprio rtp; 381 int cierror, error; 382 383 /* Perform copyin before acquiring locks if needed. */ 384 if (uap->function == RTP_SET) 385 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 386 else 387 cierror = 0; 388 389 if (uap->pid == 0) { 390 p = td->td_proc; 391 PROC_LOCK(p); 392 } else { 393 p = pfind(uap->pid); 394 if (p == NULL) 395 return (ESRCH); 396 } 397 398 switch (uap->function) { 399 case RTP_LOOKUP: 400 if ((error = p_cansee(td, p))) 401 break; 402 /* 403 * Return OUR priority if no pid specified, 404 * or if one is, report the highest priority 405 * in the process. There isn't much more you can do as 406 * there is only room to return a single priority. 407 * Note: specifying our own pid is not the same 408 * as leaving it zero. 409 */ 410 if (uap->pid == 0) { 411 pri_to_rtp(td, &rtp); 412 } else { 413 struct rtprio rtp2; 414 415 rtp.type = RTP_PRIO_IDLE; 416 rtp.prio = RTP_PRIO_MAX; 417 FOREACH_THREAD_IN_PROC(p, tdp) { 418 pri_to_rtp(tdp, &rtp2); 419 if (rtp2.type < rtp.type || 420 (rtp2.type == rtp.type && 421 rtp2.prio < rtp.prio)) { 422 rtp.type = rtp2.type; 423 rtp.prio = rtp2.prio; 424 } 425 } 426 } 427 PROC_UNLOCK(p); 428 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 429 case RTP_SET: 430 if ((error = p_cansched(td, p)) || (error = cierror)) 431 break; 432 433 /* 434 * Disallow setting rtprio in most cases if not superuser. 435 * See the comment in sys_rtprio_thread about idprio 436 * threads holding a lock. 437 */ 438 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || 439 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && 440 !unprivileged_idprio)) { 441 error = priv_check(td, PRIV_SCHED_RTPRIO); 442 if (error) 443 break; 444 } 445 446 /* 447 * If we are setting our own priority, set just our 448 * thread but if we are doing another process, 449 * do all the threads on that process. If we 450 * specify our own pid we do the latter. 451 */ 452 if (uap->pid == 0) { 453 error = rtp_to_pri(&rtp, td); 454 } else { 455 FOREACH_THREAD_IN_PROC(p, td) { 456 if ((error = rtp_to_pri(&rtp, td)) != 0) 457 break; 458 } 459 } 460 break; 461 default: 462 error = EINVAL; 463 break; 464 } 465 PROC_UNLOCK(p); 466 return (error); 467 } 468 469 int 470 rtp_to_pri(struct rtprio *rtp, struct thread *td) 471 { 472 u_char newpri; 473 u_char oldpri; 474 475 switch (RTP_PRIO_BASE(rtp->type)) { 476 case RTP_PRIO_REALTIME: 477 if (rtp->prio > RTP_PRIO_MAX) 478 return (EINVAL); 479 newpri = PRI_MIN_REALTIME + rtp->prio; 480 break; 481 case RTP_PRIO_NORMAL: 482 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) 483 return (EINVAL); 484 newpri = PRI_MIN_TIMESHARE + rtp->prio; 485 break; 486 case RTP_PRIO_IDLE: 487 if (rtp->prio > RTP_PRIO_MAX) 488 return (EINVAL); 489 newpri = PRI_MIN_IDLE + rtp->prio; 490 break; 491 default: 492 return (EINVAL); 493 } 494 495 thread_lock(td); 496 sched_class(td, rtp->type); /* XXX fix */ 497 oldpri = td->td_user_pri; 498 sched_user_prio(td, newpri); 499 if (td->td_user_pri != oldpri && (td == curthread || 500 td->td_priority == oldpri || td->td_user_pri <= PRI_MAX_REALTIME)) 501 sched_prio(td, td->td_user_pri); 502 if (TD_ON_UPILOCK(td) && oldpri != newpri) { 503 critical_enter(); 504 thread_unlock(td); 505 umtx_pi_adjust(td, oldpri); 506 critical_exit(); 507 } else 508 thread_unlock(td); 509 return (0); 510 } 511 512 void 513 pri_to_rtp(struct thread *td, struct rtprio *rtp) 514 { 515 516 