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/asan.h> 41 #include <sys/capsicum.h> 42 #include <sys/compressor.h> 43 #include <sys/eventhandler.h> 44 #include <sys/exec.h> 45 #include <sys/fcntl.h> 46 #include <sys/filedesc.h> 47 #include <sys/imgact.h> 48 #include <sys/imgact_elf.h> 49 #include <sys/kernel.h> 50 #include <sys/lock.h> 51 #include <sys/malloc.h> 52 #include <sys/mman.h> 53 #include <sys/mount.h> 54 #include <sys/mutex.h> 55 #include <sys/namei.h> 56 #include <sys/priv.h> 57 #include <sys/proc.h> 58 #include <sys/ptrace.h> 59 #include <sys/resourcevar.h> 60 #include <sys/rwlock.h> 61 #include <sys/sched.h> 62 #include <sys/sdt.h> 63 #include <sys/sf_buf.h> 64 #include <sys/shm.h> 65 #include <sys/signalvar.h> 66 #include <sys/smp.h> 67 #include <sys/stat.h> 68 #include <sys/syscallsubr.h> 69 #include <sys/sysctl.h> 70 #include <sys/sysent.h> 71 #include <sys/sysproto.h> 72 #include <sys/timers.h> 73 #include <sys/umtx.h> 74 #include <sys/vnode.h> 75 #include <sys/wait.h> 76 #ifdef KTRACE 77 #include <sys/ktrace.h> 78 #endif 79 80 #include <vm/vm.h> 81 #include <vm/vm_param.h> 82 #include <vm/pmap.h> 83 #include <vm/vm_page.h> 84 #include <vm/vm_map.h> 85 #include <vm/vm_kern.h> 86 #include <vm/vm_extern.h> 87 #include <vm/vm_object.h> 88 #include <vm/vm_pager.h> 89 90 #ifdef HWPMC_HOOKS 91 #include <sys/pmckern.h> 92 #endif 93 94 #include <machine/reg.h> 95 96 #include <security/audit/audit.h> 97 #include <security/mac/mac_framework.h> 98 99 #ifdef KDTRACE_HOOKS 100 #include <sys/dtrace_bsd.h> 101 dtrace_execexit_func_t dtrace_fasttrap_exec; 102 #endif 103 104 SDT_PROVIDER_DECLARE(proc); 105 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 106 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 107 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 108 109 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 110 111 int coredump_pack_fileinfo = 1; 112 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 113 &coredump_pack_fileinfo, 0, 114 "Enable file path packing in 'procstat -f' coredump notes"); 115 116 int coredump_pack_vmmapinfo = 1; 117 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 118 &coredump_pack_vmmapinfo, 0, 119 "Enable file path packing in 'procstat -v' coredump notes"); 120 121 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 122 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 123 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 124 static int do_execve(struct thread *td, struct image_args *args, 125 struct mac *mac_p, struct vmspace *oldvmspace); 126 127 /* XXX This should be vm_size_t. */ 128 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 129 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", 130 "Location of process' ps_strings structure"); 131 132 /* XXX This should be vm_size_t. */ 133 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 134 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", 135 "Top of process stack"); 136 137 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 138 NULL, 0, sysctl_kern_stackprot, "I", 139 "Stack memory permissions"); 140 141 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 142 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 143 &ps_arg_cache_limit, 0, 144 "Process' command line characters cache limit"); 145 146 static int disallow_high_osrel; 147 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 148 &disallow_high_osrel, 0, 149 "Disallow execution of binaries built for higher version of the world"); 150 151 static int map_at_zero = 0; 152 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 153 "Permit processes to map an object at virtual address 0."); 154 155 static int 156 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 157 { 158 struct proc *p; 159 int error; 160 161 p = curproc; 162 #ifdef SCTL_MASK32 163 if (req->flags & SCTL_MASK32) { 164 unsigned int val; 165 val = (unsigned int)p->p_sysent->sv_psstrings; 166 error = SYSCTL_OUT(req, &val, sizeof(val)); 167 } else 168 #endif 169 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 170 sizeof(p->p_sysent->sv_psstrings)); 171 return error; 172 } 173 174 static int 175 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 176 { 177 struct proc *p; 178 int error; 179 180 p = curproc; 181 #ifdef SCTL_MASK32 182 if (req->flags & SCTL_MASK32) { 183 unsigned int val; 184 val = (unsigned int)p->p_sysent->sv_usrstack; 185 error = SYSCTL_OUT(req, &val, sizeof(val)); 186 } else 187 #endif 188 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 189 sizeof(p->p_sysent->sv_usrstack)); 190 return error; 191 } 192 193 static int 194 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 195 { 196 struct proc *p; 197 198 p = curproc; 199 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 200 sizeof(p->p_sysent->sv_stackprot))); 201 } 202 203 /* 204 * Each of the items is a pointer to a `const struct execsw', hence the 205 * double pointer here. 206 */ 207 static const struct execsw **execsw; 208 209 #ifndef _SYS_SYSPROTO_H_ 210 struct execve_args { 211 char *fname; 212 char **argv; 213 char **envv; 214 }; 215 #endif 216 217 int 218 sys_execve(struct thread *td, struct execve_args *uap) 219 { 220 struct image_args args; 221 struct vmspace *oldvmspace; 222 int error; 223 224 error = pre_execve(td, &oldvmspace); 225 if (error != 0) 226 return (error); 227 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 228 uap->argv, uap->envv); 229 if (error == 0) 230 error = kern_execve(td, &args, NULL, oldvmspace); 231 post_execve(td, error, oldvmspace); 232 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 233 return (error); 234 } 235 236 #ifndef _SYS_SYSPROTO_H_ 237 struct fexecve_args { 238 int fd; 239 char **argv; 240 char **envv; 241 }; 242 #endif 243 int 244 sys_fexecve(struct thread *td, struct fexecve_args *uap) 245 { 246 struct image_args args; 247 struct vmspace *oldvmspace; 248 int error; 249 250 error = pre_execve(td, &oldvmspace); 251 if (error != 0) 252 return (error); 253 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 254 uap->argv, uap->envv); 255 if (error == 0) { 256 args.fd = uap->fd; 257 error = kern_execve(td, &args, NULL, oldvmspace); 258 } 259 post_execve(td, error, oldvmspace); 260 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 261 return (error); 262 } 263 264 #ifndef _SYS_SYSPROTO_H_ 265 struct __mac_execve_args { 266 char *fname; 267 char **argv; 268 char **envv; 269 struct mac *mac_p; 270 }; 271 #endif 272 273 int 274 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 275 { 276 #ifdef MAC 277 struct image_args args; 278 struct vmspace *oldvmspace; 