sys/kern/kern_procctl.c

   1 /*-
   2  * Copyright (c) 2014 John Baldwin
   3  * Copyright (c) 2014 The FreeBSD Foundation
   4  *
   5  * Portions of this software were developed by Konstantin Belousov
   6  * under sponsorship from the FreeBSD Foundation.
   7  *
   8  * Redistribution and use in source and binary forms, with or without
   9  * modification, are permitted provided that the following conditions
  10  * are met:
  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 the
  15  *    documentation and/or other materials provided with the distribution.
  16  *
  17  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
  18  * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
  19  * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
  20  * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
  21  * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
  22  * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
  23  * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
  24  * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
  25  * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
  26  * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
  27  * SUCH DAMAGE.
  28  */
  29
  30 #include <sys/cdefs.h>
  31 __FBSDID("$FreeBSD$");
  32
  33 #include <sys/param.h>
  34 #include <sys/systm.h>
  35 #include <sys/capability.h>
  36 #include <sys/lock.h>
  37 #include <sys/mutex.h>
  38 #include <sys/priv.h>
  39 #include <sys/proc.h>
  40 #include <sys/procctl.h>
  41 #include <sys/sx.h>
  42 #include <sys/syscallsubr.h>
  43 #include <sys/sysproto.h>
  44 #include <sys/wait.h>
  45
  46 static int
  47 protect_setchild(struct thread *td, struct proc *p, int flags)
  48 {
  49
  50         PROC_LOCK_ASSERT(p, MA_OWNED);
  51         if (p->p_flag & P_SYSTEM || p_cansched(td, p) != 0)
  52                 return (0);
  53         if (flags & PPROT_SET) {
  54                 p->p_flag |= P_PROTECTED;
  55                 if (flags & PPROT_INHERIT)
  56                         p->p_flag2 |= P2_INHERIT_PROTECTED;
  57         } else {
  58                 p->p_flag &= ~P_PROTECTED;
  59                 p->p_flag2 &= ~P2_INHERIT_PROTECTED;
  60         }
  61         return (1);
  62 }
  63
  64 static int
  65 protect_setchildren(struct thread *td, struct proc *top, int flags)
  66 {
  67         struct proc *p;
  68         int ret;
  69
  70         p = top;
  71         ret = 0;
  72         sx_assert(&proctree_lock, SX_LOCKED);
  73         for (;;) {
  74                 ret |= protect_setchild(td, p, flags);
  75                 PROC_UNLOCK(p);
  76                 /*
  77                  * If this process has children, descend to them next,
  78                  * otherwise do any siblings, and if done with this level,
  79                  * follow back up the tree (but not past top).
  80                  */
  81                 if (!LIST_EMPTY(&p->p_children))
  82                         p = LIST_FIRST(&p->p_children);
  83                 else for (;;) {
  84                         if (p == top) {
  85                                 PROC_LOCK(p);
  86                                 return (ret);
  87                         }
  88                         if (LIST_NEXT(p, p_sibling)) {
  89                                 p = LIST_NEXT(p, p_sibling);
  90                                 break;
  91                         }
  92                         p = p->p_pptr;
  93                 }
  94                 PROC_LOCK(p);
  95         }
  96 }
  97
  98 static int
  99 protect_set(struct thread *td, struct proc *p, int flags)
 100 {
 101         int error, ret;
 102
 103         switch (PPROT_OP(flags)) {
 104         case PPROT_SET:
 105         case PPROT_CLEAR:
 106                 break;
 107         default:
 108                 return (EINVAL);
 109         }
 110
 111         if ((PPROT_FLAGS(flags) & ~(PPROT_DESCEND | PPROT_INHERIT)) != 0)
 112                 return (EINVAL);
 113
 114         error = priv_check(td, PRIV_VM_MADV_PROTECT);
 115         if (error)
 116                 return (error);
 117
 118         if (flags & PPROT_DESCEND)
 119                 ret = protect_setchildren(td, p, flags);
 120         else
 121                 ret = protect_setchild(td, p, flags);
 122         if (ret == 0)
 123                 return (EPERM);
 124         return (0);
 125 }
 126
 127 static int
 128 reap_acquire(struct thread *td, struct proc *p)
 129 {
 130
 131         sx_assert(&proctree_lock, SX_XLOCKED);
 132         if (p != curproc)
 133                 return (EPERM);
 134         if ((p->p_treeflag & P_TREE_REAPER) != 0)
 135                 return (EBUSY);
 136         p->p_treeflag |= P_TREE_REAPER;
 137         /*
 138          * We do not reattach existing children and the whole tree
 139          * under them to us, since p->p_reaper already seen them.
