1 /*- 2 * SPDX-License-Identifier: BSD-3-Clause 3 * 4 * Copyright (c) 1982, 1986, 1991, 1993 5 * The Regents of the University of California. All rights reserved. 6 * (c) UNIX System Laboratories, Inc. 7 * All or some portions of this file are derived from material licensed 8 * to the University of California by American Telephone and Telegraph 9 * Co. or Unix System Laboratories, Inc. and are reproduced herein with 10 * the permission of UNIX System Laboratories, Inc. 11 * 12 * Redistribution and use in source and binary forms, with or without 13 * modification, are permitted provided that the following conditions 14 * are met: 15 * 1. Redistributions of source code must retain the above copyright 16 * notice, this list of conditions and the following disclaimer. 17 * 2. Redistributions in binary form must reproduce the above copyright 18 * notice, this list of conditions and the following disclaimer in the 19 * documentation and/or other materials provided with the distribution. 20 * 3. Neither the name of the University nor the names of its contributors 21 * may be used to endorse or promote products derived from this software 22 * without specific prior written permission. 23 * 24 * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND 25 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 26 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 27 * ARE DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE 28 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 29 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 30 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 31 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 32 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 33 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 34 * SUCH DAMAGE. 35 * 36 * @(#)kern_resource.c 8.5 (Berkeley) 1/21/94 37 */ 38 39 #include <sys/cdefs.h> 40 __FBSDID("$FreeBSD$"); 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 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures"); 70 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures"); 71 #define UIHASH(uid) (&uihashtbl[(uid) & uihash]) 72 static struct rwlock uihashtbl_lock; 73 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl; 74 static u_long uihash; /* size of hash table - 1 */ 75 76 static void calcru1(struct proc *p, struct rusage_ext *ruxp, 77 struct timeval *up, struct timeval *sp); 78 static int donice(struct thread *td, struct proc *chgp, int n); 79 static struct uidinfo *uilookup(uid_t uid); 80 static void ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td); 81 82 /* 83 * Resource controls and accounting. 84 */ 85 #ifndef _SYS_SYSPROTO_H_ 86 struct getpriority_args { 87 int which; 88 int who; 89 }; 90 #endif 91 int 92 sys_getpriority(struct thread *td, struct getpriority_args *uap) 93 { 94 95 return (kern_getpriority(td, uap->which, uap->who)); 96 } 97 98 int 99 kern_getpriority(struct thread *td, int which, int who) 100 { 101 struct proc *p; 102 struct pgrp *pg; 103 int error, low; 104 105 error = 0; 106 low = PRIO_MAX + 1; 107 switch (which) { 108 case PRIO_PROCESS: 109 if (who == 0) 110 low = td->td_proc->p_nice; 111 else { 112 p = pfind(who); 113 if (p == NULL) 114 break; 115 if (p_cansee(td, p) == 0) 116 low = p->p_nice; 117 PROC_UNLOCK(p); 118 } 119 break; 120 121 case PRIO_PGRP: 122 sx_slock(&proctree_lock); 123 if (who == 0) { 124 pg = td->td_proc->p_pgrp; 125 PGRP_LOCK(pg); 126 } else { 127 pg = pgfind(who); 128 if (pg == NULL) { 129 sx_sunlock(&proctree_lock); 130 break; 131 } 132 } 133 sx_sunlock(&proctree_lock); 134 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 135 PROC_LOCK(p); 136 if (p->p_state == PRS_NORMAL && 137 p_cansee(td, p) == 0) { 138 if (p->p_nice < low) 139 low = p->p_nice; 140 } 141 PROC_UNLOCK(p); 142 } 143 PGRP_UNLOCK(pg); 144 break; 145 146 case PRIO_USER: 147 if (who == 0) 148 who = td->td_ucred->cr_uid; 149 sx_slock(&allproc_lock); 150 FOREACH_PROC_IN_SYSTEM(p) { 151 PROC_LOCK(p); 152 if (p->p_state == PRS_NORMAL && 153 p_cansee(td, p) == 0 && 154 p->p_ucred->cr_uid == who) { 155 if (p->p_nice < low) 156 low = p->p_nice; 157 } 158 PROC_UNLOCK(p); 159 } 160 sx_sunlock(&allproc_lock); 161 break; 162 163 default: 164 error = EINVAL; 165 break; 166 } 167 if (low == PRIO_MAX + 1 && error == 0) 168 error = ESRCH; 169 td->td_retval[0] = low; 170 return (error); 171 } 172 173 #ifndef _SYS_SYSPROTO_H_ 174 struct setpriority_args { 175 int which; 176 int who; 177 int prio; 178 }; 179 #endif 180 int 181 sys_setpriority(struct thread *td, struct setpriority_args *uap) 182 { 183 184 return (kern_setpriority(td, uap->which, uap->who, uap->prio)); 185 } 186 187 int 188 kern_setpriority(struct thread *td, int which, int who, int prio) 189 { 190 struct proc *curp, *p; 191 struct pgrp *pg; 192 int found = 0, error = 0; 193 194 curp = td->td_proc; 195 switch (which) { 196 case PRIO_PROCESS: 197 if (who == 0) { 198 PROC_LOCK(curp); 199 error = donice(td, curp, prio); 200 PROC_UNLOCK(curp); 201 } else { 202 p = pfind(who); 203 if (p == NULL) 204 break; 205 error = p_cansee(td, p); 