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("$FreeBSD$"); 29 30 #include "opt_capsicum.h" 31 #include "opt_hwpmc_hooks.h" 32 #include "opt_ktrace.h" 33 #include "opt_vm.h" 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/acct.h> 38 #include <sys/capsicum.h> 39 #include <sys/eventhandler.h> 40 #include <sys/exec.h> 41 #include <sys/fcntl.h> 42 #include <sys/filedesc.h> 43 #include <sys/imgact.h> 44 #include <sys/imgact_elf.h> 45 #include <sys/kernel.h> 46 #include <sys/lock.h> 47 #include <sys/malloc.h> 48 #include <sys/mman.h> 49 #include <sys/mount.h> 50 #include <sys/mutex.h> 51 #include <sys/namei.h> 52 #include <sys/pioctl.h> 53 #include <sys/priv.h> 54 #include <sys/proc.h> 55 #include <sys/ptrace.h> 56 #include <sys/resourcevar.h> 57 #include <sys/rwlock.h> 58 #include <sys/sched.h> 59 #include <sys/sdt.h> 60 #include <sys/sf_buf.h> 61 #include <sys/shm.h> 62 #include <sys/signalvar.h> 63 #include <sys/smp.h> 64 #include <sys/stat.h> 65 #include <sys/syscallsubr.h> 66 #include <sys/sysctl.h> 67 #include <sys/sysent.h> 68 #include <sys/sysproto.h> 69 #include <sys/vnode.h> 70 #include <sys/wait.h> 71 #ifdef KTRACE 72 #include <sys/ktrace.h> 73 #endif 74 75 #include <vm/vm.h> 76 #include <vm/vm_param.h> 77 #include <vm/pmap.h> 78 #include <vm/vm_page.h> 79 #include <vm/vm_map.h> 80 #include <vm/vm_kern.h> 81 #include <vm/vm_extern.h> 82 #include <vm/vm_object.h> 83 #include <vm/vm_pager.h> 84 85 #ifdef HWPMC_HOOKS 86 #include <sys/pmckern.h> 87 #endif 88 89 #include <machine/reg.h> 90 91 #include <security/audit/audit.h> 92 #include <security/mac/mac_framework.h> 93 94 #ifdef KDTRACE_HOOKS 95 #include <sys/dtrace_bsd.h> 96 dtrace_execexit_func_t dtrace_fasttrap_exec; 97 #endif 98 99 SDT_PROVIDER_DECLARE(proc); 100 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 101 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 102 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 103 104 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 105 106 int coredump_pack_fileinfo = 1; 107 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 108 &coredump_pack_fileinfo, 0, 109 "Enable file path packing in 'procstat -f' coredump notes"); 110 111 int coredump_pack_vmmapinfo = 1; 112 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 113 &coredump_pack_vmmapinfo, 0, 114 "Enable file path packing in 'procstat -v' coredump notes"); 115 116 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 117 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 118 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 119 static int do_execve(struct thread *td, struct image_args *args, 120 struct mac *mac_p); 121 122 /* XXX This should be vm_size_t. */ 123 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 124 CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 125 126 /* XXX This should be vm_size_t. */ 127 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 128 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 129 130 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 131 NULL, 0, sysctl_kern_stackprot, "I", ""); 132 133 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 134 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 135 &ps_arg_cache_limit, 0, ""); 136 137 static int disallow_high_osrel; 138 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 139 &disallow_high_osrel, 0, 140 "Disallow execution of binaries built for higher version of the world"); 141 142 static int map_at_zero = 0; 143 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 144 "Permit processes to map an object at virtual address 0."); 145 146 static int 147 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 148 { 149 struct proc *p; 150 int error; 151 152 p = curproc; 153 #ifdef SCTL_MASK32 154 if (req->flags & SCTL_MASK32) { 155 unsigned int val; 156 val = (unsigned int)p->p_sysent->sv_psstrings; 157 error = SYSCTL_OUT(req, &val, sizeof(val)); 158 } else 159 #endif 160 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 161 sizeof(p->p_sysent->sv_psstrings)); 162 return error; 163 } 164 165 static int 166 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 167 { 168 struct proc *p; 169 int error; 170 171 p = curproc; 172 #ifdef SCTL_MASK32 173 if (req->flags & SCTL_MASK32) { 174 unsigned int val; 175 val = (unsigned int)p->p_sysent->sv_usrstack; 176 error = SYSCTL_OUT(req, &val, sizeof(val)); 177 } else 178 #endif 179 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 180 sizeof(p->p_sysent->sv_usrstack)); 181 return error; 182 } 183 184 static int 185 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 186 { 187 struct proc *p; 188 189 p = curproc; 190 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 191 sizeof(p->p_sysent->sv_stackprot))); 192 } 193 194 /* 195 * Each of the items is a pointer to a `const struct execsw', hence the 196 * double pointer here. 197 */ 198 static const struct execsw **execsw; 199 200 #ifndef _SYS_SYSPROTO_H_ 201 struct execve_args { 202 char *fname; 203 char **argv; 204 char **envv; 205 }; 206 #endif 207 208 int 209 sys_execve(struct thread *td, struct execve_args *uap) 210 { 211 struct image_args args; 212 struct vmspace *oldvmspace; 213 int error; 214 215 error = pre_execve(td, &oldvmspace); 216 if (error != 0) 217 return (error); 218 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 219 uap->argv, uap->envv); 220 if (error == 0) 221 error = kern_execve(td, &args, NULL); 222 post_execve(td, error, oldvmspace); 223 return (error); 224 } 225 226 #ifndef _SYS_SYSPROTO_H_ 227 struct fexecve_args { 228 int fd; 229 char **argv; 230 char **envv; 231 } 232 #endif 233 int 234 sys_fexecve(struct thread *td, struct fexecve_args *uap) 235 { 236 struct image_args args; 237 struct vmspace *oldvmspace; 238 int error; 239 