Now available: The Design and Implementation of the FreeBSD Operating System (Second Edition)
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
FreeBSD/Linux Kernel Cross Reference
sys/kern/kern_exec.c
Version:
- FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
SearchContext: - none - 3 - 10
1 /*- 2 * Copyright (c) 1993, David Greenman 3 * All rights reserved. 4 * 5 * Redistribution and use in source and binary forms, with or without 6 * modification, are permitted provided that the following conditions 7 * are met: 8 * 1. Redistributions of source code must retain the above copyright 9 * notice, this list of conditions and the following disclaimer. 10 * 2. Redistributions in binary form must reproduce the above copyright 11 * notice, this list of conditions and the following disclaimer in the 12 * documentation and/or other materials provided with the distribution. 13 * 14 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 15 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 16 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 17 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 18 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 19 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 20 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 21 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 22 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 23 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 24 * SUCH DAMAGE. 25 */ 26 27 #include <sys/cdefs.h> 28 __FBSDID("$FreeBSD: releng/9.0/sys/kern/kern_exec.c 226248 2011-10-11 13:18:44Z kib $"); 29 30 #include "opt_capsicum.h" 31 #include "opt_hwpmc_hooks.h" 32 #include "opt_kdtrace.h" 33 #include "opt_ktrace.h" 34 #include "opt_vm.h" 35 36 #include <sys/param.h> 37 #include <sys/capability.h> 38 #include <sys/systm.h> 39 #include <sys/capability.h> 40 #include <sys/eventhandler.h> 41 #include <sys/lock.h> 42 #include <sys/mutex.h> 43 #include <sys/sysproto.h> 44 #include <sys/signalvar.h> 45 #include <sys/kernel.h> 46 #include <sys/mount.h> 47 #include <sys/filedesc.h> 48 #include <sys/fcntl.h> 49 #include <sys/acct.h> 50 #include <sys/exec.h> 51 #include <sys/imgact.h> 52 #include <sys/imgact_elf.h> 53 #include <sys/wait.h> 54 #include <sys/malloc.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/pioctl.h> 58 #include <sys/namei.h> 59 #include <sys/resourcevar.h> 60 #include <sys/sdt.h> 61 #include <sys/sf_buf.h> 62 #include <sys/syscallsubr.h> 63 #include <sys/sysent.h> 64 #include <sys/shm.h> 65 #include <sys/sysctl.h> 66 #include <sys/vnode.h> 67 #include <sys/stat.h> 68 #ifdef KTRACE 69 #include <sys/ktrace.h> 70 #endif 71 72 #include <vm/vm.h> 73 #include <vm/vm_param.h> 74 #include <vm/pmap.h> 75 #include <vm/vm_page.h> 76 #include <vm/vm_map.h> 77 #include <vm/vm_kern.h> 78 #include <vm/vm_extern.h> 79 #include <vm/vm_object.h> 80 #include <vm/vm_pager.h> 81 82 #ifdef HWPMC_HOOKS 83 #include <sys/pmckern.h> 84 #endif 85 86 #include <machine/reg.h> 87 88 #include <security/audit/audit.h> 89 #include <security/mac/mac_framework.h> 90 91 #ifdef KDTRACE_HOOKS 92 #include <sys/dtrace_bsd.h> 93 dtrace_execexit_func_t dtrace_fasttrap_exec; 94 #endif 95 96 SDT_PROVIDER_DECLARE(proc); 97 SDT_PROBE_DEFINE(proc, kernel, , exec, exec); 98 SDT_PROBE_ARGTYPE(proc, kernel, , exec, 0, "char *"); 99 SDT_PROBE_DEFINE(proc, kernel, , exec_failure, exec-failure); 100 SDT_PROBE_ARGTYPE(proc, kernel, , exec_failure, 0, "int"); 101 SDT_PROBE_DEFINE(proc, kernel, , exec_success, exec-success); 102 SDT_PROBE_ARGTYPE(proc, kernel, , exec_success, 0, "char *"); 103 104 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 105 106 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 107 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 108 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 109 static int do_execve(struct thread *td, struct image_args *args, 110 struct mac *mac_p); 111 112 /* XXX This should be vm_size_t. */ 113 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD, 114 NULL, 0, sysctl_kern_ps_strings, "LU", ""); 115 116 /* XXX This should be vm_size_t. */ 117 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 118 CTLFLAG_CAPRD, NULL, 0, sysctl_kern_usrstack, "LU", ""); 119 120 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD, 121 NULL, 0, sysctl_kern_stackprot, "I", ""); 122 123 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 124 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 125 &ps_arg_cache_limit, 0, ""); 126 127 static int map_at_zero = 0; 128 TUNABLE_INT("security.bsd.map_at_zero", &map_at_zero); 129 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RW, &map_at_zero, 0, 130 "Permit processes to map an object at virtual address 0."); 131 132 static int 133 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 134 { 135 struct proc *p; 136 int error; 137 138 p = curproc; 139 #ifdef SCTL_MASK32 140 if (req->flags & SCTL_MASK32) { 141 unsigned int val; 142 val = (unsigned int)p->p_sysent->sv_psstrings; 143 error = SYSCTL_OUT(req, &val, sizeof(val)); 144 } else 145 #endif 146 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 147 sizeof(p->p_sysent->sv_psstrings)); 148 return error; 149 } 150 151 static int 152 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 153 { 154 struct proc *p; 155 int error; 156 157 p = curproc; 158 #ifdef SCTL_MASK32 159 if (req->flags & SCTL_MASK32) { 160 unsigned int val; 161 val = (unsigned