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 exec_args_get_begin_envv(args) - args->begin_argv); 349 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 350 args->endp - exec_args_get_begin_envv(args)); 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 imgp->proc->p_fctl0 = 0; 492 493 /* 494 * Implement image setuid/setgid. 495 * 496 * Determine new credentials before attempting image activators 497 * so that it can be used by process_exec handlers to determine 498 * credential/setid changes. 499 * 500 * Don't honor setuid/setgid if the filesystem prohibits it or if 501 * the process is being traced. 502 * 503 * We disable setuid/setgid/etc in capability mode on the basis 504 * that most setugid applications are not written with that 505 * environment in mind, and will therefore almost certainly operate 506 * incorrectly. In principle there's no reason that setugid 507 * applications might not be useful in capability mode, so we may want 508 * to reconsider this conservative design choice in the future. 509 * 510 * XXXMAC: For the time being, use NOSUID to also prohibit 511 * transitions on the file system. 512 */ 513 credential_changing = 0; 514 credential_changing |= (attr.va_mode & S_ISUID) && 515 oldcred->cr_uid != attr.va_uid; 516 credential_changing |= (attr.va_mode & S_ISGID) && 517 oldcred->cr_gid != attr.va_gid; 518 #ifdef MAC 519 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 520 interpvplabel, imgp); 521 credential_changing |= will_transition; 522 #endif 523 524 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 525 if (credential_changing) 526 imgp->proc->p_pdeathsig = 0; 527 528 if (credential_changing && 529 #ifdef CAPABILITY_MODE 530 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 531 #endif 532 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 533 (p->p_flag & P_TRACED) == 0) { 534 imgp->credential_setid = true; 535 VOP_UNLOCK(imgp->vp, 0); 536 imgp->newcred = crdup(oldcred); 537 if (attr.va_mode & S_ISUID) { 538 euip = uifind(attr.va_uid); 539 change_euid(imgp->newcred, euip); 540 } 541 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 542 if (attr.va_mode & S_ISGID) 543 change_egid(imgp->newcred, attr.va_gid); 544 /* 545 * Implement correct POSIX saved-id behavior. 546 * 547 * XXXMAC: Note that the current logic will save the 548 * uid and gid if a MAC domain transition occurs, even 549 * though maybe it shouldn't. 550 */ 551 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 552 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 553 } else { 554 /* 555 * Implement correct POSIX saved-id behavior. 556 * 557 * XXX: It's not clear that the existing behavior is 558 * POSIX-compliant. A number of sources indicate that the 559 * saved uid/gid should only be updated if the new ruid is 560 * not equal to the old ruid, or the new euid is not equal 561 * to the old euid and the new euid is not equal to the old 562 * ruid. The FreeBSD code always updates the saved uid/gid. 563 * Also, this code uses the new (replaced) euid and egid as 564 * the source, which may or may not be the right ones to use. 565 */ 566 if (oldcred->cr_svuid != oldcred->cr_uid || 567 oldcred->cr_svgid != oldcred->cr_gid) { 568 VOP_UNLOCK(imgp->vp, 0); 569 imgp->newcred = crdup(oldcred); 570 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 571 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 572 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 573 } 574 } 575 /* The new credentials are installed into the process later. */ 576 577 /* 578 * Do the best to calculate the full path to the image file. 579 */ 580 if (args->fname != NULL && args->fname[0] == '/') 581 imgp->execpath = args->fname; 582 else { 583 VOP_UNLOCK(imgp->vp, 0); 584 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 585 &imgp->freepath) != 0) 586 imgp->execpath = args->fname; 587 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 588 } 589 590 /* 591 * If the current process has a special image activator it 592 * wants to try first, call it. For example, emulating shell 593 * scripts differently. 594 */ 595 error = -1; 596 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 597 error = img_first(imgp); 598 599 /* 600 * Loop through the list of image activators, calling each one. 601 * An activator returns -1 if there is no match, 0 on success, 602 * and an error otherwise. 603 */ 604 for (i = 0; error == -1 && execsw[i]; ++i) { 605 if (execsw[i]->ex_imgact == NULL || 606 execsw[i]->ex_imgact == img_first) { 607 continue; 608 } 609 error = (*execsw[i]->ex_imgact)(imgp); 610 } 611 612 if (error) { 613 if (error == -1) { 614 if (textset == 0) 615 VOP_UNSET_TEXT(imgp->vp); 616 error = ENOEXEC; 617 } 618 goto exec_fail_dealloc; 619 } 620 621 /* 622 * Special interpreter operation, cleanup and loop up to try to 623 * activate the interpreter. 