240 error = pre_execve(td, &oldvmspace); 241 if (error != 0) 242 return (error); 243 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 244 uap->argv, uap->envv); 245 if (error == 0) { 246 args.fd = uap->fd; 247 error = kern_execve(td, &args, NULL); 248 } 249 post_execve(td, error, oldvmspace); 250 return (error); 251 } 252 253 #ifndef _SYS_SYSPROTO_H_ 254 struct __mac_execve_args { 255 char *fname; 256 char **argv; 257 char **envv; 258 struct mac *mac_p; 259 }; 260 #endif 261 262 int 263 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 264 { 265 #ifdef MAC 266 struct image_args args; 267 struct vmspace *oldvmspace; 268 int error; 269 270 error = pre_execve(td, &oldvmspace); 271 if (error != 0) 272 return (error); 273 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 274 uap->argv, uap->envv); 275 if (error == 0) 276 error = kern_execve(td, &args, uap->mac_p); 277 post_execve(td, error, oldvmspace); 278 return (error); 279 #else 280 return (ENOSYS); 281 #endif 282 } 283 284 int 285 pre_execve(struct thread *td, struct vmspace **oldvmspace) 286 { 287 struct proc *p; 288 int error; 289 290 KASSERT(td == curthread, ("non-current thread %p", td)); 291 error = 0; 292 p = td->td_proc; 293 if ((p->p_flag & P_HADTHREADS) != 0) { 294 PROC_LOCK(p); 295 if (thread_single(p, SINGLE_BOUNDARY) != 0) 296 error = ERESTART; 297 PROC_UNLOCK(p); 298 } 299 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 300 ("nested execve")); 301 *oldvmspace = p->p_vmspace; 302 return (error); 303 } 304 305 void 306 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 307 { 308 struct proc *p; 309 310 KASSERT(td == curthread, ("non-current thread %p", td)); 311 p = td->td_proc; 312 if ((p->p_flag & P_HADTHREADS) != 0) { 313 PROC_LOCK(p); 314 /* 315 * If success, we upgrade to SINGLE_EXIT state to 316 * force other threads to suicide. 317 */ 318 if (error == 0) 319 thread_single(p, SINGLE_EXIT); 320 else 321 thread_single_end(p, SINGLE_BOUNDARY); 322 PROC_UNLOCK(p); 323 } 324 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 325 KASSERT(p->p_vmspace != oldvmspace, 326 ("oldvmspace still used")); 327 vmspace_free(oldvmspace); 328 td->td_pflags &= ~TDP_EXECVMSPC; 329 } 330 } 331 332 /* 333 * XXX: kern_execve has the astonishing property of not always returning to 334 * the caller. If sufficiently bad things happen during the call to 335 * do_execve(), it can end up calling exit1(); as a result, callers must 336 * avoid doing anything which they might need to undo (e.g., allocating 337 * memory). 338 */ 339 int 340 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 341 { 342 343 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 344 args->begin_envv - args->begin_argv); 345 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 346 args->endp - args->begin_envv); 347 return (do_execve(td, args, mac_p)); 348 } 349 350 /* 351 * In-kernel implementation of execve(). All arguments are assumed to be 352 * userspace pointers from the passed thread. 353 */ 354 static int 355 do_execve(td, args, mac_p) 356 struct thread *td; 357 struct image_args *args; 358 struct mac *mac_p; 359 { 360 struct proc *p = td->td_proc; 361 struct nameidata nd; 362 struct ucred *oldcred; 363 struct uidinfo *euip = NULL; 364 register_t *stack_base; 365 int error, i; 366 struct image_params image_params, *imgp; 367 struct vattr attr; 368 int (*img_first)(struct image_params *); 369 struct pargs *oldargs = NULL, *newargs = NULL; 370 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 371 #ifdef KTRACE 372 struct vnode *tracevp = NULL; 373 struct ucred *tracecred = NULL; 374 #endif 375 struct vnode *oldtextvp = NULL, *newtextvp; 376 cap_rights_t rights; 377 int credential_changing; 378 int textset; 379 #ifdef MAC 380 struct label *interpvplabel = NULL; 381 int will_transition; 382 #endif 383 #ifdef HWPMC_HOOKS 384 struct pmckern_procexec pe; 385 #endif 386 static const char fexecv_proc_title[] = "(fexecv)"; 387 388 imgp = &image_params; 389 390 /* 391 * Lock the process and set the P_INEXEC flag to indicate that 392 * it should be left alone until we're done here. This is 393 * necessary to avoid race conditions - e.g. in ptrace() - 394 * that might allow a local user to illicitly obtain elevated 395 * privileges. 396 */ 397 PROC_LOCK(p); 398 KASSERT((p->p_flag & P_INEXEC) == 0, 399 ("%s(): process already has P_INEXEC flag", __func__)); 400 p->p_flag |= P_INEXEC; 401 PROC_UNLOCK(p); 402 403 /* 404 * Initialize part of the common data 405 */ 406 bzero(imgp, sizeof(*imgp)); 407 imgp->proc = p; 408 imgp->attr = &attr; 409 imgp->args = args; 410 oldcred = p->p_ucred; 411 412 #ifdef MAC 413 error = mac_execve_enter(imgp, mac_p); 414 if (error) 415 goto exec_fail; 416 #endif 417 418 /* 419 * Translate the file name. namei() returns a vnode pointer 420 * in ni_vp among other things. 421 * 422 * XXXAUDIT: It would be desirable to also audit the name of the 423 * interpreter if this is an interpreted binary. 424 */ 425 if (args->fname != NULL) { 426 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 427 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 428 } 429 430 SDT_PROBE1(proc, , , exec, args->fname); 431 432 interpret: 433 if (args->fname != NULL) { 434 #ifdef CAPABILITY_MODE 435 /* 436 * While capability mode can't reach this point via direct 437 * path arguments to execve(), we also don't allow 438 * interpreters to be used in capability mode (for now). 439 * Catch indirect lookups and return a permissions error. 