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 "Location of process' ps_strings structure"); 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 "Top of process stack"); 133 134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 135 NULL, 0, sysctl_kern_stackprot, "I", 136 "Stack memory permissions"); 137 138 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 139 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 140 &ps_arg_cache_limit, 0, 141 "Process' command line characters cache limit"); 142 143 static int disallow_high_osrel; 144 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 145 &disallow_high_osrel, 0, 146 "Disallow execution of binaries built for higher version of the world"); 147 148 static int map_at_zero = 0; 149 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 150 "Permit processes to map an object at virtual address 0."); 151 152 static int 153 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 154 { 155 struct proc *p; 156 int error; 157 158 p = curproc; 159 #ifdef SCTL_MASK32 160 if (req->flags & SCTL_MASK32) { 161 unsigned int val; 162 val = (unsigned int)p->p_sysent->sv_psstrings; 163 error = SYSCTL_OUT(req, &val, sizeof(val)); 164 } else 165 #endif 166 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 167 sizeof(p->p_sysent->sv_psstrings)); 168 return error; 169 } 170 171 static int 172 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 173 { 174 struct proc *p; 175 int error; 176 177 p = curproc; 178 #ifdef SCTL_MASK32 179 if (req->flags & SCTL_MASK32) { 180 unsigned int val; 181 val = (unsigned int)p->p_sysent->sv_usrstack; 182 error = SYSCTL_OUT(req, &val, sizeof(val)); 183 } else 184 #endif 185 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 186 sizeof(p->p_sysent->sv_usrstack)); 187 return error; 188 } 189 190 static int 191 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 192 { 193 struct proc *p; 194 195 p = curproc; 196 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 197 sizeof(p->p_sysent->sv_stackprot))); 198 } 199 200 /* 201 * Each of the items is a pointer to a `const struct execsw', hence the 202 * double pointer here. 203 */ 204 static const struct execsw **execsw; 205 206 #ifndef _SYS_SYSPROTO_H_ 207 struct execve_args { 208 char *fname; 209 char **argv; 210 char **envv; 211 }; 212 #endif 213 214 int 215 sys_execve(struct thread *td, struct execve_args *uap) 216 { 217 struct image_args args; 218 struct vmspace *oldvmspace; 219 int error; 220 221 error = pre_execve(td, &oldvmspace); 222 if (error != 0) 223 return (error); 224 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 225 uap->argv, uap->envv); 226 if (error == 0) 227 error = kern_execve(td, &args, NULL); 228 post_execve(td, error, oldvmspace); 229 return (error); 230 } 231 232 #ifndef _SYS_SYSPROTO_H_ 233 struct fexecve_args { 234 int fd; 235 char **argv; 236 char **envv; 237 } 238 #endif 239 int 240 sys_fexecve(struct thread *td, struct fexecve_args *uap) 241 { 242 struct image_args args; 243 struct vmspace *oldvmspace; 244 int error; 245 246 error = pre_execve(td, &oldvmspace); 247 if (error != 0) 248 return (error); 249 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 250 uap->argv, uap->envv); 251 if (error == 0) { 252 args.fd = uap->fd; 253 error = kern_execve(td, &args, NULL); 254 } 255 post_execve(td, error, oldvmspace); 256 return (error); 257 } 258 259 #ifndef _SYS_SYSPROTO_H_ 260 struct __mac_execve_args { 261 char *fname; 262 char **argv; 263 char **envv; 264 struct mac *mac_p; 265 }; 266 #endif 267 268 int 269 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 270 { 271 #ifdef MAC 272 struct image_args args; 273 struct vmspace *oldvmspace; 274 int error; 275 276 error = pre_execve(td, &oldvmspace); 277 if (error != 0) 278 return (error); 279 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 280 uap->argv, uap->envv); 281 if (error == 0) 282 error = kern_execve(td, &args, uap->mac_p); 283 post_execve(td, error, oldvmspace); 284 return (error); 285 #else 286 return (ENOSYS); 287 #endif 288 } 289 290 int 291 pre_execve(struct thread *td, struct vmspace **oldvmspace) 292 { 293 struct proc *p; 294 int error; 295 296 KASSERT(td == curthread, ("non-current thread %p", td)); 297 error = 0; 298 p = td->td_proc; 299 if ((p->p_flag & P_HADTHREADS) != 0) { 300 PROC_LOCK(p); 301 if (thread_single(p, SINGLE_BOUNDARY) != 0) 302 error = ERESTART; 303 PROC_UNLOCK(p); 304 } 305 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 306 ("nested execve")); 307 *oldvmspace = p->p_vmspace; 308 return (error); 309 } 310 311 void 312 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 313 { 314 struct proc *p; 315 316 KASSERT(td == curthread, ("non-current thread %p", td)); 317 p = td->td_proc; 318 if ((p->p_flag & P_HADTHREADS) != 0) { 319 PROC_LOCK(p); 320 /* 321 * If success, we upgrade to SINGLE_EXIT state to 322 * force other threads to suicide. 323 */ 324 if (error == EJUSTRETURN) 325 thread_single(p, SINGLE_EXIT); 326 else 327 thread_single_end(p, SINGLE_BOUNDARY); 328 PROC_UNLOCK(p); 329 } 330 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 331 KASSERT(p->p_vmspace != oldvmspace, 332 ("oldvmspace still used")); 333 vmspace_free(oldvmspace); 334 td->td_pflags &= ~TDP_EXECVMSPC; 335 } 336 } 337 338 /* 339 * XXX: kern_execve has the astonishing property of not always returning to 340 * the caller. If sufficiently bad things happen during the call to 341 * do_execve(), it can end up calling exit1(); as a result, callers must 342 * avoid doing anything which they might need to undo (e.g., allocating 343 * memory). 