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/reg.h> 60 #include <sys/resourcevar.h> 61 #include <sys/rwlock.h> 62 #include <sys/sched.h> 63 #include <sys/sdt.h> 64 #include <sys/sf_buf.h> 65 #include <sys/shm.h> 66 #include <sys/signalvar.h> 67 #include <sys/smp.h> 68 #include <sys/stat.h> 69 #include <sys/syscallsubr.h> 70 #include <sys/sysctl.h> 71 #include <sys/sysent.h> 72 #include <sys/sysproto.h> 73 #include <sys/timers.h> 74 #include <sys/umtxvar.h> 75 #include <sys/vnode.h> 76 #include <sys/wait.h> 77 #ifdef KTRACE 78 #include <sys/ktrace.h> 79 #endif 80 81 #include <vm/vm.h> 82 #include <vm/vm_param.h> 83 #include <vm/pmap.h> 84 #include <vm/vm_page.h> 85 #include <vm/vm_map.h> 86 #include <vm/vm_kern.h> 87 #include <vm/vm_extern.h> 88 #include <vm/vm_object.h> 89 #include <vm/vm_pager.h> 90 91 #ifdef HWPMC_HOOKS 92 #include <sys/pmckern.h> 93 #endif 94 95 #include <security/audit/audit.h> 96 #include <security/mac/mac_framework.h> 97 98 #ifdef KDTRACE_HOOKS 99 #include <sys/dtrace_bsd.h> 100 dtrace_execexit_func_t dtrace_fasttrap_exec; 101 #endif 102 103 SDT_PROVIDER_DECLARE(proc); 104 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 105 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 106 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 107 108 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 109 110 int coredump_pack_fileinfo = 1; 111 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 112 &coredump_pack_fileinfo, 0, 113 "Enable file path packing in 'procstat -f' coredump notes"); 114 115 int coredump_pack_vmmapinfo = 1; 116 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 117 &coredump_pack_vmmapinfo, 0, 118 "Enable file path packing in 'procstat -v' coredump notes"); 119 120 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 121 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 122 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 123 static int do_execve(struct thread *td, struct image_args *args, 124 struct mac *mac_p, struct vmspace *oldvmspace); 125 126 /* XXX This should be vm_size_t. */ 127 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 128 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", 129 "Location of process' ps_strings structure"); 130 131 /* XXX This should be vm_size_t. */ 132 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 133 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", 134 "Top of process stack"); 135 136 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 137 NULL, 0, sysctl_kern_stackprot, "I", 138 "Stack memory permissions"); 139 140 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 141 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 142 &ps_arg_cache_limit, 0, 143 "Process' command line characters cache limit"); 144 145 static int disallow_high_osrel; 146 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 147 &disallow_high_osrel, 0, 148 "Disallow execution of binaries built for higher version of the world"); 149 150 static int map_at_zero = 0; 151 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 152 "Permit processes to map an object at virtual address 0."); 153 154 static int core_dump_can_intr = 1; 155 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN, 156 &core_dump_can_intr, 0, 157 "Core dumping interruptible with SIGKILL"); 158 159 static int 160 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 161 { 162 struct proc *p; 163 int error; 164 165 p = curproc; 166 #ifdef SCTL_MASK32 167 if (req->flags & SCTL_MASK32) { 168 unsigned int val; 169 val = (unsigned int)p->p_sysent->sv_psstrings; 170 error = SYSCTL_OUT(req, &val, sizeof(val)); 171 } else 172 #endif 173 error = SYSCTL_OUT(req, &p->p_sysent->sv_psstrings, 174 sizeof(p->p_sysent->sv_psstrings)); 175 return error; 176 } 177 178 static int 179 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 180 { 181 struct proc *p; 182 int error; 183 184 p = curproc; 185 #ifdef SCTL_MASK32 186 if (req->flags & SCTL_MASK32) { 187 unsigned int val; 188 val = (unsigned int)p->p_sysent->sv_usrstack; 189 error = SYSCTL_OUT(req, &val, sizeof(val)); 190 } else 191 #endif 192 error = SYSCTL_OUT(req, &p->p_sysent->sv_usrstack, 193 sizeof(p->p_sysent->sv_usrstack)); 194 return error; 195 } 196 197 static int 198 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 199 { 200 struct proc *p; 201 202 p = curproc; 203 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 204 sizeof(p->p_sysent->sv_stackprot))); 205 } 206 207 /* 208 * Each of the items is a pointer to a `const struct execsw', hence the 209 * double pointer here. 210 */ 211 static const struct execsw **execsw; 212 213 #ifndef _SYS_SYSPROTO_H_ 214 struct execve_args { 215 char *fname; 216 char **argv; 217 char **envv; 218 }; 219 #endif 220 221 int 222 sys_execve(struct thread *td, struct execve_args *uap) 223 { 224 struct image_args args; 225 struct vmspace *oldvmspace; 226 int error; 227 228 error = pre_execve(td, &oldvmspace); 229 if (error != 0) 230 return (error); 231 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 232 uap->argv, uap->envv); 233 if (error == 0) 234 error = kern_execve(td, &args, NULL, oldvmspace); 235 post_execve(td, error, oldvmspace); 236 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 237 return (error); 238 } 239 240 #ifndef _SYS_SYSPROTO_H_ 241 struct fexecve_args { 242 int fd; 243 char **argv; 244 char **envv; 245 }; 246 #endif 247 int 248 sys_fexecve(struct thread *td, struct fexecve_args *uap) 249 { 250 struct image_args args; 251 struct vmspace *oldvmspace; 252 int error; 253 254 error = pre_execve(td, &oldvmspace); 255 if (error != 0) 256 return (error); 257 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 258 uap->argv, uap->envv); 259 if (error == 0) { 260 args.fd = uap->fd; 261 error = kern_execve(td, &args, NULL, oldvmspace); 262 } 263 post_execve(td, error, oldvmspace); 264 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 265 return (error); 266 } 267 268 #ifndef _SYS_SYSPROTO_H_ 269 struct __mac_execve_args { 270 char *fname; 271 char **argv; 272 char **envv; 273 struct mac *mac_p; 274 }; 275 #endif 276 277 int 278 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 279 { 280 #ifdef MAC 281 struct image_args args; 282 struct vmspace *oldvmspace; 283 int error; 284 285 error = pre_execve(td, &oldvmspace); 286 if (error != 0) 287 return (error); 288 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 289 uap->argv, uap->envv); 290 if (error == 0) 291 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 292 post_execve(td, error, oldvmspace); 293 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 294 return (error); 295 #else 296 return (ENOSYS); 297 #endif 298 } 299 300 int 301 pre_execve(struct thread *td, struct vmspace **oldvmspace) 302 { 303 struct proc *p; 304 int error; 305 306 KASSERT(td == curthread, ("non-current thread %p", td)); 307 error = 0; 308 p = td->td_proc; 309 if ((p->p_flag & P_HADTHREADS) != 0) { 310 PROC_LOCK(p); 311 if (thread_single(p, SINGLE_BOUNDARY) != 0) 312 error = ERESTART; 313 PROC_UNLOCK(p); 314 } 315 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 316 ("nested execve")); 317 *oldvmspace = p->p_vmspace; 318 return (error); 319 } 320 321 void 322 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 323 { 324 struct proc *p; 325 326 KASSERT(td == curthread, ("non-current thread %p", td)); 327 p = td->td_proc; 328 if ((p->p_flag & P_HADTHREADS) != 0) { 329 PROC_LOCK(p); 330 /* 331 * If success, we upgrade to SINGLE_EXIT state to 332 * force other threads to suicide. 