1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1993, David Greenman 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_capsicum.h" 33 #include "opt_hwpmc_hooks.h" 34 #include "opt_ktrace.h" 35 #include "opt_vm.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/acct.h> 40 #include <sys/capsicum.h> 41 #include <sys/eventhandler.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/filedesc.h> 45 #include <sys/imgact.h> 46 #include <sys/imgact_elf.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mman.h> 51 #include <sys/mount.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/pioctl.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/ptrace.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sched.h> 61 #include <sys/sdt.h> 62 #include <sys/sf_buf.h> 63 #include <sys/shm.h> 64 #include <sys/signalvar.h> 65 #include <sys/smp.h> 66 #include <sys/stat.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/sysproto.h> 71 #include <sys/vnode.h> 72 #include <sys/wait.h> 73 #ifdef KTRACE 74 #include <sys/ktrace.h> 75 #endif 76 77 #include <vm/vm.h> 78 #include <vm/vm_param.h> 79 #include <vm/pmap.h> 80 #include <vm/vm_page.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_kern.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_pager.h> 86 87 #ifdef HWPMC_HOOKS 88 #include <sys/pmckern.h> 89 #endif 90 91 #include <machine/reg.h> 92 93 #include <security/audit/audit.h> 94 #include <security/mac/mac_framework.h> 95 96 #ifdef KDTRACE_HOOKS 97 #include <sys/dtrace_bsd.h> 98 dtrace_execexit_func_t dtrace_fasttrap_exec; 99 #endif 100 101 SDT_PROVIDER_DECLARE(proc); 102 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 105 106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 107 108 int coredump_pack_fileinfo = 1; 109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 110 &coredump_pack_fileinfo, 0, 111 "Enable file path packing in 'procstat -f' coredump notes"); 112 113 int coredump_pack_vmmapinfo = 1; 114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 115 &coredump_pack_vmmapinfo, 0, 116 "Enable file path packing in 'procstat -v' coredump notes"); 117 118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 121 static int do_execve(struct thread *td, struct image_args *args, 122 struct mac *mac_p); 123 124 /* XXX This should be vm_size_t. */ 125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 127 128 /* XXX This should be vm_size_t. */ 129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 131 132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 133 NULL, 0, sysctl_kern_stackprot, "I", ""); 134 135 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 137 &ps_arg_cache_limit, 0, ""); 138 139 static int disallow_high_osrel; 140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 141 &disallow_high_osrel, 0, 142 "Disallow execution of binaries built for higher version of the world"); 143 144 static int map_at_zero = 0; 145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 146 "Permit processes to map an object at virtual address 0."); 147 148 EVENTHANDLER_LIST_DECLARE(process_exec); 149 150 static int 151 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 152 { 153 struct proc *p; 154 int error; 155 156 p = curproc; 157 #ifdef SCTL_MASK32 158 if (req->flags & SCTL_MASK32) { 159 unsigned int val; 160 val = (unsigned int)p->p_sysent->sv_psstrings; 161 error = SYSCTL_OUT(req, &val, sizeof(val)); 162 } else 163 #endif 164 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 165 sizeof(p->p_sysent->sv_psstrings)); 166 return error; 167 } 168 169 static int 170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 171 { 172 struct proc *p; 173 int error; 174 175 p = curproc; 176 #ifdef SCTL_MASK32 177 if (req->flags & SCTL_MASK32) { 178 unsigned int val; 179 val = (unsigned int)p->p_sysent->sv_usrstack; 180 error = SYSCTL_OUT(req, &val, sizeof(val)); 181 } else 182 #endif 183 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 184 sizeof(p->p_sysent->sv_usrstack)); 185 return error; 186 } 187 188 static int 189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 190 { 191 struct proc *p; 192 193 p = curproc; 194 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 195 sizeof(p->p_sysent->sv_stackprot))); 196 } 197 198 /* 199 * Each of the items is a pointer to a `const struct execsw', hence the 200 * double pointer here. 201 */ 202 static const struct execsw **execsw; 203 204 #ifndef _SYS_SYSPROTO_H_ 205 struct execve_args { 206 char *fname; 207 char **argv; 208 char **envv; 209 }; 210 #endif 211 212 int 213 sys_execve(struct thread *td, struct execve_args *uap) 214 { 215 struct image_args args; 216 struct vmspace *oldvmspace; 217 int error; 218 219 error = pre_execve(td, &oldvmspace); 220 if (error != 0) 221 return (error); 222 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 223 uap->argv, uap->envv); 224 if (error == 0) 225 error = kern_execve(td, &args, NULL); 226 post_execve(td, error, oldvmspace); 227 return (error); 228 } 229 230 #ifndef _SYS_SYSPROTO_H_ 231 struct fexecve_args { 232 int fd; 233 char **argv; 234 char **envv; 235 } 236 #endif 237 int 238 sys_fexecve(struct thread *td, struct fexecve_args *uap) 239 { 240 struct image_args args; 241 struct vmspace *oldvmspace; 242 int error; 243 244 error = pre_execve(td, &oldvmspace); 245 if (error != 0) 246 return (error); 247 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 248 uap->argv, uap->envv); 249 if (error == 0) { 