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 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 229 return (error); 230 } 231 232 #ifndef _SYS_SYSPROTO_H_ 233 struct fexecve_args { 234 int fd; 235 char **argv; 236 char **envv; 237 }; 238 #endif 239 int 240 sys_fexecve(struct thread *td, struct fexecve_args *uap) 241 { 242 struct image_args args; 243 struct vmspace *oldvmspace; 244 int error; 245 246 error = pre_execve(td, &oldvmspace); 247 if (error != 0) 248 return (error); 249 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 250 uap->argv, uap->envv); 251 if (error == 0) { 252 args.fd = uap->fd; 253 error = kern_execve(td, &args, NULL, oldvmspace); 254 } 255 post_execve(td, error, oldvmspace); 256 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 257 return (error); 258 } 259 260 #ifndef _SYS_SYSPROTO_H_ 261 struct __mac_execve_args { 262 char *fname; 263 char **argv; 264 char **envv; 265 struct mac *mac_p; 266 }; 267 #endif 268 269 int 270 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 271 { 272 #ifdef MAC 273 struct image_args args; 274 struct vmspace *oldvmspace; 275 int error; 276 277 error = pre_execve(td, &oldvmspace); 278 if (error != 0) 279 return (error); 280 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 281 uap->argv, uap->envv); 282 if (error == 0) 283 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 284 post_execve(td, error, oldvmspace); 285 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 286 return (error); 287 #else 288 return (ENOSYS); 289 #endif 290 } 291 292 int 293 pre_execve(struct thread *td, struct vmspace **oldvmspace) 294 { 295 struct proc *p; 296 int error; 297 298 KASSERT(td == curthread, ("non-current thread %p", td)); 299 error = 0; 300 p = td->td_proc; 301 if ((p->p_flag & P_HADTHREADS) != 0) { 302 PROC_LOCK(p); 303 if (thread_single(p, SINGLE_BOUNDARY) != 0) 304 error = ERESTART; 305 PROC_UNLOCK(p); 306 } 307 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 308 ("nested execve")); 309 *oldvmspace = p->p_vmspace; 310 return (error); 311 } 312 313 void 314 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 315 { 316 struct proc *p; 317 318 KASSERT(td == curthread, ("non-current thread %p", td)); 319 p = td->td_proc; 320 if ((p->p_flag & P_HADTHREADS) != 0) { 321 PROC_LOCK(p); 322 /* 323 * If success, we upgrade to SINGLE_EXIT state to 324 * force other threads to suicide. 325 */ 326 if (error == EJUSTRETURN) 327 thread_single(p, SINGLE_EXIT); 328 else 329 thread_single_end(p, SINGLE_BOUNDARY); 330 PROC_UNLOCK(p); 331 } 332 exec_cleanup(td, oldvmspace); 333 } 334 335 /* 336 * kern_execve() has the astonishing property of not always returning to 337 * the caller. If sufficiently bad things happen during the call to 338 * do_execve(), it can end up calling exit1(); as a result, callers must 339 * avoid doing anything which they might need to undo (e.g., allocating 340 * memory). 341 */ 342 int 343 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 344 struct vmspace *oldvmspace) 345 { 346 347 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 348 exec_args_get_begin_envv(args) - args->begin_argv); 349 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 350 args->endp - exec_args_get_begin_envv(args)); 351 return (do_execve(td, args, mac_p, oldvmspace)); 352 } 353 354 /* 355 * In-kernel implementation of execve(). All arguments are assumed to be 356 * userspace pointers from the passed thread. 357 */ 358 static int 359 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 360 struct vmspace *oldvmspace) 361 { 362 struct proc *p = td->td_proc; 363 struct nameidata nd; 364 struct ucred *oldcred; 365 struct uidinfo *euip = NULL; 366 uintptr_t stack_base; 367 struct image_params image_params, *imgp; 368 struct vattr attr; 369 int (*img_first)(struct image_params *); 370 struct pargs *oldargs = NULL, *newargs = NULL; 371 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 372 #ifdef KTRACE 373 struct vnode *tracevp = NULL; 374 struct ucred *tracecred = NULL; 375 #endif 376 struct vnode *oldtextvp = NULL, *newtextvp; 377 int credential_changing; 378 #ifdef MAC 379 struct label *interpvplabel = NULL; 380 int will_transition; 381 #endif 382 #ifdef HWPMC_HOOKS 383 struct pmckern_procexec pe; 384 #endif 385 int error, i, orig_osrel; 386 uint32_t orig_fctl0; 387 static const char fexecv_proc_title[] = "(fexecv)"; 388 389 imgp = &image_params; 390 391 /* 392 * Lock the process and set the P_INEXEC flag to indicate that 393 * it should be left alone until we're done here. This is 394 * necessary to avoid race conditions - e.g. in ptrace() - 395 * that might allow a local user to illicitly obtain elevated 396 * privileges. 397 */ 398 PROC_LOCK(p); 399 KASSERT((p->p_flag & P_INEXEC) == 0, 400 ("%s(): process already has P_INEXEC flag", __func__)); 401 p->p_flag |= P_INEXEC; 402 PROC_UNLOCK(p); 403 404 /* 405 * Initialize part of the common data 406 */ 407 bzero(imgp, sizeof(*imgp)); 408 imgp->proc = p; 409 imgp->attr = &attr; 410 imgp->args = args; 411 oldcred = p->p_ucred; 412 orig_osrel = p->p_osrel; 413 orig_fctl0 = p->p_fctl0; 414 415 #ifdef MAC 416 error = mac_execve_enter(imgp, mac_p); 417 if (error) 418 goto exec_fail; 419 #endif 420 421 /* 422 * Translate the file name. namei() returns a vnode pointer 423 * in ni_vp among other things. 424 * 425 * XXXAUDIT: It would be desirable to also audit the name of the 426 * interpreter if this is an interpreted binary. 