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/priv.h> 55 #include <sys/proc.h> 56 #include <sys/ptrace.h> 57 #include <sys/resourcevar.h> 58 #include <sys/rwlock.h> 59 #include <sys/sched.h> 60 #include <sys/sdt.h> 61 #include <sys/sf_buf.h> 62 #include <sys/shm.h> 63 #include <sys/signalvar.h> 64 #include <sys/smp.h> 65 #include <sys/stat.h> 66 #include <sys/syscallsubr.h> 67 #include <sys/sysctl.h> 68 #include <sys/sysent.h> 69 #include <sys/sysproto.h> 70 #include <sys/timers.h> 71 #include <sys/umtx.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, struct vmspace *oldvmspace); 124 125 /* XXX This should be vm_size_t. */ 126 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 127 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", 128 "Location of process' ps_strings structure"); 129 130 /* XXX This should be vm_size_t. */ 131 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 132 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", 133 "Top of process stack"); 134 135 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 136 NULL, 0, sysctl_kern_stackprot, "I", 137 "Stack memory permissions"); 138 139 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 140 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 141 &ps_arg_cache_limit, 0, 142 "Process' command line characters cache limit"); 143 144 static int disallow_high_osrel; 145 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 146 &disallow_high_osrel, 0, 147 "Disallow execution of binaries built for higher version of the world"); 148 149 static int map_at_zero = 0; 150 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 151 "Permit processes to map an object at virtual address 0."); 152 153 static int 154 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 155 { 156 struct proc *p; 157 int error; 158 159 p = curproc; 160 #ifdef SCTL_MASK32 161 if (req->flags & SCTL_MASK32) { 162 unsigned int val; 163 val = (unsigned int)p->p_sysent->sv_psstrings; 164 error = SYSCTL_OUT(req, &val, sizeof(val)); 165 } else 166 #endif 167 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 168 sizeof(p->p_sysent->sv_psstrings)); 169 return error; 170 } 171 172 static int 173 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 174 { 175 struct proc *p; 176 int error; 177 178 p = curproc; 179 #ifdef SCTL_MASK32 180 if (req->flags & SCTL_MASK32) { 181 unsigned int val; 182 val = (unsigned int)p->p_sysent->sv_usrstack; 183 error = SYSCTL_OUT(req, &val, sizeof(val)); 184 } else 185 #endif 186 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 187 sizeof(p->p_sysent->sv_usrstack)); 188 return error; 189 } 190 191 static int 192 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 193 { 194 struct proc *p; 195 196 p = curproc; 197 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 198 sizeof(p->p_sysent->sv_stackprot))); 199 } 200 201 /* 202 * Each of the items is a pointer to a `const struct execsw', hence the 203 * double pointer here. 204 */ 205 static const struct execsw **execsw; 206 207 #ifndef _SYS_SYSPROTO_H_ 208 struct execve_args { 209 char *fname; 210 char **argv; 211 char **envv; 212 }; 213 #endif 214 215 int 216 sys_execve(struct thread *td, struct execve_args *uap) 217 { 218 struct image_args args; 219 struct vmspace *oldvmspace; 220 int error; 221 222 error = pre_execve(td, &oldvmspace); 223 if (error != 0) 224 return (error); 225 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 226 uap->argv, uap->envv); 227 if (error == 0) 228 error = kern_execve(td, &args, NULL, oldvmspace); 229 post_execve(td, error, oldvmspace); 230 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 231 return (error); 232 } 233 234 #ifndef _SYS_SYSPROTO_H_ 235 struct fexecve_args { 236 int fd; 237 char **argv; 238 char **envv; 239 }; 240 #endif 241 int 242 sys_fexecve(struct thread *td, struct fexecve_args *uap) 243 { 244 struct image_args args; 245 struct vmspace *oldvmspace; 246 int error; 247 248 error = pre_execve(td, &oldvmspace); 249 if (error != 0) 250 return (error); 251 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 252 uap->argv, uap->envv); 253 if (error == 0) { 254 args.fd = uap->fd; 255 error = kern_execve(td, &args, NULL, oldvmspace); 256 } 257 post_execve(td, error, oldvmspace); 258 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 259 return (error); 260 } 261 262 #ifndef _SYS_SYSPROTO_H_ 263 struct __mac_execve_args { 264 char *fname; 265 char **argv; 266 char **envv; 267 struct mac *mac_p; 268 }; 269 #endif 270 271 int 272 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 273 { 274 #ifdef MAC 275 struct image_args args; 276 struct vmspace *oldvmspace; 277 int error; 278 279 error = pre_execve(td, &oldvmspace); 280 if (error != 0) 281 return (error); 282 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 283 uap->argv, uap->envv); 284 if (error == 0) 285 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 286 post_execve(td, error, oldvmspace); 287 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 288 return (error); 289 #else 290 return (ENOSYS); 291 #endif 292 } 293 294 int 295 pre_execve(struct thread *td, struct vmspace **oldvmspace) 296 { 297 struct proc *p; 298 int error; 299 300 KASSERT(td == curthread, ("non-current thread %p", td)); 301 error = 0; 302 p = td->td_proc; 303 if ((p->p_flag & P_HADTHREADS) != 0) { 304 PROC_LOCK(p); 305 if (thread_single(p, SINGLE_BOUNDARY) != 0) 306 error = ERESTART; 307 PROC_UNLOCK(p); 308 } 309 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 310 ("nested execve")); 311 *oldvmspace = p->p_vmspace; 312 return (error); 313 } 314 315 void 316 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 317 { 318 struct proc *p; 319 320 KASSERT(td == curthread, ("non-current thread %p", td)); 321 p = td->td_proc; 322 if ((p->p_flag & P_HADTHREADS) != 0) { 323 PROC_LOCK(p); 324 /* 325 * If success, we upgrade to SINGLE_EXIT state to 326 * force other threads to suicide. 