1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1993, David Greenman 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_capsicum.h" 33 #include "opt_hwpmc_hooks.h" 34 #include "opt_ktrace.h" 35 #include "opt_vm.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/acct.h> 40 #include <sys/capsicum.h> 41 #include <sys/eventhandler.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/filedesc.h> 45 #include <sys/imgact.h> 46 #include <sys/imgact_elf.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mman.h> 51 #include <sys/mount.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/pioctl.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/ptrace.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sched.h> 61 #include <sys/sdt.h> 62 #include <sys/sf_buf.h> 63 #include <sys/shm.h> 64 #include <sys/signalvar.h> 65 #include <sys/smp.h> 66 #include <sys/stat.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/sysproto.h> 71 #include <sys/vnode.h> 72 #include <sys/wait.h> 73 #ifdef KTRACE 74 #include <sys/ktrace.h> 75 #endif 76 77 #include <vm/vm.h> 78 #include <vm/vm_param.h> 79 #include <vm/pmap.h> 80 #include <vm/vm_page.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_kern.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_pager.h> 86 87 #ifdef HWPMC_HOOKS 88 #include <sys/pmckern.h> 89 #endif 90 91 #include <machine/reg.h> 92 93 #include <security/audit/audit.h> 94 #include <security/mac/mac_framework.h> 95 96 #ifdef KDTRACE_HOOKS 97 #include <sys/dtrace_bsd.h> 98 dtrace_execexit_func_t dtrace_fasttrap_exec; 99 #endif 100 101 SDT_PROVIDER_DECLARE(proc); 102 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 105 106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 107 108 int coredump_pack_fileinfo = 1; 109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 110 &coredump_pack_fileinfo, 0, 111 "Enable file path packing in 'procstat -f' coredump notes"); 112 113 int coredump_pack_vmmapinfo = 1; 114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 115 &coredump_pack_vmmapinfo, 0, 116 "Enable file path packing in 'procstat -v' coredump notes"); 117 118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 121 static int do_execve(struct thread *td, struct image_args *args, 122 struct mac *mac_p); 123 124 /* XXX This should be vm_size_t. */ 125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 127 128 /* XXX This should be vm_size_t. */ 129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 131 132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 133 NULL, 0, sysctl_kern_stackprot, "I", ""); 134 135 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 137 &ps_arg_cache_limit, 0, ""); 138 139 static int disallow_high_osrel; 140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 141 &disallow_high_osrel, 0, 142 "Disallow execution of binaries built for higher version of the world"); 143 144 static int map_at_zero = 0; 145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 146 "Permit processes to map an object at virtual address 0."); 147 148 EVENTHANDLER_LIST_DECLARE(process_exec); 149 150 static int 151 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 152 { 153 struct proc *p; 154 int error; 155 156 p = curproc; 157 #ifdef SCTL_MASK32 158 if (req->flags & SCTL_MASK32) { 159 unsigned int val; 160 val = (unsigned int)p->p_sysent->sv_psstrings; 161 error = SYSCTL_OUT(req, &val, sizeof(val)); 162 } else 163 #endif 164 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 165 sizeof(p->p_sysent->sv_psstrings)); 166 return error; 167 } 168 169 static int 170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 171 { 172 struct proc *p; 173 int error; 174 175 p = curproc; 176 #ifdef SCTL_MASK32 177 if (req->flags & SCTL_MASK32) { 178 unsigned int val; 179 val = (unsigned int)p->p_sysent->sv_usrstack; 180 error = SYSCTL_OUT(req, &val, sizeof(val)); 181 } else 182 #endif 183 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 184 sizeof(p->p_sysent->sv_usrstack)); 185 return error; 186 } 187 188 static int 189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 190 { 191 struct proc *p; 192 193 p = curproc; 194 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 195 sizeof(p->p_sysent->sv_stackprot))); 196 } 197 198 /* 199 * Each of the items is a pointer to a `const struct execsw', hence the 200 * double pointer here. 201 */ 202 static const struct execsw **execsw; 203 204 #ifndef _SYS_SYSPROTO_H_ 205 struct execve_args { 206 char *fname; 207 char **argv; 208 char **envv; 209 }; 210 #endif 211 212 int 213 sys_execve(struct thread *td, struct execve_args *uap) 214 { 215 struct image_args args; 216 struct vmspace *oldvmspace; 217 int error; 218 219 error = pre_execve(td, &oldvmspace); 220 if (error != 0) 221 return (error); 222 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 223 uap->argv, uap->envv); 224 if (error == 0) 225 error = kern_execve(td, &args, NULL); 226 post_execve(td, error, oldvmspace); 227 return (error); 228 } 229 230 #ifndef _SYS_SYSPROTO_H_ 231 struct fexecve_args { 232 int fd; 233 char **argv; 234 char **envv; 235 } 236 #endif 237 int 238 sys_fexecve(struct thread *td, struct fexecve_args *uap) 239 { 240 struct image_args args; 241 struct vmspace *oldvmspace; 242 int error; 243 244 error = pre_execve(td, &oldvmspace); 245 if (error != 0) 246 return (error); 247 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 248 uap->argv, uap->envv); 249 if (error == 0) { 250 args.fd = uap->fd; 251 error = kern_execve(td, &args, NULL); 252 } 253 post_execve(td, error, oldvmspace); 254 return (error); 255 } 256 257 #ifndef _SYS_SYSPROTO_H_ 258 struct __mac_execve_args { 259 char *fname; 260 char **argv; 261 char **envv; 262 struct mac *mac_p; 263 }; 264 #endif 265 266 int 267 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 268 { 269 #ifdef MAC 270 struct image_args args; 271 struct vmspace *oldvmspace; 272 int error; 273 274 error = pre_execve(td, &oldvmspace); 275 if (error != 0) 276 return (error); 277 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 278 uap->argv, uap->envv); 279 if (error == 0) 280 error = kern_execve(td, &args, uap->mac_p); 281 post_execve(td, error, oldvmspace); 282 return (error); 283 #else 284 return (ENOSYS); 285 #endif 286 } 287 288 int 289 pre_execve(struct thread *td, struct vmspace **oldvmspace) 290 { 291 struct proc *p; 292 int error; 293 294 KASSERT(td == curthread, ("non-current thread %p", td)); 295 error = 0; 296 p = td->td_proc; 297 if ((p->p_flag & P_HADTHREADS) != 0) { 298 PROC_LOCK(p); 299 if (thread_single(p, SINGLE_BOUNDARY) != 0) 300 error = ERESTART; 301 PROC_UNLOCK(p); 302 } 303 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 304 ("nested execve")); 305 *oldvmspace = p->p_vmspace; 306 return (error); 307 } 308 309 void 310 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 311 { 312 struct proc *p; 313 314 KASSERT(td == curthread, ("non-current thread %p", td)); 315 p = td->td_proc; 316 if ((p->p_flag & P_HADTHREADS) != 0) { 317 PROC_LOCK(p); 318 /* 319 * If success, we upgrade to SINGLE_EXIT state to 320 * force other threads to suicide. 321 */ 322 if (error == EJUSTRETURN) 323 thread_single(p, SINGLE_EXIT); 324 else 325 thread_single_end(p, SINGLE_BOUNDARY); 326 PROC_UNLOCK(p); 327 } 328 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 329 KASSERT(p->p_vmspace != oldvmspace, 330 ("oldvmspace still used")); 331 vmspace_free(oldvmspace); 332 td->td_pflags &= ~TDP_EXECVMSPC; 333 } 334 } 335 336 /* 337 * XXX: kern_execve has the astonishing property of not always returning to 338 * the caller. If sufficiently bad things happen during the call to 339 * do_execve(), it can end up calling exit1(); as a result, callers must 340 * avoid doing anything which they might need to undo (e.g., allocating 341 * memory). 342 */ 343 int 344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 345 { 346 347 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 348 args->begin_envv - args->begin_argv); 349 AUDIT_ARG_ENVV(args->begin_envv, args->envc, 350 args->endp - args->begin_envv); 351 return (do_execve(td, args, mac_p)); 352 } 353 354 /* 355 * In-kernel implementation of execve(). All arguments are assumed to be 356 * userspace pointers from the passed thread. 357 */ 358 static int 359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 360 { 361 struct proc *p = td->td_proc; 362 struct nameidata nd; 363 struct ucred *oldcred; 364 struct uidinfo *euip = NULL; 365 register_t *stack_base; 366 int error, i; 367 struct image_params image_params, *imgp; 368 struct vattr attr; 369 int (*img_first)(struct image_params *); 370 struct pargs *oldargs = NULL, *newargs = NULL; 371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 372 #ifdef KTRACE 373 struct vnode *tracevp = NULL; 374 struct ucred *tracecred = NULL; 375 #endif 376 struct vnode *oldtextvp = NULL, *newtextvp; 377 int credential_changing; 378 #ifdef MAC 379 struct label *interpvplabel = NULL; 380 int will_transition; 381 #endif 382 #ifdef HWPMC_HOOKS 383 struct pmckern_procexec pe; 384 #endif 385 static const char fexecv_proc_title[] = "(fexecv)"; 386 387 imgp = &image_params; 388 389 /* 390 * Lock the process and set the P_INEXEC flag to indicate that 391 * it should be left alone until we're done here. This is 392 * necessary to avoid race conditions - e.g. in ptrace() - 393 * that might allow a local user to illicitly obtain elevated 394 * privileges. 395 */ 396 PROC_LOCK(p); 397 KASSERT((p->p_flag & P_INEXEC) == 0, 398 ("%s(): process already has P_INEXEC flag", __func__)); 399 p->p_flag |= P_INEXEC; 400 PROC_UNLOCK(p); 401 402 /* 403 * Initialize part of the common data 404 */ 405 bzero(imgp, sizeof(*imgp)); 406 imgp->proc = p; 407 imgp->attr = &attr; 408 imgp->args = args; 409 oldcred = p->p_ucred; 410 411 #ifdef MAC 412 error = mac_execve_enter(imgp, mac_p); 413 if (error) 414 goto exec_fail; 415 #endif 416 417 /* 418 * Translate the file name. namei() returns a vnode pointer 419 * in ni_vp among other things. 420 * 421 * XXXAUDIT: It would be desirable to also audit the name of the 422 * interpreter if this is an interpreted binary. 423 */ 424 if (args->fname != NULL) { 425 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 426 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 427 } 428 429 SDT_PROBE1(proc, , , exec, args->fname); 430 431 interpret: 432 if (args->fname != NULL) { 433 #ifdef CAPABILITY_MODE 434 /* 435 * While capability mode can't reach this point via direct 436 * path arguments to execve(), we also don't allow 437 * interpreters to be used in capability mode (for now). 438 * Catch indirect lookups and return a permissions error. 439 */ 440 if (IN_CAPABILITY_MODE(td)) { 441 error = ECAPMODE; 442 goto exec_fail; 443 } 444 #endif 445 error = namei(&nd); 446 if (error) 447 goto exec_fail; 448 449 newtextvp = nd.ni_vp; 450 imgp->vp = newtextvp; 451 } else { 452 AUDIT_ARG_FD(args->fd); 453 /* 454 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 455 */ 456 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 457 if (error) 458 goto exec_fail; 459 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 460 AUDIT_ARG_VNODE1(newtextvp); 461 imgp->vp = newtextvp; 462 } 463 464 /* 465 * Check file permissions. Also 'opens' file and sets its vnode to 466 * text mode. 467 */ 468 error = exec_check_permissions(imgp); 469 if (error) 470 goto exec_fail_dealloc; 471 472 imgp->object = imgp->vp->v_object; 473 if (imgp->object != NULL) 474 vm_object_reference(imgp->object); 475 476 error = exec_map_first_page(imgp); 477 if (error) 478 goto exec_fail_dealloc; 479 480 imgp->proc->p_osrel = 0; 481 imgp->proc->p_fctl0 = 0; 482 483 /* 484 * Implement image setuid/setgid. 485 * 486 * Determine new credentials before attempting image activators 487 * so that it can be used by process_exec handlers to determine 488 * credential/setid changes. 