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/vnode.h> 71 #include <sys/wait.h> 72 #ifdef KTRACE 73 #include <sys/ktrace.h> 74 #endif 75 76 #include <vm/vm.h> 77 #include <vm/vm_param.h> 78 #include <vm/pmap.h> 79 #include <vm/vm_page.h> 80 #include <vm/vm_map.h> 81 #include <vm/vm_kern.h> 82 #include <vm/vm_extern.h> 83 #include <vm/vm_object.h> 84 #include <vm/vm_pager.h> 85 86 #ifdef HWPMC_HOOKS 87 #include <sys/pmckern.h> 88 #endif 89 90 #include <machine/reg.h> 91 92 #include <security/audit/audit.h> 93 #include <security/mac/mac_framework.h> 94 95 #ifdef KDTRACE_HOOKS 96 #include <sys/dtrace_bsd.h> 97 dtrace_execexit_func_t dtrace_fasttrap_exec; 98 #endif 99 100 SDT_PROVIDER_DECLARE(proc); 101 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 102 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 103 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 104 105 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 106 107 int coredump_pack_fileinfo = 1; 108 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 109 &coredump_pack_fileinfo, 0, 110 "Enable file path packing in 'procstat -f' coredump notes"); 111 112 int coredump_pack_vmmapinfo = 1; 113 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 114 &coredump_pack_vmmapinfo, 0, 115 "Enable file path packing in 'procstat -v' coredump notes"); 116 117 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 118 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 119 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 120 static int do_execve(struct thread *td, struct image_args *args, 121 struct mac *mac_p, struct vmspace *oldvmspace); 122 123 /* XXX This should be vm_size_t. */ 124 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 125 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", 126 "Location of process' ps_strings structure"); 127 128 /* XXX This should be vm_size_t. */ 129 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 130 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", 131 "Top of process stack"); 132 133 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 134 NULL, 0, sysctl_kern_stackprot, "I", 135 "Stack memory permissions"); 136 137 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 138 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 139 &ps_arg_cache_limit, 0, 140 "Process' command line characters cache limit"); 141 142 static int disallow_high_osrel; 143 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 144 &disallow_high_osrel, 0, 145 "Disallow execution of binaries built for higher version of the world"); 146 147 static int map_at_zero = 0; 148 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 149 "Permit processes to map an object at virtual address 0."); 150 151 static int 152 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 153 { 154 struct proc *p; 155 int error; 156 157 p = curproc; 158 #ifdef SCTL_MASK32 159 if (req->flags & SCTL_MASK32) { 160 unsigned int val; 161 val = (unsigned int)p->p_sysent->sv_psstrings; 162 error = SYSCTL_OUT(req, &val, sizeof(val)); 163 } else 164 #endif 165 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 166 sizeof(p->p_sysent->sv_psstrings)); 167 return error; 168 } 169 170 static int 171 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 172 { 173 struct proc *p; 174 int error; 175 176 p = curproc; 177 #ifdef SCTL_MASK32 178 if (req->flags & SCTL_MASK32) { 179 unsigned int val; 180 val = (unsigned int)p->p_sysent->sv_usrstack; 181 error = SYSCTL_OUT(req, &val, sizeof(val)); 182 } else 183 #endif 184 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 185 sizeof(p->p_sysent->sv_usrstack)); 186 return error; 187 } 188 189 static int 190 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 191 { 192 struct proc *p; 193 194 p = curproc; 195 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 196 sizeof(p->p_sysent->sv_stackprot))); 197 } 198 199 /* 200 * Each of the items is a pointer to a `const struct execsw', hence the 201 * double pointer here. 202 */ 203 static const struct execsw **execsw; 204 205 #ifndef _SYS_SYSPROTO_H_ 206 struct execve_args { 207 char *fname; 208 char **argv; 209 char **envv; 210 }; 211 #endif 212 213 int 214 sys_execve(struct thread *td, struct execve_args *uap) 215 { 216 struct image_args args; 217 struct vmspace *oldvmspace; 218 int error; 219 220 error = pre_execve(td, &oldvmspace); 221 if (error != 0) 222 return (error); 223 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 224 uap->argv, uap->envv); 225 if (error == 0) 226 error = kern_execve(td, &args, NULL, oldvmspace); 227 post_execve(td, error, oldvmspace); 228 return (error); 229 } 230 231 #ifndef _SYS_SYSPROTO_H_ 232 struct fexecve_args { 233 int fd; 234 char **argv; 235 char **envv; 236 }; 237 #endif 238 int 239 sys_fexecve(struct thread *td, struct fexecve_args *uap) 240 { 241 struct image_args args; 242 struct vmspace *oldvmspace; 243 int error; 244 245 error = pre_execve(td, &oldvmspace); 246 if (error != 0) 247 return (error); 248 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 249 uap->argv, uap->envv); 250 if (error == 0) { 251 args.fd = uap->fd; 252 error = kern_execve(td, &args, NULL, oldvmspace); 253 } 254 post_execve(td, error, oldvmspace); 255 return (error); 256 } 257 258 #ifndef _SYS_SYSPROTO_H_ 259 struct __mac_execve_args { 260 char *fname; 261 char **argv; 262 char **envv; 263 struct mac *mac_p; 264 }; 265 #endif 266 267 int 268 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 269 { 270 #ifdef MAC 271 struct image_args args; 272 struct vmspace *oldvmspace; 273 int error; 274 275 error = pre_execve(td, &oldvmspace); 276 if (error != 0) 277 return (error); 278 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 279 uap->argv, uap->envv); 280 if (error == 0) 281 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 282 post_execve(td, error, oldvmspace); 283 return (error); 284 #else 285 return (ENOSYS); 286 #endif 287 } 288 289 int 290 pre_execve(struct thread *td, struct vmspace **oldvmspace) 291 { 292 struct proc *p; 293 int error; 294 295 KASSERT(td == curthread, ("non-current thread %p", td)); 296 error = 0; 297 p = td->td_proc; 298 if ((p->p_flag & P_HADTHREADS) != 0) { 299 PROC_LOCK(p); 300 if (thread_single(p, SINGLE_BOUNDARY) != 0) 301 error = ERESTART; 302 PROC_UNLOCK(p); 303 } 304 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 305 ("nested execve")); 306 *oldvmspace = p->p_vmspace; 307 return (error); 308 } 309 310 void 311 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 312 { 313 struct proc *p; 314 315 KASSERT(td == curthread, ("non-current thread %p", td)); 316 p = td->td_proc; 317 if ((p->p_flag & P_HADTHREADS) != 0) { 318 PROC_LOCK(p); 319 /* 320 * If success, we upgrade to SINGLE_EXIT state to 321 * force other threads to suicide. 