327 */ 328 if (error == EJUSTRETURN) 329 thread_single(p, SINGLE_EXIT); 330 else 331 thread_single_end(p, SINGLE_BOUNDARY); 332 PROC_UNLOCK(p); 333 } 334 exec_cleanup(td, oldvmspace); 335 } 336 337 /* 338 * 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 struct vmspace *oldvmspace) 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, oldvmspace)); 354 } 355 356 static void 357 execve_nosetid(struct image_params *imgp) 358 { 359 imgp->credential_setid = false; 360 if (imgp->newcred != NULL) { 361 crfree(imgp->newcred); 362 imgp->newcred = NULL; 363 } 364 } 365 366 /* 367 * In-kernel implementation of execve(). All arguments are assumed to be 368 * userspace pointers from the passed thread. 369 */ 370 static int 371 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 372 struct vmspace *oldvmspace) 373 { 374 struct proc *p = td->td_proc; 375 struct nameidata nd; 376 struct ucred *oldcred; 377 struct uidinfo *euip = NULL; 378 uintptr_t stack_base; 379 struct image_params image_params, *imgp; 380 struct vattr attr; 381 int (*img_first)(struct image_params *); 382 struct pargs *oldargs = NULL, *newargs = NULL; 383 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 384 #ifdef KTRACE 385 struct ktr_io_params *kiop; 386 #endif 387 struct vnode *oldtextvp = NULL, *newtextvp; 388 int credential_changing; 389 #ifdef MAC 390 struct label *interpvplabel = NULL; 391 int will_transition; 392 #endif 393 #ifdef HWPMC_HOOKS 394 struct pmckern_procexec pe; 395 #endif 396 int error, i, orig_osrel; 397 uint32_t orig_fctl0; 398 static const char fexecv_proc_title[] = "(fexecv)"; 399 400 imgp = &image_params; 401 #ifdef KTRACE 402 kiop = NULL; 403 #endif 404 405 /* 406 * Lock the process and set the P_INEXEC flag to indicate that 407 * it should be left alone until we're done here. This is 408 * necessary to avoid race conditions - e.g. in ptrace() - 409 * that might allow a local user to illicitly obtain elevated 410 * privileges. 411 */ 412 PROC_LOCK(p); 413 KASSERT((p->p_flag & P_INEXEC) == 0, 414 ("%s(): process already has P_INEXEC flag", __func__)); 415 p->p_flag |= P_INEXEC; 416 PROC_UNLOCK(p); 417 418 /* 419 * Initialize part of the common data 420 */ 421 bzero(imgp, sizeof(*imgp)); 422 imgp->proc = p; 423 imgp->attr = &attr; 424 imgp->args = args; 425 oldcred = p->p_ucred; 426 orig_osrel = p->p_osrel; 427 orig_fctl0 = p->p_fctl0; 428 429 #ifdef MAC 430 error = mac_execve_enter(imgp, mac_p); 431 if (error) 432 goto exec_fail; 433 #endif 434 435 /* 436 * Translate the file name. namei() returns a vnode pointer 437 * in ni_vp among other things. 438 * 439 * XXXAUDIT: It would be desirable to also audit the name of the 440 * interpreter if this is an interpreted binary. 441 */ 442 if (args->fname != NULL) { 443 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 444 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 445 } 446 447 SDT_PROBE1(proc, , , exec, args->fname); 448 449 interpret: 450 if (args->fname != NULL) { 451 #ifdef CAPABILITY_MODE 452 /* 453 * While capability mode can't reach this point via direct 454 * path arguments to execve(), we also don't allow 455 * interpreters to be used in capability mode (for now). 456 * Catch indirect lookups and return a permissions error. 457 */ 458 if (IN_CAPABILITY_MODE(td)) { 459 error = ECAPMODE; 460 goto exec_fail; 461 } 462 #endif 463 error = namei(&nd); 464 if (error) 465 goto exec_fail; 466 467 newtextvp = nd.ni_vp; 468 imgp->vp = newtextvp; 469 } else { 470 AUDIT_ARG_FD(args->fd); 471 /* 472 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 473 */ 474 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 475 if (error) 476 goto exec_fail; 477 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 478 AUDIT_ARG_VNODE1(newtextvp); 479 imgp->vp = newtextvp; 480 } 481 482 /* 483 * Check file permissions. Also 'opens' file and sets its vnode to 484 * text mode. 485 */ 486 error = exec_check_permissions(imgp); 487 if (error) 488 goto exec_fail_dealloc; 489 490 imgp->object = imgp->vp->v_object; 491 if (imgp->object != NULL) 492 vm_object_reference(imgp->object); 493 494 error = exec_map_first_page(imgp); 495 if (error) 496 goto exec_fail_dealloc; 497 498 imgp->proc->p_osrel = 0; 499 imgp->proc->p_fctl0 = 0; 500 501 /* 502 * Implement image setuid/setgid. 503 * 504 * Determine new credentials before attempting image activators 505 * so that it can be used by process_exec handlers to determine 506 * credential/setid changes. 507 * 508 * Don't honor setuid/setgid if the filesystem prohibits it or if 509 * the process is being traced. 510 * 511 * We disable setuid/setgid/etc in capability mode on the basis 512 * that most setugid applications are not written with that 513 * environment in mind, and will therefore almost certainly operate 514 * incorrectly. In principle there's no reason that setugid 515 * applications might not be useful in capability mode, so we may want 516 * to reconsider this conservative design choice in the future. 517 * 518 * XXXMAC: For the time being, use NOSUID to also prohibit 519 * transitions on the file system. 520 */ 521 credential_changing = 0; 522 credential_changing |= (attr.va_mode & S_ISUID) && 523 oldcred->cr_uid != attr.va_uid; 524 credential_changing |= (attr.va_mode & S_ISGID) && 525 oldcred->cr_gid != attr.va_gid; 526 #ifdef MAC 527 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 528 interpvplabel, imgp); 529 credential_changing |= will_transition; 530 #endif 531 532 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 533 if (credential_changing) 534 imgp->proc->p_pdeathsig = 0; 535 536 if (credential_changing && 537 #ifdef CAPABILITY_MODE 538 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 539 #endif 540 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 541 (p->p_flag & P_TRACED) == 0) { 542 imgp->credential_setid = true; 543 VOP_UNLOCK(imgp->vp); 544 imgp->newcred = crdup(oldcred); 545 if (attr.va_mode & S_ISUID) { 546 euip = uifind(attr.va_uid); 547 change_euid(imgp->newcred, euip); 548 } 549 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 550 if (attr.va_mode & S_ISGID) 551 change_egid(imgp->newcred, attr.va_gid); 552 /* 553 * Implement correct POSIX saved-id behavior. 