int)p->p_sysent->sv_usrstack; 162 error = SYSCTL_OUT(req, &val, sizeof(val)); 163 } else 164 #endif 165 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 166 sizeof(p->p_sysent->sv_usrstack)); 167 return error; 168 } 169 170 static int 171 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 172 { 173 struct proc *p; 174 175 p = curproc; 176 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 177 sizeof(p->p_sysent->sv_stackprot))); 178 } 179 180 /* 181 * Each of the items is a pointer to a `const struct execsw', hence the 182 * double pointer here. 183 */ 184 static const struct execsw **execsw; 185 186 #ifndef _SYS_SYSPROTO_H_ 187 struct execve_args { 188 char *fname; 189 char **argv; 190 char **envv; 191 }; 192 #endif 193 194 int 195 sys_execve(td, uap) 196 struct thread *td; 197 struct execve_args /* { 198 char *fname; 199 char **argv; 200 char **envv; 201 } */ *uap; 202 { 203 int error; 204 struct image_args args; 205 206 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 207 uap->argv, uap->envv); 208 if (error == 0) 209 error = kern_execve(td, &args, NULL); 210 return (error); 211 } 212 213 #ifndef _SYS_SYSPROTO_H_ 214 struct fexecve_args { 215 int fd; 216 char **argv; 217 char **envv; 218 } 219 #endif 220 int 221 sys_fexecve(struct thread *td, struct fexecve_args *uap) 222 { 223 int error; 224 struct image_args args; 225 226 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 227 uap->argv, uap->envv); 228 if (error == 0) { 229 args.fd = uap->fd; 230 error = kern_execve(td, &args, NULL); 231 } 232 return (error); 233 } 234 235 #ifndef _SYS_SYSPROTO_H_ 236 struct __mac_execve_args { 237 char *fname; 238 char **argv; 239 char **envv; 240 struct mac *mac_p; 241 }; 242 #endif 243 244 int 245 sys___mac_execve(td, uap) 246 struct thread *td; 247 struct __mac_execve_args /* { 248 char *fname; 249 char **argv; 250 char **envv; 251 struct mac *mac_p; 252 } */ *uap; 253 { 254 #ifdef MAC 255 int error; 256 struct image_args args; 257 258 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 259 uap->argv, uap->envv); 260 if (error == 0) 261 error = kern_execve(td, &args, uap->mac_p); 262 return (error); 263 #else 264 return (ENOSYS); 265 #endif 266 } 267 268 /* 269 * XXX: kern_execve has the astonishing property of not always returning to 270 * the caller. If sufficiently bad things happen during the call to 271 * do_execve(), it can end up calling exit1(); as a result, callers must 272 * avoid doing anything which they might need to undo (e.g., allocating 273 * memory). 274 */ 275 int 276 kern_execve(td, args, mac_p) 277 struct thread *td; 278 struct image_args *args; 279 struct mac *mac_p; 280 { 281 struct proc *p = td->td_proc; 282 int error; 283 284 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 285 args->begin_envv - args->begin_argv); 286 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 287 args->endp - args->begin_envv); 288 if (p->p_flag & P_HADTHREADS) { 289 PROC_LOCK(p); 290 if (thread_single(SINGLE_BOUNDARY)) { 291 PROC_UNLOCK(p); 292 exec_free_args(args); 293 return (ERESTART); /* Try again later. */ 294 } 295 PROC_UNLOCK(p); 296 } 297 298 error = do_execve(td, args, mac_p); 299 300 if (p->p_flag & P_HADTHREADS) { 301 PROC_LOCK(p); 302 /* 303 * If success, we upgrade to SINGLE_EXIT state to 304 * force other threads to suicide. 305 */ 306 if (error == 0) 307 thread_single(SINGLE_EXIT); 308 else 309 thread_single_end(); 310 PROC_UNLOCK(p); 311 } 312 313 return (error); 314 } 315 316 /* 317 * In-kernel implementation of execve(). All arguments are assumed to be 318 * userspace pointers from the passed thread. 319 */ 320 static int 321 do_execve(td, args, mac_p) 322 struct thread *td; 323 struct image_args *args; 324 struct mac *mac_p; 325 { 326 struct proc *p = td->td_proc; 327 struct nameidata nd; 328 struct ucred *newcred = NULL, *oldcred; 329 struct uidinfo *euip; 330 register_t *stack_base; 331 int error, i; 332 struct image_params image_params, *imgp; 333 struct vattr attr; 334 int (*img_first)(struct image_params *); 335 struct pargs *oldargs = NULL, *newargs = NULL; 336 struct sigacts *oldsigacts, *newsigacts; 337 #ifdef KTRACE 338 struct vnode *tracevp = NULL; 339 struct ucred *tracecred = NULL; 340 #endif 341 struct vnode *textvp = NULL, *binvp = NULL; 342 int credential_changing; 343 int vfslocked; 344 int textset; 345 #ifdef MAC 346 struct label *interpvplabel = NULL; 347 int will_transition; 348 #endif 349 #ifdef HWPMC_HOOKS 350 struct pmckern_procexec pe; 351 #endif 352 static const char fexecv_proc_title[] = "(fexecv)"; 353 354 vfslocked = 0; 355 imgp = &image_params; 356 357 /* 358 * Lock the process and set the P_INEXEC flag to indicate that 359 * it should be left alone until we're done here. This is 360 * necessary to avoid race conditions - e.g. in ptrace() - 361 * that might allow a local user to illicitly obtain elevated 362 * privileges. 363 */ 364 PROC_LOCK(p); 365 KASSERT((p->p_flag & P_INEXEC) == 0, 366 ("%s(): process already has P_INEXEC flag", __func__)); 367 p->p_flag |= P_INEXEC; 368 PROC_UNLOCK(p); 369 370 /* 371 * Initialize part of the common data 372 */ 373 imgp->proc = p; 374 imgp->execlabel = NULL; 375 imgp->attr = &attr; 376 imgp->entry_addr = 0; 377 imgp->reloc_base = 0; 378 imgp->vmspace_destroyed = 0; 379 imgp->interpreted = 0; 380 imgp->opened = 0; 381 imgp->interpreter_name = NULL; 382 imgp->auxargs = NULL; 383 imgp->vp = NULL; 384 imgp->object = NULL; 385 imgp->firstpage = NULL; 386 imgp->ps_strings = 0; 387 imgp->auxarg_size = 0; 388 imgp->args = args; 389 imgp->execpath = imgp->freepath = NULL; 390 imgp->execpathp = 0; 391 imgp->canary = 0; 392 imgp->canarylen = 0; 393 imgp->pagesizes = 0; 394 imgp->pagesizeslen = 0; 395 imgp->stack_prot = 0; 396 397 #ifdef MAC 398 error = mac_execve_enter(imgp, mac_p); 399 if (error) 400 goto exec_fail; 401 #endif 402 403 imgp->image_header = NULL; 404 405 /* 406 * Translate the file name. namei() returns a vnode pointer 407 * in ni_vp amoung other things. 