333 */ 334 if (error == EJUSTRETURN) 335 thread_single(p, SINGLE_EXIT); 336 else 337 thread_single_end(p, SINGLE_BOUNDARY); 338 PROC_UNLOCK(p); 339 } 340 exec_cleanup(td, oldvmspace); 341 } 342 343 /* 344 * kern_execve() has the astonishing property of not always returning to 345 * the caller. If sufficiently bad things happen during the call to 346 * do_execve(), it can end up calling exit1(); as a result, callers must 347 * avoid doing anything which they might need to undo (e.g., allocating 348 * memory). 349 */ 350 int 351 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 352 struct vmspace *oldvmspace) 353 { 354 355 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 356 exec_args_get_begin_envv(args) - args->begin_argv); 357 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 358 args->endp - exec_args_get_begin_envv(args)); 359 return (do_execve(td, args, mac_p, oldvmspace)); 360 } 361 362 static void 363 execve_nosetid(struct image_params *imgp) 364 { 365 imgp->credential_setid = false; 366 if (imgp->newcred != NULL) { 367 crfree(imgp->newcred); 368 imgp->newcred = NULL; 369 } 370 } 371 372 /* 373 * In-kernel implementation of execve(). All arguments are assumed to be 374 * userspace pointers from the passed thread. 375 */ 376 static int 377 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 378 struct vmspace *oldvmspace) 379 { 380 struct proc *p = td->td_proc; 381 struct nameidata nd; 382 struct ucred *oldcred; 383 struct uidinfo *euip = NULL; 384 uintptr_t stack_base; 385 struct image_params image_params, *imgp; 386 struct vattr attr; 387 int (*img_first)(struct image_params *); 388 struct pargs *oldargs = NULL, *newargs = NULL; 389 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 390 #ifdef KTRACE 391 struct ktr_io_params *kiop; 392 #endif 393 struct vnode *oldtextvp = NULL, *newtextvp; 394 int credential_changing; 395 #ifdef MAC 396 struct label *interpvplabel = NULL; 397 int will_transition; 398 #endif 399 #ifdef HWPMC_HOOKS 400 struct pmckern_procexec pe; 401 #endif 402 int error, i, orig_osrel; 403 uint32_t orig_fctl0; 404 Elf_Brandinfo *orig_brandinfo; 405 static const char fexecv_proc_title[] = "(fexecv)"; 406 407 imgp = &image_params; 408 #ifdef KTRACE 409 kiop = NULL; 410 #endif 411 412 /* 413 * Lock the process and set the P_INEXEC flag to indicate that 414 * it should be left alone until we're done here. This is 415 * necessary to avoid race conditions - e.g. in ptrace() - 416 * that might allow a local user to illicitly obtain elevated 417 * privileges. 418 */ 419 PROC_LOCK(p); 420 KASSERT((p->p_flag & P_INEXEC) == 0, 421 ("%s(): process already has P_INEXEC flag", __func__)); 422 p->p_flag |= P_INEXEC; 423 PROC_UNLOCK(p); 424 425 /* 426 * Initialize part of the common data 427 */ 428 bzero(imgp, sizeof(*imgp)); 429 imgp->proc = p; 430 imgp->attr = &attr; 431 imgp->args = args; 432 oldcred = p->p_ucred; 433 orig_osrel = p->p_osrel; 434 orig_fctl0 = p->p_fctl0; 435 orig_brandinfo = p->p_elf_brandinfo; 436 437 #ifdef MAC 438 error = mac_execve_enter(imgp, mac_p); 439 if (error) 440 goto exec_fail; 441 #endif 442 443 /* 444 * Translate the file name. namei() returns a vnode pointer 445 * in ni_vp among other things. 446 * 447 * XXXAUDIT: It would be desirable to also audit the name of the 448 * interpreter if this is an interpreted binary. 449 */ 450 if (args->fname != NULL) { 451 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 452 SAVENAME | AUDITVNODE1, UIO_SYSSPACE, args->fname, td); 453 } 454 455 SDT_PROBE1(proc, , , exec, args->fname); 456 457 interpret: 458 if (args->fname != NULL) { 459 #ifdef CAPABILITY_MODE 460 /* 461 * While capability mode can't reach this point via direct 462 * path arguments to execve(), we also don't allow 463 * interpreters to be used in capability mode (for now). 464 * Catch indirect lookups and return a permissions error. 465 */ 466 if (IN_CAPABILITY_MODE(td)) { 467 error = ECAPMODE; 468 goto exec_fail; 469 } 470 #endif 471 error = namei(&nd); 472 if (error) 473 goto exec_fail; 474 475 newtextvp = nd.ni_vp; 476 imgp->vp = newtextvp; 477 } else { 478 AUDIT_ARG_FD(args->fd); 479 /* 480 * Descriptors opened only with O_EXEC or O_RDONLY are allowed. 481 */ 482 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, &newtextvp); 483 if (error) 484 goto exec_fail; 485 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 486 AUDIT_ARG_VNODE1(newtextvp); 487 imgp->vp = newtextvp; 488 } 489 490 /* 491 * Check file permissions. Also 'opens' file and sets its vnode to 492 * text mode. 493 */ 494 error = exec_check_permissions(imgp); 495 if (error) 496 goto exec_fail_dealloc; 497 498 imgp->object = imgp->vp->v_object; 499 if (imgp->object != NULL) 500 vm_object_reference(imgp->object); 501 502 error = exec_map_first_page(imgp); 503 if (error) 504 goto exec_fail_dealloc; 505 506 imgp->proc->p_osrel = 0; 507 imgp->proc->p_fctl0 = 0; 508 imgp->proc->p_elf_brandinfo = NULL; 509 510 /* 511 * Implement image setuid/setgid. 512 * 513 * Determine new credentials before attempting image activators 514 * so that it can be used by process_exec handlers to determine 515 * credential/setid changes. 516 * 517 * Don't honor setuid/setgid if the filesystem prohibits it or if 518 * the process is being traced. 519 * 520 * We disable setuid/setgid/etc in capability mode on the basis 521 * that most setugid applications are not written with that 522 * environment in mind, and will therefore almost certainly operate 523 * incorrectly. In principle there's no reason that setugid 524 * applications might not be useful in capability mode, so we may want 525 * to reconsider this conservative design choice in the future. 526 * 527 * XXXMAC: For the time being, use NOSUID to also prohibit 528 * transitions on the file system. 529 */ 530 credential_changing = 0; 531 credential_changing |= (attr.va_mode & S_ISUID) && 532 oldcred->cr_uid != attr.va_uid; 533 credential_changing |= (attr.va_mode & S_ISGID) && 534 oldcred->cr_gid != attr.va_gid; 535 #ifdef MAC 536 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 537 interpvplabel, imgp); 538 credential_changing |= will_transition; 539 #endif 540 541 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 542 if (credential_changing) 543 imgp->proc->p_pdeathsig = 0; 544 545 if (credential_changing && 546 #ifdef CAPABILITY_MODE 547 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 548 #endif 549 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 550 (p->p_flag & P_TRACED) == 0) { 551 imgp->credential_setid = true; 552 VOP_UNLOCK(imgp->vp); 553 imgp->newcred = crdup(oldcred); 554 if (attr.va_mode & S_ISUID) { 555 euip = uifind(attr.va_uid); 556 change_euid(imgp->newcred, euip); 557 } 558 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 559 if (attr.va_mode & S_ISGID) 560 change_egid(imgp->newcred, attr.va_gid); 561 /* 562 * Implement correct POSIX saved-id behavior. 