250 args.fd = uap->fd; 251 error = kern_execve(td, &args, NULL); 252 } 253 post_execve(td, error, oldvmspace); 254 return (error); 255 } 256 257 #ifndef _SYS_SYSPROTO_H_ 258 struct __mac_execve_args { 259 char *fname; 260 char **argv; 261 char **envv; 262 struct mac *mac_p; 263 }; 264 #endif 265 266 int 267 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 268 { 269 #ifdef MAC 270 struct image_args args; 271 struct vmspace *oldvmspace; 272 int error; 273 274 error = pre_execve(td, &oldvmspace); 275 if (error != 0) 276 return (error); 277 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 278 uap->argv, uap->envv); 279 if (error == 0) 280 error = kern_execve(td, &args, uap->mac_p); 281 post_execve(td, error, oldvmspace); 282 return (error); 283 #else 284 return (ENOSYS); 285 #endif 286 } 287 288 int 289 pre_execve(struct thread *td, struct vmspace **oldvmspace) 290 { 291 struct proc *p; 292 int error; 293 294 KASSERT(td == curthread, ("non-current thread %p", td)); 295 error = 0; 296 p = td->td_proc; 297 if ((p->p_flag & P_HADTHREADS) != 0) { 298 PROC_LOCK(p); 299 if (thread_single(p, SINGLE_BOUNDARY) != 0) 300 error = ERESTART; 301 PROC_UNLOCK(p); 302 } 303 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 304 ("nested execve")); 305 *oldvmspace = p->p_vmspace; 306 return (error); 307 } 308 309 void 310 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 311 { 312 struct proc *p; 313 314 KASSERT(td == curthread, ("non-current thread %p", td)); 315 p = td->td_proc; 316 if ((p->p_flag & P_HADTHREADS) != 0) { 317 PROC_LOCK(p); 318 /* 319 * If success, we upgrade to SINGLE_EXIT state to 320 * force other threads to suicide. 321 */ 322 if (error == EJUSTRETURN) 323 thread_single(p, SINGLE_EXIT); 324 else 325 thread_single_end(p, SINGLE_BOUNDARY); 326 PROC_UNLOCK(p); 327 } 328 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 329 KASSERT(p->p_vmspace != oldvmspace, 330 ("oldvmspace still used")); 331 vmspace_free(oldvmspace); 332 td->td_pflags &= ~TDP_EXECVMSPC; 333 } 334 } 335 336 /* 337 * XXX: kern_execve has the astonishing property of not always returning to 338 * the caller. If sufficiently bad things happen during the call to 339 * do_execve(), it can end up calling exit1(); as a result, callers must 340 * avoid doing anything which they might need to undo (e.g., allocating 341 * memory). 342 */ 343 int 344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 345 { 346 347 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 348 args->begin_envv - args->begin_argv); 349 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 350 args->endp - args->begin_envv); 351 return (do_execve(td, args, mac_p)); 352 } 353 354 /* 355 * In-kernel implementation of execve(). All arguments are assumed to be 356 * userspace pointers from the passed thread. 357 */ 358 static int 359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 360 { 361 struct proc *p = td->td_proc; 362 struct nameidata nd; 363 struct ucred *oldcred; 364 struct uidinfo *euip = NULL; 365 register_t *stack_base; 366 int error, i; 367 struct image_params image_params, *imgp; 368 struct vattr attr; 369 int (*img_first)(struct image_params *); 370 struct pargs *oldargs = NULL, *newargs = NULL; 371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 372 #ifdef KTRACE 373 struct vnode *tracevp = NULL; 374 struct ucred *tracecred = NULL; 375 #endif 376 struct vnode *oldtextvp = NULL, *newtextvp; 377 int credential_changing; 378 int textset; 379 #ifdef MAC 380 struct label *interpvplabel = NULL; 381 int will_transition; 382 #endif 383 #ifdef HWPMC_HOOKS 384 struct pmckern_procexec pe; 385 #endif 386 static const char fexecv_proc_title[] = "(fexecv)"; 387 388 imgp = &image_params; 389 390 /* 391 * Lock the process and set the P_INEXEC flag to indicate that 392 * it should be left alone until we're done here. This is 393 * necessary to avoid race conditions - e.g. in ptrace() - 394 * that might allow a local user to illicitly obtain elevated 395 * privileges. 396 */ 397 PROC_LOCK(p); 398 KASSERT((p->p_flag & P_INEXEC) == 0, 399 ("%s(): process already has P_INEXEC flag", __func__)); 400 p->p_flag |= P_INEXEC; 401 PROC_UNLOCK(p); 402 403 /* 404 * Initialize part of the common data 405 */ 406 bzero(imgp, sizeof(*imgp)); 407 imgp->proc = p; 408 imgp->attr = &attr; 409 imgp->args = args; 410 oldcred = p->p_ucred; 411 412 #ifdef MAC 413 error = mac_execve_enter(imgp, mac_p); 414 if (error) 415 goto exec_fail; 416 #endif 417 418 /* 419 * Translate the file name. namei() returns a vnode pointer 420 * in ni_vp among other things. 421 * 422 * XXXAUDIT: It would be desirable to also audit the name of the 423 * interpreter if this is an interpreted binary. 424 */ 425 if (args->fname != NULL) { 426 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 427 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 428 } 429 430 SDT_PROBE1(proc, , , exec, args->fname); 431 432 interpret: 433 if (args->fname != NULL) { 434 #ifdef CAPABILITY_MODE 435 /* 436 * While capability mode can't reach this point via direct 437 * path arguments to execve(), we also don't allow 438 * interpreters to be used in capability mode (for now). 439 * Catch indirect lookups and return a permissions error. 