408 * 409 * XXXAUDIT: It would be desirable to also audit the name of the 410 * interpreter if this is an interpreted binary. 411 */ 412 if (args->fname != NULL) { 413 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 414 | MPSAFE | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 415 } 416 417 SDT_PROBE(proc, kernel, , exec, args->fname, 0, 0, 0, 0 ); 418 419 interpret: 420 if (args->fname != NULL) { 421 #ifdef CAPABILITY_MODE 422 /* 423 * While capability mode can't reach this point via direct 424 * path arguments to execve(), we also don't allow 425 * interpreters to be used in capability mode (for now). 426 * Catch indirect lookups and return a permissions error. 427 */ 428 if (IN_CAPABILITY_MODE(td)) { 429 error = ECAPMODE; 430 goto exec_fail; 431 } 432 #endif 433 error = namei(&nd); 434 if (error) 435 goto exec_fail; 436 437 vfslocked = NDHASGIANT(&nd); 438 binvp = nd.ni_vp; 439 imgp->vp = binvp; 440 } else { 441 AUDIT_ARG_FD(args->fd); 442 /* 443 * Some might argue that CAP_READ and/or CAP_MMAP should also 444 * be required here; such arguments will be entertained. 445 */ 446 error = fgetvp_read(td, args->fd, CAP_FEXECVE, &binvp); 447 if (error) 448 goto exec_fail; 449 vfslocked = VFS_LOCK_GIANT(binvp->v_mount); 450 vn_lock(binvp, LK_EXCLUSIVE | LK_RETRY); 451 AUDIT_ARG_VNODE1(binvp); 452 imgp->vp = binvp; 453 } 454 455 /* 456 * Check file permissions (also 'opens' file) 457 */ 458 error = exec_check_permissions(imgp); 459 if (error) 460 goto exec_fail_dealloc; 461 462 imgp->object = imgp->vp->v_object; 463 if (imgp->object != NULL) 464 vm_object_reference(imgp->object); 465 466 /* 467 * Set VV_TEXT now so no one can write to the executable while we're 468 * activating it. 469 * 470 * Remember if this was set before and unset it in case this is not 471 * actually an executable image. 472 */ 473 textset = imgp->vp->v_vflag & VV_TEXT; 474 imgp->vp->v_vflag |= VV_TEXT; 475 476 error = exec_map_first_page(imgp); 477 if (error) 478 goto exec_fail_dealloc; 479 480 imgp->proc->p_osrel = 0; 481 /* 482 * If the current process has a special image activator it 483 * wants to try first, call it. For example, emulating shell 484 * scripts differently. 485 */ 486 error = -1; 487 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 488 error = img_first(imgp); 489 490 /* 491 * Loop through the list of image activators, calling each one. 492 * An activator returns -1 if there is no match, 0 on success, 493 * and an error otherwise. 494 */ 495 for (i = 0; error == -1 && execsw[i]; ++i) { 496 if (execsw[i]->ex_imgact == NULL || 497 execsw[i]->ex_imgact == img_first) { 498 continue; 499 } 500 error = (*execsw[i]->ex_imgact)(imgp); 501 } 502 503 if (error) { 504 if (error == -1) { 505 if (textset == 0) 506 imgp->vp->v_vflag &= ~VV_TEXT; 507 error = ENOEXEC; 508 } 509 goto exec_fail_dealloc; 510 } 511 512 /* 513 * Special interpreter operation, cleanup and loop up to try to 514 * activate the interpreter. 515 */ 516 if (imgp->interpreted) { 517 exec_unmap_first_page(imgp); 518 /* 519 * VV_TEXT needs to be unset for scripts. There is a short 520 * period before we determine that something is a script where 521 * VV_TEXT will be set. The vnode lock is held over this 522 * entire period so nothing should illegitimately be blocked. 523 */ 524 imgp->vp->v_vflag &= ~VV_TEXT; 525 /* free name buffer and old vnode */ 526 if (args->fname != NULL) 527 NDFREE(&nd, NDF_ONLY_PNBUF); 528 #ifdef MAC 529 mac_execve_interpreter_enter(binvp, &interpvplabel); 530 #endif 531 if (imgp->opened) { 532 VOP_CLOSE(binvp, FREAD, td->td_ucred, td); 533 imgp->opened = 0; 534 } 535 vput(binvp); 536 vm_object_deallocate(imgp->object); 537 imgp->object = NULL; 538 VFS_UNLOCK_GIANT(vfslocked); 539 vfslocked = 0; 540 /* set new name to that of the interpreter */ 541 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME | MPSAFE, 542 UIO_SYSSPACE, imgp->interpreter_name, td); 543 args->fname = imgp->interpreter_name; 544 goto interpret; 545 } 546 547 /* 548 * NB: We unlock the vnode here because it is believed that none 549 * of the sv_copyout_strings/sv_fixup operations require the vnode. 550 */ 551 VOP_UNLOCK(imgp->vp, 0); 552 553 /* 554 * Do the best to calculate the full path to the image file. 555 */ 556 if (imgp->auxargs != NULL && 557 ((args->fname != NULL && args->fname[0] == '/') || 558 vn_fullpath(td, imgp->vp, &imgp->execpath, &imgp->freepath) != 0)) 559 imgp->execpath = args->fname; 560 561 /* 562 * Copy out strings (args and env) and initialize stack base 563 */ 564 if (p->p_sysent->sv_copyout_strings) 565 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 566 else 567 stack_base = exec_copyout_strings(imgp); 568 569 /* 570 * If custom stack fixup routine present for this process 571 * let it do the stack setup. 