554 * 555 * XXXMAC: Note that the current logic will save the 556 * uid and gid if a MAC domain transition occurs, even 557 * though maybe it shouldn't. 558 */ 559 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 560 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 561 } else { 562 /* 563 * Implement correct POSIX saved-id behavior. 564 * 565 * XXX: It's not clear that the existing behavior is 566 * POSIX-compliant. A number of sources indicate that the 567 * saved uid/gid should only be updated if the new ruid is 568 * not equal to the old ruid, or the new euid is not equal 569 * to the old euid and the new euid is not equal to the old 570 * ruid. The FreeBSD code always updates the saved uid/gid. 571 * Also, this code uses the new (replaced) euid and egid as 572 * the source, which may or may not be the right ones to use. 573 */ 574 if (oldcred->cr_svuid != oldcred->cr_uid || 575 oldcred->cr_svgid != oldcred->cr_gid) { 576 VOP_UNLOCK(imgp->vp); 577 imgp->newcred = crdup(oldcred); 578 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 579 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 580 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 581 } 582 } 583 /* The new credentials are installed into the process later. */ 584 585 /* 586 * Do the best to calculate the full path to the image file. 587 */ 588 if (args->fname != NULL && args->fname[0] == '/') 589 imgp->execpath = args->fname; 590 else { 591 VOP_UNLOCK(imgp->vp); 592 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0) 593 imgp->execpath = args->fname; 594 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 595 } 596 597 /* 598 * If the current process has a special image activator it 599 * wants to try first, call it. For example, emulating shell 600 * scripts differently. 601 */ 602 error = -1; 603 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 604 error = img_first(imgp); 605 606 /* 607 * Loop through the list of image activators, calling each one. 608 * An activator returns -1 if there is no match, 0 on success, 609 * and an error otherwise. 610 */ 611 for (i = 0; error == -1 && execsw[i]; ++i) { 612 if (execsw[i]->ex_imgact == NULL || 613 execsw[i]->ex_imgact == img_first) { 614 continue; 615 } 616 error = (*execsw[i]->ex_imgact)(imgp); 617 } 618 619 if (error) { 620 if (error == -1) 621 error = ENOEXEC; 622 goto exec_fail_dealloc; 623 } 624 625 /* 626 * Special interpreter operation, cleanup and loop up to try to 627 * activate the interpreter. 628 */ 629 if (imgp->interpreted) { 630 exec_unmap_first_page(imgp); 631 /* 632 * The text reference needs to be removed for scripts. 633 * There is a short period before we determine that 634 * something is a script where text reference is active. 635 * The vnode lock is held over this entire period 636 * so nothing should illegitimately be blocked. 637 */ 638 MPASS(imgp->textset); 639 VOP_UNSET_TEXT_CHECKED(newtextvp); 640 imgp->textset = false; 641 /* free name buffer and old vnode */ 642 if (args->fname != NULL) 643 NDFREE(&nd, NDF_ONLY_PNBUF); 644 #ifdef MAC 645 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 646 #endif 647 if (imgp->opened) { 648 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 649 imgp->opened = 0; 650 } 651 vput(newtextvp); 652 vm_object_deallocate(imgp->object); 653 imgp->object = NULL; 654 execve_nosetid(imgp); 655 imgp->execpath = NULL; 656 free(imgp->freepath, M_TEMP); 657 imgp->freepath = NULL; 658 /* set new name to that of the interpreter */ 659 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 660 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 661 args->fname = imgp->interpreter_name; 662 goto interpret; 663 } 664 665 /* 666 * NB: We unlock the vnode here because it is believed that none 667 * of the sv_copyout_strings/sv_fixup operations require the vnode. 668 */ 669 VOP_UNLOCK(imgp->vp); 670 671 if (disallow_high_osrel && 672 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 673 error = ENOEXEC; 674 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 675 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 676 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 677 goto exec_fail_dealloc; 678 } 679 680 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 681 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 682 sys_cap_enter(td, NULL); 683 684 /* 685 * Copy out strings (args and env) and initialize stack base. 686 */ 687 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 688 if (error != 0) { 689 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 690 goto exec_fail_dealloc; 691 } 692 693 /* 694 * Stack setup. 695 */ 696 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 697 if (error != 0) { 698 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 699 goto exec_fail_dealloc; 700 } 701 702 if (args->fdp != NULL) { 703 /* Install a brand new file descriptor table. */ 704 fdinstall_remapped(td, args->fdp); 705 args->fdp = NULL; 706 } else { 707 /* 708 * Keep on using the existing file descriptor table. For 709 * security and other reasons, the file descriptor table 710 * cannot be shared after an exec. 711 */ 712 fdunshare(td); 713 pdunshare(td); 714 /* close files on exec */ 715 fdcloseexec(td); 716 } 717 718 /* 719 * Malloc things before we need locks. 720 */ 721 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 722 /* Cache arguments if they fit inside our allowance */ 723 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 724 newargs = pargs_alloc(i); 725 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 726 } 727 728 /* 729 * For security and other reasons, signal handlers cannot 730 * be shared after an exec. The new process gets a copy of the old 731 * handlers. In execsigs(), the new process will have its signals 732 * reset. 