206 if (error == 0) 207 error = donice(td, p, prio); 208 PROC_UNLOCK(p); 209 } 210 found++; 211 break; 212 213 case PRIO_PGRP: 214 sx_slock(&proctree_lock); 215 if (who == 0) { 216 pg = curp->p_pgrp; 217 PGRP_LOCK(pg); 218 } else { 219 pg = pgfind(who); 220 if (pg == NULL) { 221 sx_sunlock(&proctree_lock); 222 break; 223 } 224 } 225 sx_sunlock(&proctree_lock); 226 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 227 PROC_LOCK(p); 228 if (p->p_state == PRS_NORMAL && 229 p_cansee(td, p) == 0) { 230 error = donice(td, p, prio); 231 found++; 232 } 233 PROC_UNLOCK(p); 234 } 235 PGRP_UNLOCK(pg); 236 break; 237 238 case PRIO_USER: 239 if (who == 0) 240 who = td->td_ucred->cr_uid; 241 sx_slock(&allproc_lock); 242 FOREACH_PROC_IN_SYSTEM(p) { 243 PROC_LOCK(p); 244 if (p->p_state == PRS_NORMAL && 245 p->p_ucred->cr_uid == who && 246 p_cansee(td, p) == 0) { 247 error = donice(td, p, prio); 248 found++; 249 } 250 PROC_UNLOCK(p); 251 } 252 sx_sunlock(&allproc_lock); 253 break; 254 255 default: 256 error = EINVAL; 257 break; 258 } 259 if (found == 0 && error == 0) 260 error = ESRCH; 261 return (error); 262 } 263 264 /* 265 * Set "nice" for a (whole) process. 266 */ 267 static int 268 donice(struct thread *td, struct proc *p, int n) 269 { 270 int error; 271 272 PROC_LOCK_ASSERT(p, MA_OWNED); 273 if ((error = p_cansched(td, p))) 274 return (error); 275 if (n > PRIO_MAX) 276 n = PRIO_MAX; 277 if (n < PRIO_MIN) 278 n = PRIO_MIN; 279 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0) 280 return (EACCES); 281 sched_nice(p, n); 282 return (0); 283 } 284 285 static int unprivileged_idprio; 286 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW, 287 &unprivileged_idprio, 0, "Allow non-root users to set an idle priority"); 288 289 /* 290 * Set realtime priority for LWP. 291 */ 292 #ifndef _SYS_SYSPROTO_H_ 293 struct rtprio_thread_args { 294 int function; 295 lwpid_t lwpid; 296 struct rtprio *rtp; 297 }; 298 #endif 299 int 300 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap) 301 { 302 struct proc *p; 303 struct rtprio rtp; 304 struct thread *td1; 305 int cierror, error; 306 307 /* Perform copyin before acquiring locks if needed. */ 308 if (uap->function == RTP_SET) 309 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 310 else 311 cierror = 0; 312 313 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) { 314 p = td->td_proc; 315 td1 = td; 316 PROC_LOCK(p); 317 } else { 318 td1 = tdfind(uap->lwpid, -1); 319 if (td1 == NULL) 320 return (ESRCH); 321 p = td1->td_proc; 322 } 323 324 switch (uap->function) { 325 case RTP_LOOKUP: 326 if ((error = p_cansee(td, p))) 327 break; 328 pri_to_rtp(td1, &rtp); 329 PROC_UNLOCK(p); 330 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 331 case RTP_SET: 332 if ((error = p_cansched(td, p)) || (error = cierror)) 333 break; 334 335 /* Disallow setting rtprio in most cases if not superuser. */ 336 337 /* 338 * Realtime priority has to be restricted for reasons which 339 * should be obvious. However, for idleprio processes, there is 340 * a potential for system deadlock if an idleprio process gains 341 * a lock on a resource that other processes need (and the 342 * idleprio process can't run due to a CPU-bound normal 343 * process). Fix me! XXX 344 * 345 * This problem is not only related to idleprio process. 346 * A user level program can obtain a file lock and hold it 347 * indefinitely. Additionally, without idleprio processes it is 348 * still conceivable that a program with low priority will never 349 * get to run. In short, allowing this feature might make it 350 * easier to lock a resource indefinitely, but it is not the 351 * only thing that makes it possible. 352 */ 353 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || 354 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && 355 unprivileged_idprio == 0)) { 356 error = priv_check(td, PRIV_SCHED_RTPRIO); 357 if (error) 358 break; 359 } 360 error = rtp_to_pri(&rtp, td1); 361 break; 362 default: 363 error = EINVAL; 364 break; 365 } 366 PROC_UNLOCK(p); 367 return (error); 368 } 369 370 /* 371 * Set realtime priority. 372 */ 373 #ifndef _SYS_SYSPROTO_H_ 374 struct rtprio_args { 375 int function; 376 pid_t pid; 377 struct rtprio *rtp; 378 }; 379 #endif 380 int 381 sys_rtprio(struct thread *td, struct rtprio_args *uap) 382 { 383 struct proc *p; 384 struct thread *tdp; 385 struct rtprio rtp; 386 int cierror, error; 387 388 /* Perform copyin before acquiring locks if needed. */ 389 if (uap->function == RTP_SET) 390 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 391 else 392 cierror = 0; 393 394 if (uap->pid == 0) { 395 p = td->td_proc; 396 PROC_LOCK(p); 397 } else { 398 p = pfind(uap->pid); 399 if (p == NULL) 400 return (ESRCH); 401 } 402 403 switch (uap->function) { 404 case RTP_LOOKUP: 405 if ((error = p_cansee(td, p))) 406 break; 407 /* 408 * Return OUR priority if no pid specified, 409 * or if one is, report the highest priority 410 * in the process. There isn't much more you can do as 411 * there is only room to return a single priority. 