440 */ 441 if (IN_CAPABILITY_MODE(td)) { 442 error = ECAPMODE; 443 goto exec_fail; 444 } 445 #endif 446 error = namei(&nd); 447 if (error) 448 goto exec_fail; 449 450 newtextvp = nd.ni_vp; 451 imgp->vp = newtextvp; 452 } else { 453 AUDIT_ARG_FD(args->fd); 454 /* 455 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 456 */ 457 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 458 if (error) 459 goto exec_fail; 460 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY); 461 AUDIT_ARG_VNODE1(newtextvp); 462 imgp->vp = newtextvp; 463 } 464 465 /* 466 * Check file permissions (also 'opens' file) 467 */ 468 error = exec_check_permissions(imgp); 469 if (error) 470 goto exec_fail_dealloc; 471 472 imgp->object = imgp->vp->v_object; 473 if (imgp->object != NULL) 474 vm_object_reference(imgp->object); 475 476 /* 477 * Set VV_TEXT now so no one can write to the executable while we're 478 * activating it. 479 * 480 * Remember if this was set before and unset it in case this is not 481 * actually an executable image. 482 */ 483 textset = VOP_IS_TEXT(imgp->vp); 484 VOP_SET_TEXT(imgp->vp); 485 486 error = exec_map_first_page(imgp); 487 if (error) 488 goto exec_fail_dealloc; 489 490 imgp->proc->p_osrel = 0; 491 492 /* 493 * Implement image setuid/setgid. 494 * 495 * Determine new credentials before attempting image activators 496 * so that it can be used by process_exec handlers to determine 497 * credential/setid changes. 498 * 499 * Don't honor setuid/setgid if the filesystem prohibits it or if 500 * the process is being traced. 501 * 502 * We disable setuid/setgid/etc in capability mode on the basis 503 * that most setugid applications are not written with that 504 * environment in mind, and will therefore almost certainly operate 505 * incorrectly. In principle there's no reason that setugid 506 * applications might not be useful in capability mode, so we may want 507 * to reconsider this conservative design choice in the future. 508 * 509 * XXXMAC: For the time being, use NOSUID to also prohibit 510 * transitions on the file system. 511 */ 512 credential_changing = 0; 513 credential_changing |= (attr.va_mode & S_ISUID) && 514 oldcred->cr_uid != attr.va_uid; 515 credential_changing |= (attr.va_mode & S_ISGID) && 516 oldcred->cr_gid != attr.va_gid; 517 #ifdef MAC 518 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 519 interpvplabel, imgp); 520 credential_changing |= will_transition; 521 #endif 522 523 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 524 if (credential_changing) 525 imgp->proc->p_pdeathsig = 0; 526 527 if (credential_changing && 528 #ifdef CAPABILITY_MODE 529 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 530 #endif 531 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 532 (p->p_flag & P_TRACED) == 0) { 533 imgp->credential_setid = true; 534 VOP_UNLOCK(imgp->vp, 0); 535 imgp->newcred = crdup(oldcred); 536 if (attr.va_mode & S_ISUID) { 537 euip = uifind(attr.va_uid); 538 change_euid(imgp->newcred, euip); 539 } 540 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 541 if (attr.va_mode & S_ISGID) 542 change_egid(imgp->newcred, attr.va_gid); 543 /* 544 * Implement correct POSIX saved-id behavior. 545 * 546 * XXXMAC: Note that the current logic will save the 547 * uid and gid if a MAC domain transition occurs, even 548 * though maybe it shouldn't. 549 */ 550 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 551 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 552 } else { 553 /* 554 * Implement correct POSIX saved-id behavior. 555 * 556 * XXX: It's not clear that the existing behavior is 557 * POSIX-compliant. A number of sources indicate that the 558 * saved uid/gid should only be updated if the new ruid is 559 * not equal to the old ruid, or the new euid is not equal 560 * to the old euid and the new euid is not equal to the old 561 * ruid. The FreeBSD code always updates the saved uid/gid. 562 * Also, this code uses the new (replaced) euid and egid as 563 * the source, which may or may not be the right ones to use. 564 */ 565 if (oldcred->cr_svuid != oldcred->cr_uid || 566 oldcred->cr_svgid != oldcred->cr_gid) { 567 VOP_UNLOCK(imgp->vp, 0); 568 imgp->newcred = crdup(oldcred); 569 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 570 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 571 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 572 } 573 } 574 /* The new credentials are installed into the process later. */ 575 576 /* 577 * Do the best to calculate the full path to the image file. 578 */ 579 if (args->fname != NULL && args->fname[0] == '/') 580 imgp->execpath = args->fname; 581 else { 582 VOP_UNLOCK(imgp->vp, 0); 583 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 584 &imgp->freepath) != 0) 585 imgp->execpath = args->fname; 586 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 587 } 588 589 /* 590 * If the current process has a special image activator it 591 * wants to try first, call it. For example, emulating shell 592 * scripts differently. 593 */ 594 error = -1; 595 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 596 error = img_first(imgp); 597 598 /* 599 * Loop through the list of image activators, calling each one. 600 * An activator returns -1 if there is no match, 0 on success, 601 * and an error otherwise. 602 */ 603 for (i = 0; error == -1 && execsw[i]; ++i) { 604 if (execsw[i]->ex_imgact == NULL || 605 execsw[i]->ex_imgact == img_first) { 606 continue; 607 } 608 error = (*execsw[i]->ex_imgact)(imgp); 609 } 610 611 if (error) { 612 if (error == -1) { 613 if (textset == 0) 614 VOP_UNSET_TEXT(imgp->vp); 615 error = ENOEXEC; 616 } 617 goto exec_fail_dealloc; 618 } 619 620 /* 621 * Special interpreter operation, cleanup and loop up to try to 622 * activate the interpreter. 