1 /*- 2 * SPDX-License-Identifier: BSD-2-Clause 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 #include "opt_capsicum.h" 31 #include "opt_hwpmc_hooks.h" 32 #include "opt_ktrace.h" 33 #include "opt_vm.h" 34 35 #include <sys/param.h> 36 #include <sys/systm.h> 37 #include <sys/acct.h> 38 #include <sys/asan.h> 39 #include <sys/capsicum.h> 40 #include <sys/compressor.h> 41 #include <sys/eventhandler.h> 42 #include <sys/exec.h> 43 #include <sys/fcntl.h> 44 #include <sys/filedesc.h> 45 #include <sys/imgact.h> 46 #include <sys/imgact_elf.h> 47 #include <sys/kernel.h> 48 #include <sys/lock.h> 49 #include <sys/malloc.h> 50 #include <sys/mman.h> 51 #include <sys/mount.h> 52 #include <sys/mutex.h> 53 #include <sys/namei.h> 54 #include <sys/priv.h> 55 #include <sys/proc.h> 56 #include <sys/ptrace.h> 57 #include <sys/reg.h> 58 #include <sys/resourcevar.h> 59 #include <sys/rwlock.h> 60 #include <sys/sched.h> 61 #include <sys/sdt.h> 62 #include <sys/sf_buf.h> 63 #include <sys/shm.h> 64 #include <sys/signalvar.h> 65 #include <sys/smp.h> 66 #include <sys/stat.h> 67 #include <sys/syscallsubr.h> 68 #include <sys/sysctl.h> 69 #include <sys/sysent.h> 70 #include <sys/sysproto.h> 71 #include <sys/timers.h> 72 #include <sys/umtxvar.h> 73 #include <sys/vnode.h> 74 #include <sys/wait.h> 75 #ifdef KTRACE 76 #include <sys/ktrace.h> 77 #endif 78 79 #include <vm/vm.h> 80 #include <vm/vm_param.h> 81 #include <vm/pmap.h> 82 #include <vm/vm_page.h> 83 #include <vm/vm_map.h> 84 #include <vm/vm_kern.h> 85 #include <vm/vm_extern.h> 86 #include <vm/vm_object.h> 87 #include <vm/vm_pager.h> 88 89 #ifdef HWPMC_HOOKS 90 #include <sys/pmckern.h> 91 #endif 92 93 #include <security/audit/audit.h> 94 #include <security/mac/mac_framework.h> 95 96 #ifdef KDTRACE_HOOKS 97 #include <sys/dtrace_bsd.h> 98 dtrace_execexit_func_t dtrace_fasttrap_exec; 99 #endif 100 101 SDT_PROVIDER_DECLARE(proc); 102 SDT_PROBE_DEFINE1(proc, , , exec, "char *"); 103 SDT_PROBE_DEFINE1(proc, , , exec__failure, "int"); 104 SDT_PROBE_DEFINE1(proc, , , exec__success, "char *"); 105 106 MALLOC_DEFINE(M_PARGS, "proc-args", "Process arguments"); 107 108 int coredump_pack_fileinfo = 1; 109 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_fileinfo, CTLFLAG_RWTUN, 110 &coredump_pack_fileinfo, 0, 111 "Enable file path packing in 'procstat -f' coredump notes"); 112 113 int coredump_pack_vmmapinfo = 1; 114 SYSCTL_INT(_kern, OID_AUTO, coredump_pack_vmmapinfo, CTLFLAG_RWTUN, 115 &coredump_pack_vmmapinfo, 0, 116 "Enable file path packing in 'procstat -v' coredump notes"); 117 118 static int sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS); 119 static int sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS); 120 static int sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS); 121 static int do_execve(struct thread *td, struct image_args *args, 122 struct mac *mac_p, struct vmspace *oldvmspace); 123 124 /* XXX This should be vm_size_t. */ 125 SYSCTL_PROC(_kern, KERN_PS_STRINGS, ps_strings, CTLTYPE_ULONG|CTLFLAG_RD| 126 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_ps_strings, "LU", 127 "Location of process' ps_strings structure"); 128 129 /* XXX This should be vm_size_t. */ 130 SYSCTL_PROC(_kern, KERN_USRSTACK, usrstack, CTLTYPE_ULONG|CTLFLAG_RD| 131 CTLFLAG_CAPRD|CTLFLAG_MPSAFE, NULL, 0, sysctl_kern_usrstack, "LU", 132 "Top of process stack"); 133 134 SYSCTL_PROC(_kern, OID_AUTO, stackprot, CTLTYPE_INT|CTLFLAG_RD|CTLFLAG_MPSAFE, 135 NULL, 