412 * Note: specifying our own pid is not the same 413 * as leaving it zero. 414 */ 415 if (uap->pid == 0) { 416 pri_to_rtp(td, &rtp); 417 } else { 418 struct rtprio rtp2; 419 420 rtp.type = RTP_PRIO_IDLE; 421 rtp.prio = RTP_PRIO_MAX; 422 FOREACH_THREAD_IN_PROC(p, tdp) { 423 pri_to_rtp(tdp, &rtp2); 424 if (rtp2.type < rtp.type || 425 (rtp2.type == rtp.type && 426 rtp2.prio < rtp.prio)) { 427 rtp.type = rtp2.type; 428 rtp.prio = rtp2.prio; 429 } 430 } 431 } 432 PROC_UNLOCK(p); 433 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 434 case RTP_SET: 435 if ((error = p_cansched(td, p)) || (error = cierror)) 436 break; 437 438 /* 439 * Disallow setting rtprio in most cases if not superuser. 440 * See the comment in sys_rtprio_thread about idprio 441 * threads holding a lock. 442 */ 443 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || 444 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && 445 !unprivileged_idprio)) { 446 error = priv_check(td, PRIV_SCHED_RTPRIO); 447 if (error) 448 break; 449 } 450 451 /* 452 * If we are setting our own priority, set just our 453 * thread but if we are doing another process, 454 * do all the threads on that process. If we 455 * specify our own pid we do the latter. 456 */ 457 if (uap->pid == 0) { 458 error = rtp_to_pri(&rtp, td); 459 } else { 460 FOREACH_THREAD_IN_PROC(p, td) { 461 if ((error = rtp_to_pri(&rtp, td)) != 0) 462 break; 463 } 464 } 465 break; 466 default: 467 error = EINVAL; 468 break; 469 } 470 PROC_UNLOCK(p); 471 return (error); 472 } 473 474 int 475 rtp_to_pri(struct rtprio *rtp, struct thread *td) 476 { 477 u_char newpri, oldclass, oldpri; 478 479 switch (RTP_PRIO_BASE(rtp->type)) { 480 case RTP_PRIO_REALTIME: 481 if (rtp->prio > RTP_PRIO_MAX) 482 return (EINVAL); 483 newpri = PRI_MIN_REALTIME + rtp->prio; 484 break; 485 case RTP_PRIO_NORMAL: 486 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) 487 return (EINVAL); 488 newpri = PRI_MIN_TIMESHARE + rtp->prio; 489 break; 490 case RTP_PRIO_IDLE: 491 if (rtp->prio > RTP_PRIO_MAX) 492 return (EINVAL); 493 newpri = PRI_MIN_IDLE + rtp->prio; 494 break; 495 default: 496 return (EINVAL); 497 } 498 499 thread_lock(td); 500 oldclass = td->td_pri_class; 501 sched_class(td, rtp->type); /* XXX fix */ 502 oldpri = td->td_user_pri; 503 sched_user_prio(td, newpri); 504 if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL || 505 td->td_pri_class != RTP_PRIO_NORMAL)) 506 sched_prio(td, td->td_user_pri); 507 if (TD_ON_UPILOCK(td) && oldpri != newpri) { 508 critical_enter(); 509 thread_unlock(td); 510 umtx_pi_adjust(td, oldpri); 511 critical_exit(); 512 } else 513 thread_unlock(td); 514 return (0); 515 } 516 517 void 518 pri_to_rtp(struct thread *td, struct rtprio *rtp) 519 { 520 521 thread_lock(td); 522 switch (PRI_BASE(td->td_pri_class)) { 523 case PRI_REALTIME: 524 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME; 525 break; 526 case PRI_TIMESHARE: 527 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE; 528 break; 529 case PRI_IDLE: 530 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE; 531 break; 532 default: 533 break; 534 } 535 rtp->type = td->td_pri_class; 536 thread_unlock(td); 537 } 538 539 #if defined(COMPAT_43) 540 #ifndef _SYS_SYSPROTO_H_ 541 struct osetrlimit_args { 542 u_int which; 543 struct orlimit *rlp; 544 }; 545 #endif 546 int 547 osetrlimit(struct thread *td, struct osetrlimit_args *uap) 548 { 549 struct orlimit olim; 550 struct rlimit lim; 551 int error; 552 553 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit)))) 554 return (error); 555 lim.rlim_cur = olim.rlim_cur; 556 lim.rlim_max = olim.rlim_max; 557 error = kern_setrlimit(td, uap->which, &lim); 558 return (error); 559 } 560 561 #ifndef _SYS_SYSPROTO_H_ 562 struct ogetrlimit_args { 563 u_int which; 564 struct orlimit *rlp; 565 }; 566 #endif 567 int 568 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap) 569 { 570 struct orlimit olim; 571 struct rlimit rl; 572 int error; 573 574 if (uap->which >= RLIM_NLIMITS) 575 return (EINVAL); 576 lim_rlimit(td, uap->which, &rl); 577 578 /* 579 * XXX would be more correct to convert only RLIM_INFINITY to the 580 * old RLIM_INFINITY and fail with EOVERFLOW for other larger 581 * values. Most 64->32 and 32->16 conversions, including not 582 * unimportant ones of uids are even more broken than what we 583 * do here (they blindly truncate). We don't do this correctly 584 * here since we have little experience with EOVERFLOW yet. 585 * Elsewhere, getuid() can't fail... 586 */ 587 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur; 588 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max; 589 error = copyout(&olim, uap->rlp, sizeof(olim)); 590 return (error); 591 } 592 #endif /* COMPAT_43 */ 593 594 #ifndef _SYS_SYSPROTO_H_ 595 struct __setrlimit_args { 596 u_int which; 597 struct rlimit *rlp; 598 }; 599 #endif 600 int 601 sys_setrlimit(struct thread *td, struct __setrlimit_args *uap) 602 { 603 struct rlimit alim; 604 int error; 605 606 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit)))) 607 return (error); 608 error = kern_setrlimit(td, uap->which, &alim); 609 return (error); 610 } 611 612 static void 613 lim_cb(void *arg) 614 { 615 struct rlimit rlim; 616 struct thread *td; 617 struct proc *p; 618 619 p = arg; 620 PROC_LOCK_ASSERT(p, MA_OWNED); 621 /* 622 * Check if the process exceeds its cpu resource allocation. If 623 * it reaches the max, arrange to kill the process in ast(). 