thread_lock(td); 517 switch (PRI_BASE(td->td_pri_class)) { 518 case PRI_REALTIME: 519 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME; 520 break; 521 case PRI_TIMESHARE: 522 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE; 523 break; 524 case PRI_IDLE: 525 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE; 526 break; 527 default: 528 break; 529 } 530 rtp->type = td->td_pri_class; 531 thread_unlock(td); 532 } 533 534 #if defined(COMPAT_43) 535 #ifndef _SYS_SYSPROTO_H_ 536 struct osetrlimit_args { 537 u_int which; 538 struct orlimit *rlp; 539 }; 540 #endif 541 int 542 osetrlimit(td, uap) 543 struct thread *td; 544 register struct osetrlimit_args *uap; 545 { 546 struct orlimit olim; 547 struct rlimit lim; 548 int error; 549 550 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit)))) 551 return (error); 552 lim.rlim_cur = olim.rlim_cur; 553 lim.rlim_max = olim.rlim_max; 554 error = kern_setrlimit(td, uap->which, &lim); 555 return (error); 556 } 557 558 #ifndef _SYS_SYSPROTO_H_ 559 struct ogetrlimit_args { 560 u_int which; 561 struct orlimit *rlp; 562 }; 563 #endif 564 int 565 ogetrlimit(td, uap) 566 struct thread *td; 567 register struct ogetrlimit_args *uap; 568 { 569 struct orlimit olim; 570 struct rlimit rl; 571 struct proc *p; 572 int error; 573 574 if (uap->which >= RLIM_NLIMITS) 575 return (EINVAL); 576 p = td->td_proc; 577 PROC_LOCK(p); 578 lim_rlimit(p, uap->which, &rl); 579 PROC_UNLOCK(p); 580 581 /* 582 * XXX would be more correct to convert only RLIM_INFINITY to the 583 * old RLIM_INFINITY and fail with EOVERFLOW for other larger 584 * values. Most 64->32 and 32->16 conversions, including not 585 * unimportant ones of uids are even more broken than what we 586 * do here (they blindly truncate). We don't do this correctly 587 * here since we have little experience with EOVERFLOW yet. 588 * Elsewhere, getuid() can't fail... 589 */ 590 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur; 591 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max; 592 error = copyout(&olim, uap->rlp, sizeof(olim)); 593 return (error); 594 } 595 #endif /* COMPAT_43 */ 596 597 #ifndef _SYS_SYSPROTO_H_ 598 struct __setrlimit_args { 599 u_int which; 600 struct rlimit *rlp; 601 }; 602 #endif 603 int 604 sys_setrlimit(td, uap) 605 struct thread *td; 606 register struct __setrlimit_args *uap; 607 { 608 struct rlimit alim; 609 int error; 610 611 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit)))) 612 return (error); 613 error = kern_setrlimit(td, uap->which, &alim); 614 return (error); 615 } 616 617 static void 618 lim_cb(void *arg) 619 { 620 struct rlimit rlim; 621 struct thread *td; 622 struct proc *p; 623 624 p = arg; 625 PROC_LOCK_ASSERT(p, MA_OWNED); 626 /* 627 * Check if the process exceeds its cpu resource allocation. If 628 * it reaches the max, arrange to kill the process in ast(). 629 */ 630 if (p->p_cpulimit == RLIM_INFINITY) 631 return; 632 PROC_SLOCK(p); 633 FOREACH_THREAD_IN_PROC(p, td) { 634 ruxagg(p, td); 635 } 636 PROC_SUNLOCK(p); 637 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) { 638 lim_rlimit(p, RLIMIT_CPU, &rlim); 639 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) { 640 killproc(p, "exceeded maximum CPU limit"); 641 } else { 642 if (p->p_cpulimit < rlim.rlim_max) 643 p->p_cpulimit += 5; 644 kern_psignal(p, SIGXCPU); 645 } 646 } 647 if ((p->p_flag & P_WEXIT) == 0) 648 callout_reset(&p->p_limco, hz, lim_cb, p); 649 } 650 651 int 652 kern_setrlimit(td, which, limp) 653 struct thread *td; 654 u_int which; 655 struct rlimit *limp; 656 { 657 struct plimit *newlim, *oldlim; 658 struct proc *p; 659 register struct rlimit *alimp; 660 struct rlimit oldssiz; 661 int error; 662 663 if (which >= RLIM_NLIMITS) 664 return (EINVAL); 665 666 /* 667 * Preserve historical bugs by treating negative limits as unsigned. 