322 */ 323 if (error == EJUSTRETURN) 324 thread_single(p, SINGLE_EXIT); 325 else 326 thread_single_end(p, SINGLE_BOUNDARY); 327 PROC_UNLOCK(p); 328 } 329 exec_cleanup(td, oldvmspace); 330 } 331 332 /* 333 * kern_execve() has the astonishing property of not always returning to 334 * the caller. If sufficiently bad things happen during the call to 335 * do_execve(), it can end up calling exit1(); as a result, callers must 336 * avoid doing anything which they might need to undo (e.g., allocating 337 * memory). 338 */ 339 int 340 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 341 struct vmspace *oldvmspace) 342 { 343 344 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 345 exec_args_get_begin_envv(args) - args->begin_argv); 346 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 347 args->endp - exec_args_get_begin_envv(args)); 348 return (do_execve(td, args, mac_p, oldvmspace)); 349 } 350 351 /* 352 * In-kernel implementation of execve(). All arguments are assumed to be 353 * userspace pointers from the passed thread. 354 */ 355 static int 356 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 357 struct vmspace *oldvmspace) 358 { 359 struct proc *p = td->td_proc; 360 struct nameidata nd; 361 struct ucred *oldcred; 362 struct uidinfo *euip = NULL; 363 uintptr_t stack_base; 364 struct image_params image_params, *imgp; 365 struct vattr attr; 366 int (*img_first)(struct image_params *); 367 struct pargs *oldargs = NULL, *newargs = NULL; 368 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 369 #ifdef KTRACE 370 struct vnode *tracevp = NULL; 371 struct ucred *tracecred = NULL; 372 #endif 373 struct vnode *oldtextvp = NULL, *newtextvp; 374 int credential_changing; 375 #ifdef MAC 376 struct label *interpvplabel = NULL; 377 int will_transition; 378 #endif 379 #ifdef HWPMC_HOOKS 380 struct pmckern_procexec pe; 381 #endif 382 int error, i, orig_osrel; 383 uint32_t orig_fctl0; 384 static const char fexecv_proc_title[] = "(fexecv)"; 385 386 imgp = &image_params; 387 388 /* 389 * Lock the process and set the P_INEXEC flag to indicate that 390 * it should be left alone until we're done here. This is 391 * necessary to avoid race conditions - e.g. in ptrace() - 392 * that might allow a local user to illicitly obtain elevated 393 * privileges. 394 */ 395 PROC_LOCK(p); 396 KASSERT((p->p_flag & P_INEXEC) == 0, 397 ("%s(): process already has P_INEXEC flag", __func__)); 398 p->p_flag |= P_INEXEC; 399 PROC_UNLOCK(p); 400 401 /* 402 * Initialize part of the common data 403 */ 404 bzero(imgp, sizeof(*imgp)); 405 imgp->proc = p; 406 imgp->attr = &attr; 407 imgp->args = args; 408 oldcred = p->p_ucred; 409 orig_osrel = p->p_osrel; 410 orig_fctl0 = p->p_fctl0; 411 412 #ifdef MAC 413 error = mac_execve_enter(imgp, mac_p); 414 if (error) 415 goto exec_fail; 416 #endif 417 418 /* 419 * Translate the file name. namei() returns a vnode pointer 420 * in ni_vp among other things. 421 * 422 * XXXAUDIT: It would be desirable to also audit the name of the 423 * interpreter if this is an interpreted binary. 424 */ 425 if (args->fname != NULL) { 426 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 427 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 428 } 429 430 SDT_PROBE1(proc, , , exec, args->fname); 431 432 interpret: 433 if (args->fname != NULL) { 434 #ifdef CAPABILITY_MODE 435 /* 436 * While capability mode can't reach this point via direct 437 * path arguments to execve(), we also don't allow 438 * interpreters to be used in capability mode (for now). 439 * Catch indirect lookups and return a permissions error. 440 */ 441 if (IN_CAPABILITY_MODE(td)) { 442 error = ECAPMODE; 443 goto exec_fail; 444 } 445 #endif 446 error = namei(&nd); 447 if (error) 448 goto exec_fail; 449 450 newtextvp = nd.ni_vp; 451 imgp->vp = newtextvp; 452 } else { 453 AUDIT_ARG_FD(args->fd); 454 /* 455 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 456 */ 457 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 458 if (error) 459 goto exec_fail; 460 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 461 AUDIT_ARG_VNODE1(newtextvp); 462 imgp->vp = newtextvp; 463 } 464 465 /* 466 * Check file permissions. Also 'opens' file and sets its vnode to 467 * text mode. 468 */ 469 error = exec_check_permissions(imgp); 470 if (error) 471 goto exec_fail_dealloc; 472 473 imgp->object = imgp->vp->v_object; 474 if (imgp->object != NULL) 475 vm_object_reference(imgp->object); 476 477 error = exec_map_first_page(imgp); 478 if (error) 479 goto exec_fail_dealloc; 480 481 imgp->proc->p_osrel = 0; 482 imgp->proc->p_fctl0 = 0; 483 484 /* 485 * Implement image setuid/setgid. 486 * 487 * Determine new credentials before attempting image activators 488 * so that it can be used by process_exec handlers to determine 489 * credential/setid changes. 490 * 491 * Don't honor setuid/setgid if the filesystem prohibits it or if 492 * the process is being traced. 493 * 494 * We disable setuid/setgid/etc in capability mode on the basis 495 * that most setugid applications are not written with that 496 * environment in mind, and will therefore almost certainly operate 497 * incorrectly. In principle there's no reason that setugid 498 * applications might not be useful in capability mode, so we may want 499 * to reconsider this conservative design choice in the future. 500 * 501 * XXXMAC: For the time being, use NOSUID to also prohibit 502 * transitions on the file system. 