0, sysctl_kern_stackprot, "I", 136 "Stack memory permissions"); 137 138 u_long ps_arg_cache_limit = PAGE_SIZE / 16; 139 SYSCTL_ULONG(_kern, OID_AUTO, ps_arg_cache_limit, CTLFLAG_RW, 140 &ps_arg_cache_limit, 0, 141 "Process' command line characters cache limit"); 142 143 static int disallow_high_osrel; 144 SYSCTL_INT(_kern, OID_AUTO, disallow_high_osrel, CTLFLAG_RW, 145 &disallow_high_osrel, 0, 146 "Disallow execution of binaries built for higher version of the world"); 147 148 static int map_at_zero = 0; 149 SYSCTL_INT(_security_bsd, OID_AUTO, map_at_zero, CTLFLAG_RWTUN, &map_at_zero, 0, 150 "Permit processes to map an object at virtual address 0."); 151 152 static int core_dump_can_intr = 1; 153 SYSCTL_INT(_kern, OID_AUTO, core_dump_can_intr, CTLFLAG_RWTUN, 154 &core_dump_can_intr, 0, 155 "Core dumping interruptible with SIGKILL"); 156 157 static int 158 sysctl_kern_ps_strings(SYSCTL_HANDLER_ARGS) 159 { 160 struct proc *p; 161 vm_offset_t ps_strings; 162 163 p = curproc; 164 #ifdef SCTL_MASK32 165 if (req->flags & SCTL_MASK32) { 166 unsigned int val; 167 val = (unsigned int)PROC_PS_STRINGS(p); 168 return (SYSCTL_OUT(req, &val, sizeof(val))); 169 } 170 #endif 171 ps_strings = PROC_PS_STRINGS(p); 172 return (SYSCTL_OUT(req, &ps_strings, sizeof(ps_strings))); 173 } 174 175 static int 176 sysctl_kern_usrstack(SYSCTL_HANDLER_ARGS) 177 { 178 struct proc *p; 179 vm_offset_t val; 180 181 p = curproc; 182 #ifdef SCTL_MASK32 183 if (req->flags & SCTL_MASK32) { 184 unsigned int val32; 185 186 val32 = round_page((unsigned int)p->p_vmspace->vm_stacktop); 187 return (SYSCTL_OUT(req, &val32, sizeof(val32))); 188 } 189 #endif 190 val = round_page(p->p_vmspace->vm_stacktop); 191 return (SYSCTL_OUT(req, &val, sizeof(val))); 192 } 193 194 static int 195 sysctl_kern_stackprot(SYSCTL_HANDLER_ARGS) 196 { 197 struct proc *p; 198 199 p = curproc; 200 return (SYSCTL_OUT(req, &p->p_sysent->sv_stackprot, 201 sizeof(p->p_sysent->sv_stackprot))); 202 } 203 204 /* 205 * Each of the items is a pointer to a `const struct execsw', hence the 206 * double pointer here. 207 */ 208 static const struct execsw **execsw; 209 210 #ifndef _SYS_SYSPROTO_H_ 211 struct execve_args { 212 char *fname; 213 char **argv; 214 char **envv; 215 }; 216 #endif 217 218 int 219 sys_execve(struct thread *td, struct execve_args *uap) 220 { 221 struct image_args args; 222 struct vmspace *oldvmspace; 223 int error; 224 225 error = pre_execve(td, &oldvmspace); 226 if (error != 0) 227 return (error); 228 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 229 uap->argv, uap->envv); 230 if (error == 0) 231 error = kern_execve(td, &args, NULL, oldvmspace); 232 post_execve(td, error, oldvmspace); 233 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 234 return (error); 235 } 236 237 #ifndef _SYS_SYSPROTO_H_ 238 struct fexecve_args { 239 int fd; 240 char **argv; 241 char **envv; 242 }; 243 #endif 244 int 245 sys_fexecve(struct thread *td, struct fexecve_args *uap) 246 { 247 struct image_args args; 248 struct vmspace *oldvmspace; 249 int error; 250 251 error = pre_execve(td, &oldvmspace); 252 if (error != 0) 253 return (error); 254 error = exec_copyin_args(&args, NULL, UIO_SYSSPACE, 255 uap->argv, uap->envv); 256 if (error == 0) { 257 args.fd = uap->fd; 258 error = kern_execve(td, &args, NULL, oldvmspace); 259 } 260 post_execve(td, error, oldvmspace); 261 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 262 return (error); 263 } 264 265 #ifndef _SYS_SYSPROTO_H_ 266 struct __mac_execve_args { 267 char *fname; 268 char **argv; 269 char **envv; 270 struct mac *mac_p; 271 }; 272 #endif 273 274 int 275 sys___mac_execve(struct thread *td, struct __mac_execve_args *uap) 276 { 277 #ifdef MAC 278 struct image_args args; 279 struct vmspace *oldvmspace; 280 int error; 281 282 error = pre_execve(td, &oldvmspace); 283 if (error != 0) 284 return (error); 285 error = exec_copyin_args(&args, uap->fname, UIO_USERSPACE, 286 uap->argv, uap->envv); 287 if (error == 0) 288 error = kern_execve(td, &args, uap->mac_p, oldvmspace); 289 post_execve(td, error, oldvmspace); 290 AUDIT_SYSCALL_EXIT(error == EJUSTRETURN ? 0 : error, td); 291 return (error); 292 #else 293 return (ENOSYS); 294 #endif 295 } 296 297 int 298 pre_execve(struct thread *td, struct vmspace **oldvmspace) 299 { 300 struct proc *p; 301 int error; 302 303 KASSERT(td == curthread, ("non-current thread %p", td)); 304 error = 0; 305 p = td->td_proc; 306 if ((p->p_flag & P_HADTHREADS) != 0) { 307 PROC_LOCK(p); 308 if (thread_single(p, SINGLE_BOUNDARY) != 0) 309 error = ERESTART; 310 PROC_UNLOCK(p); 311 } 312 KASSERT(error != 0 || (td->td_pflags & TDP_EXECVMSPC) == 0, 313 ("nested execve")); 314 *oldvmspace = p->p_vmspace; 315 return (error); 316 } 317 318 void 319 post_execve(struct thread *td, int error, struct vmspace *oldvmspace) 320 { 321 struct proc *p; 322 323 KASSERT(td == curthread, ("non-current thread %p", td)); 324 p = td->td_proc; 325 if ((p->p_flag & P_HADTHREADS) != 0) { 326 PROC_LOCK(p); 327 /* 328 * If success, we upgrade to SINGLE_EXIT state to 329 * force other threads to suicide. 330 */ 331 if (error == EJUSTRETURN) 332 thread_single(p, SINGLE_EXIT); 333 else 334 thread_single_end(p, SINGLE_BOUNDARY); 335 PROC_UNLOCK(p); 336 } 337 exec_cleanup(td, oldvmspace); 338 } 339 340 /* 341 * kern_execve() has the astonishing property of not always returning to 342 * the caller. If sufficiently bad things happen during the call to 343 * do_execve(), it can end up calling exit1(); as a result, callers must 344 * avoid doing anything which they might need to undo (e.g., allocating 345 * memory). 346 */ 347 int 348 kern_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 349 struct vmspace *oldvmspace) 350 { 351 352 TSEXEC(td->td_proc->p_pid, args->begin_argv); 353 AUDIT_ARG_ARGV(args->begin_argv, args->argc, 354 exec_args_get_begin_envv(args) - args->begin_argv); 355 AUDIT_ARG_ENVV(exec_args_get_begin_envv(args), args->envc, 356 args->endp - exec_args_get_begin_envv(args)); 357 358 /* Must have at least one argument. */ 359 if (args->argc == 0) { 360 exec_free_args(args); 361 return (EINVAL); 362 } 363 return (do_execve(td, args, mac_p, oldvmspace)); 364 } 365 366 static void 367 execve_nosetid(struct image_params *imgp) 368 { 369 imgp->credential_setid = false; 370 if (imgp->newcred != NULL) { 371 crfree(imgp->newcred); 372 imgp->newcred = NULL; 373 } 374 } 375 376 /* 377 * In-kernel implementation of execve(). All arguments are assumed to be 378 * userspace pointers from the passed thread. 