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(struct thread *td, u_int which, struct rlimit *limp) 653 { 654 655 return (kern_proc_setrlimit(td, td->td_proc, which, limp)); 656 } 657 658 int 659 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which, 660 struct rlimit *limp) 661 { 662 struct plimit *newlim, *oldlim; 663 register struct rlimit *alimp; 664 struct rlimit oldssiz; 665 int error; 666 667 if (which >= RLIM_NLIMITS) 668 return (EINVAL); 669 670 /* 671 * Preserve historical bugs by treating negative limits as unsigned. 672 */ 673 if (limp->rlim_cur < 0) 674 limp->rlim_cur = RLIM_INFINITY; 675 if (limp->rlim_max < 0) 676 limp->rlim_max = RLIM_INFINITY; 677 678 oldssiz.rlim_cur = 0; 679 newlim = lim_alloc(); 680 PROC_LOCK(p); 681 oldlim = p->p_limit; 682 alimp = &oldlim->pl_rlimit[which]; 683 if (limp->rlim_cur > alimp->rlim_max || 684 limp->rlim_max > alimp->rlim_max) 685 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { 686 PROC_UNLOCK(p); 687 lim_free(newlim); 688 return (error); 689 } 690 if (limp->rlim_cur > limp->rlim_max) 691 limp->rlim_cur = limp->rlim_max; 692 lim_copy(newlim, oldlim); 693 alimp = &newlim->pl_rlimit[which]; 694 695 switch (which) { 696 697 case RLIMIT_CPU: 698 if (limp->rlim_cur != RLIM_INFINITY && 699 p->p_cpulimit == RLIM_INFINITY) 700 callout_reset(&p->p_limco, hz, lim_cb, p); 701 p->p_cpulimit = limp->rlim_cur; 702 break; 703 case RLIMIT_DATA: 704 if (limp->rlim_cur > maxdsiz) 705 limp->rlim_cur = maxdsiz; 706 if (limp->rlim_max > maxdsiz) 707 limp->rlim_max = maxdsiz; 708 break; 709 710 case RLIMIT_STACK: 711 if (limp->rlim_cur > maxssiz) 712 limp->rlim_cur = maxssiz; 713 if (limp->rlim_max > maxssiz) 714 limp->rlim_max = maxssiz; 715 oldssiz = *alimp; 716 if (p->p_sysent->sv_fixlimit != NULL) 717 p->p_sysent->sv_fixlimit(&oldssiz, 718 RLIMIT_STACK); 719 break; 720 721 case RLIMIT_NOFILE: 722 if (limp->rlim_cur > maxfilesperproc) 723 limp->rlim_cur = maxfilesperproc; 724 if (limp->rlim_max > maxfilesperproc) 725 limp->rlim_max = maxfilesperproc; 726 break; 727 728 case RLIMIT_NPROC: 729 if (limp->rlim_cur > maxprocperuid) 730 limp->rlim_cur = maxprocperuid; 731 if (limp->rlim_max > maxprocperuid) 732 limp->rlim_max = maxprocperuid; 733 if (limp->rlim_cur < 1) 734 limp->rlim_cur = 1; 735 if (limp->rlim_max < 1) 736 limp->rlim_max = 1; 737 break; 738 } 739 if (p->p_sysent->sv_fixlimit != NULL) 740 p->p_sysent->sv_fixlimit(limp, which); 741 *alimp = *limp; 742 p->p_limit = newlim; 743 PROC_UNLOCK(p); 744 lim_free(oldlim); 745 746 if (which == RLIMIT_STACK) { 747 /* 748 * Stack is allocated to the max at exec time with only 749 * "rlim_cur" bytes accessible. If stack limit is going 750 * up make more accessible, if going down make inaccessible. 751 */ 752 if (limp->rlim_cur != oldssiz.rlim_cur) { 753 vm_offset_t addr; 754 vm_size_t size; 755 vm_prot_t prot; 756 757 if (limp->rlim_cur > oldssiz.rlim_cur) { 758 prot = p->p_sysent->sv_stackprot; 759 size = limp->rlim_cur - oldssiz.rlim_cur; 760 addr = p->p_sysent->sv_usrstack - 761 limp->rlim_cur; 762 } else { 763 prot = VM_PROT_NONE; 764 size = oldssiz.rlim_cur - limp->rlim_cur; 765 addr = p->p_sysent->sv_usrstack - 766 oldssiz.rlim_cur; 767 } 768 addr = trunc_page(addr); 769 size = round_page(size); 770 (void)vm_map_protect(&p->p_vmspace->vm_map, 771 addr, addr + size, prot, FALSE); 772 } 773 } 774 775 return (0); 776 } 777 778 #ifndef _SYS_SYSPROTO_H_ 779 struct __getrlimit_args { 780 u_int which; 781 struct rlimit *rlp; 782 }; 783 #endif 784 /* ARGSUSED */ 785 int 786 sys_getrlimit(td, uap) 787 struct thread *td; 788 register struct __getrlimit_args *uap; 789 { 790 struct rlimit rlim; 791 struct proc *p; 792 int error; 793 794 if (uap->which >= RLIM_NLIMITS) 795 return (EINVAL); 796 p = td->td_proc; 797 PROC_LOCK(p); 798 lim_rlimit(p, uap->which, &rlim); 799 PROC_UNLOCK(p); 800 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); 801 return (error); 802 } 803 804 /* 805 * Transform the running time and tick information for children of proc p 806 * into user and system time usage. 807 */ 808 void 809 calccru(p, up, sp) 810 struct proc *p; 811 struct timeval *up; 812 struct timeval *sp; 813 { 814 815 PROC_LOCK_ASSERT(p, MA_OWNED); 816 calcru1(p, &p->p_crux, up, sp); 817 } 818 819 /* 820 * Transform the running time and tick information in proc p into user 821 * and system time usage. If appropriate, include the current time slice 822 * on this CPU. 823 */ 824 void 825 calcru(struct proc *p, struct timeval *up, struct timeval *sp) 826 { 827 struct thread *td; 828 uint64_t runtime, u; 829 830 PROC_LOCK_ASSERT(p, MA_OWNED); 831 PROC_SLOCK_ASSERT(p, MA_OWNED); 832 /* 833 * If we are getting stats for the current process, then add in the 834 * stats that this thread has accumulated in its current time slice. 835 * We reset the thread and CPU state as if we had performed a context 836 * switch right here. 837 */ 838 td = curthread; 839 if (td->td_proc == p) { 840 u = cpu_ticks(); 841 runtime = u - PCPU_GET(switchtime); 842 td->td_runtime += runtime; 843 td->td_incruntime += runtime; 844 PCPU_SET(switchtime, u); 845 } 846 /* Make sure the per-thread stats are current. */ 847 FOREACH_THREAD_IN_PROC(p, td) { 848 if (td->td_incruntime == 0) 849 continue; 850 ruxagg(p, td); 851 } 852 calcru1(p, &p->p_rux, up, sp); 853 } 854 855 /* Collect resource usage for a single thread. */ 856 void 857 rufetchtd(struct thread *td, struct rusage *ru) 858 { 859 struct proc *p; 860 uint64_t runtime, u; 861 862 p = td->td_proc; 863 PROC_SLOCK_ASSERT(p, MA_OWNED); 864 THREAD_LOCK_ASSERT(td, MA_OWNED); 865 /* 866 * If we are getting stats for the current thread, then add in the 867 * stats that this thread has accumulated in its current time slice. 868 * We reset the thread and CPU state as if we had performed a context 869 * switch right here. 870 */ 871 if (td == curthread) { 872 u = cpu_ticks(); 873 runtime = u - PCPU_GET(switchtime); 874 td->td_runtime += runtime; 875 td->td_incruntime += runtime; 876 PCPU_SET(switchtime, u); 877 } 878 ruxagg(p, td); 879 *ru = td->td_ru; 880 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); 881 } 882 883 static void 884 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, 885 struct timeval *sp) 886 { 887 /* {user, system, interrupt, total} {ticks, usec}: */ 888 uint64_t ut, uu, st, su, it, tt, tu; 889 890 ut = ruxp->rux_uticks; 891 st = ruxp->rux_sticks; 892 it = ruxp->rux_iticks; 893 tt = ut + st + it; 894 if (tt == 0) { 895 /* Avoid divide by zero */ 896 st = 1; 897 tt = 1; 898 } 899 tu = cputick2usec(ruxp->rux_runtime); 900 if ((int64_t)tu < 0) { 901 /* XXX: this should be an assert /phk */ 902 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", 903 (intmax_t)tu, p->p_pid, p->p_comm); 904 tu = ruxp->rux_tu; 905 } 906 907 if (tu >= ruxp->rux_tu) { 908 /* 909 * The normal case, time increased. 