503 */ 504 credential_changing = 0; 505 credential_changing |= (attr.va_mode & S_ISUID) && 506 oldcred->cr_uid != attr.va_uid; 507 credential_changing |= (attr.va_mode & S_ISGID) && 508 oldcred->cr_gid != attr.va_gid; 509 #ifdef MAC 510 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 511 interpvplabel, imgp); 512 credential_changing |= will_transition; 513 #endif 514 515 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 516 if (credential_changing) 517 imgp->proc->p_pdeathsig = 0; 518 519 if (credential_changing && 520 #ifdef CAPABILITY_MODE 521 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 522 #endif 523 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 524 (p->p_flag & P_TRACED) == 0) { 525 imgp->credential_setid = true; 526 VOP_UNLOCK(imgp->vp); 527 imgp->newcred = crdup(oldcred); 528 if (attr.va_mode & S_ISUID) { 529 euip = uifind(attr.va_uid); 530 change_euid(imgp->newcred, euip); 531 } 532 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 533 if (attr.va_mode & S_ISGID) 534 change_egid(imgp->newcred, attr.va_gid); 535 /* 536 * Implement correct POSIX saved-id behavior. 537 * 538 * XXXMAC: Note that the current logic will save the 539 * uid and gid if a MAC domain transition occurs, even 540 * though maybe it shouldn't. 541 */ 542 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 543 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 544 } else { 545 /* 546 * Implement correct POSIX saved-id behavior. 547 * 548 * XXX: It's not clear that the existing behavior is 549 * POSIX-compliant. A number of sources indicate that the 550 * saved uid/gid should only be updated if the new ruid is 551 * not equal to the old ruid, or the new euid is not equal 552 * to the old euid and the new euid is not equal to the old 553 * ruid. The FreeBSD code always updates the saved uid/gid. 554 * Also, this code uses the new (replaced) euid and egid as 555 * the source, which may or may not be the right ones to use. 556 */ 557 if (oldcred->cr_svuid != oldcred->cr_uid || 558 oldcred->cr_svgid != oldcred->cr_gid) { 559 VOP_UNLOCK(imgp->vp); 560 imgp->newcred = crdup(oldcred); 561 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 562 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 563 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 564 } 565 } 566 /* The new credentials are installed into the process later. */ 567 568 /* 569 * Do the best to calculate the full path to the image file. 570 */ 571 if (args->fname != NULL && args->fname[0] == '/') 572 imgp->execpath = args->fname; 573 else { 574 VOP_UNLOCK(imgp->vp); 575 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0) 576 imgp->execpath = args->fname; 577 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 578 } 579 580 /* 581 * If the current process has a special image activator it 582 * wants to try first, call it. For example, emulating shell 583 * scripts differently. 584 */ 585 error = -1; 586 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 587 error = img_first(imgp); 588 589 /* 590 * Loop through the list of image activators, calling each one. 591 * An activator returns -1 if there is no match, 0 on success, 592 * and an error otherwise. 593 */ 594 for (i = 0; error == -1 && execsw[i]; ++i) { 595 if (execsw[i]->ex_imgact == NULL || 596 execsw[i]->ex_imgact == img_first) { 597 continue; 598 } 599 error = (*execsw[i]->ex_imgact)(imgp); 600 } 601 602 if (error) { 603 if (error == -1) 604 error = ENOEXEC; 605 goto exec_fail_dealloc; 606 } 607 608 /* 609 * Special interpreter operation, cleanup and loop up to try to 610 * activate the interpreter. 611 */ 612 if (imgp->interpreted) { 613 exec_unmap_first_page(imgp); 614 /* 615 * The text reference needs to be removed for scripts. 616 * There is a short period before we determine that 617 * something is a script where text reference is active. 618 * The vnode lock is held over this entire period 619 * so nothing should illegitimately be blocked. 620 */ 621 MPASS(imgp->textset); 622 VOP_UNSET_TEXT_CHECKED(newtextvp); 623 imgp->textset = false; 624 /* free name buffer and old vnode */ 625 if (args->fname != NULL) 626 NDFREE(&nd, NDF_ONLY_PNBUF); 627 #ifdef MAC 628 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 629 #endif 630 if (imgp->opened) { 631 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 632 imgp->opened = 0; 633 } 634 vput(newtextvp); 635 vm_object_deallocate(imgp->object); 636 imgp->object = NULL; 637 imgp->credential_setid = false; 638 if (imgp->newcred != NULL) { 639 crfree(imgp->newcred); 640 imgp->newcred = NULL; 641 } 642 imgp->execpath = NULL; 643 free(imgp->freepath, M_TEMP); 644 imgp->freepath = NULL; 645 /* set new name to that of the interpreter */ 646 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 647 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 648 args->fname = imgp->interpreter_name; 649 goto interpret; 650 } 651 652 /* 653 * NB: We unlock the vnode here because it is believed that none 654 * of the sv_copyout_strings/sv_fixup operations require the vnode. 655 */ 656 VOP_UNLOCK(imgp->vp); 657 658 if (disallow_high_osrel && 659 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 660 error = ENOEXEC; 661 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 662 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 663 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 664 goto exec_fail_dealloc; 665 } 666 667 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 668 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 669 sys_cap_enter(td, NULL); 670 671 /* 672 * Copy out strings (args and env) and initialize stack base. 673 */ 674 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 675 if (error != 0) { 676 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 677 goto exec_fail_dealloc; 678 } 679 680 /* 681 * Stack setup. 682 */ 683 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 684 if (error != 0) { 685 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 686 goto exec_fail_dealloc; 687 } 688 689 if (args->fdp != NULL) { 690 /* Install a brand new file descriptor table. */ 691 fdinstall_remapped(td, args->fdp); 692 args->fdp = NULL; 693 } else { 694 /* 695 * Keep on using the existing file descriptor table. For 696 * security and other reasons, the file descriptor table 697 * cannot be shared after an exec. 698 */ 699 fdunshare(td); 700 /* close files on exec */ 701 fdcloseexec(td); 702 } 703 704 /* 705 * Malloc things before we need locks. 