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/asan.h> 41 #include <sys/capsicum.h> 42 #include <sys/compressor.h> 43 #include <sys/eventhandler.h> 44 #include <sys/exec.h> 45 #include <sys/fcntl.h> 46 #include <sys/filedesc.h> 47 #include <sys/imgact.h> 48 #include <sys/imgact_elf.h> 49 #include <sys/kernel.h> 50 #include <sys/lock.h> 51 #include <sys/malloc.h> 52 #include <sys/mman.h> 53 #include <sys/mount.h> 54 #include <sys/mutex.h> 55 #include <sys/namei.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/ptrace.h> 59 #include <sys/resourcevar.h> 60 #include <sys/rwlock.h> 61 #include <sys/sched.h> 62 #include <sys/sdt.h> 63 #include <sys/sf_buf.h> 64 #include <sys/shm.h> 65 #include <sys/signalvar.h> 66 #include <sys/smp.h> 67 #include <sys/stat.h> 68 #include <sys/syscallsubr.h> 69 #include <sys/sysctl.h> 70 #include <sys/sysent.h> 71 #include <sys/sysproto.h> 72 #include <sys/timers.h> 73 #include <sys/umtx.h> 74 #include <sys/vnode.h> 75 #include <sys/wait.h> 76 #ifdef KTRACE 77 #include <sys/ktrace.h> 78 #endif 79 80 #include <vm/vm.h> 81 #include <vm/vm_param.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_page.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_kern.h> 86 #include <vm/vm_extern.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_pager.h> 89 90 #ifdef HWPMC_HOOKS 91 #include <sys/pmckern.h> 92 #endif 93 94 #include <machine/reg.h> 95 96 #include <security/audit/audit.h> 97 #include <security/mac/mac_framework.h> 98 99 #ifdef KDTRACE_HOOKS 100 #include <sys/dtrace_bsd.h> 101 dtrace_execexit_func_t dtrace_fasttrap_exec; 102 #endif 103 104 SDT_PROVIDER_DECLARE(proc); 105 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 106 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 107 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 108 109 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 110 111 int coredump_pack_fileinfo = 1; 112 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 113 &coredump_pack_fileinfo, 0, 114 "Enable file path packing in 'procstat -f' coredump notes"); 115 116 int coredump_pack_vmmapinfo = 1; 117 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 118 &coredump_pack_vmmapinfo, 0, 119 "Enable file path packing in 'procstat -v' coredump notes"); 120 121 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 122 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 123 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 124 static int do_execve(struct thread *td, struct image_args *args, 125 struct mac *mac_p, struct vmspace *oldvmspace); 126 127 /* XXX This should be vm_size_t. */ 128 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 129 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", 130 "Location of process' ps_strings structure"); 131 132 /* XXX This should be vm_size_t. */ 133 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 134 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", 135 "Top of process stack"); 136 137 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 138 NULL, 0, sysctl_kern_stackprot, "I", 139 "Stack memory permissions"); 140 141 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 142 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 143 &ps_arg_cache_limit, 0, 144 "Process' command line characters cache limit"); 145 146 static int disallow_high_osrel; 147 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 148 &disallow_high_osrel, 0, 149 "Disallow execution of binaries built for higher version of the world"); 150 151 static int map_at_zero = 0; 152 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 153 "Permit processes to map an object at virtual address 0."); 154 155 static int 156 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 157 { 158 struct proc *p; 159 int error; 160 161 p = curproc; 162 #ifdef SCTL_MASK32 163 if (req->flags & SCTL_MASK32) { 164 unsigned int val; 165 val = (unsigned int)p->p_sysent->sv_psstrings; 166 error = SYSCTL_OUT(req, &val, sizeof(val)); 167 } else 168 #endif 169 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 170 sizeof(p->p_sysent->sv_psstrings)); 171 return error; 172 } 173 174 static int 175 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 176 { 177 struct proc *p; 178 int error; 179 180 p = curproc; 181 #ifdef SCTL_MASK32 182 if (req->flags & SCTL_MASK32) { 183 unsigned int val; 184 val = (unsigned int)p->p_sysent->sv_usrstack; 185 error = SYSCTL_OUT(req, &val, sizeof(val)); 186 } else 187 #endif 188 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 189 sizeof(p->p_sysent->sv_usrstack)); 190 return error; 191 } 192 193 static int 194 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 195 { 196 struct proc *p; 197 198 p = curproc; 199 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 200 sizeof(p->p_sysent->sv_stackprot))); 201 } 202 203 /* 204 * Each of the items is a pointer to a `const struct execsw', hence the 205 * double pointer here. 206 */ 207 static const struct execsw **execsw; 208 209 #ifndef _SYS_SYSPROTO_H_ 210 struct execve_args { 211 char *fname; 212 char **argv; 213 char **envv; 214 }; 215 #endif 216 217 int 218 sys_execve(struct thread *td, struct execve_args *uap) 219 { 220 struct image_args args; 221 struct vmspace *oldvmspace; 222 int error; 223 224 error = pre_execve(td, &oldvmspace); 225 if (error != 0) 226 return (error); 227 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 228 uap->argv, uap->envv); 229 if (error == 0) 230 error = kern_execve(td, &args, NULL, oldvmspace); 231 post_execve(td, error, oldvmspace); 232 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 233 return (error); 234 } 235 236 #ifndef _SYS_SYSPROTO_H_ 237 struct fexecve_args { 238 int fd; 239 char **argv; 240 char **envv; 241 }; 242 #endif 243 int 244 sys_fexecve(struct thread *td, struct fexecve_args *uap) 245 { 246 struct image_args args; 247 struct vmspace *oldvmspace; 248 int error; 249 250 error = pre_execve(td, &oldvmspace); 251 if (error != 0) 252 return (error); 253 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 254 uap->argv, uap->envv); 255 if (error == 0) { 256 args.fd = uap->fd; 257 error = kern_execve(td, &args, NULL, oldvmspace); 258 } 259 post_execve(td, error, oldvmspace); 260 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 261 return (error); 262 } 263 264 #ifndef _SYS_SYSPROTO_H_ 265 struct __mac_execve_args { 266 char *fname; 267 char **argv; 268 char **envv; 269 struct mac *mac_p; 270 }; 271 #endif 272 273 int 274 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 275 { 276 #ifdef MAC 277 struct image_args args; 278 struct vmspace *oldvmspace; 279 int error; 280 281 error = pre_execve(td, &oldvmspace); 282 if (error != 0) 283 return (error); 284 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 285 uap->argv, uap->envv); 286 if (error == 0) 287 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 288 post_execve(td, error, oldvmspace); 289 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 290 return (error); 291 #else 292 return (ENOSYS); 293 #endif 294 } 295 296 int 297 pre_execve(struct thread *td, struct vmspace **oldvmspace) 298 { 299 struct proc *p; 300 int error; 301 302 KASSERT(td == curthread, ("non-current thread %p", td)); 303 error = 0; 304 p = td->td_proc; 305 if ((p->p_flag & P_HADTHREADS) != 0) { 306 PROC_LOCK(p); 307 if (thread_single(p, SINGLE_BOUNDARY) != 0) 308 error = ERESTART; 309 PROC_UNLOCK(p); 310 } 311 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 312 ("nested execve")); 313 *oldvmspace = p->p_vmspace; 314 return (error); 315 } 316 317 void 318 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 319 { 320 struct proc *p; 321 322 KASSERT(td == curthread, ("non-current thread %p", td)); 323 p = td->td_proc; 324 if ((p->p_flag & P_HADTHREADS) != 0) { 325 PROC_LOCK(p); 326 /* 327 * If success, we upgrade to SINGLE_EXIT state to 328 * force other threads to suicide. 