1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD 3 * 4 * Copyright (c) 1993, David Greenman 5 * All rights reserved. 6 * 7 * Redistribution and use in source and binary forms, with or without 8 * modification, are permitted provided that the following conditions 9 * are met: 10 * 1. Redistributions of source code must retain the above copyright 11 * notice, this list of conditions and the following disclaimer. 12 * 2. Redistributions in binary form must reproduce the above copyright 13 * notice, this list of conditions and the following disclaimer in the 14 * documentation and/or other materials provided with the distribution. 15 * 16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND 17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE 18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE 19 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE 20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL 21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS 22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) 23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT 24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY 25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF 26 * SUCH DAMAGE. 27 */ 28 29 #include <sys/cdefs.h> 30 __FBSDID("$FreeBSD$"); 31 32 #include "opt_capsicum.h" 33 #include "opt_hwpmc_hooks.h" 34 #include "opt_ktrace.h" 35 #include "opt_vm.h" 36 37 #include <sys/param.h> 38 #include <sys/systm.h> 39 #include <sys/acct.h> 40 #include <sys/capsicum.h> 41 #include <sys/eventhandler.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/filedesc.h> 45 #include <sys/imgact.h> 46 #include <sys/imgact_elf.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mman.h> 51 #include <sys/mount.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/pioctl.h> 55 #include <sys/priv.h> 56 #include <sys/proc.h> 57 #include <sys/ptrace.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sched.h> 61 #include <sys/sdt.h> 62 #include <sys/sf_buf.h> 63 #include <sys/shm.h> 64 #include <sys/signalvar.h> 65 #include <sys/smp.h> 66 #include <sys/stat.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/sysproto.h> 71 #include <sys/vnode.h> 72 #include <sys/wait.h> 73 #ifdef KTRACE 74 #include <sys/ktrace.h> 75 #endif 76 77 #include <vm/vm.h> 78 #include <vm/vm_param.h> 79 #include <vm/pmap.h> 80 #include <vm/vm_page.h> 81 #include <vm/vm_map.h> 82 #include <vm/vm_kern.h> 83 #include <vm/vm_extern.h> 84 #include <vm/vm_object.h> 85 #include <vm/vm_pager.h> 86 87 #ifdef HWPMC_HOOKS 88 #include <sys/pmckern.h> 89 #endif 90 91 #include <machine/reg.h> 92 93 #include <security/audit/audit.h> 94 #include <security/mac/mac_framework.h> 95 96 #ifdef KDTRACE_HOOKS 97 #include <sys/dtrace_bsd.h> 98 dtrace_execexit_func_t dtrace_fasttrap_exec; 99 #endif 100 101 SDT_PROVIDER_DECLARE(proc); 102 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 105 106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 107 108 int coredump_pack_fileinfo = 1; 109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 110 &coredump_pack_fileinfo, 0, 111 "Enable file path packing in 'procstat -f' coredump notes"); 112 113 int coredump_pack_vmmapinfo = 1; 114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 115 &coredump_pack_vmmapinfo, 0, 116 "Enable file path packing in 'procstat -v' coredump notes"); 117 118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 121 static int do_execve(struct thread *td, struct image_args *args, 122 struct mac *mac_p); 123 124 /* XXX This should be vm_size_t. */ 125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", ""); 127 128 /* XXX This should be vm_size_t. */ 129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", ""); 131 132 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 133 NULL, 0, sysctl_kern_stackprot, "I", ""); 134 135 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 136 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 137 &ps_arg_cache_limit, 0, ""); 138 139 static int disallow_high_osrel; 140 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 141 &disallow_high_osrel, 0, 142 "Disallow execution of binaries built for higher version of the world"); 143 144 static int map_at_zero = 0; 145 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 146 "Permit processes to map an object at virtual address 0."); 147 148 EVENTHANDLER_LIST_DECLARE(process_exec); 149 150 static int 151 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 152 { 153 struct proc *p; 154 int error; 155 156 p = curproc; 157 #ifdef SCTL_MASK32 158 if (req->flags & SCTL_MASK32) { 159 unsigned int val; 160 val = (unsigned int)p->p_sysent->sv_psstrings; 161 error = SYSCTL_OUT(req, &val, sizeof(val)); 162 } else 163 #endif 164 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 165 sizeof(p->p_sysent->sv_psstrings)); 166 return error; 167 } 168 169 static int 170 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 171 { 172 struct proc *p; 173 int error; 174 175 p = curproc; 176 #ifdef SCTL_MASK32 177 if (req->flags & SCTL_MASK32) { 178 unsigned int val; 179 val = (unsigned int)p->p_sysent->sv_usrstack; 180 error = SYSCTL_OUT(req, &val, sizeof(val)); 181 } else 182 #endif 183 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 184 sizeof(p->p_sysent->sv_usrstack)); 185 return error; 186 } 187 188 static int 189 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 190 { 191 struct proc *p; 192 193 p = curproc; 194 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 195 sizeof(p->p_sysent->sv_stackprot))); 196 } 197 198 /* 199 * Each of the items is a pointer to a `const struct execsw', hence the 200 * double pointer here. 201 */ 202 static const struct execsw **execsw; 203 204 #ifndef _SYS_SYSPROTO_H_ 205 struct execve_args { 206 char *fname; 207 char **argv; 208 char **envv; 209 }; 210 #endif 211 212 int 213 sys_execve(struct thread *td, struct execve_args *uap) 214 { 215 struct image_args args; 216 struct vmspace *oldvmspace; 217 int error; 218 219 error = pre_execve(td, &oldvmspace); 220 if (error != 0) 221 return (error); 222 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 223 uap->argv, uap->envv); 224 if (error == 0) 225 error = kern_execve(td, &args, NULL); 226 post_execve(td, error, oldvmspace); 227 return (error); 228 } 229 230 #ifndef _SYS_SYSPROTO_H_ 231 struct fexecve_args { 232 int fd; 233 char **argv; 234 char **envv; 235 } 236 #endif 