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