329 */ 330 if (error == EJUSTRETURN) 331 thread_single(p, SINGLE_EXIT); 332 else 333 thread_single_end(p, SINGLE_BOUNDARY); 334 PROC_UNLOCK(p); 335 } 336 exec_cleanup(td, oldvmspace); 337 } 338 339 /* 340 * kern_execve() has the astonishing property of not always returning to 341 * the caller. If sufficiently bad things happen during the call to 342 * do_execve(), it can end up calling exit1(); as a result, callers must 343 * avoid doing anything which they might need to undo (e.g., allocating 344 * memory). 345 */ 346 int 347 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 348 struct vmspace *oldvmspace) 349 { 350 351 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 352 exec_args_get_begin_envv(args) - args->begin_argv); 353 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 354 args->endp - exec_args_get_begin_envv(args)); 355 return (do_execve(td, args, mac_p, oldvmspace)); 356 } 357 358 static void 359 execve_nosetid(struct image_params *imgp) 360 { 361 imgp->credential_setid = false; 362 if (imgp->newcred != NULL) { 363 crfree(imgp->newcred); 364 imgp->newcred = NULL; 365 } 366 } 367 368 /* 369 * In-kernel implementation of execve(). All arguments are assumed to be 370 * userspace pointers from the passed thread. 371 */ 372 static int 373 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 374 struct vmspace *oldvmspace) 375 { 376 struct proc *p = td->td_proc; 377 struct nameidata nd; 378 struct ucred *oldcred; 379 struct uidinfo *euip = NULL; 380 uintptr_t stack_base; 381 struct image_params image_params, *imgp; 382 struct vattr attr; 383 int (*img_first)(struct image_params *); 384 struct pargs *oldargs = NULL, *newargs = NULL; 385 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 386 #ifdef KTRACE 387 struct ktr_io_params *kiop; 388 #endif 389 struct vnode *oldtextvp = NULL, *newtextvp; 390 int credential_changing; 391 #ifdef MAC 392 struct label *interpvplabel = NULL; 393 int will_transition; 394 #endif 395 #ifdef HWPMC_HOOKS 396 struct pmckern_procexec pe; 397 #endif 398 int error, i, orig_osrel; 399 uint32_t orig_fctl0; 400 Elf_Brandinfo *orig_brandinfo; 401 static const char fexecv_proc_title[] = "(fexecv)"; 402 403 imgp = &image_params; 404 #ifdef KTRACE 405 kiop = NULL; 406 #endif 407 408 /* 409 * Lock the process and set the P_INEXEC flag to indicate that 410 * it should be left alone until we're done here. This is 411 * necessary to avoid race conditions - e.g. in ptrace() - 412 * that might allow a local user to illicitly obtain elevated 413 * privileges. 414 */ 415 PROC_LOCK(p); 416 KASSERT((p->p_flag & P_INEXEC) == 0, 417 ("%s(): process already has P_INEXEC flag", __func__)); 418 p->p_flag |= P_INEXEC; 419 PROC_UNLOCK(p); 420 421 /* 422 * Initialize part of the common data 423 */ 424 bzero(imgp, sizeof(*imgp)); 425 imgp->proc = p; 426 imgp->attr = &attr; 427 imgp->args = args; 428 oldcred = p->p_ucred; 429 orig_osrel = p->p_osrel; 430 orig_fctl0 = p->p_fctl0; 431 orig_brandinfo = p->p_elf_brandinfo; 432 433 #ifdef MAC 434 error = mac_execve_enter(imgp, mac_p); 435 if (error) 436 goto exec_fail; 437 #endif 438 439 /* 440 * Translate the file name. namei() returns a vnode pointer 441 * in ni_vp among other things. 442 * 443 * XXXAUDIT: It would be desirable to also audit the name of the 444 * interpreter if this is an interpreted binary. 445 */ 446 if (args->fname != NULL) { 447 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 448 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 449 } 450 451 SDT_PROBE1(proc, , , exec, args->fname); 452 453 interpret: 454 if (args->fname != NULL) { 455 #ifdef CAPABILITY_MODE 456 /* 457 * While capability mode can't reach this point via direct 458 * path arguments to execve(), we also don't allow 459 * interpreters to be used in capability mode (for now). 460 * Catch indirect lookups and return a permissions error. 461 */ 462 if (IN_CAPABILITY_MODE(td)) { 463 error = ECAPMODE; 464 goto exec_fail; 465 } 466 #endif 467 error = namei(&nd); 468 if (error) 469 goto exec_fail; 470 471 newtextvp = nd.ni_vp; 472 imgp->vp = newtextvp; 473 } else { 474 AUDIT_ARG_FD(args->fd); 475 /* 476 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 477 */ 478 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 479 if (error) 480 goto exec_fail; 481 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 482 AUDIT_ARG_VNODE1(newtextvp); 483 imgp->vp = newtextvp; 484 } 485 486 /* 487 * Check file permissions. Also 'opens' file and sets its vnode to 488 * text mode. 489 */ 490 error = exec_check_permissions(imgp); 491 if (error) 492 goto exec_fail_dealloc; 493 494 imgp->object = imgp->vp->v_object; 495 if (imgp->object != NULL) 496 vm_object_reference(imgp->object); 497 498 error = exec_map_first_page(imgp); 499 if (error) 500 goto exec_fail_dealloc; 501 502 imgp->proc->p_osrel = 0; 503 imgp->proc->p_fctl0 = 0; 504 imgp->proc->p_elf_brandinfo = NULL; 505 506 /* 507 * Implement image setuid/setgid. 508 * 509 * Determine new credentials before attempting image activators 510 * so that it can be used by process_exec handlers to determine 511 * credential/setid changes. 512 * 513 * Don't honor setuid/setgid if the filesystem prohibits it or if 514 * the process is being traced. 515 * 516 * We disable setuid/setgid/etc in capability mode on the basis 517 * that most setugid applications are not written with that 518 * environment in mind, and will therefore almost certainly operate 519 * incorrectly. In principle there's no reason that setugid 520 * applications might not be useful in capability mode, so we may want 521 * to reconsider this conservative design choice in the future. 522 * 523 * XXXMAC: For the time being, use NOSUID to also prohibit 524 * transitions on the file system. 