489 * 490 * Don't honor setuid/setgid if the filesystem prohibits it or if 491 * the process is being traced. 492 * 493 * We disable setuid/setgid/etc in capability mode on the basis 494 * that most setugid applications are not written with that 495 * environment in mind, and will therefore almost certainly operate 496 * incorrectly. In principle there's no reason that setugid 497 * applications might not be useful in capability mode, so we may want 498 * to reconsider this conservative design choice in the future. 499 * 500 * XXXMAC: For the time being, use NOSUID to also prohibit 501 * transitions on the file system. 502 */ 503 credential_changing = 0; 504 credential_changing |= (attr.va_mode & S_ISUID) && 505 oldcred->cr_uid != attr.va_uid; 506 credential_changing |= (attr.va_mode & S_ISGID) && 507 oldcred->cr_gid != attr.va_gid; 508 #ifdef MAC 509 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 510 interpvplabel, imgp); 511 credential_changing |= will_transition; 512 #endif 513 514 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 515 if (credential_changing) 516 imgp->proc->p_pdeathsig = 0; 517 518 if (credential_changing && 519 #ifdef CAPABILITY_MODE 520 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 521 #endif 522 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 523 (p->p_flag & P_TRACED) == 0) { 524 imgp->credential_setid = true; 525 VOP_UNLOCK(imgp->vp, 0); 526 imgp->newcred = crdup(oldcred); 527 if (attr.va_mode & S_ISUID) { 528 euip = uifind(attr.va_uid); 529 change_euid(imgp->newcred, euip); 530 } 531 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 532 if (attr.va_mode & S_ISGID) 533 change_egid(imgp->newcred, attr.va_gid); 534 /* 535 * Implement correct POSIX saved-id behavior. 536 * 537 * XXXMAC: Note that the current logic will save the 538 * uid and gid if a MAC domain transition occurs, even 539 * though maybe it shouldn't. 540 */ 541 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 542 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 543 } else { 544 /* 545 * Implement correct POSIX saved-id behavior. 546 * 547 * XXX: It's not clear that the existing behavior is 548 * POSIX-compliant. A number of sources indicate that the 549 * saved uid/gid should only be updated if the new ruid is 550 * not equal to the old ruid, or the new euid is not equal 551 * to the old euid and the new euid is not equal to the old 552 * ruid. The FreeBSD code always updates the saved uid/gid. 553 * Also, this code uses the new (replaced) euid and egid as 554 * the source, which may or may not be the right ones to use. 555 */ 556 if (oldcred->cr_svuid != oldcred->cr_uid || 557 oldcred->cr_svgid != oldcred->cr_gid) { 558 VOP_UNLOCK(imgp->vp, 0); 559 imgp->newcred = crdup(oldcred); 560 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 561 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 562 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 563 } 564 } 565 /* The new credentials are installed into the process later. */ 566 567 /* 568 * Do the best to calculate the full path to the image file. 569 */ 570 if (args->fname != NULL && args->fname[0] == '/') 571 imgp->execpath = args->fname; 572 else { 573 VOP_UNLOCK(imgp->vp, 0); 574 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 575 &imgp->freepath) != 0) 576 imgp->execpath = args->fname; 577 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 578 } 579 580 /* 581 * If the current process has a special image activator it 582 * wants to try first, call it. For example, emulating shell 583 * scripts differently. 584 */ 585 error = -1; 586 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 587 error = img_first(imgp); 588 589 /* 590 * Loop through the list of image activators, calling each one. 591 * An activator returns -1 if there is no match, 0 on success, 592 * and an error otherwise. 593 */ 594 for (i = 0; error == -1 && execsw[i]; ++i) { 595 if (execsw[i]->ex_imgact == NULL || 596 execsw[i]->ex_imgact == img_first) { 597 continue; 598 } 599 error = (*execsw[i]->ex_imgact)(imgp); 600 } 601 602 if (error) { 603 if (error == -1) 604 error = ENOEXEC; 605 goto exec_fail_dealloc; 606 } 607 608 /* 609 * Special interpreter operation, cleanup and loop up to try to 610 * activate the interpreter. 611 */ 612 if (imgp->interpreted) { 613 exec_unmap_first_page(imgp); 614 /* 615 * The text reference needs to be removed for scripts. 616 * There is a short period before we determine that 617 * something is a script where text reference is active. 618 * The vnode lock is held over this entire period 619 * so nothing should illegitimately be blocked. 620 */ 621 VOP_UNSET_TEXT_CHECKED(imgp->vp); 622 /* free name buffer and old vnode */ 623 if (args->fname != NULL) 624 NDFREE(&nd, NDF_ONLY_PNBUF); 625 #ifdef MAC 626 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 627 #endif 628 if (imgp->opened) { 629 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 630 imgp->opened = 0; 631 } 632 vput(newtextvp); 633 vm_object_deallocate(imgp->object); 634 imgp->object = NULL; 635 imgp->credential_setid = false; 636 if (imgp->newcred != NULL) { 637 crfree(imgp->newcred); 638 imgp->newcred = NULL; 639 } 640 imgp->execpath = NULL; 641 free(imgp->freepath, M_TEMP); 642 imgp->freepath = NULL; 643 /* set new name to that of the interpreter */ 644 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 645 UIO_SYSSPACE, imgp->interpreter_name, td); 646 args->fname = imgp->interpreter_name; 647 goto interpret; 648 } 649 650 /* 651 * NB: We unlock the vnode here because it is believed that none 652 * of the sv_copyout_strings/sv_fixup operations require the vnode. 