624 */ 625 if (imgp->interpreted) { 626 exec_unmap_first_page(imgp); 627 /* 628 * VV_TEXT needs to be unset for scripts. There is a short 629 * period before we determine that something is a script where 630 * VV_TEXT will be set. The vnode lock is held over this 631 * entire period so nothing should illegitimately be blocked. 632 */ 633 VOP_UNSET_TEXT(imgp->vp); 634 /* free name buffer and old vnode */ 635 if (args->fname != NULL) 636 NDFREE(&nd, NDF_ONLY_PNBUF); 637 #ifdef MAC 638 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 639 #endif 640 if (imgp->opened) { 641 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 642 imgp->opened = 0; 643 } 644 vput(newtextvp); 645 vm_object_deallocate(imgp->object); 646 imgp->object = NULL; 647 imgp->credential_setid = false; 648 if (imgp->newcred != NULL) { 649 crfree(imgp->newcred); 650 imgp->newcred = NULL; 651 } 652 imgp->execpath = NULL; 653 free(imgp->freepath, M_TEMP); 654 imgp->freepath = NULL; 655 /* set new name to that of the interpreter */ 656 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 657 UIO_SYSSPACE, imgp->interpreter_name, td); 658 args->fname = imgp->interpreter_name; 659 goto interpret; 660 } 661 662 /* 663 * NB: We unlock the vnode here because it is believed that none 664 * of the sv_copyout_strings/sv_fixup operations require the vnode. 665 */ 666 VOP_UNLOCK(imgp->vp, 0); 667 668 if (disallow_high_osrel && 669 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 670 error = ENOEXEC; 671 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 672 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 673 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 674 goto exec_fail_dealloc; 675 } 676 677 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 678 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 679 sys_cap_enter(td, NULL); 680 681 /* 682 * Copy out strings (args and env) and initialize stack base. 683 */ 684 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 685 686 /* 687 * Stack setup. 688 */ 689 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 690 if (error != 0) { 691 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 692 goto exec_fail_dealloc; 693 } 694 695 if (args->fdp != NULL) { 696 /* Install a brand new file descriptor table. */ 697 fdinstall_remapped(td, args->fdp); 698 args->fdp = NULL; 699 } else { 700 /* 701 * Keep on using the existing file descriptor table. For 702 * security and other reasons, the file descriptor table 703 * cannot be shared after an exec. 704 */ 705 fdunshare(td); 706 /* close files on exec */ 707 fdcloseexec(td); 708 } 709 710 /* 711 * Malloc things before we need locks. 712 */ 713 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 714 /* Cache arguments if they fit inside our allowance */ 715 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 716 newargs = pargs_alloc(i); 717 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 718 } 719 720 /* 721 * For security and other reasons, signal handlers cannot 722 * be shared after an exec. The new process gets a copy of the old 723 * handlers. In execsigs(), the new process will have its signals 724 * reset. 725 */ 726 if (sigacts_shared(p->p_sigacts)) { 727 oldsigacts = p->p_sigacts; 728 newsigacts = sigacts_alloc(); 729 sigacts_copy(newsigacts, oldsigacts); 730 } 731 732 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 733 734 PROC_LOCK(p); 735 if (oldsigacts) 736 p->p_sigacts = newsigacts; 737 /* Stop profiling */ 738 stopprofclock(p); 739 740 /* reset caught signals */ 741 execsigs(p); 742 743 /* name this process - nameiexec(p, ndp) */ 744 bzero(p->p_comm, sizeof(p->p_comm)); 745 if (args->fname) 746 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 747 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 748 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 749 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 750 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 751 #ifdef KTR 752 sched_clear_tdname(td); 753 #endif 754 755 /* 756 * mark as execed, wakeup the process that vforked (if any) and tell 757 * it that it now has its own resources back 758 */ 759 p->p_flag |= P_EXEC; 760 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 761 p->p_flag2 &= ~P2_NOTRACE; 762 if (p->p_flag & P_PPWAIT) { 763 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 764 cv_broadcast(&p->p_pwait); 765 /* STOPs are no longer ignored, arrange for AST */ 766 signotify(td); 767 } 768 769 /* 770 * Implement image setuid/setgid installation. 