440 */ 441 if (IN_CAPABILITY_MODE(td)) { 442 error = ECAPMODE; 443 goto exec_fail; 444 } 445 #endif 446 error = namei(&nd); 447 if (error) 448 goto exec_fail; 449 450 newtextvp = nd.ni_vp; 451 imgp->vp = newtextvp; 452 } else { 453 AUDIT_ARG_FD(args->fd); 454 /* 455 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 456 */ 457 error = fgetvp_exec(td, args->fd, 458 cap_rights_init(&rights, CAP_FEXECVE), &newtextvp); 459 if (error) 460 goto exec_fail; 461 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY); 462 AUDIT_ARG_VNODE1(newtextvp); 463 imgp->vp = newtextvp; 464 } 465 466 /* 467 * Check file permissions (also 'opens' file) 468 */ 469 error = exec_check_permissions(imgp); 470 if (error) 471 goto exec_fail_dealloc; 472 473 imgp->object = imgp->vp->v_object; 474 if (imgp->object != NULL) 475 vm_object_reference(imgp->object); 476 477 /* 478 * Set VV_TEXT now so no one can write to the executable while we're 479 * activating it. 480 * 481 * Remember if this was set before and unset it in case this is not 482 * actually an executable image. 483 */ 484 textset = VOP_IS_TEXT(imgp->vp); 485 VOP_SET_TEXT(imgp->vp); 486 487 error = exec_map_first_page(imgp); 488 if (error) 489 goto exec_fail_dealloc; 490 491 imgp->proc->p_osrel = 0; 492 493 /* 494 * Implement image setuid/setgid. 495 * 496 * Determine new credentials before attempting image activators 497 * so that it can be used by process_exec handlers to determine 498 * credential/setid changes. 499 * 500 * Don't honor setuid/setgid if the filesystem prohibits it or if 501 * the process is being traced. 502 * 503 * We disable setuid/setgid/etc in capability mode on the basis 504 * that most setugid applications are not written with that 505 * environment in mind, and will therefore almost certainly operate 506 * incorrectly. In principle there's no reason that setugid 507 * applications might not be useful in capability mode, so we may want 508 * to reconsider this conservative design choice in the future. 509 * 510 * XXXMAC: For the time being, use NOSUID to also prohibit 511 * transitions on the file system. 512 */ 513 credential_changing = 0; 514 credential_changing |= (attr.va_mode & S_ISUID) && 515 oldcred->cr_uid != attr.va_uid; 516 credential_changing |= (attr.va_mode & S_ISGID) && 517 oldcred->cr_gid != attr.va_gid; 518 #ifdef MAC 519 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 520 interpvplabel, imgp); 521 credential_changing |= will_transition; 522 #endif 523 524 if (credential_changing && 525 #ifdef CAPABILITY_MODE 526 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 527 #endif 528 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 529 (p->p_flag & P_TRACED) == 0) { 530 imgp->credential_setid = true; 531 VOP_UNLOCK(imgp->vp, 0); 532 imgp->newcred = crdup(oldcred); 533 if (attr.va_mode & S_ISUID) { 534 euip = uifind(attr.va_uid); 535 change_euid(imgp->newcred, euip); 536 } 537 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 538 if (attr.va_mode & S_ISGID) 539 change_egid(imgp->newcred, attr.va_gid); 540 /* 541 * Implement correct POSIX saved-id behavior. 542 * 543 * XXXMAC: Note that the current logic will save the 544 * uid and gid if a MAC domain transition occurs, even 545 * though maybe it shouldn't. 546 */ 547 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 548 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 549 } else { 550 /* 551 * Implement correct POSIX saved-id behavior. 552 * 553 * XXX: It's not clear that the existing behavior is 554 * POSIX-compliant. A number of sources indicate that the 555 * saved uid/gid should only be updated if the new ruid is 556 * not equal to the old ruid, or the new euid is not equal 557 * to the old euid and the new euid is not equal to the old 558 * ruid. The FreeBSD code always updates the saved uid/gid. 559 * Also, this code uses the new (replaced) euid and egid as 560 * the source, which may or may not be the right ones to use. 561 */ 562 if (oldcred->cr_svuid != oldcred->cr_uid || 563 oldcred->cr_svgid != oldcred->cr_gid) { 564 VOP_UNLOCK(imgp->vp, 0); 565 imgp->newcred = crdup(oldcred); 566 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 567 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 568 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 569 } 570 } 571 /* The new credentials are installed into the process later. */ 572 573 /* 574 * Do the best to calculate the full path to the image file. 575 */ 576 if (args->fname != NULL && args->fname[0] == '/') 577 imgp->execpath = args->fname; 578 else { 579 VOP_UNLOCK(imgp->vp, 0); 580 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 581 &imgp->freepath) != 0) 582 imgp->execpath = args->fname; 583 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 584 } 585 586 /* 587 * If the current process has a special image activator it 588 * wants to try first, call it. For example, emulating shell 589 * scripts differently. 590 */ 591 error = -1; 592 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 593 error = img_first(imgp); 594 595 /* 596 * Loop through the list of image activators, calling each one. 597 * An activator returns -1 if there is no match, 0 on success, 598 * and an error otherwise. 599 */ 600 for (i = 0; error == -1 && execsw[i]; ++i) { 601 if (execsw[i]->ex_imgact == NULL || 602 execsw[i]->ex_imgact == img_first) { 603 continue; 604 } 605 error = (*execsw[i]->ex_imgact)(imgp); 606 } 607 608 if (error) { 609 if (error == -1) { 610 if (textset == 0) 611 VOP_UNSET_TEXT(imgp->vp); 612 error = ENOEXEC; 613 } 614 goto exec_fail_dealloc; 615 } 616 617 /* 618 * Special interpreter operation, cleanup and loop up to try to 619 * activate the interpreter. 620 */ 621 if (imgp->interpreted) { 622 exec_unmap_first_page(imgp); 623 /* 624 * VV_TEXT needs to be unset for scripts. There is a short 625 * period before we determine that something is a script where 626 * VV_TEXT will be set. The vnode lock is held over this 627 * entire period so nothing should illegitimately be blocked. 