623 */ 624 if (imgp->interpreted) { 625 exec_unmap_first_page(imgp); 626 /* 627 * VV_TEXT needs to be unset for scripts. There is a short 628 * period before we determine that something is a script where 629 * VV_TEXT will be set. The vnode lock is held over this 630 * entire period so nothing should illegitimately be blocked. 631 */ 632 VOP_UNSET_TEXT(imgp->vp); 633 /* free name buffer and old vnode */ 634 if (args->fname != NULL) 635 NDFREE(&nd, NDF_ONLY_PNBUF); 636 #ifdef MAC 637 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 638 #endif 639 if (imgp->opened) { 640 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 641 imgp->opened = 0; 642 } 643 vput(newtextvp); 644 vm_object_deallocate(imgp->object); 645 imgp->object = NULL; 646 imgp->credential_setid = false; 647 if (imgp->newcred != NULL) { 648 crfree(imgp->newcred); 649 imgp->newcred = NULL; 650 } 651 imgp->execpath = NULL; 652 free(imgp->freepath, M_TEMP); 653 imgp->freepath = NULL; 654 /* set new name to that of the interpreter */ 655 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 656 UIO_SYSSPACE, imgp->interpreter_name, td); 657 args->fname = imgp->interpreter_name; 658 goto interpret; 659 } 660 661 /* 662 * NB: We unlock the vnode here because it is believed that none 663 * of the sv_copyout_strings/sv_fixup operations require the vnode. 664 */ 665 VOP_UNLOCK(imgp->vp, 0); 666 667 if (disallow_high_osrel && 668 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 669 error = ENOEXEC; 670 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 671 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 672 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 673 goto exec_fail_dealloc; 674 } 675 676 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 677 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 678 sys_cap_enter(td, NULL); 679 680 /* 681 * Copy out strings (args and env) and initialize stack base 682 */ 683 if (p->p_sysent->sv_copyout_strings) 684 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 685 else 686 stack_base = exec_copyout_strings(imgp); 687 688 /* 689 * If custom stack fixup routine present for this process 690 * let it do the stack setup. 691 * Else stuff argument count as first item on stack 692 */ 693 if (p->p_sysent->sv_fixup != NULL) 694 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 695 else 696 error = suword(--stack_base, imgp->args->argc) == 0 ? 697 0 : EFAULT; 698 if (error != 0) 699 goto exec_fail_dealloc; 700 701 if (args->fdp != NULL) { 702 /* Install a brand new file descriptor table. */ 703 fdinstall_remapped(td, args->fdp); 704 args->fdp = NULL; 705 } else { 706 /* 707 * Keep on using the existing file descriptor table. For 708 * security and other reasons, the file descriptor table 709 * cannot be shared after an exec. 710 */ 711 fdunshare(td); 712 /* close files on exec */ 713 fdcloseexec(td); 714 } 715 716 /* 717 * Malloc things before we need locks. 718 */ 719 i = imgp->args->begin_envv - imgp->args->begin_argv; 720 /* Cache arguments if they fit inside our allowance */ 721 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 722 newargs = pargs_alloc(i); 723 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 724 } 725 726 /* 727 * For security and other reasons, signal handlers cannot 728 * be shared after an exec. The new process gets a copy of the old 729 * handlers. In execsigs(), the new process will have its signals 730 * reset. 731 */ 732 if (sigacts_shared(p->p_sigacts)) { 733 oldsigacts = p->p_sigacts; 734 newsigacts = sigacts_alloc(); 735 sigacts_copy(newsigacts, oldsigacts); 736 } 737 738 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 739 740 PROC_LOCK(p); 741 if (oldsigacts) 742 p->p_sigacts = newsigacts; 743 /* Stop profiling */ 744 stopprofclock(p); 745 746 /* reset caught signals */ 747 execsigs(p); 748 749 /* name this process - nameiexec(p, ndp) */ 750 bzero(p->p_comm, sizeof(p->p_comm)); 751 if (args->fname) 752 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 753 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 754 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 755 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 756 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 757 #ifdef KTR 758 sched_clear_tdname(td); 759 #endif 760 761 /* 762 * mark as execed, wakeup the process that vforked (if any) and tell 763 * it that it now has its own resources back 764 */ 765 p->p_flag |= P_EXEC; 766 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 767 p->p_flag2 &= ~P2_NOTRACE; 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); 889 890 exec_fail_dealloc: 891 if (imgp->firstpage != NULL) 892 exec_unmap_first_page(imgp); 893 894 if (imgp->vp != NULL) { 895 if (args->fname) 896 NDFREE(&nd, NDF_ONLY_PNBUF); 897 if (imgp->opened) 898 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 899 if (error != 0) 900 vput(imgp->vp); 901 else 902 VOP_UNLOCK(imgp->vp, 0); 903 } 904 905 if (imgp->object != NULL) 906 vm_object_deallocate(imgp->object); 907 908 free(imgp->freepath, M_TEMP); 909 910 if (error == 0) { 911 if (p->p_ptevents & PTRACE_EXEC) { 912 PROC_LOCK(p); 913 if (p->p_ptevents & PTRACE_EXEC) 914 td->td_dbgflags |= TDB_EXEC; 915 PROC_UNLOCK(p); 916 } 917 918 /* 919 * Stop the process here if its stop event mask has 920 * the S_EXEC bit set. 921 */ 922 STOPEVENT(p, S_EXEC, 0); 923 } else { 924 exec_fail: 925 /* we're done here, clear P_INEXEC */ 926 PROC_LOCK(p); 927 p->p_flag &= ~P_INEXEC; 928 PROC_UNLOCK(p); 929 930 SDT_PROBE1(proc, , , exec__failure, error); 931 } 932 933 if (imgp->newcred != NULL && oldcred != NULL) 934 crfree(imgp->newcred); 935 936 #ifdef MAC 937 mac_execve_exit(imgp); 938 mac_execve_interpreter_exit(interpvplabel); 939 #endif 940 exec_free_args(args); 941 942 /* 943 * Handle deferred decrement of ref counts. 