379 */ 380 static int 381 do_execve(struct thread *td, struct image_args *args, struct mac *mac_p, 382 struct vmspace *oldvmspace) 383 { 384 struct proc *p = td->td_proc; 385 struct nameidata nd; 386 struct ucred *oldcred; 387 struct uidinfo *euip = NULL; 388 uintptr_t stack_base; 389 struct image_params image_params, *imgp; 390 struct vattr attr; 391 struct pargs *oldargs = NULL, *newargs = NULL; 392 struct sigacts *oldsigacts = NULL, *newsigacts = NULL; 393 #ifdef KTRACE 394 struct ktr_io_params *kiop; 395 #endif 396 struct vnode *oldtextvp, *newtextvp; 397 struct vnode *oldtextdvp, *newtextdvp; 398 char *oldbinname, *newbinname; 399 bool credential_changing; 400 #ifdef MAC 401 struct label *interpvplabel = NULL; 402 bool will_transition; 403 #endif 404 #ifdef HWPMC_HOOKS 405 struct pmckern_procexec pe; 406 #endif 407 int error, i, orig_osrel; 408 uint32_t orig_fctl0; 409 Elf_Brandinfo *orig_brandinfo; 410 size_t freepath_size; 411 static const char fexecv_proc_title[] = "(fexecv)"; 412 413 imgp = &image_params; 414 oldtextvp = oldtextdvp = NULL; 415 newtextvp = newtextdvp = NULL; 416 newbinname = oldbinname = NULL; 417 #ifdef KTRACE 418 kiop = NULL; 419 #endif 420 421 /* 422 * Lock the process and set the P_INEXEC flag to indicate that 423 * it should be left alone until we're done here. This is 424 * necessary to avoid race conditions - e.g. in ptrace() - 425 * that might allow a local user to illicitly obtain elevated 426 * privileges. 427 */ 428 PROC_LOCK(p); 429 KASSERT((p->p_flag & P_INEXEC) == 0, 430 ("%s(): process already has P_INEXEC flag", __func__)); 431 p->p_flag |= P_INEXEC; 432 PROC_UNLOCK(p); 433 434 /* 435 * Initialize part of the common data 436 */ 437 bzero(imgp, sizeof(*imgp)); 438 imgp->proc = p; 439 imgp->attr = &attr; 440 imgp->args = args; 441 oldcred = p->p_ucred; 442 orig_osrel = p->p_osrel; 443 orig_fctl0 = p->p_fctl0; 444 orig_brandinfo = p->p_elf_brandinfo; 445 446 #ifdef MAC 447 error = mac_execve_enter(imgp, mac_p); 448 if (error) 449 goto exec_fail; 450 #endif 451 452 SDT_PROBE1(proc, , , exec, args->fname); 453 454 interpret: 455 if (args->fname != NULL) { 456 #ifdef CAPABILITY_MODE 457 /* 458 * While capability mode can't reach this point via direct 459 * path arguments to execve(), we also don't allow 460 * interpreters to be used in capability mode (for now). 461 * Catch indirect lookups and return a permissions error. 462 */ 463 if (IN_CAPABILITY_MODE(td)) { 464 error = ECAPMODE; 465 goto exec_fail; 466 } 467 #endif 468 469 /* 470 * Translate the file name. namei() returns a vnode 471 * pointer in ni_vp among other things. 472 */ 473 NDINIT(&nd, LOOKUP, ISOPEN | LOCKLEAF | LOCKSHARED | FOLLOW | 474 AUDITVNODE1 | WANTPARENT, UIO_SYSSPACE, 475 args->fname); 476 477 error = namei(&nd); 478 if (error) 479 goto exec_fail; 480 481 newtextvp = nd.ni_vp; 482 newtextdvp = nd.ni_dvp; 483 nd.ni_dvp = NULL; 484 newbinname = malloc(nd.ni_cnd.cn_namelen + 1, M_PARGS, 485 M_WAITOK); 486 memcpy(newbinname, nd.ni_cnd.cn_nameptr, nd.ni_cnd.cn_namelen); 487 newbinname[nd.ni_cnd.cn_namelen] = '\0'; 488 imgp->vp = newtextvp; 489 490 /* 491 * Do the best to calculate the full path to the image file. 492 */ 493 if (args->fname[0] == '/') { 494 imgp->execpath = args->fname; 495 } else { 496 VOP_UNLOCK(imgp->vp); 497 freepath_size = MAXPATHLEN; 498 if (vn_fullpath_hardlink(newtextvp, newtextdvp, 499 newbinname, nd.ni_cnd.cn_namelen, &imgp->execpath, 500 &imgp->freepath, &freepath_size) != 0) 501 imgp->execpath = args->fname; 502 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 503 } 504 } else if (imgp->interpreter_vp) { 505 /* 506 * An image activator has already provided an open vnode 507 */ 508 newtextvp = imgp->interpreter_vp; 509 imgp->interpreter_vp = NULL; 510 if (vn_fullpath(newtextvp, &imgp->execpath, 511 &imgp->freepath) != 0) 512 imgp->execpath = args->fname; 513 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 514 AUDIT_ARG_VNODE1(newtextvp); 515 imgp->vp = newtextvp; 516 } else { 517 AUDIT_ARG_FD(args->fd); 518 519 /* 520 * If the descriptors was not opened with O_PATH, then 521 * we require that it was opened with O_EXEC or 522 * O_RDONLY. In either case, exec_check_permissions() 523 * below checks _current_ file access mode regardless 524 * of the permissions additionally checked at the 525 * open(2). 