628 */ 629 VOP_UNSET_TEXT(imgp->vp); 630 /* free name buffer and old vnode */ 631 if (args->fname != NULL) 632 NDFREE(&nd, NDF_ONLY_PNBUF); 633 #ifdef MAC 634 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 635 #endif 636 if (imgp->opened) { 637 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 638 imgp->opened = 0; 639 } 640 vput(newtextvp); 641 vm_object_deallocate(imgp->object); 642 imgp->object = NULL; 643 imgp->credential_setid = false; 644 if (imgp->newcred != NULL) { 645 crfree(imgp->newcred); 646 imgp->newcred = NULL; 647 } 648 imgp->execpath = NULL; 649 free(imgp->freepath, M_TEMP); 650 imgp->freepath = NULL; 651 /* set new name to that of the interpreter */ 652 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 653 UIO_SYSSPACE, imgp->interpreter_name, td); 654 args->fname = imgp->interpreter_name; 655 goto interpret; 656 } 657 658 /* 659 * NB: We unlock the vnode here because it is believed that none 660 * of the sv_copyout_strings/sv_fixup operations require the vnode. 661 */ 662 VOP_UNLOCK(imgp->vp, 0); 663 664 if (disallow_high_osrel && 665 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 666 error = ENOEXEC; 667 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 668 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 669 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 670 goto exec_fail_dealloc; 671 } 672 673 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 674 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 675 sys_cap_enter(td, NULL); 676 677 /* 678 * Copy out strings (args and env) and initialize stack base 679 */ 680 if (p->p_sysent->sv_copyout_strings) 681 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 682 else 683 stack_base = exec_copyout_strings(imgp); 684 685 /* 686 * If custom stack fixup routine present for this process 687 * let it do the stack setup. 688 * Else stuff argument count as first item on stack 689 */ 690 if (p->p_sysent->sv_fixup != NULL) 691 (*p->p_sysent->sv_fixup)(&stack_base, imgp); 692 else 693 suword(--stack_base, imgp->args->argc); 694 695 if (args->fdp != NULL) { 696 /* Install a brand new file descriptor table. */ 697 fdinstall_remapped(td, args->fdp); 698 args->fdp = NULL; 699 } else { 700 /* 701 * Keep on using the existing file descriptor table. For 702 * security and other reasons, the file descriptor table 703 * cannot be shared after an exec. 704 */ 705 fdunshare(td); 706 /* close files on exec */ 707 fdcloseexec(td); 708 } 709 710 /* 711 * Malloc things before we need locks. 712 */ 713 i = imgp->args->begin_envv - imgp->args->begin_argv; 714 /* Cache arguments if they fit inside our allowance */ 715 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 716 newargs = pargs_alloc(i); 717 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 718 } 719 720 /* 721 * For security and other reasons, signal handlers cannot 722 * be shared after an exec. The new process gets a copy of the old 723 * handlers. In execsigs(), the new process will have its signals 724 * reset. 725 */ 726 if (sigacts_shared(p->p_sigacts)) { 727 oldsigacts = p->p_sigacts; 728 newsigacts = sigacts_alloc(); 729 sigacts_copy(newsigacts, oldsigacts); 730 } 731 732 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 733 734 PROC_LOCK(p); 735 if (oldsigacts) 736 p->p_sigacts = newsigacts; 737 /* Stop profiling */ 738 stopprofclock(p); 739 740 /* reset caught signals */ 741 execsigs(p); 742 743 /* name this process - nameiexec(p, ndp) */ 744 bzero(p->p_comm, sizeof(p->p_comm)); 745 if (args->fname) 746 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 747 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 748 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 749 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 750 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 751 #ifdef KTR 752 sched_clear_tdname(td); 753 #endif 754 755 /* 756 * mark as execed, wakeup the process that vforked (if any) and tell 757 * it that it now has its own resources back 758 */ 759 p->p_flag |= P_EXEC; 760 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 761 p->p_flag2 &= ~P2_NOTRACE; 762 if (p->p_flag & P_PPWAIT) { 763 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 764 cv_broadcast(&p->p_pwait); 765 /* STOPs are no longer ignored, arrange for AST */ 766 signotify(td); 767 } 768 769 /* 770 * Implement image setuid/setgid installation. 771 */ 772 if (imgp->credential_setid) { 773 /* 774 * Turn off syscall tracing for set-id programs, except for 775 * root. Record any set-id flags first to make sure that 776 * we do not regain any tracing during a possible block. 777 */ 778 setsugid(p); 779 780 #ifdef KTRACE 781 if (p->p_tracecred != NULL && 782 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 783 ktrprocexec(p, &tracecred, &tracevp); 784 #endif 785 /* 786 * Close any file descriptors 0..2 that reference procfs, 787 * then make sure file descriptors 0..2 are in use. 788 * 789 * Both fdsetugidsafety() and fdcheckstd() may call functions 790 * taking sleepable locks, so temporarily drop our locks. 791 */ 792 PROC_UNLOCK(p); 793 VOP_UNLOCK(imgp->vp, 0); 794 fdsetugidsafety(td); 795 error = fdcheckstd(td); 796 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 797 if (error != 0) 798 goto exec_fail_dealloc; 799 PROC_LOCK(p); 800 #ifdef MAC 801 if (will_transition) { 802 mac_vnode_execve_transition(oldcred, imgp->newcred, 803 imgp->vp, interpvplabel, imgp); 804 } 805 #endif 806 } else { 807 if (oldcred->cr_uid == oldcred->cr_ruid && 808 oldcred->cr_gid == oldcred->cr_rgid) 809 p->p_flag &= ~P_SUGID; 810 } 811 /* 812 * Set the new credentials. 