1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1993, David Greenman 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_capsicum.h" 33 #include "opt_hwpmc_hooks.h" 34 #include "opt_ktrace.h" 35 #include "opt_vm.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/acct.h> 40 #include <sys/capsicum.h> 41 #include <sys/eventhandler.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/filedesc.h> 45 #include <sys/imgact.h> 46 #include <sys/imgact_elf.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mman.h> 51 #include <sys/mount.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/pioctl.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/ptrace.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sched.h> 61 #include <sys/sdt.h> 62 #include <sys/sf_buf.h> 63 #include <sys/shm.h> 64 #include <sys/signalvar.h> 65 #include <sys/smp.h> 66 #include <sys/stat.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/sysproto.h> 71 #include <sys/vnode.h> 72 #include <sys/wait.h> 73 #ifdef KTRACE 74 #include <sys/ktrace.h> 75 #endif 76 77 #include <vm/vm.h> 78 #include <vm/vm_param.h> 79 #include <vm/pmap.h> 80 #include <vm/vm_page.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_kern.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_pager.h> 86 87 #ifdef HWPMC_HOOKS 88 #include <sys/pmckern.h> 89 #endif 90 91 #include <machine/reg.h> 92 93 #include <security/audit/audit.h> 94 #include <security/mac/mac_framework.h> 95 96 #ifdef KDTRACE_HOOKS 97 #include <sys/dtrace_bsd.h> 98 dtrace_execexit_func_t dtrace_fasttrap_exec; 99 #endif 100 101 SDT_PROVIDER_DECLARE(proc); 102 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 105 106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 107 108 int coredump_pack_fileinfo = 1; 109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 110 &coredump_pack_fileinfo, 0, 111 "Enable file path packing in 'procstat -f' coredump notes"); 112 113 int coredump_pack_vmmapinfo = 1; 114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 115 &coredump_pack_vmmapinfo, 0, 116 "Enable file path packing in 'procstat -v' coredump notes"); 117 118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 121 static int do_execve(struct thread *td, struct image_args *args, 122 struct mac *mac_p); 123 124 /* XXX This should be vm_size_t. */ 125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 127 128 /* XXX This should be vm_size_t. */ 129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 131 132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 133 NULL, 0, sysctl_kern_stackprot, "I", ""); 134 135 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 137 &ps_arg_cache_limit, 0, ""); 138 139 static int disallow_high_osrel; 140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 141 &disallow_high_osrel, 0, 142 "Disallow execution of binaries built for higher version of the world"); 143 144 static int map_at_zero = 0; 145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 146 "Permit processes to map an object at virtual address 0."); 147 148 EVENTHANDLER_LIST_DECLARE(process_exec); 149 150 static int 151 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 152 { 153 struct proc *p; 154 int error; 155 156 p = curproc; 157 #ifdef SCTL_MASK32 158 if (req->flags & SCTL_MASK32) { 159 unsigned int val; 160 val = (unsigned int)p->p_sysent->sv_psstrings; 161 error = SYSCTL_OUT(req, &val, sizeof(val)); 162 } else 163 #endif 164 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 165 sizeof(p->p_sysent->sv_psstrings)); 166 return error; 167 } 168 169 static int 170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 171 { 172 struct proc *p; 173 int error; 174 175 p = curproc; 176 #ifdef SCTL_MASK32 177 if (req->flags & SCTL_MASK32) { 178 unsigned int val; 179 val = (unsigned int)p->p_sysent->sv_usrstack; 180 error = SYSCTL_OUT(req, &val, sizeof(val)); 181 } else 182 #endif 183 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 184 sizeof(p->p_sysent->sv_usrstack)); 185 return error; 186 } 187 188 static int 189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 190 { 191 struct proc *p; 192 193 p = curproc; 194 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 195 sizeof(p->p_sysent->sv_stackprot))); 196 } 197 198 /* 199 * Each of the items is a pointer to a `const struct execsw', hence the 200 * double pointer here. 201 */ 202 static const struct execsw **execsw; 203 204 #ifndef _SYS_SYSPROTO_H_ 205 struct execve_args { 206 char *fname; 207 char **argv; 208 char **envv; 209 }; 210 #endif 211 212 int 213 sys_execve(struct thread *td, struct execve_args *uap) 214 { 215 struct image_args args; 216 struct vmspace *oldvmspace; 217 int error; 218 219 error = pre_execve(td, &oldvmspace); 220 if (error != 0) 221 return (error); 222 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 223 uap->argv, uap->envv); 224 if (error == 0) 225 error = kern_execve(td, &args, NULL); 226 post_execve(td, error, oldvmspace); 227 return (error); 228 } 229 230 #ifndef _SYS_SYSPROTO_H_ 231 struct fexecve_args { 232 int fd; 233 char **argv; 234 char **envv; 235 } 236 #endif 237 int 238 sys_fexecve(struct thread *td, struct fexecve_args *uap) 239 { 240 struct image_args args; 241 struct vmspace *oldvmspace; 242 int error; 243 244 error = pre_execve(td, &oldvmspace); 245 if (error != 0) 246 return (error); 247 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 248 uap->argv, uap->envv); 249 if (error == 0) { 250 args.fd = uap->fd; 251 error = kern_execve(td, &args, NULL); 252 } 253 post_execve(td, error, oldvmspace); 254 return (error); 255 } 256 257 #ifndef _SYS_SYSPROTO_H_ 258 struct __mac_execve_args { 259 char *fname; 260 char **argv; 261 char **envv; 262 struct mac *mac_p; 263 }; 264 #endif 265 266 int 267 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 