668 */ 669 if (limp->rlim_cur < 0) 670 limp->rlim_cur = RLIM_INFINITY; 671 if (limp->rlim_max < 0) 672 limp->rlim_max = RLIM_INFINITY; 673 674 oldssiz.rlim_cur = 0; 675 p = td->td_proc; 676 newlim = lim_alloc(); 677 PROC_LOCK(p); 678 oldlim = p->p_limit; 679 alimp = &oldlim->pl_rlimit[which]; 680 if (limp->rlim_cur > alimp->rlim_max || 681 limp->rlim_max > alimp->rlim_max) 682 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { 683 PROC_UNLOCK(p); 684 lim_free(newlim); 685 return (error); 686 } 687 if (limp->rlim_cur > limp->rlim_max) 688 limp->rlim_cur = limp->rlim_max; 689 lim_copy(newlim, oldlim); 690 alimp = &newlim->pl_rlimit[which]; 691 692 switch (which) { 693 694 case RLIMIT_CPU: 695 if (limp->rlim_cur != RLIM_INFINITY && 696 p->p_cpulimit == RLIM_INFINITY) 697 callout_reset(&p->p_limco, hz, lim_cb, p); 698 p->p_cpulimit = limp->rlim_cur; 699 break; 700 case RLIMIT_DATA: 701 if (limp->rlim_cur > maxdsiz) 702 limp->rlim_cur = maxdsiz; 703 if (limp->rlim_max > maxdsiz) 704 limp->rlim_max = maxdsiz; 705 break; 706 707 case RLIMIT_STACK: 708 if (limp->rlim_cur > maxssiz) 709 limp->rlim_cur = maxssiz; 710 if (limp->rlim_max > maxssiz) 711 limp->rlim_max = maxssiz; 712 oldssiz = *alimp; 713 if (p->p_sysent->sv_fixlimit != NULL) 714 p->p_sysent->sv_fixlimit(&oldssiz, 715 RLIMIT_STACK); 716 break; 717 718 case RLIMIT_NOFILE: 719 if (limp->rlim_cur > maxfilesperproc) 720 limp->rlim_cur = maxfilesperproc; 721 if (limp->rlim_max > maxfilesperproc) 722 limp->rlim_max = maxfilesperproc; 723 break; 724 725 case RLIMIT_NPROC: 726 if (limp->rlim_cur > maxprocperuid) 727 limp->rlim_cur = maxprocperuid; 728 if (limp->rlim_max > maxprocperuid) 729 limp->rlim_max = maxprocperuid; 730 if (limp->rlim_cur < 1) 731 limp->rlim_cur = 1; 732 if (limp->rlim_max < 1) 733 limp->rlim_max = 1; 734 break; 735 } 736 if (p->p_sysent->sv_fixlimit != NULL) 737 p->p_sysent->sv_fixlimit(limp, which); 738 *alimp = *limp; 739 p->p_limit = newlim; 740 PROC_UNLOCK(p); 741 lim_free(oldlim); 742 743 if (which == RLIMIT_STACK) { 744 /* 745 * Stack is allocated to the max at exec time with only 746 * "rlim_cur" bytes accessible. If stack limit is going 747 * up make more accessible, if going down make inaccessible. 748 */ 749 if (limp->rlim_cur != oldssiz.rlim_cur) { 750 vm_offset_t addr; 751 vm_size_t size; 752 vm_prot_t prot; 753 754 if (limp->rlim_cur > oldssiz.rlim_cur) { 755 prot = p->p_sysent->sv_stackprot; 756 size = limp->rlim_cur - oldssiz.rlim_cur; 757 addr = p->p_sysent->sv_usrstack - 758 limp->rlim_cur; 759 } else { 760 prot = VM_PROT_NONE; 761 size = oldssiz.rlim_cur - limp->rlim_cur; 762 addr = p->p_sysent->sv_usrstack - 763 oldssiz.rlim_cur; 764 } 765 addr = trunc_page(addr); 766 size = round_page(size); 767 (void)vm_map_protect(&p->p_vmspace->vm_map, 768 addr, addr + size, prot, FALSE); 769 } 770 } 771 772 return (0); 773 } 774 775 #ifndef _SYS_SYSPROTO_H_ 776 struct __getrlimit_args { 777 u_int which; 778 struct rlimit *rlp; 779 }; 780 #endif 781 /* ARGSUSED */ 782 int 783 sys_getrlimit(td, uap) 784 struct thread *td; 785 register struct __getrlimit_args *uap; 786 { 787 struct rlimit rlim; 788 struct proc *p; 789 int error; 790 791 if (uap->which >= RLIM_NLIMITS) 792 return (EINVAL); 793 p = td->td_proc; 794 PROC_LOCK(p); 795 lim_rlimit(p, uap->which, &rlim); 796 PROC_UNLOCK(p); 797 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); 798 return (error); 799 } 800 801 /* 802 * Transform the running time and tick information for children of proc p 803 * into user and system time usage. 804 */ 805 void 806 calccru(p, up, sp) 807 struct proc *p; 808 struct timeval *up; 809 struct timeval *sp; 810 { 811 812 PROC_LOCK_ASSERT(p, MA_OWNED); 813 calcru1(p, &p->p_crux, up, sp); 814 } 815 816 /* 817 * Transform the running time and tick information in proc p into user 818 * and system time usage. If appropriate, include the current time slice 819 * on this CPU. 820 */ 821 void 822 calcru(struct proc *p, struct timeval *up, struct timeval *sp) 823 { 824 struct thread *td; 825 uint64_t runtime, u; 826 827 PROC_LOCK_ASSERT(p, MA_OWNED); 828 PROC_SLOCK_ASSERT(p, MA_OWNED); 829 /* 830 * If we are getting stats for the current process, then add in the 831 * stats that this thread has accumulated in its current time slice. 832 * We reset the thread and CPU state as if we had performed a context 833 * switch right here. 834 */ 835 td = curthread; 836 if (td->td_proc == p) { 837 u = cpu_ticks(); 838 runtime = u - PCPU_GET(switchtime); 839 td->td_runtime += runtime; 840 td->td_incruntime += runtime; 841 PCPU_SET(switchtime, u); 842 } 843 /* Make sure the per-thread stats are current. */ 844 FOREACH_THREAD_IN_PROC(p, td) { 845 if (td->td_incruntime == 0) 846 continue; 847 ruxagg(p, td); 848 } 849 calcru1(p, &p->p_rux, up, sp); 850 } 851 852 /* Collect resource usage for a single thread. */ 853 void 854 rufetchtd(struct thread *td, struct rusage *ru) 855 { 856 struct proc *p; 857 uint64_t runtime, u; 858 859 p = td->td_proc; 860 PROC_SLOCK_ASSERT(p, MA_OWNED); 861 THREAD_LOCK_ASSERT(td, MA_OWNED); 862 /* 863 * If we are getting stats for the current thread, then add in the 864 * stats that this thread has accumulated in its current time slice. 865 * We reset the thread and CPU state as if we had performed a context 866 * switch right here. 867 */ 868 if (td == curthread) { 869 u = cpu_ticks(); 870 runtime = u - PCPU_GET(switchtime); 871 td->td_runtime += runtime; 872 td->td_incruntime += runtime; 873 PCPU_SET(switchtime, u); 874 } 875 ruxagg(p, td); 876 *ru = td->td_ru; 877 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); 878 } 879 880 static void 881 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, 882 struct timeval *sp) 883 { 884 /* {user, system, interrupt, total} {ticks, usec}: */ 885 uint64_t ut, uu, st, su, it, tt, tu; 886 887 ut = ruxp->rux_uticks; 888 st = ruxp->rux_sticks; 889 it = ruxp->rux_iticks; 890 tt = ut + st + it; 891 if (tt == 0) { 892 /* Avoid divide by zero */ 893 st = 1; 894 tt = 1; 895 } 896 tu = cputick2usec(ruxp->rux_runtime); 897 if ((int64_t)tu < 0) { 898 /* XXX: this should be an assert /phk */ 899 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", 900 (intmax_t)tu, p->p_pid, p->p_comm); 901 tu = ruxp->rux_tu; 902 } 903 904 if (tu >= ruxp->rux_tu) { 905 /* 906 * The normal case, time increased. 907 * Enforce monotonicity of bucketed numbers. 