771 */ 772 if (imgp->credential_setid) { 773 /* 774 * Turn off syscall tracing for set-id programs, except for 775 * root. Record any set-id flags first to make sure that 776 * we do not regain any tracing during a possible block. 777 */ 778 setsugid(p); 779 780 #ifdef KTRACE 781 if (p->p_tracecred != NULL && 782 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED)) 783 ktrprocexec(p, &tracecred, &tracevp); 784 #endif 785 /* 786 * Close any file descriptors 0..2 that reference procfs, 787 * then make sure file descriptors 0..2 are in use. 788 * 789 * Both fdsetugidsafety() and fdcheckstd() may call functions 790 * taking sleepable locks, so temporarily drop our locks. 791 */ 792 PROC_UNLOCK(p); 793 VOP_UNLOCK(imgp->vp, 0); 794 fdsetugidsafety(td); 795 error = fdcheckstd(td); 796 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 797 if (error != 0) 798 goto exec_fail_dealloc; 799 PROC_LOCK(p); 800 #ifdef MAC 801 if (will_transition) { 802 mac_vnode_execve_transition(oldcred, imgp->newcred, 803 imgp->vp, interpvplabel, imgp); 804 } 805 #endif 806 } else { 807 if (oldcred->cr_uid == oldcred->cr_ruid && 808 oldcred->cr_gid == oldcred->cr_rgid) 809 p->p_flag &= ~P_SUGID; 810 } 811 /* 812 * Set the new credentials. 813 */ 814 if (imgp->newcred != NULL) { 815 proc_set_cred(p, imgp->newcred); 816 crfree(oldcred); 817 oldcred = NULL; 818 } 819 820 /* 821 * Store the vp for use in procfs. This vnode was referenced by namei 822 * or fgetvp_exec. 823 */ 824 oldtextvp = p->p_textvp; 825 p->p_textvp = newtextvp; 826 827 #ifdef KDTRACE_HOOKS 828 /* 829 * Tell the DTrace fasttrap provider about the exec if it 830 * has declared an interest. 831 */ 832 if (dtrace_fasttrap_exec) 833 dtrace_fasttrap_exec(p); 834 #endif 835 836 /* 837 * Notify others that we exec'd, and clear the P_INEXEC flag 838 * as we're now a bona fide freshly-execed process. 839 */ 840 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 841 p->p_flag &= ~P_INEXEC; 842 843 /* clear "fork but no exec" flag, as we _are_ execing */ 844 p->p_acflag &= ~AFORK; 845 846 /* 847 * Free any previous argument cache and replace it with 848 * the new argument cache, if any. 849 */ 850 oldargs = p->p_args; 851 p->p_args = newargs; 852 newargs = NULL; 853 854 PROC_UNLOCK(p); 855 856 #ifdef HWPMC_HOOKS 857 /* 858 * Check if system-wide sampling is in effect or if the 859 * current process is using PMCs. If so, do exec() time 860 * processing. This processing needs to happen AFTER the 861 * P_INEXEC flag is cleared. 862 */ 863 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 864 VOP_UNLOCK(imgp->vp, 0); 865 pe.pm_credentialschanged = credential_changing; 866 pe.pm_entryaddr = imgp->entry_addr; 867 868 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 869 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 870 } 871 #endif 872 873 /* Set values passed into the program in registers. */ 874 (*p->p_sysent->sv_setregs)(td, imgp, (u_long)(uintptr_t)stack_base); 875 876 vfs_mark_atime(imgp->vp, td->td_ucred); 877 878 SDT_PROBE1(proc, , , exec__success, args->fname); 879 880 exec_fail_dealloc: 881 if (imgp->firstpage != NULL) 882 exec_unmap_first_page(imgp); 883 884 if (imgp->vp != NULL) { 885 if (args->fname) 886 NDFREE(&nd, NDF_ONLY_PNBUF); 887 if (imgp->opened) 888 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 889 if (error != 0) 890 vput(imgp->vp); 891 else 892 VOP_UNLOCK(imgp->vp, 0); 893 } 894 895 if (imgp->object != NULL) 896 vm_object_deallocate(imgp->object); 897 898 free(imgp->freepath, M_TEMP); 899 900 if (error == 0) { 901 if (p->p_ptevents & PTRACE_EXEC) { 902 PROC_LOCK(p); 903 if (p->p_ptevents & PTRACE_EXEC) 904 td->td_dbgflags |= TDB_EXEC; 905 PROC_UNLOCK(p); 906 } 907 908 /* 909 * Stop the process here if its stop event mask has 910 * the S_EXEC bit set. 911 */ 912 STOPEVENT(p, S_EXEC, 0); 913 } else { 914 exec_fail: 915 /* we're done here, clear P_INEXEC */ 916 PROC_LOCK(p); 917 p->p_flag &= ~P_INEXEC; 918 PROC_UNLOCK(p); 919 920 SDT_PROBE1(proc, , , exec__failure, error); 921 } 922 923 if (imgp->newcred != NULL && oldcred != NULL) 924 crfree(imgp->newcred); 925 926 #ifdef MAC 927 mac_execve_exit(imgp); 928 mac_execve_interpreter_exit(interpvplabel); 929 #endif 930 exec_free_args(args); 931 932 /* 933 * Handle deferred decrement of ref counts. 934 */ 935 if (oldtextvp != NULL) 936 vrele(oldtextvp); 937 #ifdef KTRACE 938 if (tracevp != NULL) 939 vrele(tracevp); 940 if (tracecred != NULL) 941 crfree(tracecred); 942 #endif 943 pargs_drop(oldargs); 944 pargs_drop(newargs); 945 if (oldsigacts != NULL) 946 sigacts_free(oldsigacts); 947 if (euip != NULL) 948 uifree(euip); 949 950 if (error && imgp->vmspace_destroyed) { 951 /* sorry, no more process anymore. exit gracefully */ 952 exit1(td, 0, SIGABRT); 953 /* NOT REACHED */ 954 } 955 956 #ifdef KTRACE 957 if (error == 0) 958 ktrprocctor(p); 959 #endif 960 961 /* 962 * We don't want cpu_set_syscall_retval() to overwrite any of 963 * the register values put in place by exec_setregs(). 