944 */ 945 if (oldtextvp != NULL) 946 vrele(oldtextvp); 947 #ifdef KTRACE 948 if (tracevp != NULL) 949 vrele(tracevp); 950 if (tracecred != NULL) 951 crfree(tracecred); 952 #endif 953 pargs_drop(oldargs); 954 pargs_drop(newargs); 955 if (oldsigacts != NULL) 956 sigacts_free(oldsigacts); 957 if (euip != NULL) 958 uifree(euip); 959 960 if (error && imgp->vmspace_destroyed) { 961 /* sorry, no more process anymore. exit gracefully */ 962 exit1(td, 0, SIGABRT); 963 /* NOT REACHED */ 964 } 965 966 #ifdef KTRACE 967 if (error == 0) 968 ktrprocctor(p); 969 #endif 970 971 /* 972 * We don't want cpu_set_syscall_retval() to overwrite any of 973 * the register values put in place by exec_setregs(). 974 * Implementations of cpu_set_syscall_retval() will leave 975 * registers unmodified when returning EJUSTRETURN. 976 */ 977 return (error == 0 ? EJUSTRETURN : error); 978 } 979 980 int 981 exec_map_first_page(struct image_params *imgp) 982 { 983 int rv, i, after, initial_pagein; 984 vm_page_t ma[VM_INITIAL_PAGEIN]; 985 vm_object_t object; 986 987 if (imgp->firstpage != NULL) 988 exec_unmap_first_page(imgp); 989 990 object = imgp->vp->v_object; 991 if (object == NULL) 992 return (EACCES); 993 VM_OBJECT_WLOCK(object); 994 #if VM_NRESERVLEVEL > 0 995 vm_object_color(object, 0); 996 #endif 997 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 998 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 999 vm_page_xbusy(ma[0]); 1000 if (!vm_pager_has_page(object, 0, NULL, &after)) { 1001 vm_page_lock(ma[0]); 1002 vm_page_free(ma[0]); 1003 vm_page_unlock(ma[0]); 1004 VM_OBJECT_WUNLOCK(object); 1005 return (EIO); 1006 } 1007 initial_pagein = min(after, VM_INITIAL_PAGEIN); 1008 KASSERT(initial_pagein <= object->size, 1009 ("%s: initial_pagein %d object->size %ju", 1010 __func__, initial_pagein, (uintmax_t )object->size)); 1011 for (i = 1; i < initial_pagein; i++) { 1012 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1013 if (ma[i]->valid) 1014 break; 1015 if (!vm_page_tryxbusy(ma[i])) 1016 break; 1017 } else { 1018 ma[i] = vm_page_alloc(object, i, 1019 VM_ALLOC_NORMAL); 1020 if (ma[i] == NULL) 1021 break; 1022 } 1023 } 1024 initial_pagein = i; 1025 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1026 if (rv != VM_PAGER_OK) { 1027 for (i = 0; i < initial_pagein; i++) { 1028 vm_page_lock(ma[i]); 1029 vm_page_free(ma[i]); 1030 vm_page_unlock(ma[i]); 1031 } 1032 VM_OBJECT_WUNLOCK(object); 1033 return (EIO); 1034 } 1035 vm_page_xunbusy(ma[0]); 1036 for (i = 1; i < initial_pagein; i++) 1037 vm_page_readahead_finish(ma[i]); 1038 } 1039 vm_page_lock(ma[0]); 1040 vm_page_hold(ma[0]); 1041 vm_page_activate(ma[0]); 1042 vm_page_unlock(ma[0]); 1043 VM_OBJECT_WUNLOCK(object); 1044 1045 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1046 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1047 1048 return (0); 1049 } 1050 1051 void 1052 exec_unmap_first_page(struct image_params *imgp) 1053 { 1054 vm_page_t m; 1055 1056 if (imgp->firstpage != NULL) { 1057 m = sf_buf_page(imgp->firstpage); 1058 sf_buf_free(imgp->firstpage); 1059 imgp->firstpage = NULL; 1060 vm_page_lock(m); 1061 vm_page_unhold(m); 1062 vm_page_unlock(m); 1063 } 1064 } 1065 1066 /* 1067 * Destroy old address space, and allocate a new stack. 1068 * The new stack is only sgrowsiz large because it is grown 1069 * automatically on a page fault. 1070 */ 1071 int 1072 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1073 { 1074 int error; 1075 struct proc *p = imgp->proc; 1076 struct vmspace *vmspace = p->p_vmspace; 1077 vm_object_t obj; 1078 struct rlimit rlim_stack; 1079 vm_offset_t sv_minuser, stack_addr; 1080 vm_map_t map; 1081 u_long ssiz; 1082 1083 imgp->vmspace_destroyed = 1; 1084 imgp->sysent = sv; 1085 1086 /* May be called with Giant held */ 1087 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1088 1089 /* 1090 * Blow away entire process VM, if address space not shared, 1091 * otherwise, create a new VM space so that other threads are 1092 * not disrupted 1093 */ 1094 map = &vmspace->vm_map; 1095 if (map_at_zero) 1096 sv_minuser = sv->sv_minuser; 1097 else 1098 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1099 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1100 vm_map_max(map) == sv->sv_maxuser) { 1101 shmexit(vmspace); 1102 pmap_remove_pages(vmspace_pmap(vmspace)); 1103 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1104 /* An exec terminates mlockall(MCL_FUTURE). */ 1105 vm_map_lock(map); 1106 vm_map_modflags(map, 0, MAP_WIREFUTURE); 1107 vm_map_unlock(map); 1108 } else { 1109 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1110 if (error) 1111 return (error); 1112 vmspace = p->p_vmspace; 1113 map = &vmspace->vm_map; 1114 } 1115 1116 /* Map a shared page */ 1117 obj = sv->sv_shared_page_obj; 1118 if (obj != NULL) { 1119 vm_object_reference(obj); 1120 error = vm_map_fixed(map, obj, 0, 1121 sv->sv_shared_page_base, sv->sv_shared_page_len, 1122 