237 int 238 sys_fexecve(struct thread *td, struct fexecve_args *uap) 239 { 240 struct image_args args; 241 struct vmspace *oldvmspace; 242 int error; 243 244 error = pre_execve(td, &oldvmspace); 245 if (error != 0) 246 return (error); 247 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 248 uap->argv, uap->envv); 249 if (error == 0) { 250 args.fd = uap->fd; 251 error = kern_execve(td, &args, NULL); 252 } 253 post_execve(td, error, oldvmspace); 254 return (error); 255 } 256 257 #ifndef _SYS_SYSPROTO_H_ 258 struct __mac_execve_args { 259 char *fname; 260 char **argv; 261 char **envv; 262 struct mac *mac_p; 263 }; 264 #endif 265 266 int 267 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 268 { 269 #ifdef MAC 270 struct image_args args; 271 struct vmspace *oldvmspace; 272 int error; 273 274 error = pre_execve(td, &oldvmspace); 275 if (error != 0) 276 return (error); 277 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 278 uap->argv, uap->envv); 279 if (error == 0) 280 error = kern_execve(td, &args, uap->mac_p); 281 post_execve(td, error, oldvmspace); 282 return (error); 283 #else 284 return (ENOSYS); 285 #endif 286 } 287 288 int 289 pre_execve(struct thread *td, struct vmspace **oldvmspace) 290 { 291 struct proc *p; 292 int error; 293 294 KASSERT(td == curthread, ("non-current thread %p", td)); 295 error = 0; 296 p = td->td_proc; 297 if ((p->p_flag & P_HADTHREADS) != 0) { 298 PROC_LOCK(p); 299 if (thread_single(p, SINGLE_BOUNDARY) != 0) 300 error = ERESTART; 301 PROC_UNLOCK(p); 302 } 303 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 304 ("nested execve")); 305 *oldvmspace = p->p_vmspace; 306 return (error); 307 } 308 309 void 310 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 311 { 312 struct proc *p; 313 314 KASSERT(td == curthread, ("non-current thread %p", td)); 315 p = td->td_proc; 316 if ((p->p_flag & P_HADTHREADS) != 0) { 317 PROC_LOCK(p); 318 /* 319 * If success, we upgrade to SINGLE_EXIT state to 320 * force other threads to suicide. 321 */ 322 if (error == EJUSTRETURN) 323 thread_single(p, SINGLE_EXIT); 324 else 325 thread_single_end(p, SINGLE_BOUNDARY); 326 PROC_UNLOCK(p); 327 } 328 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 329 KASSERT(p->p_vmspace != oldvmspace, 330 ("oldvmspace still used")); 331 vmspace_free(oldvmspace); 332 td->td_pflags &= ~TDP_EXECVMSPC; 333 } 334 } 335 336 /* 337 * XXX: kern_execve has the astonishing property of not always returning to 338 * the caller. If sufficiently bad things happen during the call to 339 * do_execve(), it can end up calling exit1(); as a result, callers must 340 * avoid doing anything which they might need to undo (e.g., allocating 341 * memory). 342 */ 343 int 344 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 345 { 346 347 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 348 exec_args_get_begin_envv(args) - args->begin_argv); 349 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 350 args->endp - exec_args_get_begin_envv(args)); 351 return (do_execve(td, args, mac_p)); 352 } 353 354 /* 355 * In-kernel implementation of execve(). All arguments are assumed to be 356 * userspace pointers from the passed thread. 357 */ 358 static int 359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 360 { 361 struct proc *p = td->td_proc; 362 struct nameidata nd; 363 struct ucred *oldcred; 364 struct uidinfo *euip = NULL; 365 register_t *stack_base; 366 int error, i; 367 struct image_params image_params, *imgp; 368 struct vattr attr; 369 int (*img_first)(struct image_params *); 370 struct pargs *oldargs = NULL, *newargs = NULL; 371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 372 #ifdef KTRACE 373 struct vnode *tracevp = NULL; 374 struct ucred *tracecred = NULL; 375 #endif 376 struct vnode *oldtextvp = NULL, *newtextvp; 377 int credential_changing; 378 int textset; 379 #ifdef MAC 380 struct label *interpvplabel = NULL; 381 int will_transition; 382 #endif 383 #ifdef HWPMC_HOOKS 384 struct pmckern_procexec pe; 385 #endif 386 static const char fexecv_proc_title[] = "(fexecv)"; 387 388 imgp = &image_params; 389 390 /* 391 * Lock the process and set the P_INEXEC flag to indicate that 392 * it should be left alone until we're done here. This is 393 * necessary to avoid race conditions - e.g. in ptrace() - 394 * that might allow a local user to illicitly obtain elevated 395 * privileges. 396 */ 397 PROC_LOCK(p); 398 KASSERT((p->p_flag & P_INEXEC) == 0, 399 ("%s(): process already has P_INEXEC flag", __func__)); 400 p->p_flag |= P_INEXEC; 401 PROC_UNLOCK(p); 402 403 /* 404 * Initialize part of the common data 405 */ 406 bzero(imgp, sizeof(*imgp)); 407 imgp->proc = p; 408 imgp->attr = &attr; 409 imgp->args = args; 410 oldcred = p->p_ucred; 411 412 #ifdef MAC 413 error = mac_execve_enter(imgp, mac_p); 414 if (error) 415 goto exec_fail; 416 #endif 417 418 /* 419 * Translate the file name. namei() returns a vnode pointer 420 * in ni_vp among other things. 421 * 422 * XXXAUDIT: It would be desirable to also audit the name of the 423 * interpreter if this is an interpreted binary. 424 */ 425 if (args->fname != NULL) { 426 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | FOLLOW | SAVENAME 427 | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 428 } 429 430 SDT_PROBE1(proc, , , exec, args->fname); 431 432 interpret: 433 if (args->fname != NULL) { 434 #ifdef CAPABILITY_MODE 435 /* 436 * While capability mode can't reach this point via direct 437 * path arguments to execve(), we also don't allow 438 * interpreters to be used in capability mode (for now). 439 * Catch indirect lookups and return a permissions error. 440 */ 441 if (IN_CAPABILITY_MODE(td)) { 442 error = ECAPMODE; 443 goto exec_fail; 444 } 445 #endif 446 error = namei(&nd); 447 if (error) 448 goto exec_fail; 449 450 newtextvp = nd.ni_vp; 451 imgp->vp = newtextvp; 452 } else { 453 AUDIT_ARG_FD(args->fd); 454 /* 455 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 456 */ 457 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 458 if (error) 459 goto exec_fail; 460 vn_lock(newtextvp, LK_EXCLUSIVE | LK_RETRY); 461 AUDIT_ARG_VNODE1(newtextvp); 462 imgp->vp = newtextvp; 463 } 464 465 /* 466 * Check file permissions (also 'opens' file) 467 */ 468 error = exec_check_permissions(imgp); 469 if (error) 470 goto exec_fail_dealloc; 471 472 imgp->object = imgp->vp->v_object; 473 if (imgp->object != NULL) 474 vm_object_reference(imgp->object); 475 476 /* 477 * Set VV_TEXT now so no one can write to the executable while we're 478 * activating it. 479 * 480 * Remember if this was set before and unset it in case this is not 481 * actually an executable image. 