624 */ 625 if (p->p_cpulimit == RLIM_INFINITY) 626 return; 627 PROC_STATLOCK(p); 628 FOREACH_THREAD_IN_PROC(p, td) { 629 ruxagg(p, td); 630 } 631 PROC_STATUNLOCK(p); 632 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) { 633 lim_rlimit_proc(p, RLIMIT_CPU, &rlim); 634 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) { 635 killproc(p, "exceeded maximum CPU limit"); 636 } else { 637 if (p->p_cpulimit < rlim.rlim_max) 638 p->p_cpulimit += 5; 639 kern_psignal(p, SIGXCPU); 640 } 641 } 642 if ((p->p_flag & P_WEXIT) == 0) 643 callout_reset_sbt(&p->p_limco, SBT_1S, 0, 644 lim_cb, p, C_PREL(1)); 645 } 646 647 int 648 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp) 649 { 650 651 return (kern_proc_setrlimit(td, td->td_proc, which, limp)); 652 } 653 654 int 655 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which, 656 struct rlimit *limp) 657 { 658 struct plimit *newlim, *oldlim; 659 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 newlim = lim_alloc(); 676 PROC_LOCK(p); 677 oldlim = p->p_limit; 678 alimp = &oldlim->pl_rlimit[which]; 679 if (limp->rlim_cur > alimp->rlim_max || 680 limp->rlim_max > alimp->rlim_max) 681 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { 682 PROC_UNLOCK(p); 683 lim_free(newlim); 684 return (error); 685 } 686 if (limp->rlim_cur > limp->rlim_max) 687 limp->rlim_cur = limp->rlim_max; 688 lim_copy(newlim, oldlim); 689 alimp = &newlim->pl_rlimit[which]; 690 691 switch (which) { 692 case RLIMIT_CPU: 693 if (limp->rlim_cur != RLIM_INFINITY && 694 p->p_cpulimit == RLIM_INFINITY) 695 callout_reset_sbt(&p->p_limco, SBT_1S, 0, 696 lim_cb, p, C_PREL(1)); 697 p->p_cpulimit = limp->rlim_cur; 698 break; 699 case RLIMIT_DATA: 700 if (limp->rlim_cur > maxdsiz) 701 limp->rlim_cur = maxdsiz; 702 if (limp->rlim_max > maxdsiz) 703 limp->rlim_max = maxdsiz; 704 break; 705 706 case RLIMIT_STACK: 707 if (limp->rlim_cur > maxssiz) 708 limp->rlim_cur = maxssiz; 709 if (limp->rlim_max > maxssiz) 710 limp->rlim_max = maxssiz; 711 oldssiz = *alimp; 712 if (p->p_sysent->sv_fixlimit != NULL) 713 p->p_sysent->sv_fixlimit(&oldssiz, 714 RLIMIT_STACK); 715 break; 716 717 case RLIMIT_NOFILE: 718 if (limp->rlim_cur > maxfilesperproc) 719 limp->rlim_cur = maxfilesperproc; 720 if (limp->rlim_max > maxfilesperproc) 721 limp->rlim_max = maxfilesperproc; 722 break; 723 724 case RLIMIT_NPROC: 725 if (limp->rlim_cur > maxprocperuid) 726 limp->rlim_cur = maxprocperuid; 727 if (limp->rlim_max > maxprocperuid) 728 limp->rlim_max = maxprocperuid; 729 if (limp->rlim_cur < 1) 730 limp->rlim_cur = 1; 731 if (limp->rlim_max < 1) 732 limp->rlim_max = 1; 733 break; 734 } 735 if (p->p_sysent->sv_fixlimit != NULL) 736 p->p_sysent->sv_fixlimit(limp, which); 737 *alimp = *limp; 738 p->p_limit = newlim; 739 PROC_UPDATE_COW(p); 740 PROC_UNLOCK(p); 741 lim_free(oldlim); 742 743 if (which == RLIMIT_STACK && 744 /* 745 * Skip calls from exec_new_vmspace(), done when stack is 746 * not mapped yet. 747 */ 748 (td != curthread || (p->p_flag & P_INEXEC) == 0)) { 749 /* 750 * Stack is allocated to the max at exec time with only 751 * "rlim_cur" bytes accessible. If stack limit is going 752 * up make more accessible, if going down make inaccessible. 753 */ 754 if (limp->rlim_cur != oldssiz.rlim_cur) { 755 vm_offset_t addr; 756 vm_size_t size; 757 vm_prot_t prot; 758 759 if (limp->rlim_cur > oldssiz.rlim_cur) { 760 prot = p->p_sysent->sv_stackprot; 761 size = limp->rlim_cur - oldssiz.rlim_cur; 762 addr = p->p_sysent->sv_usrstack - 763 limp->rlim_cur; 764 } else { 765 prot = VM_PROT_NONE; 766 size = oldssiz.rlim_cur - limp->rlim_cur; 767 addr = p->p_sysent->sv_usrstack - 768 oldssiz.rlim_cur; 769 } 770 addr = trunc_page(addr); 771 size = round_page(size); 772 (void)vm_map_protect(&p->p_vmspace->vm_map, 773 addr, addr + size, prot, FALSE); 774 } 775 } 776 777 return (0); 778 } 779 780 #ifndef _SYS_SYSPROTO_H_ 781 struct __getrlimit_args { 782 u_int which; 783 struct rlimit *rlp; 784 }; 785 #endif 786 /* ARGSUSED */ 787 int 788 sys_getrlimit(struct thread *td, struct __getrlimit_args *uap) 789 { 790 struct rlimit rlim; 791 int error; 792 793 if (uap->which >= RLIM_NLIMITS) 794 return (EINVAL); 795 lim_rlimit(td, uap->which, &rlim); 796 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); 797 return (error); 798 } 799 800 /* 801 * Transform the running time and tick information for children of proc p 802 * into user and system time usage. 803 */ 804 void 805 calccru(struct proc *p, struct timeval *up, struct timeval *sp) 806 { 807 808 PROC_LOCK_ASSERT(p, MA_OWNED); 809 calcru1(p, &p->p_crux, up, sp); 810 } 811 812 /* 813 * Transform the running time and tick information in proc p into user 814 * and system time usage. If appropriate, include the current time slice 815 * on this CPU. 816 */ 817 void 818 calcru(struct proc *p, struct timeval *up, struct timeval *sp) 819 { 820 struct thread *td; 821 uint64_t runtime, u; 822 823 PROC_LOCK_ASSERT(p, MA_OWNED); 824 PROC_STATLOCK_ASSERT(p, MA_OWNED); 825 /* 826 * If we are getting stats for the current process, then add in the 827 * stats that this thread has accumulated in its current time slice. 