279 int error; 280 281 error = pre_execve(td, &oldvmspace); 282 if (error != 0) 283 return (error); 284 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 285 uap->argv, uap->envv); 286 if (error == 0) 287 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 288 post_execve(td, error, oldvmspace); 289 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 290 return (error); 291 #else 292 return (ENOSYS); 293 #endif 294 } 295 296 int 297 pre_execve(struct thread *td, struct vmspace **oldvmspace) 298 { 299 struct proc *p; 300 int error; 301 302 KASSERT(td == curthread, ("non-current thread %p", td)); 303 error = 0; 304 p = td->td_proc; 305 if ((p->p_flag & P_HADTHREADS) != 0) { 306 PROC_LOCK(p); 307 if (thread_single(p, SINGLE_BOUNDARY) != 0) 308 error = ERESTART; 309 PROC_UNLOCK(p); 310 } 311 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 312 ("nested execve")); 313 *oldvmspace = p->p_vmspace; 314 return (error); 315 } 316 317 void 318 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 319 { 320 struct proc *p; 321 322 KASSERT(td == curthread, ("non-current thread %p", td)); 323 p = td->td_proc; 324 if ((p->p_flag & P_HADTHREADS) != 0) { 325 PROC_LOCK(p); 326 /* 327 * If success, we upgrade to SINGLE_EXIT state to 328 * force other threads to suicide. 329 */ 330 if (error == EJUSTRETURN) 331 thread_single(p, SINGLE_EXIT); 332 else 333 thread_single_end(p, SINGLE_BOUNDARY); 334 PROC_UNLOCK(p); 335 } 336 exec_cleanup(td, oldvmspace); 337 } 338 339 /* 340 * kern_execve() has the astonishing property of not always returning to 341 * the caller. If sufficiently bad things happen during the call to 342 * do_execve(), it can end up calling exit1(); as a result, callers must 343 * avoid doing anything which they might need to undo (e.g., allocating 344 * memory). 345 */ 346 int 347 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 348 struct vmspace *oldvmspace) 349 { 350 351 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 352 exec_args_get_begin_envv(args) - args->begin_argv); 353 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 354 args->endp - exec_args_get_begin_envv(args)); 355 return (do_execve(td, args, mac_p, oldvmspace)); 356 } 357 358 /* 359 * In-kernel implementation of execve(). All arguments are assumed to be 360 * userspace pointers from the passed thread. 361 */ 362 static int 363 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 364 struct vmspace *oldvmspace) 365 { 366 struct proc *p = td->td_proc; 367 struct nameidata nd; 368 struct ucred *oldcred; 369 struct uidinfo *euip = NULL; 370 uintptr_t stack_base; 371 struct image_params image_params, *imgp; 372 struct vattr attr; 373 int (*img_first)(struct image_params *); 374 struct pargs *oldargs = NULL, *newargs = NULL; 375 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 376 #ifdef KTRACE 377 struct ktr_io_params *kiop; 378 #endif 379 struct vnode *oldtextvp = NULL, *newtextvp; 380 int credential_changing; 381 #ifdef MAC 382 struct label *interpvplabel = NULL; 383 int will_transition; 384 #endif 385 #ifdef HWPMC_HOOKS 386 struct pmckern_procexec pe; 387 #endif 388 int error, i, orig_osrel; 389 uint32_t orig_fctl0; 390 static const char fexecv_proc_title[] = "(fexecv)"; 391 392 imgp = &image_params; 393 #ifdef KTRACE 394 kiop = NULL; 395 #endif 396 397 /* 398 * Lock the process and set the P_INEXEC flag to indicate that 399 * it should be left alone until we're done here. This is 400 * necessary to avoid race conditions - e.g. in ptrace() - 401 * that might allow a local user to illicitly obtain elevated 402 * privileges. 403 */ 404 PROC_LOCK(p); 405 KASSERT((p->p_flag & P_INEXEC) == 0, 406 ("%s(): process already has P_INEXEC flag", __func__)); 407 p->p_flag |= P_INEXEC; 408 PROC_UNLOCK(p); 409 410 /* 411 * Initialize part of the common data 412 */ 413 bzero(imgp, sizeof(*imgp)); 414 imgp->proc = p; 415 imgp->attr = &attr; 416 imgp->args = args; 417 oldcred = p->p_ucred; 418 orig_osrel = p->p_osrel; 419 orig_fctl0 = p->p_fctl0; 420 421 #ifdef MAC 422 error = mac_execve_enter(imgp, mac_p); 423 if (error) 424 goto exec_fail; 425 #endif 426 427 /* 428 * Translate the file name. namei() returns a vnode pointer 429 * in ni_vp among other things. 430 * 431 * XXXAUDIT: It would be desirable to also audit the name of the 432 * interpreter if this is an interpreted binary. 433 */ 434 if (args->fname != NULL) { 435 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 436 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 437 } 438 439 SDT_PROBE1(proc, , , exec, args->fname); 440 441 interpret: 442 if (args->fname != NULL) { 443 #ifdef CAPABILITY_MODE 444 /* 445 * While capability mode can't reach this point via direct 446 * path arguments to execve(), we also don't allow 447 * interpreters to be used in capability mode (for now). 448 * Catch indirect lookups and return a permissions error. 449 */ 450 if (IN_CAPABILITY_MODE(td)) { 451 error = ECAPMODE; 452 goto exec_fail; 453 } 454 #endif 455 error = namei(&nd); 456 if (error) 457 goto exec_fail; 458 459 newtextvp = nd.ni_vp; 460 imgp->vp = newtextvp; 461 } else { 462 AUDIT_ARG_FD(args->fd); 463 /* 464 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 465 */ 466 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 467 if (error) 468 goto exec_fail; 469 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 470 AUDIT_ARG_VNODE1(newtextvp); 471 imgp->vp = newtextvp; 472 } 473 474 /* 475 * Check file permissions. Also 'opens' file and sets its vnode to 476 * text mode. 477 */ 478 error = exec_check_permissions(imgp); 479 if (error) 480 goto exec_fail_dealloc; 481 482 imgp->object = imgp->vp->v_object; 483 if (imgp->object != NULL) 484 vm_object_reference(imgp->object); 485 486 error = exec_map_first_page(imgp); 487 if (error) 488 goto exec_fail_dealloc; 489 490 imgp->proc->p_osrel = 0; 491 imgp->proc->p_fctl0 = 0; 492 493 /* 494 * Implement image setuid/setgid. 495 * 496 * Determine new credentials before attempting image activators 497 * so that it can be used by process_exec handlers to determine 498 * credential/setid changes. 499 * 500 * Don't honor setuid/setgid if the filesystem prohibits it or if 501 * the process is being traced. 502 * 503 * We disable setuid/setgid/etc in capability mode on the basis 504 * that most setugid applications are not written with that 505 * environment in mind, and will therefore almost certainly operate 506 * incorrectly. In principle there's no reason that setugid 507 * applications might not be useful in capability mode, so we may want 508 * to reconsider this conservative design choice in the future. 