482 */ 483 textset = VOP_IS_TEXT(imgp->vp); 484 VOP_SET_TEXT(imgp->vp); 485 486 error = exec_map_first_page(imgp); 487 if (error) 488 goto exec_fail_dealloc; 489 490 imgp->proc->p_osrel = 0; 491 imgp->proc->p_fctl0 = 0; 492 493 /* 494 * Implement image setuid/setgid. 495 * 496 * Determine new credentials before attempting image activators 497 * so that it can be used by process_exec handlers to determine 498 * credential/setid changes. 499 * 500 * Don't honor setuid/setgid if the filesystem prohibits it or if 501 * the process is being traced. 502 * 503 * We disable setuid/setgid/etc in capability mode on the basis 504 * that most setugid applications are not written with that 505 * environment in mind, and will therefore almost certainly operate 506 * incorrectly. In principle there's no reason that setugid 507 * applications might not be useful in capability mode, so we may want 508 * to reconsider this conservative design choice in the future. 509 * 510 * XXXMAC: For the time being, use NOSUID to also prohibit 511 * transitions on the file system. 512 */ 513 credential_changing = 0; 514 credential_changing |= (attr.va_mode & S_ISUID) && 515 oldcred->cr_uid != attr.va_uid; 516 credential_changing |= (attr.va_mode & S_ISGID) && 517 oldcred->cr_gid != attr.va_gid; 518 #ifdef MAC 519 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 520 interpvplabel, imgp); 521 credential_changing |= will_transition; 522 #endif 523 524 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 525 if (credential_changing) 526 imgp->proc->p_pdeathsig = 0; 527 528 if (credential_changing && 529 #ifdef CAPABILITY_MODE 530 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 531 #endif 532 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 533 (p->p_flag & P_TRACED) == 0) { 534 imgp->credential_setid = true; 535 VOP_UNLOCK(imgp->vp, 0); 536 imgp->newcred = crdup(oldcred); 537 if (attr.va_mode & S_ISUID) { 538 euip = uifind(attr.va_uid); 539 change_euid(imgp->newcred, euip); 540 } 541 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 542 if (attr.va_mode & S_ISGID) 543 change_egid(imgp->newcred, attr.va_gid); 544 /* 545 * Implement correct POSIX saved-id behavior. 546 * 547 * XXXMAC: Note that the current logic will save the 548 * uid and gid if a MAC domain transition occurs, even 549 * though maybe it shouldn't. 550 */ 551 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 552 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 553 } else { 554 /* 555 * Implement correct POSIX saved-id behavior. 556 * 557 * XXX: It's not clear that the existing behavior is 558 * POSIX-compliant. A number of sources indicate that the 559 * saved uid/gid should only be updated if the new ruid is 560 * not equal to the old ruid, or the new euid is not equal 561 * to the old euid and the new euid is not equal to the old 562 * ruid. The FreeBSD code always updates the saved uid/gid. 563 * Also, this code uses the new (replaced) euid and egid as 564 * the source, which may or may not be the right ones to use. 565 */ 566 if (oldcred->cr_svuid != oldcred->cr_uid || 567 oldcred->cr_svgid != oldcred->cr_gid) { 568 VOP_UNLOCK(imgp->vp, 0); 569 imgp->newcred = crdup(oldcred); 570 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 571 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 572 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 573 } 574 } 575 /* The new credentials are installed into the process later. */ 576 577 /* 578 * Do the best to calculate the full path to the image file. 579 */ 580 if (args->fname != NULL && args->fname[0] == '/') 581 imgp->execpath = args->fname; 582 else { 583 VOP_UNLOCK(imgp->vp, 0); 584 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 585 &imgp->freepath) != 0) 586 imgp->execpath = args->fname; 587 vn_lock(imgp->vp, LK_EXCLUSIVE | LK_RETRY); 588 } 589 590 /* 591 * If the current process has a special image activator it 592 * wants to try first, call it. For example, emulating shell 593 * scripts differently. 594 */ 595 error = -1; 596 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 597 error = img_first(imgp); 598 599 /* 600 * Loop through the list of image activators, calling each one. 601 * An activator returns -1 if there is no match, 0 on success, 602 * and an error otherwise. 603 */ 604 for (i = 0; error == -1 && execsw[i]; ++i) { 605 if (execsw[i]->ex_imgact == NULL || 606 execsw[i]->ex_imgact == img_first) { 607 continue; 608 } 609 error = (*execsw[i]->ex_imgact)(imgp); 610 } 611 612 if (error) { 613 if (error == -1) { 614 if (textset == 0) 615 VOP_UNSET_TEXT(imgp->vp); 616 error = ENOEXEC; 617 } 618 goto exec_fail_dealloc; 619 } 620 621 /* 622 * Special interpreter operation, cleanup and loop up to try to 623 * activate the interpreter. 624 */ 625 if (imgp->interpreted) { 626 exec_unmap_first_page(imgp); 627 /* 628 * VV_TEXT needs to be unset for scripts. There is a short 629 * period before we determine that something is a script where 630 * VV_TEXT will be set. The vnode lock is held over this 631 * entire period so nothing should illegitimately be blocked. 632 */ 633 VOP_UNSET_TEXT(imgp->vp); 634 /* free name buffer and old vnode */ 635 if (args->fname != NULL) 636 NDFREE(&nd, NDF_ONLY_PNBUF); 637 #ifdef MAC 638 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 639 #endif 640 if (imgp->opened) { 641 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 642 imgp->opened = 0; 643 } 644 vput(newtextvp); 645 vm_object_deallocate(imgp->object); 646 imgp->object = NULL; 647 imgp->credential_setid = false; 648 if (imgp->newcred != NULL) { 649 crfree(imgp->newcred); 650 imgp->newcred = NULL; 651 } 652 imgp->execpath = NULL; 653 free(imgp->freepath, M_TEMP); 654 imgp->freepath = NULL; 655 /* set new name to that of the interpreter */ 656 NDINIT(&nd, LOOKUP, LOCKLEAF | FOLLOW | SAVENAME, 657 UIO_SYSSPACE, imgp->interpreter_name, td); 658 args->fname = imgp->interpreter_name; 659 goto interpret; 660 } 661 662 /* 663 * NB: We unlock the vnode here because it is believed that none 664 * of the sv_copyout_strings/sv_fixup operations require the vnode. 