910 * Enforce monotonicity of bucketed numbers. 911 */ 912 uu = (tu * ut) / tt; 913 if (uu < ruxp->rux_uu) 914 uu = ruxp->rux_uu; 915 su = (tu * st) / tt; 916 if (su < ruxp->rux_su) 917 su = ruxp->rux_su; 918 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { 919 /* 920 * When we calibrate the cputicker, it is not uncommon to 921 * see the presumably fixed frequency increase slightly over 922 * time as a result of thermal stabilization and NTP 923 * discipline (of the reference clock). We therefore ignore 924 * a bit of backwards slop because we expect to catch up 925 * shortly. We use a 3 microsecond limit to catch low 926 * counts and a 1% limit for high counts. 927 */ 928 uu = ruxp->rux_uu; 929 su = ruxp->rux_su; 930 tu = ruxp->rux_tu; 931 } else { /* tu < ruxp->rux_tu */ 932 /* 933 * What happened here was likely that a laptop, which ran at 934 * a reduced clock frequency at boot, kicked into high gear. 935 * The wisdom of spamming this message in that case is 936 * dubious, but it might also be indicative of something 937 * serious, so lets keep it and hope laptops can be made 938 * more truthful about their CPU speed via ACPI. 939 */ 940 printf("calcru: runtime went backwards from %ju usec " 941 "to %ju usec for pid %d (%s)\n", 942 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, 943 p->p_pid, p->p_comm); 944 uu = (tu * ut) / tt; 945 su = (tu * st) / tt; 946 } 947 948 ruxp->rux_uu = uu; 949 ruxp->rux_su = su; 950 ruxp->rux_tu = tu; 951 952 up->tv_sec = uu / 1000000; 953 up->tv_usec = uu % 1000000; 954 sp->tv_sec = su / 1000000; 955 sp->tv_usec = su % 1000000; 956 } 957 958 #ifndef _SYS_SYSPROTO_H_ 959 struct getrusage_args { 960 int who; 961 struct rusage *rusage; 962 }; 963 #endif 964 int 965 sys_getrusage(td, uap) 966 register struct thread *td; 967 register struct getrusage_args *uap; 968 { 969 struct rusage ru; 970 int error; 971 972 error = kern_getrusage(td, uap->who, &ru); 973 if (error == 0) 974 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 975 return (error); 976 } 977 978 int 979 kern_getrusage(struct thread *td, int who, struct rusage *rup) 980 { 981 struct proc *p; 982 int error; 983 984 error = 0; 985 p = td->td_proc; 986 PROC_LOCK(p); 987 switch (who) { 988 case RUSAGE_SELF: 989 rufetchcalc(p, rup, &rup->ru_utime, 990 &rup->ru_stime); 991 break; 992 993 case RUSAGE_CHILDREN: 994 *rup = p->p_stats->p_cru; 995 calccru(p, &rup->ru_utime, &rup->ru_stime); 996 break; 997 998 case RUSAGE_THREAD: 999 PROC_SLOCK(p); 1000 thread_lock(td); 1001 rufetchtd(td, rup); 1002 thread_unlock(td); 1003 PROC_SUNLOCK(p); 1004 break; 1005 1006 default: 1007 error = EINVAL; 1008 } 1009 PROC_UNLOCK(p); 1010 return (error); 1011 } 1012 1013 void 1014 rucollect(struct rusage *ru, struct rusage *ru2) 1015 { 1016 long *ip, *ip2; 1017 int i; 1018 1019 if (ru->ru_maxrss < ru2->ru_maxrss) 1020 ru->ru_maxrss = ru2->ru_maxrss; 1021 ip = &ru->ru_first; 1022 ip2 = &ru2->ru_first; 1023 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) 1024 *ip++ += *ip2++; 1025 } 1026 1027 void 1028 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, 1029 struct rusage_ext *rux2) 1030 { 1031 1032 rux->rux_runtime += rux2->rux_runtime; 1033 rux->rux_uticks += rux2->rux_uticks; 1034 rux->rux_sticks += rux2->rux_sticks; 1035 rux->rux_iticks += rux2->rux_iticks; 1036 rux->rux_uu += rux2->rux_uu; 1037 rux->rux_su += rux2->rux_su; 1038 rux->rux_tu += rux2->rux_tu; 1039 