509 * 510 * XXXMAC: For the time being, use NOSUID to also prohibit 511 * transitions on the file system. 512 */ 513 credential_changing = 0; 514 credential_changing |= (attr.va_mode & S_ISUID) && 515 oldcred->cr_uid != attr.va_uid; 516 credential_changing |= (attr.va_mode & S_ISGID) && 517 oldcred->cr_gid != attr.va_gid; 518 #ifdef MAC 519 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 520 interpvplabel, imgp); 521 credential_changing |= will_transition; 522 #endif 523 524 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 525 if (credential_changing) 526 imgp->proc->p_pdeathsig = 0; 527 528 if (credential_changing && 529 #ifdef CAPABILITY_MODE 530 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 531 #endif 532 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 533 (p->p_flag & P_TRACED) == 0) { 534 imgp->credential_setid = true; 535 VOP_UNLOCK(imgp->vp); 536 imgp->newcred = crdup(oldcred); 537 if (attr.va_mode & S_ISUID) { 538 euip = uifind(attr.va_uid); 539 change_euid(imgp->newcred, euip); 540 } 541 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 542 if (attr.va_mode & S_ISGID) 543 change_egid(imgp->newcred, attr.va_gid); 544 /* 545 * Implement correct POSIX saved-id behavior. 546 * 547 * XXXMAC: Note that the current logic will save the 548 * uid and gid if a MAC domain transition occurs, even 549 * though maybe it shouldn't. 550 */ 551 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 552 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 553 } else { 554 /* 555 * Implement correct POSIX saved-id behavior. 556 * 557 * XXX: It's not clear that the existing behavior is 558 * POSIX-compliant. A number of sources indicate that the 559 * saved uid/gid should only be updated if the new ruid is 560 * not equal to the old ruid, or the new euid is not equal 561 * to the old euid and the new euid is not equal to the old 562 * ruid. The FreeBSD code always updates the saved uid/gid. 563 * Also, this code uses the new (replaced) euid and egid as 564 * the source, which may or may not be the right ones to use. 565 */ 566 if (oldcred->cr_svuid != oldcred->cr_uid || 567 oldcred->cr_svgid != oldcred->cr_gid) { 568 VOP_UNLOCK(imgp->vp); 569 imgp->newcred = crdup(oldcred); 570 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 571 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 572 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 573 } 574 } 575 /* The new credentials are installed into the process later. */ 576 577 /* 578 * Do the best to calculate the full path to the image file. 579 */ 580 if (args->fname != NULL && args->fname[0] == '/') 581 imgp->execpath = args->fname; 582 else { 583 VOP_UNLOCK(imgp->vp); 584 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0) 585 imgp->execpath = args->fname; 586 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 587 } 588 589 /* 590 * If the current process has a special image activator it 591 * wants to try first, call it. For example, emulating shell 592 * scripts differently. 593 */ 594 error = -1; 595 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 596 error = img_first(imgp); 597 598 /* 599 * Loop through the list of image activators, calling each one. 600 * An activator returns -1 if there is no match, 0 on success, 601 * and an error otherwise. 602 */ 603 for (i = 0; error == -1 && execsw[i]; ++i) { 604 if (execsw[i]->ex_imgact == NULL || 605 execsw[i]->ex_imgact == img_first) { 606 continue; 607 } 608 error = (*execsw[i]->ex_imgact)(imgp); 609 } 610 611 if (error) { 612 if (error == -1) 613 error = ENOEXEC; 614 goto exec_fail_dealloc; 615 } 616 617 /* 618 * Special interpreter operation, cleanup and loop up to try to 619 * activate the interpreter. 620 */ 621 if (imgp->interpreted) { 622 exec_unmap_first_page(imgp); 623 /* 624 * The text reference needs to be removed for scripts. 625 * There is a short period before we determine that 626 * something is a script where text reference is active. 627 * The vnode lock is held over this entire period 628 * so nothing should illegitimately be blocked. 629 */ 630 MPASS(imgp->textset); 631 VOP_UNSET_TEXT_CHECKED(newtextvp); 632 imgp->textset = false; 633 /* free name buffer and old vnode */ 634 if (args->fname != NULL) 635 NDFREE(&nd, NDF_ONLY_PNBUF); 636 #ifdef MAC 637 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 638 #endif 639 if (imgp->opened) { 640 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 641 imgp->opened = 0; 642 } 643 vput(newtextvp); 644 vm_object_deallocate(imgp->object); 645 imgp->object = NULL; 646 imgp->credential_setid = false; 647 if (imgp->newcred != NULL) { 648 crfree(imgp->newcred); 649 imgp->newcred = NULL; 650 } 651 imgp->execpath = NULL; 652 free(imgp->freepath, M_TEMP); 653 imgp->freepath = NULL; 654 /* set new name to that of the interpreter */ 655 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 656 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 657 args->fname = imgp->interpreter_name; 658 goto interpret; 659 } 660 661 /* 662 * NB: We unlock the vnode here because it is believed that none 663 * of the sv_copyout_strings/sv_fixup operations require the vnode. 664 */ 665 VOP_UNLOCK(imgp->vp); 666 667 if (disallow_high_osrel && 668 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 669 error = ENOEXEC; 670 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 671 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 672 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 673 goto exec_fail_dealloc; 674 } 675 676 /* ABI enforces the use of Capsicum. Switch into capabilities mode. */ 677 if (SV_PROC_FLAG(p, SV_CAPSICUM)) 678 sys_cap_enter(td, NULL); 679 680 /* 681 * Copy out strings (args and env) and initialize stack base. 682 */ 683 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 684 if (error != 0) { 685 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 686 goto exec_fail_dealloc; 687 } 688 689 /* 690 * Stack setup. 691 */ 692 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 693 if (error != 0) { 694 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 695 goto exec_fail_dealloc; 696 } 697 698 if (args->fdp != NULL) { 699 /* Install a brand new file descriptor table. */ 700 fdinstall_remapped(td, args->fdp); 701 args->fdp = NULL; 702 } else { 703 /* 704 * Keep on using the existing file descriptor table. For 705 * security and other reasons, the file descriptor table 706 * cannot be shared after an exec. 707 */ 708 fdunshare(td); 709 pdunshare(td); 710 /* close files on exec */ 711 fdcloseexec(td); 712 } 713 714 /* 715 * Malloc things before we need locks. 