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) { 108 109 case PRIO_PROCESS: 110 if (who == 0) 111 low = td->td_proc->p_nice; 112 else { 113 p = pfind(who); 114 if (p == NULL) 115 break; 116 if (p_cansee(td, p) == 0) 117 low = p->p_nice; 118 PROC_UNLOCK(p); 119 } 120 break; 121 122 case PRIO_PGRP: 123 sx_slock(&proctree_lock); 124 if (who == 0) { 125 pg = td->td_proc->p_pgrp; 126 PGRP_LOCK(pg); 127 } else { 128 pg = pgfind(who); 129 if (pg == NULL) { 130 sx_sunlock(&proctree_lock); 131 break; 132 } 133 } 134 sx_sunlock(&proctree_lock); 135 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 136 PROC_LOCK(p); 137 if (p->p_state == PRS_NORMAL && 138 p_cansee(td, p) == 0) { 139 if (p->p_nice < low) 140 low = p->p_nice; 141 } 142 PROC_UNLOCK(p); 143 } 144 PGRP_UNLOCK(pg); 145 break; 146 147 case PRIO_USER: 148 if (who == 0) 149 who = td->td_ucred->cr_uid; 150 sx_slock(&allproc_lock); 151 FOREACH_PROC_IN_SYSTEM(p) { 152 PROC_LOCK(p); 153 if (p->p_state == PRS_NORMAL && 154 p_cansee(td, p) == 0 && 155 p->p_ucred->cr_uid == who) { 156 if (p->p_nice < low) 157 low = p->p_nice; 158 } 159 PROC_UNLOCK(p); 160 } 161 sx_sunlock(&allproc_lock); 162 break; 163 164 default: 165 error = EINVAL; 166 break; 167 } 168 if (low == PRIO_MAX + 1 && error == 0) 169 error = ESRCH; 170 td->td_retval[0] = low; 171 return (error); 172 } 173 174 #ifndef _SYS_SYSPROTO_H_ 175 struct setpriority_args { 176 int which; 177 int who; 178 int prio; 179 }; 180 #endif 181 int 182 sys_setpriority(struct thread *td, struct setpriority_args *uap) 183 { 184 185 return (kern_setpriority(td, uap->which, uap->who, uap->prio)); 186 } 187 188 int 189 kern_setpriority(struct thread *td, int which, int who, int prio) 190 { 191 struct proc *curp, *p; 192 struct pgrp *pg; 193 int found = 0, error = 0; 194 195 curp = td->td_proc; 196 switch (which) { 197 case PRIO_PROCESS: 198 if (who == 0) { 199 PROC_LOCK(curp); 200 error = donice(td, curp, prio); 201 PROC_UNLOCK(curp); 202 } else { 203 p = pfind(who); 204 if (p == NULL) 205 break; 206 error = p_cansee(td, p); 207 if (error == 0) 208 error = donice(td, p, prio); 209 PROC_UNLOCK(p); 210 } 211 found++; 212 break; 213 214 case PRIO_PGRP: 215 sx_slock(&proctree_lock); 216 if (who == 0) { 217 pg = curp->p_pgrp; 218 PGRP_LOCK(pg); 219 } else { 220 pg = pgfind(who); 221 if (pg == NULL) { 222 sx_sunlock(&proctree_lock); 223 break; 224 } 225 } 226 sx_sunlock(&proctree_lock); 227 LIST_FOREACH(p, &pg->pg_members, p_pglist) { 228 PROC_LOCK(p); 229 if (p->p_state == PRS_NORMAL && 230 p_cansee(td, p) == 0) { 231 error = donice(td, p, prio); 232 found++; 233 } 234 PROC_UNLOCK(p); 235 } 236 PGRP_UNLOCK(pg); 237 break; 238 239 case PRIO_USER: 240 if (who == 0) 241 who = td->td_ucred->cr_uid; 242 sx_slock(&allproc_lock); 243 FOREACH_PROC_IN_SYSTEM(p) { 244 PROC_LOCK(p); 245 if (p->p_state == PRS_NORMAL && 246 p->p_ucred->cr_uid == who && 247 p_cansee(td, p) == 0) { 248 error = donice(td, p, prio); 249 found++; 250 } 251 PROC_UNLOCK(p); 252 } 253 sx_sunlock(&allproc_lock); 254 break; 255 256 default: 257 error = EINVAL; 258 break; 259 } 260 if (found == 0 && error == 0) 261 error = ESRCH; 262 return (error); 263 } 264 265 /* 266 * Set "nice" for a (whole) process. 267 */ 268 static int 269 donice(struct thread *td, struct proc *p, int n) 270 { 271 int error; 272 273 PROC_LOCK_ASSERT(p, MA_OWNED); 274 if ((error = p_cansched(td, p))) 275 return (error); 276 if (n > PRIO_MAX) 277 n = PRIO_MAX; 278 if (n < PRIO_MIN) 279 n = PRIO_MIN; 280 if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0) 281 return (EACCES); 282 sched_nice(p, n); 283 return (0); 284 } 285 286 static int unprivileged_idprio; 287 SYSCTL_INT(_security_bsd, OID_AUTO, unprivileged_idprio, CTLFLAG_RW, 288 &unprivileged_idprio, 0, "Allow non-root users to set an idle priority"); 289 290 /* 291 * Set realtime priority for LWP. 292 */ 293 #ifndef _SYS_SYSPROTO_H_ 294 struct rtprio_thread_args { 295 int function; 296 lwpid_t lwpid; 297 struct rtprio *rtp; 298 }; 299 #endif 300 int 301 sys_rtprio_thread(struct thread *td, struct rtprio_thread_args *uap) 302 { 303 struct proc *p; 304 struct rtprio rtp; 305 struct thread *td1; 306 int cierror, error; 307 308 /* Perform copyin before acquiring locks if needed. */ 309 if (uap->function == RTP_SET) 310 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 311 else 312 cierror = 0; 313 314 if (uap->lwpid == 0 || uap->lwpid == td->td_tid) { 315 p = td->td_proc; 316 td1 = td; 317 PROC_LOCK(p); 318 } else { 319 /* Only look up thread in current process */ 320 td1 = tdfind(uap->lwpid, curproc->p_pid); 321 if (td1 == NULL) 322 return (ESRCH); 323 p = td1->td_proc; 324 } 325 326 switch (uap->function) { 327 case RTP_LOOKUP: 328 if ((error = p_cansee(td, p))) 329 break; 330 pri_to_rtp(td1, &rtp); 331 PROC_UNLOCK(p); 332 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 333 case RTP_SET: 334 if ((error = p_cansched(td, p)) || (error = cierror)) 335 break; 336 337 /* Disallow setting rtprio in most cases if not superuser. */ 338 339 /* 340 * Realtime priority has to be restricted for reasons which 341 * should be obvious. However, for idleprio processes, there is 342 * a potential for system deadlock if an idleprio process gains 343 * a lock on a resource that other processes need (and the 344 * idleprio process can't run due to a CPU-bound normal 345 * process). Fix me! XXX 346 * 347 * This problem is not only related to idleprio process. 