563 * 564 * XXXMAC: Note that the current logic will save the 565 * uid and gid if a MAC domain transition occurs, even 566 * though maybe it shouldn't. 567 */ 568 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 569 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 570 } else { 571 /* 572 * Implement correct POSIX saved-id behavior. 573 * 574 * XXX: It's not clear that the existing behavior is 575 * POSIX-compliant. A number of sources indicate that the 576 * saved uid/gid should only be updated if the new ruid is 577 * not equal to the old ruid, or the new euid is not equal 578 * to the old euid and the new euid is not equal to the old 579 * ruid. The FreeBSD code always updates the saved uid/gid. 580 * Also, this code uses the new (replaced) euid and egid as 581 * the source, which may or may not be the right ones to use. 582 */ 583 if (oldcred->cr_svuid != oldcred->cr_uid || 584 oldcred->cr_svgid != oldcred->cr_gid) { 585 VOP_UNLOCK(imgp->vp); 586 imgp->newcred = crdup(oldcred); 587 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 588 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 589 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 590 } 591 } 592 /* The new credentials are installed into the process later. */ 593 594 /* 595 * Do the best to calculate the full path to the image file. 596 */ 597 if (args->fname != NULL && args->fname[0] == '/') 598 imgp->execpath = args->fname; 599 else { 600 VOP_UNLOCK(imgp->vp); 601 if (vn_fullpath(imgp->vp, &imgp->execpath, &imgp->freepath) != 0) 602 imgp->execpath = args->fname; 603 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 604 } 605 606 /* 607 * If the current process has a special image activator it 608 * wants to try first, call it. For example, emulating shell 609 * scripts differently. 610 */ 611 error = -1; 612 if ((img_first = imgp->proc->p_sysent->sv_imgact_try) != NULL) 613 error = img_first(imgp); 614 615 /* 616 * Loop through the list of image activators, calling each one. 617 * An activator returns -1 if there is no match, 0 on success, 618 * and an error otherwise. 619 */ 620 for (i = 0; error == -1 && execsw[i]; ++i) { 621 if (execsw[i]->ex_imgact == NULL || 622 execsw[i]->ex_imgact == img_first) { 623 continue; 624 } 625 error = (*execsw[i]->ex_imgact)(imgp); 626 } 627 628 if (error) { 629 if (error == -1) 630 error = ENOEXEC; 631 goto exec_fail_dealloc; 632 } 633 634 /* 635 * Special interpreter operation, cleanup and loop up to try to 636 * activate the interpreter. 637 */ 638 if (imgp->interpreted) { 639 exec_unmap_first_page(imgp); 640 /* 641 * The text reference needs to be removed for scripts. 642 * There is a short period before we determine that 643 * something is a script where text reference is active. 644 * The vnode lock is held over this entire period 645 * so nothing should illegitimately be blocked. 646 */ 647 MPASS(imgp->textset); 648 VOP_UNSET_TEXT_CHECKED(newtextvp); 649 imgp->textset = false; 650 /* free name buffer and old vnode */ 651 if (args->fname != NULL) 652 NDFREE(&nd, NDF_ONLY_PNBUF); 653 #ifdef MAC 654 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 655 #endif 656 if (imgp->opened) { 657 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 658 imgp->opened = 0; 659 } 660 vput(newtextvp); 661 vm_object_deallocate(imgp->object); 662 imgp->object = NULL; 663 execve_nosetid(imgp); 664 imgp->execpath = NULL; 665 free(imgp->freepath, M_TEMP); 666 imgp->freepath = NULL; 667 /* set new name to that of the interpreter */ 668 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 669 SAVENAME, UIO_SYSSPACE, imgp->interpreter_name, td); 670 args->fname = imgp->interpreter_name; 671 goto interpret; 672 } 673 674 /* 675 * NB: We unlock the vnode here because it is believed that none 676 * of the sv_copyout_strings/sv_fixup operations require the vnode. 677 */ 678 VOP_UNLOCK(imgp->vp); 679 680 if (disallow_high_osrel && 681 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 682 error = ENOEXEC; 683 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 684 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 685 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 686 goto exec_fail_dealloc; 687 } 688 689 /* 690 * Copy out strings (args and env) and initialize stack base. 691 */ 692 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 693 if (error != 0) { 694 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 695 goto exec_fail_dealloc; 696 } 697 698 /* 699 * Stack setup. 700 */ 701 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 702 if (error != 0) { 703 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 704 goto exec_fail_dealloc; 705 } 706 707 if (args->fdp != NULL) { 708 /* Install a brand new file descriptor table. */ 709 fdinstall_remapped(td, args->fdp); 710 args->fdp = NULL; 711 } else { 712 /* 713 * Keep on using the existing file descriptor table. For 714 * security and other reasons, the file descriptor table 715 * cannot be shared after an exec. 716 */ 717 fdunshare(td); 718 pdunshare(td); 719 /* close files on exec */ 720 fdcloseexec(td); 721 } 722 723 /* 724 * Malloc things before we need locks. 725 */ 726 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 727 /* Cache arguments if they fit inside our allowance */ 728 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 729 newargs = pargs_alloc(i); 730 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 731 } 732 733 /* 734 * For security and other reasons, signal handlers cannot 735 * be shared after an exec. The new process gets a copy of the old 736 * handlers. In execsigs(), the new process will have its signals 737 * reset. 