526 */ 527 error = fgetvp_exec(td, args->fd, &cap_fexecve_rights, 528 &newtextvp); 529 if (error != 0) 530 goto exec_fail; 531 532 if (vn_fullpath(newtextvp, &imgp->execpath, 533 &imgp->freepath) != 0) 534 imgp->execpath = args->fname; 535 vn_lock(newtextvp, LK_SHARED | LK_RETRY); 536 AUDIT_ARG_VNODE1(newtextvp); 537 imgp->vp = newtextvp; 538 } 539 540 /* 541 * Check file permissions. Also 'opens' file and sets its vnode to 542 * text mode. 543 */ 544 error = exec_check_permissions(imgp); 545 if (error) 546 goto exec_fail_dealloc; 547 548 imgp->object = imgp->vp->v_object; 549 if (imgp->object != NULL) 550 vm_object_reference(imgp->object); 551 552 error = exec_map_first_page(imgp); 553 if (error) 554 goto exec_fail_dealloc; 555 556 imgp->proc->p_osrel = 0; 557 imgp->proc->p_fctl0 = 0; 558 imgp->proc->p_elf_brandinfo = NULL; 559 560 /* 561 * Implement image setuid/setgid. 562 * 563 * Determine new credentials before attempting image activators 564 * so that it can be used by process_exec handlers to determine 565 * credential/setid changes. 566 * 567 * Don't honor setuid/setgid if the filesystem prohibits it or if 568 * the process is being traced. 569 * 570 * We disable setuid/setgid/etc in capability mode on the basis 571 * that most setugid applications are not written with that 572 * environment in mind, and will therefore almost certainly operate 573 * incorrectly. In principle there's no reason that setugid 574 * applications might not be useful in capability mode, so we may want 575 * to reconsider this conservative design choice in the future. 576 * 577 * XXXMAC: For the time being, use NOSUID to also prohibit 578 * transitions on the file system. 579 */ 580 credential_changing = false; 581 credential_changing |= (attr.va_mode & S_ISUID) && 582 oldcred->cr_uid != attr.va_uid; 583 credential_changing |= (attr.va_mode & S_ISGID) && 584 oldcred->cr_gid != attr.va_gid; 585 #ifdef MAC 586 will_transition = mac_vnode_execve_will_transition(oldcred, imgp->vp, 587 interpvplabel, imgp) != 0; 588 credential_changing |= will_transition; 589 #endif 590 591 /* Don't inherit PROC_PDEATHSIG_CTL value if setuid/setgid. */ 592 if (credential_changing) 593 imgp->proc->p_pdeathsig = 0; 594 595 if (credential_changing && 596 #ifdef CAPABILITY_MODE 597 ((oldcred->cr_flags & CRED_FLAG_CAPMODE) == 0) && 598 #endif 599 (imgp->vp->v_mount->mnt_flag & MNT_NOSUID) == 0 && 600 (p->p_flag & P_TRACED) == 0) { 601 imgp->credential_setid = true; 602 VOP_UNLOCK(imgp->vp); 603 imgp->newcred = crdup(oldcred); 604 if (attr.va_mode & S_ISUID) { 605 euip = uifind(attr.va_uid); 606 change_euid(imgp->newcred, euip); 607 } 608 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 609 if (attr.va_mode & S_ISGID) 610 change_egid(imgp->newcred, attr.va_gid); 611 /* 612 * Implement correct POSIX saved-id behavior. 