706 */ 707 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 708 /* Cache arguments if they fit inside our allowance */ 709 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 710 newargs = pargs_alloc(i); 711 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 712 } 713 714 /* 715 * For security and other reasons, signal handlers cannot 716 * be shared after an exec. The new process gets a copy of the old 717 * handlers. In execsigs(), the new process will have its signals 718 * reset. 719 */ 720 if (sigacts_shared(p->p_sigacts)) { 721 oldsigacts = p->p_sigacts; 722 newsigacts = sigacts_alloc(); 723 sigacts_copy(newsigacts, oldsigacts); 724 } 725 726 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 727 728 PROC_LOCK(p); 729 if (oldsigacts) 730 p->p_sigacts = newsigacts; 731 /* Stop profiling */ 732 stopprofclock(p); 733 734 /* reset caught signals */ 735 execsigs(p); 736 737 /* name this process - nameiexec(p, ndp) */ 738 bzero(p->p_comm, sizeof(p->p_comm)); 739 if (args->fname) 740 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 741 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 742 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 743 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 744 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 745 #ifdef KTR 746 sched_clear_tdname(td); 747 #endif 748 749 /* 750 * mark as execed, wakeup the process that vforked (if any) and tell 751 * it that it now has its own resources back 752 */ 753 p->p_flag |= P_EXEC; 754 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 755 p->p_flag2 &= ~P2_NOTRACE; 756 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 757 p->p_flag2 &= ~P2_STKGAP_DISABLE; 758 if (p->p_flag & P_PPWAIT) { 759 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 760 cv_broadcast(&p->p_pwait); 761 /* STOPs are no longer ignored, arrange for AST */ 762 signotify(td); 763 } 764 765 /* 766 * Implement image setuid/setgid installation. 767 */ 768 if (imgp->credential_setid) { 769 /* 770 * Turn off syscall tracing for set-id programs, except for 771 * root. Record any set-id flags first to make sure that 772 * we do not regain any tracing during a possible block. 773 */ 774 setsugid(p); 775 776 #ifdef KTRACE 777 if (p->p_tracecred != NULL && 778 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED)) 779 ktrprocexec(p, &tracecred, &tracevp); 780 #endif 781 /* 782 * Close any file descriptors 0..2 that reference procfs, 783 * then make sure file descriptors 0..2 are in use. 784 * 785 * Both fdsetugidsafety() and fdcheckstd() may call functions 786 * taking sleepable locks, so temporarily drop our locks. 787 */ 788 PROC_UNLOCK(p); 789 VOP_UNLOCK(imgp->vp); 790 fdsetugidsafety(td); 791 error = fdcheckstd(td); 792 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 793 if (error != 0) 794 goto exec_fail_dealloc; 795 PROC_LOCK(p); 796 #ifdef MAC 797 if (will_transition) { 798 mac_vnode_execve_transition(oldcred, imgp->newcred, 799 imgp->vp, interpvplabel, imgp); 800 } 801 #endif 802 } else { 803 if (oldcred->cr_uid == oldcred->cr_ruid && 804 oldcred->cr_gid == oldcred->cr_rgid) 805 p->p_flag &= ~P_SUGID; 806 } 807 /* 808 * Set the new credentials. 809 */ 810 if (imgp->newcred != NULL) { 811 proc_set_cred(p, imgp->newcred); 812 crfree(oldcred); 813 oldcred = NULL; 814 } 815 816 /* 817 * Store the vp for use in procfs. This vnode was referenced by namei 818 * or fgetvp_exec. 819 */ 820 oldtextvp = p->p_textvp; 821 p->p_textvp = newtextvp; 822 823 #ifdef KDTRACE_HOOKS 824 /* 825 * Tell the DTrace fasttrap provider about the exec if it 826 * has declared an interest. 827 */ 828 if (dtrace_fasttrap_exec) 829 dtrace_fasttrap_exec(p); 830 #endif 831 832 /* 833 * Notify others that we exec'd, and clear the P_INEXEC flag 834 * as we're now a bona fide freshly-execed process. 835 */ 836 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 837 p->p_flag &= ~P_INEXEC; 838 839 /* clear "fork but no exec" flag, as we _are_ execing */ 840 p->p_acflag &= ~AFORK; 841 842 /* 843 * Free any previous argument cache and replace it with 844 * the new argument cache, if any. 845 */ 846 oldargs = p->p_args; 847 p->p_args = newargs; 848 newargs = NULL; 849 850 PROC_UNLOCK(p); 851 852 #ifdef HWPMC_HOOKS 853 /* 854 * Check if system-wide sampling is in effect or if the 855 * current process is using PMCs. If so, do exec() time 856 * processing. This processing needs to happen AFTER the 857 * P_INEXEC flag is cleared. 858 */ 859 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 860 VOP_UNLOCK(imgp->vp); 861 pe.pm_credentialschanged = credential_changing; 862 pe.pm_entryaddr = imgp->entry_addr; 863 864 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 865 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 866 } 867 #endif 868 869 /* Set values passed into the program in registers. */ 870 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 871 872 VOP_MMAPPED(imgp->vp); 873 874 SDT_PROBE1(proc, , , exec__success, args->fname); 875 876 exec_fail_dealloc: 877 if (error != 0) { 878 p->p_osrel = orig_osrel; 879 p->p_fctl0 = orig_fctl0; 880 } 881 882 if (imgp->firstpage != NULL) 883 exec_unmap_first_page(imgp); 884 885 if (imgp->vp != NULL) { 886 if (args->fname) 887 NDFREE(&nd, NDF_ONLY_PNBUF); 888 if (imgp->opened) 889 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 890 if (imgp->textset) 891 VOP_UNSET_TEXT_CHECKED(imgp->vp); 892 if (error != 0) 893 vput(imgp->vp); 894 else 895 VOP_UNLOCK(imgp->vp); 896 } 897 898 if (imgp->object != NULL) 899 vm_object_deallocate(imgp->object); 900 901 free(imgp->freepath, M_TEMP); 902 903 if (error == 0) { 904 if (p->p_ptevents & PTRACE_EXEC) { 905 PROC_LOCK(p); 906 if (p->p_ptevents & PTRACE_EXEC) 907 td->td_dbgflags |= TDB_EXEC; 908 PROC_UNLOCK(p); 909 } 910 } else { 911 exec_fail: 912 /* we're done here, clear P_INEXEC */ 913 PROC_LOCK(p); 914 p->p_flag &= ~P_INEXEC; 915 PROC_UNLOCK(p); 916 917 SDT_PROBE1(proc, , , exec__failure, error); 918 } 919 920 if (imgp->newcred != NULL && oldcred != NULL) 921 crfree(imgp->newcred); 922 923 #ifdef MAC 924 mac_execve_exit(imgp); 925 mac_execve_interpreter_exit(interpvplabel); 926 #endif 927 exec_free_args(args); 928 929 /* 930 * Handle deferred decrement of ref counts. 