716 */ 717 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 718 /* Cache arguments if they fit inside our allowance */ 719 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 720 newargs = pargs_alloc(i); 721 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 722 } 723 724 /* 725 * For security and other reasons, signal handlers cannot 726 * be shared after an exec. The new process gets a copy of the old 727 * handlers. In execsigs(), the new process will have its signals 728 * reset. 729 */ 730 if (sigacts_shared(p->p_sigacts)) { 731 oldsigacts = p->p_sigacts; 732 newsigacts = sigacts_alloc(); 733 sigacts_copy(newsigacts, oldsigacts); 734 } 735 736 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 737 738 PROC_LOCK(p); 739 if (oldsigacts) 740 p->p_sigacts = newsigacts; 741 /* Stop profiling */ 742 stopprofclock(p); 743 744 /* reset caught signals */ 745 execsigs(p); 746 747 /* name this process - nameiexec(p, ndp) */ 748 bzero(p->p_comm, sizeof(p->p_comm)); 749 if (args->fname) 750 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 751 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 752 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 753 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 754 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 755 #ifdef KTR 756 sched_clear_tdname(td); 757 #endif 758 759 /* 760 * mark as execed, wakeup the process that vforked (if any) and tell 761 * it that it now has its own resources back 762 */ 763 p->p_flag |= P_EXEC; 764 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 765 p->p_flag2 &= ~P2_NOTRACE; 766 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 767 p->p_flag2 &= ~P2_STKGAP_DISABLE; 768 if (p->p_flag & P_PPWAIT) { 769 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 770 cv_broadcast(&p->p_pwait); 771 /* STOPs are no longer ignored, arrange for AST */ 772 signotify(td); 773 } 774 775 /* 776 * Implement image setuid/setgid installation. 777 */ 778 if (imgp->credential_setid) { 779 /* 780 * Turn off syscall tracing for set-id programs, except for 781 * root. Record any set-id flags first to make sure that 782 * we do not regain any tracing during a possible block. 783 */ 784 setsugid(p); 785 #ifdef KTRACE 786 kiop = ktrprocexec(p); 787 #endif 788 /* 789 * Close any file descriptors 0..2 that reference procfs, 790 * then make sure file descriptors 0..2 are in use. 791 * 792 * Both fdsetugidsafety() and fdcheckstd() may call functions 793 * taking sleepable locks, so temporarily drop our locks. 794 */ 795 PROC_UNLOCK(p); 796 VOP_UNLOCK(imgp->vp); 797 fdsetugidsafety(td); 798 error = fdcheckstd(td); 799 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 800 if (error != 0) 801 goto exec_fail_dealloc; 802 PROC_LOCK(p); 803 #ifdef MAC 804 if (will_transition) { 805 mac_vnode_execve_transition(oldcred, imgp->newcred, 806 imgp->vp, interpvplabel, imgp); 807 } 808 #endif 809 } else { 810 if (oldcred->cr_uid == oldcred->cr_ruid && 811 oldcred->cr_gid == oldcred->cr_rgid) 812 p->p_flag &= ~P_SUGID; 813 } 814 /* 815 * Set the new credentials. 816 */ 817 if (imgp->newcred != NULL) { 818 proc_set_cred(p, imgp->newcred); 819 crfree(oldcred); 820 oldcred = NULL; 821 } 822 823 /* 824 * Store the vp for use in procfs. This vnode was referenced by namei 825 * or fgetvp_exec. 826 */ 827 oldtextvp = p->p_textvp; 828 p->p_textvp = newtextvp; 829 830 #ifdef KDTRACE_HOOKS 831 /* 832 * Tell the DTrace fasttrap provider about the exec if it 833 * has declared an interest. 834 */ 835 if (dtrace_fasttrap_exec) 836 dtrace_fasttrap_exec(p); 837 #endif 838 839 /* 840 * Notify others that we exec'd, and clear the P_INEXEC flag 841 * as we're now a bona fide freshly-execed process. 842 */ 843 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 844 p->p_flag &= ~P_INEXEC; 845 846 /* clear "fork but no exec" flag, as we _are_ execing */ 847 p->p_acflag &= ~AFORK; 848 849 /* 850 * Free any previous argument cache and replace it with 851 * the new argument cache, if any. 852 */ 853 oldargs = p->p_args; 854 p->p_args = newargs; 855 newargs = NULL; 856 857 PROC_UNLOCK(p); 858 859 #ifdef HWPMC_HOOKS 860 /* 861 * Check if system-wide sampling is in effect or if the 862 * current process is using PMCs. If so, do exec() time 863 * processing. This processing needs to happen AFTER the 864 * P_INEXEC flag is cleared. 865 */ 866 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 867 VOP_UNLOCK(imgp->vp); 868 pe.pm_credentialschanged = credential_changing; 869 pe.pm_entryaddr = imgp->entry_addr; 870 871 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 872 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 873 } 874 #endif 875 876 /* Set values passed into the program in registers. */ 877 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 878 879 VOP_MMAPPED(imgp->vp); 880 881 SDT_PROBE1(proc, , , exec__success, args->fname); 882 883 exec_fail_dealloc: 884 if (error != 0) { 885 p->p_osrel = orig_osrel; 886 p->p_fctl0 = orig_fctl0; 887 } 888 889 if (imgp->firstpage != NULL) 890 exec_unmap_first_page(imgp); 891 892 if (imgp->vp != NULL) { 893 if (args->fname) 894 NDFREE(&nd, NDF_ONLY_PNBUF); 895 if (imgp->opened) 896 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 897 if (imgp->textset) 898 VOP_UNSET_TEXT_CHECKED(imgp->vp); 899 if (error != 0) 900 vput(imgp->vp); 901 else 902 VOP_UNLOCK(imgp->vp); 903 } 904 905 if (imgp->object != NULL) 906 vm_object_deallocate(imgp->object); 907 908 free(imgp->freepath, M_TEMP); 909 910 if (error == 0) { 911 if (p->p_ptevents & PTRACE_EXEC) { 912 PROC_LOCK(p); 913 if (p->p_ptevents & PTRACE_EXEC) 914 td->td_dbgflags |= TDB_EXEC; 915 PROC_UNLOCK(p); 916 } 917 } else { 918 exec_fail: 919 /* we're done here, clear P_INEXEC */ 920 PROC_LOCK(p); 921 p->p_flag &= ~P_INEXEC; 922 PROC_UNLOCK(p); 923 924 SDT_PROBE1(proc, , , exec__failure, error); 925 } 926 927 if (imgp->newcred != NULL && oldcred != NULL) 928 crfree(imgp->newcred); 929 930 #ifdef MAC 931 mac_execve_exit(imgp); 932 mac_execve_interpreter_exit(interpvplabel); 933 #endif 934 exec_free_args(args); 935 936 /* 937 * Handle deferred decrement of ref counts. 