733 */ 734 if (sigacts_shared(p->p_sigacts)) { 735 oldsigacts = p->p_sigacts; 736 newsigacts = sigacts_alloc(); 737 sigacts_copy(newsigacts, oldsigacts); 738 } 739 740 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 741 742 PROC_LOCK(p); 743 if (oldsigacts) 744 p->p_sigacts = newsigacts; 745 /* Stop profiling */ 746 stopprofclock(p); 747 748 /* reset caught signals */ 749 execsigs(p); 750 751 /* name this process - nameiexec(p, ndp) */ 752 bzero(p->p_comm, sizeof(p->p_comm)); 753 if (args->fname) 754 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 755 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 756 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 757 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 758 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 759 #ifdef KTR 760 sched_clear_tdname(td); 761 #endif 762 763 /* 764 * mark as execed, wakeup the process that vforked (if any) and tell 765 * it that it now has its own resources back 766 */ 767 p->p_flag |= P_EXEC; 768 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 769 p->p_flag2 &= ~P2_NOTRACE; 770 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 771 p->p_flag2 &= ~P2_STKGAP_DISABLE; 772 if (p->p_flag & P_PPWAIT) { 773 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 774 cv_broadcast(&p->p_pwait); 775 /* STOPs are no longer ignored, arrange for AST */ 776 signotify(td); 777 } 778 779 if (imgp->sysent->sv_setid_allowed != NULL && 780 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) 781 execve_nosetid(imgp); 782 783 /* 784 * Implement image setuid/setgid installation. 785 */ 786 if (imgp->credential_setid) { 787 /* 788 * Turn off syscall tracing for set-id programs, except for 789 * root. Record any set-id flags first to make sure that 790 * we do not regain any tracing during a possible block. 791 */ 792 setsugid(p); 793 #ifdef KTRACE 794 kiop = ktrprocexec(p); 795 #endif 796 /* 797 * Close any file descriptors 0..2 that reference procfs, 798 * then make sure file descriptors 0..2 are in use. 799 * 800 * Both fdsetugidsafety() and fdcheckstd() may call functions 801 * taking sleepable locks, so temporarily drop our locks. 802 */ 803 PROC_UNLOCK(p); 804 VOP_UNLOCK(imgp->vp); 805 fdsetugidsafety(td); 806 error = fdcheckstd(td); 807 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 808 if (error != 0) 809 goto exec_fail_dealloc; 810 PROC_LOCK(p); 811 #ifdef MAC 812 if (will_transition) { 813 mac_vnode_execve_transition(oldcred, imgp->newcred, 814 imgp->vp, interpvplabel, imgp); 815 } 816 #endif 817 } else { 818 if (oldcred->cr_uid == oldcred->cr_ruid && 819 oldcred->cr_gid == oldcred->cr_rgid) 820 p->p_flag &= ~P_SUGID; 821 } 822 /* 823 * Set the new credentials. 824 */ 825 if (imgp->newcred != NULL) { 826 proc_set_cred(p, imgp->newcred); 827 crfree(oldcred); 828 oldcred = NULL; 829 } 830 831 /* 832 * Store the vp for use in procfs. This vnode was referenced by namei 833 * or fgetvp_exec. 834 */ 835 oldtextvp = p->p_textvp; 836 p->p_textvp = newtextvp; 837 838 #ifdef KDTRACE_HOOKS 839 /* 840 * Tell the DTrace fasttrap provider about the exec if it 841 * has declared an interest. 842 */ 843 if (dtrace_fasttrap_exec) 844 dtrace_fasttrap_exec(p); 845 #endif 846 847 /* 848 * Notify others that we exec'd, and clear the P_INEXEC flag 849 * as we're now a bona fide freshly-execed process. 850 */ 851 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 852 p->p_flag &= ~P_INEXEC; 853 854 /* clear "fork but no exec" flag, as we _are_ execing */ 855 p->p_acflag &= ~AFORK; 856 857 /* 858 * Free any previous argument cache and replace it with 859 * the new argument cache, if any. 860 */ 861 oldargs = p->p_args; 862 p->p_args = newargs; 863 newargs = NULL; 864 865 PROC_UNLOCK(p); 866 867 #ifdef HWPMC_HOOKS 868 /* 869 * Check if system-wide sampling is in effect or if the 870 * current process is using PMCs. If so, do exec() time 871 * processing. This processing needs to happen AFTER the 872 * P_INEXEC flag is cleared. 873 */ 874 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 875 VOP_UNLOCK(imgp->vp); 876 pe.pm_credentialschanged = credential_changing; 877 pe.pm_entryaddr = imgp->entry_addr; 878 879 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 880 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 881 } 882 #endif 883 884 /* Set values passed into the program in registers. */ 885 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 886 887 VOP_MMAPPED(imgp->vp); 888 889 SDT_PROBE1(proc, , , exec__success, args->fname); 890 891 exec_fail_dealloc: 892 if (error != 0) { 893 p->p_osrel = orig_osrel; 894 p->p_fctl0 = orig_fctl0; 895 } 896 897 if (imgp->firstpage != NULL) 898 exec_unmap_first_page(imgp); 899 900 if (imgp->vp != NULL) { 901 if (args->fname) 902 NDFREE(&nd, NDF_ONLY_PNBUF); 903 if (imgp->opened) 904 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 905 if (imgp->textset) 906 VOP_UNSET_TEXT_CHECKED(imgp->vp); 907 if (error != 0) 908 vput(imgp->vp); 909 else 910 VOP_UNLOCK(imgp->vp); 911 } 912 913 if (imgp->object != NULL) 914 vm_object_deallocate(imgp->object); 915 916 free(imgp->freepath, M_TEMP); 917 918 if (error == 0) { 919 if (p->p_ptevents & PTRACE_EXEC) { 920 PROC_LOCK(p); 921 if (p->p_ptevents & PTRACE_EXEC) 922 td->td_dbgflags |= TDB_EXEC; 923 PROC_UNLOCK(p); 924 } 925 } else { 926 exec_fail: 927 /* we're done here, clear P_INEXEC */ 928 PROC_LOCK(p); 929 p->p_flag &= ~P_INEXEC; 930 PROC_UNLOCK(p); 931 932 SDT_PROBE1(proc, , , exec__failure, error); 933 } 934 935 if (imgp->newcred != NULL && oldcred != NULL) 936 crfree(imgp->newcred); 937 938 #ifdef MAC 939 mac_execve_exit(imgp); 940 mac_execve_interpreter_exit(interpvplabel); 941 #endif 942 exec_free_args(args); 943 944 /* 945 * Handle deferred decrement of ref counts. 946 */ 947 if (oldtextvp != NULL) 948 vrele(oldtextvp); 949 #ifdef KTRACE 950 ktr_io_params_free(kiop); 951 #endif 952 pargs_drop(oldargs); 953 pargs_drop(newargs); 954 if (oldsigacts != NULL) 955 sigacts_free(oldsigacts); 956 if (euip != NULL) 957 uifree(euip); 958 959 if (error && imgp->vmspace_destroyed) { 960 /* sorry, no more process anymore. exit gracefully */ 961 exec_cleanup(td, oldvmspace); 962 exit1(td, 0, SIGABRT); 963 /* NOT REACHED */ 964 } 965 966 #ifdef KTRACE 967 if (error == 0) 968 ktrprocctor(p); 969 #endif 970 971 /* 972 * We don't want cpu_set_syscall_retval() to overwrite any of 973 * the register values put in place by exec_setregs(). 974 * Implementations of cpu_set_syscall_retval() will leave 975 * registers unmodified when returning EJUSTRETURN. 