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/timers.h> 72#include <sys/umtx.h> 73#include <sys/vnode.h> 74#include <sys/wait.h> 75#ifdef KTRACE 76#include <sys/ktrace.h> 77#endif 78 79#include <vm/vm.h> 80#include <vm/vm_param.h> 81#include <vm/pmap.h> 82#include <vm/vm_page.h> 83#include <vm/vm_map.h> 84#include <vm/vm_kern.h> 85#include <vm/vm_extern.h> 86#include <vm/vm_object.h> 87#include <vm/vm_pager.h> 88 89#ifdef HWPMC_HOOKS 90#include <sys/pmckern.h> 91#endif 92 93#include <machine/reg.h> 94 95#include <security/audit/audit.h> 96#include <security/mac/mac_framework.h> 97 98#ifdef KDTRACE_HOOKS 99#include <sys/dtrace_bsd.h> 100dtrace_execexit_func_t dtrace_fasttrap_exec; 101#endif 102 103SDT_PROVIDER_DECLARE(proc); 104SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 105SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 106SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 107 108MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 109 110int coredump_pack_fileinfo = 1; 111SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 112 &coredump_pack_fileinfo, 0, 113 "Enable file path packing in 'procstat -f' coredump notes"); 114 115int coredump_pack_vmmapinfo = 1; 116SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 117 &coredump_pack_vmmapinfo, 0, 118 "Enable file path packing in 'procstat -v' coredump notes"); 119 120static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 121static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 122static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 123static int do_execve(struct thread *td, struct image_args *args, 124 struct mac *mac_p, struct vmspace *oldvmspace); 125 126/* XXX This should be vm_size_t. */ 127SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 128 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 129 130/* XXX This should be vm_size_t. */ 131SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 132 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 133 134SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 135 NULL, 0, sysctl_kern_stackprot, "I", ""); 136 137u_long ps_arg_cache_limit = PAGE_SIZE / 16; 138SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 139 &ps_arg_cache_limit, 0, ""); 140 141static int disallow_high_osrel; 142SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 143 &disallow_high_osrel, 0, 144 "Disallow execution of binaries built for higher version of the world"); 145 146static int map_at_zero = 0; 147SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 148 "Permit processes to map an object at virtual address 0."); 149 150EVENTHANDLER_LIST_DECLARE(process_exec); 151 152static int 153sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 154{ 155 struct proc *p; 156 int error; 157 158 p = curproc; 159#ifdef SCTL_MASK32 160 if (req->flags & SCTL_MASK32) { 161 unsigned int val; 162 val = (unsigned int)p->p_sysent->sv_psstrings; 163 error = SYSCTL_OUT(req, &val, sizeof(val)); 164 } else 165#endif 166 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 167 sizeof(p->p_sysent->sv_psstrings)); 168 return error; 169} 170 171static int 172sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 173{ 174 struct proc *p; 175 int error; 176 177 p = curproc; 178#ifdef SCTL_MASK32 179 if (req->flags & SCTL_MASK32) { 180 unsigned int val; 181 val = (unsigned int)p->p_sysent->sv_usrstack; 182 error = SYSCTL_OUT(req, &val, sizeof(val)); 183 } else 184#endif 185 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 186 sizeof(p->p_sysent->sv_usrstack)); 187 return error; 188} 189 190static int 191sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 192{ 193 struct proc *p; 194 195 p = curproc; 196 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 197 sizeof(p->p_sysent->sv_stackprot))); 198} 199 200/* 201 * Each of the items is a pointer to a `const struct execsw', hence the 202 * double pointer here. 203 */ 204static const struct execsw **execsw; 205 206#ifndef _SYS_SYSPROTO_H_ 207struct execve_args { 208 char *fname; 209 char **argv; 210 char **envv; 211}; 212#endif 213 214int 215sys_execve(struct thread *td, struct execve_args *uap) 216{ 217 struct image_args args; 218 struct vmspace *oldvmspace; 219 int error; 220 221 error = pre_execve(td, &oldvmspace); 222 if (error != 0) 223 return (error); 224 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 225 uap->argv, uap->envv); 226 if (error == 0) 227 error = kern_execve(td, &args, NULL, oldvmspace); 228 post_execve(td, error, oldvmspace); 229 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 230 return (error); 231} 232 233#ifndef _SYS_SYSPROTO_H_ 234struct fexecve_args { 235 int fd; 236 char **argv; 237 char **envv; 238} 239#endif 240int 241sys_fexecve(struct thread *td, struct fexecve_args *uap) 242{ 243 struct image_args args; 244 struct vmspace *oldvmspace; 245 int error; 246 247 error = pre_execve(td, &oldvmspace); 248 if (error != 0) 249 return (error); 250 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 251 uap->argv, uap->envv); 252 if (error == 0) { 253 args.fd = uap->fd; 254 error = kern_execve(td, &args, NULL, oldvmspace); 255 } 256 post_execve(td, error, oldvmspace); 257 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 258 return (error); 259} 260 261#ifndef _SYS_SYSPROTO_H_ 262struct __mac_execve_args { 263 char *fname; 264 char **argv; 265 char **envv; 266 struct mac *mac_p; 267}; 268#endif 269 270int 271sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 272{ 273#ifdef MAC 274 struct image_args args; 275 struct vmspace *oldvmspace; 276 int error; 277 278 error = pre_execve(td, &oldvmspace); 279 if (error != 0) 280 return (error); 281 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 282 uap->argv, uap->envv); 283 if (error == 0) 284 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 285 post_execve(td, error, oldvmspace); 286 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 287 return (error); 288#else 289 return (ENOSYS); 290#endif 291} 292 293int 294pre_execve(struct thread *td, struct vmspace **oldvmspace) 295{ 296 struct proc *p; 297 int error; 298 299 KASSERT(td == curthread, ("non-current thread %p", td)); 300 error = 0; 301 p = td->td_proc; 302 if ((p->p_flag & P_HADTHREADS) != 0) { 303 PROC_LOCK(p); 304 if (thread_single(p, SINGLE_BOUNDARY) != 0) 305 error = ERESTART; 306 PROC_UNLOCK(p); 307 } 308 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 309 ("nested execve")); 310 *oldvmspace = p->p_vmspace; 311 return (error); 312} 313 314void 315post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 316{ 317 struct proc *p; 318 319 KASSERT(td == curthread, ("non-current thread %p", td)); 320 p = td->td_proc; 321 if ((p->p_flag & P_HADTHREADS) != 0) { 322 PROC_LOCK(p); 323 /* 324 * If success, we upgrade to SINGLE_EXIT state to 325 * force other threads to suicide. 