525 */ 526 credential_changing = 0; 527 credential_changing |= (attr.va_mode & S_ISUID) && 528 oldcred->cr_uid != attr.va_uid; 529 credential_changing |= (attr.va_mode & S_ISGID) && 530 oldcred->cr_gid != attr.va_gid; 531 #ifdef MAC 532 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 533 interpvplabel, imgp); 534 credential_changing |= will_transition; 535 #endif 536 537 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 538 if (credential_changing) 539 imgp->proc->p_pdeathsig = 0; 540 541 if (credential_changing && 542 #ifdef CAPABILITY_MODE 543 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 544 #endif 545 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 546 (p->p_flag & P_TRACED) == 0) { 547 imgp->credential_setid = true; 548 VOP_UNLOCK(imgp->vp); 549 imgp->newcred = crdup(oldcred); 550 if (attr.va_mode & S_ISUID) { 551 euip = uifind(attr.va_uid); 552 change_euid(imgp->newcred, euip); 553 } 554 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 555 if (attr.va_mode & S_ISGID) 556 change_egid(imgp->newcred, attr.va_gid); 557 /* 558 * Implement correct POSIX saved-id behavior. 559 * 560 * XXXMAC: Note that the current logic will save the 561 * uid and gid if a MAC domain transition occurs, even 562 * though maybe it shouldn't. 563 */ 564 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 565 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 566 } else { 567 /* 568 * Implement correct POSIX saved-id behavior. 569 * 570 * XXX: It's not clear that the existing behavior is 571 * POSIX-compliant. A number of sources indicate that the 572 * saved uid/gid should only be updated if the new ruid is 573 * not equal to the old ruid, or the new euid is not equal 574 * to the old euid and the new euid is not equal to the old 575 * ruid. The FreeBSD code always updates the saved uid/gid. 576 * Also, this code uses the new (replaced) euid and egid as 577 * the source, which may or may not be the right ones to use. 578 */ 579 if (oldcred->cr_svuid != oldcred->cr_uid || 580 oldcred->cr_svgid != oldcred->cr_gid) { 581 VOP_UNLOCK(imgp->vp); 582 imgp->newcred = crdup(oldcred); 583 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 584 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 585 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 586 } 587 } 588 /* The new credentials are installed into the process later. */ 589 590 /* 591 * Do the best to calculate the full path to the image file. 592 */ 593 if (args->fname != NULL && args->fname[0] == '/') 594 imgp->execpath = args->fname; 595 else { 596 VOP_UNLOCK(imgp->vp); 597 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0) 598 imgp->execpath = args->fname; 599 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 600 } 601 602 /* 603 * If the current process has a special image activator it 604 * wants to try first, call it. For example, emulating shell 605 * scripts differently. 606 */ 607 error = -1; 608 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 609 error = img_first(imgp); 610 611 /* 612 * Loop through the list of image activators, calling each one. 613 * An activator returns -1 if there is no match, 0 on success, 614 * and an error otherwise. 615 */ 616 for (i = 0; error == -1 && execsw[i]; ++i) { 617 if (execsw[i]->ex_imgact == NULL || 618 execsw[i]->ex_imgact == img_first) { 619 continue; 620 } 621 error = (*execsw[i]->ex_imgact)(imgp); 622 } 623 624 if (error) { 625 if (error == -1) 626 error = ENOEXEC; 627 goto exec_fail_dealloc; 628 } 629 630 /* 631 * Special interpreter operation, cleanup and loop up to try to 632 * activate the interpreter. 633 */ 634 if (imgp->interpreted) { 635 exec_unmap_first_page(imgp); 636 /* 637 * The text reference needs to be removed for scripts. 638 * There is a short period before we determine that 639 * something is a script where text reference is active. 640 * The vnode lock is held over this entire period 641 * so nothing should illegitimately be blocked. 642 */ 643 MPASS(imgp->textset); 644 VOP_UNSET_TEXT_CHECKED(newtextvp); 645 imgp->textset = false; 646 /* free name buffer and old vnode */ 647 if (args->fname != NULL) 648 NDFREE(&nd, NDF_ONLY_PNBUF); 649 #ifdef MAC 650 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 651 #endif 652 if (imgp->opened) { 653 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 654 imgp->opened = 0; 655 } 656 vput(newtextvp); 657 vm_object_deallocate(imgp->object); 658 imgp->object = NULL; 659 execve_nosetid(imgp); 660 imgp->execpath = NULL; 661 free(imgp->freepath, M_TEMP); 662 imgp->freepath = NULL; 663 /* set new name to that of the interpreter */ 664 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 665 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 666 args->fname = imgp->interpreter_name; 667 goto interpret; 668 } 669 670 /* 671 * NB: We unlock the vnode here because it is believed that none 672 * of the sv_copyout_strings/sv_fixup operations require the vnode. 673 */ 674 VOP_UNLOCK(imgp->vp); 675 676 if (disallow_high_osrel && 677 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 678 error = ENOEXEC; 679 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 680 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 681 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 682 goto exec_fail_dealloc; 683 } 684 685 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 686 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 687 sys_cap_enter(td, NULL); 688 689 /* 690 * Copy out strings (args and env) and initialize stack base. 691 */ 692 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 693 if (error != 0) { 694 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 695 goto exec_fail_dealloc; 696 } 697 698 /* 699 * Stack setup. 700 */ 701 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 702 if (error != 0) { 703 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 704 goto exec_fail_dealloc; 705 } 706 707 if (args->fdp != NULL) { 708 /* Install a brand new file descriptor table. */ 709 fdinstall_remapped(td, args->fdp); 710 args->fdp = NULL; 711 } else { 712 /* 713 * Keep on using the existing file descriptor table. For 714 * security and other reasons, the file descriptor table 715 * cannot be shared after an exec. 716 */ 717 fdunshare(td); 718 pdunshare(td); 719 /* close files on exec */ 720 fdcloseexec(td); 721 } 722 723 /* 724 * Malloc things before we need locks. 725 */ 726 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 727 /* Cache arguments if they fit inside our allowance */ 728 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 729 newargs = pargs_alloc(i); 730 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 731 } 732 733 /* 734 * For security and other reasons, signal handlers cannot 735 * be shared after an exec. The new process gets a copy of the old 736 * handlers. In execsigs(), the new process will have its signals 737 * reset. 