572 * Else stuff argument count as first item on stack 573 */ 574 if (p->p_sysent->sv_fixup != NULL) 575 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 576 else 577 suword(--stack_base, imgp->args->argc); 578 579 /* 580 * For security and other reasons, the file descriptor table cannot 581 * be shared after an exec. 582 */ 583 fdunshare(p, td); 584 585 /* 586 * Malloc things before we need locks. 587 */ 588 newcred = crget(); 589 euip = uifind(attr.va_uid); 590 i = imgp->args->begin_envv - imgp->args->begin_argv; 591 /* Cache arguments if they fit inside our allowance */ 592 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 593 newargs = pargs_alloc(i); 594 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 595 } 596 597 /* close files on exec */ 598 fdcloseexec(td); 599 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 600 601 /* Get a reference to the vnode prior to locking the proc */ 602 VREF(binvp); 603 604 /* 605 * For security and other reasons, signal handlers cannot 606 * be shared after an exec. The new process gets a copy of the old 607 * handlers. In execsigs(), the new process will have its signals 608 * reset. 609 */ 610 PROC_LOCK(p); 611 oldcred = crcopysafe(p, newcred); 612 if (sigacts_shared(p->p_sigacts)) { 613 oldsigacts = p->p_sigacts; 614 PROC_UNLOCK(p); 615 newsigacts = sigacts_alloc(); 616 sigacts_copy(newsigacts, oldsigacts); 617 PROC_LOCK(p); 618 p->p_sigacts = newsigacts; 619 } else 620 oldsigacts = NULL; 621 622 /* Stop profiling */ 623 stopprofclock(p); 624 625 /* reset caught signals */ 626 execsigs(p); 627 628 /* name this process - nameiexec(p, ndp) */ 629 bzero(p->p_comm, sizeof(p->p_comm)); 630 if (args->fname) 631 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 632 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 633 else if (vn_commname(binvp, p->p_comm, sizeof(p->p_comm)) != 0) 634 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 635 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 636 637 /* 638 * mark as execed, wakeup the process that vforked (if any) and tell 639 * it that it now has its own resources back 640 */ 641 p->p_flag |= P_EXEC; 642 if (p->p_pptr && (p->p_flag & P_PPWAIT)) { 643 p->p_flag &= ~P_PPWAIT; 644 cv_broadcast(&p->p_pwait); 645 } 646 647 /* 648 * Implement image setuid/setgid. 649 * 650 * Don't honor setuid/setgid if the filesystem prohibits it or if 651 * the process is being traced. 652 * 653 * We disable setuid/setgid/etc in compatibility mode on the basis 654 * that most setugid applications are not written with that 655 * environment in mind, and will therefore almost certainly operate 656 * incorrectly. In principle there's no reason that setugid 657 * applications might not be useful in capability mode, so we may want 658 * to reconsider this conservative design choice in the future. 659 * 660 * XXXMAC: For the time being, use NOSUID to also prohibit 661 * transitions on the file system. 662 */ 663 credential_changing = 0; 664 credential_changing |= (attr.va_mode & S_ISUID) && oldcred->cr_uid != 665 attr.va_uid; 666 credential_changing |= (attr.va_mode & S_ISGID) && oldcred->cr_gid != 667 attr.va_gid; 668 #ifdef MAC 669 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 670 interpvplabel, imgp); 671 credential_changing |= will_transition; 672 #endif 673 674 if (credential_changing && 675 #ifdef CAPABILITY_MODE 676 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 677 #endif 678 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 679 (p->p_flag & P_TRACED) == 0) { 680 /* 681 * Turn off syscall tracing for set-id programs, except for 682 * root. Record any set-id flags first to make sure that 683 * we do not regain any tracing during a possible block. 684 */ 685 setsugid(p); 686 687 #ifdef KTRACE 688 if (priv_check_cred(oldcred, PRIV_DEBUG_DIFFCRED, 0)) 689 ktrprocexec(p, &tracecred, &tracevp); 690 #endif 691 /* 692 * Close any file descriptors 0..2 that reference procfs, 693 * then make sure file descriptors 0..2 are in use. 694 * 695 * setugidsafety() may call closef() and then pfind() 696 * which may grab the process lock. 697 * fdcheckstd() may call falloc() which may block to 698 * allocate memory, so temporarily drop the process lock. 699 */ 700 PROC_UNLOCK(p); 701 VOP_UNLOCK(imgp->vp, 0); 702 setugidsafety(td); 703 error = fdcheckstd(td); 704 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 705 if (error != 0) 706 goto done1; 707 PROC_LOCK(p); 708 /* 709 * Set the new credentials. 710 */ 711 if (attr.va_mode & S_ISUID) 712 change_euid(newcred, euip); 713 if (attr.va_mode & S_ISGID) 714 change_egid(newcred, attr.va_gid); 715 #ifdef MAC 716 if (will_transition) { 717 mac_vnode_execve_transition(oldcred, newcred, imgp->vp, 718 interpvplabel, imgp); 719 } 720 #endif 721 /* 722 * Implement correct POSIX saved-id behavior. 723 * 724 * XXXMAC: Note that the current logic will save the 725 * uid and gid if a MAC domain transition occurs, even 726 * though maybe it shouldn't. 727 */ 728 change_svuid(newcred, newcred->cr_uid); 729 change_svgid(newcred, newcred->cr_gid); 730 p->p_ucred = newcred; 731 newcred = NULL; 732 } else { 733 if (oldcred->cr_uid == oldcred->cr_ruid && 734 oldcred->cr_gid == oldcred->cr_rgid) 735 p->p_flag &= ~P_SUGID; 736 /* 737 * Implement correct POSIX saved-id behavior. 