348 * A user level program can obtain a file lock and hold it 349 * indefinitely. Additionally, without idleprio processes it is 350 * still conceivable that a program with low priority will never 351 * get to run. In short, allowing this feature might make it 352 * easier to lock a resource indefinitely, but it is not the 353 * only thing that makes it possible. 354 */ 355 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || 356 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && 357 unprivileged_idprio == 0)) { 358 error = priv_check(td, PRIV_SCHED_RTPRIO); 359 if (error) 360 break; 361 } 362 error = rtp_to_pri(&rtp, td1); 363 break; 364 default: 365 error = EINVAL; 366 break; 367 } 368 PROC_UNLOCK(p); 369 return (error); 370 } 371 372 /* 373 * Set realtime priority. 374 */ 375 #ifndef _SYS_SYSPROTO_H_ 376 struct rtprio_args { 377 int function; 378 pid_t pid; 379 struct rtprio *rtp; 380 }; 381 #endif 382 int 383 sys_rtprio(struct thread *td, struct rtprio_args *uap) 384 { 385 struct proc *p; 386 struct thread *tdp; 387 struct rtprio rtp; 388 int cierror, error; 389 390 /* Perform copyin before acquiring locks if needed. */ 391 if (uap->function == RTP_SET) 392 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio)); 393 else 394 cierror = 0; 395 396 if (uap->pid == 0) { 397 p = td->td_proc; 398 PROC_LOCK(p); 399 } else { 400 p = pfind(uap->pid); 401 if (p == NULL) 402 return (ESRCH); 403 } 404 405 switch (uap->function) { 406 case RTP_LOOKUP: 407 if ((error = p_cansee(td, p))) 408 break; 409 /* 410 * Return OUR priority if no pid specified, 411 * or if one is, report the highest priority 412 * in the process. There isn't much more you can do as 413 * there is only room to return a single priority. 414 * Note: specifying our own pid is not the same 415 * as leaving it zero. 416 */ 417 if (uap->pid == 0) { 418 pri_to_rtp(td, &rtp); 419 } else { 420 struct rtprio rtp2; 421 422 rtp.type = RTP_PRIO_IDLE; 423 rtp.prio = RTP_PRIO_MAX; 424 FOREACH_THREAD_IN_PROC(p, tdp) { 425 pri_to_rtp(tdp, &rtp2); 426 if (rtp2.type < rtp.type || 427 (rtp2.type == rtp.type && 428 rtp2.prio < rtp.prio)) { 429 rtp.type = rtp2.type; 430 rtp.prio = rtp2.prio; 431 } 432 } 433 } 434 PROC_UNLOCK(p); 435 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio))); 436 case RTP_SET: 437 if ((error = p_cansched(td, p)) || (error = cierror)) 438 break; 439 440 /* 441 * Disallow setting rtprio in most cases if not superuser. 442 * See the comment in sys_rtprio_thread about idprio 443 * threads holding a lock. 444 */ 445 if (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_REALTIME || 446 (RTP_PRIO_BASE(rtp.type) == RTP_PRIO_IDLE && 447 !unprivileged_idprio)) { 448 error = priv_check(td, PRIV_SCHED_RTPRIO); 449 if (error) 450 break; 451 } 452 453 /* 454 * If we are setting our own priority, set just our 455 * thread but if we are doing another process, 456 * do all the threads on that process. If we 457 * specify our own pid we do the latter. 458 */ 459 if (uap->pid == 0) { 460 error = rtp_to_pri(&rtp, td); 461 } else { 462 FOREACH_THREAD_IN_PROC(p, td) { 463 if ((error = rtp_to_pri(&rtp, td)) != 0) 464 break; 465 } 466 } 467 break; 468 default: 469 error = EINVAL; 470 break; 471 } 472 PROC_UNLOCK(p); 473 return (error); 474 } 475 476 int 477 rtp_to_pri(struct rtprio *rtp, struct thread *td) 478 { 479 u_char newpri, oldclass, oldpri; 480 481 switch (RTP_PRIO_BASE(rtp->type)) { 482 case RTP_PRIO_REALTIME: 483 if (rtp->prio > RTP_PRIO_MAX) 484 return (EINVAL); 485 newpri = PRI_MIN_REALTIME + rtp->prio; 486 break; 487 case RTP_PRIO_NORMAL: 488 if (rtp->prio > (PRI_MAX_TIMESHARE - PRI_MIN_TIMESHARE)) 489 return (EINVAL); 490 newpri = PRI_MIN_TIMESHARE + rtp->prio; 491 break; 492 case RTP_PRIO_IDLE: 493 if (rtp->prio > RTP_PRIO_MAX) 494 return (EINVAL); 495 newpri = PRI_MIN_IDLE + rtp->prio; 496 break; 497 default: 498 return (EINVAL); 499 } 500 501 thread_lock(td); 502 oldclass = td->td_pri_class; 503 sched_class(td, rtp->type); /* XXX fix */ 504 oldpri = td->td_user_pri; 505 sched_user_prio(td, newpri); 506 if (td->td_user_pri != oldpri && (oldclass != RTP_PRIO_NORMAL || 507 td->td_pri_class != RTP_PRIO_NORMAL)) 508 sched_prio(td, td->td_user_pri); 509 if (TD_ON_UPILOCK(td) && oldpri != newpri) { 510 critical_enter(); 511 thread_unlock(td); 512 umtx_pi_adjust(td, oldpri); 513 critical_exit(); 514 } else 515 thread_unlock(td); 516 return (0); 517 } 518 519 void 520 pri_to_rtp(struct thread *td, struct rtprio *rtp) 521 { 522 523 thread_lock(td); 524 switch (PRI_BASE(td->td_pri_class)) { 525 case PRI_REALTIME: 526 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME; 527 break; 528 case PRI_TIMESHARE: 529 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE; 530 break; 531 case PRI_IDLE: 532 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE; 533 break; 534 default: 535 break; 536 } 537 rtp->type = td->td_pri_class; 538 thread_unlock(td); 539 } 540 541 #if defined(COMPAT_43) 542 #ifndef _SYS_SYSPROTO_H_ 543 struct osetrlimit_args { 544 u_int which; 545 struct orlimit *rlp; 546 }; 547 #endif 548 int 549 osetrlimit(struct thread *td, struct osetrlimit_args *uap) 550 { 551 struct orlimit olim; 552 struct rlimit lim; 553 int error; 554 555 if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit)))) 556 return (error); 557 lim.rlim_cur = olim.rlim_cur; 558 lim.rlim_max = olim.rlim_max; 559 error = kern_setrlimit(td, uap->which, &lim); 560 return (error); 561 } 562 563 #ifndef _SYS_SYSPROTO_H_ 564 struct ogetrlimit_args { 565 u_int which; 566 struct orlimit *rlp; 567 }; 568 #endif 569 int 570 ogetrlimit(struct thread *td, struct ogetrlimit_args *uap) 571 { 572 struct orlimit olim; 573 struct rlimit rl; 574 int error; 575 576 if (uap->which >= RLIM_NLIMITS) 577 return (EINVAL); 578 lim_rlimit(td, uap->which, &rl); 579 580 /* 581 * XXX would be more correct to convert only RLIM_INFINITY to the 582 * old RLIM_INFINITY and fail with EOVERFLOW for other larger 583 * values. Most 64->32 and 32->16 conversions, including not 584 * unimportant ones of uids are even more broken than what we 585 * do here (they blindly truncate). We don't do this correctly 586 * here since we have little experience with EOVERFLOW yet. 587 * Elsewhere, getuid() can't fail... 588 */ 589 olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur; 590 olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max; 591 error = copyout(&olim, uap->rlp, sizeof(olim)); 592 return (error); 593 } 594 #endif /* COMPAT_43 */ 595 596 #ifndef _SYS_SYSPROTO_H_ 597 struct __setrlimit_args { 598 u_int which; 599 struct rlimit *rlp; 600 }; 601 #endif 602 int 603 sys_setrlimit(struct thread *td, struct __setrlimit_args *uap) 604 { 605 struct rlimit alim; 606 int error; 607 608 if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit)))) 609 return (error); 610 error = kern_setrlimit(td, uap->which, &alim); 611 return (error); 612 } 613 614 static void 615 lim_cb(void *arg) 616 { 617 struct rlimit rlim; 618 struct thread *td; 619 struct proc *p; 620 621 p = arg; 622 PROC_LOCK_ASSERT(p, MA_OWNED); 623 /* 624 * Check if the process exceeds its cpu resource allocation. If 625 * it reaches the max, arrange to kill the process in ast(). 626 */ 627 if (p->p_cpulimit == RLIM_INFINITY) 628 return; 629 PROC_STATLOCK(p); 630 FOREACH_THREAD_IN_PROC(p, td) { 631 ruxagg(p, td); 632 } 633 PROC_STATUNLOCK(p); 634 if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) { 635 lim_rlimit_proc(p, RLIMIT_CPU, &rlim); 636 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) { 637 killproc(p, "exceeded maximum CPU limit"); 638 } else { 639 if (p->p_cpulimit < rlim.rlim_max) 640 p->p_cpulimit += 5; 641 kern_psignal(p, SIGXCPU); 642 } 643 } 644 if ((p->p_flag & P_WEXIT) == 0) 645 callout_reset_sbt(&p->p_limco, SBT_1S, 0, 646 lim_cb, p, C_PREL(1)); 647 } 648 649 int 650 kern_setrlimit(struct thread *td, u_int which, struct rlimit *limp) 651 { 652 653 return (kern_proc_setrlimit(td, td->td_proc, which, limp)); 654 } 655 656 int 657 kern_proc_setrlimit(struct thread *td, struct proc *p, u_int which, 658 struct rlimit *limp) 659 { 660 struct plimit *newlim, *oldlim; 661 struct rlimit *alimp; 662 struct rlimit oldssiz; 663 int error; 664 665 if (which >= RLIM_NLIMITS) 666 return (EINVAL); 667 668 /* 669 * Preserve historical bugs by treating negative limits as unsigned. 