738 */ 739 if (sigacts_shared(p->p_sigacts)) { 740 oldsigacts = p->p_sigacts; 741 newsigacts = sigacts_alloc(); 742 sigacts_copy(newsigacts, oldsigacts); 743 } 744 745 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 746 747 PROC_LOCK(p); 748 if (oldsigacts) 749 p->p_sigacts = newsigacts; 750 /* Stop profiling */ 751 stopprofclock(p); 752 753 /* reset caught signals */ 754 execsigs(p); 755 756 /* name this process - nameiexec(p, ndp) */ 757 bzero(p->p_comm, sizeof(p->p_comm)); 758 if (args->fname) 759 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 760 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 761 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 762 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 763 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 764 #ifdef KTR 765 sched_clear_tdname(td); 766 #endif 767 768 /* 769 * mark as execed, wakeup the process that vforked (if any) and tell 770 * it that it now has its own resources back 771 */ 772 p->p_flag |= P_EXEC; 773 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 774 p->p_flag2 &= ~P2_NOTRACE; 775 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 776 p->p_flag2 &= ~P2_STKGAP_DISABLE; 777 if (p->p_flag & P_PPWAIT) { 778 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 779 cv_broadcast(&p->p_pwait); 780 /* STOPs are no longer ignored, arrange for AST */ 781 signotify(td); 782 } 783 784 if ((imgp->sysent->sv_setid_allowed != NULL && 785 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) || 786 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0) 787 execve_nosetid(imgp); 788 789 /* 790 * Implement image setuid/setgid installation. 791 */ 792 if (imgp->credential_setid) { 793 /* 794 * Turn off syscall tracing for set-id programs, except for 795 * root. Record any set-id flags first to make sure that 796 * we do not regain any tracing during a possible block. 797 */ 798 setsugid(p); 799 #ifdef KTRACE 800 kiop = ktrprocexec(p); 801 #endif 802 /* 803 * Close any file descriptors 0..2 that reference procfs, 804 * then make sure file descriptors 0..2 are in use. 805 * 806 * Both fdsetugidsafety() and fdcheckstd() may call functions 807 * taking sleepable locks, so temporarily drop our locks. 808 */ 809 PROC_UNLOCK(p); 810 VOP_UNLOCK(imgp->vp); 811 fdsetugidsafety(td); 812 error = fdcheckstd(td); 813 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 814 if (error != 0) 815 goto exec_fail_dealloc; 816 PROC_LOCK(p); 817 #ifdef MAC 818 if (will_transition) { 819 mac_vnode_execve_transition(oldcred, imgp->newcred, 820 imgp->vp, interpvplabel, imgp); 821 } 822 #endif 823 } else { 824 if (oldcred->cr_uid == oldcred->cr_ruid && 825 oldcred->cr_gid == oldcred->cr_rgid) 826 p->p_flag &= ~P_SUGID; 827 } 828 /* 829 * Set the new credentials. 830 */ 831 if (imgp->newcred != NULL) { 832 proc_set_cred(p, imgp->newcred); 833 crfree(oldcred); 834 oldcred = NULL; 835 } 836 837 /* 838 * Store the vp for use in procfs. This vnode was referenced by namei 839 * or fgetvp_exec. 840 */ 841 oldtextvp = p->p_textvp; 842 p->p_textvp = newtextvp; 843 844 #ifdef KDTRACE_HOOKS 845 /* 846 * Tell the DTrace fasttrap provider about the exec if it 847 * has declared an interest. 848 */ 849 if (dtrace_fasttrap_exec) 850 dtrace_fasttrap_exec(p); 851 #endif 852 853 /* 854 * Notify others that we exec'd, and clear the P_INEXEC flag 855 * as we're now a bona fide freshly-execed process. 856 */ 857 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 858 p->p_flag &= ~P_INEXEC; 859 860 /* clear "fork but no exec" flag, as we _are_ execing */ 861 p->p_acflag &= ~AFORK; 862 863 /* 864 * Free any previous argument cache and replace it with 865 * the new argument cache, if any. 866 */ 867 oldargs = p->p_args; 868 p->p_args = newargs; 869 newargs = NULL; 870 871 PROC_UNLOCK(p); 872 873 #ifdef HWPMC_HOOKS 874 /* 875 * Check if system-wide sampling is in effect or if the 876 * current process is using PMCs. If so, do exec() time 877 * processing. This processing needs to happen AFTER the 878 * P_INEXEC flag is cleared. 879 */ 880 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 881 VOP_UNLOCK(imgp->vp); 882 pe.pm_credentialschanged = credential_changing; 883 pe.pm_entryaddr = imgp->entry_addr; 884 885 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 886 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 887 } 888 #endif 889 890 /* Set values passed into the program in registers. */ 891 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 892 893 VOP_MMAPPED(imgp->vp); 894 895 SDT_PROBE1(proc, , , exec__success, args->fname); 896 897 exec_fail_dealloc: 898 if (error != 0) { 899 p->p_osrel = orig_osrel; 900 p->p_fctl0 = orig_fctl0; 901 p->p_elf_brandinfo = orig_brandinfo; 902 } 903 904 if (imgp->firstpage != NULL) 905 exec_unmap_first_page(imgp); 906 907 if (imgp->vp != NULL) { 908 if (args->fname) 909 NDFREE(&nd, NDF_ONLY_PNBUF); 910 if (imgp->opened) 911 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 912 if (imgp->textset) 913 VOP_UNSET_TEXT_CHECKED(imgp->vp); 914 if (error != 0) 915 vput(imgp->vp); 916 else 917 VOP_UNLOCK(imgp->vp); 918 } 919 920 if (imgp->object != NULL) 921 vm_object_deallocate(imgp->object); 922 923 free(imgp->freepath, M_TEMP); 924 925 if (error == 0) { 926 if (p->p_ptevents & PTRACE_EXEC) { 927 PROC_LOCK(p); 928 if (p->p_ptevents & PTRACE_EXEC) 929 td->td_dbgflags |= TDB_EXEC; 930 PROC_UNLOCK(p); 931 } 932 } else { 933 exec_fail: 934 /* we're done here, clear P_INEXEC */ 935 PROC_LOCK(p); 936 p->p_flag &= ~P_INEXEC; 937 PROC_UNLOCK(p); 938 939 SDT_PROBE1(proc, , , exec__failure, error); 940 } 941 942 if (imgp->newcred != NULL && oldcred != NULL) 943 crfree(imgp->newcred); 944 945 #ifdef MAC 946 mac_execve_exit(imgp); 947 mac_execve_interpreter_exit(interpvplabel); 948 #endif 949 exec_free_args(args); 950 951 /* 952 * Handle deferred decrement of ref counts. 953 */ 954 if (oldtextvp != NULL) 955 vrele(oldtextvp); 956 #ifdef KTRACE 957 ktr_io_params_free(kiop); 958 #endif 959 pargs_drop(oldargs); 960 pargs_drop(newargs); 961 if (oldsigacts != NULL) 962 sigacts_free(oldsigacts); 963 if (euip != NULL) 964 uifree(euip); 965 966 if (error && imgp->vmspace_destroyed) { 967 /* sorry, no more process anymore. exit gracefully */ 968 exec_cleanup(td, oldvmspace); 969 exit1(td, 0, SIGABRT); 970 /* NOT REACHED */ 971 } 972 973 #ifdef KTRACE 974 if (error == 0) 975 ktrprocctor(p); 976 #endif 977 978 /* 979 * We don't want cpu_set_syscall_retval() to overwrite any of 980 * the register values put in place by exec_setregs(). 981 * Implementations of cpu_set_syscall_retval() will leave 982 * registers unmodified when returning EJUSTRETURN. 983 */ 984 return (error == 0 ? EJUSTRETURN : error); 985 } 986 987 void 988 exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 989 { 990 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 991 KASSERT(td->td_proc->p_vmspace != oldvmspace, 992 ("oldvmspace still used")); 993 vmspace_free(oldvmspace); 994 td->td_pflags &= ~TDP_EXECVMSPC; 995 } 996 } 997 998 int 999 exec_map_first_page(struct image_params *imgp) 1000 { 1001 vm_object_t object; 1002 vm_page_t m; 1003 int error; 1004 1005 if (imgp->firstpage != NULL) 1006 exec_unmap_first_page(imgp); 1007 1008 object = imgp->vp->v_object; 1009 if (object == NULL) 1010 return (EACCES); 1011 #if VM_NRESERVLEVEL > 0 1012 if ((object->flags & OBJ_COLORED) == 0) { 1013 VM_OBJECT_WLOCK(object); 1014 vm_object_color(object, 0); 1015 VM_OBJECT_WUNLOCK(object); 1016 } 1017 #endif 1018 error = vm_page_grab_valid_unlocked(&m, object, 0, 1019 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1020 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1021 1022 if (error != VM_PAGER_OK) 1023 return (EIO); 1024 imgp->firstpage = sf_buf_alloc(m, 0); 1025 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1026 1027 return (0); 1028 } 1029 1030 void 1031 exec_unmap_first_page(struct image_params *imgp) 1032 { 1033 vm_page_t m; 1034 1035 if (imgp->firstpage != NULL) { 1036 m = sf_buf_page(imgp->firstpage); 1037 sf_buf_free(imgp->firstpage); 1038 imgp->firstpage = NULL; 1039 vm_page_unwire(m, PQ_ACTIVE); 1040 } 1041 } 1042 1043 void 1044 exec_onexec_old(struct thread *td) 1045 { 1046 sigfastblock_clear(td); 1047 umtx_exec(td->td_proc); 1048 } 1049 1050 /* 1051 * Destroy old address space, and allocate a new stack. 