VM_PROT_READ | VM_PROT_EXECUTE, 1123 VM_PROT_READ | VM_PROT_EXECUTE, 1124 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1125 if (error != KERN_SUCCESS) { 1126 vm_object_deallocate(obj); 1127 return (vm_mmap_to_errno(error)); 1128 } 1129 } 1130 1131 /* Allocate a new stack */ 1132 if (imgp->stack_sz != 0) { 1133 ssiz = trunc_page(imgp->stack_sz); 1134 PROC_LOCK(p); 1135 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1136 PROC_UNLOCK(p); 1137 if (ssiz > rlim_stack.rlim_max) 1138 ssiz = rlim_stack.rlim_max; 1139 if (ssiz > rlim_stack.rlim_cur) { 1140 rlim_stack.rlim_cur = ssiz; 1141 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1142 } 1143 } else if (sv->sv_maxssiz != NULL) { 1144 ssiz = *sv->sv_maxssiz; 1145 } else { 1146 ssiz = maxssiz; 1147 } 1148 stack_addr = sv->sv_usrstack - ssiz; 1149 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1150 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1151 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1152 if (error != KERN_SUCCESS) 1153 return (vm_mmap_to_errno(error)); 1154 1155 /* 1156 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1157 * are still used to enforce the stack rlimit on the process stack. 1158 */ 1159 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1160 vmspace->vm_maxsaddr = (char *)stack_addr; 1161 1162 return (0); 1163 } 1164 1165 /* 1166 * Copy out argument and environment strings from the old process address 1167 * space into the temporary string buffer. 1168 */ 1169 int 1170 exec_copyin_args(struct image_args *args, char *fname, 1171 enum uio_seg segflg, char **argv, char **envv) 1172 { 1173 u_long argp, envp; 1174 int error; 1175 size_t length; 1176 1177 bzero(args, sizeof(*args)); 1178 if (argv == NULL) 1179 return (EFAULT); 1180 1181 /* 1182 * Allocate demand-paged memory for the file name, argument, and 1183 * environment strings. 1184 */ 1185 error = exec_alloc_args(args); 1186 if (error != 0) 1187 return (error); 1188 1189 /* 1190 * Copy the file name. 1191 */ 1192 if (fname != NULL) { 1193 args->fname = args->buf; 1194 error = (segflg == UIO_SYSSPACE) ? 1195 copystr(fname, args->fname, PATH_MAX, &length) : 1196 copyinstr(fname, args->fname, PATH_MAX, &length); 1197 if (error != 0) 1198 goto err_exit; 1199 } else 1200 length = 0; 1201 1202 args->begin_argv = args->buf + length; 1203 args->endp = args->begin_argv; 1204 args->stringspace = ARG_MAX; 1205 1206 /* 1207 * extract arguments first 1208 */ 1209 for (;;) { 1210 error = fueword(argv++, &argp); 1211 if (error == -1) { 1212 error = EFAULT; 1213 goto err_exit; 1214 } 1215 if (argp == 0) 1216 break; 1217 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1218 args->stringspace, &length); 1219 if (error != 0) { 1220 if (error == ENAMETOOLONG) 1221 error = E2BIG; 1222 goto err_exit; 1223 } 1224 args->stringspace -= length; 1225 args->endp += length; 1226 args->argc++; 1227 } 1228 1229 args->begin_envv = args->endp; 1230 1231 /* 1232 * extract environment strings 1233 */ 1234 if (envv) { 1235 for (;;) { 1236 error = fueword(envv++, &envp); 1237 if (error == -1) { 1238 error = EFAULT; 1239 goto err_exit; 1240 } 1241 if (envp == 0) 1242 break; 1243 error = copyinstr((void *)(uintptr_t)envp, 1244 args->endp, args->stringspace, &length); 1245 if (error != 0) { 1246 if (error == ENAMETOOLONG) 1247 error = E2BIG; 1248 goto err_exit; 1249 } 1250 args->stringspace -= length; 1251 args->endp += length; 1252 args->envc++; 1253 } 1254 } 1255 1256 return (0); 1257 1258 err_exit: 1259 exec_free_args(args); 1260 return (error); 1261 } 1262 1263 int 1264 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1265 const void *data, size_t datalen, const int *fds, size_t fdslen) 1266 { 1267 struct filedesc *ofdp; 1268 const char *p; 1269 int *kfds; 1270 int error; 1271 1272 memset(args, '\0', sizeof(*args)); 1273 ofdp = td->td_proc->p_fd; 1274 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1275 return (E2BIG); 1276 error = exec_alloc_args(args); 1277 if (error != 0) 1278 return (error); 1279 1280 args->begin_argv = args->buf; 1281 args->stringspace = ARG_MAX; 1282 1283 if (datalen > 0) { 1284 /* 1285 * Argument buffer has been provided. Copy it into the 1286 * kernel as a single string and add a terminating null 1287 * byte. 1288 */ 1289 error = copyin(data, args->begin_argv, datalen); 1290 if (error != 0) 1291 goto err_exit; 1292 args->begin_argv[datalen] = '\0'; 1293 args->endp = args->begin_argv + datalen + 1; 1294 args->stringspace -= datalen + 1; 1295 1296 /* 1297 * Traditional argument counting. Count the number of 1298 * null bytes. 1299 */ 1300 for (p = args->begin_argv; p < args->endp; ++p) 1301 if (*p == '\0') 1302 ++args->argc; 1303 } else { 1304 /* No argument buffer provided. */ 1305 args->endp = args->begin_argv; 1306 } 1307 /* There are no environment variables. */ 1308 args->begin_envv = args->endp; 1309 1310 /* Create new file descriptor table. */ 1311 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1312 error = copyin(fds, kfds, fdslen * sizeof(int)); 1313 if (error != 0) { 1314 free(kfds, M_TEMP); 1315 goto err_exit; 1316 } 1317 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1318 free(kfds, M_TEMP); 1319 if (error != 0) 1320 goto err_exit; 1321 1322 return (0); 1323 err_exit: 1324 exec_free_args(args); 1325 return (error); 1326 } 1327 1328 struct exec_args_kva { 1329 vm_offset_t addr; 1330 u_int gen; 1331 SLIST_ENTRY(exec_args_kva) next; 1332 }; 1333 1334 