738 */ 739 if (sigacts_shared(p->p_sigacts)) { 740 oldsigacts = p->p_sigacts; 741 newsigacts = sigacts_alloc(); 742 sigacts_copy(newsigacts, oldsigacts); 743 } 744 745 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 746 747 PROC_LOCK(p); 748 if (oldsigacts) 749 p->p_sigacts = newsigacts; 750 /* Stop profiling */ 751 stopprofclock(p); 752 753 /* reset caught signals */ 754 execsigs(p); 755 756 /* name this process - nameiexec(p, ndp) */ 757 bzero(p->p_comm, sizeof(p->p_comm)); 758 if (args->fname) 759 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 760 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 761 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 762 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 763 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 764 #ifdef KTR 765 sched_clear_tdname(td); 766 #endif 767 768 /* 769 * mark as execed, wakeup the process that vforked (if any) and tell 770 * it that it now has its own resources back 771 */ 772 p->p_flag |= P_EXEC; 773 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 774 p->p_flag2 &= ~P2_NOTRACE; 775 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 776 p->p_flag2 &= ~P2_STKGAP_DISABLE; 777 if (p->p_flag & P_PPWAIT) { 778 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 779 cv_broadcast(&p->p_pwait); 780 /* STOPs are no longer ignored, arrange for AST */ 781 signotify(td); 782 } 783 784 if ((imgp->sysent->sv_setid_allowed != NULL && 785 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) || 786 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0) 787 execve_nosetid(imgp); 788 789 /* 790 * Implement image setuid/setgid installation. 791 */ 792 if (imgp->credential_setid) { 793 /* 794 * Turn off syscall tracing for set-id programs, except for 795 * root. Record any set-id flags first to make sure that 796 * we do not regain any tracing during a possible block. 797 */ 798 setsugid(p); 799 #ifdef KTRACE 800 kiop = ktrprocexec(p); 801 #endif 802 /* 803 * Close any file descriptors 0..2 that reference procfs, 804 * then make sure file descriptors 0..2 are in use. 805 * 806 * Both fdsetugidsafety() and fdcheckstd() may call functions 807 * taking sleepable locks, so temporarily drop our locks. 808 */ 809 PROC_UNLOCK(p); 810 VOP_UNLOCK(imgp->vp); 811 fdsetugidsafety(td); 812 error = fdcheckstd(td); 813 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 814 if (error != 0) 815 goto exec_fail_dealloc; 816 PROC_LOCK(p); 817 #ifdef MAC 818 if (will_transition) { 819 mac_vnode_execve_transition(oldcred, imgp->newcred, 820 imgp->vp, interpvplabel, imgp); 821 } 822 #endif 823 } else { 824 if (oldcred->cr_uid == oldcred->cr_ruid && 825 oldcred->cr_gid == oldcred->cr_rgid) 826 p->p_flag &= ~P_SUGID; 827 } 828 /* 829 * Set the new credentials. 830 */ 831 if (imgp->newcred != NULL) { 832 proc_set_cred(p, imgp->newcred); 833 crfree(oldcred); 834 oldcred = NULL; 835 } 836 837 /* 838 * Store the vp for use in procfs. This vnode was referenced by namei 839 * or fgetvp_exec. 840 */ 841 oldtextvp = p->p_textvp; 842 p->p_textvp = newtextvp; 843 844 #ifdef KDTRACE_HOOKS 845 /* 846 * Tell the DTrace fasttrap provider about the exec if it 847 * has declared an interest. 848 */ 849 if (dtrace_fasttrap_exec) 850 dtrace_fasttrap_exec(p); 851 #endif 852 853 /* 854 * Notify others that we exec'd, and clear the P_INEXEC flag 855 * as we're now a bona fide freshly-execed process. 856 */ 857 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 858 p->p_flag &= ~P_INEXEC; 859 860 /* clear "fork but no exec" flag, as we _are_ execing */ 861 p->p_acflag &= ~AFORK; 862 863 /* 864 * Free any previous argument cache and replace it with 865 * the new argument cache, if any. 866 */ 867 oldargs = p->p_args; 868 p->p_args = newargs; 869 newargs = NULL; 870 871 PROC_UNLOCK(p); 872 873 #ifdef HWPMC_HOOKS 874 /* 875 * Check if system-wide sampling is in effect or if the 876 * current process is using PMCs. If so, do exec() time 877 * processing. This processing needs to happen AFTER the 878 * P_INEXEC flag is cleared. 879 */ 880 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 881 VOP_UNLOCK(imgp->vp); 882 pe.pm_credentialschanged = credential_changing; 883 pe.pm_entryaddr = imgp->entry_addr; 884 885 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 886 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 887 } 888 #endif 889 890 /* Set values passed into the program in registers. */ 891 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 892 893 VOP_MMAPPED(imgp->vp); 894 895 SDT_PROBE1(proc, , , exec__success, args->fname); 896 897 exec_fail_dealloc: 898 if (error != 0) { 899 p->p_osrel = orig_osrel; 900 p->p_fctl0 = orig_fctl0; 901 p->p_elf_brandinfo = orig_brandinfo; 902 } 903 904 if (imgp->firstpage != NULL) 905 exec_unmap_first_page(imgp); 906 907 if (imgp->vp != NULL) { 908 if (args->fname) 909 NDFREE(&nd, NDF_ONLY_PNBUF); 910 if (imgp->opened) 911 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 912 if (imgp->textset) 913 VOP_UNSET_TEXT_CHECKED(imgp->vp); 914 if (error != 0) 915 vput(imgp->vp); 916 else 917 VOP_UNLOCK(imgp->vp); 918 } 919 920 if (imgp->object != NULL) 921 vm_object_deallocate(imgp->object); 922 923 free(imgp->freepath, M_TEMP); 924 925 if (error == 0) { 926 if (p->p_ptevents & PTRACE_EXEC) { 927 PROC_LOCK(p); 928 if (p->p_ptevents & PTRACE_EXEC) 929 td->td_dbgflags |= TDB_EXEC; 930 PROC_UNLOCK(p); 931 } 932 } else { 933 exec_fail: 934 /* we're done here, clear P_INEXEC */ 935 PROC_LOCK(p); 936 p->p_flag &= ~P_INEXEC; 937 PROC_UNLOCK(p); 938 939 SDT_PROBE1(proc, , , exec__failure, error); 940 } 941 942 if (imgp->newcred != NULL && oldcred != NULL) 943 crfree(imgp->newcred); 944 945 #ifdef MAC 946 mac_execve_exit(imgp); 947 mac_execve_interpreter_exit(interpvplabel); 948 #endif 949 exec_free_args(args); 950 951 /* 952 * Handle deferred decrement of ref counts. 953 */ 954 if (oldtextvp != NULL) 955 vrele(oldtextvp); 956 #ifdef KTRACE 957 ktr_io_params_free(kiop); 958 #endif 959 pargs_drop(oldargs); 960 pargs_drop(newargs); 961 if (oldsigacts != NULL) 962 sigacts_free(oldsigacts); 963 if (euip != NULL) 964 uifree(euip); 965 966 if (error && imgp->vmspace_destroyed) { 967 /* sorry, no more process anymore. exit gracefully */ 968 exec_cleanup(td, oldvmspace); 969 exit1(td, 0, SIGABRT); 970 /* NOT REACHED */ 971 } 972 973 #ifdef KTRACE 974 if (error == 0) 975 ktrprocctor(p); 976 #endif 977 978 /* 979 * We don't want cpu_set_syscall_retval() to overwrite any of 980 * the register values put in place by exec_setregs(). 981 * Implementations of cpu_set_syscall_retval() will leave 982 * registers unmodified when returning EJUSTRETURN. 983 */ 984 return (error == 0 ? EJUSTRETURN : error); 985 } 986 987 void 988 exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 989 { 990 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 991 KASSERT(td->td_proc->p_vmspace != oldvmspace, 992 ("oldvmspace still used")); 993 vmspace_free(oldvmspace); 994 td->td_pflags &= ~TDP_EXECVMSPC; 995 } 996 } 997 998 int 999 exec_map_first_page(struct image_params *imgp) 1000 { 1001 vm_object_t object; 1002 vm_page_t m; 1003 int error; 1004 1005 if (imgp->firstpage != NULL) 1006 exec_unmap_first_page(imgp); 1007 1008 object = imgp->vp->v_object; 1009 if (object == NULL) 1010 return (EACCES); 1011 #if VM_NRESERVLEVEL > 0 1012 if ((object->flags & OBJ_COLORED) == 0) { 1013 VM_OBJECT_WLOCK(object); 1014 vm_object_color(object, 0); 1015 VM_OBJECT_WUNLOCK(object); 1016 } 1017 #endif 1018 error = vm_page_grab_valid_unlocked(&m, object, 0, 1019 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1020 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1021 1022 if (error != VM_PAGER_OK) 1023 return (EIO); 1024 imgp->firstpage = sf_buf_alloc(m, 0); 1025 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1026 1027 return (0); 1028 } 1029 1030 void 1031 exec_unmap_first_page(struct image_params *imgp) 1032 { 1033 vm_page_t m; 1034 1035 if (imgp->firstpage != NULL) { 1036 m = sf_buf_page(imgp->firstpage); 1037 sf_buf_free(imgp->firstpage); 1038 imgp->firstpage = NULL; 1039 vm_page_unwire(m, PQ_ACTIVE); 1040 } 1041 } 1042 1043 /* 1044 * Destroy old address space, and allocate a new stack. 