976 */ 977 return (error == 0 ? EJUSTRETURN : error); 978 } 979 980 void 981 exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 982 { 983 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 984 KASSERT(td->td_proc->p_vmspace != oldvmspace, 985 ("oldvmspace still used")); 986 vmspace_free(oldvmspace); 987 td->td_pflags &= ~TDP_EXECVMSPC; 988 } 989 } 990 991 int 992 exec_map_first_page(struct image_params *imgp) 993 { 994 vm_object_t object; 995 vm_page_t m; 996 int error; 997 998 if (imgp->firstpage != NULL) 999 exec_unmap_first_page(imgp); 1000 1001 object = imgp->vp->v_object; 1002 if (object == NULL) 1003 return (EACCES); 1004 #if VM_NRESERVLEVEL > 0 1005 if ((object->flags & OBJ_COLORED) == 0) { 1006 VM_OBJECT_WLOCK(object); 1007 vm_object_color(object, 0); 1008 VM_OBJECT_WUNLOCK(object); 1009 } 1010 #endif 1011 error = vm_page_grab_valid_unlocked(&m, object, 0, 1012 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1013 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1014 1015 if (error != VM_PAGER_OK) 1016 return (EIO); 1017 imgp->firstpage = sf_buf_alloc(m, 0); 1018 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1019 1020 return (0); 1021 } 1022 1023 void 1024 exec_unmap_first_page(struct image_params *imgp) 1025 { 1026 vm_page_t m; 1027 1028 if (imgp->firstpage != NULL) { 1029 m = sf_buf_page(imgp->firstpage); 1030 sf_buf_free(imgp->firstpage); 1031 imgp->firstpage = NULL; 1032 vm_page_unwire(m, PQ_ACTIVE); 1033 } 1034 } 1035 1036 void 1037 exec_onexec_old(struct thread *td) 1038 { 1039 sigfastblock_clear(td); 1040 umtx_exec(td->td_proc); 1041 } 1042 1043 /* 1044 * Destroy old address space, and allocate a new stack. 1045 * The new stack is only sgrowsiz large because it is grown 1046 * automatically on a page fault. 1047 */ 1048 int 1049 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1050 { 1051 int error; 1052 struct proc *p = imgp->proc; 1053 struct vmspace *vmspace = p->p_vmspace; 1054 struct thread *td = curthread; 1055 vm_object_t obj; 1056 struct rlimit rlim_stack; 1057 vm_offset_t sv_minuser, stack_addr; 1058 vm_map_t map; 1059 vm_prot_t stack_prot; 1060 u_long ssiz; 1061 1062 imgp->vmspace_destroyed = 1; 1063 imgp->sysent = sv; 1064 1065 if (p->p_sysent->sv_onexec_old != NULL) 1066 p->p_sysent->sv_onexec_old(td); 1067 itimers_exec(p); 1068 if (sv->sv_onexec != NULL) 1069 sv->sv_onexec(p, imgp); 1070 1071 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1072 1073 /* 1074 * Blow away entire process VM, if address space not shared, 1075 * otherwise, create a new VM space so that other threads are 1076 * not disrupted 1077 */ 1078 map = &vmspace->vm_map; 1079 if (map_at_zero) 1080 sv_minuser = sv->sv_minuser; 1081 else 1082 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1083 if (refcount_load(&vmspace->vm_refcnt) == 1 && 1084 vm_map_min(map) == sv_minuser && 1085 vm_map_max(map) == sv->sv_maxuser && 1086 cpu_exec_vmspace_reuse(p, map)) { 1087 shmexit(vmspace); 1088 pmap_remove_pages(vmspace_pmap(vmspace)); 1089 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1090 /* 1091 * An exec terminates mlockall(MCL_FUTURE). 1092 * ASLR and W^X states must be re-evaluated. 1093 */ 1094 vm_map_lock(map); 1095 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1096 MAP_ASLR_IGNSTART | MAP_WXORX); 1097 vm_map_unlock(map); 1098 } else { 1099 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1100 if (error) 1101 return (error); 1102 vmspace = p->p_vmspace; 1103 map = &vmspace->vm_map; 1104 } 1105 map->flags |= imgp->map_flags; 1106 1107 /* Map a shared page */ 1108 obj = sv->sv_shared_page_obj; 1109 if (obj != NULL) { 1110 vm_object_reference(obj); 1111 error = vm_map_fixed(map, obj, 0, 1112 sv->sv_shared_page_base, sv->sv_shared_page_len, 1113 VM_PROT_READ | VM_PROT_EXECUTE, 1114 VM_PROT_READ | VM_PROT_EXECUTE, 1115 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1116 if (error != KERN_SUCCESS) { 1117 vm_object_deallocate(obj); 1118 return (vm_mmap_to_errno(error)); 1119 } 1120 } 1121 1122 /* Allocate a new stack */ 1123 if (imgp->stack_sz != 0) { 1124 ssiz = trunc_page(imgp->stack_sz); 1125 PROC_LOCK(p); 1126 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1127 PROC_UNLOCK(p); 1128 if (ssiz > rlim_stack.rlim_max) 1129 ssiz = rlim_stack.rlim_max; 1130 if (ssiz > rlim_stack.rlim_cur) { 1131 rlim_stack.rlim_cur = ssiz; 1132 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1133 } 1134 } else if (sv->sv_maxssiz != NULL) { 1135 ssiz = *sv->sv_maxssiz; 1136 } else { 1137 ssiz = maxssiz; 1138 } 1139 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1140 if (ssiz < imgp->eff_stack_sz) 1141 imgp->eff_stack_sz = ssiz; 1142 stack_addr = sv->sv_usrstack - ssiz; 1143 stack_prot = obj != NULL && imgp->stack_prot != 0 ? 1144 imgp->stack_prot : sv->sv_stackprot; 1145 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot, 1146 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1147 if (error != KERN_SUCCESS) { 1148 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x " 1149 "failed mach error %d errno %d\n", (uintmax_t)ssiz, 1150 stack_prot, error, vm_mmap_to_errno(error)); 1151 return (vm_mmap_to_errno(error)); 1152 } 1153 1154 /* 1155 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1156 * are still used to enforce the stack rlimit on the process stack. 1157 */ 1158 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1159 vmspace->vm_maxsaddr = (char *)stack_addr; 1160 1161 return (0); 1162 } 1163 1164 /* 1165 * Copy out argument and environment strings from the old process address 1166 * space into the temporary string buffer. 1167 */ 1168 int 1169 exec_copyin_args(struct image_args *args, const char *fname, 1170 enum uio_seg segflg, char **argv, char **envv) 1171 { 1172 u_long arg, env; 1173 int error; 1174 1175 bzero(args, sizeof(*args)); 1176 if (argv == NULL) 1177 return (EFAULT); 1178 1179 /* 1180 * Allocate demand-paged memory for the file name, argument, and 1181 * environment strings. 