931 */ 932 if (oldtextvp != NULL) 933 vrele(oldtextvp); 934 #ifdef KTRACE 935 if (tracevp != NULL) 936 vrele(tracevp); 937 if (tracecred != NULL) 938 crfree(tracecred); 939 #endif 940 pargs_drop(oldargs); 941 pargs_drop(newargs); 942 if (oldsigacts != NULL) 943 sigacts_free(oldsigacts); 944 if (euip != NULL) 945 uifree(euip); 946 947 if (error && imgp->vmspace_destroyed) { 948 /* sorry, no more process anymore. exit gracefully */ 949 exec_cleanup(td, oldvmspace); 950 exit1(td, 0, SIGABRT); 951 /* NOT REACHED */ 952 } 953 954 #ifdef KTRACE 955 if (error == 0) 956 ktrprocctor(p); 957 #endif 958 959 /* 960 * We don't want cpu_set_syscall_retval() to overwrite any of 961 * the register values put in place by exec_setregs(). 962 * Implementations of cpu_set_syscall_retval() will leave 963 * registers unmodified when returning EJUSTRETURN. 964 */ 965 return (error == 0 ? EJUSTRETURN : error); 966 } 967 968 void 969 exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 970 { 971 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 972 KASSERT(td->td_proc->p_vmspace != oldvmspace, 973 ("oldvmspace still used")); 974 vmspace_free(oldvmspace); 975 td->td_pflags &= ~TDP_EXECVMSPC; 976 } 977 } 978 979 int 980 exec_map_first_page(struct image_params *imgp) 981 { 982 vm_object_t object; 983 vm_page_t m; 984 int error; 985 986 if (imgp->firstpage != NULL) 987 exec_unmap_first_page(imgp); 988 989 object = imgp->vp->v_object; 990 if (object == NULL) 991 return (EACCES); 992 #if VM_NRESERVLEVEL > 0 993 if ((object->flags & OBJ_COLORED) == 0) { 994 VM_OBJECT_WLOCK(object); 995 vm_object_color(object, 0); 996 VM_OBJECT_WUNLOCK(object); 997 } 998 #endif 999 error = vm_page_grab_valid_unlocked(&m, object, 0, 1000 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1001 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1002 1003 if (error != VM_PAGER_OK) 1004 return (EIO); 1005 imgp->firstpage = sf_buf_alloc(m, 0); 1006 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1007 1008 return (0); 1009 } 1010 1011 void 1012 exec_unmap_first_page(struct image_params *imgp) 1013 { 1014 vm_page_t m; 1015 1016 if (imgp->firstpage != NULL) { 1017 m = sf_buf_page(imgp->firstpage); 1018 sf_buf_free(imgp->firstpage); 1019 imgp->firstpage = NULL; 1020 vm_page_unwire(m, PQ_ACTIVE); 1021 } 1022 } 1023 1024 /* 1025 * Destroy old address space, and allocate a new stack. 1026 * The new stack is only sgrowsiz large because it is grown 1027 * automatically on a page fault. 1028 */ 1029 int 1030 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1031 { 1032 int error; 1033 struct proc *p = imgp->proc; 1034 struct vmspace *vmspace = p->p_vmspace; 1035 struct thread *td = curthread; 1036 vm_object_t obj; 1037 struct rlimit rlim_stack; 1038 vm_offset_t sv_minuser, stack_addr; 1039 vm_map_t map; 1040 u_long ssiz; 1041 1042 imgp->vmspace_destroyed = 1; 1043 imgp->sysent = sv; 1044 1045 sigfastblock_clear(td); 1046 1047 /* May be called with Giant held */ 1048 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1049 1050 /* 1051 * Blow away entire process VM, if address space not shared, 1052 * otherwise, create a new VM space so that other threads are 1053 * not disrupted 1054 */ 1055 map = &vmspace->vm_map; 1056 if (map_at_zero) 1057 sv_minuser = sv->sv_minuser; 1058 else 1059 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1060 if (vmspace->vm_refcnt == 1 && vm_map_min(map) == sv_minuser && 1061 vm_map_max(map) == sv->sv_maxuser && 1062 cpu_exec_vmspace_reuse(p, map)) { 1063 shmexit(vmspace); 1064 pmap_remove_pages(vmspace_pmap(vmspace)); 1065 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1066 /* 1067 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1068 * must be re-evaluated. 1069 */ 1070 vm_map_lock(map); 1071 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1072 MAP_ASLR_IGNSTART); 1073 vm_map_unlock(map); 1074 } else { 1075 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1076 if (error) 1077 return (error); 1078 vmspace = p->p_vmspace; 1079 map = &vmspace->vm_map; 1080 } 1081 map->flags |= imgp->map_flags; 1082 1083 /* Map a shared page */ 1084 obj = sv->sv_shared_page_obj; 1085 if (obj != NULL) { 1086 vm_object_reference(obj); 1087 error = vm_map_fixed(map, obj, 0, 1088 sv->sv_shared_page_base, sv->sv_shared_page_len, 1089 VM_PROT_READ | VM_PROT_EXECUTE, 1090 VM_PROT_READ | VM_PROT_EXECUTE, 1091 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1092 if (error != KERN_SUCCESS) { 1093 vm_object_deallocate(obj); 1094 return (vm_mmap_to_errno(error)); 1095 } 1096 } 1097 1098 /* Allocate a new stack */ 1099 if (imgp->stack_sz != 0) { 1100 ssiz = trunc_page(imgp->stack_sz); 1101 PROC_LOCK(p); 1102 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1103 PROC_UNLOCK(p); 1104 if (ssiz > rlim_stack.rlim_max) 1105 ssiz = rlim_stack.rlim_max; 1106 if (ssiz > rlim_stack.rlim_cur) { 1107 rlim_stack.rlim_cur = ssiz; 1108 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1109 } 1110 } else if (sv->sv_maxssiz != NULL) { 1111 ssiz = *sv->sv_maxssiz; 1112 } else { 1113 ssiz = maxssiz; 1114 } 1115 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1116 if (ssiz < imgp->eff_stack_sz) 1117 imgp->eff_stack_sz = ssiz; 1118 stack_addr = sv->sv_usrstack - ssiz; 1119 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1120 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1121 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1122 if (error != KERN_SUCCESS) 1123 return (vm_mmap_to_errno(error)); 1124 1125 /* 1126 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1127 * are still used to enforce the stack rlimit on the process stack. 1128 */ 1129 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1130 vmspace->vm_maxsaddr = (char *)stack_addr; 1131 1132 return (0); 1133 } 1134 1135 /* 1136 * Copy out argument and environment strings from the old process address 1137 * space into the temporary string buffer. 