964 * Implementations of cpu_set_syscall_retval() will leave 965 * registers unmodified when returning EJUSTRETURN. 966 */ 967 return (error == 0 ? EJUSTRETURN : error); 968 } 969 970 int 971 exec_map_first_page(struct image_params *imgp) 972 { 973 int rv, i, after, initial_pagein; 974 vm_page_t ma[VM_INITIAL_PAGEIN]; 975 vm_object_t object; 976 977 if (imgp->firstpage != NULL) 978 exec_unmap_first_page(imgp); 979 980 object = imgp->vp->v_object; 981 if (object == NULL) 982 return (EACCES); 983 VM_OBJECT_WLOCK(object); 984 #if VM_NRESERVLEVEL > 0 985 vm_object_color(object, 0); 986 #endif 987 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 988 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 989 vm_page_xbusy(ma[0]); 990 if (!vm_pager_has_page(object, 0, NULL, &after)) { 991 vm_page_lock(ma[0]); 992 vm_page_free(ma[0]); 993 vm_page_unlock(ma[0]); 994 VM_OBJECT_WUNLOCK(object); 995 return (EIO); 996 } 997 initial_pagein = min(after, VM_INITIAL_PAGEIN); 998 KASSERT(initial_pagein <= object->size, 999 ("%s: initial_pagein %d object->size %ju", 1000 __func__, initial_pagein, (uintmax_t )object->size)); 1001 for (i = 1; i < initial_pagein; i++) { 1002 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1003 if (ma[i]->valid) 1004 break; 1005 if (!vm_page_tryxbusy(ma[i])) 1006 break; 1007 } else { 1008 ma[i] = vm_page_alloc(object, i, 1009 VM_ALLOC_NORMAL); 1010 if (ma[i] == NULL) 1011 break; 1012 } 1013 } 1014 initial_pagein = i; 1015 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1016 if (rv != VM_PAGER_OK) { 1017 for (i = 0; i < initial_pagein; i++) { 1018 vm_page_lock(ma[i]); 1019 vm_page_free(ma[i]); 1020 vm_page_unlock(ma[i]); 1021 } 1022 VM_OBJECT_WUNLOCK(object); 1023 return (EIO); 1024 } 1025 vm_page_xunbusy(ma[0]); 1026 for (i = 1; i < initial_pagein; i++) 1027 vm_page_readahead_finish(ma[i]); 1028 } 1029 vm_page_lock(ma[0]); 1030 vm_page_hold(ma[0]); 1031 vm_page_activate(ma[0]); 1032 vm_page_unlock(ma[0]); 1033 VM_OBJECT_WUNLOCK(object); 1034 1035 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1036 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1037 1038 return (0); 1039 } 1040 1041 void 1042 exec_unmap_first_page(struct image_params *imgp) 1043 { 1044 vm_page_t m; 1045 1046 if (imgp->firstpage != NULL) { 1047 m = sf_buf_page(imgp->firstpage); 1048 sf_buf_free(imgp->firstpage); 1049 imgp->firstpage = NULL; 1050 vm_page_lock(m); 1051 vm_page_unhold(m); 1052 vm_page_unlock(m); 1053 } 1054 } 1055 1056 /* 1057 * Destroy old address space, and allocate a new stack. 1058 * The new stack is only sgrowsiz large because it is grown 1059 * automatically on a page fault. 1060 */ 1061 int 1062 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1063 { 1064 int error; 1065 struct proc *p = imgp->proc; 1066 struct vmspace *vmspace = p->p_vmspace; 1067 vm_object_t obj; 1068 struct rlimit rlim_stack; 1069 vm_offset_t sv_minuser, stack_addr; 1070 vm_map_t map; 1071 u_long ssiz; 1072 1073 imgp->vmspace_destroyed = 1; 1074 imgp->sysent = sv; 1075 1076 /* May be called with Giant held */ 1077 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1078 1079 /* 1080 * Blow away entire process VM, if address space not shared, 1081 * otherwise, create a new VM space so that other threads are 1082 * not disrupted 1083 */ 1084 map = &vmspace->vm_map; 1085 if (map_at_zero) 1086 sv_minuser = sv->sv_minuser; 1087 else 1088 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1089 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1090 vm_map_max(map) == sv->sv_maxuser && 1091 cpu_exec_vmspace_reuse(p, map)) { 1092 shmexit(vmspace); 1093 pmap_remove_pages(vmspace_pmap(vmspace)); 1094 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1095 /* 1096 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1097 * must be re-evaluated. 1098 */ 1099 vm_map_lock(map); 1100 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1101 MAP_ASLR_IGNSTART); 1102 vm_map_unlock(map); 1103 } else { 1104 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1105 if (error) 1106 return (error); 1107 vmspace = p->p_vmspace; 1108 map = &vmspace->vm_map; 1109 } 1110 map->flags |= imgp->map_flags; 1111 1112 /* Map a shared page */ 1113 obj = sv->sv_shared_page_obj; 1114 if (obj != NULL) { 1115 vm_object_reference(obj); 1116 error = vm_map_fixed(map, obj, 0, 1117 sv->sv_shared_page_base, sv->sv_shared_page_len, 1118 VM_PROT_READ | VM_PROT_EXECUTE, 1119 VM_PROT_READ | VM_PROT_EXECUTE, 1120 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1121 if (error != KERN_SUCCESS) { 1122 vm_object_deallocate(obj); 1123 