sys/kern/kern_resource.c

   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/resourcevar.h>
  54 #include <sys/rwlock.h>
  55 #include <sys/sched.h>
  56 #include <sys/sx.h>
  57 #include <sys/syscallsubr.h>
  58 #include <sys/sysent.h>
  59 #include <sys/time.h>
  60 #include <sys/umtx.h>
  61
  62 #include <vm/vm.h>
  63 #include <vm/vm_param.h>
  64 #include <vm/pmap.h>
  65 #include <vm/vm_map.h>
  66
  67
  68 static MALLOC_DEFINE(M_PLIMIT, "plimit", "plimit structures");
  69 static MALLOC_DEFINE(M_UIDINFO, "uidinfo", "uidinfo structures");
  70 #define UIHASH(uid)     (&uihashtbl[(uid) & uihash])
  71 static struct rwlock uihashtbl_lock;
  72 static LIST_HEAD(uihashhead, uidinfo) *uihashtbl;
  73 static u_long uihash;           /* size of hash table - 1 */
  74
  75 static void     calcru1(struct proc *p, struct rusage_ext *ruxp,
  76                     struct timeval *up, struct timeval *sp);
  77 static int      donice(struct thread *td, struct proc *chgp, int n);
  78 static struct uidinfo *uilookup(uid_t uid);
  79
  80 /*
  81  * Resource controls and accounting.
  82  */
  83 #ifndef _SYS_SYSPROTO_H_
  84 struct getpriority_args {
  85         int     which;
  86         int     who;
  87 };
  88 #endif
  89 int
  90 getpriority(td, uap)
  91         struct thread *td;
  92         register struct getpriority_args *uap;
  93 {
  94         struct proc *p;
  95         struct pgrp *pg;
  96         int error, low;
  97
  98         error = 0;
  99         low = PRIO_MAX + 1;
 100         switch (uap->which) {
 101
 102         case PRIO_PROCESS:
 103                 if (uap->who == 0)
 104                         low = td->td_proc->p_nice;
 105                 else {
 106                         p = pfind(uap->who);
 107                         if (p == NULL)
 108                                 break;
 109                         if (p_cansee(td, p) == 0)
 110                                 low = p->p_nice;
 111                         PROC_UNLOCK(p);
 112                 }
 113                 break;
 114
 115         case PRIO_PGRP:
 116                 sx_slock(&proctree_lock);
 117                 if (uap->who == 0) {
 118                         pg = td->td_proc->p_pgrp;
 119                         PGRP_LOCK(pg);
 120                 } else {
 121                         pg = pgfind(uap->who);
 122                         if (pg == NULL) {
 123                                 sx_sunlock(&proctree_lock);
 124                                 break;
 125                         }
 126                 }
 127                 sx_sunlock(&proctree_lock);
 128                 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
 129                         PROC_LOCK(p);
 130                         if (p_cansee(td, p) == 0) {
 131                                 if (p->p_nice < low)
 132                                         low = p->p_nice;
 133                         }
 134                         PROC_UNLOCK(p);
 135                 }
 136                 PGRP_UNLOCK(pg);
 137                 break;
 138
 139         case PRIO_USER:
 140                 if (uap->who == 0)
 141                         uap->who = td->td_ucred->cr_uid;
 142                 sx_slock(&allproc_lock);
 143                 FOREACH_PROC_IN_SYSTEM(p) {
 144                         /* Do not bother to check PRS_NEW processes */
 145                         if (p->p_state == PRS_NEW)
 146                                 continue;
 147                         PROC_LOCK(p);
 148                         if (p_cansee(td, p) == 0 &&
 149                             p->p_ucred->cr_uid == uap->who) {
 150                                 if (p->p_nice < low)
 151                                         low = p->p_nice;
 152                         }
 153                         PROC_UNLOCK(p);
 154                 }
 155                 sx_sunlock(&allproc_lock);
 156                 break;
 157
 158         default:
 159                 error = EINVAL;
 160                 break;
 161         }
 162         if (low == PRIO_MAX + 1 && error == 0)
 163                 error = ESRCH;
 164         td->td_retval[0] = low;
 165         return (error);
 166 }
 167
 168 #ifndef _SYS_SYSPROTO_H_
 169 struct setpriority_args {
 170         int     which;
 171         int     who;
 172         int     prio;
 173 };
 174 #endif
 175 int
 176 setpriority(td, uap)
 177         struct thread *td;
 178         struct setpriority_args *uap;
 179 {
 180         struct proc *curp, *p;
 181         struct pgrp *pg;
 182         int found = 0, error = 0;
 183
 184         curp = td->td_proc;
 185         switch (uap->which) {
 186         case PRIO_PROCESS:
 187                 if (uap->who == 0) {
 188                         PROC_LOCK(curp);
 189                         error = donice(td, curp, uap->prio);
 190                         PROC_UNLOCK(curp);
 191                 } else {
 192                         p = pfind(uap->who);
 193                         if (p == NULL)
 194                                 break;
 195                         error = p_cansee(td, p);
 196                         if (error == 0)
 197                                 error = donice(td, p, uap->prio);
 198                         PROC_UNLOCK(p);
 199                 }
 200                 found++;
 201                 break;
 202
 203         case PRIO_PGRP:
 204                 sx_slock(&proctree_lock);
 205                 if (uap->who == 0) {
 206                         pg = curp->p_pgrp;
 207                         PGRP_LOCK(pg);
 208                 } else {
 209                         pg = pgfind(uap->who);
 210                         if (pg == NULL) {
 211                                 sx_sunlock(&proctree_lock);
 212                                 break;
 213                         }
 214                 }
 215                 sx_sunlock(&proctree_lock);
 216                 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
 217                         PROC_LOCK(p);
 218                         if (p_cansee(td, p) == 0) {
 219                                 error = donice(td, p, uap->prio);
 220                                 found++;
 221                         }
 222                         PROC_UNLOCK(p);
 223                 }
 224                 PGRP_UNLOCK(pg);
 225                 break;
 226
 227         case PRIO_USER:
 228                 if (uap->who == 0)
 229                         uap->who = td->td_ucred->cr_uid;
 230                 sx_slock(&allproc_lock);
 231                 FOREACH_PROC_IN_SYSTEM(p) {
 232                         PROC_LOCK(p);
 233                         if (p->p_ucred->cr_uid == uap->who &&
 234                             p_cansee(td, p) == 0) {
 235                                 error = donice(td, p, uap->prio);
 236                                 found++;
 237                         }
 238                         PROC_UNLOCK(p);
 239                 }
 240                 sx_sunlock(&allproc_lock);
 241                 break;
 242
 243         default:
 244                 error = EINVAL;
 245                 break;
 246         }
 247         if (found == 0 && error == 0)
 248                 error = ESRCH;
 249         return (error);
 250 }
 251
 252 /*
 253  * Set "nice" for a (whole) process.