653 */ 654 VOP_UNLOCK(imgp->vp, 0); 655 656 if (disallow_high_osrel && 657 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 658 error = ENOEXEC; 659 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 660 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 661 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 662 goto exec_fail_dealloc; 663 } 664 665 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 666 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 667 sys_cap_enter(td, NULL); 668 669 /* 670 * Copy out strings (args and env) and initialize stack base 671 */ 672 if (p->p_sysent->sv_copyout_strings) 673 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 674 else 675 stack_base = exec_copyout_strings(imgp); 676 677 /* 678 * If custom stack fixup routine present for this process 679 * let it do the stack setup. 680 * Else stuff argument count as first item on stack 681 */ 682 if (p->p_sysent->sv_fixup != NULL) 683 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 684 else 685 error = suword(--stack_base, imgp->args->argc) == 0 ? 686 0 : EFAULT; 687 if (error != 0) { 688 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 689 goto exec_fail_dealloc; 690 } 691 692 if (args->fdp != NULL) { 693 /* Install a brand new file descriptor table. */ 694 fdinstall_remapped(td, args->fdp); 695 args->fdp = NULL; 696 } else { 697 /* 698 * Keep on using the existing file descriptor table. For 699 * security and other reasons, the file descriptor table 700 * cannot be shared after an exec. 701 */ 702 fdunshare(td); 703 /* close files on exec */ 704 fdcloseexec(td); 705 } 706 707 /* 708 * Malloc things before we need locks. 709 */ 710 i = imgp->args->begin_envv - imgp->args->begin_argv; 711 /* Cache arguments if they fit inside our allowance */ 712 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 713 newargs = pargs_alloc(i); 714 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 715 } 716 717 /* 718 * For security and other reasons, signal handlers cannot 719 * be shared after an exec. The new process gets a copy of the old 720 * handlers. In execsigs(), the new process will have its signals 721 * reset. 722 */ 723 if (sigacts_shared(p->p_sigacts)) { 724 oldsigacts = p->p_sigacts; 725 newsigacts = sigacts_alloc(); 726 sigacts_copy(newsigacts, oldsigacts); 727 } 728 729 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 730 731 PROC_LOCK(p); 732 if (oldsigacts) 733 p->p_sigacts = newsigacts; 734 /* Stop profiling */ 735 stopprofclock(p); 736 737 /* reset caught signals */ 738 execsigs(p); 739 740 /* name this process - nameiexec(p, ndp) */ 741 bzero(p->p_comm, sizeof(p->p_comm)); 742 if (args->fname) 743 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 744 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 745 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 746 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 747 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 748 #ifdef KTR 749 sched_clear_tdname(td); 750 #endif 751 752 /* 753 * mark as execed, wakeup the process that vforked (if any) and tell 754 * it that it now has its own resources back 755 */ 756 p->p_flag |= P_EXEC; 757 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 758 p->p_flag2 &= ~P2_NOTRACE; 759 if (p->p_flag & P_PPWAIT) { 760 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 761 cv_broadcast(&p->p_pwait); 762 /* STOPs are no longer ignored, arrange for AST */ 763 signotify(td); 764 } 765 766 /* 767 * Implement image setuid/setgid installation. 768 */ 769 if (imgp->credential_setid) { 770 /* 771 * Turn off syscall tracing for set-id programs, except for 772 * root. Record any set-id flags first to make sure that 773 * we do not regain any tracing during a possible block. 774 */ 775 setsugid(p); 776 777 #ifdef KTRACE 778 if (p->p_tracecred != NULL && 779 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED, 0)) 780 ktrprocexec(p, &tracecred, &tracevp); 781 #endif 782 /* 783 * Close any file descriptors 0..2 that reference procfs, 784 * then make sure file descriptors 0..2 are in use. 785 * 786 * Both fdsetugidsafety() and fdcheckstd() may call functions 787 * taking sleepable locks, so temporarily drop our locks. 788 */ 789 PROC_UNLOCK(p); 790 VOP_UNLOCK(imgp->vp, 0); 791 fdsetugidsafety(td); 792 error = fdcheckstd(td); 793 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 794 if (error != 0) 795 goto exec_fail_dealloc; 796 PROC_LOCK(p); 797 #ifdef MAC 798 if (will_transition) { 799 mac_vnode_execve_transition(oldcred, imgp->newcred, 800 imgp->vp, interpvplabel, imgp); 801 } 802 #endif 803 } else { 804 if (oldcred->cr_uid == oldcred->cr_ruid && 805 oldcred->cr_gid == oldcred->cr_rgid) 806 p->p_flag &= ~P_SUGID; 807 } 808 /* 809 * Set the new credentials. 810 */ 811 if (imgp->newcred != NULL) { 812 proc_set_cred(p, imgp->newcred); 813 crfree(oldcred); 814 oldcred = NULL; 815 } 816 817 /* 818 * Store the vp for use in procfs. This vnode was referenced by namei 819 * or fgetvp_exec. 820 */ 821 oldtextvp = p->p_textvp; 822 p->p_textvp = newtextvp; 823 824 #ifdef KDTRACE_HOOKS 825 /* 826 * Tell the DTrace fasttrap provider about the exec if it 827 * has declared an interest. 828 */ 829 if (dtrace_fasttrap_exec) 830 dtrace_fasttrap_exec(p); 831 #endif 832 833 /* 834 * Notify others that we exec'd, and clear the P_INEXEC flag 835 * as we're now a bona fide freshly-execed process. 836 */ 837 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 838 p->p_flag &= ~P_INEXEC; 839 840 /* clear "fork but no exec" flag, as we _are_ execing */ 841 p->p_acflag &= ~AFORK; 842 843 /* 844 * Free any previous argument cache and replace it with 845 * the new argument cache, if any. 846 */ 847 oldargs = p->p_args; 848 p->p_args = newargs; 849 newargs = NULL; 850 851 PROC_UNLOCK(p); 852 853 #ifdef HWPMC_HOOKS 854 /* 855 * Check if system-wide sampling is in effect or if the 856 * current process is using PMCs. If so, do exec() time 857 * processing. This processing needs to happen AFTER the 858 * P_INEXEC flag is cleared. 