268 { 269 #ifdef MAC 270 struct image_args args; 271 struct vmspace *oldvmspace; 272 int error; 273 274 error = pre_execve(td, &oldvmspace); 275 if (error != 0) 276 return (error); 277 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 278 uap->argv, uap->envv); 279 if (error == 0) 280 error = kern_execve(td, &args, uap->mac_p); 281 post_execve(td, error, oldvmspace); 282 return (error); 283 #else 284 return (ENOSYS); 285 #endif 286 } 287 288 int 289 pre_execve(struct thread *td, struct vmspace **oldvmspace) 290 { 291 struct proc *p; 292 int error; 293 294 KASSERT(td == curthread, ("non-current thread %p", td)); 295 error = 0; 296 p = td->td_proc; 297 if ((p->p_flag & P_HADTHREADS) != 0) { 298 PROC_LOCK(p); 299 if (thread_single(p, SINGLE_BOUNDARY) != 0) 300 error = ERESTART; 301 PROC_UNLOCK(p); 302 } 303 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 304 ("nested execve")); 305 *oldvmspace = p->p_vmspace; 306 return (error); 307 } 308 309 void 310 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 311 { 312 struct proc *p; 313 314 KASSERT(td == curthread, ("non-current thread %p", td)); 315 p = td->td_proc; 316 if ((p->p_flag & P_HADTHREADS) != 0) { 317 PROC_LOCK(p); 318 /* 319 * If success, we upgrade to SINGLE_EXIT state to 320 * force other threads to suicide. 321 */ 322 if (error == EJUSTRETURN) 323 thread_single(p, SINGLE_EXIT); 324 else 325 thread_single_end(p, SINGLE_BOUNDARY); 326 PROC_UNLOCK(p); 327 } 328 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 329 KASSERT(p->p_vmspace != oldvmspace, 330 ("oldvmspace still used")); 331 vmspace_free(oldvmspace); 332 td->td_pflags &= ~TDP_EXECVMSPC; 333 } 334 } 335 336 /* 337 * XXX: kern_execve has the astonishing property of not always returning to 338 * the caller. If sufficiently bad things happen during the call to 339 * do_execve(), it can end up calling exit1(); as a result, callers must 340 * avoid doing anything which they might need to undo (e.g., allocating 341 * memory). 342 */ 343 int 344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 345 { 346 347 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 348 args->begin_envv - args->begin_argv); 349 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 350 args->endp - args->begin_envv); 351 return (do_execve(td, args, mac_p)); 352 } 353 354 /* 355 * In-kernel implementation of execve(). All arguments are assumed to be 356 * userspace pointers from the passed thread. 357 */ 358 static int 359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 360 { 361 struct proc *p = td->td_proc; 362 struct nameidata nd; 363 struct ucred *oldcred; 364 struct uidinfo *euip = NULL; 365 register_t *stack_base; 366 int error, i; 367 struct image_params image_params, *imgp; 368 struct vattr attr; 369 int (*img_first)(struct image_params *); 370 struct pargs *oldargs = NULL, *newargs = NULL; 371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 372 #ifdef KTRACE 373 struct vnode *tracevp = NULL; 374 struct ucred *tracecred = NULL; 375 #endif 376 struct vnode *oldtextvp = NULL, *newtextvp; 377 int credential_changing; 378 #ifdef MAC 379 struct label *interpvplabel = NULL; 380 int will_transition; 381 #endif 382 #ifdef HWPMC_HOOKS 383 struct pmckern_procexec pe; 384 #endif 385 static const char fexecv_proc_title[] = "(fexecv)"; 386 387 imgp = &image_params; 388 389 /* 390 * Lock the process and set the P_INEXEC flag to indicate that 391 * it should be left alone until we're done here. This is 392 * necessary to avoid race conditions - e.g. in ptrace() - 393 * that might allow a local user to illicitly obtain elevated 394 * privileges. 395 */ 396 PROC_LOCK(p); 397 KASSERT((p->p_flag & P_INEXEC) == 0, 398 ("%s(): process already has P_INEXEC flag", __func__)); 399 p->p_flag |= P_INEXEC; 400 PROC_UNLOCK(p); 401 402 /* 403 * Initialize part of the common data 404 */ 405 bzero(imgp, sizeof(*imgp)); 406 imgp->proc = p; 407 imgp->attr = &attr; 408 imgp->args = args; 409 oldcred = p->p_ucred; 410 411 #ifdef MAC 412 error = mac_execve_enter(imgp, mac_p); 413 if (error) 414 goto exec_fail; 415 #endif 416 417 /* 418 * Translate the file name. namei() returns a vnode pointer 419 * in ni_vp among other things. 420 * 421 * XXXAUDIT: It would be desirable to also audit the name of the 422 * interpreter if this is an interpreted binary. 423 */ 424 if (args->fname != NULL) { 425 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 426 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 427 } 428 429 SDT_PROBE1(proc, , , exec, args->fname); 430 431 interpret: 432 if (args->fname != NULL) { 433 #ifdef CAPABILITY_MODE 434 /* 435 * While capability mode can't reach this point via direct 436 * path arguments to execve(), we also don't allow 437 * interpreters to be used in capability mode (for now). 438 * Catch indirect lookups and return a permissions error. 439 */ 440 if (IN_CAPABILITY_MODE(td)) { 441 error = ECAPMODE; 442 goto exec_fail; 443 } 444 #endif 445 error = namei(&nd); 446 if (error) 447 goto exec_fail; 448 449 newtextvp = nd.ni_vp; 450 imgp->vp = newtextvp; 451 } else { 452 AUDIT_ARG_FD(args->fd); 453 /* 454 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 455 */ 456 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 457 if (error) 458 goto exec_fail; 459 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 460 AUDIT_ARG_VNODE1(newtextvp); 461 imgp->vp = newtextvp; 462 } 463 464 /* 465 * Check file permissions. Also 'opens' file and sets its vnode to 466 * text mode. 467 */ 468 error = exec_check_permissions(imgp); 469 if (error) 470 goto exec_fail_dealloc; 471 472 imgp->object = imgp->vp->v_object; 473 if (imgp->object != NULL) 474 vm_object_reference(imgp->object); 475 476 error = exec_map_first_page(imgp); 477 if (error) 478 goto exec_fail_dealloc; 479 480 imgp->proc->p_osrel = 0; 481 imgp->proc->p_fctl0 = 0; 482 483 /* 484 * Implement image setuid/setgid. 