326 */ 327 if (error == EJUSTRETURN) 328 thread_single(p, SINGLE_EXIT); 329 else 330 thread_single_end(p, SINGLE_BOUNDARY); 331 PROC_UNLOCK(p); 332 } 333 exec_cleanup(td, oldvmspace); 334} 335 336/* 337 * 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 */ 343int 344kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 345 struct vmspace *oldvmspace) 346{ 347 348 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 349 args->begin_envv - args->begin_argv); 350 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 351 args->endp - args->begin_envv); 352 return (do_execve(td, args, mac_p, oldvmspace)); 353} 354 355/* 356 * In-kernel implementation of execve(). All arguments are assumed to be 357 * userspace pointers from the passed thread. 358 */ 359static int 360do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 361 struct vmspace *oldvmspace) 362{ 363 struct proc *p = td->td_proc; 364 struct nameidata nd; 365 struct ucred *oldcred; 366 struct uidinfo *euip = NULL; 367 register_t *stack_base; 368 struct image_params image_params, *imgp; 369 struct vattr attr; 370 int (*img_first)(struct image_params *); 371 struct pargs *oldargs = NULL, *newargs = NULL; 372 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 373#ifdef KTRACE 374 struct vnode *tracevp = NULL; 375 struct ucred *tracecred = NULL; 376#endif 377 struct vnode *oldtextvp = NULL, *newtextvp; 378 int credential_changing; 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 int error, i, orig_osrel; 387 uint32_t orig_fctl0; 388 static const char fexecv_proc_title[] = "(fexecv)"; 389 390 imgp = &image_params; 391 392 /* 393 * Lock the process and set the P_INEXEC flag to indicate that 394 * it should be left alone until we're done here. This is 395 * necessary to avoid race conditions - e.g. in ptrace() - 396 * that might allow a local user to illicitly obtain elevated 397 * privileges. 398 */ 399 PROC_LOCK(p); 400 KASSERT((p->p_flag & P_INEXEC) == 0, 401 ("%s(): process already has P_INEXEC flag", __func__)); 402 p->p_flag |= P_INEXEC; 403 PROC_UNLOCK(p); 404 405 /* 406 * Initialize part of the common data 407 */ 408 bzero(imgp, sizeof(*imgp)); 409 imgp->proc = p; 410 imgp->attr = &attr; 411 imgp->args = args; 412 oldcred = p->p_ucred; 413 orig_osrel = p->p_osrel; 414 orig_fctl0 = p->p_fctl0; 415 416#ifdef MAC 417 error = mac_execve_enter(imgp, mac_p); 418 if (error) 419 goto exec_fail; 420#endif 421 422 /* 423 * Translate the file name. namei() returns a vnode pointer 424 * in ni_vp among other things. 425 * 426 * XXXAUDIT: It would be desirable to also audit the name of the 427 * interpreter if this is an interpreted binary. 428 */ 429 if (args->fname != NULL) { 430 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 431 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 432 } 433 434 SDT_PROBE1(proc, , , exec, args->fname); 435 436interpret: 437 if (args->fname != NULL) { 438#ifdef CAPABILITY_MODE 439 /* 440 * While capability mode can't reach this point via direct 441 * path arguments to execve(), we also don't allow 442 * interpreters to be used in capability mode (for now). 443 * Catch indirect lookups and return a permissions error. 444 */ 445 if (IN_CAPABILITY_MODE(td)) { 446 error = ECAPMODE; 447 goto exec_fail; 448 } 449#endif 450 error = namei(&nd); 451 if (error) 452 goto exec_fail; 453 454 newtextvp = nd.ni_vp; 455 imgp->vp = newtextvp; 456 } else { 457 AUDIT_ARG_FD(args->fd); 458 /* 459 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 460 */ 461 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 462 if (error) 463 goto exec_fail; 464 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 465 AUDIT_ARG_VNODE1(newtextvp); 466 imgp->vp = newtextvp; 467 } 468 469 /* 470 * Check file permissions. Also 'opens' file and sets its vnode to 471 * text mode. 472 */ 473 error = exec_check_permissions(imgp); 474 if (error) 475 goto exec_fail_dealloc; 476 477 imgp->object = imgp->vp->v_object; 478 if (imgp->object != NULL) 479 vm_object_reference(imgp->object); 480 481 error = exec_map_first_page(imgp); 482 if (error) 483 goto exec_fail_dealloc; 484 485 imgp->proc->p_osrel = 0; 486 imgp->proc->p_fctl0 = 0; 487 488 /* 489 * Implement image setuid/setgid. 490 * 491 * Determine new credentials before attempting image activators 492 * so that it can be used by process_exec handlers to determine 493 * credential/setid changes. 494 * 495 * Don't honor setuid/setgid if the filesystem prohibits it or if 496 * the process is being traced. 497 * 498 * We disable setuid/setgid/etc in capability mode on the basis 499 * that most setugid applications are not written with that 500 * environment in mind, and will therefore almost certainly operate 501 * incorrectly. In principle there's no reason that setugid 502 * applications might not be useful in capability mode, so we may want 503 * to reconsider this conservative design choice in the future. 504 * 505 * XXXMAC: For the time being, use NOSUID to also prohibit 506 * transitions on the file system. 