1138 */ 1139 int 1140 exec_copyin_args(struct image_args *args, const char *fname, 1141 enum uio_seg segflg, char **argv, char **envv) 1142 { 1143 u_long arg, env; 1144 int error; 1145 1146 bzero(args, sizeof(*args)); 1147 if (argv == NULL) 1148 return (EFAULT); 1149 1150 /* 1151 * Allocate demand-paged memory for the file name, argument, and 1152 * environment strings. 1153 */ 1154 error = exec_alloc_args(args); 1155 if (error != 0) 1156 return (error); 1157 1158 /* 1159 * Copy the file name. 1160 */ 1161 error = exec_args_add_fname(args, fname, segflg); 1162 if (error != 0) 1163 goto err_exit; 1164 1165 /* 1166 * extract arguments first 1167 */ 1168 for (;;) { 1169 error = fueword(argv++, &arg); 1170 if (error == -1) { 1171 error = EFAULT; 1172 goto err_exit; 1173 } 1174 if (arg == 0) 1175 break; 1176 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1177 UIO_USERSPACE); 1178 if (error != 0) 1179 goto err_exit; 1180 } 1181 1182 /* 1183 * extract environment strings 1184 */ 1185 if (envv) { 1186 for (;;) { 1187 error = fueword(envv++, &env); 1188 if (error == -1) { 1189 error = EFAULT; 1190 goto err_exit; 1191 } 1192 if (env == 0) 1193 break; 1194 error = exec_args_add_env(args, 1195 (char *)(uintptr_t)env, UIO_USERSPACE); 1196 if (error != 0) 1197 goto err_exit; 1198 } 1199 } 1200 1201 return (0); 1202 1203 err_exit: 1204 exec_free_args(args); 1205 return (error); 1206 } 1207 1208 int 1209 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1210 const void *data, size_t datalen, const int *fds, size_t fdslen) 1211 { 1212 struct filedesc *ofdp; 1213 const char *p; 1214 int *kfds; 1215 int error; 1216 1217 memset(args, '\0', sizeof(*args)); 1218 ofdp = td->td_proc->p_fd; 1219 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles) 1220 return (E2BIG); 1221 error = exec_alloc_args(args); 1222 if (error != 0) 1223 return (error); 1224 1225 args->begin_argv = args->buf; 1226 args->stringspace = ARG_MAX; 1227 1228 if (datalen > 0) { 1229 /* 1230 * Argument buffer has been provided. Copy it into the 1231 * kernel as a single string and add a terminating null 1232 * byte. 1233 */ 1234 error = copyin(data, args->begin_argv, datalen); 1235 if (error != 0) 1236 goto err_exit; 1237 args->begin_argv[datalen] = '\0'; 1238 args->endp = args->begin_argv + datalen + 1; 1239 args->stringspace -= datalen + 1; 1240 1241 /* 1242 * Traditional argument counting. Count the number of 1243 * null bytes. 1244 */ 1245 for (p = args->begin_argv; p < args->endp; ++p) 1246 if (*p == '\0') 1247 ++args->argc; 1248 } else { 1249 /* No argument buffer provided. */ 1250 args->endp = args->begin_argv; 1251 } 1252 1253 /* Create new file descriptor table. */ 1254 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1255 error = copyin(fds, kfds, fdslen * sizeof(int)); 1256 if (error != 0) { 1257 free(kfds, M_TEMP); 1258 goto err_exit; 1259 } 1260 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1261 free(kfds, M_TEMP); 1262 if (error != 0) 1263 goto err_exit; 1264 1265 return (0); 1266 err_exit: 1267 exec_free_args(args); 1268 return (error); 1269 } 1270 1271 struct exec_args_kva { 1272 vm_offset_t addr; 1273 u_int gen; 1274 SLIST_ENTRY(exec_args_kva) next; 1275 }; 1276 1277 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1278 1279 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1280 static struct mtx exec_args_kva_mtx; 1281 static u_int exec_args_gen; 1282 1283 static void 1284 exec_prealloc_args_kva(void *arg __unused) 1285 { 1286 struct exec_args_kva *argkva; 1287 u_int i; 1288 1289 SLIST_INIT(&exec_args_kva_freelist); 1290 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1291 for (i = 0; i < exec_map_entries; i++) { 1292 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1293 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1294 argkva->gen = exec_args_gen; 1295 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1296 } 1297 } 1298 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1299 1300 static vm_offset_t 1301 exec_alloc_args_kva(void **cookie) 1302 { 1303 struct exec_args_kva *argkva; 1304 1305 argkva = (void *)atomic_readandclear_ptr( 1306 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1307 if (argkva == NULL) { 1308 mtx_lock(&exec_args_kva_mtx); 1309 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1310 (void)mtx_sleep(&exec_args_kva_freelist, 1311 &exec_args_kva_mtx, 0, "execkva", 0); 1312 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1313 mtx_unlock(&exec_args_kva_mtx); 1314 } 1315 *(struct exec_args_kva **)cookie = argkva; 1316 return (argkva->addr); 1317 } 1318 1319 static void 1320 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1321 { 1322 vm_offset_t base; 1323 1324 base = argkva->addr; 1325 if (argkva->gen != gen) { 1326 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1327 MADV_FREE); 1328 argkva->gen = gen; 1329 } 1330 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1331 (uintptr_t)NULL, (uintptr_t)argkva)) { 1332 mtx_lock(&exec_args_kva_mtx); 1333 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1334 wakeup_one(&exec_args_kva_freelist); 1335 mtx_unlock(&exec_args_kva_mtx); 1336 } 1337 } 1338 1339 static void 1340 exec_free_args_kva(void *cookie) 1341 { 1342 1343 exec_release_args_kva(cookie, exec_args_gen); 1344 } 1345 1346 static void 1347 exec_args_kva_lowmem(void *arg __unused) 1348 { 1349 SLIST_HEAD(, exec_args_kva) head; 1350 struct exec_args_kva *argkva; 1351 u_int gen; 1352 int i; 1353 1354 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1355 1356 /* 1357 * Force an madvise of each KVA range. Any currently allocated ranges 1358 * will have MADV_FREE applied once they are freed. 1359 */ 1360 SLIST_INIT(&head); 1361 mtx_lock(&exec_args_kva_mtx); 1362 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1363 mtx_unlock(&exec_args_kva_mtx); 1364 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1365 SLIST_REMOVE_HEAD(&head, next); 1366 exec_release_args_kva(argkva, gen); 1367 } 1368 1369 CPU_FOREACH(i) { 1370 argkva = (void *)atomic_readandclear_ptr( 1371 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1372 if (argkva != NULL) 1373 exec_release_args_kva(argkva, gen); 1374 } 1375 } 1376 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1377 EVENTHANDLER_PRI_ANY); 1378 1379 /* 1380 * Allocate temporary demand-paged, zero-filled memory for the file name, 1381 * argument, and environment strings. 