427 */ 428 if (args->fname != NULL) { 429 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 430 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 431 } 432 433 SDT_PROBE1(proc, , , exec, args->fname); 434 435 interpret: 436 if (args->fname != NULL) { 437 #ifdef CAPABILITY_MODE 438 /* 439 * While capability mode can't reach this point via direct 440 * path arguments to execve(), we also don't allow 441 * interpreters to be used in capability mode (for now). 442 * Catch indirect lookups and return a permissions error. 443 */ 444 if (IN_CAPABILITY_MODE(td)) { 445 error = ECAPMODE; 446 goto exec_fail; 447 } 448 #endif 449 error = namei(&nd); 450 if (error) 451 goto exec_fail; 452 453 newtextvp = nd.ni_vp; 454 imgp->vp = newtextvp; 455 } else { 456 AUDIT_ARG_FD(args->fd); 457 /* 458 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 459 */ 460 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 461 if (error) 462 goto exec_fail; 463 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 464 AUDIT_ARG_VNODE1(newtextvp); 465 imgp->vp = newtextvp; 466 } 467 468 /* 469 * Check file permissions. Also 'opens' file and sets its vnode to 470 * text mode. 471 */ 472 error = exec_check_permissions(imgp); 473 if (error) 474 goto exec_fail_dealloc; 475 476 imgp->object = imgp->vp->v_object; 477 if (imgp->object != NULL) 478 vm_object_reference(imgp->object); 479 480 error = exec_map_first_page(imgp); 481 if (error) 482 goto exec_fail_dealloc; 483 484 imgp->proc->p_osrel = 0; 485 imgp->proc->p_fctl0 = 0; 486 487 /* 488 * Implement image setuid/setgid. 489 * 490 * Determine new credentials before attempting image activators 491 * so that it can be used by process_exec handlers to determine 492 * credential/setid changes. 493 * 494 * Don't honor setuid/setgid if the filesystem prohibits it or if 495 * the process is being traced. 496 * 497 * We disable setuid/setgid/etc in capability mode on the basis 498 * that most setugid applications are not written with that 499 * environment in mind, and will therefore almost certainly operate 500 * incorrectly. In principle there's no reason that setugid 501 * applications might not be useful in capability mode, so we may want 502 * to reconsider this conservative design choice in the future. 503 * 504 * XXXMAC: For the time being, use NOSUID to also prohibit 505 * transitions on the file system. 506 */ 507 credential_changing = 0; 508 credential_changing |= (attr.va_mode & S_ISUID) && 509 oldcred->cr_uid != attr.va_uid; 510 credential_changing |= (attr.va_mode & S_ISGID) && 511 oldcred->cr_gid != attr.va_gid; 512 #ifdef MAC 513 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 514 interpvplabel, imgp); 515 credential_changing |= will_transition; 516 #endif 517 518 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 519 if (credential_changing) 520 imgp->proc->p_pdeathsig = 0; 521 522 if (credential_changing && 523 #ifdef CAPABILITY_MODE 524 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 525 #endif 526 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 527 (p->p_flag & P_TRACED) == 0) { 528 imgp->credential_setid = true; 529 VOP_UNLOCK(imgp->vp); 530 imgp->newcred = crdup(oldcred); 531 if (attr.va_mode & S_ISUID) { 532 euip = uifind(attr.va_uid); 533 change_euid(imgp->newcred, euip); 534 } 535 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 536 if (attr.va_mode & S_ISGID) 537 change_egid(imgp->newcred, attr.va_gid); 538 /* 539 * Implement correct POSIX saved-id behavior. 540 * 541 * XXXMAC: Note that the current logic will save the 542 * uid and gid if a MAC domain transition occurs, even 543 * though maybe it shouldn't. 544 */ 545 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 546 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 547 } else { 548 /* 549 * Implement correct POSIX saved-id behavior. 550 * 551 * XXX: It's not clear that the existing behavior is 552 * POSIX-compliant. A number of sources indicate that the 553 * saved uid/gid should only be updated if the new ruid is 554 * not equal to the old ruid, or the new euid is not equal 555 * to the old euid and the new euid is not equal to the old 556 * ruid. The FreeBSD code always updates the saved uid/gid. 557 * Also, this code uses the new (replaced) euid and egid as 558 * the source, which may or may not be the right ones to use. 559 */ 560 if (oldcred->cr_svuid != oldcred->cr_uid || 561 oldcred->cr_svgid != oldcred->cr_gid) { 562 VOP_UNLOCK(imgp->vp); 563 imgp->newcred = crdup(oldcred); 564 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 565 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 566 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 567 } 568 } 569 /* The new credentials are installed into the process later. */ 570 571 /* 572 * Do the best to calculate the full path to the image file. 573 */ 574 if (args->fname != NULL && args->fname[0] == '/') 575 imgp->execpath = args->fname; 576 else { 577 VOP_UNLOCK(imgp->vp); 578 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0) 579 imgp->execpath = args->fname; 580 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 581 } 582 583 /* 584 * If the current process has a special image activator it 585 * wants to try first, call it. For example, emulating shell 586 * scripts differently. 587 */ 588 error = -1; 589 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 590 error = img_first(imgp); 591 592 /* 593 * Loop through the list of image activators, calling each one. 594 * An activator returns -1 if there is no match, 0 on success, 595 * and an error otherwise. 596 */ 597 for (i = 0; error == -1 && execsw[i]; ++i) { 598 if (execsw[i]->ex_imgact == NULL || 599 execsw[i]->ex_imgact == img_first) { 600 continue; 601 } 602 error = (*execsw[i]->ex_imgact)(imgp); 603 } 604 605 if (error) { 606 if (error == -1) 607 error = ENOEXEC; 608 goto exec_fail_dealloc; 609 } 610 611 /* 612 * Special interpreter operation, cleanup and loop up to try to 613 * activate the interpreter. 