507 */ 508 credential_changing = 0; 509 credential_changing |= (attr.va_mode & S_ISUID) && 510 oldcred->cr_uid != attr.va_uid; 511 credential_changing |= (attr.va_mode & S_ISGID) && 512 oldcred->cr_gid != attr.va_gid; 513#ifdef MAC 514 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 515 interpvplabel, imgp); 516 credential_changing |= will_transition; 517#endif 518 519 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 520 if (credential_changing) 521 imgp->proc->p_pdeathsig = 0; 522 523 if (credential_changing && 524#ifdef CAPABILITY_MODE 525 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 526#endif 527 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 528 (p->p_flag & P_TRACED) == 0) { 529 imgp->credential_setid = true; 530 VOP_UNLOCK(imgp->vp, 0); 531 imgp->newcred = crdup(oldcred); 532 if (attr.va_mode & S_ISUID) { 533 euip = uifind(attr.va_uid); 534 change_euid(imgp->newcred, euip); 535 } 536 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 537 if (attr.va_mode & S_ISGID) 538 change_egid(imgp->newcred, attr.va_gid); 539 /* 540 * Implement correct POSIX saved-id behavior. 541 * 542 * XXXMAC: Note that the current logic will save the 543 * uid and gid if a MAC domain transition occurs, even 544 * though maybe it shouldn't. 545 */ 546 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 547 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 548 } else { 549 /* 550 * Implement correct POSIX saved-id behavior. 551 * 552 * XXX: It's not clear that the existing behavior is 553 * POSIX-compliant. A number of sources indicate that the 554 * saved uid/gid should only be updated if the new ruid is 555 * not equal to the old ruid, or the new euid is not equal 556 * to the old euid and the new euid is not equal to the old 557 * ruid. The FreeBSD code always updates the saved uid/gid. 558 * Also, this code uses the new (replaced) euid and egid as 559 * the source, which may or may not be the right ones to use. 560 */ 561 if (oldcred->cr_svuid != oldcred->cr_uid || 562 oldcred->cr_svgid != oldcred->cr_gid) { 563 VOP_UNLOCK(imgp->vp, 0); 564 imgp->newcred = crdup(oldcred); 565 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 566 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 567 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 568 } 569 } 570 /* The new credentials are installed into the process later. */ 571 572 /* 573 * Do the best to calculate the full path to the image file. 574 */ 575 if (args->fname != NULL && args->fname[0] == '/') 576 imgp->execpath = args->fname; 577 else { 578 VOP_UNLOCK(imgp->vp, 0); 579 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 580 &imgp->freepath) != 0) 581 imgp->execpath = args->fname; 582 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 583 } 584 585 /* 586 * If the current process has a special image activator it 587 * wants to try first, call it. For example, emulating shell 588 * scripts differently. 589 */ 590 error = -1; 591 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 592 error = img_first(imgp); 593 594 /* 595 * Loop through the list of image activators, calling each one. 596 * An activator returns -1 if there is no match, 0 on success, 597 * and an error otherwise. 598 */ 599 for (i = 0; error == -1 && execsw[i]; ++i) { 600 if (execsw[i]->ex_imgact == NULL || 601 execsw[i]->ex_imgact == img_first) { 602 continue; 603 } 604 error = (*execsw[i]->ex_imgact)(imgp); 605 } 606 607 if (error) { 608 if (error == -1) 609 error = ENOEXEC; 610 goto exec_fail_dealloc; 611 } 612 613 /* 614 * Special interpreter operation, cleanup and loop up to try to 615 * activate the interpreter. 616 */ 617 if (imgp->interpreted) { 618 exec_unmap_first_page(imgp); 619 /* 620 * The text reference needs to be removed for scripts. 621 * There is a short period before we determine that 622 * something is a script where text reference is active. 623 * The vnode lock is held over this entire period 624 * so nothing should illegitimately be blocked. 625 */ 626 MPASS(imgp->textset); 627 VOP_UNSET_TEXT_CHECKED(newtextvp); 628 imgp->textset = false; 629 /* free name buffer and old vnode */ 630 if (args->fname != NULL) 631 NDFREE(&nd, NDF_ONLY_PNBUF); 632#ifdef MAC 633 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 634#endif 635 if (imgp->opened) { 636 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 637 imgp->opened = 0; 638 } 639 vput(newtextvp); 640 vm_object_deallocate(imgp->object); 641 imgp->object = NULL; 642 imgp->credential_setid = false; 643 if (imgp->newcred != NULL) { 644 crfree(imgp->newcred); 645 imgp->newcred = NULL; 646 } 647 imgp->execpath = NULL; 648 free(imgp->freepath, M_TEMP); 649 imgp->freepath = NULL; 650 /* set new name to that of the interpreter */ 651 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME, 652 UIO_SYSSPACE, imgp->interpreter_name, td); 653 args->fname = imgp->interpreter_name; 654 goto interpret; 655 } 656 657 /* 658 * NB: We unlock the vnode here because it is believed that none 659 * of the sv_copyout_strings/sv_fixup operations require the vnode. 660 */ 661 VOP_UNLOCK(imgp->vp, 0); 662 663 if (disallow_high_osrel && 664 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 665 error = ENOEXEC; 666 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 667 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 668 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 669 goto exec_fail_dealloc; 670 } 671 672 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 673 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 674 sys_cap_enter(td, NULL); 675 676 /* 677 * Copy out strings (args and env) and initialize stack base 678 */ 679 if (p->p_sysent->sv_copyout_strings) 680 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 681 else 682 stack_base = exec_copyout_strings(imgp); 683 684 /* 685 * If custom stack fixup routine present for this process 686 * let it do the stack setup. 