 254  */
 255 static int
 256 donice(struct thread *td, struct proc *p, int n)
 257 {
 258         int error;
 259
 260         PROC_LOCK_ASSERT(p, MA_OWNED);
 261         if ((error = p_cansched(td, p)))
 262                 return (error);
 263         if (n > PRIO_MAX)
 264                 n = PRIO_MAX;
 265         if (n < PRIO_MIN)
 266                 n = PRIO_MIN;
 267         if (n < p->p_nice && priv_check(td, PRIV_SCHED_SETPRIORITY) != 0)
 268                 return (EACCES);
 269         sched_nice(p, n);
 270         return (0);
 271 }
 272
 273 /*
 274  * Set realtime priority for LWP.
 275  */
 276 #ifndef _SYS_SYSPROTO_H_
 277 struct rtprio_thread_args {
 278         int             function;
 279         lwpid_t         lwpid;
 280         struct rtprio   *rtp;
 281 };
 282 #endif
 283 int
 284 rtprio_thread(struct thread *td, struct rtprio_thread_args *uap)
 285 {
 286         struct proc *p;
 287         struct rtprio rtp;
 288         struct thread *td1;
 289         int cierror, error;
 290
 291         /* Perform copyin before acquiring locks if needed. */
 292         if (uap->function == RTP_SET)
 293                 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
 294         else
 295                 cierror = 0;
 296
 297         /*
 298          * Though lwpid is unique, only current process is supported
 299          * since there is no efficient way to look up a LWP yet.
 300          */
 301         p = td->td_proc;
 302         PROC_LOCK(p);
 303
 304         switch (uap->function) {
 305         case RTP_LOOKUP:
 306                 if ((error = p_cansee(td, p)))
 307                         break;
 308                 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
 309                         td1 = td;
 310                 else
 311                         td1 = thread_find(p, uap->lwpid);
 312                 if (td1 != NULL)
 313                         pri_to_rtp(td1, &rtp);
 314                 else
 315                         error = ESRCH;
 316                 PROC_UNLOCK(p);
 317                 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
 318         case RTP_SET:
 319                 if ((error = p_cansched(td, p)) || (error = cierror))
 320                         break;
 321
 322                 /* Disallow setting rtprio in most cases if not superuser. */
 323 /*
 324  * Realtime priority has to be restricted for reasons which should be
 325  * obvious.  However, for idle priority, there is a potential for
 326  * system deadlock if an idleprio process gains a lock on a resource
 327  * that other processes need (and the idleprio process can't run
 328  * due to a CPU-bound normal process).  Fix me!  XXX
 329  */
 330 #if 0
 331                 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
 332 #else
 333                 if (rtp.type != RTP_PRIO_NORMAL) {
 334 #endif
 335                         error = priv_check(td, PRIV_SCHED_RTPRIO);
 336                         if (error)
 337                                 break;
 338                 }
 339
 340                 if (uap->lwpid == 0 || uap->lwpid == td->td_tid)
 341                         td1 = td;
 342                 else
 343                         td1 = thread_find(p, uap->lwpid);
 344                 if (td1 != NULL)
 345                         error = rtp_to_pri(&rtp, td1);
 346                 else
 347                         error = ESRCH;
 348                 break;
 349         default:
 350                 error = EINVAL;
 351                 break;
 352         }
 353         PROC_UNLOCK(p);
 354         return (error);
 355 }
 356
 357 /*
 358  * Set realtime priority.
 359  */
 360 #ifndef _SYS_SYSPROTO_H_
 361 struct rtprio_args {
 362         int             function;
 363         pid_t           pid;
 364         struct rtprio   *rtp;
 365 };
 366 #endif
 367 int
 368 rtprio(td, uap)
 369         struct thread *td;              /* curthread */
 370         register struct rtprio_args *uap;
 371 {
 372         struct proc *p;
 373         struct thread *tdp;
 374         struct rtprio rtp;
 375         int cierror, error;
 376
 377         /* Perform copyin before acquiring locks if needed. */
 378         if (uap->function == RTP_SET)
 379                 cierror = copyin(uap->rtp, &rtp, sizeof(struct rtprio));
 380         else
 381                 cierror = 0;
 382
 383         if (uap->pid == 0) {
 384                 p = td->td_proc;
 385                 PROC_LOCK(p);
 386         } else {
 387                 p = pfind(uap->pid);
 388                 if (p == NULL)
 389                         return (ESRCH);
 390         }
 391
 392         switch (uap->function) {
 393         case RTP_LOOKUP:
 394                 if ((error = p_cansee(td, p)))
 395                         break;
 396                 /*
 397                  * Return OUR priority if no pid specified,
 398                  * or if one is, report the highest priority
 399                  * in the process.  There isn't much more you can do as
 400                  * there is only room to return a single priority.
 401                  * Note: specifying our own pid is not the same
 402                  * as leaving it zero.