665 */ 666 VOP_UNLOCK(imgp->vp, 0); 667 668 if (disallow_high_osrel && 669 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 670 error = ENOEXEC; 671 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 672 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 673 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 674 goto exec_fail_dealloc; 675 } 676 677 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 678 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 679 sys_cap_enter(td, NULL); 680 681 /* 682 * Copy out strings (args and env) and initialize stack base 683 */ 684 if (p->p_sysent->sv_copyout_strings) 685 stack_base = (*p->p_sysent->sv_copyout_strings)(imgp); 686 else 687 stack_base = exec_copyout_strings(imgp); 688 689 /* 690 * If custom stack fixup routine present for this process 691 * let it do the stack setup. 692 * Else stuff argument count as first item on stack 693 */ 694 if (p->p_sysent->sv_fixup != NULL) 695 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 696 else 697 error = suword(--stack_base, imgp->args->argc) == 0 ? 698 0 : EFAULT; 699 if (error != 0) { 700 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 701 goto exec_fail_dealloc; 702 } 703 704 if (args->fdp != NULL) { 705 /* Install a brand new file descriptor table. */ 706 fdinstall_remapped(td, args->fdp); 707 args->fdp = NULL; 708 } else { 709 /* 710 * Keep on using the existing file descriptor table. For 711 * security and other reasons, the file descriptor table 712 * cannot be shared after an exec. 713 */ 714 fdunshare(td); 715 /* close files on exec */ 716 fdcloseexec(td); 717 } 718 719 /* 720 * Malloc things before we need locks. 721 */ 722 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 723 /* Cache arguments if they fit inside our allowance */ 724 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 725 newargs = pargs_alloc(i); 726 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 727 } 728 729 /* 730 * For security and other reasons, signal handlers cannot 731 * be shared after an exec. The new process gets a copy of the old 732 * handlers. In execsigs(), the new process will have its signals 733 * reset. 734 */ 735 if (sigacts_shared(p->p_sigacts)) { 736 oldsigacts = p->p_sigacts; 737 newsigacts = sigacts_alloc(); 738 sigacts_copy(newsigacts, oldsigacts); 739 } 740 741 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 742 743 PROC_LOCK(p); 744 if (oldsigacts) 745 p->p_sigacts = newsigacts; 746 /* Stop profiling */ 747 stopprofclock(p); 748 749 /* reset caught signals */ 750 execsigs(p); 751 752 /* name this process - nameiexec(p, ndp) */ 753 bzero(p->p_comm, sizeof(p->p_comm)); 754 if (args->fname) 755 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 756 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 757 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 758 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 759 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 760 #ifdef KTR 761 sched_clear_tdname(td); 762 #endif 763 764 /* 765 * mark as execed, wakeup the process that vforked (if any) and tell 766 * it that it now has its own resources back 767 */ 768 p->p_flag |= P_EXEC; 769 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 770 p->p_flag2 &= ~P2_NOTRACE; 771 if (p->p_flag & P_PPWAIT) { 772 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 773 cv_broadcast(&p->p_pwait); 774 /* STOPs are no longer ignored, arrange for AST */ 775 signotify(td); 776 } 777 778 /* 779 * Implement image setuid/setgid installation. 780 */ 781 if (imgp->credential_setid) { 782 /* 783 * Turn off syscall tracing for set-id programs, except for 784 * root. Record any set-id flags first to make sure that 785 * we do not regain any tracing during a possible block. 786 */ 787 setsugid(p); 788 789 #ifdef KTRACE 790 if (p->p_tracecred != NULL && 791 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED)) 792 ktrprocexec(p, &tracecred, &tracevp); 793 #endif 794 /* 795 * Close any file descriptors 0..2 that reference procfs, 796 * then make sure file descriptors 0..2 are in use. 797 * 798 * Both fdsetugidsafety() and fdcheckstd() may call functions 799 * taking sleepable locks, so temporarily drop our locks. 800 */ 801 PROC_UNLOCK(p); 802 VOP_UNLOCK(imgp->vp, 0); 803 fdsetugidsafety(td); 804 error = fdcheckstd(td); 805 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 806 if (error != 0) 807 goto exec_fail_dealloc; 808 PROC_LOCK(p); 809 #ifdef MAC 810 if (will_transition) { 811 mac_vnode_execve_transition(oldcred, imgp->newcred, 812 imgp->vp, interpvplabel, imgp); 813 } 814 #endif 815 } else { 816 if (oldcred->cr_uid == oldcred->cr_ruid && 817 oldcred->cr_gid == oldcred->cr_rgid) 818 p->p_flag &= ~P_SUGID; 819 } 820 /* 821 * Set the new credentials. 822 */ 823 if (imgp->newcred != NULL) { 824 proc_set_cred(p, imgp->newcred); 825 crfree(oldcred); 826 oldcred = NULL; 827 } 828 829 /* 830 * Store the vp for use in procfs. This vnode was referenced by namei 831 * or fgetvp_exec. 832 */ 833 oldtextvp = p->p_textvp; 834 p->p_textvp = newtextvp; 835 836 #ifdef KDTRACE_HOOKS 837 /* 838 * Tell the DTrace fasttrap provider about the exec if it 839 * has declared an interest. 840 */ 841 if (dtrace_fasttrap_exec) 842 dtrace_fasttrap_exec(p); 843 #endif 844 845 /* 846 * Notify others that we exec'd, and clear the P_INEXEC flag 847 * as we're now a bona fide freshly-execed process. 848 */ 849 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 850 p->p_flag &= ~P_INEXEC; 851 852 /* clear "fork but no exec" flag, as we _are_ execing */ 853 p->p_acflag &= ~AFORK; 854 855 /* 856 * Free any previous argument cache and replace it with 857 * the new argument cache, if any. 858 */ 859 oldargs = p->p_args; 860 p->p_args = newargs; 861 newargs = NULL; 862 863 PROC_UNLOCK(p); 864 865 #ifdef HWPMC_HOOKS 866 /* 867 * Check if system-wide sampling is in effect or if the 868 * current process is using PMCs. If so, do exec() time 869 * processing. This processing needs to happen AFTER the 870 * P_INEXEC flag is cleared. 