1045 * The new stack is only sgrowsiz large because it is grown 1046 * automatically on a page fault. 1047 */ 1048 int 1049 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1050 { 1051 int error; 1052 struct proc *p = imgp->proc; 1053 struct vmspace *vmspace = p->p_vmspace; 1054 struct thread *td = curthread; 1055 vm_object_t obj; 1056 struct rlimit rlim_stack; 1057 vm_offset_t sv_minuser, stack_addr; 1058 vm_map_t map; 1059 vm_prot_t stack_prot; 1060 u_long ssiz; 1061 1062 imgp->vmspace_destroyed = 1; 1063 imgp->sysent = sv; 1064 1065 sigfastblock_clear(td); 1066 umtx_exec(p); 1067 itimers_exec(p); 1068 if (sv->sv_onexec != NULL) 1069 sv->sv_onexec(p, imgp); 1070 1071 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1072 1073 /* 1074 * Blow away entire process VM, if address space not shared, 1075 * otherwise, create a new VM space so that other threads are 1076 * not disrupted 1077 */ 1078 map = &vmspace->vm_map; 1079 if (map_at_zero) 1080 sv_minuser = sv->sv_minuser; 1081 else 1082 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1083 if (refcount_load(&vmspace->vm_refcnt) == 1 && 1084 vm_map_min(map) == sv_minuser && 1085 vm_map_max(map) == sv->sv_maxuser && 1086 cpu_exec_vmspace_reuse(p, map)) { 1087 shmexit(vmspace); 1088 pmap_remove_pages(vmspace_pmap(vmspace)); 1089 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1090 /* 1091 * An exec terminates mlockall(MCL_FUTURE). 1092 * ASLR and W^X states must be re-evaluated. 1093 */ 1094 vm_map_lock(map); 1095 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1096 MAP_ASLR_IGNSTART | MAP_WXORX); 1097 vm_map_unlock(map); 1098 } else { 1099 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1100 if (error) 1101 return (error); 1102 vmspace = p->p_vmspace; 1103 map = &vmspace->vm_map; 1104 } 1105 map->flags |= imgp->map_flags; 1106 1107 /* Map a shared page */ 1108 obj = sv->sv_shared_page_obj; 1109 if (obj != NULL) { 1110 vm_object_reference(obj); 1111 error = vm_map_fixed(map, obj, 0, 1112 sv->sv_shared_page_base, sv->sv_shared_page_len, 1113 VM_PROT_READ | VM_PROT_EXECUTE, 1114 VM_PROT_READ | VM_PROT_EXECUTE, 1115 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1116 if (error != KERN_SUCCESS) { 1117 vm_object_deallocate(obj); 1118 return (vm_mmap_to_errno(error)); 1119 } 1120 } 1121 1122 /* Allocate a new stack */ 1123 if (imgp->stack_sz != 0) { 1124 ssiz = trunc_page(imgp->stack_sz); 1125 PROC_LOCK(p); 1126 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1127 PROC_UNLOCK(p); 1128 if (ssiz > rlim_stack.rlim_max) 1129 ssiz = rlim_stack.rlim_max; 1130 if (ssiz > rlim_stack.rlim_cur) { 1131 rlim_stack.rlim_cur = ssiz; 1132 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1133 } 1134 } else if (sv->sv_maxssiz != NULL) { 1135 ssiz = *sv->sv_maxssiz; 1136 } else { 1137 ssiz = maxssiz; 1138 } 1139 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1140 if (ssiz < imgp->eff_stack_sz) 1141 imgp->eff_stack_sz = ssiz; 1142 stack_addr = sv->sv_usrstack - ssiz; 1143 stack_prot = obj != NULL && imgp->stack_prot != 0 ? 1144 imgp->stack_prot : sv->sv_stackprot; 1145 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot, 1146 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1147 if (error != KERN_SUCCESS) { 1148 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x " 1149 "failed mach error %d errno %d\n", (uintmax_t)ssiz, 1150 stack_prot, error, vm_mmap_to_errno(error)); 1151 return (vm_mmap_to_errno(error)); 1152 } 1153 1154 /* 1155 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1156 * are still used to enforce the stack rlimit on the process stack. 1157 */ 1158 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1159 vmspace->vm_maxsaddr = (char *)stack_addr; 1160 1161 return (0); 1162 } 1163 1164 /* 1165 * Copy out argument and environment strings from the old process address 1166 * space into the temporary string buffer. 1167 */ 1168 int 1169 exec_copyin_args(struct image_args *args, const char *fname, 1170 enum uio_seg segflg, char **argv, char **envv) 1171 { 1172 u_long arg, env; 1173 int error; 1174 1175 bzero(args, sizeof(*args)); 1176 if (argv == NULL) 1177 return (EFAULT); 1178 1179 /* 1180 * Allocate demand-paged memory for the file name, argument, and 1181 * environment strings. 1182 */ 1183 error = exec_alloc_args(args); 1184 if (error != 0) 1185 return (error); 1186 1187 /* 1188 * Copy the file name. 1189 */ 1190 error = exec_args_add_fname(args, fname, segflg); 1191 if (error != 0) 1192 goto err_exit; 1193 1194 /* 1195 * extract arguments first 1196 */ 1197 for (;;) { 1198 error = fueword(argv++, &arg); 1199 if (error == -1) { 1200 error = EFAULT; 1201 goto err_exit; 1202 } 1203 if (arg == 0) 1204 break; 1205 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1206 UIO_USERSPACE); 1207 if (error != 0) 1208 goto err_exit; 1209 } 1210 1211 /* 1212 * extract environment strings 1213 */ 1214 if (envv) { 1215 for (;;) { 1216 error = fueword(envv++, &env); 1217 if (error == -1) { 1218 error = EFAULT; 1219 goto err_exit; 1220 } 1221 if (env == 0) 1222 break; 1223 error = exec_args_add_env(args, 1224 (char *)(uintptr_t)env, UIO_USERSPACE); 1225 if (error != 0) 1226 goto err_exit; 1227 } 1228 } 1229 1230 return (0); 1231 1232 err_exit: 1233 exec_free_args(args); 1234 return (error); 1235 } 1236 1237 int 1238 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1239 const void *data, size_t datalen, const int *fds, size_t fdslen) 1240 { 1241 struct filedesc *ofdp; 1242 const char *p; 1243 int *kfds; 1244 int error; 1245 1246 memset(args, '\0', sizeof(*args)); 1247 ofdp = td->td_proc->p_fd; 1248 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles) 1249 return (E2BIG); 1250 error = exec_alloc_args(args); 1251 if (error != 0) 1252 return (error); 1253 1254 args->begin_argv = args->buf; 1255 args->stringspace = ARG_MAX; 1256 1257 if (datalen > 0) { 1258 /* 1259 * Argument buffer has been provided. Copy it into the 1260 * kernel as a single string and add a terminating null 1261 * byte. 1262 */ 1263 error = copyin(data, args->begin_argv, datalen); 1264 if (error != 0) 1265 goto err_exit; 1266 args->begin_argv[datalen] = '\0'; 1267 args->endp = args->begin_argv + datalen + 1; 1268 args->stringspace -= datalen + 1; 1269 1270 /* 1271 * Traditional argument counting. Count the number of 1272 * null bytes. 