 403                  */
 404                 if (uap->pid == 0) {
 405                         pri_to_rtp(td, &rtp);
 406                 } else {
 407                         struct rtprio rtp2;
 408
 409                         rtp.type = RTP_PRIO_IDLE;
 410                         rtp.prio = RTP_PRIO_MAX;
 411                         FOREACH_THREAD_IN_PROC(p, tdp) {
 412                                 pri_to_rtp(tdp, &rtp2);
 413                                 if (rtp2.type <  rtp.type ||
 414                                     (rtp2.type == rtp.type &&
 415                                     rtp2.prio < rtp.prio)) {
 416                                         rtp.type = rtp2.type;
 417                                         rtp.prio = rtp2.prio;
 418                                 }
 419                         }
 420                 }
 421                 PROC_UNLOCK(p);
 422                 return (copyout(&rtp, uap->rtp, sizeof(struct rtprio)));
 423         case RTP_SET:
 424                 if ((error = p_cansched(td, p)) || (error = cierror))
 425                         break;
 426
 427                 /* Disallow setting rtprio in most cases if not superuser. */
 428 /*
 429  * Realtime priority has to be restricted for reasons which should be
 430  * obvious.  However, for idle priority, there is a potential for
 431  * system deadlock if an idleprio process gains a lock on a resource
 432  * that other processes need (and the idleprio process can't run
 433  * due to a CPU-bound normal process).  Fix me!  XXX
 434  */
 435 #if 0
 436                 if (RTP_PRIO_IS_REALTIME(rtp.type)) {
 437 #else
 438                 if (rtp.type != RTP_PRIO_NORMAL) {
 439 #endif
 440                         error = priv_check(td, PRIV_SCHED_RTPRIO);
 441                         if (error)
 442                                 break;
 443                 }
 444
 445                 /*
 446                  * If we are setting our own priority, set just our
 447                  * thread but if we are doing another process,
 448                  * do all the threads on that process. If we
 449                  * specify our own pid we do the latter.
 450                  */
 451                 if (uap->pid == 0) {
 452                         error = rtp_to_pri(&rtp, td);
 453                 } else {
 454                         FOREACH_THREAD_IN_PROC(p, td) {
 455                                 if ((error = rtp_to_pri(&rtp, td)) != 0)
 456                                         break;
 457                         }
 458                 }
 459                 break;
 460         default:
 461                 error = EINVAL;
 462                 break;
 463         }
 464         PROC_UNLOCK(p);
 465         return (error);
 466 }
 467
 468 int
 469 rtp_to_pri(struct rtprio *rtp, struct thread *td)
 470 {
 471         u_char  newpri;
 472         u_char  oldpri;
 473
 474         if (rtp->prio > RTP_PRIO_MAX)
 475                 return (EINVAL);
 476         thread_lock(td);
 477         switch (RTP_PRIO_BASE(rtp->type)) {
 478         case RTP_PRIO_REALTIME:
 479                 newpri = PRI_MIN_REALTIME + rtp->prio;
 480                 break;
 481         case RTP_PRIO_NORMAL:
 482                 newpri = PRI_MIN_TIMESHARE + rtp->prio;
 483                 break;
 484         case RTP_PRIO_IDLE:
 485                 newpri = PRI_MIN_IDLE + rtp->prio;
 486                 break;
 487         default:
 488                 thread_unlock(td);
 489                 return (EINVAL);
 490         }
 491         sched_class(td, rtp->type);     /* XXX fix */
 492         oldpri = td->td_user_pri;
 493         sched_user_prio(td, newpri);
 494         if (curthread == td)
 495                 sched_prio(curthread, td->td_user_pri); /* XXX dubious */
 496         if (TD_ON_UPILOCK(td) && oldpri != newpri) {
 497                 thread_unlock(td);
 498                 umtx_pi_adjust(td, oldpri);
 499         } else
 500                 thread_unlock(td);
 501         return (0);
 502 }
 503
 504 void
 505 pri_to_rtp(struct thread *td, struct rtprio *rtp)
 506 {
 507
 508         thread_lock(td);
 509         switch (PRI_BASE(td->td_pri_class)) {
 510         case PRI_REALTIME:
 511                 rtp->prio = td->td_base_user_pri - PRI_MIN_REALTIME;
 512                 break;
 513         case PRI_TIMESHARE:
 514                 rtp->prio = td->td_base_user_pri - PRI_MIN_TIMESHARE;
 515                 break;
 516         case PRI_IDLE:
 517                 rtp->prio = td->td_base_user_pri - PRI_MIN_IDLE;
 518                 break;
 519         default:
 520                 break;
 521         }
 522         rtp->type = td->td_pri_class;
 523         thread_unlock(td);
 524 }
 525
 526 #if defined(COMPAT_43)
 527 #ifndef _SYS_SYSPROTO_H_
 528 struct osetrlimit_args {
 529         u_int   which;
 530         struct  orlimit *rlp;
 531 };
 532 #endif
 533 int
 534 osetrlimit(td, uap)
 535         struct thread *td;
 536         register struct osetrlimit_args *uap;
 537 {
 538         struct orlimit olim;
 539         struct rlimit lim;
 540         int error;
 541
 542         if ((error = copyin(uap->rlp, &olim, sizeof(struct orlimit))))
 543                 return (error);
 544         lim.rlim_cur = olim.rlim_cur;
 545         lim.rlim_max = olim.rlim_max;
 546         error = kern_setrlimit(td, uap->which, &lim);
 547         return (error);
 548 }
 549
 550 #ifndef _SYS_SYSPROTO_H_
 551 struct ogetrlimit_args {
 552         u_int   which;
 553         struct  orlimit *rlp;
 554 };
 555 #endif
 556 int
 557 ogetrlimit(td, uap)
 558         struct thread *td;
 559         register struct ogetrlimit_args *uap;
 560 {
 561         struct orlimit olim;
 562         struct rlimit rl;
 563         struct proc *p;
 564         int error;
 565
 566         if (uap->which >= RLIM_NLIMITS)
 567                 return (EINVAL);
 568         p = td->td_proc;
 569         PROC_LOCK(p);
 570         lim_rlimit(p, uap->which, &rl);
 571         PROC_UNLOCK(p);
 572
 573         /*
 574          * XXX would be more correct to convert only RLIM_INFINITY to the
 575          * old RLIM_INFINITY and fail with EOVERFLOW for other larger
 576          * values.  Most 64->32 and 32->16 conversions, including not
 577          * unimportant ones of uids are even more broken than what we
 578          * do here (they blindly truncate).  We don't do this correctly
 579          * here since we have little experience with EOVERFLOW yet.