687 * Else stuff argument count as first item on stack 688 */ 689 if (p->p_sysent->sv_fixup != NULL) 690 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 691 else 692 error = suword(--stack_base, imgp->args->argc) == 0 ? 693 0 : EFAULT; 694 if (error != 0) { 695 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 696 goto exec_fail_dealloc; 697 } 698 699 if (args->fdp != NULL) { 700 /* Install a brand new file descriptor table. */ 701 fdinstall_remapped(td, args->fdp); 702 args->fdp = NULL; 703 } else { 704 /* 705 * Keep on using the existing file descriptor table. For 706 * security and other reasons, the file descriptor table 707 * cannot be shared after an exec. 708 */ 709 fdunshare(td); 710 /* close files on exec */ 711 fdcloseexec(td); 712 } 713 714 /* 715 * Malloc things before we need locks. 716 */ 717 i = imgp->args->begin_envv - imgp->args->begin_argv; 718 /* Cache arguments if they fit inside our allowance */ 719 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 720 newargs = pargs_alloc(i); 721 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 722 } 723 724 /* 725 * For security and other reasons, signal handlers cannot 726 * be shared after an exec. The new process gets a copy of the old 727 * handlers. In execsigs(), the new process will have its signals 728 * reset. 729 */ 730 if (sigacts_shared(p->p_sigacts)) { 731 oldsigacts = p->p_sigacts; 732 newsigacts = sigacts_alloc(); 733 sigacts_copy(newsigacts, oldsigacts); 734 } 735 736 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 737 738 PROC_LOCK(p); 739 if (oldsigacts) 740 p->p_sigacts = newsigacts; 741 /* Stop profiling */ 742 stopprofclock(p); 743 744 /* reset caught signals */ 745 execsigs(p); 746 747 /* name this process - nameiexec(p, ndp) */ 748 bzero(p->p_comm, sizeof(p->p_comm)); 749 if (args->fname) 750 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 751 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 752 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 753 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 754 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 755#ifdef KTR 756 sched_clear_tdname(td); 757#endif 758 759 /* 760 * mark as execed, wakeup the process that vforked (if any) and tell 761 * it that it now has its own resources back 762 */ 763 p->p_flag |= P_EXEC; 764 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 765 p->p_flag2 &= ~P2_NOTRACE; 766 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 767 p->p_flag2 &= ~P2_STKGAP_DISABLE; 768 if (p->p_flag & P_PPWAIT) { 769 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 770 cv_broadcast(&p->p_pwait); 771 /* STOPs are no longer ignored, arrange for AST */ 772 signotify(td); 773 } 774 775 /* 776 * Implement image setuid/setgid installation. 777 */ 778 if (imgp->credential_setid) { 779 /* 780 * Turn off syscall tracing for set-id programs, except for 781 * root. Record any set-id flags first to make sure that 782 * we do not regain any tracing during a possible block. 783 */ 784 setsugid(p); 785 786#ifdef KTRACE 787 if (p->p_tracecred != NULL && 788 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 789 ktrprocexec(p, &tracecred, &tracevp); 790#endif 791 /* 792 * Close any file descriptors 0..2 that reference procfs, 793 * then make sure file descriptors 0..2 are in use. 794 * 795 * Both fdsetugidsafety() and fdcheckstd() may call functions 796 * taking sleepable locks, so temporarily drop our locks. 797 */ 798 PROC_UNLOCK(p); 799 VOP_UNLOCK(imgp->vp, 0); 800 fdsetugidsafety(td); 801 error = fdcheckstd(td); 802 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 803 if (error != 0) 804 goto exec_fail_dealloc; 805 PROC_LOCK(p); 806#ifdef MAC 807 if (will_transition) { 808 mac_vnode_execve_transition(oldcred, imgp->newcred, 809 imgp->vp, interpvplabel, imgp); 810 } 811#endif 812 } else { 813 if (oldcred->cr_uid == oldcred->cr_ruid && 814 oldcred->cr_gid == oldcred->cr_rgid) 815 p->p_flag &= ~P_SUGID; 816 } 817 /* 818 * Set the new credentials. 819 */ 820 if (imgp->newcred != NULL) { 821 proc_set_cred(p, imgp->newcred); 822 crfree(oldcred); 823 oldcred = NULL; 824 } 825 826 /* 827 * Store the vp for use in procfs. This vnode was referenced by namei 828 * or fgetvp_exec. 829 */ 830 oldtextvp = p->p_textvp; 831 p->p_textvp = newtextvp; 832 833#ifdef KDTRACE_HOOKS 834 /* 835 * Tell the DTrace fasttrap provider about the exec if it 836 * has declared an interest. 837 */ 838 if (dtrace_fasttrap_exec) 839 dtrace_fasttrap_exec(p); 840#endif 841 842 /* 843 * Notify others that we exec'd, and clear the P_INEXEC flag 844 * as we're now a bona fide freshly-execed process. 845 */ 846 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 847 p->p_flag &= ~P_INEXEC; 848 849 /* clear "fork but no exec" flag, as we _are_ execing */ 850 p->p_acflag &= ~AFORK; 851 852 /* 853 * Free any previous argument cache and replace it with 854 * the new argument cache, if any. 855 */ 856 oldargs = p->p_args; 857 p->p_args = newargs; 858 newargs = NULL; 859 860 PROC_UNLOCK(p); 861 862#ifdef HWPMC_HOOKS 863 /* 864 * Check if system-wide sampling is in effect or if the 865 * current process is using PMCs. If so, do exec() time 866 * processing. This processing needs to happen AFTER the 867 * P_INEXEC flag is cleared. 