344 */ 345 int 346 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 347 { 348 349 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 350 exec_args_get_begin_envv(args) - args->begin_argv); 351 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 352 args->endp - exec_args_get_begin_envv(args)); 353 return (do_execve(td, args, mac_p)); 354 } 355 356 /* 357 * In-kernel implementation of execve(). All arguments are assumed to be 358 * userspace pointers from the passed thread. 359 */ 360 static int 361 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p) 362 { 363 struct proc *p = td->td_proc; 364 struct nameidata nd; 365 struct ucred *oldcred; 366 struct uidinfo *euip = NULL; 367 uintptr_t stack_base; 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 int credential_changing; 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 int error, i, orig_osrel; 387 uint32_t orig_fctl0; 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 orig_osrel = p->p_osrel; 414 orig_fctl0 = p->p_fctl0; 415 416 #ifdef MAC 417 error = mac_execve_enter(imgp, mac_p); 418 if (error) 419 goto exec_fail; 420 #endif 421 422 /* 423 * Translate the file name. namei() returns a vnode pointer 424 * in ni_vp among other things. 425 * 426 * XXXAUDIT: It would be desirable to also audit the name of the 427 * interpreter if this is an interpreted binary. 428 */ 429 if (args->fname != NULL) { 430 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 431 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 432 } 433 434 SDT_PROBE1(proc, , , exec, args->fname); 435 436 interpret: 437 if (args->fname != NULL) { 438 #ifdef CAPABILITY_MODE 439 /* 440 * While capability mode can't reach this point via direct 441 * path arguments to execve(), we also don't allow 442 * interpreters to be used in capability mode (for now). 443 * Catch indirect lookups and return a permissions error. 444 */ 445 if (IN_CAPABILITY_MODE(td)) { 446 error = ECAPMODE; 447 goto exec_fail; 448 } 449 #endif 450 error = namei(&nd); 451 if (error) 452 goto exec_fail; 453 454 newtextvp = nd.ni_vp; 455 imgp->vp = newtextvp; 456 } else { 457 AUDIT_ARG_FD(args->fd); 458 /* 459 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 460 */ 461 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 462 if (error) 463 goto exec_fail; 464 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 465 AUDIT_ARG_VNODE1(newtextvp); 466 imgp->vp = newtextvp; 467 } 468 469 /* 470 * Check file permissions. Also 'opens' file and sets its vnode to 471 * text mode. 472 */ 473 error = exec_check_permissions(imgp); 474 if (error) 475 goto exec_fail_dealloc; 476 477 imgp->object = imgp->vp->v_object; 478 if (imgp->object != NULL) 479 vm_object_reference(imgp->object); 480 481 error = exec_map_first_page(imgp); 482 if (error) 483 goto exec_fail_dealloc; 484 485 imgp->proc->p_osrel = 0; 486 imgp->proc->p_fctl0 = 0; 487 488 /* 489 * Implement image setuid/setgid. 490 * 491 * Determine new credentials before attempting image activators 492 * so that it can be used by process_exec handlers to determine 493 * credential/setid changes. 494 * 495 * Don't honor setuid/setgid if the filesystem prohibits it or if 496 * the process is being traced. 497 * 498 * We disable setuid/setgid/etc in capability mode on the basis 499 * that most setugid applications are not written with that 500 * environment in mind, and will therefore almost certainly operate 501 * incorrectly. In principle there's no reason that setugid 502 * applications might not be useful in capability mode, so we may want 503 * to reconsider this conservative design choice in the future. 504 * 505 * XXXMAC: For the time being, use NOSUID to also prohibit 506 * transitions on the file system. 507 */ 508 credential_changing = 0; 509 credential_changing |= (attr.va_mode & S_ISUID) && 510 oldcred->cr_uid != attr.va_uid; 511 credential_changing |= (attr.va_mode & S_ISGID) && 512 oldcred->cr_gid != attr.va_gid; 513 #ifdef MAC 514 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 515 interpvplabel, imgp); 516 credential_changing |= will_transition; 517 #endif 518 519 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 520 if (credential_changing) 521 imgp->proc->p_pdeathsig = 0; 522 523 if (credential_changing && 524 #ifdef CAPABILITY_MODE 525 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 526 #endif 527 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 528 (p->p_flag & P_TRACED) == 0) { 529 imgp->credential_setid = true; 530 VOP_UNLOCK(imgp->vp); 531 imgp->newcred = crdup(oldcred); 532 if (attr.va_mode & S_ISUID) { 533 euip = uifind(attr.va_uid); 534 change_euid(imgp->newcred, euip); 535 } 536 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 537 if (attr.va_mode & S_ISGID) 538 change_egid(imgp->newcred, attr.va_gid); 539 /* 540 * Implement correct POSIX saved-id behavior. 541 * 542 * XXXMAC: Note that the current logic will save the 543 * uid and gid if a MAC domain transition occurs, even 544 * though maybe it shouldn't. 545 */ 546 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 547 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 548 } else { 549 /* 550 * Implement correct POSIX saved-id behavior. 551 * 552 * XXX: It's not clear that the existing behavior is 553 * POSIX-compliant. A number of sources indicate that the 554 * saved uid/gid should only be updated if the new ruid is 555 * not equal to the old ruid, or the new euid is not equal 556 * to the old euid and the new euid is not equal to the old 557 * ruid. The FreeBSD code always updates the saved uid/gid. 558 * Also, this code uses the new (replaced) euid and egid as 559 * the source, which may or may not be the right ones to use. 560 */ 561 if (oldcred->cr_svuid != oldcred->cr_uid || 562 oldcred->cr_svgid != oldcred->cr_gid) { 563 VOP_UNLOCK(imgp->vp); 564 imgp->newcred = crdup(oldcred); 565 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 566 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 567 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 568 } 569 } 570 /* The new credentials are installed into the process later. */ 571 572 /* 573 * Do the best to calculate the full path to the image file. 574 */ 575 if (args->fname != NULL && args->fname[0] == '/') 576 imgp->execpath = args->fname; 577 else { 578 VOP_UNLOCK(imgp->vp); 579 if (vn_fullpath(td, imgp->vp, &imgp->execpath, 580 &imgp->freepath) != 0) 581 imgp->execpath = args->fname; 582 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 583 } 584 585 /* 586 * If the current process has a special image activator it 587 * wants to try first, call it. For example, emulating shell 588 * scripts differently. 