828 * We reset the thread and CPU state as if we had performed a context 829 * switch right here. 830 */ 831 td = curthread; 832 if (td->td_proc == p) { 833 u = cpu_ticks(); 834 runtime = u - PCPU_GET(switchtime); 835 td->td_runtime += runtime; 836 td->td_incruntime += runtime; 837 PCPU_SET(switchtime, u); 838 } 839 /* Make sure the per-thread stats are current. */ 840 FOREACH_THREAD_IN_PROC(p, td) { 841 if (td->td_incruntime == 0) 842 continue; 843 ruxagg(p, td); 844 } 845 calcru1(p, &p->p_rux, up, sp); 846 } 847 848 /* Collect resource usage for a single thread. */ 849 void 850 rufetchtd(struct thread *td, struct rusage *ru) 851 { 852 struct proc *p; 853 uint64_t runtime, u; 854 855 p = td->td_proc; 856 PROC_STATLOCK_ASSERT(p, MA_OWNED); 857 THREAD_LOCK_ASSERT(td, MA_OWNED); 858 /* 859 * If we are getting stats for the current thread, then add in the 860 * stats that this thread has accumulated in its current time slice. 861 * We reset the thread and CPU state as if we had performed a context 862 * switch right here. 863 */ 864 if (td == curthread) { 865 u = cpu_ticks(); 866 runtime = u - PCPU_GET(switchtime); 867 td->td_runtime += runtime; 868 td->td_incruntime += runtime; 869 PCPU_SET(switchtime, u); 870 } 871 ruxagg_locked(p, td); 872 *ru = td->td_ru; 873 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); 874 } 875 876 /* XXX: the MI version is too slow to use: */ 877 #ifndef __HAVE_INLINE_FLSLL 878 #define flsll(x) (fls((x) >> 32) != 0 ? fls((x) >> 32) + 32 : fls(x)) 879 #endif 880 881 static uint64_t 882 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c) 883 { 884 uint64_t acc, bh, bl; 885 int i, s, sa, sb; 886 887 /* 888 * Calculate (a * b) / c accurately enough without overflowing. c 889 * must be nonzero, and its top bit must be 0. a or b must be 890 * <= c, and the implementation is tuned for b <= c. 891 * 892 * The comments about times are for use in calcru1() with units of 893 * microseconds for 'a' and stathz ticks at 128 Hz for b and c. 894 * 895 * Let n be the number of top zero bits in c. Each iteration 896 * either returns, or reduces b by right shifting it by at least n. 897 * The number of iterations is at most 1 + 64 / n, and the error is 898 * at most the number of iterations. 899 * 900 * It is very unusual to need even 2 iterations. Previous 901 * implementations overflowed essentially by returning early in the 902 * first iteration, with n = 38 giving overflow at 105+ hours and 903 * n = 32 giving overlow at at 388+ days despite a more careful 904 * calculation. 388 days is a reasonable uptime, and the calculation 905 * needs to work for the uptime times the number of CPUs since 'a' 906 * is per-process. 907 */ 908 if (a >= (uint64_t)1 << 63) 909 return (0); /* Unsupported arg -- can't happen. */ 910 acc = 0; 911 for (i = 0; i < 128; i++) { 912 sa = flsll(a); 913 sb = flsll(b); 914 if (sa + sb <= 64) 915 /* Up to 105 hours on first iteration. */ 916 return (acc + (a * b) / c); 917 if (a >= c) { 918 /* 919 * This reduction is based on a = q * c + r, with the 920 * remainder r < c. 'a' may be large to start, and 921 * moving bits from b into 'a' at the end of the loop 922 * sets the top bit of 'a', so the reduction makes 923 * significant progress. 924 */ 925 acc += (a / c) * b; 926 a %= c; 927 sa = flsll(a); 928 if (sa + sb <= 64) 929 /* Up to 388 days on first iteration. */ 930 return (acc + (a * b) / c); 931 } 932 933 /* 934 * This step writes a * b as a * ((bh << s) + bl) = 935 * a * (bh << s) + a * bl = (a << s) * bh + a * bl. The 2 936 * additive terms are handled separately. Splitting in 937 * this way is linear except for rounding errors. 938 * 939 * s = 64 - sa is the maximum such that a << s fits in 64 940 * bits. Since a < c and c has at least 1 zero top bit, 941 * sa < 64 and s > 0. Thus this step makes progress by 942 * reducing b (it increases 'a', but taking remainders on 943 * the next iteration completes the reduction). 944 * 945 * Finally, the choice for s is just what is needed to keep 946 * a * bl from overflowing, so we don't need complications 947 * like a recursive call mul64_by_fraction(a, bl, c) to 948 * handle the second additive term. 949 */ 950 s = 64 - sa; 951 bh = b >> s; 952 bl = b - (bh << s); 953 acc += (a * bl) / c; 954 a <<= s; 955 b = bh; 956 } 957 return (0); /* Algorithm failure -- can't happen. */ 958 } 959 960 static void 961 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, 962 struct timeval *sp) 963 { 964 /* {user, system, interrupt, total} {ticks, usec}: */ 965 uint64_t ut, uu, st, su, it, tt, tu; 966 967 ut = ruxp->rux_uticks; 968 st = ruxp->rux_sticks; 969 it = ruxp->rux_iticks; 970 tt = ut + st + it; 971 if (tt == 0) { 972 /* Avoid divide by zero */ 973 st = 1; 974 tt = 1; 975 } 976 tu = cputick2usec(ruxp->rux_runtime); 977 if ((int64_t)tu < 0) { 978 /* XXX: this should be an assert /phk */ 979 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", 980 (intmax_t)tu, p->p_pid, p->p_comm); 981 tu = ruxp->rux_tu; 982 } 983 984 /* Subdivide tu. Avoid overflow in the multiplications. */ 985 if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) { 986 /* Up to 76 hours when stathz is 128. */ 987 uu = (tu * ut) / tt; 988 su = (tu * st) / tt; 989 } else { 990 uu = mul64_by_fraction(tu, ut, tt); 991 su = mul64_by_fraction(tu, st, tt); 992 } 993 994 if (tu >= ruxp->rux_tu) { 995 /* 996 * The normal case, time increased. 