sys/kern/kern_plimit.c

   1 /*
   2  * Copyright (c) 2006 The DragonFly Project.  All rights reserved.
   3  *
   4  * This code is derived from software contributed to The DragonFly Project
   5  * by Matthew Dillon <dillon@backplane.com>
   6  *
   7  * Redistribution and use in source and binary forms, with or without
   8  * modification, are permitted provided that the following conditions
   9  * are met:
  10  *
  11  * 1. Redistributions of source code must retain the above copyright
  12  *    notice, this list of conditions and the following disclaimer.
  13  * 2. Redistributions in binary form must reproduce the above copyright
  14  *    notice, this list of conditions and the following disclaimer in
  15  *    the documentation and/or other materials provided with the
  16  *    distribution.
  17  * 3. Neither the name of The DragonFly Project nor the names of its
  18  *    contributors may be used to endorse or promote products derived
  19  *    from this software without specific, prior written permission.
  20  *
  21  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
  22  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
  23  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS
  24  * FOR A PARTICULAR PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE
  25  * COPYRIGHT HOLDERS OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,
  26  * INCIDENTAL, SPECIAL, EXEMPLARY OR CONSEQUENTIAL DAMAGES (INCLUDING,
  27  * BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES;
  28  * LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED
  29  * AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY,
  30  * OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT
  31  * OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  32  * SUCH DAMAGE.
  33  */
  34 /*
  35  * Copyright (c) 1982, 1986, 1991, 1993
  36  *      The Regents of the University of California.  All rights reserved.
  37  * (c) UNIX System Laboratories, Inc.
  38  * All or some portions of this file are derived from material licensed
  39  * to the University of California by American Telephone and Telegraph
  40  * Co. or Unix System Laboratories, Inc. and are reproduced herein with
  41  * the permission of UNIX System Laboratories, Inc.
  42  *
  43  * Redistribution and use in source and binary forms, with or without
  44  * modification, are permitted provided that the following conditions
  45  * are met:
  46  * 1. Redistributions of source code must retain the above copyright
  47  *    notice, this list of conditions and the following disclaimer.
  48  * 2. Redistributions in binary form must reproduce the above copyright
  49  *    notice, this list of conditions and the following disclaimer in the
  50  *    documentation and/or other materials provided with the distribution.
  51  * 3. Neither the name of the University nor the names of its contributors
  52  *    may be used to endorse or promote products derived from this software
  53  *    without specific prior written permission.
  54  *
  55  * THIS SOFTWARE IS PROVIDED BY THE REGENTS AND CONTRIBUTORS ``AS IS'' AND
  56  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  57  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  58  * ARE DISCLAIMED.  IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  59  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  60  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  61  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  62  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  63  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  64  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  65  * SUCH DAMAGE.
  66  *
  67  *      @(#)kern_resource.c     8.5 (Berkeley) 1/21/94
  68  * $FreeBSD: src/sys/kern/kern_resource.c,v 1.55.2.5 2001/11/03 01:41:08 ps Exp $
  69  * $DragonFly: src/sys/kern/kern_plimit.c,v 1.3 2008/05/08 01:26:00 dillon Exp $
  70  */
  71
  72 #include <sys/resource.h>
  73 #include <sys/spinlock.h>
  74 #include <sys/proc.h>
  75 #include <sys/priv.h>
  76 #include <sys/file.h>
  77 #include <sys/lockf.h>
  78 #include <sys/kern_syscall.h>
  79
  80 #include <vm/vm_param.h>
  81 #include <vm/vm.h>
  82 #include <vm/vm_map.h>
  83
  84 #include <machine/pmap.h>
  85
  86 #include <sys/spinlock2.h>
  87
  88 static void plimit_copy(struct plimit *olimit, struct plimit *nlimit);
  89
  90 /*
  91  * Initialize proc0's plimit structure.  All later plimit structures
  92  * are inherited through fork.