589 */ 590 error = -1; 591 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 592 error = img_first(imgp); 593 594 /* 595 * Loop through the list of image activators, calling each one. 596 * An activator returns -1 if there is no match, 0 on success, 597 * and an error otherwise. 598 */ 599 for (i = 0; error == -1 && execsw[i]; ++i) { 600 if (execsw[i]->ex_imgact == NULL || 601 execsw[i]->ex_imgact == img_first) { 602 continue; 603 } 604 error = (*execsw[i]->ex_imgact)(imgp); 605 } 606 607 if (error) { 608 if (error == -1) 609 error = ENOEXEC; 610 goto exec_fail_dealloc; 611 } 612 613 /* 614 * Special interpreter operation, cleanup and loop up to try to 615 * activate the interpreter. 616 */ 617 if (imgp->interpreted) { 618 exec_unmap_first_page(imgp); 619 /* 620 * The text reference needs to be removed for scripts. 621 * There is a short period before we determine that 622 * something is a script where text reference is active. 623 * The vnode lock is held over this entire period 624 * so nothing should illegitimately be blocked. 625 */ 626 MPASS(imgp->textset); 627 VOP_UNSET_TEXT_CHECKED(newtextvp); 628 imgp->textset = false; 629 /* free name buffer and old vnode */ 630 if (args->fname != NULL) 631 NDFREE(&nd, NDF_ONLY_PNBUF); 632 #ifdef MAC 633 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 634 #endif 635 if (imgp->opened) { 636 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 637 imgp->opened = 0; 638 } 639 vput(newtextvp); 640 vm_object_deallocate(imgp->object); 641 imgp->object = NULL; 642 imgp->credential_setid = false; 643 if (imgp->newcred != NULL) { 644 crfree(imgp->newcred); 645 imgp->newcred = NULL; 646 } 647 imgp->execpath = NULL; 648 free(imgp->freepath, M_TEMP); 649 imgp->freepath = NULL; 650 /* set new name to that of the interpreter */ 651 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 652 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 653 args->fname = imgp->interpreter_name; 654 goto interpret; 655 } 656 657 /* 658 * NB: We unlock the vnode here because it is believed that none 659 * of the sv_copyout_strings/sv_fixup operations require the vnode. 660 */ 661 VOP_UNLOCK(imgp->vp); 662 663 if (disallow_high_osrel && 664 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 665 error = ENOEXEC; 666 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 667 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 668 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 669 goto exec_fail_dealloc; 670 } 671 672 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 673 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 674 sys_cap_enter(td, NULL); 675 676 /* 677 * Copy out strings (args and env) and initialize stack base. 678 */ 679 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 680 if (error != 0) { 681 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 682 goto exec_fail_dealloc; 683 } 684 685 /* 686 * Stack setup. 687 */ 688 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 689 if (error != 0) { 690 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 691 goto exec_fail_dealloc; 692 } 693 694 if (args->fdp != NULL) { 695 /* Install a brand new file descriptor table. */ 696 fdinstall_remapped(td, args->fdp); 697 args->fdp = NULL; 698 } else { 699 /* 700 * Keep on using the existing file descriptor table. For 701 * security and other reasons, the file descriptor table 702 * cannot be shared after an exec. 703 */ 704 fdunshare(td); 705 /* close files on exec */ 706 fdcloseexec(td); 707 } 708 709 /* 710 * Malloc things before we need locks. 711 */ 712 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 713 /* Cache arguments if they fit inside our allowance */ 714 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 715 newargs = pargs_alloc(i); 716 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 717 } 718 719 /* 720 * For security and other reasons, signal handlers cannot 721 * be shared after an exec. The new process gets a copy of the old 722 * handlers. In execsigs(), the new process will have its signals 723 * reset. 724 */ 725 if (sigacts_shared(p->p_sigacts)) { 726 oldsigacts = p->p_sigacts; 727 newsigacts = sigacts_alloc(); 728 sigacts_copy(newsigacts, oldsigacts); 729 } 730 731 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 732 733 PROC_LOCK(p); 734 if (oldsigacts) 735 p->p_sigacts = newsigacts; 736 /* Stop profiling */ 737 stopprofclock(p); 738 739 /* reset caught signals */ 740 execsigs(p); 741 742 /* name this process - nameiexec(p, ndp) */ 743 bzero(p->p_comm, sizeof(p->p_comm)); 744 if (args->fname) 745 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 746 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 747 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 748 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 749 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 750 #ifdef KTR 751 sched_clear_tdname(td); 752 #endif 753 754 /* 755 * mark as execed, wakeup the process that vforked (if any) and tell 756 * it that it now has its own resources back 757 */ 758 p->p_flag |= P_EXEC; 759 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 760 p->p_flag2 &= ~P2_NOTRACE; 761 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 762 p->p_flag2 &= ~P2_STKGAP_DISABLE; 763 if (p->p_flag & P_PPWAIT) { 764 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 765 cv_broadcast(&p->p_pwait); 766 /* STOPs are no longer ignored, arrange for AST */ 767 signotify(td); 768 } 769 770 /* 771 * Implement image setuid/setgid installation. 