614 */ 615 if (imgp->interpreted) { 616 exec_unmap_first_page(imgp); 617 /* 618 * The text reference needs to be removed for scripts. 619 * There is a short period before we determine that 620 * something is a script where text reference is active. 621 * The vnode lock is held over this entire period 622 * so nothing should illegitimately be blocked. 623 */ 624 MPASS(imgp->textset); 625 VOP_UNSET_TEXT_CHECKED(newtextvp); 626 imgp->textset = false; 627 /* free name buffer and old vnode */ 628 if (args->fname != NULL) 629 NDFREE(&nd, NDF_ONLY_PNBUF); 630 #ifdef MAC 631 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 632 #endif 633 if (imgp->opened) { 634 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 635 imgp->opened = 0; 636 } 637 vput(newtextvp); 638 vm_object_deallocate(imgp->object); 639 imgp->object = NULL; 640 imgp->credential_setid = false; 641 if (imgp->newcred != NULL) { 642 crfree(imgp->newcred); 643 imgp->newcred = NULL; 644 } 645 imgp->execpath = NULL; 646 free(imgp->freepath, M_TEMP); 647 imgp->freepath = NULL; 648 /* set new name to that of the interpreter */ 649 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 650 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 651 args->fname = imgp->interpreter_name; 652 goto interpret; 653 } 654 655 /* 656 * NB: We unlock the vnode here because it is believed that none 657 * of the sv_copyout_strings/sv_fixup operations require the vnode. 658 */ 659 VOP_UNLOCK(imgp->vp); 660 661 if (disallow_high_osrel && 662 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 663 error = ENOEXEC; 664 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 665 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 666 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 667 goto exec_fail_dealloc; 668 } 669 670 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 671 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 672 sys_cap_enter(td, NULL); 673 674 /* 675 * Copy out strings (args and env) and initialize stack base. 676 */ 677 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 678 if (error != 0) { 679 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 680 goto exec_fail_dealloc; 681 } 682 683 /* 684 * Stack setup. 685 */ 686 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 687 if (error != 0) { 688 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 689 goto exec_fail_dealloc; 690 } 691 692 if (args->fdp != NULL) { 693 /* Install a brand new file descriptor table. */ 694 fdinstall_remapped(td, args->fdp); 695 args->fdp = NULL; 696 } else { 697 /* 698 * Keep on using the existing file descriptor table. For 699 * security and other reasons, the file descriptor table 700 * cannot be shared after an exec. 701 */ 702 fdunshare(td); 703 pdunshare(td); 704 /* close files on exec */ 705 fdcloseexec(td); 706 } 707 708 /* 709 * Malloc things before we need locks. 710 */ 711 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 712 /* Cache arguments if they fit inside our allowance */ 713 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 714 newargs = pargs_alloc(i); 715 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 716 } 717 718 /* 719 * For security and other reasons, signal handlers cannot 720 * be shared after an exec. The new process gets a copy of the old 721 * handlers. In execsigs(), the new process will have its signals 722 * reset. 723 */ 724 if (sigacts_shared(p->p_sigacts)) { 725 oldsigacts = p->p_sigacts; 726 newsigacts = sigacts_alloc(); 727 sigacts_copy(newsigacts, oldsigacts); 728 } 729 730 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 731 732 PROC_LOCK(p); 733 if (oldsigacts) 734 p->p_sigacts = newsigacts; 735 /* Stop profiling */ 736 stopprofclock(p); 737 738 /* reset caught signals */ 739 execsigs(p); 740 741 /* name this process - nameiexec(p, ndp) */ 742 bzero(p->p_comm, sizeof(p->p_comm)); 743 if (args->fname) 744 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 745 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 746 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 747 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 748 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 749 #ifdef KTR 750 sched_clear_tdname(td); 751 #endif 752 753 /* 754 * mark as execed, wakeup the process that vforked (if any) and tell 755 * it that it now has its own resources back 756 */ 757 p->p_flag |= P_EXEC; 758 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 759 p->p_flag2 &= ~P2_NOTRACE; 760 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 761 p->p_flag2 &= ~P2_STKGAP_DISABLE; 762 if (p->p_flag & P_PPWAIT) { 763 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 764 cv_broadcast(&p->p_pwait); 765 /* STOPs are no longer ignored, arrange for AST */ 766 signotify(td); 767 } 768 769 /* 770 * Implement image setuid/setgid installation. 771 */ 772 if (imgp->credential_setid) { 773 /* 774 * Turn off syscall tracing for set-id programs, except for 775 * root. Record any set-id flags first to make sure that 776 * we do not regain any tracing during a possible block. 777 */ 778 setsugid(p); 779 780 #ifdef KTRACE 781 if (p->p_tracecred != NULL && 782 priv_check_cred(p->p_tracecred, PRIV_DEBUG_DIFFCRED)) 783 ktrprocexec(p, &tracecred, &tracevp); 784 #endif 785 /* 786 * Close any file descriptors 0..2 that reference procfs, 787 * then make sure file descriptors 0..2 are in use. 788 * 789 * Both fdsetugidsafety() and fdcheckstd() may call functions 790 * taking sleepable locks, so temporarily drop our locks. 