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1335 1336 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1337 static struct mtx exec_args_kva_mtx; 1338 static u_int exec_args_gen; 1339 1340 static void 1341 exec_prealloc_args_kva(void *arg __unused) 1342 { 1343 struct exec_args_kva *argkva; 1344 u_int i; 1345 1346 SLIST_INIT(&exec_args_kva_freelist); 1347 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1348 for (i = 0; i < exec_map_entries; i++) { 1349 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1350 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1351 argkva->gen = exec_args_gen; 1352 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1353 } 1354 } 1355 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1356 1357 static vm_offset_t 1358 exec_alloc_args_kva(void **cookie) 1359 { 1360 struct exec_args_kva *argkva; 1361 1362 argkva = (void *)atomic_readandclear_ptr( 1363 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1364 if (argkva == NULL) { 1365 mtx_lock(&exec_args_kva_mtx); 1366 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1367 (void)mtx_sleep(&exec_args_kva_freelist, 1368 &exec_args_kva_mtx, 0, "execkva", 0); 1369 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1370 mtx_unlock(&exec_args_kva_mtx); 1371 } 1372 *(struct exec_args_kva **)cookie = argkva; 1373 return (argkva->addr); 1374 } 1375 1376 static void 1377 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1378 { 1379 vm_offset_t base; 1380 1381 base = argkva->addr; 1382 if (argkva->gen != gen) { 1383 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1384 MADV_FREE); 1385 argkva->gen = gen; 1386 } 1387 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1388 (uintptr_t)NULL, (uintptr_t)argkva)) { 1389 mtx_lock(&exec_args_kva_mtx); 1390 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1391 wakeup_one(&exec_args_kva_freelist); 1392 mtx_unlock(&exec_args_kva_mtx); 1393 } 1394 } 1395 1396 static void 1397 exec_free_args_kva(void *cookie) 1398 { 1399 1400 exec_release_args_kva(cookie, exec_args_gen); 1401 } 1402 1403 static void 1404 exec_args_kva_lowmem(void *arg __unused) 1405 { 1406 SLIST_HEAD(, exec_args_kva) head; 1407 struct exec_args_kva *argkva; 1408 u_int gen; 1409 int i; 1410 1411 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1412 1413 /* 1414 * Force an madvise of each KVA range. Any currently allocated ranges 1415 * will have MADV_FREE applied once they are freed. 1416 */ 1417 SLIST_INIT(&head); 1418 mtx_lock(&exec_args_kva_mtx); 1419 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1420 mtx_unlock(&exec_args_kva_mtx); 1421 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1422 SLIST_REMOVE_HEAD(&head, next); 1423 exec_release_args_kva(argkva, gen); 1424 } 1425 1426 CPU_FOREACH(i) { 1427 argkva = (void *)atomic_readandclear_ptr( 1428 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1429 if (argkva != NULL) 1430 exec_release_args_kva(argkva, gen); 1431 } 1432 } 1433 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1434 EVENTHANDLER_PRI_ANY); 1435 1436 /* 1437 * Allocate temporary demand-paged, zero-filled memory for the file name, 1438 * argument, and environment strings. 1439 */ 1440 int 1441 exec_alloc_args(struct image_args *args) 1442 { 1443 1444 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1445 return (0); 1446 } 1447 1448 void 1449 exec_free_args(struct image_args *args) 1450 { 1451 1452 if (args->buf != NULL) { 1453 exec_free_args_kva(args->bufkva); 1454 args->buf = NULL; 1455 } 1456 if (args->fname_buf != NULL) { 1457 free(args->fname_buf, M_TEMP); 1458 args->fname_buf = NULL; 1459 } 1460 if (args->fdp != NULL) 1461 fdescfree_remapped(args->fdp); 1462 } 1463 1464 /* 1465 * Copy strings out to the new process address space, constructing new arg 1466 * and env vector tables. Return a pointer to the base so that it can be used 1467 * as the initial stack pointer. 1468 */ 1469 register_t * 1470 exec_copyout_strings(struct image_params *imgp) 1471 { 1472 int argc, envc; 1473 char **vectp; 1474 char *stringp; 1475 uintptr_t destp; 1476 register_t *stack_base; 1477 struct ps_strings *arginfo; 1478 struct proc *p; 1479 size_t execpath_len; 1480 int szsigcode, szps; 1481 char canary[sizeof(long) * 8]; 1482 1483 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1484 /* 1485 * Calculate string base and vector table pointers. 1486 * Also deal with signal trampoline code for this exec type. 1487 */ 1488 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1489 execpath_len = strlen(imgp->execpath) + 1; 1490 else 1491 execpath_len = 0; 1492 p = imgp->proc; 1493 szsigcode = 0; 1494 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1495 if (p->p_sysent->sv_sigcode_base == 0) { 1496 if (p->p_sysent->sv_szsigcode != NULL) 1497 szsigcode = *(p->p_sysent->sv_szsigcode); 1498 } 1499 destp = (uintptr_t)arginfo; 1500 1501 /* 1502 * install sigcode 1503 */ 1504 if (szsigcode != 0) { 1505 destp -= szsigcode; 1506 destp = rounddown2(destp, sizeof(void *)); 1507 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1508 } 1509 1510 /* 1511 * Copy the image path for the rtld. 1512 */ 1513 if (execpath_len != 0) { 1514 destp -= execpath_len; 1515 destp = rounddown2(destp, sizeof(void *)); 1516 imgp->execpathp = destp; 1517 copyout(imgp->execpath, (void *)destp, execpath_len); 1518 } 1519 1520 /* 1521 * Prepare the canary for SSP. 1522 */ 1523 arc4rand(canary, sizeof(canary), 0); 1524 destp -= sizeof(canary); 1525 imgp->canary = destp; 1526 copyout(canary, (void *)destp, sizeof(canary)); 1527 imgp->canarylen = sizeof(canary); 1528 1529 /* 1530 * Prepare the pagesizes array. 