859 */ 860 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 861 VOP_UNLOCK(imgp->vp, 0); 862 pe.pm_credentialschanged = credential_changing; 863 pe.pm_entryaddr = imgp->entry_addr; 864 865 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 866 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 867 } 868 #endif 869 870 /* Set values passed into the program in registers. */ 871 if (p->p_sysent->sv_setregs) 872 (*p->p_sysent->sv_setregs)(td, imgp, 873 (u_long)(uintptr_t)stack_base); 874 else 875 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 876 877 vfs_mark_atime(imgp->vp, td->td_ucred); 878 879 SDT_PROBE1(proc, , , exec__success, args->fname); 880 881 exec_fail_dealloc: 882 if (imgp->firstpage != NULL) 883 exec_unmap_first_page(imgp); 884 885 if (imgp->vp != NULL) { 886 if (args->fname) 887 NDFREE(&nd, NDF_ONLY_PNBUF); 888 if (imgp->opened) 889 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 890 if (imgp->textset) 891 VOP_UNSET_TEXT_CHECKED(imgp->vp); 892 if (error != 0) 893 vput(imgp->vp); 894 else 895 VOP_UNLOCK(imgp->vp, 0); 896 } 897 898 if (imgp->object != NULL) 899 vm_object_deallocate(imgp->object); 900 901 free(imgp->freepath, M_TEMP); 902 903 if (error == 0) { 904 if (p->p_ptevents & PTRACE_EXEC) { 905 PROC_LOCK(p); 906 if (p->p_ptevents & PTRACE_EXEC) 907 td->td_dbgflags |= TDB_EXEC; 908 PROC_UNLOCK(p); 909 } 910 911 /* 912 * Stop the process here if its stop event mask has 913 * the S_EXEC bit set. 914 */ 915 STOPEVENT(p, S_EXEC, 0); 916 } else { 917 exec_fail: 918 /* we're done here, clear P_INEXEC */ 919 PROC_LOCK(p); 920 p->p_flag &= ~P_INEXEC; 921 PROC_UNLOCK(p); 922 923 SDT_PROBE1(proc, , , exec__failure, error); 924 } 925 926 if (imgp->newcred != NULL && oldcred != NULL) 927 crfree(imgp->newcred); 928 929 #ifdef MAC 930 mac_execve_exit(imgp); 931 mac_execve_interpreter_exit(interpvplabel); 932 #endif 933 exec_free_args(args); 934 935 /* 936 * Handle deferred decrement of ref counts. 937 */ 938 if (oldtextvp != NULL) 939 vrele(oldtextvp); 940 #ifdef KTRACE 941 if (tracevp != NULL) 942 vrele(tracevp); 943 if (tracecred != NULL) 944 crfree(tracecred); 945 #endif 946 pargs_drop(oldargs); 947 pargs_drop(newargs); 948 if (oldsigacts != NULL) 949 sigacts_free(oldsigacts); 950 if (euip != NULL) 951 uifree(euip); 952 953 if (error && imgp->vmspace_destroyed) { 954 /* sorry, no more process anymore. exit gracefully */ 955 exit1(td, 0, SIGABRT); 956 /* NOT REACHED */ 957 } 958 959 #ifdef KTRACE 960 if (error == 0) 961 ktrprocctor(p); 962 #endif 963 964 /* 965 * We don't want cpu_set_syscall_retval() to overwrite any of 966 * the register values put in place by exec_setregs(). 967 * Implementations of cpu_set_syscall_retval() will leave 968 * registers unmodified when returning EJUSTRETURN. 969 */ 970 return (error == 0 ? EJUSTRETURN : error); 971 } 972 973 int 974 exec_map_first_page(struct image_params *imgp) 975 { 976 int rv, i, after, initial_pagein; 977 vm_page_t ma[VM_INITIAL_PAGEIN]; 978 vm_object_t object; 979 980 if (imgp->firstpage != NULL) 981 exec_unmap_first_page(imgp); 982 983 object = imgp->vp->v_object; 984 if (object == NULL) 985 return (EACCES); 986 VM_OBJECT_WLOCK(object); 987 #if VM_NRESERVLEVEL > 0 988 vm_object_color(object, 0); 989 #endif 990 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 991 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 992 vm_page_xbusy(ma[0]); 993 if (!vm_pager_has_page(object, 0, NULL, &after)) { 994 vm_page_lock(ma[0]); 995 vm_page_free(ma[0]); 996 vm_page_unlock(ma[0]); 997 VM_OBJECT_WUNLOCK(object); 998 return (EIO); 999 } 1000 initial_pagein = min(after, VM_INITIAL_PAGEIN); 1001 KASSERT(initial_pagein <= object->size, 1002 ("%s: initial_pagein %d object->size %ju", 1003 __func__, initial_pagein, (uintmax_t )object->size)); 1004 for (i = 1; i < initial_pagein; i++) { 1005 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1006 if (ma[i]->valid) 1007 break; 1008 if (!vm_page_tryxbusy(ma[i])) 1009 break; 1010 } else { 1011 ma[i] = vm_page_alloc(object, i, 1012 VM_ALLOC_NORMAL); 1013 if (ma[i] == NULL) 1014 break; 1015 } 1016 } 1017 initial_pagein = i; 1018 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1019 if (rv != VM_PAGER_OK) { 1020 for (i = 0; i < initial_pagein; i++) { 1021 vm_page_lock(ma[i]); 1022 vm_page_free(ma[i]); 1023 vm_page_unlock(ma[i]); 1024 } 1025 VM_OBJECT_WUNLOCK(object); 1026 return (EIO); 1027 } 1028 vm_page_xunbusy(ma[0]); 1029 for (i = 1; i < initial_pagein; i++) 1030 vm_page_readahead_finish(ma[i]); 1031 } 1032 vm_page_lock(ma[0]); 1033 vm_page_hold(ma[0]); 1034 vm_page_activate(ma[0]); 1035 vm_page_unlock(ma[0]); 1036 VM_OBJECT_WUNLOCK(object); 1037 1038 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1039 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1040 1041 return (0); 1042 } 1043 1044 void 1045 exec_unmap_first_page(struct image_params *imgp) 1046 { 1047 vm_page_t m; 1048 1049 if (imgp->firstpage != NULL) { 1050 m = sf_buf_page(imgp->firstpage); 1051 sf_buf_free(imgp->firstpage); 1052 imgp->firstpage = NULL; 1053 vm_page_lock(m); 1054 vm_page_unhold(m); 1055 vm_page_unlock(m); 1056 } 1057 } 1058 1059 /* 1060 * Destroy old address space, and allocate a new stack. 1061 * The new stack is only sgrowsiz large because it is grown 1062 * automatically on a page fault. 1063 */ 1064 int 1065 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1066 { 1067 int error; 1068 struct proc *p = imgp->proc; 1069 struct vmspace *vmspace = p->p_vmspace; 1070 vm_object_t obj; 1071 struct rlimit rlim_stack; 1072 vm_offset_t sv_minuser, stack_addr; 1073 vm_map_t map; 1074 u_long ssiz; 1075 1076 imgp->vmspace_destroyed = 1; 1077 imgp->sysent = sv; 1078 1079 /* May be called with Giant held */ 1080 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1081 1082 /* 1083 * Blow away entire process VM, if address space not shared, 1084 * otherwise, create a new VM space so that other threads are 1085 * not disrupted 1086 */ 1087 map = &vmspace->vm_map; 1088 if (map_at_zero) 1089 sv_minuser = sv->sv_minuser; 1090 else 1091 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1092 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1093 vm_map_max(map) == sv->sv_maxuser && 1094 cpu_exec_vmspace_reuse(p, map)) { 1095 shmexit(vmspace); 1096 pmap_remove_pages(vmspace_pmap(vmspace)); 1097 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1098 /* 1099 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1100 * must be re-evaluated. 