938 */ 939 if (oldtextvp != NULL) 940 vrele(oldtextvp); 941 #ifdef KTRACE 942 ktr_io_params_free(kiop); 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 vm_prot_t stack_prot; 1045 u_long ssiz; 1046 1047 imgp->vmspace_destroyed = 1; 1048 imgp->sysent = sv; 1049 1050 sigfastblock_clear(td); 1051 umtx_exec(p); 1052 itimers_exec(p); 1053 if (sv->sv_onexec != NULL) 1054 sv->sv_onexec(p, imgp); 1055 1056 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1057 1058 /* 1059 * Blow away entire process VM, if address space not shared, 1060 * otherwise, create a new VM space so that other threads are 1061 * not disrupted 1062 */ 1063 map = &vmspace->vm_map; 1064 if (map_at_zero) 1065 sv_minuser = sv->sv_minuser; 1066 else 1067 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1068 if (refcount_load(&vmspace->vm_refcnt) == 1 && 1069 vm_map_min(map) == sv_minuser && 1070 vm_map_max(map) == sv->sv_maxuser && 1071 cpu_exec_vmspace_reuse(p, map)) { 1072 shmexit(vmspace); 1073 pmap_remove_pages(vmspace_pmap(vmspace)); 1074 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1075 /* 1076 * An exec terminates mlockall(MCL_FUTURE). 1077 * ASLR and W^X states must be re-evaluated. 1078 */ 1079 vm_map_lock(map); 1080 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1081 MAP_ASLR_IGNSTART | MAP_WXORX); 1082 vm_map_unlock(map); 1083 } else { 1084 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1085 if (error) 1086 return (error); 1087 vmspace = p->p_vmspace; 1088 map = &vmspace->vm_map; 1089 } 1090 map->flags |= imgp->map_flags; 1091 1092 /* Map a shared page */ 1093 obj = sv->sv_shared_page_obj; 1094 if (obj != NULL) { 1095 vm_object_reference(obj); 1096 error = vm_map_fixed(map, obj, 0, 1097 sv->sv_shared_page_base, sv->sv_shared_page_len, 1098 VM_PROT_READ | VM_PROT_EXECUTE, 1099 VM_PROT_READ | VM_PROT_EXECUTE, 1100 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1101 if (error != KERN_SUCCESS) { 1102 vm_object_deallocate(obj); 1103 return (vm_mmap_to_errno(error)); 1104 } 1105 } 1106 1107 /* Allocate a new stack */ 1108 if (imgp->stack_sz != 0) { 1109 ssiz = trunc_page(imgp->stack_sz); 1110 PROC_LOCK(p); 1111 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1112 PROC_UNLOCK(p); 1113 if (ssiz > rlim_stack.rlim_max) 1114 ssiz = rlim_stack.rlim_max; 1115 if (ssiz > rlim_stack.rlim_cur) { 1116 rlim_stack.rlim_cur = ssiz; 1117 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1118 } 1119 } else if (sv->sv_maxssiz != NULL) { 1120 ssiz = *sv->sv_maxssiz; 1121 } else { 1122 ssiz = maxssiz; 1123 } 1124 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1125 if (ssiz < imgp->eff_stack_sz) 1126 imgp->eff_stack_sz = ssiz; 1127 stack_addr = sv->sv_usrstack - ssiz; 1128 stack_prot = obj != NULL && imgp->stack_prot != 0 ? 1129 imgp->stack_prot : sv->sv_stackprot; 1130 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot, 1131 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1132 if (error != KERN_SUCCESS) { 1133 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x " 1134 "failed mach error %d errno %d\n", (uintmax_t)ssiz, 1135 stack_prot, error, vm_mmap_to_errno(error)); 1136 return (vm_mmap_to_errno(error)); 1137 } 1138 1139 /* 1140 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1141 * are still used to enforce the stack rlimit on the process stack. 1142 */ 1143 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1144 vmspace->vm_maxsaddr = (char *)stack_addr; 1145 1146 return (0); 1147 } 1148 1149 /* 1150 * Copy out argument and environment strings from the old process address 1151 * space into the temporary string buffer. 1152 */ 1153 int 1154 exec_copyin_args(struct image_args *args, const char *fname, 1155 enum uio_seg segflg, char **argv, char **envv) 1156 { 1157 u_long arg, env; 1158 int error; 1159 1160 bzero(args, sizeof(*args)); 1161 if (argv == NULL) 1162 return (EFAULT); 1163 1164 /* 1165 * Allocate demand-paged memory for the file name, argument, and 1166 * environment strings. 1167 */ 1168 error = exec_alloc_args(args); 1169 if (error != 0) 1170 return (error); 1171 1172 /* 1173 * Copy the file name. 1174 */ 1175 error = exec_args_add_fname(args, fname, segflg); 1176 if (error != 0) 1177 goto err_exit; 1178 1179 /* 1180 * extract arguments first 1181 */ 1182 for (;;) { 1183 error = fueword(argv++, &arg); 1184 if (error == -1) { 1185 error = EFAULT; 1186 goto err_exit; 1187 } 1188 if (arg == 0) 1189 break; 1190 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1191 UIO_USERSPACE); 1192 if (error != 0) 1193 goto err_exit; 1194 } 1195 1196 /* 1197 * extract environment strings 1198 */ 1199 if (envv) { 1200 for (;;) { 1201 error = fueword(envv++, &env); 1202 if (error == -1) { 1203 error = EFAULT; 1204 goto err_exit; 1205 } 1206 if (env == 0) 1207 break; 1208 error = exec_args_add_env(args, 1209 (char *)(uintptr_t)env, UIO_USERSPACE); 1210 if (error != 0) 1211 goto err_exit; 1212 } 1213 } 1214 1215 return (0); 1216 1217 err_exit: 1218 exec_free_args(args); 1219 return (error); 1220 } 1221 1222 int 1223 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1224 const void *data, size_t datalen, const int *fds, size_t fdslen) 1225 { 1226 struct filedesc *ofdp; 1227 const char *p; 1228 int *kfds; 1229 int error; 1230 1231 memset(args, '\0', sizeof(*args)); 1232 ofdp = td->td_proc->p_fd; 1233 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles) 1234 return (E2BIG); 1235 error = exec_alloc_args(args); 1236 if (error != 0) 1237 return (error); 1238 1239 args->begin_argv = args->buf; 1240 args->stringspace = ARG_MAX; 1241 1242 if (datalen > 0) { 1243 /* 1244 * Argument buffer has been provided. Copy it into the 1245 * kernel as a single string and add a terminating null 1246 * byte. 1247 */ 1248 error = copyin(data, args->begin_argv, datalen); 1249 if (error != 0) 1250 goto err_exit; 1251 args->begin_argv[datalen] = '\0'; 1252 args->endp = args->begin_argv + datalen + 1; 1253 args->stringspace -= datalen + 1; 1254 1255 /* 1256 * Traditional argument counting. Count the number of 1257 * null bytes. 1258 */ 1259 for (p = args->begin_argv; p < args->endp; ++p) 1260 if (*p == '\0') 1261 ++args->argc; 1262 } else { 1263 /* No argument buffer provided. */ 1264 args->endp = args->begin_argv; 1265 } 1266 1267 /* Create new file descriptor table. */ 1268 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1269 error = copyin(fds, kfds, fdslen * sizeof(int)); 1270 if (error != 0) { 1271 free(kfds, M_TEMP); 1272 goto err_exit; 1273 } 1274 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1275 free(kfds, M_TEMP); 1276 if (error != 0) 1277 goto err_exit; 1278 1279 return (0); 1280 err_exit: 1281 exec_free_args(args); 1282 return (error); 1283 } 1284 1285 struct exec_args_kva { 1286 vm_offset_t addr; 1287 u_int gen; 1288 SLIST_ENTRY(exec_args_kva) next; 1289 }; 1290 1291 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1292 1293 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1294 static struct mtx exec_args_kva_mtx; 1295 static u_int exec_args_gen; 1296 1297 static void 1298 exec_prealloc_args_kva(void *arg __unused) 1299 { 1300 struct exec_args_kva *argkva; 