 580          * Elsewhere, getuid() can't fail...
 581          */
 582         olim.rlim_cur = rl.rlim_cur > 0x7fffffff ? 0x7fffffff : rl.rlim_cur;
 583         olim.rlim_max = rl.rlim_max > 0x7fffffff ? 0x7fffffff : rl.rlim_max;
 584         error = copyout(&olim, uap->rlp, sizeof(olim));
 585         return (error);
 586 }
 587 #endif /* COMPAT_43 */
 588
 589 #ifndef _SYS_SYSPROTO_H_
 590 struct __setrlimit_args {
 591         u_int   which;
 592         struct  rlimit *rlp;
 593 };
 594 #endif
 595 int
 596 setrlimit(td, uap)
 597         struct thread *td;
 598         register struct __setrlimit_args *uap;
 599 {
 600         struct rlimit alim;
 601         int error;
 602
 603         if ((error = copyin(uap->rlp, &alim, sizeof(struct rlimit))))
 604                 return (error);
 605         error = kern_setrlimit(td, uap->which, &alim);
 606         return (error);
 607 }
 608
 609 static void
 610 lim_cb(void *arg)
 611 {
 612         struct rlimit rlim;
 613         struct thread *td;
 614         struct proc *p;
 615
 616         p = arg;
 617         PROC_LOCK_ASSERT(p, MA_OWNED);
 618         /*
 619          * Check if the process exceeds its cpu resource allocation.  If
 620          * it reaches the max, arrange to kill the process in ast().
 621          */
 622         if (p->p_cpulimit == RLIM_INFINITY)
 623                 return;
 624         PROC_SLOCK(p);
 625         FOREACH_THREAD_IN_PROC(p, td) {
 626                 thread_lock(td);
 627                 ruxagg(&p->p_rux, td);
 628                 thread_unlock(td);
 629         }
 630         PROC_SUNLOCK(p);
 631         if (p->p_rux.rux_runtime > p->p_cpulimit * cpu_tickrate()) {
 632                 lim_rlimit(p, RLIMIT_CPU, &rlim);
 633                 if (p->p_rux.rux_runtime >= rlim.rlim_max * cpu_tickrate()) {
 634                         killproc(p, "exceeded maximum CPU limit");
 635                 } else {
 636                         if (p->p_cpulimit < rlim.rlim_max)
 637                                 p->p_cpulimit += 5;
 638                         psignal(p, SIGXCPU);
 639                 }
 640         }
 641         callout_reset(&p->p_limco, hz, lim_cb, p);
 642 }
 643
 644 int
 645 kern_setrlimit(td, which, limp)
 646         struct thread *td;
 647         u_int which;
 648         struct rlimit *limp;
 649 {
 650         struct plimit *newlim, *oldlim;
 651         struct proc *p;
 652         register struct rlimit *alimp;
 653         struct rlimit oldssiz;
 654         int error;
 655
 656         if (which >= RLIM_NLIMITS)
 657                 return (EINVAL);
 658
 659         /*
 660          * Preserve historical bugs by treating negative limits as unsigned.
 661          */
 662         if (limp->rlim_cur < 0)
 663                 limp->rlim_cur = RLIM_INFINITY;
 664         if (limp->rlim_max < 0)
 665                 limp->rlim_max = RLIM_INFINITY;
 666
 667         oldssiz.rlim_cur = 0;
 668         p = td->td_proc;
 669         newlim = lim_alloc();
 670         PROC_LOCK(p);
 671         oldlim = p->p_limit;
 672         alimp = &oldlim->pl_rlimit[which];
 673         if (limp->rlim_cur > alimp->rlim_max ||
 674             limp->rlim_max > alimp->rlim_max)
 675                 if ((error = priv_check(td, PRIV_PROC_SETRLIMIT))) {
 676                         PROC_UNLOCK(p);
 677                         lim_free(newlim);
 678                         return (error);
 679                 }
 680         if (limp->rlim_cur > limp->rlim_max)
 681                 limp->rlim_cur = limp->rlim_max;
 682         lim_copy(newlim, oldlim);
 683         alimp = &newlim->pl_rlimit[which];
 684
 685         switch (which) {
 686
 687         case RLIMIT_CPU:
 688                 if (limp->rlim_cur != RLIM_INFINITY &&
 689                     p->p_cpulimit == RLIM_INFINITY)
 690                         callout_reset(&p->p_limco, hz, lim_cb, p);
 691                 p->p_cpulimit = limp->rlim_cur;
 692                 break;
 693         case RLIMIT_DATA:
 694                 if (limp->rlim_cur > maxdsiz)
 695                         limp->rlim_cur = maxdsiz;
 696                 if (limp->rlim_max > maxdsiz)
 697                         limp->rlim_max = maxdsiz;
 698                 break;
 699
 700         case RLIMIT_STACK:
 701                 if (limp->rlim_cur > maxssiz)
 702                         limp->rlim_cur = maxssiz;
 703                 if (limp->rlim_max > maxssiz)
 704                         limp->rlim_max = maxssiz;
 705                 oldssiz = *alimp;
 706                 if (td->td_proc->p_sysent->sv_fixlimit != NULL)
 707                         td->td_proc->p_sysent->sv_fixlimit(&oldssiz,
 708                             RLIMIT_STACK);
 709                 break;
 710
 711         case RLIMIT_NOFILE:
 712                 if (limp->rlim_cur > maxfilesperproc)
 713                         limp->rlim_cur = maxfilesperproc;
 714                 if (limp->rlim_max > maxfilesperproc)
 715                         limp->rlim_max = maxfilesperproc;
 716                 break;
 717
 718         case RLIMIT_NPROC:
 719                 if (limp->rlim_cur > maxprocperuid)
 720                         limp->rlim_cur = maxprocperuid;
 721                 if (limp->rlim_max > maxprocperuid)
 722                         limp->rlim_max = maxprocperuid;
 723                 if (limp->rlim_cur < 1)
 724                         limp->rlim_cur = 1;
 725                 if (limp->rlim_max < 1)
 726                         limp->rlim_max = 1;
 727                 break;
 728         }
 729         if (td->td_proc->p_sysent->sv_fixlimit != NULL)
 730                 td->td_proc->p_sysent->sv_fixlimit(limp, which);
 731         *alimp = *limp;
 732         p->p_limit = newlim;
 733         PROC_UNLOCK(p);
 734         lim_free(oldlim);
 735
 736         if (which == RLIMIT_STACK) {
 737                 /*
 738                  * Stack is allocated to the max at exec time with only
 739                  * "rlim_cur" bytes accessible.  If stack limit is going
 740                  * up make more accessible, if going down make inaccessible.