908 */ 909 uu = (tu * ut) / tt; 910 if (uu < ruxp->rux_uu) 911 uu = ruxp->rux_uu; 912 su = (tu * st) / tt; 913 if (su < ruxp->rux_su) 914 su = ruxp->rux_su; 915 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { 916 /* 917 * When we calibrate the cputicker, it is not uncommon to 918 * see the presumably fixed frequency increase slightly over 919 * time as a result of thermal stabilization and NTP 920 * discipline (of the reference clock). We therefore ignore 921 * a bit of backwards slop because we expect to catch up 922 * shortly. We use a 3 microsecond limit to catch low 923 * counts and a 1% limit for high counts. 924 */ 925 uu = ruxp->rux_uu; 926 su = ruxp->rux_su; 927 tu = ruxp->rux_tu; 928 } else { /* tu < ruxp->rux_tu */ 929 /* 930 * What happened here was likely that a laptop, which ran at 931 * a reduced clock frequency at boot, kicked into high gear. 932 * The wisdom of spamming this message in that case is 933 * dubious, but it might also be indicative of something 934 * serious, so lets keep it and hope laptops can be made 935 * more truthful about their CPU speed via ACPI. 936 */ 937 printf("calcru: runtime went backwards from %ju usec " 938 "to %ju usec for pid %d (%s)\n", 939 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, 940 p->p_pid, p->p_comm); 941 uu = (tu * ut) / tt; 942 su = (tu * st) / tt; 943 } 944 945 ruxp->rux_uu = uu; 946 ruxp->rux_su = su; 947 ruxp->rux_tu = tu; 948 949 up->tv_sec = uu / 1000000; 950 up->tv_usec = uu % 1000000; 951 sp->tv_sec = su / 1000000; 952 sp->tv_usec = su % 1000000; 953 } 954 955 #ifndef _SYS_SYSPROTO_H_ 956 struct getrusage_args { 957 int who; 958 struct rusage *rusage; 959 }; 960 #endif 961 int 962 sys_getrusage(td, uap) 963 register struct thread *td; 964 register struct getrusage_args *uap; 965 { 966 struct rusage ru; 967 int error; 968 969 error = kern_getrusage(td, uap->who, &ru); 970 if (error == 0) 971 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 972 return (error); 973 } 974 975 int 976 kern_getrusage(struct thread *td, int who, struct rusage *rup) 977 { 978 struct proc *p; 979 int error; 980 981 error = 0; 982 p = td->td_proc; 983 PROC_LOCK(p); 984 switch (who) { 985 case RUSAGE_SELF: 986 rufetchcalc(p, rup, &rup->ru_utime, 987 &rup->ru_stime); 988 break; 989 990 case RUSAGE_CHILDREN: 991 *rup = p->p_stats->p_cru; 992 calccru(p, &rup->ru_utime, &rup->ru_stime); 993 break; 994 995 case RUSAGE_THREAD: 996 PROC_SLOCK(p); 997 thread_lock(td); 998 rufetchtd(td, rup); 999 thread_unlock(td); 1000 PROC_SUNLOCK(p); 1001 break; 1002 1003 default: 1004 error = EINVAL; 1005 } 1006 PROC_UNLOCK(p); 1007 return (error); 1008 } 1009 1010 void 1011 rucollect(struct rusage *ru, struct rusage *ru2) 1012 { 1013 long *ip, *ip2; 1014 int i; 1015 1016 if (ru->ru_maxrss < ru2->ru_maxrss) 1017 ru->ru_maxrss = ru2->ru_maxrss; 1018 ip = &ru->ru_first; 1019 ip2 = &ru2->ru_first; 1020 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) 1021 *ip++ += *ip2++; 1022 } 1023 1024 void 1025 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, 1026 struct rusage_ext *rux2) 1027 { 1028 1029 rux->rux_runtime += rux2->rux_runtime; 1030 rux->rux_uticks += rux2->rux_uticks; 1031 rux->rux_sticks += rux2->rux_sticks; 1032 rux->rux_iticks += rux2->rux_iticks; 1033 rux->rux_uu += rux2->rux_uu; 1034 rux->rux_su += rux2->rux_su; 1035 rux->rux_tu += rux2->rux_tu; 1036 rucollect(ru, ru2); 1037 } 1038 1039 /* 1040 * Aggregate