772 */ 773 if (imgp->credential_setid) { 774 /* 775 * Turn off syscall tracing for set-id programs, except for 776 * root. Record any set-id flags first to make sure that 777 * we do not regain any tracing during a possible block. 778 */ 779 setsugid(p); 780 781 #ifdef KTRACE 782 if (p->p_tracecred != NULL && 783 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED)) 784 ktrprocexec(p, &tracecred, &tracevp); 785 #endif 786 /* 787 * Close any file descriptors 0..2 that reference procfs, 788 * then make sure file descriptors 0..2 are in use. 789 * 790 * Both fdsetugidsafety() and fdcheckstd() may call functions 791 * taking sleepable locks, so temporarily drop our locks. 792 */ 793 PROC_UNLOCK(p); 794 VOP_UNLOCK(imgp->vp); 795 fdsetugidsafety(td); 796 error = fdcheckstd(td); 797 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 798 if (error != 0) 799 goto exec_fail_dealloc; 800 PROC_LOCK(p); 801 #ifdef MAC 802 if (will_transition) { 803 mac_vnode_execve_transition(oldcred, imgp->newcred, 804 imgp->vp, interpvplabel, imgp); 805 } 806 #endif 807 } else { 808 if (oldcred->cr_uid == oldcred->cr_ruid && 809 oldcred->cr_gid == oldcred->cr_rgid) 810 p->p_flag &= ~P_SUGID; 811 } 812 /* 813 * Set the new credentials. 814 */ 815 if (imgp->newcred != NULL) { 816 proc_set_cred(p, imgp->newcred); 817 crfree(oldcred); 818 oldcred = NULL; 819 } 820 821 /* 822 * Store the vp for use in procfs. This vnode was referenced by namei 823 * or fgetvp_exec. 824 */ 825 oldtextvp = p->p_textvp; 826 p->p_textvp = newtextvp; 827 828 #ifdef KDTRACE_HOOKS 829 /* 830 * Tell the DTrace fasttrap provider about the exec if it 831 * has declared an interest. 832 */ 833 if (dtrace_fasttrap_exec) 834 dtrace_fasttrap_exec(p); 835 #endif 836 837 /* 838 * Notify others that we exec'd, and clear the P_INEXEC flag 839 * as we're now a bona fide freshly-execed process. 840 */ 841 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 842 p->p_flag &= ~P_INEXEC; 843 844 /* clear "fork but no exec" flag, as we _are_ execing */ 845 p->p_acflag &= ~AFORK; 846 847 /* 848 * Free any previous argument cache and replace it with 849 * the new argument cache, if any. 850 */ 851 oldargs = p->p_args; 852 p->p_args = newargs; 853 newargs = NULL; 854 855 PROC_UNLOCK(p); 856 857 #ifdef HWPMC_HOOKS 858 /* 859 * Check if system-wide sampling is in effect or if the 860 * current process is using PMCs. If so, do exec() time 861 * processing. This processing needs to happen AFTER the 862 * P_INEXEC flag is cleared. 863 */ 864 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 865 VOP_UNLOCK(imgp->vp); 866 pe.pm_credentialschanged = credential_changing; 867 pe.pm_entryaddr = imgp->entry_addr; 868 869 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 870 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 871 } 872 #endif 873 874 /* Set values passed into the program in registers. */ 875 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 876 877 VOP_MMAPPED(imgp->vp); 878 879 SDT_PROBE1(proc, , , exec__success, args->fname); 880 881 exec_fail_dealloc: 882 if (error != 0) { 883 p->p_osrel = orig_osrel; 884 p->p_fctl0 = orig_fctl0; 885 } 886 887 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 (imgp->textset) 896 VOP_UNSET_TEXT_CHECKED(imgp->vp); 897 if (error != 0) 898 vput(imgp->vp); 899 else 900 VOP_UNLOCK(imgp->vp); 901 } 902 903 if (imgp->object != NULL) 904 vm_object_deallocate(imgp->object); 905 906 free(imgp->freepath, M_TEMP); 907 908 if (error == 0) { 909 if (p->p_ptevents & PTRACE_EXEC) { 910 PROC_LOCK(p); 911 if (p->p_ptevents & PTRACE_EXEC) 912 td->td_dbgflags |= TDB_EXEC; 913 PROC_UNLOCK(p); 914 } 915 916 /* 917 * Stop the process here if its stop event mask has 918 * the S_EXEC bit set. 919 */ 920 STOPEVENT(p, S_EXEC, 0); 921 } else { 922 exec_fail: 923 /* we're done here, clear P_INEXEC */ 924 PROC_LOCK(p); 925 p->p_flag &= ~P_INEXEC; 926 PROC_UNLOCK(p); 927 928 SDT_PROBE1(proc, , , exec__failure, error); 929 } 930 931 if (imgp->newcred != NULL && oldcred != NULL) 932 crfree(imgp->newcred); 933 934 #ifdef MAC 935 mac_execve_exit(imgp); 936 mac_execve_interpreter_exit(interpvplabel); 937 #endif 938 exec_free_args(args); 939 940 /* 941 * Handle deferred decrement of ref counts. 942 */ 943 if (oldtextvp != NULL) 944 vrele(oldtextvp); 945 #ifdef KTRACE 946 if (tracevp != NULL) 947 vrele(tracevp); 948 if (tracecred != NULL) 949 crfree(tracecred); 950 #endif 951 pargs_drop(oldargs); 952 pargs_drop(newargs); 953 if (oldsigacts != NULL) 954 sigacts_free(oldsigacts); 955 if (euip != NULL) 956 uifree(euip); 957 958 if (error && imgp->vmspace_destroyed) { 959 /* sorry, no more process anymore. exit gracefully */ 960 exit1(td, 0, SIGABRT); 961 /* NOT REACHED */ 962 } 963 964 #ifdef KTRACE 965 if (error == 0) 966 ktrprocctor(p); 967 #endif 968 969 /* 970 * We don't want cpu_set_syscall_retval() to overwrite any of 971 * the register values put in place by exec_setregs(). 972 * Implementations of cpu_set_syscall_retval() will leave 973 * registers unmodified when returning EJUSTRETURN. 