868 */ 869 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 870 VOP_UNLOCK(imgp->vp, 0); 871 pe.pm_credentialschanged = credential_changing; 872 pe.pm_entryaddr = imgp->entry_addr; 873 874 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 875 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 876 } 877#endif 878 879 /* Set values passed into the program in registers. */ 880 if (p->p_sysent->sv_setregs) 881 (*p->p_sysent->sv_setregs)(td, imgp, 882 (u_long)(uintptr_t)stack_base); 883 else 884 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 885 886 vfs_mark_atime(imgp->vp, td->td_ucred); 887 888 SDT_PROBE1(proc, , , exec__success, args->fname); 889 890exec_fail_dealloc: 891 if (error != 0) { 892 p->p_osrel = orig_osrel; 893 p->p_fctl0 = orig_fctl0; 894 } 895 896 if (imgp->firstpage != NULL) 897 exec_unmap_first_page(imgp); 898 899 if (imgp->vp != NULL) { 900 if (args->fname) 901 NDFREE(&nd, NDF_ONLY_PNBUF); 902 if (imgp->opened) 903 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 904 if (imgp->textset) 905 VOP_UNSET_TEXT_CHECKED(imgp->vp); 906 if (error != 0) 907 vput(imgp->vp); 908 else 909 VOP_UNLOCK(imgp->vp, 0); 910 } 911 912 if (imgp->object != NULL) 913 vm_object_deallocate(imgp->object); 914 915 free(imgp->freepath, M_TEMP); 916 917 if (error == 0) { 918 if (p->p_ptevents & PTRACE_EXEC) { 919 PROC_LOCK(p); 920 if (p->p_ptevents & PTRACE_EXEC) 921 td->td_dbgflags |= TDB_EXEC; 922 PROC_UNLOCK(p); 923 } 924 925 /* 926 * Stop the process here if its stop event mask has 927 * the S_EXEC bit set. 928 */ 929 STOPEVENT(p, S_EXEC, 0); 930 } else { 931exec_fail: 932 /* we're done here, clear P_INEXEC */ 933 PROC_LOCK(p); 934 p->p_flag &= ~P_INEXEC; 935 PROC_UNLOCK(p); 936 937 SDT_PROBE1(proc, , , exec__failure, error); 938 } 939 940 if (imgp->newcred != NULL && oldcred != NULL) 941 crfree(imgp->newcred); 942 943#ifdef MAC 944 mac_execve_exit(imgp); 945 mac_execve_interpreter_exit(interpvplabel); 946#endif 947 exec_free_args(args); 948 949 /* 950 * Handle deferred decrement of ref counts. 951 */ 952 if (oldtextvp != NULL) 953 vrele(oldtextvp); 954#ifdef KTRACE 955 if (tracevp != NULL) 956 vrele(tracevp); 957 if (tracecred != NULL) 958 crfree(tracecred); 959#endif 960 pargs_drop(oldargs); 961 pargs_drop(newargs); 962 if (oldsigacts != NULL) 963 sigacts_free(oldsigacts); 964 if (euip != NULL) 965 uifree(euip); 966 967 if (error && imgp->vmspace_destroyed) { 968 /* sorry, no more process anymore. exit gracefully */ 969 exec_cleanup(td, oldvmspace); 970 exit1(td, 0, SIGABRT); 971 /* NOT REACHED */ 972 } 973 974#ifdef KTRACE 975 if (error == 0) 976 ktrprocctor(p); 977#endif 978 979 /* 980 * We don't want cpu_set_syscall_retval() to overwrite any of 981 * the register values put in place by exec_setregs(). 982 * Implementations of cpu_set_syscall_retval() will leave 983 * registers unmodified when returning EJUSTRETURN. 984 */ 985 return (error == 0 ? EJUSTRETURN : error); 986} 987 988void 989exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 990{ 991 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 992 KASSERT(td->td_proc->p_vmspace != oldvmspace, 993 ("oldvmspace still used")); 994 vmspace_free(oldvmspace); 995 td->td_pflags &= ~TDP_EXECVMSPC; 996 } 997} 998 999int 1000exec_map_first_page(struct image_params *imgp) 1001{ 1002 int rv, i, after, initial_pagein; 1003 vm_page_t ma[VM_INITIAL_PAGEIN]; 1004 vm_object_t object; 1005 1006 if (imgp->firstpage != NULL) 1007 exec_unmap_first_page(imgp); 1008 1009 object = imgp->vp->v_object; 1010 if (object == NULL) 1011 return (EACCES); 1012 VM_OBJECT_WLOCK(object); 1013#if VM_NRESERVLEVEL > 0 1014 vm_object_color(object, 0); 1015#endif 1016 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 1017 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 1018 vm_page_xbusy(ma[0]); 1019 if (!vm_pager_has_page(object, 0, NULL, &after)) { 1020 vm_page_lock(ma[0]); 1021 if (!vm_page_wired(ma[0])) 1022 vm_page_free(ma[0]); 1023 else 1024 vm_page_xunbusy_maybelocked(ma[0]); 1025 vm_page_unlock(ma[0]); 1026 VM_OBJECT_WUNLOCK(object); 1027 return (EIO); 1028 } 1029 initial_pagein = min(after, VM_INITIAL_PAGEIN); 1030 KASSERT(initial_pagein <= object->size, 1031 ("%s: initial_pagein %d object->size %ju", 1032 __func__, initial_pagein, (uintmax_t )object->size)); 1033 for (i = 1; i < initial_pagein; i++) { 1034 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1035 if (ma[i]->valid) 1036 break; 1037 if (!vm_page_tryxbusy(ma[i])) 1038 break; 1039 } else { 1040 ma[i] = vm_page_alloc(object, i, 1041 VM_ALLOC_NORMAL); 1042 if (ma[i] == NULL) 1043 break; 1044 } 1045 } 1046 initial_pagein = i; 1047 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1048 if (rv != VM_PAGER_OK) { 1049 for (i = 0; i < initial_pagein; i++) { 1050 vm_page_lock(ma[i]); 1051 if (!vm_page_wired(ma[i])) 1052 vm_page_free(ma[i]); 1053 else 1054 vm_page_xunbusy_maybelocked(ma[i]); 1055 vm_page_unlock(ma[i]); 1056 } 1057 VM_OBJECT_WUNLOCK(object); 1058 return (EIO); 1059 } 1060 vm_page_xunbusy(ma[0]); 1061 for (i = 1; i < initial_pagein; i++) 1062 vm_page_readahead_finish(ma[i]); 1063 } 1064 vm_page_lock(ma[0]); 1065 vm_page_hold(ma[0]); 1066 vm_page_activate(ma[0]); 1067 vm_page_unlock(ma[0]); 1068 VM_OBJECT_WUNLOCK(object); 1069 1070 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1071 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1072 1073 return (0); 1074} 1075 1076void 1077exec_unmap_first_page(struct image_params *imgp) 1078{ 1079 vm_page_t m; 1080 1081 if (imgp->firstpage != NULL) { 1082 m = sf_buf_page(imgp->firstpage); 1083 sf_buf_free(imgp->firstpage); 1084 imgp->firstpage = NULL; 1085 vm_page_lock(m); 1086 vm_page_unhold(m); 1087 vm_page_unlock(m); 1088 } 1089} 1090 1091/* 1092 * Destroy old address space, and allocate a new stack. 1093 * The new stack is only sgrowsiz large because it is grown 1094 * automatically on a page fault. 1095 */ 1096int 1097exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1098{ 1099 int error; 1100 struct proc *p = imgp->proc; 1101 struct vmspace *vmspace = p->p_vmspace; 1102 vm_object_t obj; 1103 struct rlimit rlim_stack; 1104 vm_offset_t sv_minuser, stack_addr; 1105 vm_map_t map; 1106 u_long ssiz; 1107 1108 imgp->vmspace_destroyed = 1; 1109 imgp->sysent = sv; 1110 1111 umtx_exec(p); 1112 itimers_exec(p); 1113 1114 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1115 1116 /* 1117 * Blow away entire process VM, if address space not shared, 1118 * otherwise, create a new VM space so that other threads are 1119 * not disrupted 1120 */ 1121 map = &vmspace->vm_map; 1122 if (map_at_zero) 1123 sv_minuser = sv->sv_minuser; 1124 else 1125 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1126 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1127 vm_map_max(map) == sv->sv_maxuser && 1128 cpu_exec_vmspace_reuse(p, map)) { 1129 shmexit(vmspace); 1130 pmap_remove_pages(vmspace_pmap(vmspace)); 1131 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1132 /* 1133 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1134 * must be re-evaluated. 