485 * 486 * Determine new credentials before attempting image activators 487 * so that it can be used by process_exec handlers to determine 488 * credential/setid changes. 489 * 490 * Don't honor setuid/setgid if the filesystem prohibits it or if 491 * the process is being traced. 492 * 493 * We disable setuid/setgid/etc in capability mode on the basis 494 * that most setugid applications are not written with that 495 * environment in mind, and will therefore almost certainly operate 496 * incorrectly. In principle there's no reason that setugid 497 * applications might not be useful in capability mode, so we may want 498 * to reconsider this conservative design choice in the future. 499 * 500 * XXXMAC: For the time being, use NOSUID to also prohibit 501 * transitions on the file system. 502 */ 503 credential_changing = 0; 504 credential_changing |= (attr.va_mode & S_ISUID) && 505 oldcred->cr_uid != attr.va_uid; 506 credential_changing |= (attr.va_mode & S_ISGID) && 507 oldcred->cr_gid != attr.va_gid; 508 #ifdef MAC 509 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 510 interpvplabel, imgp); 511 credential_changing |= will_transition; 512 #endif 513 514 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 515 if (credential_changing) 516 imgp->proc->p_pdeathsig = 0; 517 518 if (credential_changing && 519 #ifdef CAPABILITY_MODE 520 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 521 #endif 522 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 523 (p->p_flag & P_TRACED) == 0) { 524 imgp->credential_setid = true; 525 VOP_UNLOCK(imgp->vp, 0); 526 imgp->newcred = crdup(oldcred); 527 if (attr.va_mode & S_ISUID) { 528 euip = uifind(attr.va_uid); 529 change_euid(imgp->newcred, euip); 530 } 531 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 532 if (attr.va_mode & S_ISGID) 533 change_egid(imgp->newcred, attr.va_gid); 534 /* 535 * Implement correct POSIX saved-id behavior. 536 * 537 * XXXMAC: Note that the current logic will save the 538 * uid and gid if a MAC domain transition occurs, even 539 * though maybe it shouldn't. 540 */ 541 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 542 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 543 } else { 544 /* 545 * Implement correct POSIX saved-id behavior. 546 * 547 * XXX: It's not clear that the existing behavior is 548 * POSIX-compliant. A number of sources indicate that the 549 * saved uid/gid should only be updated if the new ruid is 550 * not equal to the old ruid, or the new euid is not equal 551 * to the old euid and the new euid is not equal to the old 552 * ruid. The FreeBSD code always updates the saved uid/gid. 553 * Also, this code uses the new (replaced) euid and egid as 554 * the source, which may or may not be the right ones to use. 555 */ 556 if (oldcred->cr_svuid != oldcred->cr_uid || 557 oldcred->cr_svgid != oldcred->cr_gid) { 558 VOP_UNLOCK(imgp->vp, 0); 559 imgp->newcred = crdup(oldcred); 560 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 561 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 562 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 563 } 564 } 565 /* The new credentials are installed into the process later. */ 566 567 /* 568 * Do the best to calculate the full path to the image file. 569 */ 570 if (args->fname != NULL && args->fname[0] == '/') 571 imgp->execpath = args->fname; 572 else { 573 VOP_UNLOCK(imgp->vp, 0); 574 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 575 &imgp->freepath) != 0) 576 imgp->execpath = args->fname; 577 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 578 } 579 580 /* 581 * If the current process has a special image activator it 582 * wants to try first, call it. For example, emulating shell 583 * scripts differently. 584 */ 585 error = -1; 586 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 587 error = img_first(imgp); 588 589 /* 590 * Loop through the list of image activators, calling each one. 591 * An activator returns -1 if there is no match, 0 on success, 592 * and an error otherwise. 593 */ 594 for (i = 0; error == -1 && execsw[i]; ++i) { 595 if (execsw[i]->ex_imgact == NULL || 596 execsw[i]->ex_imgact == img_first) { 597 continue; 598 } 599 error = (*execsw[i]->ex_imgact)(imgp); 600 } 601 602 if (error) { 603 if (error == -1) 604 error = ENOEXEC; 605 goto exec_fail_dealloc; 606 } 607 608 /* 609 * Special interpreter operation, cleanup and loop up to try to 610 * activate the interpreter. 611 */ 612 if (imgp->interpreted) { 613 exec_unmap_first_page(imgp); 614 /* 615 * The text reference needs to be removed for scripts. 616 * There is a short period before we determine that 617 * something is a script where text reference is active. 618 * The vnode lock is held over this entire period 619 * so nothing should illegitimately be blocked. 620 */ 621 MPASS(imgp->textset); 622 VOP_UNSET_TEXT_CHECKED(newtextvp); 623 imgp->textset = false; 624 /* free name buffer and old vnode */ 625 if (args->fname != NULL) 626 NDFREE(&nd, NDF_ONLY_PNBUF); 627 #ifdef MAC 628 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 629 #endif 630 if (imgp->opened) { 631 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 632 imgp->opened = 0; 633 } 634 vput(newtextvp); 635 vm_object_deallocate(imgp->object); 636 imgp->object = NULL; 637 imgp->credential_setid = false; 638 if (imgp->newcred != NULL) { 639 crfree(imgp->newcred); 640 imgp->newcred = NULL; 641 } 642 imgp->execpath = NULL; 643 free(imgp->freepath, M_TEMP); 644 imgp->freepath = NULL; 645 /* set new name to that of the interpreter */ 646 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 647 UIO_SYSSPACE, imgp->interpreter_name, td); 648 args->fname = imgp->interpreter_name; 649 goto interpret; 650 } 651 652 /* 653 * NB: We unlock the vnode here because it is believed that none 654 * of the sv_copyout_strings/sv_fixup operations require the vnode. 