791 */ 792 PROC_UNLOCK(p); 793 VOP_UNLOCK(imgp->vp); 794 fdsetugidsafety(td); 795 error = fdcheckstd(td); 796 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 797 if (error != 0) 798 goto exec_fail_dealloc; 799 PROC_LOCK(p); 800 #ifdef MAC 801 if (will_transition) { 802 mac_vnode_execve_transition(oldcred, imgp->newcred, 803 imgp->vp, interpvplabel, imgp); 804 } 805 #endif 806 } else { 807 if (oldcred->cr_uid == oldcred->cr_ruid && 808 oldcred->cr_gid == oldcred->cr_rgid) 809 p->p_flag &= ~P_SUGID; 810 } 811 /* 812 * Set the new credentials. 813 */ 814 if (imgp->newcred != NULL) { 815 proc_set_cred(p, imgp->newcred); 816 crfree(oldcred); 817 oldcred = NULL; 818 } 819 820 /* 821 * Store the vp for use in procfs. This vnode was referenced by namei 822 * or fgetvp_exec. 823 */ 824 oldtextvp = p->p_textvp; 825 p->p_textvp = newtextvp; 826 827 #ifdef KDTRACE_HOOKS 828 /* 829 * Tell the DTrace fasttrap provider about the exec if it 830 * has declared an interest. 831 */ 832 if (dtrace_fasttrap_exec) 833 dtrace_fasttrap_exec(p); 834 #endif 835 836 /* 837 * Notify others that we exec'd, and clear the P_INEXEC flag 838 * as we're now a bona fide freshly-execed process. 839 */ 840 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 841 p->p_flag &= ~P_INEXEC; 842 843 /* clear "fork but no exec" flag, as we _are_ execing */ 844 p->p_acflag &= ~AFORK; 845 846 /* 847 * Free any previous argument cache and replace it with 848 * the new argument cache, if any. 849 */ 850 oldargs = p->p_args; 851 p->p_args = newargs; 852 newargs = NULL; 853 854 PROC_UNLOCK(p); 855 856 #ifdef HWPMC_HOOKS 857 /* 858 * Check if system-wide sampling is in effect or if the 859 * current process is using PMCs. If so, do exec() time 860 * processing. This processing needs to happen AFTER the 861 * P_INEXEC flag is cleared. 862 */ 863 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 864 VOP_UNLOCK(imgp->vp); 865 pe.pm_credentialschanged = credential_changing; 866 pe.pm_entryaddr = imgp->entry_addr; 867 868 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 869 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 870 } 871 #endif 872 873 /* Set values passed into the program in registers. */ 874 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 875 876 VOP_MMAPPED(imgp->vp); 877 878 SDT_PROBE1(proc, , , exec__success, args->fname); 879 880 exec_fail_dealloc: 881 if (error != 0) { 882 p->p_osrel = orig_osrel; 883 p->p_fctl0 = orig_fctl0; 884 } 885 886 if (imgp->firstpage != NULL) 887 exec_unmap_first_page(imgp); 888 889 if (imgp->vp != NULL) { 890 if (args->fname) 891 NDFREE(&nd, NDF_ONLY_PNBUF); 892 if (imgp->opened) 893 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 894 if (imgp->textset) 895 VOP_UNSET_TEXT_CHECKED(imgp->vp); 896 if (error != 0) 897 vput(imgp->vp); 898 else 899 VOP_UNLOCK(imgp->vp); 900 } 901 902 if (imgp->object != NULL) 903 vm_object_deallocate(imgp->object); 904 905 free(imgp->freepath, M_TEMP); 906 907 if (error == 0) { 908 if (p->p_ptevents & PTRACE_EXEC) { 909 PROC_LOCK(p); 910 if (p->p_ptevents & PTRACE_EXEC) 911 td->td_dbgflags |= TDB_EXEC; 912 PROC_UNLOCK(p); 913 } 914 } else { 915 exec_fail: 916 /* we're done here, clear P_INEXEC */ 917 PROC_LOCK(p); 918 p->p_flag &= ~P_INEXEC; 919 PROC_UNLOCK(p); 920 921 SDT_PROBE1(proc, , , exec__failure, error); 922 } 923 924 if (imgp->newcred != NULL && oldcred != NULL) 925 crfree(imgp->newcred); 926 927 #ifdef MAC 928 mac_execve_exit(imgp); 929 mac_execve_interpreter_exit(interpvplabel); 930 #endif 931 exec_free_args(args); 932 933 /* 934 * Handle deferred decrement of ref counts. 935 */ 936 if (oldtextvp != NULL) 937 vrele(oldtextvp); 938 #ifdef KTRACE 939 if (tracevp != NULL) 940 vrele(tracevp); 941 if (tracecred != NULL) 942 crfree(tracecred); 943 #endif 944 pargs_drop(oldargs); 945 pargs_drop(newargs); 946 if (oldsigacts != NULL) 947 sigacts_free(oldsigacts); 948 if (euip != NULL) 949 uifree(euip); 950 951 if (error && imgp->vmspace_destroyed) { 952 /* sorry, no more process anymore. exit gracefully */ 953 exec_cleanup(td, oldvmspace); 954 exit1(td, 0, SIGABRT); 955 /* NOT REACHED */ 956 } 957 958 #ifdef KTRACE 959 if (error == 0) 960 ktrprocctor(p); 961 #endif 962 963 /* 964 * We don't want cpu_set_syscall_retval() to overwrite any of 965 * the register values put in place by exec_setregs(). 966 * Implementations of cpu_set_syscall_retval() will leave 967 * registers unmodified when returning EJUSTRETURN. 968 */ 969 return (error == 0 ? EJUSTRETURN : error); 970 } 971 972 void 973 exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 974 { 975 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 976 KASSERT(td->td_proc->p_vmspace != oldvmspace, 977 ("oldvmspace still used")); 978 vmspace_free(oldvmspace); 979 td->td_pflags &= ~TDP_EXECVMSPC; 980 } 981 } 982 983 int 984 exec_map_first_page(struct image_params *imgp) 985 { 986 vm_object_t object; 987 vm_page_t m; 988 int error; 989 990 if (imgp->firstpage != NULL) 991 exec_unmap_first_page(imgp); 992 993 object = imgp->vp->v_object; 994 if (object == NULL) 995 return (EACCES); 996 #if VM_NRESERVLEVEL > 0 997 if ((object->flags & OBJ_COLORED) == 0) { 998 VM_OBJECT_WLOCK(object); 999 vm_object_color(object, 0); 1000 VM_OBJECT_WUNLOCK(object); 1001 } 1002 #endif 1003 error = vm_page_grab_valid_unlocked(&m, object, 0, 1004 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1005 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1006 1007 if (error != VM_PAGER_OK) 1008 return (EIO); 1009 imgp->firstpage = sf_buf_alloc(m, 0); 1010 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1011 1012 return (0); 1013 } 1014 1015 void 1016 exec_unmap_first_page(struct image_params *imgp) 1017 { 1018 vm_page_t m; 1019 1020 if (imgp->firstpage != NULL) { 1021 m = sf_buf_page(imgp->firstpage); 1022 sf_buf_free(imgp->firstpage); 1023 imgp->firstpage = NULL; 1024 vm_page_unwire(m, PQ_ACTIVE); 1025 } 1026 } 1027 1028 /* 1029 * Destroy old address space, and allocate a new stack. 