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