871 */ 872 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 873 VOP_UNLOCK(imgp->vp, 0); 874 pe.pm_credentialschanged = credential_changing; 875 pe.pm_entryaddr = imgp->entry_addr; 876 877 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 878 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 879 } 880 #endif 881 882 /* Set values passed into the program in registers. */ 883 if (p->p_sysent->sv_setregs) 884 (*p->p_sysent->sv_setregs)(td, imgp, 885 (u_long)(uintptr_t)stack_base); 886 else 887 exec_setregs(td, imgp, (u_long)(uintptr_t)stack_base); 888 889 vfs_mark_atime(imgp->vp, td->td_ucred); 890 891 SDT_PROBE1(proc, , , exec__success, args->fname); 892 893 exec_fail_dealloc: 894 if (imgp->firstpage != NULL) 895 exec_unmap_first_page(imgp); 896 897 if (imgp->vp != NULL) { 898 if (args->fname) 899 NDFREE(&nd, NDF_ONLY_PNBUF); 900 if (imgp->opened) 901 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 902 if (error != 0) 903 vput(imgp->vp); 904 else 905 VOP_UNLOCK(imgp->vp, 0); 906 } 907 908 if (imgp->object != NULL) 909 vm_object_deallocate(imgp->object); 910 911 free(imgp->freepath, M_TEMP); 912 913 if (error == 0) { 914 if (p->p_ptevents & PTRACE_EXEC) { 915 PROC_LOCK(p); 916 if (p->p_ptevents & PTRACE_EXEC) 917 td->td_dbgflags |= TDB_EXEC; 918 PROC_UNLOCK(p); 919 } 920 921 /* 922 * Stop the process here if its stop event mask has 923 * the S_EXEC bit set. 924 */ 925 STOPEVENT(p, S_EXEC, 0); 926 } else { 927 exec_fail: 928 /* we're done here, clear P_INEXEC */ 929 PROC_LOCK(p); 930 p->p_flag &= ~P_INEXEC; 931 PROC_UNLOCK(p); 932 933 SDT_PROBE1(proc, , , exec__failure, error); 934 } 935 936 if (imgp->newcred != NULL && oldcred != NULL) 937 crfree(imgp->newcred); 938 939 #ifdef MAC 940 mac_execve_exit(imgp); 941 mac_execve_interpreter_exit(interpvplabel); 942 #endif 943 exec_free_args(args); 944 945 /* 946 * Handle deferred decrement of ref counts. 947 */ 948 if (oldtextvp != NULL) 949 vrele(oldtextvp); 950 #ifdef KTRACE 951 if (tracevp != NULL) 952 vrele(tracevp); 953 if (tracecred != NULL) 954 crfree(tracecred); 955 #endif 956 pargs_drop(oldargs); 957 pargs_drop(newargs); 958 if (oldsigacts != NULL) 959 sigacts_free(oldsigacts); 960 if (euip != NULL) 961 uifree(euip); 962 963 if (error && imgp->vmspace_destroyed) { 964 /* sorry, no more process anymore. exit gracefully */ 965 exit1(td, 0, SIGABRT); 966 /* NOT REACHED */ 967 } 968 969 #ifdef KTRACE 970 if (error == 0) 971 ktrprocctor(p); 972 #endif 973 974 /* 975 * We don't want cpu_set_syscall_retval() to overwrite any of 976 * the register values put in place by exec_setregs(). 977 * Implementations of cpu_set_syscall_retval() will leave 978 * registers unmodified when returning EJUSTRETURN. 979 */ 980 return (error == 0 ? EJUSTRETURN : error); 981 } 982 983 int 984 exec_map_first_page(struct image_params *imgp) 985 { 986 int rv, i, after, initial_pagein; 987 vm_page_t ma[VM_INITIAL_PAGEIN]; 988 vm_object_t object; 989 990 if (imgp->firstpage != NULL) 991 exec_unmap_first_page(imgp); 992 993 object = imgp->vp->v_object; 994 if (object == NULL) 995 return (EACCES); 996 VM_OBJECT_WLOCK(object); 997 #if VM_NRESERVLEVEL > 0 998 vm_object_color(object, 0); 999 #endif 1000 ma[0] = vm_page_grab(object, 0, VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY); 1001 if (ma[0]->valid != VM_PAGE_BITS_ALL) { 1002 vm_page_xbusy(ma[0]); 1003 if (!vm_pager_has_page(object, 0, NULL, &after)) { 1004 vm_page_lock(ma[0]); 1005 vm_page_free(ma[0]); 1006 vm_page_unlock(ma[0]); 1007 VM_OBJECT_WUNLOCK(object); 1008 return (EIO); 1009 } 1010 initial_pagein = min(after, VM_INITIAL_PAGEIN); 1011 KASSERT(initial_pagein <= object->size, 1012 ("%s: initial_pagein %d object->size %ju", 1013 __func__, initial_pagein, (uintmax_t )object->size)); 1014 for (i = 1; i < initial_pagein; i++) { 1015 if ((ma[i] = vm_page_next(ma[i - 1])) != NULL) { 1016 if (ma[i]->valid) 1017 break; 1018 if (!vm_page_tryxbusy(ma[i])) 1019 break; 1020 } else { 1021 ma[i] = vm_page_alloc(object, i, 1022 VM_ALLOC_NORMAL); 1023 if (ma[i] == NULL) 1024 break; 1025 } 1026 } 1027 initial_pagein = i; 1028 rv = vm_pager_get_pages(object, ma, initial_pagein, NULL, NULL); 1029 if (rv != VM_PAGER_OK) { 1030 for (i = 0; i < initial_pagein; i++) { 1031 vm_page_lock(ma[i]); 1032 vm_page_free(ma[i]); 1033 vm_page_unlock(ma[i]); 1034 } 1035 VM_OBJECT_WUNLOCK(object); 1036 return (EIO); 1037 } 1038 vm_page_xunbusy(ma[0]); 1039 for (i = 1; i < initial_pagein; i++) 1040 vm_page_readahead_finish(ma[i]); 1041 } 1042 vm_page_lock(ma[0]); 1043 vm_page_hold(ma[0]); 1044 vm_page_activate(ma[0]); 1045 vm_page_unlock(ma[0]); 1046 VM_OBJECT_WUNLOCK(object); 1047 1048 imgp->firstpage = sf_buf_alloc(ma[0], 0); 1049 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1050 1051 return (0); 1052 } 1053 1054 void 1055 exec_unmap_first_page(struct image_params *imgp) 1056 { 1057 vm_page_t m; 1058 1059 if (imgp->firstpage != NULL) { 1060 m = sf_buf_page(imgp->firstpage); 1061 sf_buf_free(imgp->firstpage); 1062 imgp->firstpage = NULL; 1063 vm_page_lock(m); 1064 vm_page_unhold(m); 1065 vm_page_unlock(m); 1066 } 1067 } 1068 1069 /* 1070 * Destroy old address space, and allocate a new stack. 1071 * The new stack is only sgrowsiz large because it is grown 1072 * automatically on a page fault. 1073 */ 1074 int 1075 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1076 { 1077 int error; 1078 struct proc *p = imgp->proc; 1079 struct vmspace *vmspace = p->p_vmspace; 1080 vm_object_t obj; 1081 struct rlimit rlim_stack; 1082 vm_offset_t sv_minuser, stack_addr; 1083 vm_map_t map; 1084 u_long ssiz; 1085 1086 imgp->vmspace_destroyed = 1; 1087 imgp->sysent = sv; 1088 1089 /* May be called with Giant held */ 1090 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1091 1092 /* 1093 * Blow away entire process VM, if address space not shared, 1094 * otherwise, create a new VM space so that other threads are 1095 * not disrupted 1096 */ 1097 map = &vmspace->vm_map; 1098 if (map_at_zero) 1099 sv_minuser = sv->sv_minuser; 1100 else 1101 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1102 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1103 vm_map_max(map) == sv->sv_maxuser && 1104 cpu_exec_vmspace_reuse(p, map)) { 1105 shmexit(vmspace); 1106 pmap_remove_pages(vmspace_pmap(vmspace)); 1107 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1108 /* 1109 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1110 * must be re-evaluated. 