1273 */ 1274 for (p = args->begin_argv; p < args->endp; ++p) 1275 if (*p == '\0') 1276 ++args->argc; 1277 } else { 1278 /* No argument buffer provided. */ 1279 args->endp = args->begin_argv; 1280 } 1281 1282 /* Create new file descriptor table. */ 1283 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1284 error = copyin(fds, kfds, fdslen * sizeof(int)); 1285 if (error != 0) { 1286 free(kfds, M_TEMP); 1287 goto err_exit; 1288 } 1289 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1290 free(kfds, M_TEMP); 1291 if (error != 0) 1292 goto err_exit; 1293 1294 return (0); 1295 err_exit: 1296 exec_free_args(args); 1297 return (error); 1298 } 1299 1300 struct exec_args_kva { 1301 vm_offset_t addr; 1302 u_int gen; 1303 SLIST_ENTRY(exec_args_kva) next; 1304 }; 1305 1306 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1307 1308 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1309 static struct mtx exec_args_kva_mtx; 1310 static u_int exec_args_gen; 1311 1312 static void 1313 exec_prealloc_args_kva(void *arg __unused) 1314 { 1315 struct exec_args_kva *argkva; 1316 u_int i; 1317 1318 SLIST_INIT(&exec_args_kva_freelist); 1319 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1320 for (i = 0; i < exec_map_entries; i++) { 1321 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1322 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1323 argkva->gen = exec_args_gen; 1324 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1325 } 1326 } 1327 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1328 1329 static vm_offset_t 1330 exec_alloc_args_kva(void **cookie) 1331 { 1332 struct exec_args_kva *argkva; 1333 1334 argkva = (void *)atomic_readandclear_ptr( 1335 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1336 if (argkva == NULL) { 1337 mtx_lock(&exec_args_kva_mtx); 1338 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1339 (void)mtx_sleep(&exec_args_kva_freelist, 1340 &exec_args_kva_mtx, 0, "execkva", 0); 1341 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1342 mtx_unlock(&exec_args_kva_mtx); 1343 } 1344 kasan_mark((void *)argkva->addr, exec_map_entry_size, 1345 exec_map_entry_size, 0); 1346 *(struct exec_args_kva **)cookie = argkva; 1347 return (argkva->addr); 1348 } 1349 1350 static void 1351 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1352 { 1353 vm_offset_t base; 1354 1355 base = argkva->addr; 1356 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size, 1357 KASAN_EXEC_ARGS_FREED); 1358 if (argkva->gen != gen) { 1359 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1360 MADV_FREE); 1361 argkva->gen = gen; 1362 } 1363 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1364 (uintptr_t)NULL, (uintptr_t)argkva)) { 1365 mtx_lock(&exec_args_kva_mtx); 1366 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1367 wakeup_one(&exec_args_kva_freelist); 1368 mtx_unlock(&exec_args_kva_mtx); 1369 } 1370 } 1371 1372 static void 1373 exec_free_args_kva(void *cookie) 1374 { 1375 1376 exec_release_args_kva(cookie, exec_args_gen); 1377 } 1378 1379 static void 1380 exec_args_kva_lowmem(void *arg __unused) 1381 { 1382 SLIST_HEAD(, exec_args_kva) head; 1383 struct exec_args_kva *argkva; 1384 u_int gen; 1385 int i; 1386 1387 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1388 1389 /* 1390 * Force an madvise of each KVA range. Any currently allocated ranges 1391 * will have MADV_FREE applied once they are freed. 1392 */ 1393 SLIST_INIT(&head); 1394 mtx_lock(&exec_args_kva_mtx); 1395 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1396 mtx_unlock(&exec_args_kva_mtx); 1397 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1398 SLIST_REMOVE_HEAD(&head, next); 1399 exec_release_args_kva(argkva, gen); 1400 } 1401 1402 CPU_FOREACH(i) { 1403 argkva = (void *)atomic_readandclear_ptr( 1404 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1405 if (argkva != NULL) 1406 exec_release_args_kva(argkva, gen); 1407 } 1408 } 1409 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1410 EVENTHANDLER_PRI_ANY); 1411 1412 /* 1413 * Allocate temporary demand-paged, zero-filled memory for the file name, 1414 * argument, and environment strings. 1415 */ 1416 int 1417 exec_alloc_args(struct image_args *args) 1418 { 1419 1420 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1421 return (0); 1422 } 1423 1424 void 1425 exec_free_args(struct image_args *args) 1426 { 1427 1428 if (args->buf != NULL) { 1429 exec_free_args_kva(args->bufkva); 1430 args->buf = NULL; 1431 } 1432 if (args->fname_buf != NULL) { 1433 free(args->fname_buf, M_TEMP); 1434 args->fname_buf = NULL; 1435 } 1436 if (args->fdp != NULL) 1437 fdescfree_remapped(args->fdp); 1438 } 1439 1440 /* 1441 * A set to functions to fill struct image args. 1442 * 1443 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1444 * fname) before the other functions. All exec_args_add_arg() calls must 1445 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1446 * may be called any time after exec_args_add_fname(). 1447 * 1448 * exec_args_add_fname() - install path to be executed 1449 * exec_args_add_arg() - append an argument string 1450 * exec_args_add_env() - append an env string 1451 * exec_args_adjust_args() - adjust location of the argument list to 1452 * allow new arguments to be prepended 1453 */ 1454 int 1455 exec_args_add_fname(struct image_args *args, const char *fname, 1456 enum uio_seg segflg) 1457 { 1458 int error; 1459 size_t length; 1460 1461 KASSERT(args->fname == NULL, ("fname already appended")); 1462 KASSERT(args->endp == NULL, ("already appending to args")); 1463 1464 if (fname != NULL) { 1465 args->fname = args->buf; 1466 error = segflg == UIO_SYSSPACE ? 1467 copystr(fname, args->fname, PATH_MAX, &length) : 1468 copyinstr(fname, args->fname, PATH_MAX, &length); 1469 if (error != 0) 1470 return (error == ENAMETOOLONG ? E2BIG : error); 1471 } else 1472 length = 0; 1473 1474 /* Set up for _arg_*()/_env_*() */ 1475 args->endp = args->buf + length; 1476 /* begin_argv must be set and kept updated */ 1477 args->begin_argv = args->endp; 1478 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1479 ("too little space remaining for arguments %zu < %zu", 1480 exec_map_entry_size - length, (size_t)ARG_MAX)); 1481 args->stringspace = ARG_MAX; 1482 1483 return (0); 1484 } 1485 1486 static int 1487 exec_args_add_str(struct image_args *args, const char *str, 1488 enum uio_seg segflg, int *countp) 1489 { 1490 int error; 1491 size_t length; 1492 1493 KASSERT(args->endp != NULL, ("endp not initialized")); 1494 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1495 1496 error = (segflg == UIO_SYSSPACE) ? 