613 * 614 * XXXMAC: Note that the current logic will save the 615 * uid and gid if a MAC domain transition occurs, even 616 * though maybe it shouldn't. 617 */ 618 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 619 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 620 } else { 621 /* 622 * Implement correct POSIX saved-id behavior. 623 * 624 * XXX: It's not clear that the existing behavior is 625 * POSIX-compliant. A number of sources indicate that the 626 * saved uid/gid should only be updated if the new ruid is 627 * not equal to the old ruid, or the new euid is not equal 628 * to the old euid and the new euid is not equal to the old 629 * ruid. The FreeBSD code always updates the saved uid/gid. 630 * Also, this code uses the new (replaced) euid and egid as 631 * the source, which may or may not be the right ones to use. 632 */ 633 if (oldcred->cr_svuid != oldcred->cr_uid || 634 oldcred->cr_svgid != oldcred->cr_gid) { 635 VOP_UNLOCK(imgp->vp); 636 imgp->newcred = crdup(oldcred); 637 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 638 change_svuid(imgp->newcred, imgp->newcred->cr_uid); 639 change_svgid(imgp->newcred, imgp->newcred->cr_gid); 640 } 641 } 642 /* The new credentials are installed into the process later. */ 643 644 /* 645 * Loop through the list of image activators, calling each one. 646 * An activator returns -1 if there is no match, 0 on success, 647 * and an error otherwise. 648 */ 649 error = -1; 650 for (i = 0; error == -1 && execsw[i]; ++i) { 651 if (execsw[i]->ex_imgact == NULL) 652 continue; 653 error = (*execsw[i]->ex_imgact)(imgp); 654 } 655 656 if (error) { 657 if (error == -1) 658 error = ENOEXEC; 659 goto exec_fail_dealloc; 660 } 661 662 /* 663 * Special interpreter operation, cleanup and loop up to try to 664 * activate the interpreter. 665 */ 666 if (imgp->interpreted) { 667 exec_unmap_first_page(imgp); 668 /* 669 * The text reference needs to be removed for scripts. 670 * There is a short period before we determine that 671 * something is a script where text reference is active. 672 * The vnode lock is held over this entire period 673 * so nothing should illegitimately be blocked. 674 */ 675 MPASS(imgp->textset); 676 VOP_UNSET_TEXT_CHECKED(newtextvp); 677 imgp->textset = false; 678 /* free name buffer and old vnode */ 679 #ifdef MAC 680 mac_execve_interpreter_enter(newtextvp, &interpvplabel); 681 #endif 682 if (imgp->opened) { 683 VOP_CLOSE(newtextvp, FREAD, td->td_ucred, td); 684 imgp->opened = false; 685 } 686 vput(newtextvp); 687 imgp->vp = newtextvp = NULL; 688 if (args->fname != NULL) { 689 if (newtextdvp != NULL) { 690 vrele(newtextdvp); 691 newtextdvp = NULL; 692 } 693 NDFREE_PNBUF(&nd); 694 free(newbinname, M_PARGS); 695 newbinname = NULL; 696 } 697 vm_object_deallocate(imgp->object); 698 imgp->object = NULL; 699 execve_nosetid(imgp); 700 imgp->execpath = NULL; 701 free(imgp->freepath, M_TEMP); 702 imgp->freepath = NULL; 703 /* set new name to that of the interpreter */ 704 if (imgp->interpreter_vp) { 705 args->fname = NULL; 706 } else { 707 args->fname = imgp->interpreter_name; 708 } 709 goto interpret; 710 } 711 712 /* 713 * NB: We unlock the vnode here because it is believed that none 714 * of the sv_copyout_strings/sv_fixup operations require the vnode. 715 */ 716 VOP_UNLOCK(imgp->vp); 717 718 if (disallow_high_osrel && 719 P_OSREL_MAJOR(p->p_osrel) > P_OSREL_MAJOR(__FreeBSD_version)) { 720 error = ENOEXEC; 721 uprintf("Osrel %d for image %s too high\n", p->p_osrel, 722 imgp->execpath != NULL ? imgp->execpath : "<unresolved>"); 723 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 724 goto exec_fail_dealloc; 725 } 726 727 /* 728 * Copy out strings (args and env) and initialize stack base. 729 */ 730 error = (*p->p_sysent->sv_copyout_strings)(imgp, &stack_base); 731 if (error != 0) { 732 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 733 goto exec_fail_dealloc; 734 } 735 736 /* 737 * Stack setup. 