1030 * The new stack is only sgrowsiz large because it is grown 1031 * automatically on a page fault. 1032 */ 1033 int 1034 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1035 { 1036 int error; 1037 struct proc *p = imgp->proc; 1038 struct vmspace *vmspace = p->p_vmspace; 1039 struct thread *td = curthread; 1040 vm_object_t obj; 1041 struct rlimit rlim_stack; 1042 vm_offset_t sv_minuser, stack_addr; 1043 vm_map_t map; 1044 u_long ssiz; 1045 1046 imgp->vmspace_destroyed = 1; 1047 imgp->sysent = sv; 1048 1049 sigfastblock_clear(td); 1050 1051 /* May be called with Giant held */ 1052 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1053 1054 /* 1055 * Blow away entire process VM, if address space not shared, 1056 * otherwise, create a new VM space so that other threads are 1057 * not disrupted 1058 */ 1059 map = &vmspace->vm_map; 1060 if (map_at_zero) 1061 sv_minuser = sv->sv_minuser; 1062 else 1063 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1064 if (refcount_load(&vmspace->vm_refcnt) == 1 && 1065 vm_map_min(map) == sv_minuser && 1066 vm_map_max(map) == sv->sv_maxuser && 1067 cpu_exec_vmspace_reuse(p, map)) { 1068 shmexit(vmspace); 1069 pmap_remove_pages(vmspace_pmap(vmspace)); 1070 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1071 /* 1072 * An exec terminates mlockall(MCL_FUTURE), ASLR state 1073 * must be re-evaluated. 1074 */ 1075 vm_map_lock(map); 1076 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1077 MAP_ASLR_IGNSTART); 1078 vm_map_unlock(map); 1079 } else { 1080 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1081 if (error) 1082 return (error); 1083 vmspace = p->p_vmspace; 1084 map = &vmspace->vm_map; 1085 } 1086 map->flags |= imgp->map_flags; 1087 1088 /* Map a shared page */ 1089 obj = sv->sv_shared_page_obj; 1090 if (obj != NULL) { 1091 vm_object_reference(obj); 1092 error = vm_map_fixed(map, obj, 0, 1093 sv->sv_shared_page_base, sv->sv_shared_page_len, 1094 VM_PROT_READ | VM_PROT_EXECUTE, 1095 VM_PROT_READ | VM_PROT_EXECUTE, 1096 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1097 if (error != KERN_SUCCESS) { 1098 vm_object_deallocate(obj); 1099 return (vm_mmap_to_errno(error)); 1100 } 1101 } 1102 1103 /* Allocate a new stack */ 1104 if (imgp->stack_sz != 0) { 1105 ssiz = trunc_page(imgp->stack_sz); 1106 PROC_LOCK(p); 1107 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1108 PROC_UNLOCK(p); 1109 if (ssiz > rlim_stack.rlim_max) 1110 ssiz = rlim_stack.rlim_max; 1111 if (ssiz > rlim_stack.rlim_cur) { 1112 rlim_stack.rlim_cur = ssiz; 1113 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1114 } 1115 } else if (sv->sv_maxssiz != NULL) { 1116 ssiz = *sv->sv_maxssiz; 1117 } else { 1118 ssiz = maxssiz; 1119 } 1120 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1121 if (ssiz < imgp->eff_stack_sz) 1122 imgp->eff_stack_sz = ssiz; 1123 stack_addr = sv->sv_usrstack - ssiz; 1124 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, 1125 obj != NULL && imgp->stack_prot != 0 ? imgp->stack_prot : 1126 sv->sv_stackprot, VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1127 if (error != KERN_SUCCESS) 1128 return (vm_mmap_to_errno(error)); 1129 1130 /* 1131 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1132 * are still used to enforce the stack rlimit on the process stack. 1133 */ 1134 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1135 vmspace->vm_maxsaddr = (char *)stack_addr; 1136 1137 return (0); 1138 } 1139 1140 /* 1141 * Copy out argument and environment strings from the old process address 1142 * space into the temporary string buffer. 1143 */ 1144 int 1145 exec_copyin_args(struct image_args *args, const char *fname, 1146 enum uio_seg segflg, char **argv, char **envv) 1147 { 1148 u_long arg, env; 1149 int error; 1150 1151 bzero(args, sizeof(*args)); 1152 if (argv == NULL) 1153 return (EFAULT); 1154 1155 /* 1156 * Allocate demand-paged memory for the file name, argument, and 1157 * environment strings. 1158 */ 1159 error = exec_alloc_args(args); 1160 if (error != 0) 1161 return (error); 1162 1163 /* 1164 * Copy the file name. 1165 */ 1166 error = exec_args_add_fname(args, fname, segflg); 1167 if (error != 0) 1168 goto err_exit; 1169 1170 /* 1171 * extract arguments first 1172 */ 1173 for (;;) { 1174 error = fueword(argv++, &arg); 1175 if (error == -1) { 1176 error = EFAULT; 1177 goto err_exit; 1178 } 1179 if (arg == 0) 1180 break; 1181 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1182 UIO_USERSPACE); 1183 if (error != 0) 1184 goto err_exit; 1185 } 1186 1187 /* 1188 * extract environment strings 1189 */ 1190 if (envv) { 1191 for (;;) { 1192 error = fueword(envv++, &env); 1193 if (error == -1) { 1194 error = EFAULT; 1195 goto err_exit; 1196 } 1197 if (env == 0) 1198 break; 1199 error = exec_args_add_env(args, 1200 (char *)(uintptr_t)env, UIO_USERSPACE); 1201 if (error != 0) 1202 goto err_exit; 1203 } 1204 } 1205 1206 return (0); 1207 1208 err_exit: 1209 exec_free_args(args); 1210 return (error); 1211 } 1212 1213 int 1214 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1215 const void *data, size_t datalen, const int *fds, size_t fdslen) 1216 { 1217 struct filedesc *ofdp; 1218 const char *p; 1219 int *kfds; 1220 int error; 1221 1222 memset(args, '\0', sizeof(*args)); 1223 ofdp = td->td_proc->p_fd; 1224 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles) 1225 return (E2BIG); 1226 error = exec_alloc_args(args); 1227 if (error != 0) 1228 return (error); 1229 1230 args->begin_argv = args->buf; 1231 args->stringspace = ARG_MAX; 1232 1233 if (datalen > 0) { 1234 /* 1235 * Argument buffer has been provided. Copy it into the 1236 * kernel as a single string and add a terminating null 1237 * byte. 1238 */ 1239 error = copyin(data, args->begin_argv, datalen); 1240 if (error != 0) 1241 goto err_exit; 1242 args->begin_argv[datalen] = '\0'; 1243 args->endp = args->begin_argv + datalen + 1; 1244 args->stringspace -= datalen + 1; 1245 1246 /* 1247 * Traditional argument counting. Count the number of 1248 * null bytes. 