 140          */
 141         return (0);
 142 }
 143
 144 static int
 145 reap_release(struct thread *td, struct proc *p)
 146 {
 147
 148         sx_assert(&proctree_lock, SX_XLOCKED);
 149         if (p != curproc)
 150                 return (EPERM);
 151         if (p == initproc)
 152                 return (EINVAL);
 153         if ((p->p_treeflag & P_TREE_REAPER) == 0)
 154                 return (EINVAL);
 155         reaper_abandon_children(p, false);
 156         return (0);
 157 }
 158
 159 static int
 160 reap_status(struct thread *td, struct proc *p,
 161     struct procctl_reaper_status *rs)
 162 {
 163         struct proc *reap, *p2, *first_p;
 164
 165         sx_assert(&proctree_lock, SX_LOCKED);
 166         bzero(rs, sizeof(*rs));
 167         if ((p->p_treeflag & P_TREE_REAPER) == 0) {
 168                 reap = p->p_reaper;
 169         } else {
 170                 reap = p;
 171                 rs->rs_flags |= REAPER_STATUS_OWNED;
 172         }
 173         if (reap == initproc)
 174                 rs->rs_flags |= REAPER_STATUS_REALINIT;
 175         rs->rs_reaper = reap->p_pid;
 176         rs->rs_descendants = 0;
 177         rs->rs_children = 0;
 178         if (!LIST_EMPTY(&reap->p_reaplist)) {
 179                 first_p = LIST_FIRST(&reap->p_children);
 180                 if (first_p == NULL)
 181                         first_p = LIST_FIRST(&reap->p_reaplist);
 182                 rs->rs_pid = first_p->p_pid;
 183                 LIST_FOREACH(p2, &reap->p_reaplist, p_reapsibling) {
 184                         if (proc_realparent(p2) == reap)
 185                                 rs->rs_children++;
 186                         rs->rs_descendants++;
 187                 }
 188         } else {
 189                 rs->rs_pid = -1;
 190         }
 191         return (0);
 192 }
 193
 194 static int
 195 reap_getpids(struct thread *td, struct proc *p, struct procctl_reaper_pids *rp)
 196 {
 197         struct proc *reap, *p2;
 198         struct procctl_reaper_pidinfo *pi, *pip;
 199         u_int i, n;
 200         int error;
 201
 202         sx_assert(&proctree_lock, SX_LOCKED);
 203         PROC_UNLOCK(p);
 204         reap = (p->p_treeflag & P_TREE_REAPER) == 0 ? p->p_reaper : p;
 205         n = i = 0;
 206         error = 0;
 207         LIST_FOREACH(p2, &reap->p_reaplist, p_reapsibling)
 208                 n++;
 209         sx_unlock(&proctree_lock);
 210         if (rp->rp_count < n)
 211                 n = rp->rp_count;
 212         pi = malloc(n * sizeof(*pi), M_TEMP, M_WAITOK);
 213         sx_slock(&proctree_lock);
 214         LIST_FOREACH(p2, &reap->p_reaplist, p_reapsibling) {
 215                 if (i == n)
 216                         break;
 217                 pip = &pi[i];
 218                 bzero(pip, sizeof(*pip));
 219                 pip->pi_pid = p2->p_pid;
 220                 pip->pi_subtree = p2->p_reapsubtree;
 221                 pip->pi_flags = REAPER_PIDINFO_VALID;
 222                 if (proc_realparent(p2) == reap)
 223                         pip->pi_flags |= REAPER_PIDINFO_CHILD;
 224                 if ((p2->p_treeflag & P_TREE_REAPER) != 0)
 225                         pip->pi_flags |= REAPER_PIDINFO_REAPER;
 226                 i++;
 227         }
 228         sx_sunlock(&proctree_lock);
 229         error = copyout(pi, rp->rp_pids, i * sizeof(*pi));
 230         free(pi, M_TEMP);
 231         sx_slock(&proctree_lock);
 232         PROC_LOCK(p);
 233         return (error);
 234 }
 235
 236 static void
 237 reap_kill_proc(struct thread *td, struct proc *p2, ksiginfo_t *ksi,
 238     struct procctl_reaper_kill *rk, int *error)
 239 {
 240         int error1;
 241
 242         PROC_LOCK(p2);
 243         error1 = p_cansignal(td, p2, rk->rk_sig);
 244         if (error1 == 0) {
 245                 pksignal(p2, rk->rk_sig, ksi);
 246                 rk->rk_killed++;