tick counts into the proc's rusage_ext. 1041 */ 1042 static void 1043 ruxagg_locked(struct rusage_ext *rux, struct thread *td) 1044 { 1045 1046 THREAD_LOCK_ASSERT(td, MA_OWNED); 1047 PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED); 1048 rux->rux_runtime += td->td_incruntime; 1049 rux->rux_uticks += td->td_uticks; 1050 rux->rux_sticks += td->td_sticks; 1051 rux->rux_iticks += td->td_iticks; 1052 } 1053 1054 void 1055 ruxagg(struct proc *p, struct thread *td) 1056 { 1057 1058 thread_lock(td); 1059 ruxagg_locked(&p->p_rux, td); 1060 ruxagg_locked(&td->td_rux, td); 1061 td->td_incruntime = 0; 1062 td->td_uticks = 0; 1063 td->td_iticks = 0; 1064 td->td_sticks = 0; 1065 thread_unlock(td); 1066 } 1067 1068 /* 1069 * Update the rusage_ext structure and fetch a valid aggregate rusage 1070 * for proc p if storage for one is supplied. 1071 */ 1072 void 1073 rufetch(struct proc *p, struct rusage *ru) 1074 { 1075 struct thread *td; 1076 1077 PROC_SLOCK_ASSERT(p, MA_OWNED); 1078 1079 *ru = p->p_ru; 1080 if (p->p_numthreads > 0) { 1081 FOREACH_THREAD_IN_PROC(p, td) { 1082 ruxagg(p, td); 1083 rucollect(ru, &td->td_ru); 1084 } 1085 } 1086 } 1087 1088 /* 1089 * Atomically perform a rufetch and a calcru together. 1090 * Consumers, can safely assume the calcru is executed only once 1091 * rufetch is completed. 1092 */ 1093 void 1094 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, 1095 struct timeval *sp) 1096 { 1097 1098 PROC_SLOCK(p); 1099 rufetch(p, ru); 1100 calcru(p, up, sp); 1101 PROC_SUNLOCK(p); 1102 } 1103 1104 /* 1105 * Allocate a new resource limits structure and initialize its 1106 * reference count and mutex pointer. 1107 */ 1108 struct plimit * 1109 lim_alloc() 1110 { 1111 struct plimit *limp; 1112 1113 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); 1114 refcount_init(&limp->pl_refcnt, 1); 1115 return (limp); 1116 } 1117 1118 struct plimit * 1119 lim_hold(limp) 1120 struct plimit *limp; 1121 { 1122 1123 refcount_acquire(&limp->pl_refcnt); 1124 return (limp); 1125 } 1126 1127 void 1128 lim_fork(struct proc *p1, struct proc *p2) 1129 { 1130 1131 PROC_LOCK_ASSERT(p1, MA_OWNED); 1132 PROC_LOCK_ASSERT(p2, MA_OWNED); 1133 1134 p2->p_limit = lim_hold(p1->p_limit); 1135 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); 1136 if (p1->p_cpulimit != RLIM_INFINITY) 1137 callout_reset(&p2->p_limco, hz, lim_cb, p2); 1138 } 1139 1140 void 1141 lim_free(limp) 1142 struct plimit *limp; 1143 { 1144 1145 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow")); 1146 if (refcount_release(&limp->pl_refcnt)) 1147 free((void *)limp, M_PLIMIT); 1148 } 1149 1150 /* 1151 * Make a copy of the plimit structure. 1152 * We share these structures copy-on-write after fork. 1153 */ 1154 void 1155 lim_copy(dst, src) 1156 struct plimit *dst, *src; 1157 { 1158 1159 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit")); 1160 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); 1161 } 1162 1163 /* 1164 * Return the hard limit for a particular system resource. The 1165 * which parameter specifies the index into the rlimit array. 1166 */ 1167 rlim_t 1168 lim_max(struct proc *p, int which) 1169 { 1170 struct rlimit rl; 1171 1172 lim_rlimit(p, which, &rl); 1173 return (rl.rlim_max); 1174 } 1175 1176 /* 1177 * Return the current (soft) limit for a particular system resource. 