1182 */ 1183 error = exec_alloc_args(args); 1184 if (error != 0) 1185 return (error); 1186 1187 /* 1188 * Copy the file name. 1189 */ 1190 error = exec_args_add_fname(args, fname, segflg); 1191 if (error != 0) 1192 goto err_exit; 1193 1194 /* 1195 * extract arguments first 1196 */ 1197 for (;;) { 1198 error = fueword(argv++, &arg); 1199 if (error == -1) { 1200 error = EFAULT; 1201 goto err_exit; 1202 } 1203 if (arg == 0) 1204 break; 1205 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1206 UIO_USERSPACE); 1207 if (error != 0) 1208 goto err_exit; 1209 } 1210 1211 /* 1212 * extract environment strings 1213 */ 1214 if (envv) { 1215 for (;;) { 1216 error = fueword(envv++, &env); 1217 if (error == -1) { 1218 error = EFAULT; 1219 goto err_exit; 1220 } 1221 if (env == 0) 1222 break; 1223 error = exec_args_add_env(args, 1224 (char *)(uintptr_t)env, UIO_USERSPACE); 1225 if (error != 0) 1226 goto err_exit; 1227 } 1228 } 1229 1230 return (0); 1231 1232 err_exit: 1233 exec_free_args(args); 1234 return (error); 1235 } 1236 1237 int 1238 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1239 const void *data, size_t datalen, const int *fds, size_t fdslen) 1240 { 1241 struct filedesc *ofdp; 1242 const char *p; 1243 int *kfds; 1244 int error; 1245 1246 memset(args, '\0', sizeof(*args)); 1247 ofdp = td->td_proc->p_fd; 1248 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles) 1249 return (E2BIG); 1250 error = exec_alloc_args(args); 1251 if (error != 0) 1252 return (error); 1253 1254 args->begin_argv = args->buf; 1255 args->stringspace = ARG_MAX; 1256 1257 if (datalen > 0) { 1258 /* 1259 * Argument buffer has been provided. Copy it into the 1260 * kernel as a single string and add a terminating null 1261 * byte. 1262 */ 1263 error = copyin(data, args->begin_argv, datalen); 1264 if (error != 0) 1265 goto err_exit; 1266 args->begin_argv[datalen] = '\0'; 1267 args->endp = args->begin_argv + datalen + 1; 1268 args->stringspace -= datalen + 1; 1269 1270 /* 1271 * Traditional argument counting. Count the number of 1272 * null bytes. 1273 */ 1274 for (p = args->begin_argv; p < args->endp; ++p) 1275 if (*p == '\0') 1276 ++args->argc; 1277 } else { 1278 /* No argument buffer provided. */ 1279 args->endp = args->begin_argv; 1280 } 1281 1282 /* Create new file descriptor table. */ 1283 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1284 error = copyin(fds, kfds, fdslen * sizeof(int)); 1285 if (error != 0) { 1286 free(kfds, M_TEMP); 1287 goto err_exit; 1288 } 1289 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1290 free(kfds, M_TEMP); 1291 if (error != 0) 1292 goto err_exit; 1293 1294 return (0); 1295 err_exit: 1296 exec_free_args(args); 1297 return (error); 1298 } 1299 1300 struct exec_args_kva { 1301 vm_offset_t addr; 1302 u_int gen; 1303 SLIST_ENTRY(exec_args_kva) next; 1304 }; 1305 1306 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1307 1308 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1309 static struct mtx exec_args_kva_mtx; 1310 static u_int exec_args_gen; 1311 1312 static void 1313 exec_prealloc_args_kva(void *arg __unused) 1314 { 1315 struct exec_args_kva *argkva; 1316 u_int i; 1317 1318 SLIST_INIT(&exec_args_kva_freelist); 1319 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1320 for (i = 0; i < exec_map_entries; i++) { 1321 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1322 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1323 argkva->gen = exec_args_gen; 1324 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1325 } 1326 } 1327 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1328 1329 static vm_offset_t 1330 exec_alloc_args_kva(void **cookie) 1331 { 1332 struct exec_args_kva *argkva; 1333 1334 argkva = (void *)atomic_readandclear_ptr( 1335 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1336 if (argkva == NULL) { 1337 mtx_lock(&exec_args_kva_mtx); 1338 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1339 (void)mtx_sleep(&exec_args_kva_freelist, 1340 &exec_args_kva_mtx, 0, "execkva", 0); 1341 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1342 mtx_unlock(&exec_args_kva_mtx); 1343 } 1344 *(struct exec_args_kva **)cookie = argkva; 1345 return (argkva->addr); 1346 } 1347 1348 static void 1349 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1350 { 1351 vm_offset_t base; 1352 1353 base = argkva->addr; 1354 if (argkva->gen != gen) { 1355 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1356 MADV_FREE); 1357 argkva->gen = gen; 1358 } 1359 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1360 (uintptr_t)NULL, (uintptr_t)argkva)) { 1361 mtx_lock(&exec_args_kva_mtx); 1362 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1363 wakeup_one(&exec_args_kva_freelist); 1364 mtx_unlock(&exec_args_kva_mtx); 1365 } 1366 } 1367 1368 static void 1369 exec_free_args_kva(void *cookie) 1370 { 1371 1372 exec_release_args_kva(cookie, exec_args_gen); 1373 } 1374 1375 static void 1376 exec_args_kva_lowmem(void *arg __unused) 1377 { 1378 SLIST_HEAD(, exec_args_kva) head; 1379 struct exec_args_kva *argkva; 1380 u_int gen; 1381 int i; 1382 1383 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1384 1385 /* 1386 * Force an madvise of each KVA range. Any currently allocated ranges 1387 * will have MADV_FREE applied once they are freed. 1388 */ 1389 SLIST_INIT(&head); 1390 mtx_lock(&exec_args_kva_mtx); 1391 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1392 mtx_unlock(&exec_args_kva_mtx); 1393 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1394 SLIST_REMOVE_HEAD(&head, next); 1395 exec_release_args_kva(argkva, gen); 1396 } 1397 1398 CPU_FOREACH(i) { 1399 argkva = (void *)atomic_readandclear_ptr( 1400 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1401 if (argkva != NULL) 1402 exec_release_args_kva(argkva, gen); 1403 } 1404 } 1405 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1406 EVENTHANDLER_PRI_ANY); 1407 1408 /* 1409 * Allocate temporary demand-paged, zero-filled memory for the file name, 1410 * argument, and environment strings. 