670 */ 671 if (limp->rlim_cur < 0) 672 limp->rlim_cur = RLIM_INFINITY; 673 if (limp->rlim_max < 0) 674 limp->rlim_max = RLIM_INFINITY; 675 676 oldssiz.rlim_cur = 0; 677 newlim = lim_alloc(); 678 PROC_LOCK(p); 679 oldlim = p->p_limit; 680 alimp = &oldlim->pl_rlimit[which]; 681 if (limp->rlim_cur > alimp->rlim_max || 682 limp->rlim_max > alimp->rlim_max) 683 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) { 684 PROC_UNLOCK(p); 685 lim_free(newlim); 686 return (error); 687 } 688 if (limp->rlim_cur > limp->rlim_max) 689 limp->rlim_cur = limp->rlim_max; 690 lim_copy(newlim, oldlim); 691 alimp = &newlim->pl_rlimit[which]; 692 693 switch (which) { 694 695 case RLIMIT_CPU: 696 if (limp->rlim_cur != RLIM_INFINITY && 697 p->p_cpulimit == RLIM_INFINITY) 698 callout_reset_sbt(&p->p_limco, SBT_1S, 0, 699 lim_cb, p, C_PREL(1)); 700 p->p_cpulimit = limp->rlim_cur; 701 break; 702 case RLIMIT_DATA: 703 if (limp->rlim_cur > maxdsiz) 704 limp->rlim_cur = maxdsiz; 705 if (limp->rlim_max > maxdsiz) 706 limp->rlim_max = maxdsiz; 707 break; 708 709 case RLIMIT_STACK: 710 if (limp->rlim_cur > maxssiz) 711 limp->rlim_cur = maxssiz; 712 if (limp->rlim_max > maxssiz) 713 limp->rlim_max = maxssiz; 714 oldssiz = *alimp; 715 if (p->p_sysent->sv_fixlimit != NULL) 716 p->p_sysent->sv_fixlimit(&oldssiz, 717 RLIMIT_STACK); 718 break; 719 720 case RLIMIT_NOFILE: 721 if (limp->rlim_cur > maxfilesperproc) 722 limp->rlim_cur = maxfilesperproc; 723 if (limp->rlim_max > maxfilesperproc) 724 limp->rlim_max = maxfilesperproc; 725 break; 726 727 case RLIMIT_NPROC: 728 if (limp->rlim_cur > maxprocperuid) 729 limp->rlim_cur = maxprocperuid; 730 if (limp->rlim_max > maxprocperuid) 731 limp->rlim_max = maxprocperuid; 732 if (limp->rlim_cur < 1) 733 limp->rlim_cur = 1; 734 if (limp->rlim_max < 1) 735 limp->rlim_max = 1; 736 break; 737 } 738 if (p->p_sysent->sv_fixlimit != NULL) 739 p->p_sysent->sv_fixlimit(limp, which); 740 *alimp = *limp; 741 p->p_limit = newlim; 742 PROC_UPDATE_COW(p); 743 PROC_UNLOCK(p); 744 lim_free(oldlim); 745 746 if (which == RLIMIT_STACK && 747 /* 748 * Skip calls from exec_new_vmspace(), done when stack is 749 * not mapped yet. 750 */ 751 (td != curthread || (p->p_flag & P_INEXEC) == 0)) { 752 /* 753 * Stack is allocated to the max at exec time with only 754 * "rlim_cur" bytes accessible. If stack limit is going 755 * up make more accessible, if going down make inaccessible. 756 */ 757 if (limp->rlim_cur != oldssiz.rlim_cur) { 758 vm_offset_t addr; 759 vm_size_t size; 760 vm_prot_t prot; 761 762 if (limp->rlim_cur > oldssiz.rlim_cur) { 763 prot = p->p_sysent->sv_stackprot; 764 size = limp->rlim_cur - oldssiz.rlim_cur; 765 addr = p->p_sysent->sv_usrstack - 766 limp->rlim_cur; 767 } else { 768 prot = VM_PROT_NONE; 769 size = oldssiz.rlim_cur - limp->rlim_cur; 770 addr = p->p_sysent->sv_usrstack - 771 oldssiz.rlim_cur; 772 } 773 addr = trunc_page(addr); 774 size = round_page(size); 775 (void)vm_map_protect(&p->p_vmspace->vm_map, 776 addr, addr + size, prot, FALSE); 777 } 778 } 779 780 return (0); 781 } 782 783 #ifndef _SYS_SYSPROTO_H_ 784 struct __getrlimit_args { 785 u_int which; 786 struct rlimit *rlp; 787 }; 788 #endif 789 /* ARGSUSED */ 790 int 791 sys_getrlimit(struct thread *td, struct __getrlimit_args *uap) 792 { 793 struct rlimit rlim; 794 int error; 795 796 if (uap->which >= RLIM_NLIMITS) 797 return (EINVAL); 798 lim_rlimit(td, uap->which, &rlim); 799 error = copyout(&rlim, uap->rlp, sizeof(struct rlimit)); 800 return (error); 801 } 802 803 /* 804 * Transform the running time and tick information for children of proc p 805 * into user and system time usage. 806 */ 807 void 808 calccru(struct proc *p, struct timeval *up, struct timeval *sp) 809 { 810 811 PROC_LOCK_ASSERT(p, MA_OWNED); 812 calcru1(p, &p->p_crux, up, sp); 813 } 814 815 /* 816 * Transform the running time and tick information in proc p into user 817 * and system time usage. If appropriate, include the current time slice 818 * on this CPU. 819 */ 820 void 821 calcru(struct proc *p, struct timeval *up, struct timeval *sp) 822 { 823 struct thread *td; 824 uint64_t runtime, u; 825 826 PROC_LOCK_ASSERT(p, MA_OWNED); 827 PROC_STATLOCK_ASSERT(p, MA_OWNED); 828 /* 829 * If we are getting stats for the current process, then add in the 830 * stats that this thread has accumulated in its current time slice. 831 * We reset the thread and CPU state as if we had performed a context 832 * switch right here. 833 */ 834 td = curthread; 835 if (td->td_proc == p) { 836 u = cpu_ticks(); 837 runtime = u - PCPU_GET(switchtime); 838 td->td_runtime += runtime; 839 td->td_incruntime += runtime; 840 PCPU_SET(switchtime, u); 841 } 842 /* Make sure the per-thread stats are current. */ 843 FOREACH_THREAD_IN_PROC(p, td) { 844 if (td->td_incruntime == 0) 845 continue; 846 ruxagg(p, td); 847 } 848 calcru1(p, &p->p_rux, up, sp); 849 } 850 851 /* Collect resource usage for a single thread. */ 852 void 853 rufetchtd(struct thread *td, struct rusage *ru) 854 { 855 struct proc *p; 856 uint64_t runtime, u; 857 858 p = td->td_proc; 859 PROC_STATLOCK_ASSERT(p, MA_OWNED); 860 THREAD_LOCK_ASSERT(td, MA_OWNED); 861 /* 862 * If we are getting stats for the current thread, then add in the 863 * stats that this thread has accumulated in its current time slice. 