738 */ 739 error = (*p->p_sysent->sv_fixup)(&stack_base, imgp); 740 if (error != 0) { 741 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 742 goto exec_fail_dealloc; 743 } 744 745 /* 746 * For security and other reasons, the file descriptor table cannot be 747 * shared after an exec. 748 */ 749 fdunshare(td); 750 pdunshare(td); 751 /* close files on exec */ 752 fdcloseexec(td); 753 754 /* 755 * Malloc things before we need locks. 756 */ 757 i = exec_args_get_begin_envv(imgp->args) - imgp->args->begin_argv; 758 /* Cache arguments if they fit inside our allowance */ 759 if (ps_arg_cache_limit >= i + sizeof(struct pargs)) { 760 newargs = pargs_alloc(i); 761 bcopy(imgp->args->begin_argv, newargs->ar_args, i); 762 } 763 764 /* 765 * For security and other reasons, signal handlers cannot 766 * be shared after an exec. The new process gets a copy of the old 767 * handlers. In execsigs(), the new process will have its signals 768 * reset. 769 */ 770 if (sigacts_shared(p->p_sigacts)) { 771 oldsigacts = p->p_sigacts; 772 newsigacts = sigacts_alloc(); 773 sigacts_copy(newsigacts, oldsigacts); 774 } 775 776 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 777 778 PROC_LOCK(p); 779 if (oldsigacts) 780 p->p_sigacts = newsigacts; 781 /* Stop profiling */ 782 stopprofclock(p); 783 784 /* reset caught signals */ 785 execsigs(p); 786 787 /* name this process - nameiexec(p, ndp) */ 788 bzero(p->p_comm, sizeof(p->p_comm)); 789 if (args->fname) 790 bcopy(nd.ni_cnd.cn_nameptr, p->p_comm, 791 min(nd.ni_cnd.cn_namelen, MAXCOMLEN)); 792 else if (vn_commname(newtextvp, p->p_comm, sizeof(p->p_comm)) != 0) 793 bcopy(fexecv_proc_title, p->p_comm, sizeof(fexecv_proc_title)); 794 bcopy(p->p_comm, td->td_name, sizeof(td->td_name)); 795 #ifdef KTR 796 sched_clear_tdname(td); 797 #endif 798 799 /* 800 * mark as execed, wakeup the process that vforked (if any) and tell 801 * it that it now has its own resources back 802 */ 803 p->p_flag |= P_EXEC; 804 if ((p->p_flag2 & P2_NOTRACE_EXEC) == 0) 805 p->p_flag2 &= ~P2_NOTRACE; 806 if ((p->p_flag2 & P2_STKGAP_DISABLE_EXEC) == 0) 807 p->p_flag2 &= ~P2_STKGAP_DISABLE; 808 if (p->p_flag & P_PPWAIT) { 809 p->p_flag &= ~(P_PPWAIT | P_PPTRACE); 810 cv_broadcast(&p->p_pwait); 811 /* STOPs are no longer ignored, arrange for AST */ 812 signotify(td); 813 } 814 815 if ((imgp->sysent->sv_setid_allowed != NULL && 816 !(*imgp->sysent->sv_setid_allowed)(td, imgp)) || 817 (p->p_flag2 & P2_NO_NEW_PRIVS) != 0) 818 execve_nosetid(imgp); 819 820 /* 821 * Implement image setuid/setgid installation. 822 */ 823 if (imgp->credential_setid) { 824 /* 825 * Turn off syscall tracing for set-id programs, except for 826 * root. Record any set-id flags first to make sure that 827 * we do not regain any tracing during a possible block. 828 */ 829 setsugid(p); 830 #ifdef KTRACE 831 kiop = ktrprocexec(p); 832 #endif 833 /* 834 * Close any file descriptors 0..2 that reference procfs, 835 * then make sure file descriptors 0..2 are in use. 836 * 837 * Both fdsetugidsafety() and fdcheckstd() may call functions 838 * taking sleepable locks, so temporarily drop our locks. 