1101 */ 1102 vm_map_lock(map); 1103 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1104 MAP_ASLR_IGNSTART); 1105 vm_map_unlock(map); 1106 } else { 1107 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1108 if (error) 1109 return (error); 1110 vmspace = p->p_vmspace; 1111 map = &vmspace->vm_map; 1112 } 1113 map->flags |= imgp->map_flags; 1114 1115 /* Map a shared page */ 1116 obj = sv->sv_shared_page_obj; 1117 if (obj != NULL) { 1118 vm_object_reference(obj); 1119 error = vm_map_fixed(map, obj, 0, 1120 sv->sv_shared_page_base, sv->sv_shared_page_len, 1121 VM_PROT_READ | VM_PROT_EXECUTE, 1122 VM_PROT_READ | VM_PROT_EXECUTE, 1123 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1124 if (error != KERN_SUCCESS) { 1125 vm_object_deallocate(obj); 1126 return (vm_mmap_to_errno(error)); 1127 } 1128 } 1129 1130 /* Allocate a new stack */ 1131 if (imgp->stack_sz != 0) { 1132 ssiz = trunc_page(imgp->stack_sz); 1133 PROC_LOCK(p); 1134 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1135 PROC_UNLOCK(p); 1136 if (ssiz > rlim_stack.rlim_max) 1137 ssiz = rlim_stack.rlim_max; 1138 if (ssiz > rlim_stack.rlim_cur) { 1139 rlim_stack.rlim_cur = ssiz; 1140 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1141 } 1142 } else if (sv->sv_maxssiz != NULL) { 1143 ssiz = *sv->sv_maxssiz; 1144 } else { 1145 ssiz = maxssiz; 1146 } 1147 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1148 if (ssiz < imgp->eff_stack_sz) 1149 imgp->eff_stack_sz = ssiz; 1150 stack_addr = sv->sv_usrstack - ssiz; 1151 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1152 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1153 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1154 if (error != KERN_SUCCESS) 1155 return (vm_mmap_to_errno(error)); 1156 1157 /* 1158 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1159 * are still used to enforce the stack rlimit on the process stack. 1160 */ 1161 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1162 vmspace->vm_maxsaddr = (char *)stack_addr; 1163 1164 return (0); 1165 } 1166 1167 /* 1168 * Copy out argument and environment strings from the old process address 1169 * space into the temporary string buffer. 1170 */ 1171 int 1172 exec_copyin_args(struct image_args *args, char *fname, 1173 enum uio_seg segflg, char **argv, char **envv) 1174 { 1175 u_long argp, envp; 1176 int error; 1177 size_t length; 1178 1179 bzero(args, sizeof(*args)); 1180 if (argv == NULL) 1181 return (EFAULT); 1182 1183 /* 1184 * Allocate demand-paged memory for the file name, argument, and 1185 * environment strings. 1186 */ 1187 error = exec_alloc_args(args); 1188 if (error != 0) 1189 return (error); 1190 1191 /* 1192 * Copy the file name. 1193 */ 1194 if (fname != NULL) { 1195 args->fname = args->buf; 1196 error = (segflg == UIO_SYSSPACE) ? 1197 copystr(fname, args->fname, PATH_MAX, &length) : 1198 copyinstr(fname, args->fname, PATH_MAX, &length); 1199 if (error != 0) 1200 goto err_exit; 1201 } else 1202 length = 0; 1203 1204 args->begin_argv = args->buf + length; 1205 args->endp = args->begin_argv; 1206 args->stringspace = ARG_MAX; 1207 1208 /* 1209 * extract arguments first 1210 */ 1211 for (;;) { 1212 error = fueword(argv++, &argp); 1213 if (error == -1) { 1214 error = EFAULT; 1215 goto err_exit; 1216 } 1217 if (argp == 0) 1218 break; 1219 error = copyinstr((void *)(uintptr_t)argp, args->endp, 1220 args->stringspace, &length); 1221 if (error != 0) { 1222 if (error == ENAMETOOLONG) 1223 error = E2BIG; 1224 goto err_exit; 1225 } 1226 args->stringspace -= length; 1227 args->endp += length; 1228 args->argc++; 1229 } 1230 1231 args->begin_envv = args->endp; 1232 1233 /* 1234 * extract environment strings 1235 */ 1236 if (envv) { 1237 for (;;) { 1238 error = fueword(envv++, &envp); 1239 if (error == -1) { 1240 error = EFAULT; 1241 goto err_exit; 1242 } 1243 if (envp == 0) 1244 break; 1245 error = copyinstr((void *)(uintptr_t)envp, 1246 args->endp, args->stringspace, &length); 1247 if (error != 0) { 1248 if (error == ENAMETOOLONG) 1249 error = E2BIG; 1250 goto err_exit; 1251 } 1252 args->stringspace -= length; 1253 args->endp += length; 1254 args->envc++; 1255 } 1256 } 1257 1258 return (0); 1259 1260 err_exit: 1261 exec_free_args(args); 1262 return (error); 1263 } 1264 1265 int 1266 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1267 const void *data, size_t datalen, const int *fds, size_t fdslen) 1268 { 1269 struct filedesc *ofdp; 1270 const char *p; 1271 int *kfds; 1272 int error; 1273 1274 memset(args, '\0', sizeof(*args)); 1275 ofdp = td->td_proc->p_fd; 1276 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1277 return (E2BIG); 1278 error = exec_alloc_args(args); 1279 if (error != 0) 1280 return (error); 1281 1282 args->begin_argv = args->buf; 1283 args->stringspace = ARG_MAX; 1284 1285 if (datalen > 0) { 1286 /* 1287 * Argument buffer has been provided. Copy it into the 1288 * kernel as a single string and add a terminating null 1289 * byte. 1290 */ 1291 error = copyin(data, args->begin_argv, datalen); 1292 if (error != 0) 1293 goto err_exit; 1294 args->begin_argv[datalen] = '\0'; 1295 args->endp = args->begin_argv + datalen + 1; 1296 args->stringspace -= datalen + 1; 1297 1298 /* 1299 * Traditional argument counting. Count the number of 1300 * null bytes. 