738 * 739 * XXX: It's not clear that the existing behavior is 740 * POSIX-compliant. A number of sources indicate that the 741 * saved uid/gid should only be updated if the new ruid is 742 * not equal to the old ruid, or the new euid is not equal 743 * to the old euid and the new euid is not equal to the old 744 * ruid. The FreeBSD code always updates the saved uid/gid. 745 * Also, this code uses the new (replaced) euid and egid as 746 * the source, which may or may not be the right ones to use. 747 */ 748 if (oldcred->cr_svuid != oldcred->cr_uid || 749 oldcred->cr_svgid != oldcred->cr_gid) { 750 change_svuid(newcred, newcred->cr_uid); 751 change_svgid(newcred, newcred->cr_gid); 752 p->p_ucred = newcred; 753 newcred = NULL; 754 } 755 } 756 757 /* 758 * Store the vp for use in procfs. This vnode was referenced prior 759 * to locking the proc lock. 760 */ 761 textvp = p->p_textvp; 762 p->p_textvp = binvp; 763 764 #ifdef KDTRACE_HOOKS 765 /* 766 * Tell the DTrace fasttrap provider about the exec if it 767 * has declared an interest. 768 */ 769 if (dtrace_fasttrap_exec) 770 dtrace_fasttrap_exec(p); 771 #endif 772 773 /* 774 * Notify others that we exec'd, and clear the P_INEXEC flag 775 * as we're now a bona fide freshly-execed process. 776 */ 777 KNOTE_LOCKED(&p->p_klist, NOTE_EXEC); 778 p->p_flag &= ~P_INEXEC; 779 780 /* clear "fork but no exec" flag, as we _are_ execing */ 781 p->p_acflag &= ~AFORK; 782 783 /* 784 * Free any previous argument cache and replace it with 785 * the new argument cache, if any. 786 */ 787 oldargs = p->p_args; 788 p->p_args = newargs; 789 newargs = NULL; 790 791 #ifdef HWPMC_HOOKS 792 /* 793 * Check if system-wide sampling is in effect or if the 794 * current process is using PMCs. If so, do exec() time 795 * processing. This processing needs to happen AFTER the 796 * P_INEXEC flag is cleared. 797 * 798 * The proc lock needs to be released before taking the PMC 799 * SX. 800 */ 801 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 802 PROC_UNLOCK(p); 803 VOP_UNLOCK(imgp->vp, 0); 804 pe.pm_credentialschanged = credential_changing; 805 pe.pm_entryaddr = imgp->entry_addr; 806 807 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 808 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 809 } else 810 PROC_UNLOCK(p); 811 #else /* !HWPMC_HOOKS */ 812 PROC_UNLOCK(p); 813 #endif 814 815 /* Set values passed into the program in registers. */ 816 if (p->p_sysent->sv_setregs) 817 (*p->p_sysent->sv_setregs)(td, imgp, 818 (u_long)(uintptr_t)stack_base); 819 else 820 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 821 822 vfs_mark_atime(imgp->vp, td->td_ucred); 823 824 SDT_PROBE(proc, kernel, , exec_success, args->fname, 0, 0, 0, 0); 825 826 done1: 827 /* 828 * Free any resources malloc'd earlier that we didn't use. 829 */ 830 uifree(euip); 831 if (newcred == NULL) 832 crfree(oldcred); 833 else 834 crfree(newcred); 835 VOP_UNLOCK(imgp->vp, 0); 836 837 /* 838 * Handle deferred decrement of ref counts. 839 */ 840 if (textvp != NULL) { 841 int tvfslocked; 842 843 tvfslocked = VFS_LOCK_GIANT(textvp->v_mount); 844 vrele(textvp); 845 VFS_UNLOCK_GIANT(tvfslocked); 846 } 847 if (binvp && error != 0) 848 vrele(binvp); 849 #ifdef KTRACE 850 if (tracevp != NULL) { 851 int tvfslocked; 852 853 tvfslocked = VFS_LOCK_GIANT(tracevp->v_mount); 854 vrele(tracevp); 855 VFS_UNLOCK_GIANT(tvfslocked); 856 } 857 if (tracecred != NULL) 858 crfree(tracecred); 859 #endif 860 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 861 pargs_drop(oldargs); 862 pargs_drop(newargs); 863 if (oldsigacts != NULL) 864 sigacts_free(oldsigacts); 865 866 exec_fail_dealloc: 867 868 /* 869 * free various allocated resources 870 */ 871 if (imgp->firstpage != NULL) 872 exec_unmap_first_page(imgp); 873 874 if (imgp->vp != NULL) { 875 if (args->fname) 876 NDFREE(&nd, NDF_ONLY_PNBUF); 877 if (imgp->opened) 878 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 879 vput(imgp->vp); 880 } 881 882 if (imgp->object != NULL) 883 vm_object_deallocate(imgp->object); 884 885 free(imgp->freepath, M_TEMP); 886 887 if (error == 0) { 888 PROC_LOCK(p); 889 td->td_dbgflags |= TDB_EXEC; 890 PROC_UNLOCK(p); 891 892 /* 893 * Stop the process here if its stop event mask has 894 * the S_EXEC bit set. 895 */ 896 STOPEVENT(p, S_EXEC, 0); 897 goto done2; 898 } 899 900 exec_fail: 901 /* we're done here, clear P_INEXEC */ 902 PROC_LOCK(p); 903 p->p_flag &= ~P_INEXEC; 904 PROC_UNLOCK(p); 905 906 SDT_PROBE(proc, kernel, , exec_failure, error, 0, 0, 0, 0); 907 908 done2: 909 #ifdef MAC 910 mac_execve_exit(imgp); 911 mac_execve_interpreter_exit(interpvplabel); 912 #endif 913 VFS_UNLOCK_GIANT(vfslocked); 914 exec_free_args(args); 915 916 if (error && imgp->vmspace_destroyed) { 917 /* sorry, no more process anymore. exit gracefully */ 918 exit1(td, W_EXITCODE(0, SIGABRT)); 919 /* NOT REACHED */ 920 } 921 922 #ifdef KTRACE 923 if (error == 0) 924 ktrprocctor(p); 925 #endif 926 927 return (error); 928 } 929 930 int 931 exec_map_first_page(imgp) 932 struct image_params *imgp; 933 { 934 int rv, i; 935 int initial_pagein; 936 vm_page_t ma[VM_INITIAL_PAGEIN]; 937 vm_object_t object; 938 939 if (imgp->firstpage != NULL) 940 exec_unmap_first_page(imgp); 941 942 object = imgp->vp->v_object; 943 if (object == NULL) 944 return (EACCES); 945 VM_OBJECT_LOCK(object); 946 #if VM_NRESERVLEVEL > 0 947 if ((object->flags & OBJ_COLORED) == 0) { 948 object->flags |= OBJ_COLORED; 949 object->pg_color = 0; 950 } 951 #endif 952 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_RETRY); 953 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 954 initial_pagein = VM_INITIAL_PAGEIN; 955 if (initial_pagein > object->size) 956 initial_pagein = object->size; 957 for (i = 1; i < initial_pagein; i++) { 958 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 