 741                  */
 742                 if (limp->rlim_cur != oldssiz.rlim_cur) {
 743                         vm_offset_t addr;
 744                         vm_size_t size;
 745                         vm_prot_t prot;
 746
 747                         if (limp->rlim_cur > oldssiz.rlim_cur) {
 748                                 prot = p->p_sysent->sv_stackprot;
 749                                 size = limp->rlim_cur - oldssiz.rlim_cur;
 750                                 addr = p->p_sysent->sv_usrstack -
 751                                     limp->rlim_cur;
 752                         } else {
 753                                 prot = VM_PROT_NONE;
 754                                 size = oldssiz.rlim_cur - limp->rlim_cur;
 755                                 addr = p->p_sysent->sv_usrstack -
 756                                     oldssiz.rlim_cur;
 757                         }
 758                         addr = trunc_page(addr);
 759                         size = round_page(size);
 760                         (void)vm_map_protect(&p->p_vmspace->vm_map,
 761                             addr, addr + size, prot, FALSE);
 762                 }
 763         }
 764
 765         return (0);
 766 }
 767
 768 #ifndef _SYS_SYSPROTO_H_
 769 struct __getrlimit_args {
 770         u_int   which;
 771         struct  rlimit *rlp;
 772 };
 773 #endif
 774 /* ARGSUSED */
 775 int
 776 getrlimit(td, uap)
 777         struct thread *td;
 778         register struct __getrlimit_args *uap;
 779 {
 780         struct rlimit rlim;
 781         struct proc *p;
 782         int error;
 783
 784         if (uap->which >= RLIM_NLIMITS)
 785                 return (EINVAL);
 786         p = td->td_proc;
 787         PROC_LOCK(p);
 788         lim_rlimit(p, uap->which, &rlim);
 789         PROC_UNLOCK(p);
 790         error = copyout(&rlim, uap->rlp, sizeof(struct rlimit));
 791         return (error);
 792 }
 793
 794 /*
 795  * Transform the running time and tick information for children of proc p
 796  * into user and system time usage.
 797  */
 798 void
 799 calccru(p, up, sp)
 800         struct proc *p;
 801         struct timeval *up;
 802         struct timeval *sp;
 803 {
 804
 805         PROC_LOCK_ASSERT(p, MA_OWNED);
 806         calcru1(p, &p->p_crux, up, sp);
 807 }
 808
 809 /*
 810  * Transform the running time and tick information in proc p into user
 811  * and system time usage.  If appropriate, include the current time slice
 812  * on this CPU.
 813  */
 814 void
 815 calcru(struct proc *p, struct timeval *up, struct timeval *sp)
 816 {
 817         struct thread *td;
 818         uint64_t u;
 819
 820         PROC_LOCK_ASSERT(p, MA_OWNED);
 821         PROC_SLOCK_ASSERT(p, MA_OWNED);
 822         /*
 823          * If we are getting stats for the current process, then add in the
 824          * stats that this thread has accumulated in its current time slice.
 825          * We reset the thread and CPU state as if we had performed a context
 826          * switch right here.
 827          */
 828         td = curthread;
 829         if (td->td_proc == p) {
 830                 u = cpu_ticks();
 831                 p->p_rux.rux_runtime += u - PCPU_GET(switchtime);
 832                 PCPU_SET(switchtime, u);
 833         }
 834         /* Make sure the per-thread stats are current. */
 835         FOREACH_THREAD_IN_PROC(p, td) {
 836                 if (td->td_incruntime == 0)
 837                         continue;
 838                 thread_lock(td);
 839                 ruxagg(&p->p_rux, td);
 840                 thread_unlock(td);
 841         }
 842         calcru1(p, &p->p_rux, up, sp);
 843 }
 844
 845 static void
 846 calcru1(struct proc *p, struct rusage_ext *ruxp, struct timeval *up,
 847     struct timeval *sp)
 848 {
 849         /* {user, system, interrupt, total} {ticks, usec}: */
 850         u_int64_t ut, uu, st, su, it, tt, tu;
 851
 852         ut = ruxp->rux_uticks;
 853         st = ruxp->rux_sticks;
 854         it = ruxp->rux_iticks;
 855         tt = ut + st + it;
 856         if (tt == 0) {
 857                 /* Avoid divide by zero */
 858                 st = 1;
 859                 tt = 1;
 860         }
 861         tu = cputick2usec(ruxp->rux_runtime);
 862         if ((int64_t)tu < 0) {
 863                 /* XXX: this should be an assert /phk */
 864                 printf("calcru: negative runtime of %jd usec for pid %d (%s)\n",
 865                     (intmax_t)tu, p->p_pid, p->p_comm);
 866                 tu = ruxp->rux_tu;
 867         }
 868
 869         if (tu >= ruxp->rux_tu) {
 870                 /*
 871                  * The normal case, time increased.