  93  */
  94 void
  95 plimit_init0(struct plimit *limit)
  96 {
  97         int i;
  98         rlim_t lim;
  99
 100         for (i = 0; i < RLIM_NLIMITS; ++i) {
 101                 limit->pl_rlimit[i].rlim_cur = RLIM_INFINITY;
 102                 limit->pl_rlimit[i].rlim_max = RLIM_INFINITY;
 103         }
 104         limit->pl_rlimit[RLIMIT_NOFILE].rlim_cur = maxfiles;
 105         limit->pl_rlimit[RLIMIT_NOFILE].rlim_max = maxfiles;
 106         limit->pl_rlimit[RLIMIT_NPROC].rlim_cur = maxproc;
 107         limit->pl_rlimit[RLIMIT_NPROC].rlim_max = maxproc;
 108         lim = ptoa((rlim_t)vmstats.v_free_count);
 109         limit->pl_rlimit[RLIMIT_RSS].rlim_max = lim;
 110         limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_max = lim;
 111         limit->pl_rlimit[RLIMIT_MEMLOCK].rlim_cur = lim / 3;
 112         limit->p_cpulimit = RLIM_INFINITY;
 113         limit->p_refcnt = 1;
 114         spin_init(&limit->p_spin);
 115 }
 116
 117 /*
 118  * Return a plimit for use by a new forked process given the one
 119  * contained in the parent process.
 120  *
 121  * MPSAFE
 122  */
 123 struct plimit *
 124 plimit_fork(struct proc *p1)
 125 {
 126         struct plimit *olimit = p1->p_limit;
 127         struct plimit *nlimit = NULL;
 128         struct plimit *rlimit;
 129
 130         /*
 131          * If we are exclusive (but not threaded-exclusive), but have only
 132          * one reference, we can convert the structure to copy-on-write
 133          * again.
 134          *
 135          * If we were threaded but are no longer threaded we can do the same
 136          * thing.
 137          */
 138         if (olimit->p_exclusive == 1) {
 139                 KKASSERT(olimit->p_refcnt == 1);
 140                 olimit->p_exclusive = 0;
 141         } else if (olimit->p_exclusive == 2 && p1->p_nthreads == 1) {
 142                 KKASSERT(olimit->p_refcnt == 1);
 143                 olimit->p_exclusive = 0;
 144         }
 145
 146         /*
 147          * Take a short-cut that requires limited spin locks.  If we aren't
 148          * exclusive we will not be threaded and we can just bump the ref
 149          * count.  If that is true and we also have only one ref then there
 150          * can be no other accessors.
 151          */
 152         if (olimit->p_exclusive == 0) {
 153                 if (olimit->p_refcnt == 1) {
 154                         ++olimit->p_refcnt;
 155                 } else {
 156                         spin_lock_wr(&olimit->p_spin);
 157                         ++olimit->p_refcnt;
 158                         spin_unlock_wr(&olimit->p_spin);
 159                 }
 160                 return(olimit);
 161         }
 162
 163         /*
 164          * Full-blown code-up.
 165          */
 166         nlimit = NULL;
 167         spin_lock_wr(&olimit->p_spin);
 168
 169         for (;;) {
 170                 if (olimit->p_exclusive == 0) {
 171                         ++olimit->p_refcnt;
 172                         rlimit = olimit;
 173                         break;
 174                 }
 175                 if (nlimit) {
 176                         plimit_copy(olimit, nlimit);
 177                         rlimit = nlimit;
 178                         nlimit = NULL;
 179                         break;
 180                 }
 181                 spin_unlock_wr(&olimit->p_spin);
 182                 nlimit = kmalloc(sizeof(*nlimit), M_SUBPROC, M_WAITOK);
 183                 spin_lock_wr(&olimit->p_spin);
 184         }
 185         spin_unlock_wr(&olimit->p_spin);
 186         if (nlimit)
 187                 kfree(nlimit, M_SUBPROC);
 188         return(rlimit);
 189 }
 190
 191 /*
 192  * This routine is called when a new LWP is created for a process.  We
 193  * must force exclusivity (=2) so p->p_limit remains stable.