1111 */ 1112 vm_map_lock(map); 1113 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1114 MAP_ASLR_IGNSTART); 1115 vm_map_unlock(map); 1116 } else { 1117 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1118 if (error) 1119 return (error); 1120 vmspace = p->p_vmspace; 1121 map = &vmspace->vm_map; 1122 } 1123 map->flags |= imgp->map_flags; 1124 1125 /* Map a shared page */ 1126 obj = sv->sv_shared_page_obj; 1127 if (obj != NULL) { 1128 vm_object_reference(obj); 1129 error = vm_map_fixed(map, obj, 0, 1130 sv->sv_shared_page_base, sv->sv_shared_page_len, 1131 VM_PROT_READ | VM_PROT_EXECUTE, 1132 VM_PROT_READ | VM_PROT_EXECUTE, 1133 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1134 if (error != KERN_SUCCESS) { 1135 vm_object_deallocate(obj); 1136 return (vm_mmap_to_errno(error)); 1137 } 1138 } 1139 1140 /* Allocate a new stack */ 1141 if (imgp->stack_sz != 0) { 1142 ssiz = trunc_page(imgp->stack_sz); 1143 PROC_LOCK(p); 1144 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1145 PROC_UNLOCK(p); 1146 if (ssiz > rlim_stack.rlim_max) 1147 ssiz = rlim_stack.rlim_max; 1148 if (ssiz > rlim_stack.rlim_cur) { 1149 rlim_stack.rlim_cur = ssiz; 1150 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1151 } 1152 } else if (sv->sv_maxssiz != NULL) { 1153 ssiz = *sv->sv_maxssiz; 1154 } else { 1155 ssiz = maxssiz; 1156 } 1157 stack_addr = sv->sv_usrstack - ssiz; 1158 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1159 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1160 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1161 if (error != KERN_SUCCESS) 1162 return (vm_mmap_to_errno(error)); 1163 1164 /* 1165 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1166 * are still used to enforce the stack rlimit on the process stack. 1167 */ 1168 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1169 vmspace->vm_maxsaddr = (char *)stack_addr; 1170 1171 return (0); 1172 } 1173 1174 /* 1175 * Copy out argument and environment strings from the old process address 1176 * space into the temporary string buffer. 1177 */ 1178 int 1179 exec_copyin_args(struct image_args *args, const char *fname, 1180 enum uio_seg segflg, char **argv, char **envv) 1181 { 1182 u_long arg, env; 1183 int error; 1184 1185 bzero(args, sizeof(*args)); 1186 if (argv == NULL) 1187 return (EFAULT); 1188 1189 /* 1190 * Allocate demand-paged memory for the file name, argument, and 1191 * environment strings. 1192 */ 1193 error = exec_alloc_args(args); 1194 if (error != 0) 1195 return (error); 1196 1197 /* 1198 * Copy the file name. 1199 */ 1200 error = exec_args_add_fname(args, fname, segflg); 1201 if (error != 0) 1202 goto err_exit; 1203 1204 /* 1205 * extract arguments first 1206 */ 1207 for (;;) { 1208 error = fueword(argv++, &arg); 1209 if (error == -1) { 1210 error = EFAULT; 1211 goto err_exit; 1212 } 1213 if (arg == 0) 1214 break; 1215 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1216 UIO_USERSPACE); 1217 if (error != 0) 1218 goto err_exit; 1219 } 1220 1221 /* 1222 * extract environment strings 1223 */ 1224 if (envv) { 1225 for (;;) { 1226 error = fueword(envv++, &env); 1227 if (error == -1) { 1228 error = EFAULT; 1229 goto err_exit; 1230 } 1231 if (env == 0) 1232 break; 1233 error = exec_args_add_env(args, 1234 (char *)(uintptr_t)env, UIO_USERSPACE); 1235 if (error != 0) 1236 goto err_exit; 1237 } 1238 } 1239 1240 return (0); 1241 1242 err_exit: 1243 exec_free_args(args); 1244 return (error); 1245 } 1246 1247 int 1248 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1249 const void *data, size_t datalen, const int *fds, size_t fdslen) 1250 { 1251 struct filedesc *ofdp; 1252 const char *p; 1253 int *kfds; 1254 int error; 1255 1256 memset(args, '\0', sizeof(*args)); 1257 ofdp = td->td_proc->p_fd; 1258 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1259 return (E2BIG); 1260 error = exec_alloc_args(args); 1261 if (error != 0) 1262 return (error); 1263 1264 args->begin_argv = args->buf; 1265 args->stringspace = ARG_MAX; 1266 1267 if (datalen > 0) { 1268 /* 1269 * Argument buffer has been provided. Copy it into the 1270 * kernel as a single string and add a terminating null 1271 * byte. 1272 */ 1273 error = copyin(data, args->begin_argv, datalen); 1274 if (error != 0) 1275 goto err_exit; 1276 args->begin_argv[datalen] = '\0'; 1277 args->endp = args->begin_argv + datalen + 1; 1278 args->stringspace -= datalen + 1; 1279 1280 /* 1281 * Traditional argument counting. Count the number of 1282 * null bytes. 1283 */ 1284 for (p = args->begin_argv; p < args->endp; ++p) 1285 if (*p == '\0') 1286 ++args->argc; 1287 } else { 1288 /* No argument buffer provided. */ 1289 args->endp = args->begin_argv; 1290 } 1291 1292 /* Create new file descriptor table. */ 1293 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1294 error = copyin(fds, kfds, fdslen * sizeof(int)); 1295 if (error != 0) { 1296 free(kfds, M_TEMP); 1297 goto err_exit; 1298 } 1299 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1300 free(kfds, M_TEMP); 1301 if (error != 0) 1302 goto err_exit; 1303 1304 return (0); 1305 err_exit: 1306 exec_free_args(args); 1307 return (error); 1308 } 1309 1310 struct exec_args_kva { 1311 vm_offset_t addr; 1312 u_int gen; 1313 SLIST_ENTRY(exec_args_kva) next; 1314 }; 1315 1316 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1317 1318 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1319 static struct mtx exec_args_kva_mtx; 1320 static u_int exec_args_gen; 1321 1322 static void 1323 exec_prealloc_args_kva(void *arg __unused) 1324 { 1325 struct exec_args_kva *argkva; 1326 u_int i; 1327 1328 SLIST_INIT(&exec_args_kva_freelist); 1329 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1330 for (i = 0; i < exec_map_entries; i++) { 1331 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1332 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1333 argkva->gen = exec_args_gen; 1334 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1335 } 1336 } 1337 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1338 1339 static vm_offset_t 1340 exec_alloc_args_kva(void **cookie) 1341 { 1342 struct exec_args_kva *argkva; 1343 1344 argkva = (void *)atomic_readandclear_ptr( 1345 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1346 if (argkva == NULL) { 1347 mtx_lock(&exec_args_kva_mtx); 1348 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1349 (void)mtx_sleep(&exec_args_kva_freelist, 1350 &exec_args_kva_mtx, 0, "execkva", 0); 1351 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1352 mtx_unlock(&exec_args_kva_mtx); 1353 } 1354 *(struct exec_args_kva **)cookie = argkva; 1355 return (argkva->addr); 1356 } 1357 1358 static void 1359 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1360 { 1361 vm_offset_t base; 1362 1363 base = argkva->addr; 1364 if (argkva->gen != gen) { 1365 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1366 MADV_FREE); 1367 argkva->gen = gen; 1368 } 1369 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1370 (uintptr_t)NULL, (uintptr_t)argkva)) { 1371 mtx_lock(&exec_args_kva_mtx); 1372 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1373 wakeup_one(&exec_args_kva_freelist); 1374 mtx_unlock(&exec_args_kva_mtx); 1375 } 1376 } 1377 1378 static void 1379 exec_free_args_kva(void *cookie) 1380 { 1381 1382 exec_release_args_kva(cookie, exec_args_gen); 1383 } 1384 1385 static void 1386 exec_args_kva_lowmem(void *arg __unused) 1387 { 1388 SLIST_HEAD(, exec_args_kva) head; 1389 struct exec_args_kva *argkva; 1390 u_int gen; 1391 int i; 1392 1393 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1394 1395 /* 1396 * Force an madvise of each KVA range. Any currently allocated ranges 1397 * will have MADV_FREE applied once they are freed. 