1052 * The new stack is only sgrowsiz large because it is grown 1053 * automatically on a page fault. 1054 */ 1055 int 1056 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1057 { 1058 int error; 1059 struct proc *p = imgp->proc; 1060 struct vmspace *vmspace = p->p_vmspace; 1061 struct thread *td = curthread; 1062 vm_object_t obj; 1063 struct rlimit rlim_stack; 1064 vm_offset_t sv_minuser, stack_addr; 1065 vm_map_t map; 1066 vm_prot_t stack_prot; 1067 u_long ssiz; 1068 1069 imgp->vmspace_destroyed = 1; 1070 imgp->sysent = sv; 1071 1072 if (p->p_sysent->sv_onexec_old != NULL) 1073 p->p_sysent->sv_onexec_old(td); 1074 itimers_exec(p); 1075 1076 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1077 1078 /* 1079 * Blow away entire process VM, if address space not shared, 1080 * otherwise, create a new VM space so that other threads are 1081 * not disrupted 1082 */ 1083 map = &vmspace->vm_map; 1084 if (map_at_zero) 1085 sv_minuser = sv->sv_minuser; 1086 else 1087 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1088 if (refcount_load(&vmspace->vm_refcnt) == 1 && 1089 vm_map_min(map) == sv_minuser && 1090 vm_map_max(map) == sv->sv_maxuser && 1091 cpu_exec_vmspace_reuse(p, map)) { 1092 shmexit(vmspace); 1093 pmap_remove_pages(vmspace_pmap(vmspace)); 1094 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1095 /* 1096 * An exec terminates mlockall(MCL_FUTURE). 1097 * ASLR and W^X states must be re-evaluated. 1098 */ 1099 vm_map_lock(map); 1100 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1101 MAP_ASLR_IGNSTART | MAP_WXORX); 1102 vm_map_unlock(map); 1103 } else { 1104 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1105 if (error) 1106 return (error); 1107 vmspace = p->p_vmspace; 1108 map = &vmspace->vm_map; 1109 } 1110 map->flags |= imgp->map_flags; 1111 1112 /* Map a shared page */ 1113 obj = sv->sv_shared_page_obj; 1114 if (obj != NULL) { 1115 vm_object_reference(obj); 1116 error = vm_map_fixed(map, obj, 0, 1117 sv->sv_shared_page_base, sv->sv_shared_page_len, 1118 VM_PROT_READ | VM_PROT_EXECUTE, 1119 VM_PROT_READ | VM_PROT_EXECUTE, 1120 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1121 if (error != KERN_SUCCESS) { 1122 vm_object_deallocate(obj); 1123 return (vm_mmap_to_errno(error)); 1124 } 1125 } 1126 1127 /* Allocate a new stack */ 1128 if (imgp->stack_sz != 0) { 1129 ssiz = trunc_page(imgp->stack_sz); 1130 PROC_LOCK(p); 1131 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1132 PROC_UNLOCK(p); 1133 if (ssiz > rlim_stack.rlim_max) 1134 ssiz = rlim_stack.rlim_max; 1135 if (ssiz > rlim_stack.rlim_cur) { 1136 rlim_stack.rlim_cur = ssiz; 1137 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1138 } 1139 } else if (sv->sv_maxssiz != NULL) { 1140 ssiz = *sv->sv_maxssiz; 1141 } else { 1142 ssiz = maxssiz; 1143 } 1144 imgp->eff_stack_sz = lim_cur(curthread, RLIMIT_STACK); 1145 if (ssiz < imgp->eff_stack_sz) 1146 imgp->eff_stack_sz = ssiz; 1147 stack_addr = sv->sv_usrstack - ssiz; 1148 stack_prot = obj != NULL && imgp->stack_prot != 0 ? 1149 imgp->stack_prot : sv->sv_stackprot; 1150 error = vm_map_stack(map, stack_addr, (vm_size_t)ssiz, stack_prot, 1151 VM_PROT_ALL, MAP_STACK_GROWS_DOWN); 1152 if (error != KERN_SUCCESS) { 1153 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x " 1154 "failed mach error %d errno %d\n", (uintmax_t)ssiz, 1155 stack_prot, error, vm_mmap_to_errno(error)); 1156 return (vm_mmap_to_errno(error)); 1157 } 1158 1159 /* 1160 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1161 * are still used to enforce the stack rlimit on the process stack. 1162 */ 1163 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1164 vmspace->vm_maxsaddr = (char *)stack_addr; 1165 1166 return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0); 1167 } 1168 1169 /* 1170 * Copy out argument and environment strings from the old process address 1171 * space into the temporary string buffer. 1172 */ 1173 int 1174 exec_copyin_args(struct image_args *args, const char *fname, 1175 enum uio_seg segflg, char **argv, char **envv) 1176 { 1177 u_long arg, env; 1178 int error; 1179 1180 bzero(args, sizeof(*args)); 1181 if (argv == NULL) 1182 return (EFAULT); 1183 1184 /* 1185 * Allocate demand-paged memory for the file name, argument, and 1186 * environment strings. 1187 */ 1188 error = exec_alloc_args(args); 1189 if (error != 0) 1190 return (error); 1191 1192 /* 1193 * Copy the file name. 1194 */ 1195 error = exec_args_add_fname(args, fname, segflg); 1196 if (error != 0) 1197 goto err_exit; 1198 1199 /* 1200 * extract arguments first 1201 */ 1202 for (;;) { 1203 error = fueword(argv++, &arg); 1204 if (error == -1) { 1205 error = EFAULT; 1206 goto err_exit; 1207 } 1208 if (arg == 0) 1209 break; 1210 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1211 UIO_USERSPACE); 1212 if (error != 0) 1213 goto err_exit; 1214 } 1215 1216 /* 1217 * extract environment strings 1218 */ 1219 if (envv) { 1220 for (;;) { 1221 error = fueword(envv++, &env); 1222 if (error == -1) { 1223 error = EFAULT; 1224 goto err_exit; 1225 } 1226 if (env == 0) 1227 break; 1228 error = exec_args_add_env(args, 1229 (char *)(uintptr_t)env, UIO_USERSPACE); 1230 if (error != 0) 1231 goto err_exit; 1232 } 1233 } 1234 1235 return (0); 1236 1237 err_exit: 1238 exec_free_args(args); 1239 return (error); 1240 } 1241 1242 int 1243 exec_copyin_data_fds(struct thread *td, struct image_args *args, 1244 const void *data, size_t datalen, const int *fds, size_t fdslen) 1245 { 1246 struct filedesc *ofdp; 1247 const char *p; 1248 int *kfds; 1249 int error; 1250 1251 memset(args, '\0', sizeof(*args)); 1252 ofdp = td->td_proc->p_fd; 1253 if (datalen >= ARG_MAX || fdslen >= ofdp->fd_nfiles) 1254 return (E2BIG); 1255 error = exec_alloc_args(args); 1256 if (error != 0) 1257 return (error); 1258 1259 args->begin_argv = args->buf; 1260 args->stringspace = ARG_MAX; 1261 1262 if (datalen > 0) { 1263 /* 1264 * Argument buffer has been provided. Copy it into the 1265 * kernel as a single string and add a terminating null 1266 * byte. 1267 */ 1268 error = copyin(data, args->begin_argv, datalen); 1269 if (error != 0) 1270 goto err_exit; 1271 args->begin_argv[datalen] = '\0'; 1272 args->endp = args->begin_argv + datalen + 1; 1273 args->stringspace -= datalen + 1; 1274 1275 /* 1276 * Traditional argument counting. Count the number of 1277 * null bytes. 