655 */ 656 VOP_UNLOCK(imgp->vp, 0); 657 658 if (disallow_high_osrel && 659 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 660 error = ENOEXEC; 661 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 662 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 663 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 664 goto exec_fail_dealloc; 665 } 666 667 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 668 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 669 sys_cap_enter(td, NULL); 670 671 /* 672 * Copy out strings (args and env) and initialize stack base 673 */ 674 if (p->p_sysent->sv_copyout_strings) 675 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 676 else 677 stack_base = exec_copyout_strings(imgp); 678 679 /* 680 * If custom stack fixup routine present for this process 681 * let it do the stack setup. 682 * Else stuff argument count as first item on stack 683 */ 684 if (p->p_sysent->sv_fixup != NULL) 685 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 686 else 687 error = suword(--stack_base, imgp->args->argc) == 0 ? 688 0 : EFAULT; 689 if (error != 0) { 690 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 691 goto exec_fail_dealloc; 692 } 693 694 if (args->fdp != NULL) { 695 /* Install a brand new file descriptor table. */ 696 fdinstall_remapped(td, args->fdp); 697 args->fdp = NULL; 698 } else { 699 /* 700 * Keep on using the existing file descriptor table. For 701 * security and other reasons, the file descriptor table 702 * cannot be shared after an exec. 703 */ 704 fdunshare(td); 705 /* close files on exec */ 706 fdcloseexec(td); 707 } 708 709 /* 710 * Malloc things before we need locks. 711 */ 712 i = imgp->args->begin_envv - imgp->args->begin_argv; 713 /* Cache arguments if they fit inside our allowance */ 714 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 715 newargs = pargs_alloc(i); 716 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 717 } 718 719 /* 720 * For security and other reasons, signal handlers cannot 721 * be shared after an exec. The new process gets a copy of the old 722 * handlers. In execsigs(), the new process will have its signals 723 * reset. 724 */ 725 if (sigacts_shared(p->p_sigacts)) { 726 oldsigacts = p->p_sigacts; 727 newsigacts = sigacts_alloc(); 728 sigacts_copy(newsigacts, oldsigacts); 729 } 730 731 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 732 733 PROC_LOCK(p); 734 if (oldsigacts) 735 p->p_sigacts = newsigacts; 736 /* Stop profiling */ 737 stopprofclock(p); 738 739 /* reset caught signals */ 740 execsigs(p); 741 742 /* name this process - nameiexec(p, ndp) */ 743 bzero(p->p_comm, sizeof(p->p_comm)); 744 if (args->fname) 745 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 746 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 747 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 748 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 749 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 750 #ifdef KTR 751 sched_clear_tdname(td); 752 #endif 753 754 /* 755 * mark as execed, wakeup the process that vforked (if any) and tell 756 * it that it now has its own resources back 757 */ 758 p->p_flag |= P_EXEC; 759 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 760 p->p_flag2 &= ~P2_NOTRACE; 761 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 762 p->p_flag2 &= ~P2_STKGAP_DISABLE; 763 if (p->p_flag & P_PPWAIT) { 764 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 765 cv_broadcast(&p->p_pwait); 766 /* STOPs are no longer ignored, arrange for AST */ 767 signotify(td); 768 } 769 770 /* 771 * Implement image setuid/setgid installation. 772 */ 773 if (imgp->credential_setid) { 774 /* 775 * Turn off syscall tracing for set-id programs, except for 776 * root. Record any set-id flags first to make sure that 777 * we do not regain any tracing during a possible block. 778 */ 779 setsugid(p); 780 781 #ifdef KTRACE 782 if (p->p_tracecred != NULL && 783 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 784 ktrprocexec(p, &tracecred, &tracevp); 785 #endif 786 /* 787 * Close any file descriptors 0..2 that reference procfs, 788 * then make sure file descriptors 0..2 are in use. 789 * 790 * Both fdsetugidsafety() and fdcheckstd() may call functions 791 * taking sleepable locks, so temporarily drop our locks. 792 */ 793 PROC_UNLOCK(p); 794 VOP_UNLOCK(imgp->vp, 0); 795 fdsetugidsafety(td); 796 error = fdcheckstd(td); 797 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 798 if (error != 0) 799 goto exec_fail_dealloc; 800 PROC_LOCK(p); 801 #ifdef MAC 802 if (will_transition) { 803 mac_vnode_execve_transition(oldcred, imgp->newcred, 804 imgp->vp, interpvplabel, imgp); 805 } 806 #endif 807 } else { 808 if (oldcred->cr_uid == oldcred->cr_ruid && 809 oldcred->cr_gid == oldcred->cr_rgid) 810 p->p_flag &= ~P_SUGID; 811 } 812 /* 813 * Set the new credentials. 814 */ 815 if (imgp->newcred != NULL) { 816 proc_set_cred(p, imgp->newcred); 817 crfree(oldcred); 818 oldcred = NULL; 819 } 820 821 /* 822 * Store the vp for use in procfs. This vnode was referenced by namei 823 * or fgetvp_exec. 824 */ 825 oldtextvp = p->p_textvp; 826 p->p_textvp = newtextvp; 827 828 #ifdef KDTRACE_HOOKS 829 /* 830 * Tell the DTrace fasttrap provider about the exec if it 831 * has declared an interest. 832 */ 833 if (dtrace_fasttrap_exec) 834 dtrace_fasttrap_exec(p); 835 #endif 836 837 /* 838 * Notify others that we exec'd, and clear the P_INEXEC flag 839 * as we're now a bona fide freshly-execed process. 