1249 */ 1250 for (p = args->begin_argv; p < args->endp; ++p) 1251 if (*p == '\0') 1252 ++args->argc; 1253 } else { 1254 /* No argument buffer provided. */ 1255 args->endp = args->begin_argv; 1256 } 1257 1258 /* Create new file descriptor table. */ 1259 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1260 error = copyin(fds, kfds, fdslen * sizeof(int)); 1261 if (error != 0) { 1262 free(kfds, M_TEMP); 1263 goto err_exit; 1264 } 1265 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1266 free(kfds, M_TEMP); 1267 if (error != 0) 1268 goto err_exit; 1269 1270 return (0); 1271 err_exit: 1272 exec_free_args(args); 1273 return (error); 1274 } 1275 1276 struct exec_args_kva { 1277 vm_offset_t addr; 1278 u_int gen; 1279 SLIST_ENTRY(exec_args_kva) next; 1280 }; 1281 1282 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1283 1284 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1285 static struct mtx exec_args_kva_mtx; 1286 static u_int exec_args_gen; 1287 1288 static void 1289 exec_prealloc_args_kva(void *arg __unused) 1290 { 1291 struct exec_args_kva *argkva; 1292 u_int i; 1293 1294 SLIST_INIT(&exec_args_kva_freelist); 1295 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1296 for (i = 0; i < exec_map_entries; i++) { 1297 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1298 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1299 argkva->gen = exec_args_gen; 1300 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1301 } 1302 } 1303 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1304 1305 static vm_offset_t 1306 exec_alloc_args_kva(void **cookie) 1307 { 1308 struct exec_args_kva *argkva; 1309 1310 argkva = (void *)atomic_readandclear_ptr( 1311 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1312 if (argkva == NULL) { 1313 mtx_lock(&exec_args_kva_mtx); 1314 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1315 (void)mtx_sleep(&exec_args_kva_freelist, 1316 &exec_args_kva_mtx, 0, "execkva", 0); 1317 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1318 mtx_unlock(&exec_args_kva_mtx); 1319 } 1320 *(struct exec_args_kva **)cookie = argkva; 1321 return (argkva->addr); 1322 } 1323 1324 static void 1325 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1326 { 1327 vm_offset_t base; 1328 1329 base = argkva->addr; 1330 if (argkva->gen != gen) { 1331 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1332 MADV_FREE); 1333 argkva->gen = gen; 1334 } 1335 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1336 (uintptr_t)NULL, (uintptr_t)argkva)) { 1337 mtx_lock(&exec_args_kva_mtx); 1338 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1339 wakeup_one(&exec_args_kva_freelist); 1340 mtx_unlock(&exec_args_kva_mtx); 1341 } 1342 } 1343 1344 static void 1345 exec_free_args_kva(void *cookie) 1346 { 1347 1348 exec_release_args_kva(cookie, exec_args_gen); 1349 } 1350 1351 static void 1352 exec_args_kva_lowmem(void *arg __unused) 1353 { 1354 SLIST_HEAD(, exec_args_kva) head; 1355 struct exec_args_kva *argkva; 1356 u_int gen; 1357 int i; 1358 1359 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1360 1361 /* 1362 * Force an madvise of each KVA range. Any currently allocated ranges 1363 * will have MADV_FREE applied once they are freed. 1364 */ 1365 SLIST_INIT(&head); 1366 mtx_lock(&exec_args_kva_mtx); 1367 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1368 mtx_unlock(&exec_args_kva_mtx); 1369 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1370 SLIST_REMOVE_HEAD(&head, next); 1371 exec_release_args_kva(argkva, gen); 1372 } 1373 1374 CPU_FOREACH(i) { 1375 argkva = (void *)atomic_readandclear_ptr( 1376 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1377 if (argkva != NULL) 1378 exec_release_args_kva(argkva, gen); 1379 } 1380 } 1381 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1382 EVENTHANDLER_PRI_ANY); 1383 1384 /* 1385 * Allocate temporary demand-paged, zero-filled memory for the file name, 1386 * argument, and environment strings. 1387 */ 1388 int 1389 exec_alloc_args(struct image_args *args) 1390 { 1391 1392 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1393 return (0); 1394 } 1395 1396 void 1397 exec_free_args(struct image_args *args) 1398 { 1399 1400 if (args->buf != NULL) { 1401 exec_free_args_kva(args->bufkva); 1402 args->buf = NULL; 1403 } 1404 if (args->fname_buf != NULL) { 1405 free(args->fname_buf, M_TEMP); 1406 args->fname_buf = NULL; 1407 } 1408 if (args->fdp != NULL) 1409 fdescfree_remapped(args->fdp); 1410 } 1411 1412 /* 1413 * A set to functions to fill struct image args. 1414 * 1415 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1416 * fname) before the other functions. All exec_args_add_arg() calls must 1417 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1418 * may be called any time after exec_args_add_fname(). 