1382 */ 1383 int 1384 exec_alloc_args(struct image_args *args) 1385 { 1386 1387 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1388 return (0); 1389 } 1390 1391 void 1392 exec_free_args(struct image_args *args) 1393 { 1394 1395 if (args->buf != NULL) { 1396 exec_free_args_kva(args->bufkva); 1397 args->buf = NULL; 1398 } 1399 if (args->fname_buf != NULL) { 1400 free(args->fname_buf, M_TEMP); 1401 args->fname_buf = NULL; 1402 } 1403 if (args->fdp != NULL) 1404 fdescfree_remapped(args->fdp); 1405 } 1406 1407 /* 1408 * A set to functions to fill struct image args. 1409 * 1410 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1411 * fname) before the other functions. All exec_args_add_arg() calls must 1412 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1413 * may be called any time after exec_args_add_fname(). 1414 * 1415 * exec_args_add_fname() - install path to be executed 1416 * exec_args_add_arg() - append an argument string 1417 * exec_args_add_env() - append an env string 1418 * exec_args_adjust_args() - adjust location of the argument list to 1419 * allow new arguments to be prepended 1420 */ 1421 int 1422 exec_args_add_fname(struct image_args *args, const char *fname, 1423 enum uio_seg segflg) 1424 { 1425 int error; 1426 size_t length; 1427 1428 KASSERT(args->fname == NULL, ("fname already appended")); 1429 KASSERT(args->endp == NULL, ("already appending to args")); 1430 1431 if (fname != NULL) { 1432 args->fname = args->buf; 1433 error = segflg == UIO_SYSSPACE ? 1434 copystr(fname, args->fname, PATH_MAX, &length) : 1435 copyinstr(fname, args->fname, PATH_MAX, &length); 1436 if (error != 0) 1437 return (error == ENAMETOOLONG ? E2BIG : error); 1438 } else 1439 length = 0; 1440 1441 /* Set up for _arg_*()/_env_*() */ 1442 args->endp = args->buf + length; 1443 /* begin_argv must be set and kept updated */ 1444 args->begin_argv = args->endp; 1445 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1446 ("too little space remaining for arguments %zu < %zu", 1447 exec_map_entry_size - length, (size_t)ARG_MAX)); 1448 args->stringspace = ARG_MAX; 1449 1450 return (0); 1451 } 1452 1453 static int 1454 exec_args_add_str(struct image_args *args, const char *str, 1455 enum uio_seg segflg, int *countp) 1456 { 1457 int error; 1458 size_t length; 1459 1460 KASSERT(args->endp != NULL, ("endp not initialized")); 1461 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1462 1463 error = (segflg == UIO_SYSSPACE) ? 1464 copystr(str, args->endp, args->stringspace, &length) : 1465 copyinstr(str, args->endp, args->stringspace, &length); 1466 if (error != 0) 1467 return (error == ENAMETOOLONG ? E2BIG : error); 1468 args->stringspace -= length; 1469 args->endp += length; 1470 (*countp)++; 1471 1472 return (0); 1473 } 1474 1475 int 1476 exec_args_add_arg(struct image_args *args, const char *argp, 1477 enum uio_seg segflg) 1478 { 1479 1480 KASSERT(args->envc == 0, ("appending args after env")); 1481 1482 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1483 } 1484 1485 int 1486 exec_args_add_env(struct image_args *args, const char *envp, 1487 enum uio_seg segflg) 1488 { 1489 1490 if (args->envc == 0) 1491 args->begin_envv = args->endp; 1492 1493 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1494 } 1495 1496 int 1497 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1498 { 1499 ssize_t offset; 1500 1501 KASSERT(args->endp != NULL, ("endp not initialized")); 1502 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1503 1504 offset = extend - consume; 1505 if (args->stringspace < offset) 1506 return (E2BIG); 1507 memmove(args->begin_argv + extend, args->begin_argv + consume, 1508 args->endp - args->begin_argv + consume); 1509 if (args->envc > 0) 1510 args->begin_envv += offset; 1511 args->endp += offset; 1512 args->stringspace -= offset; 1513 return (0); 1514 } 1515 1516 char * 1517 exec_args_get_begin_envv(struct image_args *args) 1518 { 1519 1520 KASSERT(args->endp != NULL, ("endp not initialized")); 1521 1522 if (args->envc > 0) 1523 return (args->begin_envv); 1524 return (args->endp); 1525 } 1526 1527 /* 1528 * Copy strings out to the new process address space, constructing new arg 1529 * and env vector tables. Return a pointer to the base so that it can be used 1530 * as the initial stack pointer. 1531 */ 1532 int 1533 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1534 { 1535 int argc, envc; 1536 char **vectp; 1537 char *stringp; 1538 uintptr_t destp, ustringp; 1539 struct ps_strings *arginfo; 1540 struct proc *p; 1541 size_t execpath_len; 1542 int error, szsigcode, szps; 1543 char canary[sizeof(long) * 8]; 1544 1545 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1546 /* 1547 * Calculate string base and vector table pointers. 1548 * Also deal with signal trampoline code for this exec type. 1549 */ 1550 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1551 execpath_len = strlen(imgp->execpath) + 1; 1552 else 1553 execpath_len = 0; 1554 p = imgp->proc; 1555 szsigcode = 0; 1556 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1557 imgp->ps_strings = arginfo; 1558 if (p->p_sysent->sv_sigcode_base == 0) { 1559 if (p->p_sysent->sv_szsigcode != NULL) 1560 szsigcode = *(p->p_sysent->sv_szsigcode); 1561 } 1562 destp = (uintptr_t)arginfo; 1563 1564 /* 1565 * install sigcode 1566 */ 1567 if (szsigcode != 0) { 1568 destp -= szsigcode; 1569 destp = rounddown2(destp, sizeof(void *)); 1570 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1571 szsigcode); 1572 if (error != 0) 1573 return (error); 1574 } 1575 1576 /* 1577 * Copy the image path for the rtld. 1578 */ 1579 if (execpath_len != 0) { 1580 destp -= execpath_len; 1581 destp = rounddown2(destp, sizeof(void *)); 1582 imgp->execpathp = (void *)destp; 1583 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1584 if (error != 0) 1585 return (error); 1586 } 1587 1588 /* 1589 * Prepare the canary for SSP. 1590 */ 1591 arc4rand(canary, sizeof(canary), 0); 1592 destp -= sizeof(canary); 1593 imgp->canary = (void *)destp; 1594 error = copyout(canary, imgp->canary, sizeof(canary)); 1595 if (error != 0) 1596 return (error); 1597 imgp->canarylen = sizeof(canary); 1598 1599 /* 1600 * Prepare the pagesizes array. 