1178 * The which parameter which specifies the index into the rlimit array 1179 */ 1180 rlim_t 1181 lim_cur(struct proc *p, int which) 1182 { 1183 struct rlimit rl; 1184 1185 lim_rlimit(p, which, &rl); 1186 return (rl.rlim_cur); 1187 } 1188 1189 /* 1190 * Return a copy of the entire rlimit structure for the system limit 1191 * specified by 'which' in the rlimit structure pointed to by 'rlp'. 1192 */ 1193 void 1194 lim_rlimit(struct proc *p, int which, struct rlimit *rlp) 1195 { 1196 1197 PROC_LOCK_ASSERT(p, MA_OWNED); 1198 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1199 ("request for invalid resource limit")); 1200 *rlp = p->p_limit->pl_rlimit[which]; 1201 if (p->p_sysent->sv_fixlimit != NULL) 1202 p->p_sysent->sv_fixlimit(rlp, which); 1203 } 1204 1205 void 1206 uihashinit() 1207 { 1208 1209 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); 1210 rw_init(&uihashtbl_lock, "uidinfo hash"); 1211 } 1212 1213 /* 1214 * Look up a uidinfo struct for the parameter uid. 1215 * uihashtbl_lock must be locked. 1216 */ 1217 static struct uidinfo * 1218 uilookup(uid) 1219 uid_t uid; 1220 { 1221 struct uihashhead *uipp; 1222 struct uidinfo *uip; 1223 1224 rw_assert(&uihashtbl_lock, RA_LOCKED); 1225 uipp = UIHASH(uid); 1226 LIST_FOREACH(uip, uipp, ui_hash) 1227 if (uip->ui_uid == uid) 1228 break; 1229 1230 return (uip); 1231 } 1232 1233 /* 1234 * Find or allocate a struct uidinfo for a particular uid. 1235 * Increase refcount on uidinfo struct returned. 1236 * uifree() should be called on a struct uidinfo when released. 1237 */ 1238 struct uidinfo * 1239 uifind(uid) 1240 uid_t uid; 1241 { 1242 struct uidinfo *old_uip, *uip; 1243 1244 rw_rlock(&uihashtbl_lock); 1245 uip = uilookup(uid); 1246 if (uip == NULL) { 1247 rw_runlock(&uihashtbl_lock); 1248 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO); 1249 racct_create(&uip->ui_racct); 1250 rw_wlock(&uihashtbl_lock); 1251 /* 1252 * There's a chance someone created our uidinfo while we 1253 * were in malloc and not holding the lock, so we have to 1254 * make sure we don't insert a duplicate uidinfo. 1255 */ 1256 if ((old_uip = uilookup(uid)) != NULL) { 1257 /* Someone else beat us to it. */ 1258 racct_destroy(&uip->ui_racct); 1259 free(uip, M_UIDINFO); 1260 uip = old_uip; 1261 } else { 1262 refcount_init(&uip->ui_ref, 0); 1263 uip->ui_uid = uid; 1264 mtx_init(&uip->ui_vmsize_mtx, "ui_vmsize", NULL, 1265 MTX_DEF); 1266 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash); 1267 } 1268 } 1269 uihold(uip); 1270 rw_unlock(&uihashtbl_lock); 1271 return (uip); 1272 } 1273 1274 /* 1275 * Place another refcount on a uidinfo struct. 1276 */ 1277 void 1278 uihold(uip) 1279 struct uidinfo *uip; 1280 { 1281 1282 refcount_acquire(&uip->ui_ref); 1283 } 1284 1285 /*- 1286 * Since uidinfo structs have a long lifetime, we use an 1287 * opportunistic refcounting scheme to avoid locking the lookup hash 1288 * for each release. 1289 * 1290 * If the refcount hits 0, we need to free the structure, 1291 * which means we need to lock the hash. 1292 * Optimal case: 1293 * After locking the struct and lowering the refcount, if we find 1294 * that we don't need to free, simply unlock and return. 1295 * Suboptimal case: 1296 * If refcount lowering results in need to free, bump the count 1297 * back up, lose the lock and acquire the locks in the proper 1298 * order to try again. 