840 */ 841 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 842 p->p_flag &= ~P_INEXEC; 843 844 /* clear "fork but no exec" flag, as we _are_ execing */ 845 p->p_acflag &= ~AFORK; 846 847 /* 848 * Free any previous argument cache and replace it with 849 * the new argument cache, if any. 850 */ 851 oldargs = p->p_args; 852 p->p_args = newargs; 853 newargs = NULL; 854 855 PROC_UNLOCK(p); 856 857 #ifdef HWPMC_HOOKS 858 /* 859 * Check if system-wide sampling is in effect or if the 860 * current process is using PMCs. If so, do exec() time 861 * processing. This processing needs to happen AFTER the 862 * P_INEXEC flag is cleared. 863 */ 864 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 865 VOP_UNLOCK(imgp->vp, 0); 866 pe.pm_credentialschanged = credential_changing; 867 pe.pm_entryaddr = imgp->entry_addr; 868 869 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 870 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 871 } 872 #endif 873 874 /* Set values passed into the program in registers. */ 875 if (p->p_sysent->sv_setregs) 876 (*p->p_sysent->sv_setregs)(td, imgp, 877 (u_long)(uintptr_t)stack_base); 878 else 879 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 880 881 vfs_mark_atime(imgp->vp, td->td_ucred); 882 883 SDT_PROBE1(proc, , , exec__success, args->fname); 884 885 exec_fail_dealloc: 886 if (imgp->firstpage != NULL) 887 exec_unmap_first_page(imgp); 888 889 if (imgp->vp != NULL) { 890 if (args->fname) 891 NDFREE(&nd, NDF_ONLY_PNBUF); 892 if (imgp->opened) 893 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 894 if (imgp->textset) 895 VOP_UNSET_TEXT_CHECKED(imgp->vp); 896 if (error != 0) 897 vput(imgp->vp); 898 else 899 VOP_UNLOCK(imgp->vp, 0); 900 } 901 902 if (imgp->object != NULL) 903 vm_object_deallocate(imgp->object); 904 905 free(imgp->freepath, M_TEMP); 906 907 if (error == 0) { 908 if (p->p_ptevents & PTRACE_EXEC) { 909 PROC_LOCK(p); 910 if (p->p_ptevents & PTRACE_EXEC) 911 td->td_dbgflags |= TDB_EXEC; 912 PROC_UNLOCK(p); 913 } 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 /* 969 * We don't want cpu_set_syscall_retval() to overwrite any of 970 * the register values put in place by exec_setregs(). 971 * Implementations of cpu_set_syscall_retval() will leave 972 * registers unmodified when returning EJUSTRETURN. 973 */ 974 return (error == 0 ? EJUSTRETURN : error); 975 } 976 977 int 978 exec_map_first_page(struct image_params *imgp) 979 { 980 int rv, i, after, initial_pagein; 981 vm_page_t ma[VM_INITIAL_PAGEIN]; 982 vm_object_t object; 983 984 if (imgp->firstpage != NULL) 985 exec_unmap_first_page(imgp); 986 987 object = imgp->vp->v_object; 988 if (object == NULL) 989 return (EACCES); 990 VM_OBJECT_WLOCK(object); 991 #if VM_NRESERVLEVEL > 0 992 vm_object_color(object, 0); 993 #endif 994 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 995 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 996 vm_page_xbusy(ma[0]); 997 if (!vm_pager_has_page(object, 0, NULL, &after)) { 998 vm_page_lock(ma[0]); 999 vm_page_free(ma[0]); 1000 vm_page_unlock(ma[0]); 1001 VM_OBJECT_WUNLOCK(object); 1002 return (EIO); 1003 } 1004 initial_pagein = min(after, VM_INITIAL_PAGEIN); 1005 KASSERT(initial_pagein <= object->size, 1006 ("%s: initial_pagein %d object->size %ju", 1007 __func__, initial_pagein, (uintmax_t )object->size)); 1008 for (i = 1; i < initial_pagein; i++) { 1009 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1010 if (ma[i]->valid) 1011 break; 1012 if (!vm_page_tryxbusy(ma[i])) 1013 break; 1014 } else { 1015 ma[i] = vm_page_alloc(object, i, 1016 VM_ALLOC_NORMAL); 1017 if (ma[i] == NULL) 1018 break; 1019 } 1020 } 1021 initial_pagein = i; 1022 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1023 if (rv != VM_PAGER_OK) { 1024 for (i = 0; i < initial_pagein; i++) { 1025 vm_page_lock(ma[i]); 1026 vm_page_free(ma[i]); 1027 vm_page_unlock(ma[i]); 1028 } 1029 VM_OBJECT_WUNLOCK(object); 1030 return (EIO); 1031 } 1032 vm_page_xunbusy(ma[0]); 1033 for (i = 1; i < initial_pagein; i++) 1034 vm_page_readahead_finish(ma[i]); 1035 } 1036 vm_page_lock(ma[0]); 1037 vm_page_hold(ma[0]); 1038 vm_page_activate(ma[0]); 1039 vm_page_unlock(ma[0]); 1040 VM_OBJECT_WUNLOCK(object); 1041 1042 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1043 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1044 1045 return (0); 1046 } 1047 1048 void 1049 exec_unmap_first_page(struct image_params *imgp) 1050 { 1051 vm_page_t m; 1052 1053 if (imgp->firstpage != NULL) { 1054 m = sf_buf_page(imgp->firstpage); 1055 sf_buf_free(imgp->firstpage); 1056 imgp->firstpage = NULL; 1057 vm_page_lock(m); 1058 vm_page_unhold(m); 1059 vm_page_unlock(m); 1060 } 1061 } 1062 1063 /* 1064 * Destroy old address space, and allocate a new stack. 1065 * The new stack is only sgrowsiz large because it is grown 1066 * automatically on a page fault. 1067 */ 1068 int 1069 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1070 { 1071 int error; 1072 struct proc *p = imgp->proc; 1073 struct vmspace *vmspace = p->p_vmspace; 1074 vm_object_t obj; 1075 struct rlimit rlim_stack; 1076 vm_offset_t sv_minuser, stack_addr; 1077 vm_map_t map; 1078 u_long ssiz; 1079 1080 imgp->vmspace_destroyed = 1; 1081 imgp->sysent = sv; 1082 1083 /* May be called with Giant held */ 1084 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1085 1086 /* 1087 * Blow away entire process VM, if address space not shared, 1088 * otherwise, create a new VM space so that other threads are 1089 * not disrupted 1090 */ 1091 map = &vmspace->vm_map; 1092 if (map_at_zero) 1093 sv_minuser = sv->sv_minuser; 1094 else 1095 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1096 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1097 vm_map_max(map) == sv->sv_maxuser && 1098 cpu_exec_vmspace_reuse(p, map)) { 1099 shmexit(vmspace); 1100 pmap_remove_pages(vmspace_pmap(vmspace)); 1101 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1102 /* 1103 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1104 * must be re-evaluated. 