 247                 *error = error1;
 248         } else if (*error == ESRCH) {
 249                 rk->rk_fpid = p2->p_pid;
 250                 *error = error1;
 251         }
 252         PROC_UNLOCK(p2);
 253 }
 254
 255 struct reap_kill_tracker {
 256         struct proc *parent;
 257         TAILQ_ENTRY(reap_kill_tracker) link;
 258 };
 259
 260 TAILQ_HEAD(reap_kill_tracker_head, reap_kill_tracker);
 261
 262 static void
 263 reap_kill_sched(struct reap_kill_tracker_head *tracker, struct proc *p2)
 264 {
 265         struct reap_kill_tracker *t;
 266
 267         t = malloc(sizeof(struct reap_kill_tracker), M_TEMP, M_WAITOK);
 268         t->parent = p2;
 269         TAILQ_INSERT_TAIL(tracker, t, link);
 270 }
 271
 272 static int
 273 reap_kill(struct thread *td, struct proc *p, struct procctl_reaper_kill *rk)
 274 {
 275         struct proc *reap, *p2;
 276         ksiginfo_t ksi;
 277         struct reap_kill_tracker_head tracker;
 278         struct reap_kill_tracker *t;
 279         int error;
 280
 281         sx_assert(&proctree_lock, SX_LOCKED);
 282         if (IN_CAPABILITY_MODE(td))
 283                 return (ECAPMODE);
 284         if (rk->rk_sig <= 0 || rk->rk_sig > _SIG_MAXSIG ||
 285             (rk->rk_flags & ~(REAPER_KILL_CHILDREN |
 286             REAPER_KILL_SUBTREE)) != 0 || (rk->rk_flags &
 287             (REAPER_KILL_CHILDREN | REAPER_KILL_SUBTREE)) ==
 288             (REAPER_KILL_CHILDREN | REAPER_KILL_SUBTREE))
 289                 return (EINVAL);
 290         PROC_UNLOCK(p);
 291         reap = (p->p_treeflag & P_TREE_REAPER) == 0 ? p->p_reaper : p;
 292         ksiginfo_init(&ksi);
 293         ksi.ksi_signo = rk->rk_sig;
 294         ksi.ksi_code = SI_USER;
 295         ksi.ksi_pid = td->td_proc->p_pid;
 296         ksi.ksi_uid = td->td_ucred->cr_ruid;
 297         error = ESRCH;
 298         rk->rk_killed = 0;
 299         rk->rk_fpid = -1;
 300         if ((rk->rk_flags & REAPER_KILL_CHILDREN) != 0) {
 301                 for (p2 = LIST_FIRST(&reap->p_children); p2 != NULL;
 302                     p2 = LIST_NEXT(p2, p_sibling)) {
 303                         reap_kill_proc(td, p2, &ksi, rk, &error);
 304                         /*
 305                          * Do not end the loop on error, signal
 306                          * everything we can.
 307                          */
 308                 }
 309         } else {
 310                 TAILQ_INIT(&tracker);
 311                 reap_kill_sched(&tracker, reap);
 312                 while ((t = TAILQ_FIRST(&tracker)) != NULL) {
 313                         MPASS((t->parent->p_treeflag & P_TREE_REAPER) != 0);
 314                         TAILQ_REMOVE(&tracker, t, link);
 315                         for (p2 = LIST_FIRST(&t->parent->p_reaplist); p2 != NULL;
 316                             p2 = LIST_NEXT(p2, p_reapsibling)) {
 317                                 if (t->parent == reap &&
 318                                     (rk->rk_flags & REAPER_KILL_SUBTREE) != 0 &&
 319                                     p2->p_reapsubtree != rk->rk_subtree)
 320                                         continue;
 321                                 if ((p2->p_treeflag & P_TREE_REAPER) != 0)
 322                                         reap_kill_sched(&tracker, p2);
 323                                 reap_kill_proc(td, p2, &ksi, rk, &error);
 324                         }
 325                         free(t, M_TEMP);
 326                 }
 327         }
 328         PROC_LOCK(p);
 329         return (error);
 330 }
 331
 332 static int
 333 trace_ctl(struct thread *td, struct proc *p, int state)
 334 {
 335
 336         PROC_LOCK_ASSERT(p, MA_OWNED);
 337
 338         /*
 339          * Ktrace changes p_traceflag from or to zero under the
 340          * process lock, so the test does not need to acquire ktrace
 341          * mutex.