864 * We reset the thread and CPU state as if we had performed a context 865 * switch right here. 866 */ 867 if (td == curthread) { 868 u = cpu_ticks(); 869 runtime = u - PCPU_GET(switchtime); 870 td->td_runtime += runtime; 871 td->td_incruntime += runtime; 872 PCPU_SET(switchtime, u); 873 } 874 ruxagg_locked(p, td); 875 *ru = td->td_ru; 876 calcru1(p, &td->td_rux, &ru->ru_utime, &ru->ru_stime); 877 } 878 879 /* XXX: the MI version is too slow to use: */ 880 #ifndef __HAVE_INLINE_FLSLL 881 #define flsll(x) (fls((x) >> 32) != 0 ? fls((x) >> 32) + 32 : fls(x)) 882 #endif 883 884 static uint64_t 885 mul64_by_fraction(uint64_t a, uint64_t b, uint64_t c) 886 { 887 uint64_t acc, bh, bl; 888 int i, s, sa, sb; 889 890 /* 891 * Calculate (a * b) / c accurately enough without overflowing. c 892 * must be nonzero, and its top bit must be 0. a or b must be 893 * <= c, and the implementation is tuned for b <= c. 894 * 895 * The comments about times are for use in calcru1() with units of 896 * microseconds for 'a' and stathz ticks at 128 Hz for b and c. 897 * 898 * Let n be the number of top zero bits in c. Each iteration 899 * either returns, or reduces b by right shifting it by at least n. 900 * The number of iterations is at most 1 + 64 / n, and the error is 901 * at most the number of iterations. 902 * 903 * It is very unusual to need even 2 iterations. Previous 904 * implementations overflowed essentially by returning early in the 905 * first iteration, with n = 38 giving overflow at 105+ hours and 906 * n = 32 giving overlow at at 388+ days despite a more careful 907 * calculation. 388 days is a reasonable uptime, and the calculation 908 * needs to work for the uptime times the number of CPUs since 'a' 909 * is per-process. 910 */ 911 if (a >= (uint64_t)1 << 63) 912 return (0); /* Unsupported arg -- can't happen. */ 913 acc = 0; 914 for (i = 0; i < 128; i++) { 915 sa = flsll(a); 916 sb = flsll(b); 917 if (sa + sb <= 64) 918 /* Up to 105 hours on first iteration. */ 919 return (acc + (a * b) / c); 920 if (a >= c) { 921 /* 922 * This reduction is based on a = q * c + r, with the 923 * remainder r < c. 'a' may be large to start, and 924 * moving bits from b into 'a' at the end of the loop 925 * sets the top bit of 'a', so the reduction makes 926 * significant progress. 927 */ 928 acc += (a / c) * b; 929 a %= c; 930 sa = flsll(a); 931 if (sa + sb <= 64) 932 /* Up to 388 days on first iteration. */ 933 return (acc + (a * b) / c); 934 } 935 936 /* 937 * This step writes a * b as a * ((bh << s) + bl) = 938 * a * (bh << s) + a * bl = (a << s) * bh + a * bl. The 2 939 * additive terms are handled separately. Splitting in 940 * this way is linear except for rounding errors. 941 * 942 * s = 64 - sa is the maximum such that a << s fits in 64 943 * bits. Since a < c and c has at least 1 zero top bit, 944 * sa < 64 and s > 0. Thus this step makes progress by 945 * reducing b (it increases 'a', but taking remainders on 946 * the next iteration completes the reduction). 947 * 948 * Finally, the choice for s is just what is needed to keep 949 * a * bl from overflowing, so we don't need complications 950 * like a recursive call mul64_by_fraction(a, bl, c) to 951 * handle the second additive term. 952 */ 953 s = 64 - sa; 954 bh = b >> s; 955 bl = b - (bh << s); 956 acc += (a * bl) / c; 957 a <<= s; 958 b = bh; 959 } 960 return (0); /* Algorithm failure -- can't happen. */ 961 } 962 963 static void 964 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up, 965 struct timeval *sp) 966 { 967 /* {user, system, interrupt, total} {ticks, usec}: */ 968 uint64_t ut, uu, st, su, it, tt, tu; 969 970 ut = ruxp->rux_uticks; 971 st = ruxp->rux_sticks; 972 it = ruxp->rux_iticks; 973 tt = ut + st + it; 974 if (tt == 0) { 975 /* Avoid divide by zero */ 976 st = 1; 977 tt = 1; 978 } 979 tu = cputick2usec(ruxp->rux_runtime); 980 if ((int64_t)tu < 0) { 981 /* XXX: this should be an assert /phk */ 982 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n", 983 (intmax_t)tu, p->p_pid, p->p_comm); 984 tu = ruxp->rux_tu; 985 } 986 987 /* Subdivide tu. Avoid overflow in the multiplications. */ 988 if (__predict_true(tu <= ((uint64_t)1 << 38) && tt <= (1 << 26))) { 989 /* Up to 76 hours when stathz is 128. */ 990 uu = (tu * ut) / tt; 991 su = (tu * st) / tt; 992 } else { 993 uu = mul64_by_fraction(tu, ut, tt); 994 su = mul64_by_fraction(tu, st, tt); 995 } 996 997 if (tu >= ruxp->rux_tu) { 998 /* 999 * The normal case, time increased. 1000 * Enforce monotonicity of bucketed numbers. 