1301 */ 1302 for (p = args->begin_argv; p < args->endp; ++p) 1303 if (*p == '\0') 1304 ++args->argc; 1305 } else { 1306 /* No argument buffer provided. */ 1307 args->endp = args->begin_argv; 1308 } 1309 /* There are no environment variables. */ 1310 args->begin_envv = args->endp; 1311 1312 /* Create new file descriptor table. */ 1313 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1314 error = copyin(fds, kfds, fdslen * sizeof(int)); 1315 if (error != 0) { 1316 free(kfds, M_TEMP); 1317 goto err_exit; 1318 } 1319 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1320 free(kfds, M_TEMP); 1321 if (error != 0) 1322 goto err_exit; 1323 1324 return (0); 1325 err_exit: 1326 exec_free_args(args); 1327 return (error); 1328 } 1329 1330 struct exec_args_kva { 1331 vm_offset_t addr; 1332 u_int gen; 1333 SLIST_ENTRY(exec_args_kva) next; 1334 }; 1335 1336 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1337 1338 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1339 static struct mtx exec_args_kva_mtx; 1340 static u_int exec_args_gen; 1341 1342 static void 1343 exec_prealloc_args_kva(void *arg __unused) 1344 { 1345 struct exec_args_kva *argkva; 1346 u_int i; 1347 1348 SLIST_INIT(&exec_args_kva_freelist); 1349 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1350 for (i = 0; i < exec_map_entries; i++) { 1351 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1352 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1353 argkva->gen = exec_args_gen; 1354 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1355 } 1356 } 1357 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1358 1359 static vm_offset_t 1360 exec_alloc_args_kva(void **cookie) 1361 { 1362 struct exec_args_kva *argkva; 1363 1364 argkva = (void *)atomic_readandclear_ptr( 1365 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1366 if (argkva == NULL) { 1367 mtx_lock(&exec_args_kva_mtx); 1368 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1369 (void)mtx_sleep(&exec_args_kva_freelist, 1370 &exec_args_kva_mtx, 0, "execkva", 0); 1371 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1372 mtx_unlock(&exec_args_kva_mtx); 1373 } 1374 *(struct exec_args_kva **)cookie = argkva; 1375 return (argkva->addr); 1376 } 1377 1378 static void 1379 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1380 { 1381 vm_offset_t base; 1382 1383 base = argkva->addr; 1384 if (argkva->gen != gen) { 1385 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1386 MADV_FREE); 1387 argkva->gen = gen; 1388 } 1389 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1390 (uintptr_t)NULL, (uintptr_t)argkva)) { 1391 mtx_lock(&exec_args_kva_mtx); 1392 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1393 wakeup_one(&exec_args_kva_freelist); 1394 mtx_unlock(&exec_args_kva_mtx); 1395 } 1396 } 1397 1398 static void 1399 exec_free_args_kva(void *cookie) 1400 { 1401 1402 exec_release_args_kva(cookie, exec_args_gen); 1403 } 1404 1405 static void 1406 exec_args_kva_lowmem(void *arg __unused) 1407 { 1408 SLIST_HEAD(, exec_args_kva) head; 1409 struct exec_args_kva *argkva; 1410 u_int gen; 1411 int i; 1412 1413 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1414 1415 /* 1416 * Force an madvise of each KVA range. Any currently allocated ranges 1417 * will have MADV_FREE applied once they are freed. 1418 */ 1419 SLIST_INIT(&head); 1420 mtx_lock(&exec_args_kva_mtx); 1421 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1422 mtx_unlock(&exec_args_kva_mtx); 1423 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1424 SLIST_REMOVE_HEAD(&head, next); 1425 exec_release_args_kva(argkva, gen); 1426 } 1427 1428 CPU_FOREACH(i) { 1429 argkva = (void *)atomic_readandclear_ptr( 1430 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1431 if (argkva != NULL) 1432 exec_release_args_kva(argkva, gen); 1433 } 1434 } 1435 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1436 EVENTHANDLER_PRI_ANY); 1437 1438 /* 1439 * Allocate temporary demand-paged, zero-filled memory for the file name, 1440 * argument, and environment strings. 1441 */ 1442 int 1443 exec_alloc_args(struct image_args *args) 1444 { 1445 1446 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1447 return (0); 1448 } 1449 1450 void 1451 exec_free_args(struct image_args *args) 1452 { 1453 1454 if (args->buf != NULL) { 1455 exec_free_args_kva(args->bufkva); 1456 args->buf = NULL; 1457 } 1458 if (args->fname_buf != NULL) { 1459 free(args->fname_buf, M_TEMP); 1460 args->fname_buf = NULL; 1461 } 1462 if (args->fdp != NULL) 1463 fdescfree_remapped(args->fdp); 1464 } 1465 1466 /* 1467 * Copy strings out to the new process address space, constructing new arg 1468 * and env vector tables. Return a pointer to the base so that it can be used 1469 * as the initial stack pointer. 1470 */ 1471 register_t * 1472 exec_copyout_strings(struct image_params *imgp) 1473 { 1474 int argc, envc; 1475 char **vectp; 1476 char *stringp; 1477 uintptr_t destp; 1478 register_t *stack_base; 1479 struct ps_strings *arginfo; 1480 struct proc *p; 1481 size_t execpath_len; 1482 int szsigcode, szps; 1483 char canary[sizeof(long) * 8]; 1484 1485 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1486 /* 1487 * Calculate string base and vector table pointers. 1488 * Also deal with signal trampoline code for this exec type. 1489 */ 1490 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1491 execpath_len = strlen(imgp->execpath) + 1; 1492 else 1493 execpath_len = 0; 1494 p = imgp->proc; 1495 szsigcode = 0; 1496 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1497 if (p->p_sysent->sv_sigcode_base == 0) { 1498 if (p->p_sysent->sv_szsigcode != NULL) 1499 szsigcode = *(p->p_sysent->sv_szsigcode); 1500 } 1501 destp = (uintptr_t)arginfo; 1502 1503 /* 1504 * install sigcode 1505 */ 1506 if (szsigcode != 0) { 1507 destp -= szsigcode; 1508 destp = rounddown2(destp, sizeof(void *)); 1509 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1510 } 1511 1512 /* 1513 * Copy the image path for the rtld. 1514 */ 1515 if (execpath_len != 0) { 1516 destp -= execpath_len; 1517 destp = rounddown2(destp, sizeof(void *)); 1518 imgp->execpathp = destp; 1519 copyout(imgp->execpath, (void *)destp, execpath_len); 1520 } 1521 1522 /* 1523 * Prepare the canary for SSP. 1524 */ 1525 arc4rand(canary, sizeof(canary), 0); 1526 destp -= sizeof(canary); 1527 imgp->canary = destp; 1528 copyout(canary, (void *)destp, sizeof(canary)); 1529 imgp->canarylen = sizeof(canary); 1530 1531 /* 1532 * Prepare the pagesizes array. 