rucollect(ru, ru2); 1040 } 1041 1042 /* 1043 * Aggregate tick counts into the proc's rusage_ext. 1044 */ 1045 static void 1046 ruxagg_locked(struct rusage_ext *rux, struct thread *td) 1047 { 1048 1049 THREAD_LOCK_ASSERT(td, MA_OWNED); 1050 PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED); 1051 rux->rux_runtime += td->td_incruntime; 1052 rux->rux_uticks += td->td_uticks; 1053 rux->rux_sticks += td->td_sticks; 1054 rux->rux_iticks += td->td_iticks; 1055 } 1056 1057 void 1058 ruxagg(struct proc *p, struct thread *td) 1059 { 1060 1061 thread_lock(td); 1062 ruxagg_locked(&p->p_rux, td); 1063 ruxagg_locked(&td->td_rux, td); 1064 td->td_incruntime = 0; 1065 td->td_uticks = 0; 1066 td->td_iticks = 0; 1067 td->td_sticks = 0; 1068 thread_unlock(td); 1069 } 1070 1071 /* 1072 * Update the rusage_ext structure and fetch a valid aggregate rusage 1073 * for proc p if storage for one is supplied. 1074 */ 1075 void 1076 rufetch(struct proc *p, struct rusage *ru) 1077 { 1078 struct thread *td; 1079 1080 PROC_SLOCK_ASSERT(p, MA_OWNED); 1081 1082 *ru = p->p_ru; 1083 if (p->p_numthreads > 0) { 1084 FOREACH_THREAD_IN_PROC(p, td) { 1085 ruxagg(p, td); 1086 rucollect(ru, &td->td_ru); 1087 } 1088 } 1089 } 1090 1091 /* 1092 * Atomically perform a rufetch and a calcru together. 1093 * Consumers, can safely assume the calcru is executed only once 1094 * rufetch is completed. 1095 */ 1096 void 1097 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, 1098 struct timeval *sp) 1099 { 1100 1101 PROC_SLOCK(p); 1102 rufetch(p, ru); 1103 calcru(p, up, sp); 1104 PROC_SUNLOCK(p); 1105 } 1106 1107 /* 1108 * Allocate a new resource limits structure and initialize its 1109 * reference count and mutex pointer. 1110 */ 1111 struct plimit * 1112 lim_alloc() 1113 { 1114 struct plimit *limp; 1115 1116 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); 1117 refcount_init(&limp->pl_refcnt, 1); 1118 return (limp); 1119 } 1120 1121 struct plimit * 1122 lim_hold(limp) 1123 struct plimit *limp; 1124 { 1125 1126 refcount_acquire(&limp->pl_refcnt); 1127 return (limp); 1128 } 1129 1130 void 1131 lim_fork(struct proc *p1, struct proc *p2) 1132 { 1133 1134 PROC_LOCK_ASSERT(p1, MA_OWNED); 1135 PROC_LOCK_ASSERT(p2, MA_OWNED); 1136 1137 p2->p_limit = lim_hold(p1->p_limit); 1138 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); 1139 if (p1->p_cpulimit != RLIM_INFINITY) 1140 callout_reset(&p2->p_limco, hz, lim_cb, p2); 1141 } 1142 1143 void 1144 lim_free(limp) 1145 struct plimit *limp; 1146 { 1147 1148 KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow")); 1149 if (refcount_release(&limp->pl_refcnt)) 1150 free((void *)limp, M_PLIMIT); 1151 } 1152 1153 /* 1154 * Make a copy of the plimit structure. 1155 * We share these structures copy-on-write after fork. 1156 */ 1157 void 1158 lim_copy(dst, src) 1159 struct plimit *dst, *src; 1160 { 1161 1162 KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit")); 1163 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); 1164 } 1165 1166 /* 1167 * Return the hard limit for a particular system resource. The 1168 * which parameter specifies the index into the rlimit array. 1169 */ 1170 rlim_t 1171 lim_max(struct proc *p, int which) 1172 { 1173 struct rlimit rl; 1174 1175 lim_rlimit(p, which, &rl); 1176 return (rl.rlim_max); 1177 } 1178 1179 /* 1180 * Return the current (soft) limit for a particular system resource. 