997 * Enforce monotonicity of bucketed numbers. 998 */ 999 if (uu < ruxp->rux_uu) 1000 uu = ruxp->rux_uu; 1001 if (su < ruxp->rux_su) 1002 su = ruxp->rux_su; 1003 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { 1004 /* 1005 * When we calibrate the cputicker, it is not uncommon to 1006 * see the presumably fixed frequency increase slightly over 1007 * time as a result of thermal stabilization and NTP 1008 * discipline (of the reference clock). We therefore ignore 1009 * a bit of backwards slop because we expect to catch up 1010 * shortly. We use a 3 microsecond limit to catch low 1011 * counts and a 1% limit for high counts. 1012 */ 1013 uu = ruxp->rux_uu; 1014 su = ruxp->rux_su; 1015 tu = ruxp->rux_tu; 1016 } else { /* tu < ruxp->rux_tu */ 1017 /* 1018 * What happened here was likely that a laptop, which ran at 1019 * a reduced clock frequency at boot, kicked into high gear. 1020 * The wisdom of spamming this message in that case is 1021 * dubious, but it might also be indicative of something 1022 * serious, so lets keep it and hope laptops can be made 1023 * more truthful about their CPU speed via ACPI. 1024 */ 1025 printf("calcru: runtime went backwards from %ju usec " 1026 "to %ju usec for pid %d (%s)\n", 1027 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, 1028 p->p_pid, p->p_comm); 1029 } 1030 1031 ruxp->rux_uu = uu; 1032 ruxp->rux_su = su; 1033 ruxp->rux_tu = tu; 1034 1035 up->tv_sec = uu / 1000000; 1036 up->tv_usec = uu % 1000000; 1037 sp->tv_sec = su / 1000000; 1038 sp->tv_usec = su % 1000000; 1039 } 1040 1041 #ifndef _SYS_SYSPROTO_H_ 1042 struct getrusage_args { 1043 int who; 1044 struct rusage *rusage; 1045 }; 1046 #endif 1047 int 1048 sys_getrusage(struct thread *td, struct getrusage_args *uap) 1049 { 1050 struct rusage ru; 1051 int error; 1052 1053 error = kern_getrusage(td, uap->who, &ru); 1054 if (error == 0) 1055 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 1056 return (error); 1057 } 1058 1059 int 1060 kern_getrusage(struct thread *td, int who, struct rusage *rup) 1061 { 1062 struct proc *p; 1063 int error; 1064 1065 error = 0; 1066 p = td->td_proc; 1067 PROC_LOCK(p); 1068 switch (who) { 1069 case RUSAGE_SELF: 1070 rufetchcalc(p, rup, &rup->ru_utime, 1071 &rup->ru_stime); 1072 break; 1073 1074 case RUSAGE_CHILDREN: 1075 *rup = p->p_stats->p_cru; 1076 calccru(p, &rup->ru_utime, &rup->ru_stime); 1077 break; 1078 1079 case RUSAGE_THREAD: 1080 PROC_STATLOCK(p); 1081 thread_lock(td); 1082 rufetchtd(td, rup); 1083 thread_unlock(td); 1084 PROC_STATUNLOCK(p); 1085 break; 1086 1087 default: 1088 error = EINVAL; 1089 } 1090 PROC_UNLOCK(p); 1091 return (error); 1092 } 1093 1094 void 1095 rucollect(struct rusage *ru, struct rusage *ru2) 1096 { 1097 long *ip, *ip2; 1098 int i; 1099 1100 if (ru->ru_maxrss < ru2->ru_maxrss) 1101 ru->ru_maxrss = ru2->ru_maxrss; 1102 ip = &ru->ru_first; 1103 ip2 = &ru2->ru_first; 1104 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) 1105 *ip++ += *ip2++; 1106 } 1107 1108 void 1109 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, 1110 struct rusage_ext *rux2) 1111 { 1112 1113 rux->rux_runtime += rux2->rux_runtime; 1114 rux->rux_uticks += rux2->rux_uticks; 1115 rux->rux_sticks += rux2->rux_sticks; 1116 rux->rux_iticks += rux2->rux_iticks; 1117 rux->rux_uu += rux2->rux_uu; 1118 rux->rux_su += rux2->rux_su; 1119 rux->rux_tu += rux2->rux_tu; 1120 rucollect(ru, ru2); 1121 } 1122 1123 /* 1124 * Aggregate tick counts into the proc's rusage_ext. 1125 */ 1126 static void 1127 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td) 1128 { 1129 1130 rux->rux_runtime += td->td_incruntime; 1131 rux->rux_uticks += td->td_uticks; 1132 rux->rux_sticks += td->td_sticks; 1133 rux->rux_iticks += td->td_iticks; 1134 } 1135 1136 void 1137 ruxagg_locked(struct proc *p, struct thread *td) 1138 { 1139 THREAD_LOCK_ASSERT(td, MA_OWNED); 1140 PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED); 1141 1142 ruxagg_ext_locked(&p->p_rux, td); 1143 ruxagg_ext_locked(&td->td_rux, td); 1144 td->td_incruntime = 0; 1145 td->td_uticks = 0; 1146 td->td_iticks = 0; 1147 td->td_sticks = 0; 1148 } 1149 1150 void 1151 ruxagg(struct proc *p, struct thread *td) 1152 { 1153 1154 thread_lock(td); 1155 ruxagg_locked(p, td); 1156 thread_unlock(td); 1157 } 1158 1159 /* 1160 * Update the rusage_ext structure and fetch a valid aggregate rusage 1161 * for proc p if storage for one is supplied. 