1105 */ 1106 vm_map_lock(map); 1107 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1108 MAP_ASLR_IGNSTART); 1109 vm_map_unlock(map); 1110 } else { 1111 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1112 if (error) 1113 return (error); 1114 vmspace = p->p_vmspace; 1115 map = &vmspace->vm_map; 1116 } 1117 map->flags |= imgp->map_flags; 1118 1119 /* Map a shared page */ 1120 obj = sv->sv_shared_page_obj; 1121 if (obj != NULL) { 1122 vm_object_reference(obj); 1123 error = vm_map_fixed(map, obj, 0, 1124 sv->sv_shared_page_base, sv->sv_shared_page_len, 1125 VM_PROT_READ | VM_PROT_EXECUTE, 1126 VM_PROT_READ | VM_PROT_EXECUTE, 1127 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1128 if (error != KERN_SUCCESS) { 1129 vm_object_deallocate(obj); 1130 return (vm_mmap_to_errno(error)); 1131 } 1132 } 1133 1134 /* Allocate a new stack */ 1135 if (imgp->stack_sz != 0) { 1136 ssiz = trunc_page(imgp->stack_sz); 1137 PROC_LOCK(p); 1138 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1139 PROC_UNLOCK(p); 1140 if (ssiz > rlim_stack.rlim_max) 1141 ssiz = rlim_stack.rlim_max; 1142 if (ssiz > rlim_stack.rlim_cur) { 1143 rlim_stack.rlim_cur = ssiz; 1144 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1145 } 1146 } else if (sv->sv_maxssiz != NULL) { 1147 ssiz = *sv->sv_maxssiz; 1148 } else { 1149 ssiz = maxssiz; 1150 } 1151 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1152 if (ssiz < imgp->eff_stack_sz) 1153 imgp->eff_stack_sz = ssiz; 1154 stack_addr = sv->sv_usrstack - ssiz; 1155 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1156 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1157 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1158 if (error != KERN_SUCCESS) 1159 return (vm_mmap_to_errno(error)); 1160 1161 /* 1162 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1163 * are still used to enforce the stack rlimit on the process stack. 1164 */ 1165 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1166 vmspace->vm_maxsaddr = (char *)stack_addr; 1167 1168 return (0); 1169 } 1170 1171 /* 1172 * Copy out argument and environment strings from the old process address 1173 * space into the temporary string buffer. 1174 */ 1175 int 1176 exec_copyin_args(struct image_args *args, char *fname, 1177 enum uio_seg segflg, char **argv, char **envv) 1178 { 1179 u_long argp, envp; 1180 int error; 1181 size_t length; 1182 1183 bzero(args, sizeof(*args)); 1184 if (argv == NULL) 1185 return (EFAULT); 1186 1187 /* 1188 * Allocate demand-paged memory for the file name, argument, and 1189 * environment strings. 1190 */ 1191 error = exec_alloc_args(args); 1192 if (error != 0) 1193 return (error); 1194 1195 /* 1196 * Copy the file name. 1197 */ 1198 if (fname != NULL) { 1199 args->fname = args->buf; 1200 error = (segflg == UIO_SYSSPACE) ? 1201 copystr(fname, args->fname, PATH_MAX, &length) : 1202 copyinstr(fname, args->fname, PATH_MAX, &length); 1203 if (error != 0) 1204 goto err_exit; 1205 } else 1206 length = 0; 1207 1208 args->begin_argv = args->buf + length; 1209 args->endp = args->begin_argv; 1210 args->stringspace = ARG_MAX; 1211 1212 /* 1213 * extract arguments first 1214 */ 1215 for (;;) { 1216 error = fueword(argv++, &argp); 1217 if (error == -1) { 1218 error = EFAULT; 1219 goto err_exit; 1220 } 1221 if (argp == 0) 1222 break; 1223 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1224 args->stringspace, &length); 1225 if (error != 0) { 1226 if (error == ENAMETOOLONG) 1227 error = E2BIG; 1228 goto err_exit; 1229 } 1230 args->stringspace -= length; 1231 args->endp += length; 1232 args->argc++; 1233 } 1234 1235 args->begin_envv = args->endp; 1236 1237 /* 1238 * extract environment strings 1239 */ 1240 if (envv) { 1241 for (;;) { 1242 error = fueword(envv++, &envp); 1243 if (error == -1) { 1244 error = EFAULT; 1245 goto err_exit; 1246 } 1247 if (envp == 0) 1248 break; 1249 error = copyinstr((void *)(uintptr_t)envp, 1250 args->endp, args->stringspace, &length); 1251 if (error != 0) { 1252 if (error == ENAMETOOLONG) 1253 error = E2BIG; 1254 goto err_exit; 1255 } 1256 args->stringspace -= length; 1257 args->endp += length; 1258 args->envc++; 1259 } 1260 } 1261 1262 return (0); 1263 1264 err_exit: 1265 exec_free_args(args); 1266 return (error); 1267 } 1268 1269 int 1270 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1271 const void *data, size_t datalen, const int *fds, size_t fdslen) 1272 { 1273 struct filedesc *ofdp; 1274 const char *p; 1275 int *kfds; 1276 int error; 1277 1278 memset(args, '\0', sizeof(*args)); 1279 ofdp = td->td_proc->p_fd; 1280 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1281 return (E2BIG); 1282 error = exec_alloc_args(args); 1283 if (error != 0) 1284 return (error); 1285 1286 args->begin_argv = args->buf; 1287 args->stringspace = ARG_MAX; 1288 1289 if (datalen > 0) { 1290 /* 1291 * Argument buffer has been provided. Copy it into the 1292 * kernel as a single string and add a terminating null 1293 * byte. 1294 */ 1295 error = copyin(data, args->begin_argv, datalen); 1296 if (error != 0) 1297 goto err_exit; 1298 args->begin_argv[datalen] = '\0'; 1299 args->endp = args->begin_argv + datalen + 1; 1300 args->stringspace -= datalen + 1; 1301 1302 /* 1303 * Traditional argument counting. Count the number of 1304 * null bytes. 