1181 * The which parameter which specifies the index into the rlimit array 1182 */ 1183 rlim_t 1184 lim_cur(struct proc *p, int which) 1185 { 1186 struct rlimit rl; 1187 1188 lim_rlimit(p, which, &rl); 1189 return (rl.rlim_cur); 1190 } 1191 1192 /* 1193 * Return a copy of the entire rlimit structure for the system limit 1194 * specified by 'which' in the rlimit structure pointed to by 'rlp'. 1195 */ 1196 void 1197 lim_rlimit(struct proc *p, int which, struct rlimit *rlp) 1198 { 1199 1200 PROC_LOCK_ASSERT(p, MA_OWNED); 1201 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1202 ("request for invalid resource limit")); 1203 *rlp = p->p_limit->pl_rlimit[which]; 1204 if (p->p_sysent->sv_fixlimit != NULL) 1205 p->p_sysent->sv_fixlimit(rlp, which); 1206 } 1207 1208 void 1209 uihashinit() 1210 { 1211 1212 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); 1213 rw_init(&uihashtbl_lock, "uidinfo hash"); 1214 } 1215 1216 /* 1217 * Look up a uidinfo struct for the parameter uid. 1218 * uihashtbl_lock must be locked. 1219 */ 1220 static struct uidinfo * 1221 uilookup(uid) 1222 uid_t uid; 1223 { 1224 struct uihashhead *uipp; 1225 struct uidinfo *uip; 1226 1227 rw_assert(&uihashtbl_lock, RA_LOCKED); 1228 uipp = UIHASH(uid); 1229 LIST_FOREACH(uip, uipp, ui_hash) 1230 if (uip->ui_uid == uid) 1231 break; 1232 1233 return (uip); 1234 } 1235 1236 /* 1237 * Find or allocate a struct uidinfo for a particular uid. 1238 * Increase refcount on uidinfo struct returned. 1239 * uifree() should be called on a struct uidinfo when released. 1240 */ 1241 struct uidinfo * 1242 uifind(uid) 1243 uid_t uid; 1244 { 1245 struct uidinfo *old_uip, *uip; 1246 1247 rw_rlock(&uihashtbl_lock); 1248 uip = uilookup(uid); 1249 if (uip == NULL) { 1250 rw_runlock(&uihashtbl_lock); 1251 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO); 1252 racct_create(&uip->ui_racct); 1253 rw_wlock(&uihashtbl_lock); 1254 /* 1255 * There's a chance someone created our uidinfo while we 1256 * were in malloc and not holding the lock, so we have to 1257 * make sure we don't insert a duplicate uidinfo. 1258 */ 1259 if ((old_uip = uilookup(uid)) != NULL) { 1260 /* Someone else beat us to it. */ 1261 racct_destroy(&uip->ui_racct); 1262 free(uip, M_UIDINFO); 1263 uip = old_uip; 1264 } else { 1265 refcount_init(&uip->ui_ref, 0); 1266 uip->ui_uid = uid; 1267 mtx_init(&uip->ui_vmsize_mtx, "ui_vmsize", NULL, 1268 MTX_DEF); 1269 LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash); 1270 } 1271 } 1272 uihold(uip); 1273 rw_unlock(&uihashtbl_lock); 1274 return (uip); 1275 } 1276 1277 /* 1278 * Place another refcount on a uidinfo struct. 1279 */ 1280 void 1281 uihold(uip) 1282 struct uidinfo *uip; 1283 { 1284 1285 refcount_acquire(&uip->ui_ref); 1286 } 1287 1288 /*- 1289 * Since uidinfo structs have a long lifetime, we use an 1290 * opportunistic refcounting scheme to avoid locking the lookup hash 1291 * for each release. 1292 * 1293 * If the refcount hits 0, we need to free the structure, 1294 * which means we need to lock the hash. 1295 * Optimal case: 1296 * After locking the struct and lowering the refcount, if we find 1297 * that we don't need to free, simply unlock and return. 1298 * Suboptimal case: 1299 * If refcount lowering results in need to free, bump the count 1300 * back up, lose the lock and acquire the locks in the proper 1301 * order to try again. 