 194  *
 195  * LWPs share the same process structure so this does not bump refcnt.
 196  */
 197 void
 198 plimit_lwp_fork(struct proc *p)
 199 {
 200         struct plimit *olimit;
 201
 202         for (;;) {
 203                 olimit = p->p_limit;
 204                 if (olimit->p_exclusive == 2) {
 205                         KKASSERT(olimit->p_refcnt == 1);
 206                         break;
 207                 }
 208                 if (olimit->p_refcnt == 1) {
 209                         olimit->p_exclusive = 2;
 210                         break;
 211                 }
 212                 plimit_modify(p, -1, NULL);
 213         }
 214 }
 215
 216 /*
 217  * This routine is called to fixup a proces's p_limit structure prior
 218  * to it being modified.  If index >= 0 the specified modification is also
 219  * made.
 220  *
 221  * This routine must make the limit structure exclusive.  A later fork
 222  * will convert it back to copy-on-write if possible.
 223  *
 224  * We can count on p->p_limit being stable since if we had created any
 225  * threads it will have already been made exclusive (=2).
 226  *
 227  * MPSAFE
 228  */
 229 void
 230 plimit_modify(struct proc *p, int index, struct rlimit *rlim)
 231 {
 232         struct plimit *olimit;
 233         struct plimit *nlimit;
 234         struct plimit *rlimit;
 235
 236         /*
 237          * Shortcut.  If we are not threaded we may be able to trivially
 238          * set the structure to exclusive access without needing to acquire
 239          * any spinlocks.   The p_limit structure will be stable.
 240          */
 241         olimit = p->p_limit;
 242         if (p->p_nthreads == 1) {
 243                 if (olimit->p_exclusive == 0 && olimit->p_refcnt == 1)
 244                         olimit->p_exclusive = 1;
 245                 if (olimit->p_exclusive) {
 246                         if (index >= 0)
 247                                 p->p_limit->pl_rlimit[index] = *rlim;
 248                         return;
 249                 }
 250         }
 251
 252         /*
 253          * Full-blown code-up.  Make a copy if we aren't exclusive.  If
 254          * we have only one ref we can safely convert the structure to
 255          * exclusive without copying.
 256          */
 257         nlimit = NULL;
 258         spin_lock_wr(&olimit->p_spin);
 259
 260         for (;;) {
 261                 if (olimit->p_refcnt == 1) {
 262                         if (olimit->p_exclusive == 0)
 263                                 olimit->p_exclusive = 1;
 264                         rlimit = olimit;
 265                         break;
 266                 }
 267                 KKASSERT(olimit->p_exclusive == 0);
 268                 if (nlimit) {
 269                         plimit_copy(olimit, nlimit);
 270                         nlimit->p_exclusive = 1;
 271                         p->p_limit = nlimit;
 272                         rlimit = nlimit;
 273                         nlimit = NULL;
 274                         break;
 275                 }
 276                 spin_unlock_wr(&olimit->p_spin);
 277                 nlimit = kmalloc(sizeof(*nlimit), M_SUBPROC, M_WAITOK);
 278                 spin_lock_wr(&olimit->p_spin);
 279         }
 280         if (index >= 0)
 281                 rlimit->pl_rlimit[index] = *rlim;
 282         spin_unlock_wr(&olimit->p_spin);
 283         if (nlimit)
 284                 kfree(nlimit, M_SUBPROC);
 285 }
 286
 287 /*
 288  * Destroy a process's plimit structure.