1601 */ 1602 destp -= szps; 1603 destp = rounddown2(destp, sizeof(void *)); 1604 imgp->pagesizes = (void *)destp; 1605 error = copyout(pagesizes, imgp->pagesizes, szps); 1606 if (error != 0) 1607 return (error); 1608 imgp->pagesizeslen = szps; 1609 1610 /* 1611 * Allocate room for the argument and environment strings. 1612 */ 1613 destp -= ARG_MAX - imgp->args->stringspace; 1614 destp = rounddown2(destp, sizeof(void *)); 1615 ustringp = destp; 1616 1617 if (imgp->sysent->sv_stackgap != NULL) 1618 imgp->sysent->sv_stackgap(imgp, &destp); 1619 1620 if (imgp->auxargs) { 1621 /* 1622 * Allocate room on the stack for the ELF auxargs 1623 * array. It has up to AT_COUNT entries. 1624 */ 1625 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1626 destp = rounddown2(destp, sizeof(void *)); 1627 } 1628 1629 vectp = (char **)destp; 1630 1631 /* 1632 * Allocate room for the argv[] and env vectors including the 1633 * terminating NULL pointers. 1634 */ 1635 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1636 1637 /* 1638 * vectp also becomes our initial stack base 1639 */ 1640 *stack_base = (uintptr_t)vectp; 1641 1642 stringp = imgp->args->begin_argv; 1643 argc = imgp->args->argc; 1644 envc = imgp->args->envc; 1645 1646 /* 1647 * Copy out strings - arguments and environment. 1648 */ 1649 error = copyout(stringp, (void *)ustringp, 1650 ARG_MAX - imgp->args->stringspace); 1651 if (error != 0) 1652 return (error); 1653 1654 /* 1655 * Fill in "ps_strings" struct for ps, w, etc. 1656 */ 1657 imgp->argv = vectp; 1658 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1659 suword32(&arginfo->ps_nargvstr, argc) != 0) 1660 return (EFAULT); 1661 1662 /* 1663 * Fill in argument portion of vector table. 1664 */ 1665 for (; argc > 0; --argc) { 1666 if (suword(vectp++, ustringp) != 0) 1667 return (EFAULT); 1668 while (*stringp++ != 0) 1669 ustringp++; 1670 ustringp++; 1671 } 1672 1673 /* a null vector table pointer separates the argp's from the envp's */ 1674 if (suword(vectp++, 0) != 0) 1675 return (EFAULT); 1676 1677 imgp->envv = vectp; 1678 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1679 suword32(&arginfo->ps_nenvstr, envc) != 0) 1680 return (EFAULT); 1681 1682 /* 1683 * Fill in environment portion of vector table. 1684 */ 1685 for (; envc > 0; --envc) { 1686 if (suword(vectp++, ustringp) != 0) 1687 return (EFAULT); 1688 while (*stringp++ != 0) 1689 ustringp++; 1690 ustringp++; 1691 } 1692 1693 /* end of vector table is a null pointer */ 1694 if (suword(vectp, 0) != 0) 1695 return (EFAULT); 1696 1697 if (imgp->auxargs) { 1698 vectp++; 1699 error = imgp->sysent->sv_copyout_auxargs(imgp, 1700 (uintptr_t)vectp); 1701 if (error != 0) 1702 return (error); 1703 } 1704 1705 return (0); 1706 } 1707 1708 /* 1709 * Check permissions of file to execute. 1710 * Called with imgp->vp locked. 1711 * Return 0 for success or error code on failure. 1712 */ 1713 int 1714 exec_check_permissions(struct image_params *imgp) 1715 { 1716 struct vnode *vp = imgp->vp; 1717 struct vattr *attr = imgp->attr; 1718 struct thread *td; 1719 int error; 1720 1721 td = curthread; 1722 1723 /* Get file attributes */ 1724 error = VOP_GETATTR(vp, attr, td->td_ucred); 1725 if (error) 1726 return (error); 1727 1728 #ifdef MAC 1729 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1730 if (error) 1731 return (error); 1732 #endif 1733 1734 /* 1735 * 1) Check if file execution is disabled for the filesystem that 1736 * this file resides on. 1737 * 2) Ensure that at least one execute bit is on. Otherwise, a 1738 * privileged user will always succeed, and we don't want this 1739 * to happen unless the file really is executable. 1740 * 3) Ensure that the file is a regular file. 1741 */ 1742 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1743 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1744 (attr->va_type != VREG)) 1745 return (EACCES); 1746 1747 /* 1748 * Zero length files can't be exec'd 1749 */ 1750 if (attr->va_size == 0) 1751 return (ENOEXEC); 1752 1753 /* 1754 * Check for execute permission to file based on current credentials. 1755 */ 1756 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1757 if (error) 1758 return (error); 1759 1760 /* 1761 * Check number of open-for-writes on the file and deny execution 1762 * if there are any. 1763 * 1764 * Add a text reference now so no one can write to the 1765 * executable while we're activating it. 1766 * 1767 * Remember if this was set before and unset it in case this is not 1768 * actually an executable image. 1769 */ 1770 error = VOP_SET_TEXT(vp); 1771 if (error != 0) 1772 return (error); 1773 imgp->textset = true; 1774 1775 /* 1776 * Call filesystem specific open routine (which does nothing in the 1777 * general case). 1778 */ 1779 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1780 if (error == 0) 1781 imgp->opened = 1; 1782 return (error); 1783 } 1784 1785 /* 1786 * Exec handler registration 1787 */ 1788 int 1789 exec_register(const struct execsw *execsw_arg) 1790 { 1791 const struct execsw **es, **xs, **newexecsw; 1792 u_int count = 2; /* New slot and trailing NULL */ 1793 1794 if (execsw) 1795 for (es = execsw; *es; es++) 1796 count++; 1797 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1798 xs = newexecsw; 1799 if (execsw) 1800 for (es = execsw; *es; es++) 1801 *xs++ = *es; 1802 *xs++ = execsw_arg; 1803 *xs = NULL; 1804 if (execsw) 1805 free(execsw, M_TEMP); 1806 execsw = newexecsw; 1807 return (0); 1808 } 1809 1810 int 1811 exec_unregister(const struct execsw *execsw_arg) 1812 { 1813 const struct execsw **es, **xs, **newexecsw; 1814 int count = 1; 1815 1816 if (execsw == NULL) 1817 panic("unregister with no handlers left?\n"); 1818 1819 for (es = execsw; *es; es++) { 1820 if (*es == execsw_arg) 1821 break; 1822 } 1823 if (*es == NULL) 1824 return (ENOENT); 1825 for (es = execsw; *es; es++) 1826 if (*es != execsw_arg) 1827 count++; 1828 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1829 xs = newexecsw; 1830 for (es = execsw; *es; es++) 1831 if (*es != execsw_arg) 1832 *xs++ = *es; 1833 *xs = NULL; 1834 if (execsw) 1835 free(execsw, M_TEMP); 1836 execsw = newexecsw; 1837 return (0); 1838 } 1839