1419 * 1420 * exec_args_add_fname() - install path to be executed 1421 * exec_args_add_arg() - append an argument string 1422 * exec_args_add_env() - append an env string 1423 * exec_args_adjust_args() - adjust location of the argument list to 1424 * allow new arguments to be prepended 1425 */ 1426 int 1427 exec_args_add_fname(struct image_args *args, const char *fname, 1428 enum uio_seg segflg) 1429 { 1430 int error; 1431 size_t length; 1432 1433 KASSERT(args->fname == NULL, ("fname already appended")); 1434 KASSERT(args->endp == NULL, ("already appending to args")); 1435 1436 if (fname != NULL) { 1437 args->fname = args->buf; 1438 error = segflg == UIO_SYSSPACE ? 1439 copystr(fname, args->fname, PATH_MAX, &length) : 1440 copyinstr(fname, args->fname, PATH_MAX, &length); 1441 if (error != 0) 1442 return (error == ENAMETOOLONG ? E2BIG : error); 1443 } else 1444 length = 0; 1445 1446 /* Set up for _arg_*()/_env_*() */ 1447 args->endp = args->buf + length; 1448 /* begin_argv must be set and kept updated */ 1449 args->begin_argv = args->endp; 1450 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1451 ("too little space remaining for arguments %zu < %zu", 1452 exec_map_entry_size - length, (size_t)ARG_MAX)); 1453 args->stringspace = ARG_MAX; 1454 1455 return (0); 1456 } 1457 1458 static int 1459 exec_args_add_str(struct image_args *args, const char *str, 1460 enum uio_seg segflg, int *countp) 1461 { 1462 int error; 1463 size_t length; 1464 1465 KASSERT(args->endp != NULL, ("endp not initialized")); 1466 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1467 1468 error = (segflg == UIO_SYSSPACE) ? 1469 copystr(str, args->endp, args->stringspace, &length) : 1470 copyinstr(str, args->endp, args->stringspace, &length); 1471 if (error != 0) 1472 return (error == ENAMETOOLONG ? E2BIG : error); 1473 args->stringspace -= length; 1474 args->endp += length; 1475 (*countp)++; 1476 1477 return (0); 1478 } 1479 1480 int 1481 exec_args_add_arg(struct image_args *args, const char *argp, 1482 enum uio_seg segflg) 1483 { 1484 1485 KASSERT(args->envc == 0, ("appending args after env")); 1486 1487 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1488 } 1489 1490 int 1491 exec_args_add_env(struct image_args *args, const char *envp, 1492 enum uio_seg segflg) 1493 { 1494 1495 if (args->envc == 0) 1496 args->begin_envv = args->endp; 1497 1498 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1499 } 1500 1501 int 1502 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1503 { 1504 ssize_t offset; 1505 1506 KASSERT(args->endp != NULL, ("endp not initialized")); 1507 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1508 1509 offset = extend - consume; 1510 if (args->stringspace < offset) 1511 return (E2BIG); 1512 memmove(args->begin_argv + extend, args->begin_argv + consume, 1513 args->endp - args->begin_argv + consume); 1514 if (args->envc > 0) 1515 args->begin_envv += offset; 1516 args->endp += offset; 1517 args->stringspace -= offset; 1518 return (0); 1519 } 1520 1521 char * 1522 exec_args_get_begin_envv(struct image_args *args) 1523 { 1524 1525 KASSERT(args->endp != NULL, ("endp not initialized")); 1526 1527 if (args->envc > 0) 1528 return (args->begin_envv); 1529 return (args->endp); 1530 } 1531 1532 /* 1533 * Copy strings out to the new process address space, constructing new arg 1534 * and env vector tables. Return a pointer to the base so that it can be used 1535 * as the initial stack pointer. 1536 */ 1537 int 1538 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1539 { 1540 int argc, envc; 1541 char **vectp; 1542 char *stringp; 1543 uintptr_t destp, ustringp; 1544 struct ps_strings *arginfo; 1545 struct proc *p; 1546 size_t execpath_len; 1547 int error, szsigcode, szps; 1548 char canary[sizeof(long) * 8]; 1549 1550 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1551 /* 1552 * Calculate string base and vector table pointers. 1553 * Also deal with signal trampoline code for this exec type. 1554 */ 1555 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1556 execpath_len = strlen(imgp->execpath) + 1; 1557 else 1558 execpath_len = 0; 1559 p = imgp->proc; 1560 szsigcode = 0; 1561 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1562 imgp->ps_strings = arginfo; 1563 if (p->p_sysent->sv_sigcode_base == 0) { 1564 if (p->p_sysent->sv_szsigcode != NULL) 1565 szsigcode = *(p->p_sysent->sv_szsigcode); 1566 } 1567 destp = (uintptr_t)arginfo; 1568 1569 /* 1570 * install sigcode 1571 */ 1572 if (szsigcode != 0) { 1573 destp -= szsigcode; 1574 destp = rounddown2(destp, sizeof(void *)); 1575 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1576 szsigcode); 1577 if (error != 0) 1578 return (error); 1579 } 1580 1581 /* 1582 * Copy the image path for the rtld. 1583 */ 1584 if (execpath_len != 0) { 1585 destp -= execpath_len; 1586 destp = rounddown2(destp, sizeof(void *)); 1587 imgp->execpathp = (void *)destp; 1588 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1589 if (error != 0) 1590 return (error); 1591 } 1592 1593 /* 1594 * Prepare the canary for SSP. 1595 */ 1596 arc4rand(canary, sizeof(canary), 0); 1597 destp -= sizeof(canary); 1598 imgp->canary = (void *)destp; 1599 error = copyout(canary, imgp->canary, sizeof(canary)); 1600 if (error != 0) 1601 return (error); 1602 imgp->canarylen = sizeof(canary); 1603 1604 /* 1605 * Prepare the pagesizes array. 1606 */ 1607 destp -= szps; 1608 destp = rounddown2(destp, sizeof(void *)); 1609 imgp->pagesizes = (void *)destp; 1610 error = copyout(pagesizes, imgp->pagesizes, szps); 1611 if (error != 0) 1612 return (error); 1613 imgp->pagesizeslen = szps; 1614 1615 /* 1616 * Allocate room for the argument and environment strings. 