1162 */ 1163 void 1164 rufetch(struct proc *p, struct rusage *ru) 1165 { 1166 struct thread *td; 1167 1168 PROC_STATLOCK_ASSERT(p, MA_OWNED); 1169 1170 *ru = p->p_ru; 1171 if (p->p_numthreads > 0) { 1172 FOREACH_THREAD_IN_PROC(p, td) { 1173 ruxagg(p, td); 1174 rucollect(ru, &td->td_ru); 1175 } 1176 } 1177 } 1178 1179 /* 1180 * Atomically perform a rufetch and a calcru together. 1181 * Consumers, can safely assume the calcru is executed only once 1182 * rufetch is completed. 1183 */ 1184 void 1185 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, 1186 struct timeval *sp) 1187 { 1188 1189 PROC_STATLOCK(p); 1190 rufetch(p, ru); 1191 calcru(p, up, sp); 1192 PROC_STATUNLOCK(p); 1193 } 1194 1195 /* 1196 * Allocate a new resource limits structure and initialize its 1197 * reference count and mutex pointer. 1198 */ 1199 struct plimit * 1200 lim_alloc() 1201 { 1202 struct plimit *limp; 1203 1204 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); 1205 refcount_init(&limp->pl_refcnt, 1); 1206 return (limp); 1207 } 1208 1209 struct plimit * 1210 lim_hold(struct plimit *limp) 1211 { 1212 1213 refcount_acquire(&limp->pl_refcnt); 1214 return (limp); 1215 } 1216 1217 void 1218 lim_fork(struct proc *p1, struct proc *p2) 1219 { 1220 1221 PROC_LOCK_ASSERT(p1, MA_OWNED); 1222 PROC_LOCK_ASSERT(p2, MA_OWNED); 1223 1224 p2->p_limit = lim_hold(p1->p_limit); 1225 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); 1226 if (p1->p_cpulimit != RLIM_INFINITY) 1227 callout_reset_sbt(&p2->p_limco, SBT_1S, 0, 1228 lim_cb, p2, C_PREL(1)); 1229 } 1230 1231 void 1232 lim_free(struct plimit *limp) 1233 { 1234 1235 if (refcount_release(&limp->pl_refcnt)) 1236 free((void *)limp, M_PLIMIT); 1237 } 1238 1239 void 1240 lim_freen(struct plimit *limp, int n) 1241 { 1242 1243 if (refcount_releasen(&limp->pl_refcnt, n)) 1244 free((void *)limp, M_PLIMIT); 1245 } 1246 1247 /* 1248 * Make a copy of the plimit structure. 1249 * We share these structures copy-on-write after fork. 1250 */ 1251 void 1252 lim_copy(struct plimit *dst, struct plimit *src) 1253 { 1254 1255 KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit")); 1256 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); 1257 } 1258 1259 /* 1260 * Return the hard limit for a particular system resource. The 1261 * which parameter specifies the index into the rlimit array. 1262 */ 1263 rlim_t 1264 lim_max(struct thread *td, int which) 1265 { 1266 struct rlimit rl; 1267 1268 lim_rlimit(td, which, &rl); 1269 return (rl.rlim_max); 1270 } 1271 1272 rlim_t 1273 lim_max_proc(struct proc *p, int which) 1274 { 1275 struct rlimit rl; 1276 1277 lim_rlimit_proc(p, which, &rl); 1278 return (rl.rlim_max); 1279 } 1280 1281 /* 1282 * Return the current (soft) limit for a particular system resource. 1283 * The which parameter which specifies the index into the rlimit array 1284 */ 1285 rlim_t 1286 (lim_cur)(struct thread *td, int which) 1287 { 1288 struct rlimit rl; 1289 1290 lim_rlimit(td, which, &rl); 1291 return (rl.rlim_cur); 1292 } 1293 1294 rlim_t 1295 lim_cur_proc(struct proc *p, int which) 1296 { 1297 struct rlimit rl; 1298 1299 lim_rlimit_proc(p, which, &rl); 1300 return (rl.rlim_cur); 1301 } 1302 1303 /* 1304 * Return a copy of the entire rlimit structure for the system limit 1305 * specified by 'which' in the rlimit structure pointed to by 'rlp'. 1306 */ 1307 void 1308 lim_rlimit(struct thread *td, int which, struct rlimit *rlp) 1309 { 1310 struct proc *p = td->td_proc; 1311 1312 MPASS(td == curthread); 1313 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1314 ("request for invalid resource limit")); 1315 *rlp = td->td_limit->pl_rlimit[which]; 1316 if (p->p_sysent->sv_fixlimit != NULL) 1317 p->p_sysent->sv_fixlimit(rlp, which); 1318 } 1319 1320 void 1321 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp) 1322 { 1323 1324 PROC_LOCK_ASSERT(p, MA_OWNED); 1325 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1326 ("request for invalid resource limit")); 1327 *rlp = p->p_limit->pl_rlimit[which]; 1328 if (p->p_sysent->sv_fixlimit != NULL) 1329 p->p_sysent->sv_fixlimit(rlp, which); 1330 } 1331 1332 void 1333 uihashinit() 1334 { 1335 1336 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); 1337 rw_init(&uihashtbl_lock, "uidinfo hash"); 1338 } 1339 1340 /* 1341 * Look up a uidinfo struct for the parameter uid. 1342 * uihashtbl_lock must be locked. 1343 * Increase refcount on uidinfo struct returned. 1344 */ 1345 static struct uidinfo * 1346 uilookup(uid_t uid) 1347 { 1348 struct uihashhead *uipp; 1349 struct uidinfo *uip; 1350 1351 rw_assert(&uihashtbl_lock, RA_LOCKED); 1352 uipp = UIHASH(uid); 1353 LIST_FOREACH(uip, uipp, ui_hash) 1354 if (uip->ui_uid == uid) { 1355 uihold(uip); 1356 break; 1357 } 1358 1359 return (uip); 1360 } 1361 1362 /* 1363 * Find or allocate a struct uidinfo for a particular uid. 1364 * Returns with uidinfo struct referenced. 1365 * uifree() should be called on a struct uidinfo when released. 