959 if (ma[i]->valid) 960 break; 961 if ((ma[i]->oflags & VPO_BUSY) || ma[i]->busy) 962 break; 963 vm_page_busy(ma[i]); 964 } else { 965 ma[i] = vm_page_alloc(object, i, 966 VM_ALLOC_NORMAL | VM_ALLOC_IFNOTCACHED); 967 if (ma[i] == NULL) 968 break; 969 } 970 } 971 initial_pagein = i; 972 rv = vm_pager_get_pages(object, ma, initial_pagein, 0); 973 ma[0] = vm_page_lookup(object, 0); 974 if ((rv != VM_PAGER_OK) || (ma[0] == NULL)) { 975 if (ma[0] != NULL) { 976 vm_page_lock(ma[0]); 977 vm_page_free(ma[0]); 978 vm_page_unlock(ma[0]); 979 } 980 VM_OBJECT_UNLOCK(object); 981 return (EIO); 982 } 983 } 984 vm_page_lock(ma[0]); 985 vm_page_hold(ma[0]); 986 vm_page_unlock(ma[0]); 987 vm_page_wakeup(ma[0]); 988 VM_OBJECT_UNLOCK(object); 989 990 imgp->firstpage = sf_buf_alloc(ma[0], 0); 991 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 992 993 return (0); 994 } 995 996 void 997 exec_unmap_first_page(imgp) 998 struct image_params *imgp; 999 { 1000 vm_page_t m; 1001 1002 if (imgp->firstpage != NULL) { 1003 m = sf_buf_page(imgp->firstpage); 1004 sf_buf_free(imgp->firstpage); 1005 imgp->firstpage = NULL; 1006 vm_page_lock(m); 1007 vm_page_unhold(m); 1008 vm_page_unlock(m); 1009 } 1010 } 1011 1012 /* 1013 * Destroy old address space, and allocate a new stack 1014 * The new stack is only SGROWSIZ large because it is grown 1015 * automatically in trap.c. 1016 */ 1017 int 1018 exec_new_vmspace(imgp, sv) 1019 struct image_params *imgp; 1020 struct sysentvec *sv; 1021 { 1022 int error; 1023 struct proc *p = imgp->proc; 1024 struct vmspace *vmspace = p->p_vmspace; 1025 vm_object_t obj; 1026 vm_offset_t sv_minuser, stack_addr; 1027 vm_map_t map; 1028 u_long ssiz; 1029 1030 imgp->vmspace_destroyed = 1; 1031 imgp->sysent = sv; 1032 1033 /* May be called with Giant held */ 1034 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1035 1036 /* 1037 * Blow away entire process VM, if address space not shared, 1038 * otherwise, create a new VM space so that other threads are 1039 * not disrupted 1040 */ 1041 map = &vmspace->vm_map; 1042 if (map_at_zero) 1043 sv_minuser = sv->sv_minuser; 1044 else 1045 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1046 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1047 vm_map_max(map) == sv->sv_maxuser) { 1048 shmexit(vmspace); 1049 pmap_remove_pages(vmspace_pmap(vmspace)); 1050 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1051 } else { 1052 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1053 if (error) 1054 return (error); 1055 vmspace = p->p_vmspace; 1056 map = &vmspace->vm_map; 1057 } 1058 1059 /* Map a shared page */ 1060 obj = sv->sv_shared_page_obj; 1061 if (obj != NULL) { 1062 vm_object_reference(obj); 1063 error = vm_map_fixed(map, obj, 0, 1064 sv->sv_shared_page_base, sv->sv_shared_page_len, 1065 VM_PROT_READ | VM_PROT_EXECUTE, VM_PROT_ALL, 1066 MAP_COPY_ON_WRITE | MAP_ACC_NO_CHARGE); 1067 if (error) { 1068 vm_object_deallocate(obj); 1069 return (error); 1070 } 1071 } 1072 1073 /* Allocate a new stack */ 1074 if (sv->sv_maxssiz != NULL) 1075 ssiz = *sv->sv_maxssiz; 1076 else 1077 ssiz = maxssiz; 1078 stack_addr = sv->sv_usrstack - ssiz; 1079 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1080 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1081 sv->sv_stackprot, 1082 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1083 if (error) 1084 return (error); 1085 1086 #ifdef __ia64__ 1087 /* Allocate a new register stack */ 1088 stack_addr = IA64_BACKINGSTORE; 1089 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1090 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_UP); 1091 if (error) 1092 return (error); 1093 #endif 1094 1095 /* vm_ssize and vm_maxsaddr are somewhat antiquated concepts in the 1096 * VM_STACK case, but they are still used to monitor the size of the 1097 * process stack so we can check the stack rlimit. 1098 */ 1099 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1100 vmspace->vm_maxsaddr = (char *)sv->sv_usrstack - ssiz; 1101 1102 return (0); 1103 } 1104 1105 /* 1106 * Copy out argument and environment strings from the old process address 1107 * space into the temporary string buffer. 1108 */ 1109 int 1110 exec_copyin_args(struct image_args *args, char *fname, 1111 enum uio_seg segflg, char **argv, char **envv) 1112 { 1113 char *argp, *envp; 1114 int error; 1115 size_t length; 1116 1117 bzero(args, sizeof(*args)); 1118 if (argv == NULL) 1119 return (EFAULT); 1120 1121 /* 1122 * Allocate demand-paged memory for the file name, argument, and 1123 * environment strings. 1124 */ 1125 error = exec_alloc_args(args); 1126 if (error != 0) 1127 return (error); 1128 1129 /* 1130 * Copy the file name. 1131 */ 1132 if (fname != NULL) { 1133 args->fname = args->buf; 1134 error = (segflg == UIO_SYSSPACE) ? 