 289  *
 290  * MPSAFE
 291  */
 292 void
 293 plimit_free(struct proc *p)
 294 {
 295         struct plimit *limit;
 296
 297         if ((limit = p->p_limit) != NULL) {
 298                 p->p_limit = NULL;
 299
 300                 if (limit->p_refcnt == 1) {
 301                         limit->p_refcnt = -999;
 302                         kfree(limit, M_SUBPROC);
 303                 } else {
 304                         spin_lock_wr(&limit->p_spin);
 305                         if (--limit->p_refcnt == 0) {
 306                                 spin_unlock_wr(&limit->p_spin);
 307                                 kfree(limit, M_SUBPROC);
 308                         } else {
 309                                 spin_unlock_wr(&limit->p_spin);
 310                         }
 311                 }
 312         }
 313 }
 314
 315 /*
 316  * Modify a resource limit (from system call)
 317  *
 318  * MPSAFE
 319  */
 320 int
 321 kern_setrlimit(u_int which, struct rlimit *limp)
 322 {
 323         struct proc *p = curproc;
 324         struct plimit *limit;
 325         struct rlimit *alimp;
 326         int error;
 327
 328         if (which >= RLIM_NLIMITS)
 329                 return (EINVAL);
 330
 331         /*
 332          * We will be modifying a resource, make a copy if necessary.
 333          */
 334         plimit_modify(p, -1, NULL);
 335         limit = p->p_limit;
 336         alimp = &limit->pl_rlimit[which];
 337
 338         /*
 339          * Preserve historical bugs by treating negative limits as unsigned.
 340          */
 341         if (limp->rlim_cur < 0)
 342                 limp->rlim_cur = RLIM_INFINITY;
 343         if (limp->rlim_max < 0)
 344                 limp->rlim_max = RLIM_INFINITY;
 345
 346         spin_lock_rd(&limit->p_spin);
 347         if (limp->rlim_cur > alimp->rlim_max ||
 348             limp->rlim_max > alimp->rlim_max) {
 349                 spin_unlock_rd(&limit->p_spin);
 350                 error = priv_check_cred(p->p_ucred, PRIV_PROC_SETRLIMIT, 0);
 351                 if (error)
 352                         return (error);
 353         } else {
 354                 spin_unlock_rd(&limit->p_spin);
 355         }
 356         if (limp->rlim_cur > limp->rlim_max)
 357                 limp->rlim_cur = limp->rlim_max;
 358
 359         switch (which) {
 360         case RLIMIT_CPU:
 361                 spin_lock_wr(&limit->p_spin);
 362                 if (limp->rlim_cur > RLIM_INFINITY / (rlim_t)1000000)
 363                         limit->p_cpulimit = RLIM_INFINITY;
 364                 else
 365                         limit->p_cpulimit = (rlim_t)1000000 * limp->rlim_cur;
 366                 spin_unlock_wr(&limit->p_spin);
 367                 break;
 368         case RLIMIT_DATA:
 369                 if (limp->rlim_cur > maxdsiz)
 370                         limp->rlim_cur = maxdsiz;
 371                 if (limp->rlim_max > maxdsiz)
 372                         limp->rlim_max = maxdsiz;
 373                 break;
 374
 375         case RLIMIT_STACK:
 376                 if (limp->rlim_cur > maxssiz)
 377                         limp->rlim_cur = maxssiz;
 378                 if (limp->rlim_max > maxssiz)
 379                         limp->rlim_max = maxssiz;
 380                 /*
 381                  * Stack is allocated to the max at exec time with only
 382                  * "rlim_cur" bytes accessible.  If stack limit is going
 383                  * up make more accessible, if going down make inaccessible.