return (vm_mmap_to_errno(error)); 1124 } 1125 } 1126 1127 /* Allocate a new stack */ 1128 if (imgp->stack_sz != 0) { 1129 ssiz = trunc_page(imgp->stack_sz); 1130 PROC_LOCK(p); 1131 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1132 PROC_UNLOCK(p); 1133 if (ssiz > rlim_stack.rlim_max) 1134 ssiz = rlim_stack.rlim_max; 1135 if (ssiz > rlim_stack.rlim_cur) { 1136 rlim_stack.rlim_cur = ssiz; 1137 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1138 } 1139 } else if (sv->sv_maxssiz != NULL) { 1140 ssiz = *sv->sv_maxssiz; 1141 } else { 1142 ssiz = maxssiz; 1143 } 1144 stack_addr = sv->sv_usrstack - ssiz; 1145 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1146 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1147 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1148 if (error != KERN_SUCCESS) 1149 return (vm_mmap_to_errno(error)); 1150 1151 /* 1152 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1153 * are still used to enforce the stack rlimit on the process stack. 1154 */ 1155 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1156 vmspace->vm_maxsaddr = (char *)stack_addr; 1157 1158 return (0); 1159 } 1160 1161 /* 1162 * Copy out argument and environment strings from the old process address 1163 * space into the temporary string buffer. 1164 */ 1165 int 1166 exec_copyin_args(struct image_args *args, const char *fname, 1167 enum uio_seg segflg, char **argv, char **envv) 1168 { 1169 u_long arg, env; 1170 int error; 1171 1172 bzero(args, sizeof(*args)); 1173 if (argv == NULL) 1174 return (EFAULT); 1175 1176 /* 1177 * Allocate demand-paged memory for the file name, argument, and 1178 * environment strings. 1179 */ 1180 error = exec_alloc_args(args); 1181 if (error != 0) 1182 return (error); 1183 1184 /* 1185 * Copy the file name. 1186 */ 1187 error = exec_args_add_fname(args, fname, segflg); 1188 if (error != 0) 1189 goto err_exit; 1190 1191 /* 1192 * extract arguments first 1193 */ 1194 for (;;) { 1195 error = fueword(argv++, &arg); 1196 if (error == -1) { 1197 error = EFAULT; 1198 goto err_exit; 1199 } 1200 if (arg == 0) 1201 break; 1202 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1203 UIO_USERSPACE); 1204 if (error != 0) 1205 goto err_exit; 1206 } 1207 1208 /* 1209 * extract environment strings 1210 */ 1211 if (envv) { 1212 for (;;) { 1213 error = fueword(envv++, &env); 1214 if (error == -1) { 1215 error = EFAULT; 1216 goto err_exit; 1217 } 1218 if (env == 0) 1219 break; 1220 error = exec_args_add_env(args, 1221 (char *)(uintptr_t)env, UIO_USERSPACE); 1222 if (error != 0) 1223 goto err_exit; 1224 } 1225 } 1226 1227 return (0); 1228 1229 err_exit: 1230 exec_free_args(args); 1231 return (error); 1232 } 1233 1234 int 1235 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1236 const void *data, size_t datalen, const int *fds, size_t fdslen) 1237 { 1238 struct filedesc *ofdp; 1239 const char *p; 1240 int *kfds; 1241 int error; 1242 1243 memset(args, '\0', sizeof(*args)); 1244 ofdp = td->td_proc->p_fd; 1245 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1246 return (E2BIG); 1247 error = exec_alloc_args(args); 1248 if (error != 0) 1249 return (error); 1250 1251 args->begin_argv = args->buf; 1252 args->stringspace = ARG_MAX; 1253 1254 if (datalen > 0) { 1255 /* 1256 * Argument buffer has been provided. Copy it into the 1257 * kernel as a single string and add a terminating null 1258 * byte. 1259 */ 1260 error = copyin(data, args->begin_argv, datalen); 1261 if (error != 0) 1262 goto err_exit; 1263 args->begin_argv[datalen] = '\0'; 1264 args->endp = args->begin_argv + datalen + 1; 1265 args->stringspace -= datalen + 1; 1266 1267 /* 1268 * Traditional argument counting. Count the number of 1269 * null bytes. 1270 */ 1271 for (p = args->begin_argv; p < args->endp; ++p) 1272 if (*p == '\0') 1273 ++args->argc; 1274 } else { 1275 /* No argument buffer provided. */ 1276 args->endp = args->begin_argv; 1277 } 1278 1279 /* Create new file descriptor table. */ 1280 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1281 error = copyin(fds, kfds, fdslen * sizeof(int)); 1282 if (error != 0) { 1283 free(kfds, M_TEMP); 1284 goto err_exit; 1285 } 1286 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1287 free(kfds, M_TEMP); 1288 if (error != 0) 1289 goto err_exit; 1290 1291 return (0); 1292 err_exit: 1293 exec_free_args(args); 1294 return (error); 1295 } 1296 1297 struct exec_args_kva { 1298 vm_offset_t addr; 1299 u_int gen; 1300 SLIST_ENTRY(exec_args_kva) next; 1301 }; 1302 1303 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1304 1305 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1306 static struct mtx exec_args_kva_mtx; 1307 static u_int exec_args_gen; 1308 1309 static void 1310 exec_prealloc_args_kva(void *arg __unused) 1311 { 1312 struct exec_args_kva *argkva; 1313 u_int i; 1314 1315 SLIST_INIT(&exec_args_kva_freelist); 1316 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1317 for (i = 0; i < exec_map_entries; i++) { 1318 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1319 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1320 argkva->gen = exec_args_gen; 1321 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1322 } 1323 } 1324 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1325 1326 static vm_offset_t 1327 exec_alloc_args_kva(void **cookie) 1328 { 1329 struct exec_args_kva *argkva; 1330 1331 argkva = (void *)atomic_readandclear_ptr( 1332 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1333 if (argkva == NULL) { 1334 mtx_lock(&exec_args_kva_mtx); 1335 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1336 (void)mtx_sleep(&exec_args_kva_freelist, 1337 &exec_args_kva_mtx, 0, "execkva", 0); 1338 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1339 mtx_unlock(&exec_args_kva_mtx); 1340 } 1341 *(struct exec_args_kva **)cookie = argkva; 1342 return (argkva->addr); 1343 } 1344 1345 static void 1346 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1347 { 1348 vm_offset_t base; 1349 1350 base = argkva->addr; 1351 if (argkva->gen != gen) { 1352 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1353 MADV_FREE); 1354 argkva->gen = gen; 1355 } 1356 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1357 (uintptr_t)NULL, (uintptr_t)argkva)) { 1358 mtx_lock(&exec_args_kva_mtx); 1359 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1360 wakeup_one(&exec_args_kva_freelist); 1361 mtx_unlock(&exec_args_kva_mtx); 1362 } 1363 } 1364 1365 static void 1366 exec_free_args_kva(void *cookie) 1367 { 1368 1369 exec_release_args_kva(cookie, exec_args_gen); 1370 } 1371 1372 static void 1373 exec_args_kva_lowmem(void *arg __unused) 1374 { 1375 SLIST_HEAD(, exec_args_kva) head; 1376 struct exec_args_kva *argkva; 1377 u_int gen; 1378 int i; 1379 1380 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1381 1382 /* 1383 * Force an madvise of each KVA range. Any currently allocated ranges 1384 * will have MADV_FREE applied once they are freed. 1385 */ 1386 SLIST_INIT(&head); 1387 mtx_lock(&exec_args_kva_mtx); 1388 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1389 mtx_unlock(&exec_args_kva_mtx); 1390 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1391 SLIST_REMOVE_HEAD(&head, next); 1392 exec_release_args_kva(argkva, gen); 1393 } 1394 1395 CPU_FOREACH(i) { 1396 argkva = (void *)atomic_readandclear_ptr( 1397 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1398 if (argkva != NULL) 1399 exec_release_args_kva(argkva, gen); 1400 } 1401 } 1402 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1403 EVENTHANDLER_PRI_ANY); 1404 1405 /* 1406 * Allocate temporary demand-paged, zero-filled memory for the file name, 1407 * argument, and environment strings. 1408 */ 1409 int 1410 exec_alloc_args(struct image_args *args) 1411 { 1412 1413 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1414 return (0); 1415 } 1416 1417 void 1418 exec_free_args(struct image_args *args) 1419 { 1420 1421 if (args->buf != NULL) { 1422 exec_free_args_kva(args->bufkva); 1423 args->buf = NULL; 1424 } 1425 if (args->fname_buf != NULL) { 1426 free(args->fname_buf, M_TEMP); 1427 args->fname_buf = NULL; 1428 } 1429 if (args->fdp != NULL) 1430 fdescfree_remapped(args->fdp); 1431 } 1432 1433 /* 1434 * A set to functions to fill struct image args. 1435 * 1436 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1437 * fname) before the other functions. All exec_args_add_arg() calls must 1438 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1439 * may be called any time after exec_args_add_fname(). 1440 * 1441 * exec_args_add_fname() - install path to be executed 1442 * exec_args_add_arg() - append an argument string 1443 * exec_args_add_env() - append an env string 1444 * exec_args_adjust_args() - adjust location of the argument list to 1445 * allow new arguments to be prepended 1446 */ 1447 int 1448 exec_args_add_fname(struct image_args *args, const char *fname, 1449 enum uio_seg segflg) 1450 { 1451 int error; 1452 size_t length; 1453 1454 KASSERT(args->fname == NULL, ("fname already appended")); 1455 KASSERT(args->endp == NULL, ("already appending to args")); 1456 1457 if (fname != NULL) { 1458 args->fname = args->buf; 1459 error = segflg == UIO_SYSSPACE ? 