1001 */ 1002 if (uu < ruxp->rux_uu) 1003 uu = ruxp->rux_uu; 1004 if (su < ruxp->rux_su) 1005 su = ruxp->rux_su; 1006 } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) { 1007 /* 1008 * When we calibrate the cputicker, it is not uncommon to 1009 * see the presumably fixed frequency increase slightly over 1010 * time as a result of thermal stabilization and NTP 1011 * discipline (of the reference clock). We therefore ignore 1012 * a bit of backwards slop because we expect to catch up 1013 * shortly. We use a 3 microsecond limit to catch low 1014 * counts and a 1% limit for high counts. 1015 */ 1016 uu = ruxp->rux_uu; 1017 su = ruxp->rux_su; 1018 tu = ruxp->rux_tu; 1019 } else { /* tu < ruxp->rux_tu */ 1020 /* 1021 * What happened here was likely that a laptop, which ran at 1022 * a reduced clock frequency at boot, kicked into high gear. 1023 * The wisdom of spamming this message in that case is 1024 * dubious, but it might also be indicative of something 1025 * serious, so lets keep it and hope laptops can be made 1026 * more truthful about their CPU speed via ACPI. 1027 */ 1028 printf("calcru: runtime went backwards from %ju usec " 1029 "to %ju usec for pid %d (%s)\n", 1030 (uintmax_t)ruxp->rux_tu, (uintmax_t)tu, 1031 p->p_pid, p->p_comm); 1032 } 1033 1034 ruxp->rux_uu = uu; 1035 ruxp->rux_su = su; 1036 ruxp->rux_tu = tu; 1037 1038 up->tv_sec = uu / 1000000; 1039 up->tv_usec = uu % 1000000; 1040 sp->tv_sec = su / 1000000; 1041 sp->tv_usec = su % 1000000; 1042 } 1043 1044 #ifndef _SYS_SYSPROTO_H_ 1045 struct getrusage_args { 1046 int who; 1047 struct rusage *rusage; 1048 }; 1049 #endif 1050 int 1051 sys_getrusage(struct thread *td, struct getrusage_args *uap) 1052 { 1053 struct rusage ru; 1054 int error; 1055 1056 error = kern_getrusage(td, uap->who, &ru); 1057 if (error == 0) 1058 error = copyout(&ru, uap->rusage, sizeof(struct rusage)); 1059 return (error); 1060 } 1061 1062 int 1063 kern_getrusage(struct thread *td, int who, struct rusage *rup) 1064 { 1065 struct proc *p; 1066 int error; 1067 1068 error = 0; 1069 p = td->td_proc; 1070 PROC_LOCK(p); 1071 switch (who) { 1072 case RUSAGE_SELF: 1073 rufetchcalc(p, rup, &rup->ru_utime, 1074 &rup->ru_stime); 1075 break; 1076 1077 case RUSAGE_CHILDREN: 1078 *rup = p->p_stats->p_cru; 1079 calccru(p, &rup->ru_utime, &rup->ru_stime); 1080 break; 1081 1082 case RUSAGE_THREAD: 1083 PROC_STATLOCK(p); 1084 thread_lock(td); 1085 rufetchtd(td, rup); 1086 thread_unlock(td); 1087 PROC_STATUNLOCK(p); 1088 break; 1089 1090 default: 1091 error = EINVAL; 1092 } 1093 PROC_UNLOCK(p); 1094 return (error); 1095 } 1096 1097 void 1098 rucollect(struct rusage *ru, struct rusage *ru2) 1099 { 1100 long *ip, *ip2; 1101 int i; 1102 1103 if (ru->ru_maxrss < ru2->ru_maxrss) 1104 ru->ru_maxrss = ru2->ru_maxrss; 1105 ip = &ru->ru_first; 1106 ip2 = &ru2->ru_first; 1107 for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--) 1108 *ip++ += *ip2++; 1109 } 1110 1111 void 1112 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2, 1113 struct rusage_ext *rux2) 1114 { 1115 1116 rux->rux_runtime += rux2->rux_runtime; 1117 rux->rux_uticks += rux2->rux_uticks; 1118 rux->rux_sticks += rux2->rux_sticks; 1119 rux->rux_iticks += rux2->rux_iticks; 1120 rux->rux_uu += rux2->rux_uu; 1121 rux->rux_su += rux2->rux_su; 1122 rux->rux_tu += rux2->rux_tu; 1123 rucollect(ru, ru2); 1124 } 1125 1126 /* 1127 * Aggregate tick counts into the proc's rusage_ext. 1128 */ 1129 static void 1130 ruxagg_ext_locked(struct rusage_ext *rux, struct thread *td) 1131 { 1132 1133 rux->rux_runtime += td->td_incruntime; 1134 rux->rux_uticks += td->td_uticks; 1135 rux->rux_sticks += td->td_sticks; 1136 rux->rux_iticks += td->td_iticks; 1137 } 1138 1139 void 1140 ruxagg_locked(struct proc *p, struct thread *td) 1141 { 1142 THREAD_LOCK_ASSERT(td, MA_OWNED); 1143 PROC_STATLOCK_ASSERT(td->td_proc, MA_OWNED); 1144 1145 ruxagg_ext_locked(&p->p_rux, td); 1146 ruxagg_ext_locked(&td->td_rux, td); 1147 td->td_incruntime = 0; 1148 td->td_uticks = 0; 1149 td->td_iticks = 0; 1150 td->td_sticks = 0; 1151 } 1152 1153 void 1154 ruxagg(struct proc *p, struct thread *td) 1155 { 1156 1157 thread_lock(td); 1158 ruxagg_locked(p, td); 1159 thread_unlock(td); 1160 } 1161 1162 /* 1163 * Update the rusage_ext structure and fetch a valid aggregate rusage 1164 * for proc p if storage for one is supplied. 1165 */ 1166 void 1167 rufetch(struct proc *p, struct rusage *ru) 1168 { 1169 struct thread *td; 1170 1171 PROC_STATLOCK_ASSERT(p, MA_OWNED); 1172 1173 *ru = p->p_ru; 1174 if (p->p_numthreads > 0) { 1175 FOREACH_THREAD_IN_PROC(p, td) { 1176 ruxagg(p, td); 1177 rucollect(ru, &td->td_ru); 1178 } 1179 } 1180 } 1181 1182 /* 1183 * Atomically perform a rufetch and a calcru together. 1184 * Consumers, can safely assume the calcru is executed only once 1185 * rufetch is completed. 