 872                  * Enforce monotonicity of bucketed numbers.
 873                  */
 874                 uu = (tu * ut) / tt;
 875                 if (uu < ruxp->rux_uu)
 876                         uu = ruxp->rux_uu;
 877                 su = (tu * st) / tt;
 878                 if (su < ruxp->rux_su)
 879                         su = ruxp->rux_su;
 880         } else if (tu + 3 > ruxp->rux_tu || 101 * tu > 100 * ruxp->rux_tu) {
 881                 /*
 882                  * When we calibrate the cputicker, it is not uncommon to
 883                  * see the presumably fixed frequency increase slightly over
 884                  * time as a result of thermal stabilization and NTP
 885                  * discipline (of the reference clock).  We therefore ignore
 886                  * a bit of backwards slop because we  expect to catch up
 887                  * shortly.  We use a 3 microsecond limit to catch low
 888                  * counts and a 1% limit for high counts.
 889                  */
 890                 uu = ruxp->rux_uu;
 891                 su = ruxp->rux_su;
 892                 tu = ruxp->rux_tu;
 893         } else { /* tu < ruxp->rux_tu */
 894                 /*
 895                  * What happened here was likely that a laptop, which ran at
 896                  * a reduced clock frequency at boot, kicked into high gear.
 897                  * The wisdom of spamming this message in that case is
 898                  * dubious, but it might also be indicative of something
 899                  * serious, so lets keep it and hope laptops can be made
 900                  * more truthful about their CPU speed via ACPI.
 901                  */
 902                 printf("calcru: runtime went backwards from %ju usec "
 903                     "to %ju usec for pid %d (%s)\n",
 904                     (uintmax_t)ruxp->rux_tu, (uintmax_t)tu,
 905                     p->p_pid, p->p_comm);
 906                 uu = (tu * ut) / tt;
 907                 su = (tu * st) / tt;
 908         }
 909
 910         ruxp->rux_uu = uu;
 911         ruxp->rux_su = su;
 912         ruxp->rux_tu = tu;
 913
 914         up->tv_sec = uu / 1000000;
 915         up->tv_usec = uu % 1000000;
 916         sp->tv_sec = su / 1000000;
 917         sp->tv_usec = su % 1000000;
 918 }
 919
 920 #ifndef _SYS_SYSPROTO_H_
 921 struct getrusage_args {
 922         int     who;
 923         struct  rusage *rusage;
 924 };
 925 #endif
 926 int
 927 getrusage(td, uap)
 928         register struct thread *td;
 929         register struct getrusage_args *uap;
 930 {
 931         struct rusage ru;
 932         int error;
 933
 934         error = kern_getrusage(td, uap->who, &ru);
 935         if (error == 0)
 936                 error = copyout(&ru, uap->rusage, sizeof(struct rusage));
 937         return (error);
 938 }
 939
 940 int
 941 kern_getrusage(td, who, rup)
 942         struct thread *td;
 943         int who;
 944         struct rusage *rup;
 945 {
 946         struct proc *p;
 947         int error;
 948
 949         error = 0;
 950         p = td->td_proc;
 951         PROC_LOCK(p);
 952         switch (who) {
 953         case RUSAGE_SELF:
 954                 rufetchcalc(p, rup, &rup->ru_utime,
 955                     &rup->ru_stime);
 956                 break;
 957
 958         case RUSAGE_CHILDREN:
 959                 *rup = p->p_stats->p_cru;
 960                 calccru(p, &rup->ru_utime, &rup->ru_stime);
 961                 break;
 962
 963         default:
 964                 error = EINVAL;
 965         }
 966         PROC_UNLOCK(p);
 967         return (error);
 968 }
 969
 970 void
 971 rucollect(struct rusage *ru, struct rusage *ru2)
 972 {
 973         long *ip, *ip2;
 974         int i;
 975
 976         if (ru->ru_maxrss < ru2->ru_maxrss)
 977                 ru->ru_maxrss = ru2->ru_maxrss;
 978         ip = &ru->ru_first;
 979         ip2 = &ru2->ru_first;
 980         for (i = &ru->ru_last - &ru->ru_first; i >= 0; i--)
 981                 *ip++ += *ip2++;
 982 }
 983
 984 void
 985 ruadd(struct rusage *ru, struct rusage_ext *rux, struct rusage *ru2,
 986     struct rusage_ext *rux2)
 987 {
 988
 989         rux->rux_runtime += rux2->rux_runtime;
 990         rux->rux_uticks += rux2->rux_uticks;
 991         rux->rux_sticks += rux2->rux_sticks;
 992         rux->rux_iticks += rux2->rux_iticks;
 993         rux->rux_uu += rux2->rux_uu;
 994         rux->rux_su += rux2->rux_su;
 995         rux->rux_tu += rux2->rux_tu;
 996         rucollect(ru, ru2);
 997 }
 998
 999 /*
1000  * Aggregate tick counts into the proc's rusage_ext.
1001  */
1002 void
1003 ruxagg(struct rusage_ext *rux, struct thread *td)
1004 {
1005
1006         THREAD_LOCK_ASSERT(td, MA_OWNED);
1007         PROC_SLOCK_ASSERT(td->td_proc, MA_OWNED);
1008         rux->rux_runtime += td->td_incruntime;
1009         rux->rux_uticks += td->td_uticks;
1010         rux->rux_sticks += td->td_sticks;
1011         rux->rux_iticks += td->td_iticks;
1012         td->td_incruntime = 0;
1013         td->td_uticks = 0;
1014         td->td_iticks = 0;
1015         td->td_sticks = 0;
1016 }
1017
1018 /*
1019  * Update the rusage_ext structure and fetch a valid aggregate rusage
1020  * for proc p if storage for one is supplied.