1302 */ 1303 void 1304 uifree(uip) 1305 struct uidinfo *uip; 1306 { 1307 int old; 1308 1309 /* Prepare for optimal case. */ 1310 old = uip->ui_ref; 1311 if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1)) 1312 return; 1313 1314 /* Prepare for suboptimal case. */ 1315 rw_wlock(&uihashtbl_lock); 1316 if (refcount_release(&uip->ui_ref)) { 1317 racct_destroy(&uip->ui_racct); 1318 LIST_REMOVE(uip, ui_hash); 1319 rw_wunlock(&uihashtbl_lock); 1320 if (uip->ui_sbsize != 0) 1321 printf("freeing uidinfo: uid = %d, sbsize = %ld\n", 1322 uip->ui_uid, uip->ui_sbsize); 1323 if (uip->ui_proccnt != 0) 1324 printf("freeing uidinfo: uid = %d, proccnt = %ld\n", 1325 uip->ui_uid, uip->ui_proccnt); 1326 if (uip->ui_vmsize != 0) 1327 printf("freeing uidinfo: uid = %d, swapuse = %lld\n", 1328 uip->ui_uid, (unsigned long long)uip->ui_vmsize); 1329 mtx_destroy(&uip->ui_vmsize_mtx); 1330 free(uip, M_UIDINFO); 1331 return; 1332 } 1333 /* 1334 * Someone added a reference between atomic_cmpset_int() and 1335 * rw_wlock(&uihashtbl_lock). 1336 */ 1337 rw_wunlock(&uihashtbl_lock); 1338 } 1339 1340 void 1341 ui_racct_foreach(void (*callback)(struct racct *racct, 1342 void *arg2, void *arg3), void *arg2, void *arg3) 1343 { 1344 struct uidinfo *uip; 1345 struct uihashhead *uih; 1346 1347 rw_rlock(&uihashtbl_lock); 1348 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { 1349 LIST_FOREACH(uip, uih, ui_hash) { 1350 (callback)(uip->ui_racct, arg2, arg3); 1351 } 1352 } 1353 rw_runlock(&uihashtbl_lock); 1354 } 1355 1356 /* 1357 * Change the count associated with number of processes 1358 * a given user is using. When 'max' is 0, don't enforce a limit 1359 */ 1360 int 1361 chgproccnt(uip, diff, max) 1362 struct uidinfo *uip; 1363 int diff; 1364 rlim_t max; 1365 { 1366 1367 /* Don't allow them to exceed max, but allow subtraction. */ 1368 if (diff > 0 && max != 0) { 1369 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) { 1370 atomic_subtract_long(&uip->ui_proccnt, (long)diff); 1371 return (0); 1372 } 1373 } else { 1374 atomic_add_long(&uip->ui_proccnt, (long)diff); 1375 if (uip->ui_proccnt < 0) 1376 printf("negative proccnt for uid = %d\n", uip->ui_uid); 1377 } 1378 return (1); 1379 } 1380 1381 /* 1382 * Change the total socket buffer size a user has used. 1383 */ 1384 int 1385 chgsbsize(uip, hiwat, to, max) 1386 struct uidinfo *uip; 1387 u_int *hiwat; 1388 u_int to; 1389 rlim_t max; 1390 { 1391 int diff; 1392 1393 diff = to - *hiwat; 1394 if (diff > 0) { 1395 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) { 1396 atomic_subtract_long(&uip->ui_sbsize, (long)diff); 1397 return (0); 1398 } 1399 } else { 1400 atomic_add_long(&uip->ui_sbsize, (long)diff); 1401 if (uip->ui_sbsize < 0) 1402 printf("negative sbsize for uid = %d\n", uip->ui_uid); 1403 } 1404 *hiwat = to; 1405 return (1); 1406 } 1407 1408 /* 1409 * Change the count associated with number of pseudo-terminals 1410 * a given user is using. When 'max' is 0, don't enforce a limit 1411 */ 1412 int 1413 chgptscnt(uip, diff, max) 1414 struct uidinfo *uip; 1415 int diff; 1416 rlim_t max; 1417 { 1418 1419 /* Don't allow them to exceed max, but allow subtraction. */ 1420 if (diff > 0 && max != 0) { 1421 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) { 1422 atomic_subtract_long(&uip->ui_ptscnt, (long)diff); 1423 return (0); 1424 } 1425 } else { 1426 atomic_add_long(&uip->ui_ptscnt, (long)diff); 1427 if (uip->ui_ptscnt < 0) 1428 printf("negative ptscnt for uid = %d\n", uip->ui_uid); 1429 } 1430 return (1); 1431 } 1432