1398 */ 1399 SLIST_INIT(&head); 1400 mtx_lock(&exec_args_kva_mtx); 1401 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1402 mtx_unlock(&exec_args_kva_mtx); 1403 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1404 SLIST_REMOVE_HEAD(&head, next); 1405 exec_release_args_kva(argkva, gen); 1406 } 1407 1408 CPU_FOREACH(i) { 1409 argkva = (void *)atomic_readandclear_ptr( 1410 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1411 if (argkva != NULL) 1412 exec_release_args_kva(argkva, gen); 1413 } 1414 } 1415 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1416 EVENTHANDLER_PRI_ANY); 1417 1418 /* 1419 * Allocate temporary demand-paged, zero-filled memory for the file name, 1420 * argument, and environment strings. 1421 */ 1422 int 1423 exec_alloc_args(struct image_args *args) 1424 { 1425 1426 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1427 return (0); 1428 } 1429 1430 void 1431 exec_free_args(struct image_args *args) 1432 { 1433 1434 if (args->buf != NULL) { 1435 exec_free_args_kva(args->bufkva); 1436 args->buf = NULL; 1437 } 1438 if (args->fname_buf != NULL) { 1439 free(args->fname_buf, M_TEMP); 1440 args->fname_buf = NULL; 1441 } 1442 if (args->fdp != NULL) 1443 fdescfree_remapped(args->fdp); 1444 } 1445 1446 /* 1447 * A set to functions to fill struct image args. 1448 * 1449 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1450 * fname) before the other functions. All exec_args_add_arg() calls must 1451 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1452 * may be called any time after exec_args_add_fname(). 1453 * 1454 * exec_args_add_fname() - install path to be executed 1455 * exec_args_add_arg() - append an argument string 1456 * exec_args_add_env() - append an env string 1457 * exec_args_adjust_args() - adjust location of the argument list to 1458 * allow new arguments to be prepended 1459 */ 1460 int 1461 exec_args_add_fname(struct image_args *args, const char *fname, 1462 enum uio_seg segflg) 1463 { 1464 int error; 1465 size_t length; 1466 1467 KASSERT(args->fname == NULL, ("fname already appended")); 1468 KASSERT(args->endp == NULL, ("already appending to args")); 1469 1470 if (fname != NULL) { 1471 args->fname = args->buf; 1472 error = segflg == UIO_SYSSPACE ? 1473 copystr(fname, args->fname, PATH_MAX, &length) : 1474 copyinstr(fname, args->fname, PATH_MAX, &length); 1475 if (error != 0) 1476 return (error == ENAMETOOLONG ? E2BIG : error); 1477 } else 1478 length = 0; 1479 1480 /* Set up for _arg_*()/_env_*() */ 1481 args->endp = args->buf + length; 1482 /* begin_argv must be set and kept updated */ 1483 args->begin_argv = args->endp; 1484 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1485 ("too little space remaining for arguments %zu < %zu", 1486 exec_map_entry_size - length, (size_t)ARG_MAX)); 1487 args->stringspace = ARG_MAX; 1488 1489 return (0); 1490 } 1491 1492 static int 1493 exec_args_add_str(struct image_args *args, const char *str, 1494 enum uio_seg segflg, int *countp) 1495 { 1496 int error; 1497 size_t length; 1498 1499 KASSERT(args->endp != NULL, ("endp not initialized")); 1500 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1501 1502 error = (segflg == UIO_SYSSPACE) ? 1503 copystr(str, args->endp, args->stringspace, &length) : 1504 copyinstr(str, args->endp, args->stringspace, &length); 1505 if (error != 0) 1506 return (error == ENAMETOOLONG ? E2BIG : error); 1507 args->stringspace -= length; 1508 args->endp += length; 1509 (*countp)++; 1510 1511 return (0); 1512 } 1513 1514 int 1515 exec_args_add_arg(struct image_args *args, const char *argp, 1516 enum uio_seg segflg) 1517 { 1518 1519 KASSERT(args->envc == 0, ("appending args after env")); 1520 1521 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1522 } 1523 1524 int 1525 exec_args_add_env(struct image_args *args, const char *envp, 1526 enum uio_seg segflg) 1527 { 1528 1529 if (args->envc == 0) 1530 args->begin_envv = args->endp; 1531 1532 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1533 } 1534 1535 int 1536 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1537 { 1538 ssize_t offset; 1539 1540 KASSERT(args->endp != NULL, ("endp not initialized")); 1541 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1542 1543 offset = extend - consume; 1544 if (args->stringspace < offset) 1545 return (E2BIG); 1546 memmove(args->begin_argv + extend, args->begin_argv + consume, 1547 args->endp - args->begin_argv + consume); 1548 if (args->envc > 0) 1549 args->begin_envv += offset; 1550 args->endp += offset; 1551 args->stringspace -= offset; 1552 return (0); 1553 } 1554 1555 char * 1556 exec_args_get_begin_envv(struct image_args *args) 1557 { 1558 1559 KASSERT(args->endp != NULL, ("endp not initialized")); 1560 1561 if (args->envc > 0) 1562 return (args->begin_envv); 1563 return (args->endp); 1564 } 1565 1566 /* 1567 * Copy strings out to the new process address space, constructing new arg 1568 * and env vector tables. Return a pointer to the base so that it can be used 1569 * as the initial stack pointer. 1570 */ 1571 register_t * 1572 exec_copyout_strings(struct image_params *imgp) 1573 { 1574 int argc, envc; 1575 char **vectp; 1576 char *stringp; 1577 uintptr_t destp; 1578 register_t *stack_base; 1579 struct ps_strings *arginfo; 1580 struct proc *p; 1581 size_t execpath_len; 1582 int szsigcode, szps; 1583 char canary[sizeof(long) * 8]; 1584 1585 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1586 /* 1587 * Calculate string base and vector table pointers. 