1278 */ 1279 for (p = args->begin_argv; p < args->endp; ++p) 1280 if (*p == '\0') 1281 ++args->argc; 1282 } else { 1283 /* No argument buffer provided. */ 1284 args->endp = args->begin_argv; 1285 } 1286 1287 /* Create new file descriptor table. */ 1288 kfds = malloc(fdslen * sizeof(int), M_TEMP, M_WAITOK); 1289 error = copyin(fds, kfds, fdslen * sizeof(int)); 1290 if (error != 0) { 1291 free(kfds, M_TEMP); 1292 goto err_exit; 1293 } 1294 error = fdcopy_remapped(ofdp, kfds, fdslen, &args->fdp); 1295 free(kfds, M_TEMP); 1296 if (error != 0) 1297 goto err_exit; 1298 1299 return (0); 1300 err_exit: 1301 exec_free_args(args); 1302 return (error); 1303 } 1304 1305 struct exec_args_kva { 1306 vm_offset_t addr; 1307 u_int gen; 1308 SLIST_ENTRY(exec_args_kva) next; 1309 }; 1310 1311 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1312 1313 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1314 static struct mtx exec_args_kva_mtx; 1315 static u_int exec_args_gen; 1316 1317 static void 1318 exec_prealloc_args_kva(void *arg __unused) 1319 { 1320 struct exec_args_kva *argkva; 1321 u_int i; 1322 1323 SLIST_INIT(&exec_args_kva_freelist); 1324 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1325 for (i = 0; i < exec_map_entries; i++) { 1326 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1327 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1328 argkva->gen = exec_args_gen; 1329 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1330 } 1331 } 1332 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1333 1334 static vm_offset_t 1335 exec_alloc_args_kva(void **cookie) 1336 { 1337 struct exec_args_kva *argkva; 1338 1339 argkva = (void *)atomic_readandclear_ptr( 1340 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1341 if (argkva == NULL) { 1342 mtx_lock(&exec_args_kva_mtx); 1343 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1344 (void)mtx_sleep(&exec_args_kva_freelist, 1345 &exec_args_kva_mtx, 0, "execkva", 0); 1346 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1347 mtx_unlock(&exec_args_kva_mtx); 1348 } 1349 kasan_mark((void *)argkva->addr, exec_map_entry_size, 1350 exec_map_entry_size, 0); 1351 *(struct exec_args_kva **)cookie = argkva; 1352 return (argkva->addr); 1353 } 1354 1355 static void 1356 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1357 { 1358 vm_offset_t base; 1359 1360 base = argkva->addr; 1361 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size, 1362 KASAN_EXEC_ARGS_FREED); 1363 if (argkva->gen != gen) { 1364 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1365 MADV_FREE); 1366 argkva->gen = gen; 1367 } 1368 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1369 (uintptr_t)NULL, (uintptr_t)argkva)) { 1370 mtx_lock(&exec_args_kva_mtx); 1371 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1372 wakeup_one(&exec_args_kva_freelist); 1373 mtx_unlock(&exec_args_kva_mtx); 1374 } 1375 } 1376 1377 static void 1378 exec_free_args_kva(void *cookie) 1379 { 1380 1381 exec_release_args_kva(cookie, exec_args_gen); 1382 } 1383 1384 static void 1385 exec_args_kva_lowmem(void *arg __unused) 1386 { 1387 SLIST_HEAD(, exec_args_kva) head; 1388 struct exec_args_kva *argkva; 1389 u_int gen; 1390 int i; 1391 1392 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1393 1394 /* 1395 * Force an madvise of each KVA range. Any currently allocated ranges 1396 * will have MADV_FREE applied once they are freed. 1397 */ 1398 SLIST_INIT(&head); 1399 mtx_lock(&exec_args_kva_mtx); 1400 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1401 mtx_unlock(&exec_args_kva_mtx); 1402 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1403 SLIST_REMOVE_HEAD(&head, next); 1404 exec_release_args_kva(argkva, gen); 1405 } 1406 1407 CPU_FOREACH(i) { 1408 argkva = (void *)atomic_readandclear_ptr( 1409 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1410 if (argkva != NULL) 1411 exec_release_args_kva(argkva, gen); 1412 } 1413 } 1414 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1415 EVENTHANDLER_PRI_ANY); 1416 1417 /* 1418 * Allocate temporary demand-paged, zero-filled memory for the file name, 1419 * argument, and environment strings. 1420 */ 1421 int 1422 exec_alloc_args(struct image_args *args) 1423 { 1424 1425 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1426 return (0); 1427 } 1428 1429 void 1430 exec_free_args(struct image_args *args) 1431 { 1432 1433 if (args->buf != NULL) { 1434 exec_free_args_kva(args->bufkva); 1435 args->buf = NULL; 1436 } 1437 if (args->fname_buf != NULL) { 1438 free(args->fname_buf, M_TEMP); 1439 args->fname_buf = NULL; 1440 } 1441 if (args->fdp != NULL) 1442 fdescfree_remapped(args->fdp); 1443 } 1444 1445 /* 1446 * A set to functions to fill struct image args. 1447 * 1448 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1449 * fname) before the other functions. All exec_args_add_arg() calls must 1450 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1451 * may be called any time after exec_args_add_fname(). 1452 * 1453 * exec_args_add_fname() - install path to be executed 1454 * exec_args_add_arg() - append an argument string 1455 * exec_args_add_env() - append an env string 1456 * exec_args_adjust_args() - adjust location of the argument list to 1457 * allow new arguments to be prepended 1458 */ 1459 int 1460 exec_args_add_fname(struct image_args *args, const char *fname, 1461 enum uio_seg segflg) 1462 { 1463 int error; 1464 size_t length; 1465 1466 KASSERT(args->fname == NULL, ("fname already appended")); 1467 KASSERT(args->endp == NULL, ("already appending to args")); 1468 1469 if (fname != NULL) { 1470 args->fname = args->buf; 1471 error = segflg == UIO_SYSSPACE ? 1472 copystr(fname, args->fname, PATH_MAX, &length) : 1473 copyinstr(fname, args->fname, PATH_MAX, &length); 1474 if (error != 0) 1475 return (error == ENAMETOOLONG ? E2BIG : error); 1476 } else 1477 length = 0; 1478 1479 /* Set up for _arg_*()/_env_*() */ 1480 args->endp = args->buf + length; 1481 /* begin_argv must be set and kept updated */ 1482 args->begin_argv = args->endp; 1483 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1484 ("too little space remaining for arguments %zu < %zu", 1485 exec_map_entry_size - length, (size_t)ARG_MAX)); 1486 args->stringspace = ARG_MAX; 1487 1488 return (0); 1489 } 1490 1491 static int 1492 exec_args_add_str(struct image_args *args, const char *str, 1493 enum uio_seg segflg, int *countp) 1494 { 1495 int error; 1496 size_t length; 1497 1498 KASSERT(args->endp != NULL, ("endp not initialized")); 1499 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1500 1501 error = (segflg == UIO_SYSSPACE) ? 