813 */ 814 if (imgp->newcred != NULL) { 815 proc_set_cred(p, imgp->newcred); 816 crfree(oldcred); 817 oldcred = NULL; 818 } 819 820 /* 821 * Store the vp for use in procfs. This vnode was referenced by namei 822 * or fgetvp_exec. 823 */ 824 oldtextvp = p->p_textvp; 825 p->p_textvp = newtextvp; 826 827 #ifdef KDTRACE_HOOKS 828 /* 829 * Tell the DTrace fasttrap provider about the exec if it 830 * has declared an interest. 831 */ 832 if (dtrace_fasttrap_exec) 833 dtrace_fasttrap_exec(p); 834 #endif 835 836 /* 837 * Notify others that we exec'd, and clear the P_INEXEC flag 838 * as we're now a bona fide freshly-execed process. 839 */ 840 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 841 p->p_flag &= ~P_INEXEC; 842 843 /* clear "fork but no exec" flag, as we _are_ execing */ 844 p->p_acflag &= ~AFORK; 845 846 /* 847 * Free any previous argument cache and replace it with 848 * the new argument cache, if any. 849 */ 850 oldargs = p->p_args; 851 p->p_args = newargs; 852 newargs = NULL; 853 854 #ifdef HWPMC_HOOKS 855 /* 856 * Check if system-wide sampling is in effect or if the 857 * current process is using PMCs. If so, do exec() time 858 * processing. This processing needs to happen AFTER the 859 * P_INEXEC flag is cleared. 860 * 861 * The proc lock needs to be released before taking the PMC 862 * SX. 863 */ 864 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 865 PROC_UNLOCK(p); 866 VOP_UNLOCK(imgp->vp, 0); 867 pe.pm_credentialschanged = credential_changing; 868 pe.pm_entryaddr = imgp->entry_addr; 869 870 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 871 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 872 } else 873 PROC_UNLOCK(p); 874 #else /* !HWPMC_HOOKS */ 875 PROC_UNLOCK(p); 876 #endif 877 878 /* Set values passed into the program in registers. */ 879 if (p->p_sysent->sv_setregs) 880 (*p->p_sysent->sv_setregs)(td, imgp, 881 (u_long)(uintptr_t)stack_base); 882 else 883 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 884 885 vfs_mark_atime(imgp->vp, td->td_ucred); 886 887 SDT_PROBE1(proc, , , exec__success, args->fname); 888 889 exec_fail_dealloc: 890 if (imgp->firstpage != NULL) 891 exec_unmap_first_page(imgp); 892 893 if (imgp->vp != NULL) { 894 if (args->fname) 895 NDFREE(&nd, NDF_ONLY_PNBUF); 896 if (imgp->opened) 897 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 898 if (error != 0) 899 vput(imgp->vp); 900 else 901 VOP_UNLOCK(imgp->vp, 0); 902 } 903 904 if (imgp->object != NULL) 905 vm_object_deallocate(imgp->object); 906 907 free(imgp->freepath, M_TEMP); 908 909 if (error == 0) { 910 PROC_LOCK(p); 911 if (p->p_ptevents & PTRACE_EXEC) 912 td->td_dbgflags |= TDB_EXEC; 913 PROC_UNLOCK(p); 914 915 /* 916 * Stop the process here if its stop event mask has 917 * the S_EXEC bit set. 918 */ 919 STOPEVENT(p, S_EXEC, 0); 920 } else { 921 exec_fail: 922 /* we're done here, clear P_INEXEC */ 923 PROC_LOCK(p); 924 p->p_flag &= ~P_INEXEC; 925 PROC_UNLOCK(p); 926 927 SDT_PROBE1(proc, , , exec__failure, error); 928 } 929 930 if (imgp->newcred != NULL && oldcred != NULL) 931 crfree(imgp->newcred); 932 933 #ifdef MAC 934 mac_execve_exit(imgp); 935 mac_execve_interpreter_exit(interpvplabel); 936 #endif 937 exec_free_args(args); 938 939 /* 940 * Handle deferred decrement of ref counts. 941 */ 942 if (oldtextvp != NULL) 943 vrele(oldtextvp); 944 #ifdef KTRACE 945 if (tracevp != NULL) 946 vrele(tracevp); 947 if (tracecred != NULL) 948 crfree(tracecred); 949 #endif 950 pargs_drop(oldargs); 951 pargs_drop(newargs); 952 if (oldsigacts != NULL) 953 sigacts_free(oldsigacts); 954 if (euip != NULL) 955 uifree(euip); 956 957 if (error && imgp->vmspace_destroyed) { 958 /* sorry, no more process anymore. exit gracefully */ 959 exit1(td, 0, SIGABRT); 960 /* NOT REACHED */ 961 } 962 963 #ifdef KTRACE 964 if (error == 0) 965 ktrprocctor(p); 966 #endif 967 968 return (error); 969 } 970 971 int 972 exec_map_first_page(imgp) 973 struct image_params *imgp; 974 { 975 int rv, i, after, initial_pagein; 976 vm_page_t ma[VM_INITIAL_PAGEIN]; 977 vm_object_t object; 978 979 if (imgp->firstpage != NULL) 980 exec_unmap_first_page(imgp); 981 982 object = imgp->vp->v_object; 983 if (object == NULL) 984 return (EACCES); 985 VM_OBJECT_WLOCK(object); 986 #if VM_NRESERVLEVEL > 0 987 vm_object_color(object, 0); 988 #endif 989 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 990 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 991 vm_page_xbusy(ma[0]); 992 if (!vm_pager_has_page(object, 0, NULL, &after)) { 993 vm_page_lock(ma[0]); 994 vm_page_free(ma[0]); 995 vm_page_unlock(ma[0]); 996 VM_OBJECT_WUNLOCK(object); 997 return (EIO); 998 } 999 initial_pagein = min(after, VM_INITIAL_PAGEIN); 1000 KASSERT(initial_pagein <= object->size, 1001 ("%s: initial_pagein %d object->size %ju", 1002 __func__, initial_pagein, (uintmax_t )object->size)); 1003 for (i = 1; i < initial_pagein; i++) { 1004 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1005 if (ma[i]->valid) 1006 break; 1007 if (vm_page_tryxbusy(ma[i])) 1008 break; 1009 } else { 1010 ma[i] = vm_page_alloc(object, i, 1011 VM_ALLOC_NORMAL); 1012 if (ma[i] == NULL) 1013 break; 1014 } 1015 } 1016 initial_pagein = i; 1017 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1018 if (rv != VM_PAGER_OK) { 1019 for (i = 0; i < initial_pagein; i++) { 1020 vm_page_lock(ma[i]); 1021 vm_page_free(ma[i]); 1022 vm_page_unlock(ma[i]); 1023 } 1024 VM_OBJECT_WUNLOCK(object); 1025 return (EIO); 1026 } 1027 vm_page_xunbusy(ma[0]); 1028 for (i = 1; i < initial_pagein; i++) 1029 vm_page_readahead_finish(ma[i]); 1030 } 1031 vm_page_lock(ma[0]); 1032 vm_page_hold(ma[0]); 1033 vm_page_activate(ma[0]); 1034 vm_page_unlock(ma[0]); 1035 VM_OBJECT_WUNLOCK(object); 1036 1037 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1038 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1039 1040 return (0); 1041 } 1042 1043 void 1044 exec_unmap_first_page(imgp) 1045 struct image_params *imgp; 1046 { 1047 vm_page_t m; 1048 1049 if (imgp->firstpage != NULL) { 1050 m = sf_buf_page(imgp->firstpage); 1051 sf_buf_free(imgp->firstpage); 1052 imgp->firstpage = NULL; 1053 vm_page_lock(m); 1054 vm_page_unhold(m); 1055 vm_page_unlock(m); 1056 } 1057 } 1058 1059 /* 1060 * Destroy old address space, and allocate a new stack 1061 * The new stack is only SGROWSIZ large because it is grown 1062 * automatically in trap.c. 1063 */ 1064 int 1065 exec_new_vmspace(imgp, sv) 1066 struct image_params *imgp; 1067 struct sysentvec *sv; 1068 { 1069 int error; 1070 struct proc *p = imgp->proc; 1071 struct vmspace *vmspace = p->p_vmspace; 1072 vm_object_t obj; 1073 struct rlimit rlim_stack; 1074 vm_offset_t sv_minuser, stack_addr; 1075 vm_map_t map; 1076 u_long ssiz; 1077 1078 imgp->vmspace_destroyed = 1; 1079 imgp->sysent = sv; 1080 1081 /* May be called with Giant held */ 1082 EVENTHANDLER_INVOKE(process_exec, p, imgp); 1083 1084 /* 1085 * Blow away entire process VM, if address space not shared, 1086 * otherwise, create a new VM space so that other threads are 1087 * not disrupted 1088 */ 1089 map = &vmspace->vm_map; 1090 if (map_at_zero) 1091 sv_minuser = sv->sv_minuser; 1092 else 1093 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1094 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1095 vm_map_max(map) == sv->sv_maxuser) { 1096 shmexit(vmspace); 1097 pmap_remove_pages(vmspace_pmap(vmspace)); 1098 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1099 } else { 1100 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1101 if (error) 1102 return (error); 1103 vmspace = p->p_vmspace; 1104 map = &vmspace->vm_map; 1105 } 1106 1107 /* Map a shared page */ 1108 obj = sv->sv_shared_page_obj; 1109 if (obj != NULL) { 1110 vm_object_reference(obj); 1111 error = vm_map_fixed(map, obj, 0, 1112 sv->sv_shared_page_base, sv->sv_shared_page_len, 1113 VM_PROT_READ | VM_PROT_EXECUTE, 1114 VM_PROT_READ | VM_PROT_EXECUTE, 1115 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1116 if (error) { 1117 vm_object_deallocate(obj); 1118 return (error); 1119 } 1120 } 1121 1122 /* Allocate a new stack */ 1123 if (imgp->stack_sz != 0) { 1124 ssiz = trunc_page(imgp->stack_sz); 1125 PROC_LOCK(p); 1126 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1127 PROC_UNLOCK(p); 1128 if (ssiz > rlim_stack.rlim_max) 1129 ssiz = rlim_stack.rlim_max; 1130 if (ssiz > rlim_stack.rlim_cur) { 1131 rlim_stack.rlim_cur = ssiz; 1132 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1133 } 1134 } else if (sv->sv_maxssiz != NULL) { 1135 ssiz = *sv->sv_maxssiz; 1136 } else { 1137 ssiz = maxssiz; 1138 } 1139 stack_addr = sv->sv_usrstack - ssiz; 1140 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1141 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1142 sv->sv_stackprot, 1143 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1144 if (error) 1145 return (error); 1146 1147 /* 1148 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1149 * are still used to enforce the stack rlimit on the process stack. 1150 */ 1151 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1152 vmspace->vm_maxsaddr = (char *)stack_addr; 1153 1154 return (0); 1155 } 1156 1157 /* 1158 * Copy out argument and environment strings from the old process address 1159 * space into the temporary string buffer. 1160 */ 1161 int 1162 exec_copyin_args(struct image_args *args, char *fname, 1163 enum uio_seg segflg, char **argv, char **envv) 1164 { 1165 u_long argp, envp; 1166 int error; 1167 size_t length; 1168 1169 bzero(args, sizeof(*args)); 1170 if (argv == NULL) 1171 return (EFAULT); 1172 1173 /* 1174 * Allocate demand-paged memory for the file name, argument, and 1175 * environment strings. 1176 */ 1177 error = exec_alloc_args(args); 1178 if (error != 0) 1179 return (error); 1180 1181 /* 1182 * Copy the file name. 1183 */ 1184 if (fname != NULL) { 1185 args->fname = args->buf; 1186 error = (segflg == UIO_SYSSPACE) ? 1187 copystr(fname, args->fname, PATH_MAX, &length) : 1188 copyinstr(fname, args->fname, PATH_MAX, &length); 1189 if (error != 0) 1190 goto err_exit; 1191 } else 1192 length = 0; 1193 1194 args->begin_argv = args->buf + length; 1195 args->endp = args->begin_argv; 1196 args->stringspace = ARG_MAX; 1197 1198 /* 1199 * extract arguments first 1200 */ 1201 for (;;) { 1202 error = fueword(argv++, &argp); 1203 if (error == -1) { 1204 error = EFAULT; 1205 goto err_exit; 1206 } 1207 if (argp == 0) 1208 break; 1209 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1210 args->stringspace, &length); 1211 if (error != 0) { 1212 if (error == ENAMETOOLONG) 1213 error = E2BIG; 1214 goto err_exit; 1215 } 1216 args->stringspace -= length; 1217 args->endp += length; 1218 args->argc++; 1219 } 1220 1221 args->begin_envv = args->endp; 1222 1223 /* 1224 * extract environment strings 1225 */ 1226 if (envv) { 1227 for (;;) { 1228 error = fueword(envv++, &envp); 1229 if (error == -1) { 1230 error = EFAULT; 1231 goto err_exit; 1232 } 1233 if (envp == 0) 1234 break; 1235 error = copyinstr((void *)(uintptr_t)envp, 1236 args->endp, args->stringspace, &length); 1237 if (error != 0) { 1238 if (error == ENAMETOOLONG) 1239 error = E2BIG; 1240 goto err_exit; 1241 } 1242 args->stringspace -= length; 1243 args->endp += length; 1244 args->envc++; 1245 } 1246 } 1247 1248 return (0); 1249 1250 err_exit: 1251 exec_free_args(args); 1252 return (error); 1253 } 1254 1255 int 1256 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1257 const void *data, size_t datalen, const int *fds, size_t fdslen) 1258 { 1259 struct filedesc *ofdp; 1260 const char *p; 1261 int *kfds; 1262 int error; 1263 1264 memset(args, '\0', sizeof(*args)); 1265 ofdp = td->td_proc->p_fd; 1266 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1267 return (E2BIG); 1268 error = exec_alloc_args(args); 1269 if (error != 0) 1270 return (error); 1271 1272 args->begin_argv = args->buf; 1273 args->stringspace = ARG_MAX; 1274 1275 if (datalen > 0) { 1276 /* 1277 * Argument buffer has been provided. Copy it into the 1278 * kernel as a single string and add a terminating null 1279 * byte. 1280 */ 1281 error = copyin(data, args->begin_argv, datalen); 1282 if (error != 0) 1283 goto err_exit; 1284 args->begin_argv[datalen] = '\0'; 1285 args->endp = args->begin_argv + datalen + 1; 1286 args->stringspace -= datalen + 1; 1287 1288 /* 1289 * Traditional argument counting. Count the number of 1290 * null bytes. 