974 */ 975 return (error == 0 ? EJUSTRETURN : error); 976 } 977 978 int 979 exec_map_first_page(struct image_params *imgp) 980 { 981 vm_object_t object; 982 vm_page_t m; 983 int error; 984 985 if (imgp->firstpage != NULL) 986 exec_unmap_first_page(imgp); 987 988 object = imgp->vp->v_object; 989 if (object == NULL) 990 return (EACCES); 991 #if VM_NRESERVLEVEL > 0 992 if ((object->flags & OBJ_COLORED) == 0) { 993 VM_OBJECT_WLOCK(object); 994 vm_object_color(object, 0); 995 VM_OBJECT_WUNLOCK(object); 996 } 997 #endif 998 error = vm_page_grab_valid_unlocked(&m, object, 0, 999 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1000 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1001 1002 if (error != VM_PAGER_OK) 1003 return (EIO); 1004 imgp->firstpage = sf_buf_alloc(m, 0); 1005 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1006 1007 return (0); 1008 } 1009 1010 void 1011 exec_unmap_first_page(struct image_params *imgp) 1012 { 1013 vm_page_t m; 1014 1015 if (imgp->firstpage != NULL) { 1016 m = sf_buf_page(imgp->firstpage); 1017 sf_buf_free(imgp->firstpage); 1018 imgp->firstpage = NULL; 1019 vm_page_unwire(m, PQ_ACTIVE); 1020 } 1021 } 1022 1023 /* 1024 * Destroy old address space, and allocate a new stack. 1025 * The new stack is only sgrowsiz large because it is grown 1026 * automatically on a page fault. 1027 */ 1028 int 1029 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1030 { 1031 int error; 1032 struct proc *p = imgp->proc; 1033 struct vmspace *vmspace = p->p_vmspace; 1034 struct thread *td = curthread; 1035 vm_object_t obj; 1036 struct rlimit rlim_stack; 1037 vm_offset_t sv_minuser, stack_addr; 1038 vm_map_t map; 1039 u_long ssiz; 1040 1041 imgp->vmspace_destroyed = 1; 1042 imgp->sysent = sv; 1043 1044 sigfastblock_clear(td); 1045 1046 /* May be called with Giant held */ 1047 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1048 1049 /* 1050 * Blow away entire process VM, if address space not shared, 1051 * otherwise, create a new VM space so that other threads are 1052 * not disrupted 1053 */ 1054 map = &vmspace->vm_map; 1055 if (map_at_zero) 1056 sv_minuser = sv->sv_minuser; 1057 else 1058 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1059 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1060 vm_map_max(map) == sv->sv_maxuser && 1061 cpu_exec_vmspace_reuse(p, map)) { 1062 shmexit(vmspace); 1063 pmap_remove_pages(vmspace_pmap(vmspace)); 1064 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1065 /* 1066 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1067 * must be re-evaluated. 1068 */ 1069 vm_map_lock(map); 1070 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1071 MAP_ASLR_IGNSTART); 1072 vm_map_unlock(map); 1073 } else { 1074 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1075 if (error) 1076 return (error); 1077 vmspace = p->p_vmspace; 1078 map = &vmspace->vm_map; 1079 } 1080 map->flags |= imgp->map_flags; 1081 1082 /* Map a shared page */ 1083 obj = sv->sv_shared_page_obj; 1084 if (obj != NULL) { 1085 vm_object_reference(obj); 1086 error = vm_map_fixed(map, obj, 0, 1087 sv->sv_shared_page_base, sv->sv_shared_page_len, 1088 VM_PROT_READ | VM_PROT_EXECUTE, 1089 VM_PROT_READ | VM_PROT_EXECUTE, 1090 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1091 if (error != KERN_SUCCESS) { 1092 vm_object_deallocate(obj); 1093 return (vm_mmap_to_errno(error)); 1094 } 1095 } 1096 1097 /* Allocate a new stack */ 1098 if (imgp->stack_sz != 0) { 1099 ssiz = trunc_page(imgp->stack_sz); 1100 PROC_LOCK(p); 1101 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1102 PROC_UNLOCK(p); 1103 if (ssiz > rlim_stack.rlim_max) 1104 ssiz = rlim_stack.rlim_max; 1105 if (ssiz > rlim_stack.rlim_cur) { 1106 rlim_stack.rlim_cur = ssiz; 1107 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1108 } 1109 } else if (sv->sv_maxssiz != NULL) { 1110 ssiz = *sv->sv_maxssiz; 1111 } else { 1112 ssiz = maxssiz; 1113 } 1114 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1115 if (ssiz < imgp->eff_stack_sz) 1116 imgp->eff_stack_sz = ssiz; 1117 stack_addr = sv->sv_usrstack - ssiz; 1118 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1119 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1120 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1121 if (error != KERN_SUCCESS) 1122 return (vm_mmap_to_errno(error)); 1123 1124 /* 1125 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1126 * are still used to enforce the stack rlimit on the process stack. 1127 */ 1128 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1129 vmspace->vm_maxsaddr = (char *)stack_addr; 1130 1131 return (0); 1132 } 1133 1134 /* 1135 * Copy out argument and environment strings from the old process address 1136 * space into the temporary string buffer. 1137 */ 1138 int 1139 exec_copyin_args(struct image_args *args, const char *fname, 1140 enum uio_seg segflg, char **argv, char **envv) 1141 { 1142 u_long arg, env; 1143 int error; 1144 1145 bzero(args, sizeof(*args)); 1146 if (argv == NULL) 1147 return (EFAULT); 1148 1149 /* 1150 * Allocate demand-paged memory for the file name, argument, and 1151 * environment strings. 1152 */ 1153 error = exec_alloc_args(args); 1154 if (error != 0) 1155 return (error); 1156 1157 /* 1158 * Copy the file name. 1159 */ 1160 error = exec_args_add_fname(args, fname, segflg); 1161 if (error != 0) 1162 goto err_exit; 1163 1164 /* 1165 * extract arguments first 1166 */ 1167 for (;;) { 1168 error = fueword(argv++, &arg); 1169 if (error == -1) { 1170 error = EFAULT; 1171 goto err_exit; 1172 } 1173 if (arg == 0) 1174 break; 1175 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1176 UIO_USERSPACE); 1177 if (error != 0) 1178 goto err_exit; 1179 } 1180 1181 /* 1182 * extract environment strings 1183 */ 1184 if (envv) { 1185 for (;;) { 1186 error = fueword(envv++, &env); 1187 if (error == -1) { 1188 error = EFAULT; 1189 goto err_exit; 1190 } 1191 if (env == 0) 1192 break; 1193 error = exec_args_add_env(args, 1194 (char *)(uintptr_t)env, UIO_USERSPACE); 1195 if (error != 0) 1196 goto err_exit; 1197 } 1198 } 1199 1200 return (0); 1201 1202 err_exit: 1203 exec_free_args(args); 1204 return (error); 1205 } 1206 1207 int 1208 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1209 const void *data, size_t datalen, const int *fds, size_t fdslen) 1210 { 1211 struct filedesc *ofdp; 1212 const char *p; 1213 int *kfds; 1214 int error; 1215 1216 memset(args, '\0', sizeof(*args)); 1217 ofdp = td->td_proc->p_fd; 1218 if (datalen >= ARG_MAX || fdslen > ofdp->fd_lastfile + 1) 1219 return (E2BIG); 1220 error = exec_alloc_args(args); 1221 if (error != 0) 1222 return (error); 1223 1224 args->begin_argv = args->buf; 1225 args->stringspace = ARG_MAX; 1226 1227 if (datalen > 0) { 1228 /* 1229 * Argument buffer has been provided. Copy it into the 1230 * kernel as a single string and add a terminating null 1231 * byte. 1232 */ 1233 error = copyin(data, args->begin_argv, datalen); 1234 if (error != 0) 1235 goto err_exit; 1236 args->begin_argv[datalen] = '\0'; 1237 args->endp = args->begin_argv + datalen + 1; 1238 args->stringspace -= datalen + 1; 1239 1240 /* 1241 * Traditional argument counting. Count the number of 1242 * null bytes. 1243 */ 1244 for (p = args->begin_argv; p < args->endp; ++p) 1245 if (*p == '\0') 1246 ++args->argc; 1247 } else { 1248 /* No argument buffer provided. */ 1249 args->endp = args->begin_argv; 1250 } 1251 1252 /* Create new file descriptor table. */ 1253 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1254 error = copyin(fds, kfds, fdslen * sizeof(int)); 1255 if (error != 0) { 1256 free(kfds, M_TEMP); 1257 goto err_exit; 1258 } 1259 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1260 free(kfds, M_TEMP); 1261 if (error != 0) 1262 goto err_exit; 1263 1264 return (0); 1265 err_exit: 1266 exec_free_args(args); 1267 return (error); 1268 } 1269 1270 struct exec_args_kva { 1271 vm_offset_t addr; 1272 u_int gen; 1273 SLIST_ENTRY(exec_args_kva) next; 1274 }; 1275 1276 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1277 1278 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1279 static struct mtx exec_args_kva_mtx; 1280 static u_int exec_args_gen; 1281 1282 static void 1283 exec_prealloc_args_kva(void *arg __unused) 1284 { 1285 struct exec_args_kva *argkva; 1286 u_int i; 1287 1288 SLIST_INIT(&exec_args_kva_freelist); 1289 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1290 for (i = 0; i < exec_map_entries; i++) { 1291 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1292 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1293 argkva->gen = exec_args_gen; 1294 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1295 } 1296 } 1297 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1298 1299 static vm_offset_t 1300 exec_alloc_args_kva(void **cookie) 1301 { 1302 struct exec_args_kva *argkva; 1303 1304 argkva = (void *)atomic_readandclear_ptr( 1305 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1306 if (argkva == NULL) { 1307 mtx_lock(&exec_args_kva_mtx); 1308 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1309 (void)mtx_sleep(&exec_args_kva_freelist, 1310 &exec_args_kva_mtx, 0, "execkva", 0); 1311 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1312 mtx_unlock(&exec_args_kva_mtx); 1313 } 1314 *(struct exec_args_kva **)cookie = argkva; 1315 return (argkva->addr); 1316 } 1317 1318 static void 1319 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1320 { 1321 vm_offset_t base; 1322 1323 base = argkva->addr; 1324 if (argkva->gen != gen) { 1325 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1326 MADV_FREE); 1327 argkva->gen = gen; 1328 } 1329 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1330 (uintptr_t)NULL, (uintptr_t)argkva)) { 1331 mtx_lock(&exec_args_kva_mtx); 1332 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1333 wakeup_one(&exec_args_kva_freelist); 1334 mtx_unlock(&exec_args_kva_mtx); 1335 } 1336 } 1337 1338 static void 1339 exec_free_args_kva(void *cookie) 1340 { 1341 1342 exec_release_args_kva(cookie, exec_args_gen); 1343 } 1344 1345 static void 1346 exec_args_kva_lowmem(void *arg __unused) 1347 { 1348 SLIST_HEAD(, exec_args_kva) head; 1349 struct exec_args_kva *argkva; 1350 u_int gen; 1351 int i; 1352 1353 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1354 1355 /* 1356 * Force an madvise of each KVA range. Any currently allocated ranges 1357 * will have MADV_FREE applied once they are freed. 1358 */ 1359 SLIST_INIT(&head); 1360 mtx_lock(&exec_args_kva_mtx); 1361 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1362 mtx_unlock(&exec_args_kva_mtx); 1363 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1364 SLIST_REMOVE_HEAD(&head, next); 1365 exec_release_args_kva(argkva, gen); 1366 } 1367 1368 CPU_FOREACH(i) { 1369 argkva = (void *)atomic_readandclear_ptr( 1370 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1371 if (argkva != NULL) 1372 exec_release_args_kva(argkva, gen); 1373 } 1374 } 1375 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1376 EVENTHANDLER_PRI_ANY); 1377 1378 /* 1379 * Allocate temporary demand-paged, zero-filled memory for the file name, 1380 * argument, and environment strings. 1381 */ 1382 int 1383 exec_alloc_args(struct image_args *args) 1384 { 1385 1386 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1387 return (0); 1388 } 1389 1390 void 1391 exec_free_args(struct image_args *args) 1392 { 1393 1394 if (args->buf != NULL) { 1395 exec_free_args_kva(args->bufkva); 1396 args->buf = NULL; 1397 } 1398 if (args->fname_buf != NULL) { 1399 free(args->fname_buf, M_TEMP); 1400 args->fname_buf = NULL; 1401 } 1402 if (args->fdp != NULL) 1403 fdescfree_remapped(args->fdp); 1404 } 1405 1406 /* 1407 * A set to functions to fill struct image args. 1408 * 1409 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1410 * fname) before the other functions. All exec_args_add_arg() calls must 1411 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1412 * may be called any time after exec_args_add_fname(). 