1588 * Also deal with signal trampoline code for this exec type. 1589 */ 1590 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1591 execpath_len = strlen(imgp->execpath) + 1; 1592 else 1593 execpath_len = 0; 1594 p = imgp->proc; 1595 szsigcode = 0; 1596 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1597 if (p->p_sysent->sv_sigcode_base == 0) { 1598 if (p->p_sysent->sv_szsigcode != NULL) 1599 szsigcode = *(p->p_sysent->sv_szsigcode); 1600 } 1601 destp = (uintptr_t)arginfo; 1602 1603 /* 1604 * install sigcode 1605 */ 1606 if (szsigcode != 0) { 1607 destp -= szsigcode; 1608 destp = rounddown2(destp, sizeof(void *)); 1609 copyout(p->p_sysent->sv_sigcode, (void *)destp, szsigcode); 1610 } 1611 1612 /* 1613 * Copy the image path for the rtld. 1614 */ 1615 if (execpath_len != 0) { 1616 destp -= execpath_len; 1617 destp = rounddown2(destp, sizeof(void *)); 1618 imgp->execpathp = destp; 1619 copyout(imgp->execpath, (void *)destp, execpath_len); 1620 } 1621 1622 /* 1623 * Prepare the canary for SSP. 1624 */ 1625 arc4rand(canary, sizeof(canary), 0); 1626 destp -= sizeof(canary); 1627 imgp->canary = destp; 1628 copyout(canary, (void *)destp, sizeof(canary)); 1629 imgp->canarylen = sizeof(canary); 1630 1631 /* 1632 * Prepare the pagesizes array. 1633 */ 1634 destp -= szps; 1635 destp = rounddown2(destp, sizeof(void *)); 1636 imgp->pagesizes = destp; 1637 copyout(pagesizes, (void *)destp, szps); 1638 imgp->pagesizeslen = szps; 1639 1640 destp -= ARG_MAX - imgp->args->stringspace; 1641 destp = rounddown2(destp, sizeof(void *)); 1642 1643 vectp = (char **)destp; 1644 if (imgp->auxargs) { 1645 /* 1646 * Allocate room on the stack for the ELF auxargs 1647 * array. It has up to AT_COUNT entries. 1648 */ 1649 vectp -= howmany(AT_COUNT * sizeof(Elf_Auxinfo), 1650 sizeof(*vectp)); 1651 } 1652 1653 /* 1654 * Allocate room for the argv[] and env vectors including the 1655 * terminating NULL pointers. 1656 */ 1657 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1658 1659 /* 1660 * vectp also becomes our initial stack base 1661 */ 1662 stack_base = (register_t *)vectp; 1663 1664 stringp = imgp->args->begin_argv; 1665 argc = imgp->args->argc; 1666 envc = imgp->args->envc; 1667 1668 /* 1669 * Copy out strings - arguments and environment. 1670 */ 1671 copyout(stringp, (void *)destp, ARG_MAX - imgp->args->stringspace); 1672 1673 /* 1674 * Fill in "ps_strings" struct for ps, w, etc. 1675 */ 1676 suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp); 1677 suword32(&arginfo->ps_nargvstr, argc); 1678 1679 /* 1680 * Fill in argument portion of vector table. 1681 */ 1682 for (; argc > 0; --argc) { 1683 suword(vectp++, (long)(intptr_t)destp); 1684 while (*stringp++ != 0) 1685 destp++; 1686 destp++; 1687 } 1688 1689 /* a null vector table pointer separates the argp's from the envp's */ 1690 suword(vectp++, 0); 1691 1692 suword(&arginfo->ps_envstr, (long)(intptr_t)vectp); 1693 suword32(&arginfo->ps_nenvstr, envc); 1694 1695 /* 1696 * Fill in environment portion of vector table. 1697 */ 1698 for (; envc > 0; --envc) { 1699 suword(vectp++, (long)(intptr_t)destp); 1700 while (*stringp++ != 0) 1701 destp++; 1702 destp++; 1703 } 1704 1705 /* end of vector table is a null pointer */ 1706 suword(vectp, 0); 1707 1708 return (stack_base); 1709 } 1710 1711 /* 1712 * Check permissions of file to execute. 1713 * Called with imgp->vp locked. 1714 * Return 0 for success or error code on failure. 1715 */ 1716 int 1717 exec_check_permissions(struct image_params *imgp) 1718 { 1719 struct vnode *vp = imgp->vp; 1720 struct vattr *attr = imgp->attr; 1721 struct thread *td; 1722 int error, writecount; 1723 1724 td = curthread; 1725 1726 /* Get file attributes */ 1727 error = VOP_GETATTR(vp, attr, td->td_ucred); 1728 if (error) 1729 return (error); 1730 1731 #ifdef MAC 1732 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1733 if (error) 1734 return (error); 1735 #endif 1736 1737 /* 1738 * 1) Check if file execution is disabled for the filesystem that 1739 * this file resides on. 1740 * 2) Ensure that at least one execute bit is on. Otherwise, a 1741 * privileged user will always succeed, and we don't want this 1742 * to happen unless the file really is executable. 1743 * 3) Ensure that the file is a regular file. 1744 */ 1745 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1746 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1747 (attr->va_type != VREG)) 1748 return (EACCES); 1749 1750 /* 1751 * Zero length files can't be exec'd 1752 */ 1753 if (attr->va_size == 0) 1754 return (ENOEXEC); 1755 1756 /* 1757 * Check for execute permission to file based on current credentials. 1758 */ 1759 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1760 if (error) 1761 return (error); 1762 1763 /* 1764 * Check number of open-for-writes on the file and deny execution 1765 * if there are any. 1766 */ 1767 error = VOP_GET_WRITECOUNT(vp, &writecount); 1768 if (error != 0) 1769 return (error); 1770 if (writecount != 0) 1771 return (ETXTBSY); 1772 1773 /* 1774 * Call filesystem specific open routine (which does nothing in the 1775 * general case). 1776 */ 1777 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1778 if (error == 0) 1779 imgp->opened = 1; 1780 return (error); 1781 } 1782 1783 /* 1784 * Exec handler registration 1785 */ 1786 int 1787 exec_register(const struct execsw *execsw_arg) 1788 { 1789 const struct execsw **es, **xs, **newexecsw; 1790 u_int count = 2; /* New slot and trailing NULL */ 1791 1792 if (execsw) 1793 for (es = execsw; *es; es++) 1794 count++; 1795 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1796 xs = newexecsw; 1797 if (execsw) 1798 for (es = execsw; *es; es++) 1799 *xs++ = *es; 1800 *xs++ = execsw_arg; 1801 *xs = NULL; 1802 if (execsw) 1803 free(execsw, M_TEMP); 1804 execsw = newexecsw; 1805 return (0); 1806 } 1807 1808 int 1809 exec_unregister(const struct execsw *execsw_arg) 1810 { 1811 const struct execsw **es, **xs, **newexecsw; 1812 int count = 1; 1813 1814 if (execsw == NULL) 1815 panic("unregister with no handlers left?\n"); 1816 1817 for (es = execsw; *es; es++) { 1818 if (*es == execsw_arg) 1819 break; 1820 } 1821 if (*es == NULL) 1822 return (ENOENT); 1823 for (es = execsw; *es; es++) 1824 if (*es != execsw_arg) 1825 count++; 1826 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1827 xs = newexecsw; 1828 for (es = execsw; *es; es++) 1829 if (*es != execsw_arg) 1830 *xs++ = *es; 1831 *xs = NULL; 1832 if (execsw) 1833 free(execsw, M_TEMP); 1834 execsw = newexecsw; 1835 return (0); 1836 } 1837