1305 */ 1306 for (p = args->begin_argv; p < args->endp; ++p) 1307 if (*p == '\0') 1308 ++args->argc; 1309 } else { 1310 /* No argument buffer provided. */ 1311 args->endp = args->begin_argv; 1312 } 1313 /* There are no environment variables. */ 1314 args->begin_envv = args->endp; 1315 1316 /* Create new file descriptor table. */ 1317 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1318 error = copyin(fds, kfds, fdslen * sizeof(int)); 1319 if (error != 0) { 1320 free(kfds, M_TEMP); 1321 goto err_exit; 1322 } 1323 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1324 free(kfds, M_TEMP); 1325 if (error != 0) 1326 goto err_exit; 1327 1328 return (0); 1329 err_exit: 1330 exec_free_args(args); 1331 return (error); 1332 } 1333 1334 struct exec_args_kva { 1335 vm_offset_t addr; 1336 u_int gen; 1337 SLIST_ENTRY(exec_args_kva) next; 1338 }; 1339 1340 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1341 1342 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1343 static struct mtx exec_args_kva_mtx; 1344 static u_int exec_args_gen; 1345 1346 static void 1347 exec_prealloc_args_kva(void *arg __unused) 1348 { 1349 struct exec_args_kva *argkva; 1350 u_int i; 1351 1352 SLIST_INIT(&exec_args_kva_freelist); 1353 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1354 for (i = 0; i < exec_map_entries; i++) { 1355 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1356 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1357 argkva->gen = exec_args_gen; 1358 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1359 } 1360 } 1361 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1362 1363 static vm_offset_t 1364 exec_alloc_args_kva(void **cookie) 1365 { 1366 struct exec_args_kva *argkva; 1367 1368 argkva = (void *)atomic_readandclear_ptr( 1369 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1370 if (argkva == NULL) { 1371 mtx_lock(&exec_args_kva_mtx); 1372 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1373 (void)mtx_sleep(&exec_args_kva_freelist, 1374 &exec_args_kva_mtx, 0, "execkva", 0); 1375 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1376 mtx_unlock(&exec_args_kva_mtx); 1377 } 1378 *(struct exec_args_kva **)cookie = argkva; 1379 return (argkva->addr); 1380 } 1381 1382 static void 1383 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1384 { 1385 vm_offset_t base; 1386 1387 base = argkva->addr; 1388 if (argkva->gen != gen) { 1389 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1390 MADV_FREE); 1391 argkva->gen = gen; 1392 } 1393 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1394 (uintptr_t)NULL, (uintptr_t)argkva)) { 1395 mtx_lock(&exec_args_kva_mtx); 1396 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1397 wakeup_one(&exec_args_kva_freelist); 1398 mtx_unlock(&exec_args_kva_mtx); 1399 } 1400 } 1401 1402 static void 1403 exec_free_args_kva(void *cookie) 1404 { 1405 1406 exec_release_args_kva(cookie, exec_args_gen); 1407 } 1408 1409 static void 1410 exec_args_kva_lowmem(void *arg __unused) 1411 { 1412 SLIST_HEAD(, exec_args_kva) head; 1413 struct exec_args_kva *argkva; 1414 u_int gen; 1415 int i; 1416 1417 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1418 1419 /* 1420 * Force an madvise of each KVA range. Any currently allocated ranges 1421 * will have MADV_FREE applied once they are freed. 1422 */ 1423 SLIST_INIT(&head); 1424 mtx_lock(&exec_args_kva_mtx); 1425 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1426 mtx_unlock(&exec_args_kva_mtx); 1427 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1428 SLIST_REMOVE_HEAD(&head, next); 1429 exec_release_args_kva(argkva, gen); 1430 } 1431 1432 CPU_FOREACH(i) { 1433 argkva = (void *)atomic_readandclear_ptr( 1434 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1435 if (argkva != NULL) 1436 exec_release_args_kva(argkva, gen); 1437 } 1438 } 1439 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1440 EVENTHANDLER_PRI_ANY); 1441 1442 /* 1443 * Allocate temporary demand-paged, zero-filled memory for the file name, 1444 * argument, and environment strings. 1445 */ 1446 int 1447 exec_alloc_args(struct image_args *args) 1448 { 1449 1450 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1451 return (0); 1452 } 1453 1454 void 1455 exec_free_args(struct image_args *args) 1456 { 1457 1458 if (args->buf != NULL) { 1459 exec_free_args_kva(args->bufkva); 1460 args->buf = NULL; 1461 } 1462 if (args->fname_buf != NULL) { 1463 free(args->fname_buf, M_TEMP); 1464 args->fname_buf = NULL; 1465 } 1466 if (args->fdp != NULL) 1467 fdescfree_remapped(args->fdp); 1468 } 1469 1470 /* 1471 * Copy strings out to the new process address space, constructing new arg 1472 * and env vector tables. Return a pointer to the base so that it can be used 1473 * as the initial stack pointer. 1474 */ 1475 register_t * 1476 exec_copyout_strings(struct image_params *imgp) 1477 { 1478 int argc, envc; 1479 char **vectp; 1480 char *stringp; 1481 uintptr_t destp; 1482 register_t *stack_base; 1483 struct ps_strings *arginfo; 1484 struct proc *p; 1485 size_t execpath_len; 1486 int szsigcode, szps; 1487 char canary[sizeof(long) * 8]; 1488 1489 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1490 /* 1491 * Calculate string base and vector table pointers. 1492 * Also deal with signal trampoline code for this exec type. 1493 */ 1494 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1495 execpath_len = strlen(imgp->execpath) + 1; 1496 else 1497 execpath_len = 0; 1498 p = imgp->proc; 1499 szsigcode = 0; 1500 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1501 if (p->p_sysent->sv_sigcode_base == 0) { 1502 if (p->p_sysent->sv_szsigcode != NULL) 1503 szsigcode = *(p->p_sysent->sv_szsigcode); 1504 } 1505 destp = (uintptr_t)arginfo; 1506 1507 /* 1508 * install sigcode 1509 */ 1510 if (szsigcode != 0) { 1511 destp -= szsigcode; 1512 destp = rounddown2(destp, sizeof(void *)); 1513 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1514 } 1515 1516 /* 1517 * Copy the image path for the rtld. 1518 */ 1519 if (execpath_len != 0) { 1520 destp -= execpath_len; 1521 destp = rounddown2(destp, sizeof(void *)); 1522 imgp->execpathp = destp; 1523 copyout(imgp->execpath, (void *)destp, execpath_len); 1524 } 1525 1526 /* 1527 * Prepare the canary for SSP. 1528 */ 1529 arc4rand(canary, sizeof(canary), 0); 1530 destp -= sizeof(canary); 1531 imgp->canary = destp; 1532 copyout(canary, (void *)destp, sizeof(canary)); 1533 imgp->canarylen = sizeof(canary); 1534 1535 /* 1536 * Prepare the pagesizes array. 