1533 */ 1534 destp -= szps; 1535 destp = rounddown2(destp, sizeof(void *)); 1536 imgp->pagesizes = destp; 1537 copyout(pagesizes, (void *)destp, szps); 1538 imgp->pagesizeslen = szps; 1539 1540 destp -= ARG_MAX - imgp->args->stringspace; 1541 destp = rounddown2(destp, sizeof(void *)); 1542 1543 vectp = (char **)destp; 1544 if (imgp->sysent->sv_stackgap != NULL) 1545 imgp->sysent->sv_stackgap(imgp, (u_long *)&vectp); 1546 1547 if (imgp->auxargs) { 1548 /* 1549 * Allocate room on the stack for the ELF auxargs 1550 * array. It has up to AT_COUNT entries. 1551 */ 1552 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 1553 sizeof(*vectp)); 1554 } 1555 1556 /* 1557 * Allocate room for the argv[] and env vectors including the 1558 * terminating NULL pointers. 1559 */ 1560 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1561 1562 /* 1563 * vectp also becomes our initial stack base 1564 */ 1565 stack_base = (register_t *)vectp; 1566 1567 stringp = imgp->args->begin_argv; 1568 argc = imgp->args->argc; 1569 envc = imgp->args->envc; 1570 1571 /* 1572 * Copy out strings - arguments and environment. 1573 */ 1574 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1575 1576 /* 1577 * Fill in "ps_strings" struct for ps, w, etc. 1578 */ 1579 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1580 suword32(&arginfo->ps_nargvstr, argc); 1581 1582 /* 1583 * Fill in argument portion of vector table. 1584 */ 1585 for (; argc > 0; --argc) { 1586 suword(vectp++, (long)(intptr_t)destp); 1587 while (*stringp++ != 0) 1588 destp++; 1589 destp++; 1590 } 1591 1592 /* a null vector table pointer separates the argp's from the envp's */ 1593 suword(vectp++, 0); 1594 1595 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1596 suword32(&arginfo->ps_nenvstr, envc); 1597 1598 /* 1599 * Fill in environment portion of vector table. 1600 */ 1601 for (; envc > 0; --envc) { 1602 suword(vectp++, (long)(intptr_t)destp); 1603 while (*stringp++ != 0) 1604 destp++; 1605 destp++; 1606 } 1607 1608 /* end of vector table is a null pointer */ 1609 suword(vectp, 0); 1610 1611 return (stack_base); 1612 } 1613 1614 /* 1615 * Check permissions of file to execute. 1616 * Called with imgp->vp locked. 1617 * Return 0 for success or error code on failure. 1618 */ 1619 int 1620 exec_check_permissions(struct image_params *imgp) 1621 { 1622 struct vnode *vp = imgp->vp; 1623 struct vattr *attr = imgp->attr; 1624 struct thread *td; 1625 int error; 1626 1627 td = curthread; 1628 1629 /* Get file attributes */ 1630 error = VOP_GETATTR(vp, attr, td->td_ucred); 1631 if (error) 1632 return (error); 1633 1634 #ifdef MAC 1635 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1636 if (error) 1637 return (error); 1638 #endif 1639 1640 /* 1641 * 1) Check if file execution is disabled for the filesystem that 1642 * this file resides on. 1643 * 2) Ensure that at least one execute bit is on. Otherwise, a 1644 * privileged user will always succeed, and we don't want this 1645 * to happen unless the file really is executable. 1646 * 3) Ensure that the file is a regular file. 1647 */ 1648 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1649 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1650 (attr->va_type != VREG)) 1651 return (EACCES); 1652 1653 /* 1654 * Zero length files can't be exec'd 1655 */ 1656 if (attr->va_size == 0) 1657 return (ENOEXEC); 1658 1659 /* 1660 * Check for execute permission to file based on current credentials. 1661 */ 1662 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1663 if (error) 1664 return (error); 1665 1666 /* 1667 * Check number of open-for-writes on the file and deny execution 1668 * if there are any. 1669 * 1670 * Add a text reference now so no one can write to the 1671 * executable while we're activating it. 1672 * 1673 * Remember if this was set before and unset it in case this is not 1674 * actually an executable image. 1675 */ 1676 error = VOP_SET_TEXT(vp); 1677 if (error != 0) 1678 return (error); 1679 imgp->textset = true; 1680 1681 /* 1682 * Call filesystem specific open routine (which does nothing in the 1683 * general case). 1684 */ 1685 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1686 if (error == 0) 1687 imgp->opened = 1; 1688 return (error); 1689 } 1690 1691 /* 1692 * Exec handler registration 1693 */ 1694 int 1695 exec_register(const struct execsw *execsw_arg) 1696 { 1697 const struct execsw **es, **xs, **newexecsw; 1698 u_int count = 2; /* New slot and trailing NULL */ 1699 1700 if (execsw) 1701 for (es = execsw; *es; es++) 1702 count++; 1703 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1704 xs = newexecsw; 1705 if (execsw) 1706 for (es = execsw; *es; es++) 1707 *xs++ = *es; 1708 *xs++ = execsw_arg; 1709 *xs = NULL; 1710 if (execsw) 1711 free(execsw, M_TEMP); 1712 execsw = newexecsw; 1713 return (0); 1714 } 1715 1716 int 1717 exec_unregister(const struct execsw *execsw_arg) 1718 { 1719 const struct execsw **es, **xs, **newexecsw; 1720 int count = 1; 1721 1722 if (execsw == NULL) 1723 panic("unregister with no handlers left?\n"); 1724 1725 for (es = execsw; *es; es++) { 1726 if (*es == execsw_arg) 1727 break; 1728 } 1729 if (*es == NULL) 1730 return (ENOENT); 1731 for (es = execsw; *es; es++) 1732 if (*es != execsw_arg) 1733 count++; 1734 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1735 xs = newexecsw; 1736 for (es = execsw; *es; es++) 1737 if (*es != execsw_arg) 1738 *xs++ = *es; 1739 *xs = NULL; 1740 if (execsw) 1741 free(execsw, M_TEMP); 1742 execsw = newexecsw; 1743 return (0); 1744 } 1745