1021  */
1022 void
1023 rufetch(struct proc *p, struct rusage *ru)
1024 {
1025         struct thread *td;
1026
1027         PROC_SLOCK_ASSERT(p, MA_OWNED);
1028
1029         *ru = p->p_ru;
1030         if (p->p_numthreads > 0)  {
1031                 FOREACH_THREAD_IN_PROC(p, td) {
1032                         thread_lock(td);
1033                         ruxagg(&p->p_rux, td);
1034                         thread_unlock(td);
1035                         rucollect(ru, &td->td_ru);
1036                 }
1037         }
1038 }
1039
1040 /*
1041  * Atomically perform a rufetch and a calcru together.
1042  * Consumers, can safely assume the calcru is executed only once
1043  * rufetch is completed.
1044  */
1045 void
1046 rufetchcalc(struct proc *p, struct rusage *ru, struct timeval *up,
1047     struct timeval *sp)
1048 {
1049
1050         PROC_SLOCK(p);
1051         rufetch(p, ru);
1052         calcru(p, up, sp);
1053         PROC_SUNLOCK(p);
1054 }
1055
1056 /*
1057  * Allocate a new resource limits structure and initialize its
1058  * reference count and mutex pointer.
1059  */
1060 struct plimit *
1061 lim_alloc()
1062 {
1063         struct plimit *limp;
1064
1065         limp = malloc(sizeof(struct plimit), M_PLIMIT, M_WAITOK);
1066         refcount_init(&limp->pl_refcnt, 1);
1067         return (limp);
1068 }
1069
1070 struct plimit *
1071 lim_hold(limp)
1072         struct plimit *limp;
1073 {
1074
1075         refcount_acquire(&limp->pl_refcnt);
1076         return (limp);
1077 }
1078
1079 void
1080 lim_fork(struct proc *p1, struct proc *p2)
1081 {
1082         p2->p_limit = lim_hold(p1->p_limit);
1083         callout_init_mtx(&p2->p_limco, &p2->p_mtx, 0);
1084         if (p1->p_cpulimit != RLIM_INFINITY)
1085                 callout_reset(&p2->p_limco, hz, lim_cb, p2);
1086 }
1087
1088 void
1089 lim_free(limp)
1090         struct plimit *limp;
1091 {
1092
1093         KASSERT(limp->pl_refcnt > 0, ("plimit refcnt underflow"));
1094         if (refcount_release(&limp->pl_refcnt))
1095                 free((void *)limp, M_PLIMIT);
1096 }
1097
1098 /*
1099  * Make a copy of the plimit structure.
1100  * We share these structures copy-on-write after fork.
1101  */
1102 void
1103 lim_copy(dst, src)
1104         struct plimit *dst, *src;
1105 {
1106
1107         KASSERT(dst->pl_refcnt == 1, ("lim_copy to shared limit"));
1108         bcopy(src->pl_rlimit, dst->pl_rlimit, sizeof(src->pl_rlimit));
1109 }
1110
1111 /*
1112  * Return the hard limit for a particular system resource.  The
1113  * which parameter specifies the index into the rlimit array.
1114  */
1115 rlim_t
1116 lim_max(struct proc *p, int which)
1117 {
1118         struct rlimit rl;
1119
1120         lim_rlimit(p, which, &rl);
1121         return (rl.rlim_max);
1122 }
1123
1124 /*
1125  * Return the current (soft) limit for a particular system resource.
1126  * The which parameter which specifies the index into the rlimit array
1127  */
1128 rlim_t
1129 lim_cur(struct proc *p, int which)
1130 {
1131         struct rlimit rl;
1132
1133         lim_rlimit(p, which, &rl);
1134         return (rl.rlim_cur);
1135 }
1136
1137 /*
1138  * Return a copy of the entire rlimit structure for the system limit
1139  * specified by 'which' in the rlimit structure pointed to by 'rlp'.
1140  */
1141 void
1142 lim_rlimit(struct proc *p, int which, struct rlimit *rlp)
1143 {
1144
1145         PROC_LOCK_ASSERT(p, MA_OWNED);
1146         KASSERT(which >= 0 && which < RLIM_NLIMITS,
1147             ("request for invalid resource limit"));
1148         *rlp = p->p_limit->pl_rlimit[which];
1149         if (p->p_sysent->sv_fixlimit != NULL)
1150                 p->p_sysent->sv_fixlimit(rlp, which);
1151 }
1152
1153 /*
1154  * Find the uidinfo structure for a uid.  This structure is used to
1155  * track the total resource consumption (process count, socket buffer
1156  * size, etc.) for the uid and impose limits.
1157  */
1158 void
1159 uihashinit()
1160 {
1161
1162         uihashtbl = hashinit(maxproc / 16, M_UIDINFO, &uihash);
1163         rw_init(&uihashtbl_lock, "uidinfo hash");
1164 }
1165
1166 /*
1167  * Look up a uidinfo struct for the parameter uid.
1168  * uihashtbl_lock must be locked.
1169  */
1170 static struct uidinfo *
1171 uilookup(uid)
1172         uid_t uid;
1173 {
1174         struct uihashhead *uipp;
1175         struct uidinfo *uip;
1176
1177         rw_assert(&uihashtbl_lock, RA_LOCKED);
1178         uipp = UIHASH(uid);
1179         LIST_FOREACH(uip, uipp, ui_hash)
1180                 if (uip->ui_uid == uid)
1181                         break;
1182
1183         return (uip);
1184 }
1185
1186 /*
1187  * Find or allocate a struct uidinfo for a particular uid.
1188  * Increase refcount on uidinfo struct returned.
1189  * uifree() should be called on a struct uidinfo when released.