1186 */ 1187 void 1188 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up, 1189 struct timeval *sp) 1190 { 1191 1192 PROC_STATLOCK(p); 1193 rufetch(p, ru); 1194 calcru(p, up, sp); 1195 PROC_STATUNLOCK(p); 1196 } 1197 1198 /* 1199 * Allocate a new resource limits structure and initialize its 1200 * reference count and mutex pointer. 1201 */ 1202 struct plimit * 1203 lim_alloc() 1204 { 1205 struct plimit *limp; 1206 1207 limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK); 1208 refcount_init(&limp->pl_refcnt, 1); 1209 return (limp); 1210 } 1211 1212 struct plimit * 1213 lim_hold(struct plimit *limp) 1214 { 1215 1216 refcount_acquire(&limp->pl_refcnt); 1217 return (limp); 1218 } 1219 1220 void 1221 lim_fork(struct proc *p1, struct proc *p2) 1222 { 1223 1224 PROC_LOCK_ASSERT(p1, MA_OWNED); 1225 PROC_LOCK_ASSERT(p2, MA_OWNED); 1226 1227 p2->p_limit = lim_hold(p1->p_limit); 1228 callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0); 1229 if (p1->p_cpulimit != RLIM_INFINITY) 1230 callout_reset_sbt(&p2->p_limco, SBT_1S, 0, 1231 lim_cb, p2, C_PREL(1)); 1232 } 1233 1234 void 1235 lim_free(struct plimit *limp) 1236 { 1237 1238 if (refcount_release(&limp->pl_refcnt)) 1239 free((void *)limp, M_PLIMIT); 1240 } 1241 1242 /* 1243 * Make a copy of the plimit structure. 1244 * We share these structures copy-on-write after fork. 1245 */ 1246 void 1247 lim_copy(struct plimit *dst, struct plimit *src) 1248 { 1249 1250 KASSERT(dst->pl_refcnt <= 1, ("lim_copy to shared limit")); 1251 bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit)); 1252 } 1253 1254 /* 1255 * Return the hard limit for a particular system resource. The 1256 * which parameter specifies the index into the rlimit array. 1257 */ 1258 rlim_t 1259 lim_max(struct thread *td, int which) 1260 { 1261 struct rlimit rl; 1262 1263 lim_rlimit(td, which, &rl); 1264 return (rl.rlim_max); 1265 } 1266 1267 rlim_t 1268 lim_max_proc(struct proc *p, int which) 1269 { 1270 struct rlimit rl; 1271 1272 lim_rlimit_proc(p, which, &rl); 1273 return (rl.rlim_max); 1274 } 1275 1276 /* 1277 * Return the current (soft) limit for a particular system resource. 1278 * The which parameter which specifies the index into the rlimit array 1279 */ 1280 rlim_t 1281 (lim_cur)(struct thread *td, int which) 1282 { 1283 struct rlimit rl; 1284 1285 lim_rlimit(td, which, &rl); 1286 return (rl.rlim_cur); 1287 } 1288 1289 rlim_t 1290 lim_cur_proc(struct proc *p, int which) 1291 { 1292 struct rlimit rl; 1293 1294 lim_rlimit_proc(p, which, &rl); 1295 return (rl.rlim_cur); 1296 } 1297 1298 /* 1299 * Return a copy of the entire rlimit structure for the system limit 1300 * specified by 'which' in the rlimit structure pointed to by 'rlp'. 1301 */ 1302 void 1303 lim_rlimit(struct thread *td, int which, struct rlimit *rlp) 1304 { 1305 struct proc *p = td->td_proc; 1306 1307 MPASS(td == curthread); 1308 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1309 ("request for invalid resource limit")); 1310 *rlp = td->td_limit->pl_rlimit[which]; 1311 if (p->p_sysent->sv_fixlimit != NULL) 1312 p->p_sysent->sv_fixlimit(rlp, which); 1313 } 1314 1315 void 1316 lim_rlimit_proc(struct proc *p, int which, struct rlimit *rlp) 1317 { 1318 1319 PROC_LOCK_ASSERT(p, MA_OWNED); 1320 KASSERT(which >= 0 && which < RLIM_NLIMITS, 1321 ("request for invalid resource limit")); 1322 *rlp = p->p_limit->pl_rlimit[which]; 1323 if (p->p_sysent->sv_fixlimit != NULL) 1324 p->p_sysent->sv_fixlimit(rlp, which); 1325 } 1326 1327 void 1328 uihashinit() 1329 { 1330 1331 uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash); 1332 rw_init(&uihashtbl_lock, "uidinfo hash"); 1333 } 1334 1335 /* 1336 * Look up a uidinfo struct for the parameter uid. 1337 * uihashtbl_lock must be locked. 1338 * Increase refcount on uidinfo struct returned. 1339 */ 1340 static struct uidinfo * 1341 uilookup(uid_t uid) 1342 { 1343 struct uihashhead *uipp; 1344 struct uidinfo *uip; 1345 1346 rw_assert(&uihashtbl_lock, RA_LOCKED); 1347 uipp = UIHASH(uid); 1348 LIST_FOREACH(uip, uipp, ui_hash) 1349 if (uip->ui_uid == uid) { 1350 uihold(uip); 1351 break; 1352 } 1353 1354 return (uip); 1355 } 1356 1357 /* 1358 * Find or allocate a struct uidinfo for a particular uid. 1359 * Returns with uidinfo struct referenced. 1360 * uifree() should be called on a struct uidinfo when released. 1361 */ 1362 struct uidinfo * 1363 uifind(uid_t uid) 1364 { 1365 struct uidinfo *new_uip, *uip; 1366 struct ucred *cred; 1367 1368 cred = curthread->td_ucred; 1369 if (cred->cr_uidinfo->ui_uid == uid) { 1370 uip = cred->cr_uidinfo; 1371 uihold(uip); 1372 return (uip); 1373 } else if (cred->cr_ruidinfo->ui_uid == uid) { 1374 uip = cred->cr_ruidinfo; 1375 uihold(uip); 1376 return (uip); 1377 } 1378 1379 rw_rlock(&uihashtbl_lock); 1380 uip = uilookup(uid); 1381 rw_runlock(&uihashtbl_lock); 1382 if (uip != NULL) 1383 return (uip); 1384 1385 new_uip = malloc(sizeof(*new_uip), M_UIDINFO, M_WAITOK | M_ZERO); 1386 racct_create(&new_uip->ui_racct); 1387 refcount_init(&new_uip->ui_ref, 1); 1388 new_uip->ui_uid = uid; 1389 1390 rw_wlock(&uihashtbl_lock); 1391 /* 1392 * There's a chance someone created our uidinfo while we 1393 * were in malloc and not holding the lock, so we have to 1394 * make sure we don't insert a duplicate uidinfo. 