1502 copystr(str, args->endp, args->stringspace, &length) : 1503 copyinstr(str, args->endp, args->stringspace, &length); 1504 if (error != 0) 1505 return (error == ENAMETOOLONG ? E2BIG : error); 1506 args->stringspace -= length; 1507 args->endp += length; 1508 (*countp)++; 1509 1510 return (0); 1511 } 1512 1513 int 1514 exec_args_add_arg(struct image_args *args, const char *argp, 1515 enum uio_seg segflg) 1516 { 1517 1518 KASSERT(args->envc == 0, ("appending args after env")); 1519 1520 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1521 } 1522 1523 int 1524 exec_args_add_env(struct image_args *args, const char *envp, 1525 enum uio_seg segflg) 1526 { 1527 1528 if (args->envc == 0) 1529 args->begin_envv = args->endp; 1530 1531 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1532 } 1533 1534 int 1535 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1536 { 1537 ssize_t offset; 1538 1539 KASSERT(args->endp != NULL, ("endp not initialized")); 1540 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1541 1542 offset = extend - consume; 1543 if (args->stringspace < offset) 1544 return (E2BIG); 1545 memmove(args->begin_argv + extend, args->begin_argv + consume, 1546 args->endp - args->begin_argv + consume); 1547 if (args->envc > 0) 1548 args->begin_envv += offset; 1549 args->endp += offset; 1550 args->stringspace -= offset; 1551 return (0); 1552 } 1553 1554 char * 1555 exec_args_get_begin_envv(struct image_args *args) 1556 { 1557 1558 KASSERT(args->endp != NULL, ("endp not initialized")); 1559 1560 if (args->envc > 0) 1561 return (args->begin_envv); 1562 return (args->endp); 1563 } 1564 1565 void 1566 exec_stackgap(struct image_params *imgp, uintptr_t *dp) 1567 { 1568 if (imgp->sysent->sv_stackgap == NULL || 1569 (imgp->proc->p_fctl0 & (NT_FREEBSD_FCTL_ASLR_DISABLE | 1570 NT_FREEBSD_FCTL_ASG_DISABLE)) != 0 || 1571 (imgp->map_flags & MAP_ASLR) == 0) 1572 return; 1573 imgp->sysent->sv_stackgap(imgp, dp); 1574 } 1575 1576 /* 1577 * Copy strings out to the new process address space, constructing new arg 1578 * and env vector tables. Return a pointer to the base so that it can be used 1579 * as the initial stack pointer. 1580 */ 1581 int 1582 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1583 { 1584 int argc, envc; 1585 char **vectp; 1586 char *stringp; 1587 uintptr_t destp, ustringp; 1588 struct ps_strings *arginfo; 1589 struct proc *p; 1590 size_t execpath_len; 1591 int error, szsigcode, szps; 1592 char canary[sizeof(long) * 8]; 1593 1594 szps = sizeof(pagesizes[0]) * MAXPAGESIZES; 1595 /* 1596 * Calculate string base and vector table pointers. 1597 * Also deal with signal trampoline code for this exec type. 1598 */ 1599 if (imgp->execpath != NULL && imgp->auxargs != NULL) 1600 execpath_len = strlen(imgp->execpath) + 1; 1601 else 1602 execpath_len = 0; 1603 p = imgp->proc; 1604 szsigcode = 0; 1605 arginfo = (struct ps_strings *)p->p_sysent->sv_psstrings; 1606 imgp->ps_strings = arginfo; 1607 if (p->p_sysent->sv_sigcode_base == 0) { 1608 if (p->p_sysent->sv_szsigcode != NULL) 1609 szsigcode = *(p->p_sysent->sv_szsigcode); 1610 } 1611 destp = (uintptr_t)arginfo; 1612 1613 /* 1614 * install sigcode 1615 */ 1616 if (szsigcode != 0) { 1617 destp -= szsigcode; 1618 destp = rounddown2(destp, sizeof(void *)); 1619 error = copyout(p->p_sysent->sv_sigcode, (void *)destp, 1620 szsigcode); 1621 if (error != 0) 1622 return (error); 1623 } 1624 1625 /* 1626 * Copy the image path for the rtld. 1627 */ 1628 if (execpath_len != 0) { 1629 destp -= execpath_len; 1630 destp = rounddown2(destp, sizeof(void *)); 1631 imgp->execpathp = (void *)destp; 1632 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1633 if (error != 0) 1634 return (error); 1635 } 1636 1637 /* 1638 * Prepare the canary for SSP. 1639 */ 1640 arc4rand(canary, sizeof(canary), 0); 1641 destp -= sizeof(canary); 1642 imgp->canary = (void *)destp; 1643 error = copyout(canary, imgp->canary, sizeof(canary)); 1644 if (error != 0) 1645 return (error); 1646 imgp->canarylen = sizeof(canary); 1647 1648 /* 1649 * Prepare the pagesizes array. 1650 */ 1651 destp -= szps; 1652 destp = rounddown2(destp, sizeof(void *)); 1653 imgp->pagesizes = (void *)destp; 1654 error = copyout(pagesizes, imgp->pagesizes, szps); 1655 if (error != 0) 1656 return (error); 1657 imgp->pagesizeslen = szps; 1658 1659 /* 1660 * Allocate room for the argument and environment strings. 1661 */ 1662 destp -= ARG_MAX - imgp->args->stringspace; 1663 destp = rounddown2(destp, sizeof(void *)); 1664 ustringp = destp; 1665 1666 exec_stackgap(imgp, &destp); 1667 1668 if (imgp->auxargs) { 1669 /* 1670 * Allocate room on the stack for the ELF auxargs 1671 * array. It has up to AT_COUNT entries. 1672 */ 1673 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1674 destp = rounddown2(destp, sizeof(void *)); 1675 } 1676 1677 vectp = (char **)destp; 1678 1679 /* 1680 * Allocate room for the argv[] and env vectors including the 1681 * terminating NULL pointers. 1682 */ 1683 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1684 1685 /* 1686 * vectp also becomes our initial stack base 1687 */ 1688 *stack_base = (uintptr_t)vectp; 1689 1690 stringp = imgp->args->begin_argv; 1691 argc = imgp->args->argc; 1692 envc = imgp->args->envc; 1693 1694 /* 1695 * Copy out strings - arguments and environment. 1696 */ 1697 error = copyout(stringp, (void *)ustringp, 1698 ARG_MAX - imgp->args->stringspace); 1699 if (error != 0) 1700 return (error); 1701 1702 /* 1703 * Fill in "ps_strings" struct for ps, w, etc. 1704 */ 1705 imgp->argv = vectp; 1706 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1707 suword32(&arginfo->ps_nargvstr, argc) != 0) 1708 return (EFAULT); 1709 1710 /* 1711 * Fill in argument portion of vector table. 1712 */ 1713 for (; argc > 0; --argc) { 1714 if (suword(vectp++, ustringp) != 0) 1715 return (EFAULT); 1716 while (*stringp++ != 0) 1717 ustringp++; 1718 ustringp++; 1719 } 1720 1721 /* a null vector table pointer separates the argp's from the envp's */ 1722 if (suword(vectp++, 0) != 0) 1723 return (EFAULT); 1724 1725 imgp->envv = vectp; 1726 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1727 suword32(&arginfo->ps_nenvstr, envc) != 0) 1728 return (EFAULT); 1729 1730 /* 1731 * Fill in environment portion of vector table. 1732 */ 1733 for (; envc > 0; --envc) { 1734 if (suword(vectp++, ustringp) != 0) 1735 return (EFAULT); 1736 while (*stringp++ != 0) 1737 ustringp++; 1738 ustringp++; 1739 } 1740 1741 /* end of vector table is a null pointer */ 1742 if (suword(vectp, 0) != 0) 1743 return (EFAULT); 1744 1745 if (imgp->auxargs) { 1746 vectp++; 1747 error = imgp->sysent->sv_copyout_auxargs(imgp, 1748 (uintptr_t)vectp); 1749 if (error != 0) 1750 return (error); 1751 } 1752 1753 return (0); 1754 } 1755 1756 /* 1757 * Check permissions of file to execute. 1758 * Called with imgp->vp locked. 1759 * Return 0 for success or error code on failure. 1760 */ 1761 int 1762 exec_check_permissions(struct image_params *imgp) 1763 { 1764 struct vnode *vp = imgp->vp; 1765 struct vattr *attr = imgp->attr; 1766 struct thread *td; 1767 int error; 1768 1769 td = curthread; 1770 1771 /* Get file attributes */ 1772 error = VOP_GETATTR(vp, attr, td->td_ucred); 1773 if (error) 1774 return (error); 1775 1776 #ifdef MAC 1777 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1778 if (error) 1779 return (error); 1780 #endif 1781 1782 /* 1783 * 1) Check if file execution is disabled for the filesystem that 1784 * this file resides on. 1785 * 2) Ensure that at least one execute bit is on. Otherwise, a 1786 * privileged user will always succeed, and we don't want this 1787 * to happen unless the file really is executable. 