 384                  */
 385                 spin_lock_rd(&limit->p_spin);
 386                 if (limp->rlim_cur != alimp->rlim_cur) {
 387                         vm_offset_t addr;
 388                         vm_size_t size;
 389                         vm_prot_t prot;
 390
 391                         if (limp->rlim_cur > alimp->rlim_cur) {
 392                                 prot = VM_PROT_ALL;
 393                                 size = limp->rlim_cur - alimp->rlim_cur;
 394                                 addr = USRSTACK - limp->rlim_cur;
 395                         } else {
 396                                 prot = VM_PROT_NONE;
 397                                 size = alimp->rlim_cur - limp->rlim_cur;
 398                                 addr = USRSTACK - alimp->rlim_cur;
 399                         }
 400                         spin_unlock_rd(&limit->p_spin);
 401                         addr = trunc_page(addr);
 402                         size = round_page(size);
 403                         (void) vm_map_protect(&p->p_vmspace->vm_map,
 404                                               addr, addr+size, prot, FALSE);
 405                 } else {
 406                         spin_unlock_rd(&limit->p_spin);
 407                 }
 408                 break;
 409
 410         case RLIMIT_NOFILE:
 411                 if (limp->rlim_cur > maxfilesperproc)
 412                         limp->rlim_cur = maxfilesperproc;
 413                 if (limp->rlim_max > maxfilesperproc)
 414                         limp->rlim_max = maxfilesperproc;
 415                 break;
 416
 417         case RLIMIT_NPROC:
 418                 if (limp->rlim_cur > maxprocperuid)
 419                         limp->rlim_cur = maxprocperuid;
 420                 if (limp->rlim_max > maxprocperuid)
 421                         limp->rlim_max = maxprocperuid;
 422                 if (limp->rlim_cur < 1)
 423                         limp->rlim_cur = 1;
 424                 if (limp->rlim_max < 1)
 425                         limp->rlim_max = 1;
 426                 break;
 427         case RLIMIT_POSIXLOCKS:
 428                 if (limp->rlim_cur > maxposixlocksperuid)
 429                         limp->rlim_cur = maxposixlocksperuid;
 430                 if (limp->rlim_max > maxposixlocksperuid)
 431                         limp->rlim_max = maxposixlocksperuid;
 432                 break;
 433         }
 434         spin_lock_wr(&limit->p_spin);
 435         *alimp = *limp;
 436         spin_unlock_wr(&limit->p_spin);
 437         return (0);
 438 }
 439
 440 /*
 441  * The rlimit indexed by which is returned in the second argument.
 442  *
 443  * MPSAFE
 444  */
 445 int
 446 kern_getrlimit(u_int which, struct rlimit *limp)
 447 {
 448         struct proc *p = curproc;
 449         struct plimit *limit;
 450
 451         if (which >= RLIM_NLIMITS)
 452                 return (EINVAL);
 453
 454         limit = p->p_limit;
 455         spin_lock_rd(&limit->p_spin);
 456         *limp = p->p_rlimit[which];
 457         spin_unlock_rd(&limit->p_spin);
 458         return (0);
 459 }
 460
 461 /*
 462  * Determine if the cpu limit has been reached and return an operations
 463  * code for the caller to perform.
 464  *
 465  * MPSAFE
 466  */
 467 int
 468 plimit_testcpulimit(struct plimit *limit, u_int64_t ttime)
 469 {
 470         struct rlimit *rlim;
 471         int mode;
 472
 473         mode = PLIMIT_TESTCPU_OK;
 474         if (limit->p_cpulimit != RLIM_INFINITY) {
 475                 spin_lock_rd(&limit->p_spin);
 476                 if (ttime > limit->p_cpulimit) {
 477                         rlim = &limit->pl_rlimit[RLIMIT_CPU];
 478                         if (ttime / (rlim_t)1000000 >= rlim->rlim_max + 5) {
 479                                 mode = PLIMIT_TESTCPU_KILL;
 480                         } else {
 481                                 mode = PLIMIT_TESTCPU_XCPU;
 482                         }
 483                 }
 484                 spin_unlock_rd(&limit->p_spin);
 485         }
 486         return(mode);
 487 }
 488
 489 /*
 490  * Helper routine to copy olimit to nlimit and initialize nlimit for
 491  * use.  nlimit's reference count will be set to 1 and its exclusive bit
 492  * will be cleared.
 493  *
 494  * MPSAFE
 495  */
 496 static
 497 void
 498 plimit_copy(struct plimit *olimit, struct plimit *nlimit)
 499 {
 500         *nlimit = *olimit;
 501
 502         spin_init(&nlimit->p_spin);
 503         nlimit->p_refcnt = 1;
 504         nlimit->p_exclusive = 0;
 505 }
 506