1460 copystr(fname, args->fname, PATH_MAX, &length) : 1461 copyinstr(fname, args->fname, PATH_MAX, &length); 1462 if (error != 0) 1463 return (error == ENAMETOOLONG ? E2BIG : error); 1464 } else 1465 length = 0; 1466 1467 /* Set up for _arg_*()/_env_*() */ 1468 args->endp = args->buf + length; 1469 /* begin_argv must be set and kept updated */ 1470 args->begin_argv = args->endp; 1471 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1472 ("too little space remaining for arguments %zu < %zu", 1473 exec_map_entry_size - length, (size_t)ARG_MAX)); 1474 args->stringspace = ARG_MAX; 1475 1476 return (0); 1477 } 1478 1479 static int 1480 exec_args_add_str(struct image_args *args, const char *str, 1481 enum uio_seg segflg, int *countp) 1482 { 1483 int error; 1484 size_t length; 1485 1486 KASSERT(args->endp != NULL, ("endp not initialized")); 1487 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1488 1489 error = (segflg == UIO_SYSSPACE) ? 1490 copystr(str, args->endp, args->stringspace, &length) : 1491 copyinstr(str, args->endp, args->stringspace, &length); 1492 if (error != 0) 1493 return (error == ENAMETOOLONG ? E2BIG : error); 1494 args->stringspace -= length; 1495 args->endp += length; 1496 (*countp)++; 1497 1498 return (0); 1499 } 1500 1501 int 1502 exec_args_add_arg(struct image_args *args, const char *argp, 1503 enum uio_seg segflg) 1504 { 1505 1506 KASSERT(args->envc == 0, ("appending args after env")); 1507 1508 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1509 } 1510 1511 int 1512 exec_args_add_env(struct image_args *args, const char *envp, 1513 enum uio_seg segflg) 1514 { 1515 1516 if (args->envc == 0) 1517 args->begin_envv = args->endp; 1518 1519 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1520 } 1521 1522 int 1523 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1524 { 1525 ssize_t offset; 1526 1527 KASSERT(args->endp != NULL, ("endp not initialized")); 1528 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1529 1530 offset = extend - consume; 1531 if (args->stringspace < offset) 1532 return (E2BIG); 1533 memmove(args->begin_argv + extend, args->begin_argv + consume, 1534 args->endp - args->begin_argv + consume); 1535 if (args->envc > 0) 1536 args->begin_envv += offset; 1537 args->endp += offset; 1538 args->stringspace -= offset; 1539 return (0); 1540 } 1541 1542 char * 1543 exec_args_get_begin_envv(struct image_args *args) 1544 { 1545 1546 KASSERT(args->endp != NULL, ("endp not initialized")); 1547 1548 if (args->envc > 0) 1549 return (args->begin_envv); 1550 return (args->endp); 1551 } 1552 1553 /* 1554 * Copy strings out to the new process address space, constructing new arg 1555 * and env vector tables. Return a pointer to the base so that it can be used 1556 * as the initial stack pointer. 1557 */ 1558 register_t * 1559 exec_copyout_strings(struct image_params *imgp) 1560 { 1561 int argc, envc; 1562 char **vectp; 1563 char *stringp; 1564 uintptr_t destp; 1565 register_t *stack_base; 1566 struct ps_strings *arginfo; 1567 struct proc *p; 1568 size_t execpath_len; 1569 int szsigcode, szps; 1570 char canary[sizeof(long) * 8]; 1571 1572 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1573 /* 1574 * Calculate string base and vector table pointers. 1575 * Also deal with signal trampoline code for this exec type. 1576 */ 1577 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1578 execpath_len = strlen(imgp->execpath) + 1; 1579 else 1580 execpath_len = 0; 1581 p = imgp->proc; 1582 szsigcode = 0; 1583 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1584 if (p->p_sysent->sv_sigcode_base == 0) { 1585 if (p->p_sysent->sv_szsigcode != NULL) 1586 szsigcode = *(p->p_sysent->sv_szsigcode); 1587 } 1588 destp = (uintptr_t)arginfo; 1589 1590 /* 1591 * install sigcode 1592 */ 1593 if (szsigcode != 0) { 1594 destp -= szsigcode; 1595 destp = rounddown2(destp, sizeof(void *)); 1596 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1597 } 1598 1599 /* 1600 * Copy the image path for the rtld. 1601 */ 1602 if (execpath_len != 0) { 1603 destp -= execpath_len; 1604 destp = rounddown2(destp, sizeof(void *)); 1605 imgp->execpathp = destp; 1606 copyout(imgp->execpath, (void *)destp, execpath_len); 1607 } 1608 1609 /* 1610 * Prepare the canary for SSP. 1611 */ 1612 arc4rand(canary, sizeof(canary), 0); 1613 destp -= sizeof(canary); 1614 imgp->canary = destp; 1615 copyout(canary, (void *)destp, sizeof(canary)); 1616 imgp->canarylen = sizeof(canary); 1617 1618 /* 1619 * Prepare the pagesizes array. 1620 */ 1621 destp -= szps; 1622 destp = rounddown2(destp, sizeof(void *)); 1623 imgp->pagesizes = destp; 1624 copyout(pagesizes, (void *)destp, szps); 1625 imgp->pagesizeslen = szps; 1626 1627 destp -= ARG_MAX - imgp->args->stringspace; 1628 destp = rounddown2(destp, sizeof(void *)); 1629 1630 vectp = (char **)destp; 1631 if (imgp->auxargs) { 1632 /* 1633 * Allocate room on the stack for the ELF auxargs 1634 * array. It has up to AT_COUNT entries. 