1411 */ 1412 int 1413 exec_alloc_args(struct image_args *args) 1414 { 1415 1416 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1417 return (0); 1418 } 1419 1420 void 1421 exec_free_args(struct image_args *args) 1422 { 1423 1424 if (args->buf != NULL) { 1425 exec_free_args_kva(args->bufkva); 1426 args->buf = NULL; 1427 } 1428 if (args->fname_buf != NULL) { 1429 free(args->fname_buf, M_TEMP); 1430 args->fname_buf = NULL; 1431 } 1432 if (args->fdp != NULL) 1433 fdescfree_remapped(args->fdp); 1434 } 1435 1436 /* 1437 * A set to functions to fill struct image args. 1438 * 1439 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1440 * fname) before the other functions. All exec_args_add_arg() calls must 1441 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1442 * may be called any time after exec_args_add_fname(). 1443 * 1444 * exec_args_add_fname() - install path to be executed 1445 * exec_args_add_arg() - append an argument string 1446 * exec_args_add_env() - append an env string 1447 * exec_args_adjust_args() - adjust location of the argument list to 1448 * allow new arguments to be prepended 1449 */ 1450 int 1451 exec_args_add_fname(struct image_args *args, const char *fname, 1452 enum uio_seg segflg) 1453 { 1454 int error; 1455 size_t length; 1456 1457 KASSERT(args->fname == NULL, ("fname already appended")); 1458 KASSERT(args->endp == NULL, ("already appending to args")); 1459 1460 if (fname != NULL) { 1461 args->fname = args->buf; 1462 error = segflg == UIO_SYSSPACE ? 1463 copystr(fname, args->fname, PATH_MAX, &length) : 1464 copyinstr(fname, args->fname, PATH_MAX, &length); 1465 if (error != 0) 1466 return (error == ENAMETOOLONG ? E2BIG : error); 1467 } else 1468 length = 0; 1469 1470 /* Set up for _arg_*()/_env_*() */ 1471 args->endp = args->buf + length; 1472 /* begin_argv must be set and kept updated */ 1473 args->begin_argv = args->endp; 1474 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1475 ("too little space remaining for arguments %zu < %zu", 1476 exec_map_entry_size - length, (size_t)ARG_MAX)); 1477 args->stringspace = ARG_MAX; 1478 1479 return (0); 1480 } 1481 1482 static int 1483 exec_args_add_str(struct image_args *args, const char *str, 1484 enum uio_seg segflg, int *countp) 1485 { 1486 int error; 1487 size_t length; 1488 1489 KASSERT(args->endp != NULL, ("endp not initialized")); 1490 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1491 1492 error = (segflg == UIO_SYSSPACE) ? 1493 copystr(str, args->endp, args->stringspace, &length) : 1494 copyinstr(str, args->endp, args->stringspace, &length); 1495 if (error != 0) 1496 return (error == ENAMETOOLONG ? E2BIG : error); 1497 args->stringspace -= length; 1498 args->endp += length; 1499 (*countp)++; 1500 1501 return (0); 1502 } 1503 1504 int 1505 exec_args_add_arg(struct image_args *args, const char *argp, 1506 enum uio_seg segflg) 1507 { 1508 1509 KASSERT(args->envc == 0, ("appending args after env")); 1510 1511 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1512 } 1513 1514 int 1515 exec_args_add_env(struct image_args *args, const char *envp, 1516 enum uio_seg segflg) 1517 { 1518 1519 if (args->envc == 0) 1520 args->begin_envv = args->endp; 1521 1522 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1523 } 1524 1525 int 1526 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1527 { 1528 ssize_t offset; 1529 1530 KASSERT(args->endp != NULL, ("endp not initialized")); 1531 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1532 1533 offset = extend - consume; 1534 if (args->stringspace < offset) 1535 return (E2BIG); 1536 memmove(args->begin_argv + extend, args->begin_argv + consume, 1537 args->endp - args->begin_argv + consume); 1538 if (args->envc > 0) 1539 args->begin_envv += offset; 1540 args->endp += offset; 1541 args->stringspace -= offset; 1542 return (0); 1543 } 1544 1545 char * 1546 exec_args_get_begin_envv(struct image_args *args) 1547 { 1548 1549 KASSERT(args->endp != NULL, ("endp not initialized")); 1550 1551 if (args->envc > 0) 1552 return (args->begin_envv); 1553 return (args->endp); 1554 } 1555 1556 void 1557 exec_stackgap(struct image_params *imgp, uintptr_t *dp) 1558 { 1559 if (imgp->sysent->sv_stackgap == NULL || 1560 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE | 1561 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 || 1562 (imgp->map_flags & MAP_ASLR) == 0) 1563 return; 1564 imgp->sysent->sv_stackgap(imgp, dp); 1565 } 1566 1567 /* 1568 * Copy strings out to the new process address space, constructing new arg 1569 * and env vector tables. Return a pointer to the base so that it can be used 1570 * as the initial stack pointer. 1571 */ 1572 int 1573 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1574 { 1575 int argc, envc; 1576 char **vectp; 1577 char *stringp; 1578 uintptr_t destp, ustringp; 1579 struct ps_strings *arginfo; 1580 struct proc *p; 1581 size_t execpath_len; 1582 int error, szsigcode, szps; 1583 char canary[sizeof(long) * 8]; 1584 1585 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1586 /* 1587 * Calculate string base and vector table pointers. 1588 * Also deal with signal trampoline code for this exec type. 1589 */ 1590 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1591 execpath_len = strlen(imgp->execpath) + 1; 1592 else 1593 execpath_len = 0; 1594 p = imgp->proc; 1595 szsigcode = 0; 1596 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1597 imgp->ps_strings = arginfo; 1598 if (p->p_sysent->sv_sigcode_base == 0) { 1599 if (p->p_sysent->sv_szsigcode != NULL) 1600 szsigcode = *(p->p_sysent->sv_szsigcode); 1601 } 1602 destp = (uintptr_t)arginfo; 1603 1604 /* 1605 * install sigcode 1606 */ 1607 if (szsigcode != 0) { 1608 destp -= szsigcode; 1609 destp = rounddown2(destp, sizeof(void *)); 1610 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1611 szsigcode); 1612 if (error != 0) 1613 return (error); 1614 } 1615 1616 /* 1617 * Copy the image path for the rtld. 1618 */ 1619 if (execpath_len != 0) { 1620 destp -= execpath_len; 1621 destp = rounddown2(destp, sizeof(void *)); 1622 imgp->execpathp = (void *)destp; 1623 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1624 if (error != 0) 1625 return (error); 1626 } 1627 1628 /* 1629 * Prepare the canary for SSP. 1630 */ 1631 arc4rand(canary, sizeof(canary), 0); 1632 destp -= sizeof(canary); 1633 imgp->canary = (void *)destp; 1634 error = copyout(canary, imgp->canary, sizeof(canary)); 1635 if (error != 0) 1636 return (error); 1637 imgp->canarylen = sizeof(canary); 1638 1639 /* 1640 * Prepare the pagesizes array. 