1617 */ 1618 destp -= ARG_MAX - imgp->args->stringspace; 1619 destp = rounddown2(destp, sizeof(void *)); 1620 ustringp = destp; 1621 1622 if (imgp->sysent->sv_stackgap != NULL) 1623 imgp->sysent->sv_stackgap(imgp, &destp); 1624 1625 if (imgp->auxargs) { 1626 /* 1627 * Allocate room on the stack for the ELF auxargs 1628 * array. It has up to AT_COUNT entries. 1629 */ 1630 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1631 destp = rounddown2(destp, sizeof(void *)); 1632 } 1633 1634 vectp = (char **)destp; 1635 1636 /* 1637 * Allocate room for the argv[] and env vectors including the 1638 * terminating NULL pointers. 1639 */ 1640 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1641 1642 /* 1643 * vectp also becomes our initial stack base 1644 */ 1645 *stack_base = (uintptr_t)vectp; 1646 1647 stringp = imgp->args->begin_argv; 1648 argc = imgp->args->argc; 1649 envc = imgp->args->envc; 1650 1651 /* 1652 * Copy out strings - arguments and environment. 1653 */ 1654 error = copyout(stringp, (void *)ustringp, 1655 ARG_MAX - imgp->args->stringspace); 1656 if (error != 0) 1657 return (error); 1658 1659 /* 1660 * Fill in "ps_strings" struct for ps, w, etc. 1661 */ 1662 imgp->argv = vectp; 1663 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1664 suword32(&arginfo->ps_nargvstr, argc) != 0) 1665 return (EFAULT); 1666 1667 /* 1668 * Fill in argument portion of vector table. 1669 */ 1670 for (; argc > 0; --argc) { 1671 if (suword(vectp++, ustringp) != 0) 1672 return (EFAULT); 1673 while (*stringp++ != 0) 1674 ustringp++; 1675 ustringp++; 1676 } 1677 1678 /* a null vector table pointer separates the argp's from the envp's */ 1679 if (suword(vectp++, 0) != 0) 1680 return (EFAULT); 1681 1682 imgp->envv = vectp; 1683 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1684 suword32(&arginfo->ps_nenvstr, envc) != 0) 1685 return (EFAULT); 1686 1687 /* 1688 * Fill in environment portion of vector table. 1689 */ 1690 for (; envc > 0; --envc) { 1691 if (suword(vectp++, ustringp) != 0) 1692 return (EFAULT); 1693 while (*stringp++ != 0) 1694 ustringp++; 1695 ustringp++; 1696 } 1697 1698 /* end of vector table is a null pointer */ 1699 if (suword(vectp, 0) != 0) 1700 return (EFAULT); 1701 1702 if (imgp->auxargs) { 1703 vectp++; 1704 error = imgp->sysent->sv_copyout_auxargs(imgp, 1705 (uintptr_t)vectp); 1706 if (error != 0) 1707 return (error); 1708 } 1709 1710 return (0); 1711 } 1712 1713 /* 1714 * Check permissions of file to execute. 1715 * Called with imgp->vp locked. 1716 * Return 0 for success or error code on failure. 1717 */ 1718 int 1719 exec_check_permissions(struct image_params *imgp) 1720 { 1721 struct vnode *vp = imgp->vp; 1722 struct vattr *attr = imgp->attr; 1723 struct thread *td; 1724 int error; 1725 1726 td = curthread; 1727 1728 /* Get file attributes */ 1729 error = VOP_GETATTR(vp, attr, td->td_ucred); 1730 if (error) 1731 return (error); 1732 1733 #ifdef MAC 1734 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1735 if (error) 1736 return (error); 1737 #endif 1738 1739 /* 1740 * 1) Check if file execution is disabled for the filesystem that 1741 * this file resides on. 1742 * 2) Ensure that at least one execute bit is on. Otherwise, a 1743 * privileged user will always succeed, and we don't want this 1744 * to happen unless the file really is executable. 1745 * 3) Ensure that the file is a regular file. 1746 */ 1747 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1748 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1749 (attr->va_type != VREG)) 1750 return (EACCES); 1751 1752 /* 1753 * Zero length files can't be exec'd 1754 */ 1755 if (attr->va_size == 0) 1756 return (ENOEXEC); 1757 1758 /* 1759 * Check for execute permission to file based on current credentials. 1760 */ 1761 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1762 if (error) 1763 return (error); 1764 1765 /* 1766 * Check number of open-for-writes on the file and deny execution 1767 * if there are any. 1768 * 1769 * Add a text reference now so no one can write to the 1770 * executable while we're activating it. 1771 * 1772 * Remember if this was set before and unset it in case this is not 1773 * actually an executable image. 1774 */ 1775 error = VOP_SET_TEXT(vp); 1776 if (error != 0) 1777 return (error); 1778 imgp->textset = true; 1779 1780 /* 1781 * Call filesystem specific open routine (which does nothing in the 1782 * general case). 1783 */ 1784 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1785 if (error == 0) 1786 imgp->opened = 1; 1787 return (error); 1788 } 1789 1790 /* 1791 * Exec handler registration 1792 */ 1793 int 1794 exec_register(const struct execsw *execsw_arg) 1795 { 1796 const struct execsw **es, **xs, **newexecsw; 1797 u_int count = 2; /* New slot and trailing NULL */ 1798 1799 if (execsw) 1800 for (es = execsw; *es; es++) 1801 count++; 1802 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1803 xs = newexecsw; 1804 if (execsw) 1805 for (es = execsw; *es; es++) 1806 *xs++ = *es; 1807 *xs++ = execsw_arg; 1808 *xs = NULL; 1809 if (execsw) 1810 free(execsw, M_TEMP); 1811 execsw = newexecsw; 1812 return (0); 1813 } 1814 1815 int 1816 exec_unregister(const struct execsw *execsw_arg) 1817 { 1818 const struct execsw **es, **xs, **newexecsw; 1819 int count = 1; 1820 1821 if (execsw == NULL) 1822 panic("unregister with no handlers left?\n"); 1823 1824 for (es = execsw; *es; es++) { 1825 if (*es == execsw_arg) 1826 break; 1827 } 1828 if (*es == NULL) 1829 return (ENOENT); 1830 for (es = execsw; *es; es++) 1831 if (*es != execsw_arg) 1832 count++; 1833 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1834 xs = newexecsw; 1835 for (es = execsw; *es; es++) 1836 if (*es != execsw_arg) 1837 *xs++ = *es; 1838 *xs = NULL; 1839 if (execsw) 1840 free(execsw, M_TEMP); 1841 execsw = newexecsw; 1842 return (0); 1843 } 1844