1135 copystr(fname, args->fname, PATH_MAX, &length) : 1136 copyinstr(fname, args->fname, PATH_MAX, &length); 1137 if (error != 0) 1138 goto err_exit; 1139 } else 1140 length = 0; 1141 1142 args->begin_argv = args->buf + length; 1143 args->endp = args->begin_argv; 1144 args->stringspace = ARG_MAX; 1145 1146 /* 1147 * extract arguments first 1148 */ 1149 while ((argp = (caddr_t) (intptr_t) fuword(argv++))) { 1150 if (argp == (caddr_t) -1) { 1151 error = EFAULT; 1152 goto err_exit; 1153 } 1154 if ((error = copyinstr(argp, args->endp, 1155 args->stringspace, &length))) { 1156 if (error == ENAMETOOLONG) 1157 error = E2BIG; 1158 goto err_exit; 1159 } 1160 args->stringspace -= length; 1161 args->endp += length; 1162 args->argc++; 1163 } 1164 1165 args->begin_envv = args->endp; 1166 1167 /* 1168 * extract environment strings 1169 */ 1170 if (envv) { 1171 while ((envp = (caddr_t)(intptr_t)fuword(envv++))) { 1172 if (envp == (caddr_t)-1) { 1173 error = EFAULT; 1174 goto err_exit; 1175 } 1176 if ((error = copyinstr(envp, args->endp, 1177 args->stringspace, &length))) { 1178 if (error == ENAMETOOLONG) 1179 error = E2BIG; 1180 goto err_exit; 1181 } 1182 args->stringspace -= length; 1183 args->endp += length; 1184 args->envc++; 1185 } 1186 } 1187 1188 return (0); 1189 1190 err_exit: 1191 exec_free_args(args); 1192 return (error); 1193 } 1194 1195 /* 1196 * Allocate temporary demand-paged, zero-filled memory for the file name, 1197 * argument, and environment strings. Returns zero if the allocation succeeds 1198 * and ENOMEM otherwise. 1199 */ 1200 int 1201 exec_alloc_args(struct image_args *args) 1202 { 1203 1204 args->buf = (char *)kmem_alloc_wait(exec_map, PATH_MAX + ARG_MAX); 1205 return (args->buf != NULL ? 0 : ENOMEM); 1206 } 1207 1208 void 1209 exec_free_args(struct image_args *args) 1210 { 1211 1212 if (args->buf != NULL) { 1213 kmem_free_wakeup(exec_map, (vm_offset_t)args->buf, 1214 PATH_MAX + ARG_MAX); 1215 args->buf = NULL; 1216 } 1217 if (args->fname_buf != NULL) { 1218 free(args->fname_buf, M_TEMP); 1219 args->fname_buf = NULL; 1220 } 1221 } 1222 1223 /* 1224 * Copy strings out to the new process address space, constructing new arg 1225 * and env vector tables. Return a pointer to the base so that it can be used 1226 * as the initial stack pointer. 1227 */ 1228 register_t * 1229 exec_copyout_strings(imgp) 1230 struct image_params *imgp; 1231 { 1232 int argc, envc; 1233 char **vectp; 1234 char *stringp, *destp; 1235 register_t *stack_base; 1236 struct ps_strings *arginfo; 1237 struct proc *p; 1238 size_t execpath_len; 1239 int szsigcode, szps; 1240 char canary[sizeof(long) * 8]; 1241 1242 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1243 /* 1244 * Calculate string base and vector table pointers. 1245 * Also deal with signal trampoline code for this exec type. 1246 */ 1247 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1248 execpath_len = strlen(imgp->execpath) + 1; 1249 else 1250 execpath_len = 0; 1251 p = imgp->proc; 1252 szsigcode = 0; 1253 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1254 if (p->p_sysent->sv_sigcode_base == 0) { 1255 if (p->p_sysent->sv_szsigcode != NULL) 1256 szsigcode = *(p->p_sysent->sv_szsigcode); 1257 } 1258 destp = (caddr_t)arginfo - szsigcode - SPARE_USRSPACE - 1259 roundup(execpath_len, sizeof(char *)) - 1260 roundup(sizeof(canary), sizeof(char *)) - 1261 roundup(szps, sizeof(char *)) - 1262 roundup((ARG_MAX - imgp->args->stringspace), sizeof(char *)); 1263 1264 /* 1265 * install sigcode 1266 */ 1267 if (szsigcode != 0) 1268 copyout(p->p_sysent->sv_sigcode, ((caddr_t)arginfo - 1269 szsigcode), szsigcode); 1270 1271 /* 1272 * Copy the image path for the rtld. 1273 */ 1274 if (execpath_len != 0) { 1275 imgp->execpathp = (uintptr_t)arginfo - szsigcode - execpath_len; 1276 copyout(imgp->execpath, (void *)imgp->execpathp, 1277 execpath_len); 1278 } 1279 1280 /* 1281 * Prepare the canary for SSP. 1282 */ 1283 arc4rand(canary, sizeof(canary), 0); 1284 imgp->canary = (uintptr_t)arginfo - szsigcode - execpath_len - 1285 sizeof(canary); 1286 copyout(canary, (void *)imgp->canary, sizeof(canary)); 1287 imgp->canarylen = sizeof(canary); 1288 1289 /* 1290 * Prepare the pagesizes array. 1291 */ 1292 imgp->pagesizes = (uintptr_t)arginfo - szsigcode - execpath_len - 1293 roundup(sizeof(canary), sizeof(char *)) - szps; 1294 copyout(pagesizes, (void *)imgp->pagesizes, szps); 1295 imgp->pagesizeslen = szps; 1296 1297 /* 1298 * If we have a valid auxargs ptr, prepare some room 1299 * on the stack. 1300 */ 1301 if (imgp->auxargs) { 1302 /* 1303 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1304 * lower compatibility. 1305 */ 1306 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1307 (AT_COUNT * 2); 1308 /* 1309 * The '+ 2' is for the null pointers at the end of each of 1310 * the arg and env vector sets,and imgp->auxarg_size is room 1311 * for argument of Runtime loader. 1312 */ 1313 vectp = (char **)(destp - (imgp->args->argc + 1314 imgp->args->envc + 2 + imgp->auxarg_size) 1315 * sizeof(char *)); 1316 } else { 1317 /* 1318 * The '+ 2' is for the null pointers at the end of each of 1319 * the arg and env vector sets 1320 */ 1321 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc + 2) * 1322 sizeof(char *)); 1323 } 1324 1325 /* 1326 * vectp also becomes our initial stack base 1327 */ 1328 stack_base = (register_t *)vectp; 1329 1330 stringp = imgp->args->begin_argv; 1331 argc = imgp->args->argc; 1332 envc = imgp->args->envc; 1333 1334 /* 1335 * Copy out strings - arguments and environment. 1336 */ 1337 copyout(stringp, destp, ARG_MAX - imgp->args->stringspace); 1338 1339 /* 1340 * Fill in "ps_strings" struct for ps, w, etc. 1341 */ 1342 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1343 suword32(&arginfo->ps_nargvstr, argc); 1344 1345 /* 1346 * Fill in argument portion of vector table. 1347 */ 1348 for (; argc > 0; --argc) { 1349 suword(vectp++, (long)(intptr_t)destp); 1350 while (*stringp++ != 0) 1351 destp++; 1352 destp++; 1353 } 1354 1355 /* a null vector table pointer separates the argp's from the envp's */ 1356 suword(vectp++, 0); 1357 1358 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1359 suword32(&arginfo->ps_nenvstr, envc); 1360 1361 /* 1362 * Fill in environment portion of vector table. 