1366 */ 1367 struct uidinfo * 1368 uifind(uid_t uid) 1369 { 1370 struct uidinfo *new_uip, *uip; 1371 struct ucred *cred; 1372 1373 cred = curthread->td_ucred; 1374 if (cred->cr_uidinfo->ui_uid == uid) { 1375 uip = cred->cr_uidinfo; 1376 uihold(uip); 1377 return (uip); 1378 } else if (cred->cr_ruidinfo->ui_uid == uid) { 1379 uip = cred->cr_ruidinfo; 1380 uihold(uip); 1381 return (uip); 1382 } 1383 1384 rw_rlock(&uihashtbl_lock); 1385 uip = uilookup(uid); 1386 rw_runlock(&uihashtbl_lock); 1387 if (uip != NULL) 1388 return (uip); 1389 1390 new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO); 1391 racct_create(&new_uip->ui_racct); 1392 refcount_init(&new_uip->ui_ref, 1); 1393 new_uip->ui_uid = uid; 1394 1395 rw_wlock(&uihashtbl_lock); 1396 /* 1397 * There's a chance someone created our uidinfo while we 1398 * were in malloc and not holding the lock, so we have to 1399 * make sure we don't insert a duplicate uidinfo. 1400 */ 1401 if ((uip = uilookup(uid)) == NULL) { 1402 LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash); 1403 rw_wunlock(&uihashtbl_lock); 1404 uip = new_uip; 1405 } else { 1406 rw_wunlock(&uihashtbl_lock); 1407 racct_destroy(&new_uip->ui_racct); 1408 free(new_uip, M_UIDINFO); 1409 } 1410 return (uip); 1411 } 1412 1413 /* 1414 * Place another refcount on a uidinfo struct. 1415 */ 1416 void 1417 uihold(struct uidinfo *uip) 1418 { 1419 1420 refcount_acquire(&uip->ui_ref); 1421 } 1422 1423 /*- 1424 * Since uidinfo structs have a long lifetime, we use an 1425 * opportunistic refcounting scheme to avoid locking the lookup hash 1426 * for each release. 1427 * 1428 * If the refcount hits 0, we need to free the structure, 1429 * which means we need to lock the hash. 1430 * Optimal case: 1431 * After locking the struct and lowering the refcount, if we find 1432 * that we don't need to free, simply unlock and return. 1433 * Suboptimal case: 1434 * If refcount lowering results in need to free, bump the count 1435 * back up, lose the lock and acquire the locks in the proper 1436 * order to try again. 1437 */ 1438 void 1439 uifree(struct uidinfo *uip) 1440 { 1441 1442 if (refcount_release_if_not_last(&uip->ui_ref)) 1443 return; 1444 1445 rw_wlock(&uihashtbl_lock); 1446 if (refcount_release(&uip->ui_ref) == 0) { 1447 rw_wunlock(&uihashtbl_lock); 1448 return; 1449 } 1450 1451 racct_destroy(&uip->ui_racct); 1452 LIST_REMOVE(uip, ui_hash); 1453 rw_wunlock(&uihashtbl_lock); 1454 1455 if (uip->ui_sbsize != 0) 1456 printf("freeing uidinfo: uid = %d, sbsize = %ld\n", 1457 uip->ui_uid, uip->ui_sbsize); 1458 if (uip->ui_proccnt != 0) 1459 printf("freeing uidinfo: uid = %d, proccnt = %ld\n", 1460 uip->ui_uid, uip->ui_proccnt); 1461 if (uip->ui_vmsize != 0) 1462 printf("freeing uidinfo: uid = %d, swapuse = %lld\n", 1463 uip->ui_uid, (unsigned long long)uip->ui_vmsize); 1464 free(uip, M_UIDINFO); 1465 } 1466 1467 #ifdef RACCT 1468 void 1469 ui_racct_foreach(void (*callback)(struct racct *racct, 1470 void *arg2, void *arg3), void (*pre)(void), void (*post)(void), 1471 void *arg2, void *arg3) 1472 { 1473 struct uidinfo *uip; 1474 struct uihashhead *uih; 1475 1476 rw_rlock(&uihashtbl_lock); 1477 if (pre != NULL) 1478 (pre)(); 1479 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { 1480 LIST_FOREACH(uip, uih, ui_hash) { 1481 (callback)(uip->ui_racct, arg2, arg3); 1482 } 1483 } 1484 if (post != NULL) 1485 (post)(); 1486 rw_runlock(&uihashtbl_lock); 1487 } 1488 #endif 1489 1490 static inline int 1491 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name) 1492 { 1493 long new; 1494 1495 /* Don't allow them to exceed max, but allow subtraction. */ 1496 new = atomic_fetchadd_long(limit, (long)diff) + diff; 1497 if (diff > 0 && max != 0) { 1498 if (new < 0 || new > max) { 1499 atomic_subtract_long(limit, (long)diff); 1500 return (0); 1501 } 1502 } else if (new < 0) 1503 printf("negative %s for uid = %d\n", name, uip->ui_uid); 1504 return (1); 1505 } 1506 1507 /* 1508 * Change the count associated with number of processes 1509 * a given user is using. When 'max' is 0, don't enforce a limit 1510 */ 1511 int 1512 chgproccnt(struct uidinfo *uip, int diff, rlim_t max) 1513 { 1514 1515 return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt")); 1516 } 1517 1518 /* 1519 * Change the total socket buffer size a user has used. 1520 */ 1521 int 1522 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max) 1523 { 1524 int diff, rv; 1525 1526 diff = to - *hiwat; 1527 if (diff > 0 && max == 0) { 1528 rv = 0; 1529 } else { 1530 rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize"); 1531 if (rv != 0) 1532 *hiwat = to; 1533 } 1534 return (rv); 1535 } 1536 1537 /* 1538 * Change the count associated with number of pseudo-terminals 1539 * a given user is using. When 'max' is 0, don't enforce a limit 1540 */ 1541 int 1542 chgptscnt(struct uidinfo *uip, int diff, rlim_t max) 1543 { 1544 1545 return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt")); 1546 } 1547 1548 int 1549 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max) 1550 { 1551 1552 return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt")); 1553 } 1554 1555 int 1556 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max) 1557 { 1558 1559 return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt")); 1560 } 1561