1301 u_int i; 1302 1303 SLIST_INIT(&exec_args_kva_freelist); 1304 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1305 for (i = 0; i < exec_map_entries; i++) { 1306 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1307 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1308 argkva->gen = exec_args_gen; 1309 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1310 } 1311 } 1312 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1313 1314 static vm_offset_t 1315 exec_alloc_args_kva(void **cookie) 1316 { 1317 struct exec_args_kva *argkva; 1318 1319 argkva = (void *)atomic_readandclear_ptr( 1320 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1321 if (argkva == NULL) { 1322 mtx_lock(&exec_args_kva_mtx); 1323 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1324 (void)mtx_sleep(&exec_args_kva_freelist, 1325 &exec_args_kva_mtx, 0, "execkva", 0); 1326 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1327 mtx_unlock(&exec_args_kva_mtx); 1328 } 1329 kasan_mark((void *)argkva->addr, exec_map_entry_size, 1330 exec_map_entry_size, 0); 1331 *(struct exec_args_kva **)cookie = argkva; 1332 return (argkva->addr); 1333 } 1334 1335 static void 1336 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1337 { 1338 vm_offset_t base; 1339 1340 base = argkva->addr; 1341 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size, 1342 KASAN_EXEC_ARGS_FREED); 1343 if (argkva->gen != gen) { 1344 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1345 MADV_FREE); 1346 argkva->gen = gen; 1347 } 1348 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1349 (uintptr_t)NULL, (uintptr_t)argkva)) { 1350 mtx_lock(&exec_args_kva_mtx); 1351 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1352 wakeup_one(&exec_args_kva_freelist); 1353 mtx_unlock(&exec_args_kva_mtx); 1354 } 1355 } 1356 1357 static void 1358 exec_free_args_kva(void *cookie) 1359 { 1360 1361 exec_release_args_kva(cookie, exec_args_gen); 1362 } 1363 1364 static void 1365 exec_args_kva_lowmem(void *arg __unused) 1366 { 1367 SLIST_HEAD(, exec_args_kva) head; 1368 struct exec_args_kva *argkva; 1369 u_int gen; 1370 int i; 1371 1372 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1373 1374 /* 1375 * Force an madvise of each KVA range. Any currently allocated ranges 1376 * will have MADV_FREE applied once they are freed. 1377 */ 1378 SLIST_INIT(&head); 1379 mtx_lock(&exec_args_kva_mtx); 1380 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1381 mtx_unlock(&exec_args_kva_mtx); 1382 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1383 SLIST_REMOVE_HEAD(&head, next); 1384 exec_release_args_kva(argkva, gen); 1385 } 1386 1387 CPU_FOREACH(i) { 1388 argkva = (void *)atomic_readandclear_ptr( 1389 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1390 if (argkva != NULL) 1391 exec_release_args_kva(argkva, gen); 1392 } 1393 } 1394 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1395 EVENTHANDLER_PRI_ANY); 1396 1397 /* 1398 * Allocate temporary demand-paged, zero-filled memory for the file name, 1399 * argument, and environment strings. 1400 */ 1401 int 1402 exec_alloc_args(struct image_args *args) 1403 { 1404 1405 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1406 return (0); 1407 } 1408 1409 void 1410 exec_free_args(struct image_args *args) 1411 { 1412 1413 if (args->buf != NULL) { 1414 exec_free_args_kva(args->bufkva); 1415 args->buf = NULL; 1416 } 1417 if (args->fname_buf != NULL) { 1418 free(args->fname_buf, M_TEMP); 1419 args->fname_buf = NULL; 1420 } 1421 if (args->fdp != NULL) 1422 fdescfree_remapped(args->fdp); 1423 } 1424 1425 /* 1426 * A set to functions to fill struct image args. 1427 * 1428 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1429 * fname) before the other functions. All exec_args_add_arg() calls must 1430 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1431 * may be called any time after exec_args_add_fname(). 1432 * 1433 * exec_args_add_fname() - install path to be executed 1434 * exec_args_add_arg() - append an argument string 1435 * exec_args_add_env() - append an env string 1436 * exec_args_adjust_args() - adjust location of the argument list to 1437 * allow new arguments to be prepended 1438 */ 1439 int 1440 exec_args_add_fname(struct image_args *args, const char *fname, 1441 enum uio_seg segflg) 1442 { 1443 int error; 1444 size_t length; 1445 1446 KASSERT(args->fname == NULL, ("fname already appended")); 1447 KASSERT(args->endp == NULL, ("already appending to args")); 1448 1449 if (fname != NULL) { 1450 args->fname = args->buf; 1451 error = segflg == UIO_SYSSPACE ? 1452 copystr(fname, args->fname, PATH_MAX, &length) : 1453 copyinstr(fname, args->fname, PATH_MAX, &length); 1454 if (error != 0) 1455 return (error == ENAMETOOLONG ? E2BIG : error); 1456 } else 1457 length = 0; 1458 1459 /* Set up for _arg_*()/_env_*() */ 1460 args->endp = args->buf + length; 1461 /* begin_argv must be set and kept updated */ 1462 args->begin_argv = args->endp; 1463 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1464 ("too little space remaining for arguments %zu < %zu", 1465 exec_map_entry_size - length, (size_t)ARG_MAX)); 1466 args->stringspace = ARG_MAX; 1467 1468 return (0); 1469 } 1470 1471 static int 1472 exec_args_add_str(struct image_args *args, const char *str, 1473 enum uio_seg segflg, int *countp) 1474 { 1475 int error; 1476 size_t length; 1477 1478 KASSERT(args->endp != NULL, ("endp not initialized")); 1479 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1480 1481 error = (segflg == UIO_SYSSPACE) ? 1482 copystr(str, args->endp, args->stringspace, &length) : 1483 copyinstr(str, args->endp, args->stringspace, &length); 1484 if (error != 0) 1485 return (error == ENAMETOOLONG ? E2BIG : error); 1486 args->stringspace -= length; 1487 args->endp += length; 1488 (*countp)++; 1489 1490 return (0); 1491 } 1492 1493 int 1494 exec_args_add_arg(struct image_args *args, const char *argp, 1495 enum uio_seg segflg) 1496 { 1497 1498 KASSERT(args->envc == 0, ("appending args after env")); 1499 1500 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1501 } 1502 1503 int 1504 exec_args_add_env(struct image_args *args, const char *envp, 1505 enum uio_seg segflg) 1506 { 1507 1508 if (args->envc == 0) 1509 args->begin_envv = args->endp; 1510 1511 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1512 } 1513 1514 int 1515 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1516 { 1517 ssize_t offset; 1518 1519 KASSERT(args->endp != NULL, ("endp not initialized")); 1520 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1521 1522 offset = extend - consume; 1523 if (args->stringspace < offset) 1524 return (E2BIG); 1525 memmove(args->begin_argv + extend, args->begin_argv + consume, 1526 args->endp - args->begin_argv + consume); 1527 if (args->envc > 0) 1528 args->begin_envv += offset; 1529 args->endp += offset; 1530 args->stringspace -= offset; 1531 return (0); 1532 } 1533 1534 char * 1535 exec_args_get_begin_envv(struct image_args *args) 1536 { 1537 1538 KASSERT(args->endp != NULL, ("endp not initialized")); 1539 1540 if (args->envc > 0) 1541 return (args->begin_envv); 1542 return (args->endp); 1543 } 1544 1545 void 1546 exec_stackgap(struct image_params *imgp, uintptr_t *dp) 1547 { 1548 if (imgp->sysent->sv_stackgap == NULL || 1549 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE | 1550 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 || 1551 (imgp->map_flags & MAP_ASLR) == 0) 1552 return; 1553 imgp->sysent->sv_stackgap(imgp, dp); 1554 } 1555 1556 /* 1557 * Copy strings out to the new process address space, constructing new arg 1558 * and env vector tables. Return a pointer to the base so that it can be used 1559 * as the initial stack pointer. 1560 */ 1561 int 1562 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1563 { 1564 int argc, envc; 1565 char **vectp; 1566 char *stringp; 1567 uintptr_t destp, ustringp; 1568 struct ps_strings *arginfo; 1569 struct proc *p; 1570 size_t execpath_len; 1571 int error, szsigcode, szps; 1572 char canary[sizeof(long) * 8]; 1573 1574 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1575 /* 1576 * Calculate string base and vector table pointers. 1577 * Also deal with signal trampoline code for this exec type. 1578 */ 1579 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1580 execpath_len = strlen(imgp->execpath) + 1; 1581 else 1582 execpath_len = 0; 1583 p = imgp->proc; 1584 szsigcode = 0; 1585 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1586 imgp->ps_strings = arginfo; 1587 if (p->p_sysent->sv_sigcode_base == 0) { 1588 if (p->p_sysent->sv_szsigcode != NULL) 1589 szsigcode = *(p->p_sysent->sv_szsigcode); 1590 } 1591 destp = (uintptr_t)arginfo; 1592 1593 /* 1594 * install sigcode 1595 */ 1596 if (szsigcode != 0) { 1597 destp -= szsigcode; 1598 destp = rounddown2(destp, sizeof(void *)); 1599 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1600 szsigcode); 1601 if (error != 0) 1602 return (error); 1603 } 1604 1605 /* 1606 * Copy the image path for the rtld. 1607 */ 1608 if (execpath_len != 0) { 1609 destp -= execpath_len; 1610 destp = rounddown2(destp, sizeof(void *)); 1611 imgp->execpathp = (void *)destp; 1612 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1613 if (error != 0) 1614 return (error); 1615 } 1616 1617 /* 1618 * Prepare the canary for SSP. 1619 */ 1620 arc4rand(canary, sizeof(canary), 0); 1621 destp -= sizeof(canary); 1622 imgp->canary = (void *)destp; 1623 error = copyout(canary, imgp->canary, sizeof(canary)); 1624 if (error != 0) 1625 return (error); 1626 imgp->canarylen = sizeof(canary); 1627 1628 /* 1629 * Prepare the pagesizes array. 1630 */ 1631 destp -= szps; 1632 destp = rounddown2(destp, sizeof(void *)); 1633 imgp->pagesizes = (void *)destp; 1634 error = copyout(pagesizes, imgp->pagesizes, szps); 1635 if (error != 0) 1636 return (error); 1637 imgp->pagesizeslen = szps; 1638 1639 /* 1640 * Allocate room for the argument and environment strings. 1641 */ 1642 destp -= ARG_MAX - imgp->args->stringspace; 1643 destp = rounddown2(destp, sizeof(void *)); 1644 ustringp = destp; 1645 1646 exec_stackgap(imgp, &destp); 1647 1648 if (imgp->auxargs) { 1649 /* 1650 * Allocate room on the stack for the ELF auxargs 1651 * array. It has up to AT_COUNT entries. 1652 */ 1653 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1654 destp = rounddown2(destp, sizeof(void *)); 1655 } 1656 1657 vectp = (char **)destp; 1658 1659 /* 1660 * Allocate room for the argv[] and env vectors including the 1661 * terminating NULL pointers. 1662 */ 1663 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1664 1665 /* 1666 * vectp also becomes our initial stack base 1667 */ 1668 *stack_base = (uintptr_t)vectp; 1669 1670 stringp = imgp->args->begin_argv; 1671 argc = imgp->args->argc; 1672 envc = imgp->args->envc; 1673 1674 /* 1675 * Copy out strings - arguments and environment. 1676 */ 1677 error = copyout(stringp, (void *)ustringp, 1678 ARG_MAX - imgp->args->stringspace); 1679 if (error != 0) 1680 return (error); 1681 1682 /* 1683 * Fill in "ps_strings" struct for ps, w, etc. 1684 */ 1685 imgp->argv = vectp; 1686 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1687 suword32(&arginfo->ps_nargvstr, argc) != 0) 1688 return (EFAULT); 1689 1690 /* 1691 * Fill in argument portion of vector table. 1692 */ 1693 for (; argc > 0; --argc) { 1694 if (suword(vectp++, ustringp) != 0) 1695 return (EFAULT); 1696 while (*stringp++ != 0) 1697 ustringp++; 1698 ustringp++; 1699 } 1700 1701 /* a null vector table pointer separates the argp's from the envp's */ 1702 if (suword(vectp++, 0) != 0) 1703 return (EFAULT); 1704 1705 imgp->envv = vectp; 1706 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1707 suword32(&arginfo->ps_nenvstr, envc) != 0) 1708 return (EFAULT); 1709 1710 /* 1711 * Fill in environment portion of vector table. 1712 */ 1713 for (; envc > 0; --envc) { 1714 if (suword(vectp++, ustringp) != 0) 1715 return (EFAULT); 1716 while (*stringp++ != 0) 1717 ustringp++; 1718 ustringp++; 1719 } 1720 1721 /* end of vector table is a null pointer */ 1722 if (suword(vectp, 0) != 0) 1723 return (EFAULT); 1724 1725 if (imgp->auxargs) { 1726 vectp++; 1727 error = imgp->sysent->sv_copyout_auxargs(imgp, 1728 (uintptr_t)vectp); 1729 if (error != 0) 1730 return (error); 1731 } 1732 1733 return (0); 1734 } 1735 1736 /* 1737 * Check permissions of file to execute. 1738 * Called with imgp->vp locked. 1739 * Return 0 for success or error code on failure. 1740 */ 1741 int 1742 exec_check_permissions(struct image_params *imgp) 1743 { 1744 struct vnode *vp = imgp->vp; 1745 struct vattr *attr = imgp->attr; 1746 struct thread *td; 1747 int error; 1748 1749 td = curthread; 1750 1751 /* Get file attributes */ 1752 error = VOP_GETATTR(vp, attr, td->td_ucred); 1753 if (error) 1754 return (error); 1755 1756 #ifdef MAC 1757 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1758 if (error) 1759 return (error); 1760 #endif 1761 1762 /* 1763 * 1) Check if file execution is disabled for the filesystem that 1764 * this file resides on. 1765 * 2) Ensure that at least one execute bit is on. Otherwise, a 1766 * privileged user will always succeed, and we don't want this 1767 * to happen unless the file really is executable. 