1291 */ 1292 for (p = args->begin_argv; p < args->endp; ++p) 1293 if (*p == '\0') 1294 ++args->argc; 1295 } else { 1296 /* No argument buffer provided. */ 1297 args->endp = args->begin_argv; 1298 } 1299 /* There are no environment variables. */ 1300 args->begin_envv = args->endp; 1301 1302 /* Create new file descriptor table. */ 1303 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1304 error = copyin(fds, kfds, fdslen * sizeof(int)); 1305 if (error != 0) { 1306 free(kfds, M_TEMP); 1307 goto err_exit; 1308 } 1309 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1310 free(kfds, M_TEMP); 1311 if (error != 0) 1312 goto err_exit; 1313 1314 return (0); 1315 err_exit: 1316 exec_free_args(args); 1317 return (error); 1318 } 1319 1320 struct exec_args_kva { 1321 vm_offset_t addr; 1322 SLIST_ENTRY(exec_args_kva) next; 1323 }; 1324 1325 static DPCPU_DEFINE(struct exec_args_kva *, exec_args_kva); 1326 1327 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1328 static struct mtx exec_args_kva_mtx; 1329 1330 static void 1331 exec_prealloc_args_kva(void *arg __unused) 1332 { 1333 struct exec_args_kva *argkva; 1334 u_int i; 1335 1336 SLIST_INIT(&exec_args_kva_freelist); 1337 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1338 for (i = 0; i < exec_map_entries; i++) { 1339 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1340 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1341 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1342 } 1343 } 1344 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1345 1346 static vm_offset_t 1347 exec_alloc_args_kva(void **cookie) 1348 { 1349 struct exec_args_kva *argkva; 1350 1351 argkva = (void *)atomic_readandclear_ptr( 1352 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1353 if (argkva == NULL) { 1354 mtx_lock(&exec_args_kva_mtx); 1355 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1356 (void)mtx_sleep(&exec_args_kva_freelist, 1357 &exec_args_kva_mtx, 0, "execkva", 0); 1358 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1359 mtx_unlock(&exec_args_kva_mtx); 1360 } 1361 *(struct exec_args_kva **)cookie = argkva; 1362 return (argkva->addr); 1363 } 1364 1365 static void 1366 exec_free_args_kva(void *cookie) 1367 { 1368 struct exec_args_kva *argkva; 1369 vm_offset_t base; 1370 1371 argkva = cookie; 1372 base = argkva->addr; 1373 1374 vm_map_madvise(exec_map, base, base + exec_map_entry_size, MADV_FREE); 1375 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1376 (uintptr_t)NULL, (uintptr_t)argkva)) { 1377 mtx_lock(&exec_args_kva_mtx); 1378 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1379 wakeup_one(&exec_args_kva_freelist); 1380 mtx_unlock(&exec_args_kva_mtx); 1381 } 1382 } 1383 1384 /* 1385 * Allocate temporary demand-paged, zero-filled memory for the file name, 1386 * argument, and environment strings. 1387 */ 1388 int 1389 exec_alloc_args(struct image_args *args) 1390 { 1391 1392 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1393 return (0); 1394 } 1395 1396 void 1397 exec_free_args(struct image_args *args) 1398 { 1399 1400 if (args->buf != NULL) { 1401 exec_free_args_kva(args->bufkva); 1402 args->buf = NULL; 1403 } 1404 if (args->fname_buf != NULL) { 1405 free(args->fname_buf, M_TEMP); 1406 args->fname_buf = NULL; 1407 } 1408 if (args->fdp != NULL) 1409 fdescfree_remapped(args->fdp); 1410 } 1411 1412 /* 1413 * Copy strings out to the new process address space, constructing new arg 1414 * and env vector tables. Return a pointer to the base so that it can be used 1415 * as the initial stack pointer. 1416 */ 1417 register_t * 1418 exec_copyout_strings(imgp) 1419 struct image_params *imgp; 1420 { 1421 int argc, envc; 1422 char **vectp; 1423 char *stringp; 1424 uintptr_t destp; 1425 register_t *stack_base; 1426 struct ps_strings *arginfo; 1427 struct proc *p; 1428 size_t execpath_len; 1429 int szsigcode, szps; 1430 char canary[sizeof(long) * 8]; 1431 1432 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1433 /* 1434 * Calculate string base and vector table pointers. 1435 * Also deal with signal trampoline code for this exec type. 1436 */ 1437 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1438 execpath_len = strlen(imgp->execpath) + 1; 1439 else 1440 execpath_len = 0; 1441 p = imgp->proc; 1442 szsigcode = 0; 1443 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1444 if (p->p_sysent->sv_sigcode_base == 0) { 1445 if (p->p_sysent->sv_szsigcode != NULL) 1446 szsigcode = *(p->p_sysent->sv_szsigcode); 1447 } 1448 destp = (uintptr_t)arginfo; 1449 1450 /* 1451 * install sigcode 1452 */ 1453 if (szsigcode != 0) { 1454 destp -= szsigcode; 1455 destp = rounddown2(destp, sizeof(void *)); 1456 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1457 } 1458 1459 /* 1460 * Copy the image path for the rtld. 1461 */ 1462 if (execpath_len != 0) { 1463 destp -= execpath_len; 1464 imgp->execpathp = destp; 1465 copyout(imgp->execpath, (void *)destp, execpath_len); 1466 } 1467 1468 /* 1469 * Prepare the canary for SSP. 1470 */ 1471 arc4rand(canary, sizeof(canary), 0); 1472 destp -= sizeof(canary); 1473 imgp->canary = destp; 1474 copyout(canary, (void *)destp, sizeof(canary)); 1475 imgp->canarylen = sizeof(canary); 1476 1477 /* 1478 * Prepare the pagesizes array. 