1531 */ 1532 destp -= szps; 1533 destp = rounddown2(destp, sizeof(void *)); 1534 imgp->pagesizes = destp; 1535 copyout(pagesizes, (void *)destp, szps); 1536 imgp->pagesizeslen = szps; 1537 1538 destp -= ARG_MAX - imgp->args->stringspace; 1539 destp = rounddown2(destp, sizeof(void *)); 1540 1541 vectp = (char **)destp; 1542 if (imgp->auxargs) { 1543 /* 1544 * Allocate room on the stack for the ELF auxargs 1545 * array. It has up to AT_COUNT entries. 1546 */ 1547 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 1548 sizeof(*vectp)); 1549 } 1550 1551 /* 1552 * Allocate room for the argv[] and env vectors including the 1553 * terminating NULL pointers. 1554 */ 1555 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1556 1557 /* 1558 * vectp also becomes our initial stack base 1559 */ 1560 stack_base = (register_t *)vectp; 1561 1562 stringp = imgp->args->begin_argv; 1563 argc = imgp->args->argc; 1564 envc = imgp->args->envc; 1565 1566 /* 1567 * Copy out strings - arguments and environment. 1568 */ 1569 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1570 1571 /* 1572 * Fill in "ps_strings" struct for ps, w, etc. 1573 */ 1574 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1575 suword32(&arginfo->ps_nargvstr, argc); 1576 1577 /* 1578 * Fill in argument portion of vector table. 1579 */ 1580 for (; argc > 0; --argc) { 1581 suword(vectp++, (long)(intptr_t)destp); 1582 while (*stringp++ != 0) 1583 destp++; 1584 destp++; 1585 } 1586 1587 /* a null vector table pointer separates the argp's from the envp's */ 1588 suword(vectp++, 0); 1589 1590 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1591 suword32(&arginfo->ps_nenvstr, envc); 1592 1593 /* 1594 * Fill in environment portion of vector table. 1595 */ 1596 for (; envc > 0; --envc) { 1597 suword(vectp++, (long)(intptr_t)destp); 1598 while (*stringp++ != 0) 1599 destp++; 1600 destp++; 1601 } 1602 1603 /* end of vector table is a null pointer */ 1604 suword(vectp, 0); 1605 1606 return (stack_base); 1607 } 1608 1609 /* 1610 * Check permissions of file to execute. 1611 * Called with imgp->vp locked. 1612 * Return 0 for success or error code on failure. 1613 */ 1614 int 1615 exec_check_permissions(struct image_params *imgp) 1616 { 1617 struct vnode *vp = imgp->vp; 1618 struct vattr *attr = imgp->attr; 1619 struct thread *td; 1620 int error, writecount; 1621 1622 td = curthread; 1623 1624 /* Get file attributes */ 1625 error = VOP_GETATTR(vp, attr, td->td_ucred); 1626 if (error) 1627 return (error); 1628 1629 #ifdef MAC 1630 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1631 if (error) 1632 return (error); 1633 #endif 1634 1635 /* 1636 * 1) Check if file execution is disabled for the filesystem that 1637 * this file resides on. 1638 * 2) Ensure that at least one execute bit is on. Otherwise, a 1639 * privileged user will always succeed, and we don't want this 1640 * to happen unless the file really is executable. 1641 * 3) Ensure that the file is a regular file. 1642 */ 1643 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1644 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1645 (attr->va_type != VREG)) 1646 return (EACCES); 1647 1648 /* 1649 * Zero length files can't be exec'd 1650 */ 1651 if (attr->va_size == 0) 1652 return (ENOEXEC); 1653 1654 /* 1655 * Check for execute permission to file based on current credentials. 1656 */ 1657 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1658 if (error) 1659 return (error); 1660 1661 /* 1662 * Check number of open-for-writes on the file and deny execution 1663 * if there are any. 1664 */ 1665 error = VOP_GET_WRITECOUNT(vp, &writecount); 1666 if (error != 0) 1667 return (error); 1668 if (writecount != 0) 1669 return (ETXTBSY); 1670 1671 /* 1672 * Call filesystem specific open routine (which does nothing in the 1673 * general case). 1674 */ 1675 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1676 if (error == 0) 1677 imgp->opened = 1; 1678 return (error); 1679 } 1680 1681 /* 1682 * Exec handler registration 1683 */ 1684 int 1685 exec_register(const struct execsw *execsw_arg) 1686 { 1687 const struct execsw **es, **xs, **newexecsw; 1688 u_int count = 2; /* New slot and trailing NULL */ 1689 1690 if (execsw) 1691 for (es = execsw; *es; es++) 1692 count++; 1693 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1694 xs = newexecsw; 1695 if (execsw) 1696 for (es = execsw; *es; es++) 1697 *xs++ = *es; 1698 *xs++ = execsw_arg; 1699 *xs = NULL; 1700 if (execsw) 1701 free(execsw, M_TEMP); 1702 execsw = newexecsw; 1703 return (0); 1704 } 1705 1706 int 1707 exec_unregister(const struct execsw *execsw_arg) 1708 { 1709 const struct execsw **es, **xs, **newexecsw; 1710 int count = 1; 1711 1712 if (execsw == NULL) 1713 panic("unregister with no handlers left?\n"); 1714 1715 for (es = execsw; *es; es++) { 1716 if (*es == execsw_arg) 1717 break; 1718 } 1719 if (*es == NULL) 1720 return (ENOENT); 1721 for (es = execsw; *es; es++) 1722 if (*es != execsw_arg) 1723 count++; 1724 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1725 xs = newexecsw; 1726 for (es = execsw; *es; es++) 1727 if (*es != execsw_arg) 1728 *xs++ = *es; 1729 *xs = NULL; 1730 if (execsw) 1731 free(execsw, M_TEMP); 1732 execsw = newexecsw; 1733 return (0); 1734 } 1735