1768 * 3) Ensure that the file is a regular file. 1769 */ 1770 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1771 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1772 (attr->va_type != VREG)) 1773 return (EACCES); 1774 1775 /* 1776 * Zero length files can't be exec'd 1777 */ 1778 if (attr->va_size == 0) 1779 return (ENOEXEC); 1780 1781 /* 1782 * Check for execute permission to file based on current credentials. 1783 */ 1784 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1785 if (error) 1786 return (error); 1787 1788 /* 1789 * Check number of open-for-writes on the file and deny execution 1790 * if there are any. 1791 * 1792 * Add a text reference now so no one can write to the 1793 * executable while we're activating it. 1794 * 1795 * Remember if this was set before and unset it in case this is not 1796 * actually an executable image. 1797 */ 1798 error = VOP_SET_TEXT(vp); 1799 if (error != 0) 1800 return (error); 1801 imgp->textset = true; 1802 1803 /* 1804 * Call filesystem specific open routine (which does nothing in the 1805 * general case). 1806 */ 1807 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1808 if (error == 0) 1809 imgp->opened = 1; 1810 return (error); 1811 } 1812 1813 /* 1814 * Exec handler registration 1815 */ 1816 int 1817 exec_register(const struct execsw *execsw_arg) 1818 { 1819 const struct execsw **es, **xs, **newexecsw; 1820 u_int count = 2; /* New slot and trailing NULL */ 1821 1822 if (execsw) 1823 for (es = execsw; *es; es++) 1824 count++; 1825 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1826 xs = newexecsw; 1827 if (execsw) 1828 for (es = execsw; *es; es++) 1829 *xs++ = *es; 1830 *xs++ = execsw_arg; 1831 *xs = NULL; 1832 if (execsw) 1833 free(execsw, M_TEMP); 1834 execsw = newexecsw; 1835 return (0); 1836 } 1837 1838 int 1839 exec_unregister(const struct execsw *execsw_arg) 1840 { 1841 const struct execsw **es, **xs, **newexecsw; 1842 int count = 1; 1843 1844 if (execsw == NULL) 1845 panic("unregister with no handlers left?\n"); 1846 1847 for (es = execsw; *es; es++) { 1848 if (*es == execsw_arg) 1849 break; 1850 } 1851 if (*es == NULL) 1852 return (ENOENT); 1853 for (es = execsw; *es; es++) 1854 if (*es != execsw_arg) 1855 count++; 1856 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1857 xs = newexecsw; 1858 for (es = execsw; *es; es++) 1859 if (*es != execsw_arg) 1860 *xs++ = *es; 1861 *xs = NULL; 1862 if (execsw) 1863 free(execsw, M_TEMP); 1864 execsw = newexecsw; 1865 return (0); 1866 } 1867 1868 /* 1869 * Write out a core segment to the compression stream. 1870 */ 1871 static int 1872 compress_chunk(struct coredump_params *cp, char *base, char *buf, u_int len) 1873 { 1874 u_int chunk_len; 1875 int error; 1876 1877 while (len > 0) { 1878 chunk_len = MIN(len, CORE_BUF_SIZE); 1879 1880 /* 1881 * We can get EFAULT error here. 1882 * In that case zero out the current chunk of the segment. 1883 */ 1884 error = copyin(base, buf, chunk_len); 1885 if (error != 0) 1886 bzero(buf, chunk_len); 1887 error = compressor_write(cp->comp, buf, chunk_len); 1888 if (error != 0) 1889 break; 1890 base += chunk_len; 1891 len -= chunk_len; 1892 } 1893 return (error); 1894 } 1895 1896 int 1897 core_write(struct coredump_params *cp, const void *base, size_t len, 1898 off_t offset, enum uio_seg seg, size_t *resid) 1899 { 1900 1901 return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base), 1902 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1903 cp->active_cred, cp->file_cred, resid, cp->td)); 1904 } 1905 1906 int 1907 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp, 1908 void *tmpbuf) 1909 { 1910 vm_map_t map; 1911 struct mount *mp; 1912 size_t resid, runlen; 1913 int error; 1914 bool success; 1915 1916 KASSERT((uintptr_t)base % PAGE_SIZE == 0, 1917 ("%s: user address %p is not page-aligned", __func__, base)); 1918 1919 if (cp->comp != NULL) 1920 return (compress_chunk(cp, base, tmpbuf, len)); 1921 1922 map = &cp->td->td_proc->p_vmspace->vm_map; 1923 for (; len > 0; base += runlen, offset += runlen, len -= runlen) { 1924 /* 1925 * Attempt to page in all virtual pages in the range. If a 1926 * virtual page is not backed by the pager, it is represented as 1927 * a hole in the file. This can occur with zero-filled 1928 * anonymous memory or truncated files, for example. 1929 */ 1930 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) { 1931 error = vm_fault(map, (uintptr_t)base + runlen, 1932 VM_PROT_READ, VM_FAULT_NOFILL, NULL); 1933 if (runlen == 0) 1934 success = error == KERN_SUCCESS; 1935 else if ((error == KERN_SUCCESS) != success) 1936 break; 1937 } 1938 1939 if (success) { 1940 error = core_write(cp, base, runlen, offset, 1941 UIO_USERSPACE, &resid); 1942 if (error != 0) { 1943 if (error != EFAULT) 1944 break; 1945 1946 /* 1947 * EFAULT may be returned if the user mapping 1948 * could not be accessed, e.g., because a mapped 1949 * file has been truncated. Skip the page if no 1950 * progress was made, to protect against a 1951 * hypothetical scenario where vm_fault() was 1952 * successful but core_write() returns EFAULT 1953 * anyway. 1954 */ 1955 runlen -= resid; 1956 if (runlen == 0) { 1957 success = false; 1958 runlen = PAGE_SIZE; 1959 } 1960 } 1961 } 1962 if (!success) { 1963 error = vn_start_write(cp->vp, &mp, V_WAIT); 1964 if (error != 0) 1965 break; 1966 vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY); 1967 error = vn_truncate_locked(cp->vp, offset + runlen, 1968 false, cp->td->td_ucred); 1969 VOP_UNLOCK(cp->vp); 1970 vn_finished_write(mp); 1971 if (error != 0) 1972 break; 1973 } 1974 } 1975 return (error); 1976 } 1977 1978 /* 1979 * Drain into a core file. 1980 */ 1981 int 1982 sbuf_drain_core_output(void *arg, const char *data, int len) 1983 { 1984 struct coredump_params *cp; 1985 struct proc *p; 1986 int error, locked; 1987 1988 cp = arg; 1989 p = cp->td->td_proc; 1990 1991 /* 1992 * Some kern_proc out routines that print to this sbuf may 1993 * call us with the process lock held. Draining with the 1994 * non-sleepable lock held is unsafe. The lock is needed for 1995 * those routines when dumping a live process. In our case we 1996 * can safely release the lock before draining and acquire 1997 * again after. 1998 */ 1999 locked = PROC_LOCKED(p); 2000 if (locked) 2001 PROC_UNLOCK(p); 2002 if (cp->comp != NULL) 2003 error = compressor_write(cp->comp, __DECONST(char *, data), len); 2004 else 2005 error = core_write(cp, __DECONST(void *, data), len, cp->offset, 2006 UIO_SYSSPACE, NULL); 2007 if (locked) 2008 PROC_LOCK(p); 2009 if (error != 0) 2010 return (-error); 2011 cp->offset += len; 2012 return (len); 2013 } 2014