1395 */ 1396 if ((uip = uilookup(uid)) == NULL) { 1397 LIST_INSERT_HEAD(UIHASH(uid), new_uip, ui_hash); 1398 rw_wunlock(&uihashtbl_lock); 1399 uip = new_uip; 1400 } else { 1401 rw_wunlock(&uihashtbl_lock); 1402 racct_destroy(&new_uip->ui_racct); 1403 free(new_uip, M_UIDINFO); 1404 } 1405 return (uip); 1406 } 1407 1408 /* 1409 * Place another refcount on a uidinfo struct. 1410 */ 1411 void 1412 uihold(struct uidinfo *uip) 1413 { 1414 1415 refcount_acquire(&uip->ui_ref); 1416 } 1417 1418 /*- 1419 * Since uidinfo structs have a long lifetime, we use an 1420 * opportunistic refcounting scheme to avoid locking the lookup hash 1421 * for each release. 1422 * 1423 * If the refcount hits 0, we need to free the structure, 1424 * which means we need to lock the hash. 1425 * Optimal case: 1426 * After locking the struct and lowering the refcount, if we find 1427 * that we don't need to free, simply unlock and return. 1428 * Suboptimal case: 1429 * If refcount lowering results in need to free, bump the count 1430 * back up, lose the lock and acquire the locks in the proper 1431 * order to try again. 1432 */ 1433 void 1434 uifree(struct uidinfo *uip) 1435 { 1436 1437 if (refcount_release_if_not_last(&uip->ui_ref)) 1438 return; 1439 1440 rw_wlock(&uihashtbl_lock); 1441 if (refcount_release(&uip->ui_ref) == 0) { 1442 rw_wunlock(&uihashtbl_lock); 1443 return; 1444 } 1445 1446 racct_destroy(&uip->ui_racct); 1447 LIST_REMOVE(uip, ui_hash); 1448 rw_wunlock(&uihashtbl_lock); 1449 1450 if (uip->ui_sbsize != 0) 1451 printf("freeing uidinfo: uid = %d, sbsize = %ld\n", 1452 uip->ui_uid, uip->ui_sbsize); 1453 if (uip->ui_proccnt != 0) 1454 printf("freeing uidinfo: uid = %d, proccnt = %ld\n", 1455 uip->ui_uid, uip->ui_proccnt); 1456 if (uip->ui_vmsize != 0) 1457 printf("freeing uidinfo: uid = %d, swapuse = %lld\n", 1458 uip->ui_uid, (unsigned long long)uip->ui_vmsize); 1459 free(uip, M_UIDINFO); 1460 } 1461 1462 #ifdef RACCT 1463 void 1464 ui_racct_foreach(void (*callback)(struct racct *racct, 1465 void *arg2, void *arg3), void (*pre)(void), void (*post)(void), 1466 void *arg2, void *arg3) 1467 { 1468 struct uidinfo *uip; 1469 struct uihashhead *uih; 1470 1471 rw_rlock(&uihashtbl_lock); 1472 if (pre != NULL) 1473 (pre)(); 1474 for (uih = &uihashtbl[uihash]; uih >= uihashtbl; uih--) { 1475 LIST_FOREACH(uip, uih, ui_hash) { 1476 (callback)(uip->ui_racct, arg2, arg3); 1477 } 1478 } 1479 if (post != NULL) 1480 (post)(); 1481 rw_runlock(&uihashtbl_lock); 1482 } 1483 #endif 1484 1485 static inline int 1486 chglimit(struct uidinfo *uip, long *limit, int diff, rlim_t max, const char *name) 1487 { 1488 long new; 1489 1490 /* Don't allow them to exceed max, but allow subtraction. */ 1491 new = atomic_fetchadd_long(limit, (long)diff) + diff; 1492 if (diff > 0 && max != 0) { 1493 if (new < 0 || new > max) { 1494 atomic_subtract_long(limit, (long)diff); 1495 return (0); 1496 } 1497 } else if (new < 0) 1498 printf("negative %s for uid = %d\n", name, uip->ui_uid); 1499 return (1); 1500 } 1501 1502 /* 1503 * Change the count associated with number of processes 1504 * a given user is using. When 'max' is 0, don't enforce a limit 1505 */ 1506 int 1507 chgproccnt(struct uidinfo *uip, int diff, rlim_t max) 1508 { 1509 1510 return (chglimit(uip, &uip->ui_proccnt, diff, max, "proccnt")); 1511 } 1512 1513 /* 1514 * Change the total socket buffer size a user has used. 1515 */ 1516 int 1517 chgsbsize(struct uidinfo *uip, u_int *hiwat, u_int to, rlim_t max) 1518 { 1519 int diff, rv; 1520 1521 diff = to - *hiwat; 1522 if (diff > 0 && max == 0) { 1523 rv = 0; 1524 } else { 1525 rv = chglimit(uip, &uip->ui_sbsize, diff, max, "sbsize"); 1526 if (rv != 0) 1527 *hiwat = to; 1528 } 1529 return (rv); 1530 } 1531 1532 /* 1533 * Change the count associated with number of pseudo-terminals 1534 * a given user is using. When 'max' is 0, don't enforce a limit 1535 */ 1536 int 1537 chgptscnt(struct uidinfo *uip, int diff, rlim_t max) 1538 { 1539 1540 return (chglimit(uip, &uip->ui_ptscnt, diff, max, "ptscnt")); 1541 } 1542 1543 int 1544 chgkqcnt(struct uidinfo *uip, int diff, rlim_t max) 1545 { 1546 1547 return (chglimit(uip, &uip->ui_kqcnt, diff, max, "kqcnt")); 1548 } 1549 1550 int 1551 chgumtxcnt(struct uidinfo *uip, int diff, rlim_t max) 1552 { 1553 1554 return (chglimit(uip, &uip->ui_umtxcnt, diff, max, "umtxcnt")); 1555 } 1556