1479 */ 1480 destp -= szps; 1481 destp = rounddown2(destp, sizeof(void *)); 1482 imgp->pagesizes = destp; 1483 copyout(pagesizes, (void *)destp, szps); 1484 imgp->pagesizeslen = szps; 1485 1486 destp -= ARG_MAX - imgp->args->stringspace; 1487 destp = rounddown2(destp, sizeof(void *)); 1488 1489 /* 1490 * If we have a valid auxargs ptr, prepare some room 1491 * on the stack. 1492 */ 1493 if (imgp->auxargs) { 1494 /* 1495 * 'AT_COUNT*2' is size for the ELF Auxargs data. This is for 1496 * lower compatibility. 1497 */ 1498 imgp->auxarg_size = (imgp->auxarg_size) ? imgp->auxarg_size : 1499 (AT_COUNT * 2); 1500 /* 1501 * The '+ 2' is for the null pointers at the end of each of 1502 * the arg and env vector sets,and imgp->auxarg_size is room 1503 * for argument of Runtime loader. 1504 */ 1505 vectp = (char **)(destp - (imgp->args->argc + 1506 imgp->args->envc + 2 + imgp->auxarg_size) 1507 * sizeof(char *)); 1508 } else { 1509 /* 1510 * The '+ 2' is for the null pointers at the end of each of 1511 * the arg and env vector sets 1512 */ 1513 vectp = (char **)(destp - (imgp->args->argc + imgp->args->envc 1514 + 2) * sizeof(char *)); 1515 } 1516 1517 /* 1518 * vectp also becomes our initial stack base 1519 */ 1520 stack_base = (register_t *)vectp; 1521 1522 stringp = imgp->args->begin_argv; 1523 argc = imgp->args->argc; 1524 envc = imgp->args->envc; 1525 1526 /* 1527 * Copy out strings - arguments and environment. 1528 */ 1529 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1530 1531 /* 1532 * Fill in "ps_strings" struct for ps, w, etc. 1533 */ 1534 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1535 suword32(&arginfo->ps_nargvstr, argc); 1536 1537 /* 1538 * Fill in argument portion of vector table. 1539 */ 1540 for (; argc > 0; --argc) { 1541 suword(vectp++, (long)(intptr_t)destp); 1542 while (*stringp++ != 0) 1543 destp++; 1544 destp++; 1545 } 1546 1547 /* a null vector table pointer separates the argp's from the envp's */ 1548 suword(vectp++, 0); 1549 1550 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1551 suword32(&arginfo->ps_nenvstr, envc); 1552 1553 /* 1554 * Fill in environment portion of vector table. 1555 */ 1556 for (; envc > 0; --envc) { 1557 suword(vectp++, (long)(intptr_t)destp); 1558 while (*stringp++ != 0) 1559 destp++; 1560 destp++; 1561 } 1562 1563 /* end of vector table is a null pointer */ 1564 suword(vectp, 0); 1565 1566 return (stack_base); 1567 } 1568 1569 /* 1570 * Check permissions of file to execute. 1571 * Called with imgp->vp locked. 1572 * Return 0 for success or error code on failure. 1573 */ 1574 int 1575 exec_check_permissions(imgp) 1576 struct image_params *imgp; 1577 { 1578 struct vnode *vp = imgp->vp; 1579 struct vattr *attr = imgp->attr; 1580 struct thread *td; 1581 int error, writecount; 1582 1583 td = curthread; 1584 1585 /* Get file attributes */ 1586 error = VOP_GETATTR(vp, attr, td->td_ucred); 1587 if (error) 1588 return (error); 1589 1590 #ifdef MAC 1591 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1592 if (error) 1593 return (error); 1594 #endif 1595 1596 /* 1597 * 1) Check if file execution is disabled for the filesystem that 1598 * this file resides on. 1599 * 2) Ensure that at least one execute bit is on. Otherwise, a 1600 * privileged user will always succeed, and we don't want this 1601 * to happen unless the file really is executable. 1602 * 3) Ensure that the file is a regular file. 1603 */ 1604 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1605 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1606 (attr->va_type != VREG)) 1607 return (EACCES); 1608 1609 /* 1610 * Zero length files can't be exec'd 1611 */ 1612 if (attr->va_size == 0) 1613 return (ENOEXEC); 1614 1615 /* 1616 * Check for execute permission to file based on current credentials. 1617 */ 1618 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1619 if (error) 1620 return (error); 1621 1622 /* 1623 * Check number of open-for-writes on the file and deny execution 1624 * if there are any. 1625 */ 1626 error = VOP_GET_WRITECOUNT(vp, &writecount); 1627 if (error != 0) 1628 return (error); 1629 if (writecount != 0) 1630 return (ETXTBSY); 1631 1632 /* 1633 * Call filesystem specific open routine (which does nothing in the 1634 * general case). 1635 */ 1636 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1637 if (error == 0) 1638 imgp->opened = 1; 1639 return (error); 1640 } 1641 1642 /* 1643 * Exec handler registration 1644 */ 1645 int 1646 exec_register(execsw_arg) 1647 const struct execsw *execsw_arg; 1648 { 1649 const struct execsw **es, **xs, **newexecsw; 1650 int count = 2; /* New slot and trailing NULL */ 1651 1652 if (execsw) 1653 for (es = execsw; *es; es++) 1654 count++; 1655 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1656 xs = newexecsw; 1657 if (execsw) 1658 for (es = execsw; *es; es++) 1659 *xs++ = *es; 1660 *xs++ = execsw_arg; 1661 *xs = NULL; 1662 if (execsw) 1663 free(execsw, M_TEMP); 1664 execsw = newexecsw; 1665 return (0); 1666 } 1667 1668 int 1669 exec_unregister(execsw_arg) 1670 const struct execsw *execsw_arg; 1671 { 1672 const struct execsw **es, **xs, **newexecsw; 1673 int count = 1; 1674 1675 if (execsw == NULL) 1676 panic("unregister with no handlers left?\n"); 1677 1678 for (es = execsw; *es; es++) { 1679 if (*es == execsw_arg) 1680 break; 1681 } 1682 if (*es == NULL) 1683 return (ENOENT); 1684 for (es = execsw; *es; es++) 1685 if (*es != execsw_arg) 1686 count++; 1687 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1688 xs = newexecsw; 1689 for (es = execsw; *es; es++) 1690 if (*es != execsw_arg) 1691 *xs++ = *es; 1692 *xs = NULL; 1693 if (execsw) 1694 free(execsw, M_TEMP); 1695 execsw = newexecsw; 1696 return (0); 1697 } 1698