1135 */ 1136 vm_map_lock(map); 1137 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1138 MAP_ASLR_IGNSTART); 1139 vm_map_unlock(map); 1140 } else { 1141 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1142 if (error) 1143 return (error); 1144 vmspace = p->p_vmspace; 1145 map = &vmspace->vm_map; 1146 } 1147 map->flags |= imgp->map_flags; 1148 1149 /* Map a shared page */ 1150 obj = sv->sv_shared_page_obj; 1151 if (obj != NULL) { 1152 vm_object_reference(obj); 1153 error = vm_map_fixed(map, obj, 0, 1154 sv->sv_shared_page_base, sv->sv_shared_page_len, 1155 VM_PROT_READ | VM_PROT_EXECUTE, 1156 VM_PROT_READ | VM_PROT_EXECUTE, 1157 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1158 if (error != KERN_SUCCESS) { 1159 vm_object_deallocate(obj); 1160 return (vm_mmap_to_errno(error)); 1161 } 1162 } 1163 1164 /* Allocate a new stack */ 1165 if (imgp->stack_sz != 0) { 1166 ssiz = trunc_page(imgp->stack_sz); 1167 PROC_LOCK(p); 1168 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1169 PROC_UNLOCK(p); 1170 if (ssiz > rlim_stack.rlim_max) 1171 ssiz = rlim_stack.rlim_max; 1172 if (ssiz > rlim_stack.rlim_cur) { 1173 rlim_stack.rlim_cur = ssiz; 1174 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1175 } 1176 } else if (sv->sv_maxssiz != NULL) { 1177 ssiz = *sv->sv_maxssiz; 1178 } else { 1179 ssiz = maxssiz; 1180 } 1181 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1182 if (ssiz < imgp->eff_stack_sz) 1183 imgp->eff_stack_sz = ssiz; 1184 stack_addr = sv->sv_usrstack - ssiz; 1185 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1186 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1187 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1188 if (error != KERN_SUCCESS) 1189 return (vm_mmap_to_errno(error)); 1190 1191 /* 1192 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1193 * are still used to enforce the stack rlimit on the process stack. 1194 */ 1195 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1196 vmspace->vm_maxsaddr = (char *)stack_addr; 1197 1198 return (0); 1199} 1200 1201/* 1202 * Copy out argument and environment strings from the old process address 1203 * space into the temporary string buffer. 1204 */ 1205int 1206exec_copyin_args(struct image_args *args, char *fname, 1207 enum uio_seg segflg, char **argv, char **envv) 1208{ 1209 u_long argp, envp; 1210 int error; 1211 size_t length; 1212 1213 bzero(args, sizeof(*args)); 1214 if (argv == NULL) 1215 return (EFAULT); 1216 1217 /* 1218 * Allocate demand-paged memory for the file name, argument, and 1219 * environment strings. 1220 */ 1221 error = exec_alloc_args(args); 1222 if (error != 0) 1223 return (error); 1224 1225 /* 1226 * Copy the file name. 1227 */ 1228 if (fname != NULL) { 1229 args->fname = args->buf; 1230 error = (segflg == UIO_SYSSPACE) ? 1231 copystr(fname, args->fname, PATH_MAX, &length) : 1232 copyinstr(fname, args->fname, PATH_MAX, &length); 1233 if (error != 0) 1234 goto err_exit; 1235 } else 1236 length = 0; 1237 1238 args->begin_argv = args->buf + length; 1239 args->endp = args->begin_argv; 1240 args->stringspace = ARG_MAX; 1241 1242 /* 1243 * extract arguments first 1244 */ 1245 for (;;) { 1246 error = fueword(argv++, &argp); 1247 if (error == -1) { 1248 error = EFAULT; 1249 goto err_exit; 1250 } 1251 if (argp == 0) 1252 break; 1253 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1254 args->stringspace, &length); 1255 if (error != 0) { 1256 if (error == ENAMETOOLONG) 1257 error = E2BIG; 1258 goto err_exit; 1259 } 1260 args->stringspace -= length; 1261 args->endp += length; 1262 args->argc++; 1263 } 1264 1265 args->begin_envv = args->endp; 1266 1267 /* 1268 * extract environment strings 1269 */ 1270 if (envv) { 1271 for (;;) { 1272 error = fueword(envv++, &envp); 1273 if (error == -1) { 1274 error = EFAULT; 1275 goto err_exit; 1276 } 1277 if (envp == 0) 1278 break; 1279 error = copyinstr((void *)(uintptr_t)envp, 1280 args->endp, args->stringspace, &length); 1281 if (error != 0) { 1282 if (error == ENAMETOOLONG) 1283 error = E2BIG; 1284 goto err_exit; 1285 } 1286 args->stringspace -= length; 1287 args->endp += length; 1288 args->envc++; 1289 } 1290 } 1291 1292 return (0); 1293 1294err_exit: 1295 exec_free_args(args); 1296 return (error); 1297} 1298 1299int 1300exec_copyin_data_fds(struct thread *td, struct image_args *args, 1301 const void *data, size_t datalen, const int *fds, size_t fdslen) 1302{ 1303 struct filedesc *ofdp; 1304 const char *p; 1305 int *kfds; 1306 int error; 1307 1308 memset(args, '\0', sizeof(*args)); 1309 ofdp = td->td_proc->p_fd; 1310 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1311 return (E2BIG); 1312 error = exec_alloc_args(args); 1313 if (error != 0) 1314 return (error); 1315 1316 args->begin_argv = args->buf; 1317 args->stringspace = ARG_MAX; 1318 1319 if (datalen > 0) { 1320 /* 1321 * Argument buffer has been provided. Copy it into the 1322 * kernel as a single string and add a terminating null 1323 * byte. 1324 */ 1325 error = copyin(data, args->begin_argv, datalen); 1326 if (error != 0) 1327 goto err_exit; 1328 args->begin_argv[datalen] = '\0'; 1329 args->endp = args->begin_argv + datalen + 1; 1330 args->stringspace -= datalen + 1; 1331 1332 /* 1333 * Traditional argument counting. Count the number of 1334 * null bytes. 