1299 */ 1300 void 1301 uifree(uip) 1302 struct uidinfo *uip; 1303 { 1304 int old; 1305 1306 /* Prepare for optimal case. */ 1307 old = uip->ui_ref; 1308 if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1)) 1309 return; 1310 1311 /* Prepare for suboptimal case. */ 1312 rw_wlock(&uihashtbl_lock); 1313 if (refcount_release(&uip->ui_ref)) { 1314 racct_destroy(&uip->ui_racct); 1315 LIST_REMOVE(uip, ui_hash); 1316 rw_wunlock(&uihashtbl_lock); 1317 if (uip->ui_sbsize != 0) 1318 printf("freeing uidinfo: uid = %d, sbsize = %ld\n", 1319 uip->ui_uid, uip->ui_sbsize); 1320 if (uip->ui_proccnt != 0) 1321 printf("freeing uidinfo: uid = %d, proccnt = %ld\n", 1322 uip->ui_uid, uip->ui_proccnt); 1323 if (uip->ui_vmsize != 0) 1324 printf("freeing uidinfo: uid = %d, swapuse = %lld\n", 1325 uip->ui_uid, (unsigned long long)uip->ui_vmsize); 1326 mtx_destroy(&uip->ui_vmsize_mtx); 1327 free(uip, M_UIDINFO); 1328 return; 1329 } 1330 /* 1331 * Someone added a reference between atomic_cmpset_int() and 1332 * rw_wlock(&uihashtbl_lock). 1333 */ 1334 rw_wunlock(&uihashtbl_lock); 1335 } 1336 1337 void 1338 ui_racct_foreach(void (*callback)(struct racct *racct, 1339 void *arg2, void *arg3), void *arg2, void *arg3) 1340 { 1341 struct uidinfo *uip; 1342 struct uihashhead *uih; 1343 1344 rw_rlock(&uihashtbl_lock); 1345 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { 1346 LIST_FOREACH(uip, uih, ui_hash) { 1347 (callback)(uip->ui_racct, arg2, arg3); 1348 } 1349 } 1350 rw_runlock(&uihashtbl_lock); 1351 } 1352 1353 /* 1354 * Change the count associated with number of processes 1355 * a given user is using. When 'max' is 0, don't enforce a limit 1356 */ 1357 int 1358 chgproccnt(uip, diff, max) 1359 struct uidinfo *uip; 1360 int diff; 1361 rlim_t max; 1362 { 1363 1364 /* Don't allow them to exceed max, but allow subtraction. */ 1365 if (diff > 0 && max != 0) { 1366 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) { 1367 atomic_subtract_long(&uip->ui_proccnt, (long)diff); 1368 return (0); 1369 } 1370 } else { 1371 atomic_add_long(&uip->ui_proccnt, (long)diff); 1372 if (uip->ui_proccnt < 0) 1373 printf("negative proccnt for uid = %d\n", uip->ui_uid); 1374 } 1375 return (1); 1376 } 1377 1378 /* 1379 * Change the total socket buffer size a user has used. 1380 */ 1381 int 1382 chgsbsize(uip, hiwat, to, max) 1383 struct uidinfo *uip; 1384 u_int *hiwat; 1385 u_int to; 1386 rlim_t max; 1387 { 1388 int diff; 1389 1390 diff = to - *hiwat; 1391 if (diff > 0) { 1392 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) { 1393 atomic_subtract_long(&uip->ui_sbsize, (long)diff); 1394 return (0); 1395 } 1396 } else { 1397 atomic_add_long(&uip->ui_sbsize, (long)diff); 1398 if (uip->ui_sbsize < 0) 1399 printf("negative sbsize for uid = %d\n", uip->ui_uid); 1400 } 1401 *hiwat = to; 1402 return (1); 1403 } 1404 1405 /* 1406 * Change the count associated with number of pseudo-terminals 1407 * a given user is using. When 'max' is 0, don't enforce a limit 1408 */ 1409 int 1410 chgptscnt(uip, diff, max) 1411 struct uidinfo *uip; 1412 int diff; 1413 rlim_t max; 1414 { 1415 1416 /* Don't allow them to exceed max, but allow subtraction. */ 1417 if (diff > 0 && max != 0) { 1418 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) { 1419 atomic_subtract_long(&uip->ui_ptscnt, (long)diff); 1420 return (0); 1421 } 1422 } else { 1423 atomic_add_long(&uip->ui_ptscnt, (long)diff); 1424 if (uip->ui_ptscnt < 0) 1425 printf("negative ptscnt for uid = %d\n", uip->ui_uid); 1426 } 1427 return (1); 1428 } 1429