1413 * 1414 * exec_args_add_fname() - install path to be executed 1415 * exec_args_add_arg() - append an argument string 1416 * exec_args_add_env() - append an env string 1417 * exec_args_adjust_args() - adjust location of the argument list to 1418 * allow new arguments to be prepended 1419 */ 1420 int 1421 exec_args_add_fname(struct image_args *args, const char *fname, 1422 enum uio_seg segflg) 1423 { 1424 int error; 1425 size_t length; 1426 1427 KASSERT(args->fname == NULL, ("fname already appended")); 1428 KASSERT(args->endp == NULL, ("already appending to args")); 1429 1430 if (fname != NULL) { 1431 args->fname = args->buf; 1432 error = segflg == UIO_SYSSPACE ? 1433 copystr(fname, args->fname, PATH_MAX, &length) : 1434 copyinstr(fname, args->fname, PATH_MAX, &length); 1435 if (error != 0) 1436 return (error == ENAMETOOLONG ? E2BIG : error); 1437 } else 1438 length = 0; 1439 1440 /* Set up for _arg_*()/_env_*() */ 1441 args->endp = args->buf + length; 1442 /* begin_argv must be set and kept updated */ 1443 args->begin_argv = args->endp; 1444 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1445 ("too little space remaining for arguments %zu < %zu", 1446 exec_map_entry_size - length, (size_t)ARG_MAX)); 1447 args->stringspace = ARG_MAX; 1448 1449 return (0); 1450 } 1451 1452 static int 1453 exec_args_add_str(struct image_args *args, const char *str, 1454 enum uio_seg segflg, int *countp) 1455 { 1456 int error; 1457 size_t length; 1458 1459 KASSERT(args->endp != NULL, ("endp not initialized")); 1460 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1461 1462 error = (segflg == UIO_SYSSPACE) ? 1463 copystr(str, args->endp, args->stringspace, &length) : 1464 copyinstr(str, args->endp, args->stringspace, &length); 1465 if (error != 0) 1466 return (error == ENAMETOOLONG ? E2BIG : error); 1467 args->stringspace -= length; 1468 args->endp += length; 1469 (*countp)++; 1470 1471 return (0); 1472 } 1473 1474 int 1475 exec_args_add_arg(struct image_args *args, const char *argp, 1476 enum uio_seg segflg) 1477 { 1478 1479 KASSERT(args->envc == 0, ("appending args after env")); 1480 1481 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1482 } 1483 1484 int 1485 exec_args_add_env(struct image_args *args, const char *envp, 1486 enum uio_seg segflg) 1487 { 1488 1489 if (args->envc == 0) 1490 args->begin_envv = args->endp; 1491 1492 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1493 } 1494 1495 int 1496 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1497 { 1498 ssize_t offset; 1499 1500 KASSERT(args->endp != NULL, ("endp not initialized")); 1501 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1502 1503 offset = extend - consume; 1504 if (args->stringspace < offset) 1505 return (E2BIG); 1506 memmove(args->begin_argv + extend, args->begin_argv + consume, 1507 args->endp - args->begin_argv + consume); 1508 if (args->envc > 0) 1509 args->begin_envv += offset; 1510 args->endp += offset; 1511 args->stringspace -= offset; 1512 return (0); 1513 } 1514 1515 char * 1516 exec_args_get_begin_envv(struct image_args *args) 1517 { 1518 1519 KASSERT(args->endp != NULL, ("endp not initialized")); 1520 1521 if (args->envc > 0) 1522 return (args->begin_envv); 1523 return (args->endp); 1524 } 1525 1526 /* 1527 * Copy strings out to the new process address space, constructing new arg 1528 * and env vector tables. Return a pointer to the base so that it can be used 1529 * as the initial stack pointer. 1530 */ 1531 int 1532 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1533 { 1534 int argc, envc; 1535 char **vectp; 1536 char *stringp; 1537 uintptr_t destp, ustringp; 1538 struct ps_strings *arginfo; 1539 struct proc *p; 1540 size_t execpath_len; 1541 int error, szsigcode, szps; 1542 char canary[sizeof(long) * 8]; 1543 1544 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1545 /* 1546 * Calculate string base and vector table pointers. 1547 * Also deal with signal trampoline code for this exec type. 1548 */ 1549 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1550 execpath_len = strlen(imgp->execpath) + 1; 1551 else 1552 execpath_len = 0; 1553 p = imgp->proc; 1554 szsigcode = 0; 1555 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1556 if (p->p_sysent->sv_sigcode_base == 0) { 1557 if (p->p_sysent->sv_szsigcode != NULL) 1558 szsigcode = *(p->p_sysent->sv_szsigcode); 1559 } 1560 destp = (uintptr_t)arginfo; 1561 1562 /* 1563 * install sigcode 1564 */ 1565 if (szsigcode != 0) { 1566 destp -= szsigcode; 1567 destp = rounddown2(destp, sizeof(void *)); 1568 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1569 szsigcode); 1570 if (error != 0) 1571 return (error); 1572 } 1573 1574 /* 1575 * Copy the image path for the rtld. 1576 */ 1577 if (execpath_len != 0) { 1578 destp -= execpath_len; 1579 destp = rounddown2(destp, sizeof(void *)); 1580 imgp->execpathp = destp; 1581 error = copyout(imgp->execpath, (void *)destp, execpath_len); 1582 if (error != 0) 1583 return (error); 1584 } 1585 1586 /* 1587 * Prepare the canary for SSP. 1588 */ 1589 arc4rand(canary, sizeof(canary), 0); 1590 destp -= sizeof(canary); 1591 imgp->canary = destp; 1592 error = copyout(canary, (void *)destp, sizeof(canary)); 1593 if (error != 0) 1594 return (error); 1595 imgp->canarylen = sizeof(canary); 1596 1597 /* 1598 * Prepare the pagesizes array. 1599 */ 1600 destp -= szps; 1601 destp = rounddown2(destp, sizeof(void *)); 1602 imgp->pagesizes = destp; 1603 error = copyout(pagesizes, (void *)destp, szps); 1604 if (error != 0) 1605 return (error); 1606 imgp->pagesizeslen = szps; 1607 1608 /* 1609 * Allocate room for the argument and environment strings. 