1335 */ 1336 for (p = args->begin_argv; p < args->endp; ++p) 1337 if (*p == '\0') 1338 ++args->argc; 1339 } else { 1340 /* No argument buffer provided. */ 1341 args->endp = args->begin_argv; 1342 } 1343 /* There are no environment variables. */ 1344 args->begin_envv = args->endp; 1345 1346 /* Create new file descriptor table. */ 1347 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1348 error = copyin(fds, kfds, fdslen * sizeof(int)); 1349 if (error != 0) { 1350 free(kfds, M_TEMP); 1351 goto err_exit; 1352 } 1353 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1354 free(kfds, M_TEMP); 1355 if (error != 0) 1356 goto err_exit; 1357 1358 return (0); 1359err_exit: 1360 exec_free_args(args); 1361 return (error); 1362} 1363 1364struct exec_args_kva { 1365 vm_offset_t addr; 1366 u_int gen; 1367 SLIST_ENTRY(exec_args_kva) next; 1368}; 1369 1370DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1371 1372static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1373static struct mtx exec_args_kva_mtx; 1374static u_int exec_args_gen; 1375 1376static void 1377exec_prealloc_args_kva(void *arg __unused) 1378{ 1379 struct exec_args_kva *argkva; 1380 u_int i; 1381 1382 SLIST_INIT(&exec_args_kva_freelist); 1383 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1384 for (i = 0; i < exec_map_entries; i++) { 1385 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1386 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1387 argkva->gen = exec_args_gen; 1388 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1389 } 1390} 1391SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1392 1393static vm_offset_t 1394exec_alloc_args_kva(void **cookie) 1395{ 1396 struct exec_args_kva *argkva; 1397 1398 argkva = (void *)atomic_readandclear_ptr( 1399 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1400 if (argkva == NULL) { 1401 mtx_lock(&exec_args_kva_mtx); 1402 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1403 (void)mtx_sleep(&exec_args_kva_freelist, 1404 &exec_args_kva_mtx, 0, "execkva", 0); 1405 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1406 mtx_unlock(&exec_args_kva_mtx); 1407 } 1408 *(struct exec_args_kva **)cookie = argkva; 1409 return (argkva->addr); 1410} 1411 1412static void 1413exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1414{ 1415 vm_offset_t base; 1416 1417 base = argkva->addr; 1418 if (argkva->gen != gen) { 1419 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1420 MADV_FREE); 1421 argkva->gen = gen; 1422 } 1423 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1424 (uintptr_t)NULL, (uintptr_t)argkva)) { 1425 mtx_lock(&exec_args_kva_mtx); 1426 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1427 wakeup_one(&exec_args_kva_freelist); 1428 mtx_unlock(&exec_args_kva_mtx); 1429 } 1430} 1431 1432static void 1433exec_free_args_kva(void *cookie) 1434{ 1435 1436 exec_release_args_kva(cookie, exec_args_gen); 1437} 1438 1439static void 1440exec_args_kva_lowmem(void *arg __unused) 1441{ 1442 SLIST_HEAD(, exec_args_kva) head; 1443 struct exec_args_kva *argkva; 1444 u_int gen; 1445 int i; 1446 1447 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1448 1449 /* 1450 * Force an madvise of each KVA range. Any currently allocated ranges 1451 * will have MADV_FREE applied once they are freed. 1452 */ 1453 SLIST_INIT(&head); 1454 mtx_lock(&exec_args_kva_mtx); 1455 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1456 mtx_unlock(&exec_args_kva_mtx); 1457 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1458 SLIST_REMOVE_HEAD(&head, next); 1459 exec_release_args_kva(argkva, gen); 1460 } 1461 1462 CPU_FOREACH(i) { 1463 argkva = (void *)atomic_readandclear_ptr( 1464 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1465 if (argkva != NULL) 1466 exec_release_args_kva(argkva, gen); 1467 } 1468} 1469EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1470 EVENTHANDLER_PRI_ANY); 1471 1472/* 1473 * Allocate temporary demand-paged, zero-filled memory for the file name, 1474 * argument, and environment strings. 1475 */ 1476int 1477exec_alloc_args(struct image_args *args) 1478{ 1479 1480 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1481 return (0); 1482} 1483 1484void 1485exec_free_args(struct image_args *args) 1486{ 1487 1488 if (args->buf != NULL) { 1489 exec_free_args_kva(args->bufkva); 1490 args->buf = NULL; 1491 } 1492 if (args->fname_buf != NULL) { 1493 free(args->fname_buf, M_TEMP); 1494 args->fname_buf = NULL; 1495 } 1496 if (args->fdp != NULL) 1497 fdescfree_remapped(args->fdp); 1498} 1499 1500void 1501exec_stackgap(struct image_params *imgp, uintptr_t *dp) 1502{ 1503 if (imgp->sysent->sv_stackgap == NULL || 1504 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE | 1505 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 || 1506 (imgp->map_flags & MAP_ASLR) == 0) 1507 return; 1508 imgp->sysent->sv_stackgap(imgp, (u_long *)dp); 1509} 1510 1511/* 1512 * Copy strings out to the new process address space, constructing new arg 1513 * and env vector tables. Return a pointer to the base so that it can be used 1514 * as the initial stack pointer. 1515 */ 1516register_t * 1517exec_copyout_strings(struct image_params *imgp) 1518{ 1519 int argc, envc; 1520 char **vectp; 1521 char *stringp; 1522 uintptr_t destp; 1523 register_t *stack_base; 1524 struct ps_strings *arginfo; 1525 struct proc *p; 1526 size_t execpath_len; 1527 int szsigcode, szps; 1528 char canary[sizeof(long) * 8]; 1529 1530 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1531 /* 1532 * Calculate string base and vector table pointers. 1533 * Also deal with signal trampoline code for this exec type. 1534 */ 1535 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1536 execpath_len = strlen(imgp->execpath) + 1; 1537 else 1538 execpath_len = 0; 1539 p = imgp->proc; 1540 szsigcode = 0; 1541 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1542 if (p->p_sysent->sv_sigcode_base == 0) { 1543 if (p->p_sysent->sv_szsigcode != NULL) 1544 szsigcode = *(p->p_sysent->sv_szsigcode); 1545 } 1546 destp = (uintptr_t)arginfo; 1547 1548 /* 1549 * install sigcode 1550 */ 1551 if (szsigcode != 0) { 1552 destp -= szsigcode; 1553 destp = rounddown2(destp, sizeof(void *)); 1554 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1555 } 1556 1557 /* 1558 * Copy the image path for the rtld. 1559 */ 1560 if (execpath_len != 0) { 1561 destp -= execpath_len; 1562 destp = rounddown2(destp, sizeof(void *)); 1563 imgp->execpathp = destp; 1564 copyout(imgp->execpath, (void *)destp, execpath_len); 1565 } 1566 1567 /* 1568 * Prepare the canary for SSP. 1569 */ 1570 arc4rand(canary, sizeof(canary), 0); 1571 destp -= sizeof(canary); 1572 imgp->canary = destp; 1573 copyout(canary, (void *)destp, sizeof(canary)); 1574 imgp->canarylen = sizeof(canary); 1575 1576 /* 1577 * Prepare the pagesizes array. 