 342          */
 343         if ((p->p_flag & P_TRACED) != 0 || p->p_traceflag != 0)
 344                 return (EBUSY);
 345
 346         switch (state) {
 347         case PROC_TRACE_CTL_ENABLE:
 348                 if (td->td_proc != p)
 349                         return (EPERM);
 350                 p->p_flag2 &= ~(P2_NOTRACE | P2_NOTRACE_EXEC);
 351                 break;
 352         case PROC_TRACE_CTL_DISABLE_EXEC:
 353                 p->p_flag2 |= P2_NOTRACE_EXEC | P2_NOTRACE;
 354                 break;
 355         case PROC_TRACE_CTL_DISABLE:
 356                 if ((p->p_flag2 & P2_NOTRACE_EXEC) != 0) {
 357                         KASSERT((p->p_flag2 & P2_NOTRACE) != 0,
 358                             ("dandling P2_NOTRACE_EXEC"));
 359                         if (td->td_proc != p)
 360                                 return (EPERM);
 361                         p->p_flag2 &= ~P2_NOTRACE_EXEC;
 362                 } else {
 363                         p->p_flag2 |= P2_NOTRACE;
 364                 }
 365                 break;
 366         default:
 367                 return (EINVAL);
 368         }
 369         return (0);
 370 }
 371
 372 static int
 373 trace_status(struct thread *td, struct proc *p, int *data)
 374 {
 375
 376         if ((p->p_flag2 & P2_NOTRACE) != 0) {
 377                 KASSERT((p->p_flag & P_TRACED) == 0,
 378                     ("%d traced but tracing disabled", p->p_pid));
 379                 *data = -1;
 380         } else if ((p->p_flag & P_TRACED) != 0) {
 381                 *data = p->p_pptr->p_pid;
 382         } else {
 383                 *data = 0;
 384         }
 385         return (0);
 386 }
 387
 388 #ifndef _SYS_SYSPROTO_H_
 389 struct procctl_args {
 390         idtype_t idtype;
 391         id_t    id;
 392         int     com;
 393         void    *data;
 394 };
 395 #endif
 396 /* ARGSUSED */
 397 int
 398 sys_procctl(struct thread *td, struct procctl_args *uap)
 399 {
 400         void *data;
 401         union {
 402                 struct procctl_reaper_status rs;
 403                 struct procctl_reaper_pids rp;
 404                 struct procctl_reaper_kill rk;
 405         } x;
 406         int error, error1, flags;
 407
 408         switch (uap->com) {
 409         case PROC_SPROTECT:
 410         case PROC_TRACE_CTL:
 411                 error = copyin(uap->data, &flags, sizeof(flags));
 412                 if (error != 0)
 413                         return (error);
 414                 data = &flags;
 415                 break;
 416         case PROC_REAP_ACQUIRE:
 417         case PROC_REAP_RELEASE:
 418                 if (uap->data != NULL)
 419                         return (EINVAL);
 420                 data = NULL;
 421                 break;
 422         case PROC_REAP_STATUS:
 423                 data = &x.rs;
 424                 break;
 425         case PROC_REAP_GETPIDS:
 426                 error = copyin(uap->data, &x.rp, sizeof(x.rp));
 427                 if (error != 0)
 428                         return (error);
 429                 data = &x.rp;
 430                 break;
 431         case PROC_REAP_KILL:
 432                 error = copyin(uap->data, &x.rk, sizeof(x.rk));
 433                 if (error != 0)
 434                         return (error);
 435                 data = &x.rk;
 436                 break;
 437         case PROC_TRACE_STATUS:
 438                 data = &flags;
 439                 break;
 440         default:
 441                 return (EINVAL);
 442         }
 443         error = kern_procctl(td, uap->idtype, uap->id, uap->com, data);
 444         switch (uap->com) {
 445         case PROC_REAP_STATUS:
 446                 if (error == 0)
 447                         error = copyout(&x.rs, uap->data, sizeof(x.rs));
 448                 break;
 449         case PROC_REAP_KILL:
 450                 error1 = copyout(&x.rk, uap->data, sizeof(x.rk));
 451                 if (error == 0)
 452                         error = error1;
 453                 break;
 454         case PROC_TRACE_STATUS:
 455                 if (error == 0)
 456                         error = copyout(&flags, uap->data, sizeof(flags));
 457                 break;
 458         }
 459         return (error);
 460 }
 461
 462 static int
 463 kern_procctl_single(struct thread *td, struct proc *p, int com, void *data)
 464 {
 465
 466         PROC_LOCK_ASSERT(p, MA_OWNED);
 467         switch (com) {
 468         case PROC_SPROTECT:
 469                 return (protect_set(td, p, *(int *)data));
 470         case PROC_REAP_ACQUIRE:
 471                 return (reap_acquire(td, p));
 472         case PROC_REAP_RELEASE:
 473                 return (reap_release(td, p));
 474         case PROC_REAP_STATUS:
 475                 return (reap_status(td, p, data));
 476         case PROC_REAP_GETPIDS:
 477                 return (reap_getpids(td, p, data));
 478         case PROC_REAP_KILL:
 479                 return (reap_kill(td, p, data));
 480         case PROC_TRACE_CTL:
 481                 return (trace_ctl(td, p, *(int *)data));
 482         case PROC_TRACE_STATUS:
 483                 return (trace_status(td, p, data));
 484         default:
 485                 return (EINVAL);
 486         }
 487 }
 488
 489 int
 490 kern_procctl(struct thread *td, idtype_t idtype, id_t id, int com, void *data)
 491 {
 492         struct pgrp *pg;
 493         struct proc *p;
 494         int error, first_error, ok;
 495         bool tree_locked;
 496
 497         switch (com) {
 498         case PROC_REAP_ACQUIRE:
 499         case PROC_REAP_RELEASE:
 500         case PROC_REAP_STATUS:
 501         case PROC_REAP_GETPIDS:
 502         case PROC_REAP_KILL:
 503         case PROC_TRACE_STATUS:
 504                 if (idtype != P_PID)
 505                         return (EINVAL);
 506         }
 507
 508         switch (com) {
 509         case PROC_SPROTECT:
 510         case PROC_REAP_STATUS:
 511         case PROC_REAP_GETPIDS:
 512         case PROC_REAP_KILL:
 513         case PROC_TRACE_CTL:
 514                 sx_slock(&proctree_lock);
 515                 tree_locked = true;
 516                 break;
 517         case PROC_REAP_ACQUIRE:
 518         case PROC_REAP_RELEASE:
 519                 sx_xlock(&proctree_lock);
 520                 tree_locked = true;
 521                 break;
 522         case PROC_TRACE_STATUS:
 523                 tree_locked = false;
 524                 break;
 525         default:
 526                 return (EINVAL);
 527         }
 528
 529         switch (idtype) {
 530         case P_PID:
 531                 p = pfind(id);
 532                 if (p == NULL) {
 533                         error = ESRCH;
 534                         break;
 535                 }
 536                 error = p_cansee(td, p);
 537                 if (error == 0)
 538                         error = kern_procctl_single(td, p, com, data);
 539                 PROC_UNLOCK(p);
 540                 break;
 541         case P_PGID:
 542                 /*
 543                  * Attempt to apply the operation to all members of the
 544                  * group.  Ignore processes in the group that can't be
 545                  * seen.  Ignore errors so long as at least one process is
 546                  * able to complete the request successfully.
 547                  */
 548                 pg = pgfind(id);
 549                 if (pg == NULL) {
 550                         error = ESRCH;
 551                         break;
 552                 }
 553                 PGRP_UNLOCK(pg);
 554                 ok = 0;
 555                 first_error = 0;
 556                 LIST_FOREACH(p, &pg->pg_members, p_pglist) {
 557                         PROC_LOCK(p);
 558                         if (p->p_state == PRS_NEW || p_cansee(td, p) != 0) {
 559                                 PROC_UNLOCK(p);
 560                                 continue;
 561                         }
 562                         error = kern_procctl_single(td, p, com, data);
 563                         PROC_UNLOCK(p);
 564                         if (error == 0)
 565                                 ok = 1;
 566                         else if (first_error == 0)
 567                                 first_error = error;
 568                 }
 569                 if (ok)
 570                         error = 0;
 571                 else if (first_error != 0)
 572                         error = first_error;
 573                 else
 574                         /*
 575                          * Was not able to see any processes in the
 576                          * process group.
 577                          */
 578                         error = ESRCH;
 579                 break;
 580         default:
 581                 error = EINVAL;
 582                 break;
 583         }
 584         if (tree_locked)
 585                 sx_unlock(&proctree_lock);
 586         return (error);
 587 }