1497 copystr(str, args->endp, args->stringspace, &length) : 1498 copyinstr(str, args->endp, args->stringspace, &length); 1499 if (error != 0) 1500 return (error == ENAMETOOLONG ? E2BIG : error); 1501 args->stringspace -= length; 1502 args->endp += length; 1503 (*countp)++; 1504 1505 return (0); 1506 } 1507 1508 int 1509 exec_args_add_arg(struct image_args *args, const char *argp, 1510 enum uio_seg segflg) 1511 { 1512 1513 KASSERT(args->envc == 0, ("appending args after env")); 1514 1515 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1516 } 1517 1518 int 1519 exec_args_add_env(struct image_args *args, const char *envp, 1520 enum uio_seg segflg) 1521 { 1522 1523 if (args->envc == 0) 1524 args->begin_envv = args->endp; 1525 1526 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1527 } 1528 1529 int 1530 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1531 { 1532 ssize_t offset; 1533 1534 KASSERT(args->endp != NULL, ("endp not initialized")); 1535 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1536 1537 offset = extend - consume; 1538 if (args->stringspace < offset) 1539 return (E2BIG); 1540 memmove(args->begin_argv + extend, args->begin_argv + consume, 1541 args->endp - args->begin_argv + consume); 1542 if (args->envc > 0) 1543 args->begin_envv += offset; 1544 args->endp += offset; 1545 args->stringspace -= offset; 1546 return (0); 1547 } 1548 1549 char * 1550 exec_args_get_begin_envv(struct image_args *args) 1551 { 1552 1553 KASSERT(args->endp != NULL, ("endp not initialized")); 1554 1555 if (args->envc > 0) 1556 return (args->begin_envv); 1557 return (args->endp); 1558 } 1559 1560 void 1561 exec_stackgap(struct image_params *imgp, uintptr_t *dp) 1562 { 1563 if (imgp->sysent->sv_stackgap == NULL || 1564 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE | 1565 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 || 1566 (imgp->map_flags & MAP_ASLR) == 0) 1567 return; 1568 imgp->sysent->sv_stackgap(imgp, dp); 1569 } 1570 1571 /* 1572 * Copy strings out to the new process address space, constructing new arg 1573 * and env vector tables. Return a pointer to the base so that it can be used 1574 * as the initial stack pointer. 1575 */ 1576 int 1577 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1578 { 1579 int argc, envc; 1580 char **vectp; 1581 char *stringp; 1582 uintptr_t destp, ustringp; 1583 struct ps_strings *arginfo; 1584 struct proc *p; 1585 size_t execpath_len; 1586 int error, szsigcode, szps; 1587 char canary[sizeof(long) * 8]; 1588 1589 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1590 /* 1591 * Calculate string base and vector table pointers. 1592 * Also deal with signal trampoline code for this exec type. 1593 */ 1594 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1595 execpath_len = strlen(imgp->execpath) + 1; 1596 else 1597 execpath_len = 0; 1598 p = imgp->proc; 1599 szsigcode = 0; 1600 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1601 imgp->ps_strings = arginfo; 1602 if (p->p_sysent->sv_sigcode_base == 0) { 1603 if (p->p_sysent->sv_szsigcode != NULL) 1604 szsigcode = *(p->p_sysent->sv_szsigcode); 1605 } 1606 destp = (uintptr_t)arginfo; 1607 1608 /* 1609 * install sigcode 1610 */ 1611 if (szsigcode != 0) { 1612 destp -= szsigcode; 1613 destp = rounddown2(destp, sizeof(void *)); 1614 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1615 szsigcode); 1616 if (error != 0) 1617 return (error); 1618 } 1619 1620 /* 1621 * Copy the image path for the rtld. 1622 */ 1623 if (execpath_len != 0) { 1624 destp -= execpath_len; 1625 destp = rounddown2(destp, sizeof(void *)); 1626 imgp->execpathp = (void *)destp; 1627 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1628 if (error != 0) 1629 return (error); 1630 } 1631 1632 /* 1633 * Prepare the canary for SSP. 1634 */ 1635 arc4rand(canary, sizeof(canary), 0); 1636 destp -= sizeof(canary); 1637 imgp->canary = (void *)destp; 1638 error = copyout(canary, imgp->canary, sizeof(canary)); 1639 if (error != 0) 1640 return (error); 1641 imgp->canarylen = sizeof(canary); 1642 1643 /* 1644 * Prepare the pagesizes array. 1645 */ 1646 destp -= szps; 1647 destp = rounddown2(destp, sizeof(void *)); 1648 imgp->pagesizes = (void *)destp; 1649 error = copyout(pagesizes, imgp->pagesizes, szps); 1650 if (error != 0) 1651 return (error); 1652 imgp->pagesizeslen = szps; 1653 1654 /* 1655 * Allocate room for the argument and environment strings. 1656 */ 1657 destp -= ARG_MAX - imgp->args->stringspace; 1658 destp = rounddown2(destp, sizeof(void *)); 1659 ustringp = destp; 1660 1661 exec_stackgap(imgp, &destp); 1662 1663 if (imgp->auxargs) { 1664 /* 1665 * Allocate room on the stack for the ELF auxargs 1666 * array. It has up to AT_COUNT entries. 1667 */ 1668 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1669 destp = rounddown2(destp, sizeof(void *)); 1670 } 1671 1672 vectp = (char **)destp; 1673 1674 /* 1675 * Allocate room for the argv[] and env vectors including the 1676 * terminating NULL pointers. 1677 */ 1678 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1679 1680 /* 1681 * vectp also becomes our initial stack base 1682 */ 1683 *stack_base = (uintptr_t)vectp; 1684 1685 stringp = imgp->args->begin_argv; 1686 argc = imgp->args->argc; 1687 envc = imgp->args->envc; 1688 1689 /* 1690 * Copy out strings - arguments and environment. 1691 */ 1692 error = copyout(stringp, (void *)ustringp, 1693 ARG_MAX - imgp->args->stringspace); 1694 if (error != 0) 1695 return (error); 1696 1697 /* 1698 * Fill in "ps_strings" struct for ps, w, etc. 1699 */ 1700 imgp->argv = vectp; 1701 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1702 suword32(&arginfo->ps_nargvstr, argc) != 0) 1703 return (EFAULT); 1704 1705 /* 1706 * Fill in argument portion of vector table. 1707 */ 1708 for (; argc > 0; --argc) { 1709 if (suword(vectp++, ustringp) != 0) 1710 return (EFAULT); 1711 while (*stringp++ != 0) 1712 ustringp++; 1713 ustringp++; 1714 } 1715 1716 /* a null vector table pointer separates the argp's from the envp's */ 1717 if (suword(vectp++, 0) != 0) 1718 return (EFAULT); 1719 1720 imgp->envv = vectp; 1721 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1722 suword32(&arginfo->ps_nenvstr, envc) != 0) 1723 return (EFAULT); 1724 1725 /* 1726 * Fill in environment portion of vector table. 1727 */ 1728 for (; envc > 0; --envc) { 1729 if (suword(vectp++, ustringp) != 0) 1730 return (EFAULT); 1731 while (*stringp++ != 0) 1732 ustringp++; 1733 ustringp++; 1734 } 1735 1736 /* end of vector table is a null pointer */ 1737 if (suword(vectp, 0) != 0) 1738 return (EFAULT); 1739 1740 if (imgp->auxargs) { 1741 vectp++; 1742 error = imgp->sysent->sv_copyout_auxargs(imgp, 1743 (uintptr_t)vectp); 1744 if (error != 0) 1745 return (error); 1746 } 1747 1748 return (0); 1749 } 1750 1751 /* 1752 * Check permissions of file to execute. 1753 * Called with imgp->vp locked. 1754 * Return 0 for success or error code on failure. 1755 */ 1756 int 1757 exec_check_permissions(struct image_params *imgp) 1758 { 1759 struct vnode *vp = imgp->vp; 1760 struct vattr *attr = imgp->attr; 1761 struct thread *td; 1762 int error; 1763 1764 td = curthread; 1765 1766 /* Get file attributes */ 1767 error = VOP_GETATTR(vp, attr, td->td_ucred); 1768 if (error) 1769 return (error); 1770 1771 #ifdef MAC 1772 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1773 if (error) 1774 return (error); 1775 #endif 1776 1777 /* 1778 * 1) Check if file execution is disabled for the filesystem that 1779 * this file resides on. 1780 * 2) Ensure that at least one execute bit is on. Otherwise, a 1781 * privileged user will always succeed, and we don't want this 1782 * to happen unless the file really is executable. 