1578 */ 1579 destp -= szps; 1580 destp = rounddown2(destp, sizeof(void *)); 1581 imgp->pagesizes = destp; 1582 copyout(pagesizes, (void *)destp, szps); 1583 imgp->pagesizeslen = szps; 1584 1585 destp -= ARG_MAX - imgp->args->stringspace; 1586 destp = rounddown2(destp, sizeof(void *)); 1587 1588 exec_stackgap(imgp, &destp); 1589 vectp = (char **)destp; 1590 1591 if (imgp->auxargs) { 1592 /* 1593 * Allocate room on the stack for the ELF auxargs 1594 * array. It has up to AT_COUNT entries. 1595 */ 1596 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 1597 sizeof(*vectp)); 1598 } 1599 1600 /* 1601 * Allocate room for the argv[] and env vectors including the 1602 * terminating NULL pointers. 1603 */ 1604 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1605 1606 /* 1607 * vectp also becomes our initial stack base 1608 */ 1609 stack_base = (register_t *)vectp; 1610 1611 stringp = imgp->args->begin_argv; 1612 argc = imgp->args->argc; 1613 envc = imgp->args->envc; 1614 1615 /* 1616 * Copy out strings - arguments and environment. 1617 */ 1618 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1619 1620 /* 1621 * Fill in "ps_strings" struct for ps, w, etc. 1622 */ 1623 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1624 suword32(&arginfo->ps_nargvstr, argc); 1625 1626 /* 1627 * Fill in argument portion of vector table. 1628 */ 1629 for (; argc > 0; --argc) { 1630 suword(vectp++, (long)(intptr_t)destp); 1631 while (*stringp++ != 0) 1632 destp++; 1633 destp++; 1634 } 1635 1636 /* a null vector table pointer separates the argp's from the envp's */ 1637 suword(vectp++, 0); 1638 1639 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1640 suword32(&arginfo->ps_nenvstr, envc); 1641 1642 /* 1643 * Fill in environment portion of vector table. 1644 */ 1645 for (; envc > 0; --envc) { 1646 suword(vectp++, (long)(intptr_t)destp); 1647 while (*stringp++ != 0) 1648 destp++; 1649 destp++; 1650 } 1651 1652 /* end of vector table is a null pointer */ 1653 suword(vectp, 0); 1654 1655 return (stack_base); 1656} 1657 1658/* 1659 * Check permissions of file to execute. 1660 * Called with imgp->vp locked. 1661 * Return 0 for success or error code on failure. 1662 */ 1663int 1664exec_check_permissions(struct image_params *imgp) 1665{ 1666 struct vnode *vp = imgp->vp; 1667 struct vattr *attr = imgp->attr; 1668 struct thread *td; 1669 int error; 1670 1671 td = curthread; 1672 1673 /* Get file attributes */ 1674 error = VOP_GETATTR(vp, attr, td->td_ucred); 1675 if (error) 1676 return (error); 1677 1678#ifdef MAC 1679 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1680 if (error) 1681 return (error); 1682#endif 1683 1684 /* 1685 * 1) Check if file execution is disabled for the filesystem that 1686 * this file resides on. 1687 * 2) Ensure that at least one execute bit is on. Otherwise, a 1688 * privileged user will always succeed, and we don't want this 1689 * to happen unless the file really is executable. 1690 * 3) Ensure that the file is a regular file. 1691 */ 1692 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1693 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1694 (attr->va_type != VREG)) 1695 return (EACCES); 1696 1697 /* 1698 * Zero length files can't be exec'd 1699 */ 1700 if (attr->va_size == 0) 1701 return (ENOEXEC); 1702 1703 /* 1704 * Check for execute permission to file based on current credentials. 1705 */ 1706 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1707 if (error) 1708 return (error); 1709 1710 /* 1711 * Check number of open-for-writes on the file and deny execution 1712 * if there are any. 1713 * 1714 * Add a text reference now so no one can write to the 1715 * executable while we're activating it. 1716 * 1717 * Remember if this was set before and unset it in case this is not 1718 * actually an executable image. 1719 */ 1720 error = VOP_SET_TEXT(vp); 1721 if (error != 0) 1722 return (error); 1723 imgp->textset = true; 1724 1725 /* 1726 * Call filesystem specific open routine (which does nothing in the 1727 * general case). 1728 */ 1729 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1730 if (error == 0) 1731 imgp->opened = 1; 1732 return (error); 1733} 1734 1735/* 1736 * Exec handler registration 1737 */ 1738int 1739exec_register(const struct execsw *execsw_arg) 1740{ 1741 const struct execsw **es, **xs, **newexecsw; 1742 u_int count = 2; /* New slot and trailing NULL */ 1743 1744 if (execsw) 1745 for (es = execsw; *es; es++) 1746 count++; 1747 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1748 xs = newexecsw; 1749 if (execsw) 1750 for (es = execsw; *es; es++) 1751 *xs++ = *es; 1752 *xs++ = execsw_arg; 1753 *xs = NULL; 1754 if (execsw) 1755 free(execsw, M_TEMP); 1756 execsw = newexecsw; 1757 return (0); 1758} 1759 1760int 1761exec_unregister(const struct execsw *execsw_arg) 1762{ 1763 const struct execsw **es, **xs, **newexecsw; 1764 int count = 1; 1765 1766 if (execsw == NULL) 1767 panic("unregister with no handlers left?\n"); 1768 1769 for (es = execsw; *es; es++) { 1770 if (*es == execsw_arg) 1771 break; 1772 } 1773 if (*es == NULL) 1774 return (ENOENT); 1775 for (es = execsw; *es; es++) 1776 if (*es != execsw_arg) 1777 count++; 1778 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1779 xs = newexecsw; 1780 for (es = execsw; *es; es++) 1781 if (*es != execsw_arg) 1782 *xs++ = *es; 1783 *xs = NULL; 1784 if (execsw) 1785 free(execsw, M_TEMP); 1786 execsw = newexecsw; 1787 return (0); 1788} 1789