1610 */ 1611 destp -= ARG_MAX - imgp->args->stringspace; 1612 destp = rounddown2(destp, sizeof(void *)); 1613 ustringp = destp; 1614 1615 if (imgp->sysent->sv_stackgap != NULL) 1616 imgp->sysent->sv_stackgap(imgp, &destp); 1617 1618 if (imgp->auxargs) { 1619 /* 1620 * Allocate room on the stack for the ELF auxargs 1621 * array. It has up to AT_COUNT entries. 1622 */ 1623 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1624 destp = rounddown2(destp, sizeof(void *)); 1625 } 1626 1627 vectp = (char **)destp; 1628 1629 /* 1630 * Allocate room for the argv[] and env vectors including the 1631 * terminating NULL pointers. 1632 */ 1633 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1634 1635 /* 1636 * vectp also becomes our initial stack base 1637 */ 1638 *stack_base = (uintptr_t)vectp; 1639 1640 stringp = imgp->args->begin_argv; 1641 argc = imgp->args->argc; 1642 envc = imgp->args->envc; 1643 1644 /* 1645 * Copy out strings - arguments and environment. 1646 */ 1647 error = copyout(stringp, (void *)ustringp, 1648 ARG_MAX - imgp->args->stringspace); 1649 if (error != 0) 1650 return (error); 1651 1652 /* 1653 * Fill in "ps_strings" struct for ps, w, etc. 1654 */ 1655 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1656 suword32(&arginfo->ps_nargvstr, argc) != 0) 1657 return (EFAULT); 1658 1659 /* 1660 * Fill in argument portion of vector table. 1661 */ 1662 for (; argc > 0; --argc) { 1663 if (suword(vectp++, ustringp) != 0) 1664 return (EFAULT); 1665 while (*stringp++ != 0) 1666 ustringp++; 1667 ustringp++; 1668 } 1669 1670 /* a null vector table pointer separates the argp's from the envp's */ 1671 if (suword(vectp++, 0) != 0) 1672 return (EFAULT); 1673 1674 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1675 suword32(&arginfo->ps_nenvstr, envc) != 0) 1676 return (EFAULT); 1677 1678 /* 1679 * Fill in environment portion of vector table. 1680 */ 1681 for (; envc > 0; --envc) { 1682 if (suword(vectp++, ustringp) != 0) 1683 return (EFAULT); 1684 while (*stringp++ != 0) 1685 ustringp++; 1686 ustringp++; 1687 } 1688 1689 /* end of vector table is a null pointer */ 1690 if (suword(vectp, 0) != 0) 1691 return (EFAULT); 1692 1693 if (imgp->auxargs) { 1694 vectp++; 1695 error = imgp->sysent->sv_copyout_auxargs(imgp, 1696 (uintptr_t)vectp); 1697 if (error != 0) 1698 return (error); 1699 } 1700 1701 return (0); 1702 } 1703 1704 /* 1705 * Check permissions of file to execute. 1706 * Called with imgp->vp locked. 1707 * Return 0 for success or error code on failure. 1708 */ 1709 int 1710 exec_check_permissions(struct image_params *imgp) 1711 { 1712 struct vnode *vp = imgp->vp; 1713 struct vattr *attr = imgp->attr; 1714 struct thread *td; 1715 int error; 1716 1717 td = curthread; 1718 1719 /* Get file attributes */ 1720 error = VOP_GETATTR(vp, attr, td->td_ucred); 1721 if (error) 1722 return (error); 1723 1724 #ifdef MAC 1725 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1726 if (error) 1727 return (error); 1728 #endif 1729 1730 /* 1731 * 1) Check if file execution is disabled for the filesystem that 1732 * this file resides on. 1733 * 2) Ensure that at least one execute bit is on. Otherwise, a 1734 * privileged user will always succeed, and we don't want this 1735 * to happen unless the file really is executable. 1736 * 3) Ensure that the file is a regular file. 1737 */ 1738 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1739 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1740 (attr->va_type != VREG)) 1741 return (EACCES); 1742 1743 /* 1744 * Zero length files can't be exec'd 1745 */ 1746 if (attr->va_size == 0) 1747 return (ENOEXEC); 1748 1749 /* 1750 * Check for execute permission to file based on current credentials. 1751 */ 1752 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1753 if (error) 1754 return (error); 1755 1756 /* 1757 * Check number of open-for-writes on the file and deny execution 1758 * if there are any. 1759 * 1760 * Add a text reference now so no one can write to the 1761 * executable while we're activating it. 1762 * 1763 * Remember if this was set before and unset it in case this is not 1764 * actually an executable image. 1765 */ 1766 error = VOP_SET_TEXT(vp); 1767 if (error != 0) 1768 return (error); 1769 imgp->textset = true; 1770 1771 /* 1772 * Call filesystem specific open routine (which does nothing in the 1773 * general case). 1774 */ 1775 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1776 if (error == 0) 1777 imgp->opened = 1; 1778 return (error); 1779 } 1780 1781 /* 1782 * Exec handler registration 1783 */ 1784 int 1785 exec_register(const struct execsw *execsw_arg) 1786 { 1787 const struct execsw **es, **xs, **newexecsw; 1788 u_int count = 2; /* New slot and trailing NULL */ 1789 1790 if (execsw) 1791 for (es = execsw; *es; es++) 1792 count++; 1793 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1794 xs = newexecsw; 1795 if (execsw) 1796 for (es = execsw; *es; es++) 1797 *xs++ = *es; 1798 *xs++ = execsw_arg; 1799 *xs = NULL; 1800 if (execsw) 1801 free(execsw, M_TEMP); 1802 execsw = newexecsw; 1803 return (0); 1804 } 1805 1806 int 1807 exec_unregister(const struct execsw *execsw_arg) 1808 { 1809 const struct execsw **es, **xs, **newexecsw; 1810 int count = 1; 1811 1812 if (execsw == NULL) 1813 panic("unregister with no handlers left?\n"); 1814 1815 for (es = execsw; *es; es++) { 1816 if (*es == execsw_arg) 1817 break; 1818 } 1819 if (*es == NULL) 1820 return (ENOENT); 1821 for (es = execsw; *es; es++) 1822 if (*es != execsw_arg) 1823 count++; 1824 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1825 xs = newexecsw; 1826 for (es = execsw; *es; es++) 1827 if (*es != execsw_arg) 1828 *xs++ = *es; 1829 *xs = NULL; 1830 if (execsw) 1831 free(execsw, M_TEMP); 1832 execsw = newexecsw; 1833 return (0); 1834 } 1835