1641 */ 1642 destp -= szps; 1643 destp = rounddown2(destp, sizeof(void *)); 1644 imgp->pagesizes = (void *)destp; 1645 error = copyout(pagesizes, imgp->pagesizes, szps); 1646 if (error != 0) 1647 return (error); 1648 imgp->pagesizeslen = szps; 1649 1650 /* 1651 * Allocate room for the argument and environment strings. 1652 */ 1653 destp -= ARG_MAX - imgp->args->stringspace; 1654 destp = rounddown2(destp, sizeof(void *)); 1655 ustringp = destp; 1656 1657 exec_stackgap(imgp, &destp); 1658 1659 if (imgp->auxargs) { 1660 /* 1661 * Allocate room on the stack for the ELF auxargs 1662 * array. It has up to AT_COUNT entries. 1663 */ 1664 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1665 destp = rounddown2(destp, sizeof(void *)); 1666 } 1667 1668 vectp = (char **)destp; 1669 1670 /* 1671 * Allocate room for the argv[] and env vectors including the 1672 * terminating NULL pointers. 1673 */ 1674 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1675 1676 /* 1677 * vectp also becomes our initial stack base 1678 */ 1679 *stack_base = (uintptr_t)vectp; 1680 1681 stringp = imgp->args->begin_argv; 1682 argc = imgp->args->argc; 1683 envc = imgp->args->envc; 1684 1685 /* 1686 * Copy out strings - arguments and environment. 1687 */ 1688 error = copyout(stringp, (void *)ustringp, 1689 ARG_MAX - imgp->args->stringspace); 1690 if (error != 0) 1691 return (error); 1692 1693 /* 1694 * Fill in "ps_strings" struct for ps, w, etc. 1695 */ 1696 imgp->argv = vectp; 1697 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1698 suword32(&arginfo->ps_nargvstr, argc) != 0) 1699 return (EFAULT); 1700 1701 /* 1702 * Fill in argument portion of vector table. 1703 */ 1704 for (; argc > 0; --argc) { 1705 if (suword(vectp++, ustringp) != 0) 1706 return (EFAULT); 1707 while (*stringp++ != 0) 1708 ustringp++; 1709 ustringp++; 1710 } 1711 1712 /* a null vector table pointer separates the argp's from the envp's */ 1713 if (suword(vectp++, 0) != 0) 1714 return (EFAULT); 1715 1716 imgp->envv = vectp; 1717 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1718 suword32(&arginfo->ps_nenvstr, envc) != 0) 1719 return (EFAULT); 1720 1721 /* 1722 * Fill in environment portion of vector table. 1723 */ 1724 for (; envc > 0; --envc) { 1725 if (suword(vectp++, ustringp) != 0) 1726 return (EFAULT); 1727 while (*stringp++ != 0) 1728 ustringp++; 1729 ustringp++; 1730 } 1731 1732 /* end of vector table is a null pointer */ 1733 if (suword(vectp, 0) != 0) 1734 return (EFAULT); 1735 1736 if (imgp->auxargs) { 1737 vectp++; 1738 error = imgp->sysent->sv_copyout_auxargs(imgp, 1739 (uintptr_t)vectp); 1740 if (error != 0) 1741 return (error); 1742 } 1743 1744 return (0); 1745 } 1746 1747 /* 1748 * Check permissions of file to execute. 1749 * Called with imgp->vp locked. 1750 * Return 0 for success or error code on failure. 1751 */ 1752 int 1753 exec_check_permissions(struct image_params *imgp) 1754 { 1755 struct vnode *vp = imgp->vp; 1756 struct vattr *attr = imgp->attr; 1757 struct thread *td; 1758 int error; 1759 1760 td = curthread; 1761 1762 /* Get file attributes */ 1763 error = VOP_GETATTR(vp, attr, td->td_ucred); 1764 if (error) 1765 return (error); 1766 1767 #ifdef MAC 1768 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1769 if (error) 1770 return (error); 1771 #endif 1772 1773 /* 1774 * 1) Check if file execution is disabled for the filesystem that 1775 * this file resides on. 1776 * 2) Ensure that at least one execute bit is on. Otherwise, a 1777 * privileged user will always succeed, and we don't want this 1778 * to happen unless the file really is executable. 1779 * 3) Ensure that the file is a regular file. 1780 */ 1781 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1782 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1783 (attr->va_type != VREG)) 1784 return (EACCES); 1785 1786 /* 1787 * Zero length files can't be exec'd 1788 */ 1789 if (attr->va_size == 0) 1790 return (ENOEXEC); 1791 1792 /* 1793 * Check for execute permission to file based on current credentials. 1794 */ 1795 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1796 if (error) 1797 return (error); 1798 1799 /* 1800 * Check number of open-for-writes on the file and deny execution 1801 * if there are any. 1802 * 1803 * Add a text reference now so no one can write to the 1804 * executable while we're activating it. 1805 * 1806 * Remember if this was set before and unset it in case this is not 1807 * actually an executable image. 1808 */ 1809 error = VOP_SET_TEXT(vp); 1810 if (error != 0) 1811 return (error); 1812 imgp->textset = true; 1813 1814 /* 1815 * Call filesystem specific open routine (which does nothing in the 1816 * general case). 1817 */ 1818 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1819 if (error == 0) 1820 imgp->opened = 1; 1821 return (error); 1822 } 1823 1824 /* 1825 * Exec handler registration 1826 */ 1827 int 1828 exec_register(const struct execsw *execsw_arg) 1829 { 1830 const struct execsw **es, **xs, **newexecsw; 1831 u_int count = 2; /* New slot and trailing NULL */ 1832 1833 if (execsw) 1834 for (es = execsw; *es; es++) 1835 count++; 1836 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1837 xs = newexecsw; 1838 if (execsw) 1839 for (es = execsw; *es; es++) 1840 *xs++ = *es; 1841 *xs++ = execsw_arg; 1842 *xs = NULL; 1843 if (execsw) 1844 free(execsw, M_TEMP); 1845 execsw = newexecsw; 1846 return (0); 1847 } 1848 1849 int 1850 exec_unregister(const struct execsw *execsw_arg) 1851 { 1852 const struct execsw **es, **xs, **newexecsw; 1853 int count = 1; 1854 1855 if (execsw == NULL) 1856 panic("unregister with no handlers left?\n"); 1857 1858 for (es = execsw; *es; es++) { 1859 if (*es == execsw_arg) 1860 break; 1861 } 1862 if (*es == NULL) 1863 return (ENOENT); 1864 for (es = execsw; *es; es++) 1865 if (*es != execsw_arg) 1866 count++; 1867 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1868 xs = newexecsw; 1869 for (es = execsw; *es; es++) 1870 if (*es != execsw_arg) 1871 *xs++ = *es; 1872 *xs = NULL; 1873 if (execsw) 1874 free(execsw, M_TEMP); 1875 execsw = newexecsw; 1876 return (0); 1877 } 1878