1783 * 3) Ensure that the file is a regular file. 1784 */ 1785 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1786 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1787 (attr->va_type != VREG)) 1788 return (EACCES); 1789 1790 /* 1791 * Zero length files can't be exec'd 1792 */ 1793 if (attr->va_size == 0) 1794 return (ENOEXEC); 1795 1796 /* 1797 * Check for execute permission to file based on current credentials. 1798 */ 1799 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1800 if (error) 1801 return (error); 1802 1803 /* 1804 * Check number of open-for-writes on the file and deny execution 1805 * if there are any. 1806 * 1807 * Add a text reference now so no one can write to the 1808 * executable while we're activating it. 1809 * 1810 * Remember if this was set before and unset it in case this is not 1811 * actually an executable image. 1812 */ 1813 error = VOP_SET_TEXT(vp); 1814 if (error != 0) 1815 return (error); 1816 imgp->textset = true; 1817 1818 /* 1819 * Call filesystem specific open routine (which does nothing in the 1820 * general case). 1821 */ 1822 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1823 if (error == 0) 1824 imgp->opened = 1; 1825 return (error); 1826 } 1827 1828 /* 1829 * Exec handler registration 1830 */ 1831 int 1832 exec_register(const struct execsw *execsw_arg) 1833 { 1834 const struct execsw **es, **xs, **newexecsw; 1835 u_int count = 2; /* New slot and trailing NULL */ 1836 1837 if (execsw) 1838 for (es = execsw; *es; es++) 1839 count++; 1840 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1841 xs = newexecsw; 1842 if (execsw) 1843 for (es = execsw; *es; es++) 1844 *xs++ = *es; 1845 *xs++ = execsw_arg; 1846 *xs = NULL; 1847 if (execsw) 1848 free(execsw, M_TEMP); 1849 execsw = newexecsw; 1850 return (0); 1851 } 1852 1853 int 1854 exec_unregister(const struct execsw *execsw_arg) 1855 { 1856 const struct execsw **es, **xs, **newexecsw; 1857 int count = 1; 1858 1859 if (execsw == NULL) 1860 panic("unregister with no handlers left?\n"); 1861 1862 for (es = execsw; *es; es++) { 1863 if (*es == execsw_arg) 1864 break; 1865 } 1866 if (*es == NULL) 1867 return (ENOENT); 1868 for (es = execsw; *es; es++) 1869 if (*es != execsw_arg) 1870 count++; 1871 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1872 xs = newexecsw; 1873 for (es = execsw; *es; es++) 1874 if (*es != execsw_arg) 1875 *xs++ = *es; 1876 *xs = NULL; 1877 if (execsw) 1878 free(execsw, M_TEMP); 1879 execsw = newexecsw; 1880 return (0); 1881 } 1882 1883 /* 1884 * Write out a core segment to the compression stream. 1885 */ 1886 static int 1887 compress_chunk(struct coredump_params *cp, char *base, char *buf, u_int len) 1888 { 1889 u_int chunk_len; 1890 int error; 1891 1892 while (len > 0) { 1893 chunk_len = MIN(len, CORE_BUF_SIZE); 1894 1895 /* 1896 * We can get EFAULT error here. 1897 * In that case zero out the current chunk of the segment. 1898 */ 1899 error = copyin(base, buf, chunk_len); 1900 if (error != 0) 1901 bzero(buf, chunk_len); 1902 error = compressor_write(cp->comp, buf, chunk_len); 1903 if (error != 0) 1904 break; 1905 base += chunk_len; 1906 len -= chunk_len; 1907 } 1908 return (error); 1909 } 1910 1911 int 1912 core_write(struct coredump_params *cp, const void *base, size_t len, 1913 off_t offset, enum uio_seg seg, size_t *resid) 1914 { 1915 1916 return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base), 1917 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1918 cp->active_cred, cp->file_cred, resid, cp->td)); 1919 } 1920 1921 int 1922 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp, 1923 void *tmpbuf) 1924 { 1925 vm_map_t map; 1926 struct mount *mp; 1927 size_t resid, runlen; 1928 int error; 1929 bool success; 1930 1931 KASSERT((uintptr_t)base % PAGE_SIZE == 0, 1932 ("%s: user address %p is not page-aligned", __func__, base)); 1933 1934 if (cp->comp != NULL) 1935 return (compress_chunk(cp, base, tmpbuf, len)); 1936 1937 map = &cp->td->td_proc->p_vmspace->vm_map; 1938 for (; len > 0; base += runlen, offset += runlen, len -= runlen) { 1939 /* 1940 * Attempt to page in all virtual pages in the range. If a 1941 * virtual page is not backed by the pager, it is represented as 1942 * a hole in the file. This can occur with zero-filled 1943 * anonymous memory or truncated files, for example. 1944 */ 1945 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) { 1946 error = vm_fault(map, (uintptr_t)base + runlen, 1947 VM_PROT_READ, VM_FAULT_NOFILL, NULL); 1948 if (runlen == 0) 1949 success = error == KERN_SUCCESS; 1950 else if ((error == KERN_SUCCESS) != success) 1951 break; 1952 } 1953 1954 if (success) { 1955 error = core_write(cp, base, runlen, offset, 1956 UIO_USERSPACE, &resid); 1957 if (error != 0) { 1958 if (error != EFAULT) 1959 break; 1960 1961 /* 1962 * EFAULT may be returned if the user mapping 1963 * could not be accessed, e.g., because a mapped 1964 * file has been truncated. Skip the page if no 1965 * progress was made, to protect against a 1966 * hypothetical scenario where vm_fault() was 1967 * successful but core_write() returns EFAULT 1968 * anyway. 1969 */ 1970 runlen -= resid; 1971 if (runlen == 0) { 1972 success = false; 1973 runlen = PAGE_SIZE; 1974 } 1975 } 1976 } 1977 if (!success) { 1978 error = vn_start_write(cp->vp, &mp, V_WAIT); 1979 if (error != 0) 1980 break; 1981 vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY); 1982 error = vn_truncate_locked(cp->vp, offset + runlen, 1983 false, cp->td->td_ucred); 1984 VOP_UNLOCK(cp->vp); 1985 vn_finished_write(mp); 1986 if (error != 0) 1987 break; 1988 } 1989 } 1990 return (error); 1991 } 1992 1993 /* 1994 * Drain into a core file. 1995 */ 1996 int 1997 sbuf_drain_core_output(void *arg, const char *data, int len) 1998 { 1999 struct coredump_params *cp; 2000 struct proc *p; 2001 int error, locked; 2002 2003 cp = arg; 2004 p = cp->td->td_proc; 2005 2006 /* 2007 * Some kern_proc out routines that print to this sbuf may 2008 * call us with the process lock held. Draining with the 2009 * non-sleepable lock held is unsafe. The lock is needed for 2010 * those routines when dumping a live process. In our case we 2011 * can safely release the lock before draining and acquire 2012 * again after. 2013 */ 2014 locked = PROC_LOCKED(p); 2015 if (locked) 2016 PROC_UNLOCK(p); 2017 if (cp->comp != NULL) 2018 error = compressor_write(cp->comp, __DECONST(char *, data), len); 2019 else 2020 error = core_write(cp, __DECONST(void *, data), len, cp->offset, 2021 UIO_SYSSPACE, NULL); 2022 if (locked) 2023 PROC_LOCK(p); 2024 if (error != 0) 2025 return (-error); 2026 cp->offset += len; 2027 return (len); 2028 } 2029