1363 */ 1364 for (; envc > 0; --envc) { 1365 suword(vectp++, (long)(intptr_t)destp); 1366 while (*stringp++ != 0) 1367 destp++; 1368 destp++; 1369 } 1370 1371 /* end of vector table is a null pointer */ 1372 suword(vectp, 0); 1373 1374 return (stack_base); 1375 } 1376 1377 /* 1378 * Check permissions of file to execute. 1379 * Called with imgp->vp locked. 1380 * Return 0 for success or error code on failure. 1381 */ 1382 int 1383 exec_check_permissions(imgp) 1384 struct image_params *imgp; 1385 { 1386 struct vnode *vp = imgp->vp; 1387 struct vattr *attr = imgp->attr; 1388 struct thread *td; 1389 int error; 1390 1391 td = curthread; 1392 1393 /* Get file attributes */ 1394 error = VOP_GETATTR(vp, attr, td->td_ucred); 1395 if (error) 1396 return (error); 1397 1398 #ifdef MAC 1399 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1400 if (error) 1401 return (error); 1402 #endif 1403 1404 /* 1405 * 1) Check if file execution is disabled for the filesystem that 1406 * this file resides on. 1407 * 2) Ensure that at least one execute bit is on. Otherwise, a 1408 * privileged user will always succeed, and we don't want this 1409 * to happen unless the file really is executable. 1410 * 3) Ensure that the file is a regular file. 1411 */ 1412 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1413 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1414 (attr->va_type != VREG)) 1415 return (EACCES); 1416 1417 /* 1418 * Zero length files can't be exec'd 1419 */ 1420 if (attr->va_size == 0) 1421 return (ENOEXEC); 1422 1423 /* 1424 * Check for execute permission to file based on current credentials. 1425 */ 1426 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1427 if (error) 1428 return (error); 1429 1430 /* 1431 * Check number of open-for-writes on the file and deny execution 1432 * if there are any. 1433 */ 1434 if (vp->v_writecount) 1435 return (ETXTBSY); 1436 1437 /* 1438 * Call filesystem specific open routine (which does nothing in the 1439 * general case). 1440 */ 1441 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1442 if (error == 0) 1443 imgp->opened = 1; 1444 return (error); 1445 } 1446 1447 /* 1448 * Exec handler registration 1449 */ 1450 int 1451 exec_register(execsw_arg) 1452 const struct execsw *execsw_arg; 1453 { 1454 const struct execsw **es, **xs, **newexecsw; 1455 int count = 2; /* New slot and trailing NULL */ 1456 1457 if (execsw) 1458 for (es = execsw; *es; es++) 1459 count++; 1460 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1461 if (newexecsw == NULL) 1462 return (ENOMEM); 1463 xs = newexecsw; 1464 if (execsw) 1465 for (es = execsw; *es; es++) 1466 *xs++ = *es; 1467 *xs++ = execsw_arg; 1468 *xs = NULL; 1469 if (execsw) 1470 free(execsw, M_TEMP); 1471 execsw = newexecsw; 1472 return (0); 1473 } 1474 1475 int 1476 exec_unregister(execsw_arg) 1477 const struct execsw *execsw_arg; 1478 { 1479 const struct execsw **es, **xs, **newexecsw; 1480 int count = 1; 1481 1482 if (execsw == NULL) 1483 panic("unregister with no handlers left?\n"); 1484 1485 for (es = execsw; *es; es++) { 1486 if (*es == execsw_arg) 1487 break; 1488 } 1489 if (*es == NULL) 1490 return (ENOENT); 1491 for (es = execsw; *es; es++) 1492 if (*es != execsw_arg) 1493 count++; 1494 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1495 if (newexecsw == NULL) 1496 return (ENOMEM); 1497 xs = newexecsw; 1498 for (es = execsw; *es; es++) 1499 if (*es != execsw_arg) 1500 *xs++ = *es; 1501 *xs = NULL; 1502 if (execsw) 1503 free(execsw, M_TEMP); 1504 execsw = newexecsw; 1505 return (0); 1506 } 1507 1508 static vm_object_t shared_page_obj; 1509 static int shared_page_free; 1510 1511 int 1512 shared_page_fill(int size, int align, const char *data) 1513 { 1514 vm_page_t m; 1515 struct sf_buf *s; 1516 vm_offset_t sk; 1517 int res; 1518 1519 VM_OBJECT_LOCK(shared_page_obj); 1520 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY); 1521 res = roundup(shared_page_free, align); 1522 if (res + size >= IDX_TO_OFF(shared_page_obj->size)) 1523 res = -1; 1524 else { 1525 VM_OBJECT_UNLOCK(shared_page_obj); 1526 s = sf_buf_alloc(m, SFB_DEFAULT); 1527 sk = sf_buf_kva(s); 1528 bcopy(data, (void *)(sk + res), size); 1529 shared_page_free = res + size; 1530 sf_buf_free(s); 1531 VM_OBJECT_LOCK(shared_page_obj); 1532 } 1533 vm_page_wakeup(m); 1534 VM_OBJECT_UNLOCK(shared_page_obj); 1535 return (res); 1536 } 1537 1538 static void 1539 shared_page_init(void *dummy __unused) 1540 { 1541 vm_page_t m; 1542 1543 shared_page_obj = vm_pager_allocate(OBJT_PHYS, 0, PAGE_SIZE, 1544 VM_PROT_DEFAULT, 0, NULL); 1545 VM_OBJECT_LOCK(shared_page_obj); 1546 m = vm_page_grab(shared_page_obj, 0, VM_ALLOC_RETRY | VM_ALLOC_NOBUSY | 1547 VM_ALLOC_ZERO); 1548 m->valid = VM_PAGE_BITS_ALL; 1549 VM_OBJECT_UNLOCK(shared_page_obj); 1550 } 1551 1552 SYSINIT(shp, SI_SUB_EXEC, SI_ORDER_FIRST, (sysinit_cfunc_t)shared_page_init, 1553 NULL); 1554 1555 void 1556 exec_sysvec_init(void *param) 1557 { 1558 struct sysentvec *sv; 1559 1560 sv = (struct sysentvec *)param; 1561 1562 if ((sv->sv_flags & SV_SHP) == 0) 1563 return; 1564 sv->sv_shared_page_obj = shared_page_obj; 1565 sv->sv_sigcode_base = sv->sv_shared_page_base + 1566 shared_page_fill(*(sv->sv_szsigcode), 16, sv->sv_sigcode); 1567 }
Cache object: 7c09dbcfff7ab7870d687432a49343a6
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]