1537 */ 1538 destp -= szps; 1539 destp = rounddown2(destp, sizeof(void *)); 1540 imgp->pagesizes = destp; 1541 copyout(pagesizes, (void *)destp, szps); 1542 imgp->pagesizeslen = szps; 1543 1544 destp -= ARG_MAX - imgp->args->stringspace; 1545 destp = rounddown2(destp, sizeof(void *)); 1546 1547 vectp = (char **)destp; 1548 if (imgp->sysent->sv_stackgap != NULL) 1549 imgp->sysent->sv_stackgap(imgp, (u_long *)&vectp); 1550 1551 if (imgp->auxargs) { 1552 /* 1553 * Allocate room on the stack for the ELF auxargs 1554 * array. It has up to AT_COUNT entries. 1555 */ 1556 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 1557 sizeof(*vectp)); 1558 } 1559 1560 /* 1561 * Allocate room for the argv[] and env vectors including the 1562 * terminating NULL pointers. 1563 */ 1564 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1565 1566 /* 1567 * vectp also becomes our initial stack base 1568 */ 1569 stack_base = (register_t *)vectp; 1570 1571 stringp = imgp->args->begin_argv; 1572 argc = imgp->args->argc; 1573 envc = imgp->args->envc; 1574 1575 /* 1576 * Copy out strings - arguments and environment. 1577 */ 1578 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1579 1580 /* 1581 * Fill in "ps_strings" struct for ps, w, etc. 1582 */ 1583 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1584 suword32(&arginfo->ps_nargvstr, argc); 1585 1586 /* 1587 * Fill in argument portion of vector table. 1588 */ 1589 for (; argc > 0; --argc) { 1590 suword(vectp++, (long)(intptr_t)destp); 1591 while (*stringp++ != 0) 1592 destp++; 1593 destp++; 1594 } 1595 1596 /* a null vector table pointer separates the argp's from the envp's */ 1597 suword(vectp++, 0); 1598 1599 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1600 suword32(&arginfo->ps_nenvstr, envc); 1601 1602 /* 1603 * Fill in environment portion of vector table. 1604 */ 1605 for (; envc > 0; --envc) { 1606 suword(vectp++, (long)(intptr_t)destp); 1607 while (*stringp++ != 0) 1608 destp++; 1609 destp++; 1610 } 1611 1612 /* end of vector table is a null pointer */ 1613 suword(vectp, 0); 1614 1615 return (stack_base); 1616 } 1617 1618 /* 1619 * Check permissions of file to execute. 1620 * Called with imgp->vp locked. 1621 * Return 0 for success or error code on failure. 1622 */ 1623 int 1624 exec_check_permissions(struct image_params *imgp) 1625 { 1626 struct vnode *vp = imgp->vp; 1627 struct vattr *attr = imgp->attr; 1628 struct thread *td; 1629 int error; 1630 1631 td = curthread; 1632 1633 /* Get file attributes */ 1634 error = VOP_GETATTR(vp, attr, td->td_ucred); 1635 if (error) 1636 return (error); 1637 1638 #ifdef MAC 1639 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1640 if (error) 1641 return (error); 1642 #endif 1643 1644 /* 1645 * 1) Check if file execution is disabled for the filesystem that 1646 * this file resides on. 1647 * 2) Ensure that at least one execute bit is on. Otherwise, a 1648 * privileged user will always succeed, and we don't want this 1649 * to happen unless the file really is executable. 1650 * 3) Ensure that the file is a regular file. 1651 */ 1652 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1653 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1654 (attr->va_type != VREG)) 1655 return (EACCES); 1656 1657 /* 1658 * Zero length files can't be exec'd 1659 */ 1660 if (attr->va_size == 0) 1661 return (ENOEXEC); 1662 1663 /* 1664 * Check for execute permission to file based on current credentials. 1665 */ 1666 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1667 if (error) 1668 return (error); 1669 1670 /* 1671 * Check number of open-for-writes on the file and deny execution 1672 * if there are any. 1673 * 1674 * Add a text reference now so no one can write to the 1675 * executable while we're activating it. 1676 * 1677 * Remember if this was set before and unset it in case this is not 1678 * actually an executable image. 1679 */ 1680 error = VOP_SET_TEXT(vp); 1681 if (error != 0) 1682 return (error); 1683 imgp->textset = true; 1684 1685 /* 1686 * Call filesystem specific open routine (which does nothing in the 1687 * general case). 1688 */ 1689 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1690 if (error == 0) 1691 imgp->opened = 1; 1692 return (error); 1693 } 1694 1695 /* 1696 * Exec handler registration 1697 */ 1698 int 1699 exec_register(const struct execsw *execsw_arg) 1700 { 1701 const struct execsw **es, **xs, **newexecsw; 1702 u_int count = 2; /* New slot and trailing NULL */ 1703 1704 if (execsw) 1705 for (es = execsw; *es; es++) 1706 count++; 1707 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1708 xs = newexecsw; 1709 if (execsw) 1710 for (es = execsw; *es; es++) 1711 *xs++ = *es; 1712 *xs++ = execsw_arg; 1713 *xs = NULL; 1714 if (execsw) 1715 free(execsw, M_TEMP); 1716 execsw = newexecsw; 1717 return (0); 1718 } 1719 1720 int 1721 exec_unregister(const struct execsw *execsw_arg) 1722 { 1723 const struct execsw **es, **xs, **newexecsw; 1724 int count = 1; 1725 1726 if (execsw == NULL) 1727 panic("unregister with no handlers left?\n"); 1728 1729 for (es = execsw; *es; es++) { 1730 if (*es == execsw_arg) 1731 break; 1732 } 1733 if (*es == NULL) 1734 return (ENOENT); 1735 for (es = execsw; *es; es++) 1736 if (*es != execsw_arg) 1737 count++; 1738 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1739 xs = newexecsw; 1740 for (es = execsw; *es; es++) 1741 if (*es != execsw_arg) 1742 *xs++ = *es; 1743 *xs = NULL; 1744 if (execsw) 1745 free(execsw, M_TEMP); 1746 execsw = newexecsw; 1747 return (0); 1748 } 1749