1190  */
1191 struct uidinfo *
1192 uifind(uid)
1193         uid_t uid;
1194 {
1195         struct uidinfo *old_uip, *uip;
1196
1197         rw_rlock(&uihashtbl_lock);
1198         uip = uilookup(uid);
1199         if (uip == NULL) {
1200                 rw_runlock(&uihashtbl_lock);
1201                 uip = malloc(sizeof(*uip), M_UIDINFO, M_WAITOK | M_ZERO);
1202                 rw_wlock(&uihashtbl_lock);
1203                 /*
1204                  * There's a chance someone created our uidinfo while we
1205                  * were in malloc and not holding the lock, so we have to
1206                  * make sure we don't insert a duplicate uidinfo.
1207                  */
1208                 if ((old_uip = uilookup(uid)) != NULL) {
1209                         /* Someone else beat us to it. */
1210                         free(uip, M_UIDINFO);
1211                         uip = old_uip;
1212                 } else {
1213                         refcount_init(&uip->ui_ref, 0);
1214                         uip->ui_uid = uid;
1215                         LIST_INSERT_HEAD(UIHASH(uid), uip, ui_hash);
1216                 }
1217         }
1218         uihold(uip);
1219         rw_unlock(&uihashtbl_lock);
1220         return (uip);
1221 }
1222
1223 /*
1224  * Place another refcount on a uidinfo struct.
1225  */
1226 void
1227 uihold(uip)
1228         struct uidinfo *uip;
1229 {
1230
1231         refcount_acquire(&uip->ui_ref);
1232 }
1233
1234 /*-
1235  * Since uidinfo structs have a long lifetime, we use an
1236  * opportunistic refcounting scheme to avoid locking the lookup hash
1237  * for each release.
1238  *
1239  * If the refcount hits 0, we need to free the structure,
1240  * which means we need to lock the hash.
1241  * Optimal case:
1242  *   After locking the struct and lowering the refcount, if we find
1243  *   that we don't need to free, simply unlock and return.
1244  * Suboptimal case:
1245  *   If refcount lowering results in need to free, bump the count
1246  *   back up, lose the lock and acquire the locks in the proper
1247  *   order to try again.
1248  */
1249 void
1250 uifree(uip)
1251         struct uidinfo *uip;
1252 {
1253         int old;
1254
1255         /* Prepare for optimal case. */
1256         old = uip->ui_ref;
1257         if (old > 1 && atomic_cmpset_int(&uip->ui_ref, old, old - 1))
1258                 return;
1259
1260         /* Prepare for suboptimal case. */
1261         rw_wlock(&uihashtbl_lock);
1262         if (refcount_release(&uip->ui_ref)) {
1263                 LIST_REMOVE(uip, ui_hash);
1264                 rw_wunlock(&uihashtbl_lock);
1265                 if (uip->ui_sbsize != 0)
1266                         printf("freeing uidinfo: uid = %d, sbsize = %ld\n",
1267                             uip->ui_uid, uip->ui_sbsize);
1268                 if (uip->ui_proccnt != 0)
1269                         printf("freeing uidinfo: uid = %d, proccnt = %ld\n",
1270                             uip->ui_uid, uip->ui_proccnt);
1271                 FREE(uip, M_UIDINFO);
1272                 return;
1273         }
1274         /*
1275          * Someone added a reference between atomic_cmpset_int() and
1276          * rw_wlock(&uihashtbl_lock).
1277          */
1278         rw_wunlock(&uihashtbl_lock);
1279 }
1280
1281 /*
1282  * Change the count associated with number of processes
1283  * a given user is using.  When 'max' is 0, don't enforce a limit
1284  */
1285 int
1286 chgproccnt(uip, diff, max)
1287         struct  uidinfo *uip;
1288         int     diff;
1289         rlim_t  max;
1290 {
1291
1292         /* Don't allow them to exceed max, but allow subtraction. */
1293         if (diff > 0 && max != 0) {
1294                 if (atomic_fetchadd_long(&uip->ui_proccnt, (long)diff) + diff > max) {
1295                         atomic_subtract_long(&uip->ui_proccnt, (long)diff);
1296                         return (0);
1297                 }
1298         } else {
1299                 atomic_add_long(&uip->ui_proccnt, (long)diff);
1300                 if (uip->ui_proccnt < 0)
1301                         printf("negative proccnt for uid = %d\n", uip->ui_uid);
1302         }
1303         return (1);
1304 }
1305
1306 /*
1307  * Change the total socket buffer size a user has used.
1308  */
1309 int
1310 chgsbsize(uip, hiwat, to, max)
1311         struct  uidinfo *uip;
1312         u_int  *hiwat;
1313         u_int   to;
1314         rlim_t  max;
1315 {
1316         int diff;
1317
1318         diff = to - *hiwat;
1319         if (diff > 0) {
1320                 if (atomic_fetchadd_long(&uip->ui_sbsize, (long)diff) + diff > max) {
1321                         atomic_subtract_long(&uip->ui_sbsize, (long)diff);
1322                         return (0);
1323                 }
1324         } else {
1325                 atomic_add_long(&uip->ui_sbsize, (long)diff);
1326                 if (uip->ui_sbsize < 0)
1327                         printf("negative sbsize for uid = %d\n", uip->ui_uid);
1328         }
1329         *hiwat = to;
1330         return (1);
1331 }
1332
1333 /*
1334  * Change the count associated with number of pseudo-terminals
1335  * a given user is using.  When 'max' is 0, don't enforce a limit
1336  */
1337 int
1338 chgptscnt(uip, diff, max)
1339         struct  uidinfo *uip;
1340         int     diff;
1341         rlim_t  max;
1342 {
1343
1344         /* Don't allow them to exceed max, but allow subtraction. */
1345         if (diff > 0 && max != 0) {
1346                 if (atomic_fetchadd_long(&uip->ui_ptscnt, (long)diff) + diff > max) {
1347                         atomic_subtract_long(&uip->ui_ptscnt, (long)diff);
1348                         return (0);
1349                 }
1350         } else {
1351                 atomic_add_long(&uip->ui_ptscnt, (long)diff);
1352                 if (uip->ui_ptscnt < 0)
1353                         printf("negative ptscnt for uid = %d\n", uip->ui_uid);
1354         }
1355         return (1);
1356 }