1788 * 3) Ensure that the file is a regular file. 1789 */ 1790 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1791 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1792 (attr->va_type != VREG)) 1793 return (EACCES); 1794 1795 /* 1796 * Zero length files can't be exec'd 1797 */ 1798 if (attr->va_size == 0) 1799 return (ENOEXEC); 1800 1801 /* 1802 * Check for execute permission to file based on current credentials. 1803 */ 1804 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1805 if (error) 1806 return (error); 1807 1808 /* 1809 * Check number of open-for-writes on the file and deny execution 1810 * if there are any. 1811 * 1812 * Add a text reference now so no one can write to the 1813 * executable while we're activating it. 1814 * 1815 * Remember if this was set before and unset it in case this is not 1816 * actually an executable image. 1817 */ 1818 error = VOP_SET_TEXT(vp); 1819 if (error != 0) 1820 return (error); 1821 imgp->textset = true; 1822 1823 /* 1824 * Call filesystem specific open routine (which does nothing in the 1825 * general case). 1826 */ 1827 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1828 if (error == 0) 1829 imgp->opened = 1; 1830 return (error); 1831 } 1832 1833 /* 1834 * Exec handler registration 1835 */ 1836 int 1837 exec_register(const struct execsw *execsw_arg) 1838 { 1839 const struct execsw **es, **xs, **newexecsw; 1840 u_int count = 2; /* New slot and trailing NULL */ 1841 1842 if (execsw) 1843 for (es = execsw; *es; es++) 1844 count++; 1845 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1846 xs = newexecsw; 1847 if (execsw) 1848 for (es = execsw; *es; es++) 1849 *xs++ = *es; 1850 *xs++ = execsw_arg; 1851 *xs = NULL; 1852 if (execsw) 1853 free(execsw, M_TEMP); 1854 execsw = newexecsw; 1855 return (0); 1856 } 1857 1858 int 1859 exec_unregister(const struct execsw *execsw_arg) 1860 { 1861 const struct execsw **es, **xs, **newexecsw; 1862 int count = 1; 1863 1864 if (execsw == NULL) 1865 panic("unregister with no handlers left?\n"); 1866 1867 for (es = execsw; *es; es++) { 1868 if (*es == execsw_arg) 1869 break; 1870 } 1871 if (*es == NULL) 1872 return (ENOENT); 1873 for (es = execsw; *es; es++) 1874 if (*es != execsw_arg) 1875 count++; 1876 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1877 xs = newexecsw; 1878 for (es = execsw; *es; es++) 1879 if (*es != execsw_arg) 1880 *xs++ = *es; 1881 *xs = NULL; 1882 if (execsw) 1883 free(execsw, M_TEMP); 1884 execsw = newexecsw; 1885 return (0); 1886 } 1887 1888 /* 1889 * Write out a core segment to the compression stream. 1890 */ 1891 static int 1892 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len) 1893 { 1894 size_t chunk_len; 1895 int error; 1896 1897 while (len > 0) { 1898 chunk_len = MIN(len, CORE_BUF_SIZE); 1899 1900 /* 1901 * We can get EFAULT error here. 1902 * In that case zero out the current chunk of the segment. 1903 */ 1904 error = copyin(base, buf, chunk_len); 1905 if (error != 0) 1906 bzero(buf, chunk_len); 1907 error = compressor_write(cp->comp, buf, chunk_len); 1908 if (error != 0) 1909 break; 1910 base += chunk_len; 1911 len -= chunk_len; 1912 } 1913 return (error); 1914 } 1915 1916 int 1917 core_write(struct coredump_params *cp, const void *base, size_t len, 1918 off_t offset, enum uio_seg seg, size_t *resid) 1919 { 1920 1921 return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base), 1922 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1923 cp->active_cred, cp->file_cred, resid, cp->td)); 1924 } 1925 1926 int 1927 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp, 1928 void *tmpbuf) 1929 { 1930 vm_map_t map; 1931 struct mount *mp; 1932 size_t resid, runlen; 1933 int error; 1934 bool success; 1935 1936 KASSERT((uintptr_t)base % PAGE_SIZE == 0, 1937 ("%s: user address %p is not page-aligned", __func__, base)); 1938 1939 if (cp->comp != NULL) 1940 return (compress_chunk(cp, base, tmpbuf, len)); 1941 1942 map = &cp->td->td_proc->p_vmspace->vm_map; 1943 for (; len > 0; base += runlen, offset += runlen, len -= runlen) { 1944 /* 1945 * Attempt to page in all virtual pages in the range. If a 1946 * virtual page is not backed by the pager, it is represented as 1947 * a hole in the file. This can occur with zero-filled 1948 * anonymous memory or truncated files, for example. 1949 */ 1950 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) { 1951 if (core_dump_can_intr && curproc_sigkilled()) 1952 return (EINTR); 1953 error = vm_fault(map, (uintptr_t)base + runlen, 1954 VM_PROT_READ, VM_FAULT_NOFILL, NULL); 1955 if (runlen == 0) 1956 success = error == KERN_SUCCESS; 1957 else if ((error == KERN_SUCCESS) != success) 1958 break; 1959 } 1960 1961 if (success) { 1962 error = core_write(cp, base, runlen, offset, 1963 UIO_USERSPACE, &resid); 1964 if (error != 0) { 1965 if (error != EFAULT) 1966 break; 1967 1968 /* 1969 * EFAULT may be returned if the user mapping 1970 * could not be accessed, e.g., because a mapped 1971 * file has been truncated. Skip the page if no 1972 * progress was made, to protect against a 1973 * hypothetical scenario where vm_fault() was 1974 * successful but core_write() returns EFAULT 1975 * anyway. 1976 */ 1977 runlen -= resid; 1978 if (runlen == 0) { 1979 success = false; 1980 runlen = PAGE_SIZE; 1981 } 1982 } 1983 } 1984 if (!success) { 1985 error = vn_start_write(cp->vp, &mp, V_WAIT); 1986 if (error != 0) 1987 break; 1988 vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY); 1989 error = vn_truncate_locked(cp->vp, offset + runlen, 1990 false, cp->td->td_ucred); 1991 VOP_UNLOCK(cp->vp); 1992 vn_finished_write(mp); 1993 if (error != 0) 1994 break; 1995 } 1996 } 1997 return (error); 1998 } 1999 2000 /* 2001 * Drain into a core file. 2002 */ 2003 int 2004 sbuf_drain_core_output(void *arg, const char *data, int len) 2005 { 2006 struct coredump_params *cp; 2007 struct proc *p; 2008 int error, locked; 2009 2010 cp = arg; 2011 p = cp->td->td_proc; 2012 2013 /* 2014 * Some kern_proc out routines that print to this sbuf may 2015 * call us with the process lock held. Draining with the 2016 * non-sleepable lock held is unsafe. The lock is needed for 2017 * those routines when dumping a live process. In our case we 2018 * can safely release the lock before draining and acquire 2019 * again after. 2020 */ 2021 locked = PROC_LOCKED(p); 2022 if (locked) 2023 PROC_UNLOCK(p); 2024 if (cp->comp != NULL) 2025 error = compressor_write(cp->comp, __DECONST(char *, data), len); 2026 else 2027 error = core_write(cp, __DECONST(void *, data), len, cp->offset, 2028 UIO_SYSSPACE, NULL); 2029 if (locked) 2030 PROC_LOCK(p); 2031 if (error != 0) 2032 return (-error); 2033 cp->offset += len; 2034 return (len); 2035 } 2036