1635 */ 1636 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 1637 sizeof(*vectp)); 1638 } 1639 1640 /* 1641 * Allocate room for the argv[] and env vectors including the 1642 * terminating NULL pointers. 1643 */ 1644 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1645 1646 /* 1647 * vectp also becomes our initial stack base 1648 */ 1649 stack_base = (register_t *)vectp; 1650 1651 stringp = imgp->args->begin_argv; 1652 argc = imgp->args->argc; 1653 envc = imgp->args->envc; 1654 1655 /* 1656 * Copy out strings - arguments and environment. 1657 */ 1658 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1659 1660 /* 1661 * Fill in "ps_strings" struct for ps, w, etc. 1662 */ 1663 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1664 suword32(&arginfo->ps_nargvstr, argc); 1665 1666 /* 1667 * Fill in argument portion of vector table. 1668 */ 1669 for (; argc > 0; --argc) { 1670 suword(vectp++, (long)(intptr_t)destp); 1671 while (*stringp++ != 0) 1672 destp++; 1673 destp++; 1674 } 1675 1676 /* a null vector table pointer separates the argp's from the envp's */ 1677 suword(vectp++, 0); 1678 1679 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1680 suword32(&arginfo->ps_nenvstr, envc); 1681 1682 /* 1683 * Fill in environment portion of vector table. 1684 */ 1685 for (; envc > 0; --envc) { 1686 suword(vectp++, (long)(intptr_t)destp); 1687 while (*stringp++ != 0) 1688 destp++; 1689 destp++; 1690 } 1691 1692 /* end of vector table is a null pointer */ 1693 suword(vectp, 0); 1694 1695 return (stack_base); 1696 } 1697 1698 /* 1699 * Check permissions of file to execute. 1700 * Called with imgp->vp locked. 1701 * Return 0 for success or error code on failure. 1702 */ 1703 int 1704 exec_check_permissions(struct image_params *imgp) 1705 { 1706 struct vnode *vp = imgp->vp; 1707 struct vattr *attr = imgp->attr; 1708 struct thread *td; 1709 int error, writecount; 1710 1711 td = curthread; 1712 1713 /* Get file attributes */ 1714 error = VOP_GETATTR(vp, attr, td->td_ucred); 1715 if (error) 1716 return (error); 1717 1718 #ifdef MAC 1719 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1720 if (error) 1721 return (error); 1722 #endif 1723 1724 /* 1725 * 1) Check if file execution is disabled for the filesystem that 1726 * this file resides on. 1727 * 2) Ensure that at least one execute bit is on. Otherwise, a 1728 * privileged user will always succeed, and we don't want this 1729 * to happen unless the file really is executable. 1730 * 3) Ensure that the file is a regular file. 1731 */ 1732 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1733 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1734 (attr->va_type != VREG)) 1735 return (EACCES); 1736 1737 /* 1738 * Zero length files can't be exec'd 1739 */ 1740 if (attr->va_size == 0) 1741 return (ENOEXEC); 1742 1743 /* 1744 * Check for execute permission to file based on current credentials. 1745 */ 1746 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1747 if (error) 1748 return (error); 1749 1750 /* 1751 * Check number of open-for-writes on the file and deny execution 1752 * if there are any. 1753 */ 1754 error = VOP_GET_WRITECOUNT(vp, &writecount); 1755 if (error != 0) 1756 return (error); 1757 if (writecount != 0) 1758 return (ETXTBSY); 1759 1760 /* 1761 * Call filesystem specific open routine (which does nothing in the 1762 * general case). 1763 */ 1764 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1765 if (error == 0) 1766 imgp->opened = 1; 1767 return (error); 1768 } 1769 1770 /* 1771 * Exec handler registration 1772 */ 1773 int 1774 exec_register(const struct execsw *execsw_arg) 1775 { 1776 const struct execsw **es, **xs, **newexecsw; 1777 u_int count = 2; /* New slot and trailing NULL */ 1778 1779 if (execsw) 1780 for (es = execsw; *es; es++) 1781 count++; 1782 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1783 xs = newexecsw; 1784 if (execsw) 1785 for (es = execsw; *es; es++) 1786 *xs++ = *es; 1787 *xs++ = execsw_arg; 1788 *xs = NULL; 1789 if (execsw) 1790 free(execsw, M_TEMP); 1791 execsw = newexecsw; 1792 return (0); 1793 } 1794 1795 int 1796 exec_unregister(const struct execsw *execsw_arg) 1797 { 1798 const struct execsw **es, **xs, **newexecsw; 1799 int count = 1; 1800 1801 if (execsw == NULL) 1802 panic("unregister with no handlers left?\n"); 1803 1804 for (es = execsw; *es; es++) { 1805 if (*es == execsw_arg) 1806 break; 1807 } 1808 if (*es == NULL) 1809 return (ENOENT); 1810 for (es = execsw; *es; es++) 1811 if (*es != execsw_arg) 1812 count++; 1813 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1814 xs = newexecsw; 1815 for (es = execsw; *es; es++) 1816 if (*es != execsw_arg) 1817 *xs++ = *es; 1818 *xs = NULL; 1819 if (execsw) 1820 free(execsw, M_TEMP); 1821 execsw = newexecsw; 1822 return (0); 1823 } 1824