839 */ 840 PROC_UNLOCK(p); 841 VOP_UNLOCK(imgp->vp); 842 fdsetugidsafety(td); 843 error = fdcheckstd(td); 844 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 845 if (error != 0) 846 goto exec_fail_dealloc; 847 PROC_LOCK(p); 848 #ifdef MAC 849 if (will_transition) { 850 mac_vnode_execve_transition(oldcred, imgp->newcred, 851 imgp->vp, interpvplabel, imgp); 852 } 853 #endif 854 } else { 855 if (oldcred->cr_uid == oldcred->cr_ruid && 856 oldcred->cr_gid == oldcred->cr_rgid) 857 p->p_flag &= ~P_SUGID; 858 } 859 /* 860 * Set the new credentials. 861 */ 862 if (imgp->newcred != NULL) { 863 proc_set_cred(p, imgp->newcred); 864 crfree(oldcred); 865 oldcred = NULL; 866 } 867 868 /* 869 * Store the vp for use in kern.proc.pathname. This vnode was 870 * referenced by namei() or by fexecve variant of fname handling. 871 */ 872 oldtextvp = p->p_textvp; 873 p->p_textvp = newtextvp; 874 oldtextdvp = p->p_textdvp; 875 p->p_textdvp = newtextdvp; 876 newtextdvp = NULL; 877 oldbinname = p->p_binname; 878 p->p_binname = newbinname; 879 newbinname = NULL; 880 881 #ifdef KDTRACE_HOOKS 882 /* 883 * Tell the DTrace fasttrap provider about the exec if it 884 * has declared an interest. 885 */ 886 if (dtrace_fasttrap_exec) 887 dtrace_fasttrap_exec(p); 888 #endif 889 890 /* 891 * Notify others that we exec'd, and clear the P_INEXEC flag 892 * as we're now a bona fide freshly-execed process. 893 */ 894 KNOTE_LOCKED(p->p_klist, NOTE_EXEC); 895 p->p_flag &= ~P_INEXEC; 896 897 /* clear "fork but no exec" flag, as we _are_ execing */ 898 p->p_acflag &= ~AFORK; 899 900 /* 901 * Free any previous argument cache and replace it with 902 * the new argument cache, if any. 903 */ 904 oldargs = p->p_args; 905 p->p_args = newargs; 906 newargs = NULL; 907 908 PROC_UNLOCK(p); 909 910 #ifdef HWPMC_HOOKS 911 /* 912 * Check if system-wide sampling is in effect or if the 913 * current process is using PMCs. If so, do exec() time 914 * processing. This processing needs to happen AFTER the 915 * P_INEXEC flag is cleared. 916 */ 917 if (PMC_SYSTEM_SAMPLING_ACTIVE() || PMC_PROC_IS_USING_PMCS(p)) { 918 VOP_UNLOCK(imgp->vp); 919 pe.pm_credentialschanged = credential_changing; 920 pe.pm_baseaddr = imgp->reloc_base; 921 pe.pm_dynaddr = imgp->et_dyn_addr; 922 923 PMC_CALL_HOOK_X(td, PMC_FN_PROCESS_EXEC, (void *) &pe); 924 vn_lock(imgp->vp, LK_SHARED | LK_RETRY); 925 } 926 #endif 927 928 /* Set values passed into the program in registers. */ 929 (*p->p_sysent->sv_setregs)(td, imgp, stack_base); 930 931 VOP_MMAPPED(imgp->vp); 932 933 SDT_PROBE1(proc, , , exec__success, args->fname); 934 935 exec_fail_dealloc: 936 if (error != 0) { 937 p->p_osrel = orig_osrel; 938 p->p_fctl0 = orig_fctl0; 939 p->p_elf_brandinfo = orig_brandinfo; 940 } 941 942 if (imgp->firstpage != NULL) 943 exec_unmap_first_page(imgp); 944 945 if (imgp->vp != NULL) { 946 if (imgp->opened) 947 VOP_CLOSE(imgp->vp, FREAD, td->td_ucred, td); 948 if (imgp->textset) 949 VOP_UNSET_TEXT_CHECKED(imgp->vp); 950 if (error != 0) 951 vput(imgp->vp); 952 else 953 VOP_UNLOCK(imgp->vp); 954 if (args->fname != NULL) 955 NDFREE_PNBUF(&nd); 956 if (newtextdvp != NULL) 957 vrele(newtextdvp); 958 free(newbinname, M_PARGS); 959 } 960 961 if (imgp->object != NULL) 962 vm_object_deallocate(imgp->object); 963 964 free(imgp->freepath, M_TEMP); 965 966 if (error == 0) { 967 if (p->p_ptevents & PTRACE_EXEC) { 968 PROC_LOCK(p); 969 if (p->p_ptevents & PTRACE_EXEC) 970 td->td_dbgflags |= TDB_EXEC; 971 PROC_UNLOCK(p); 972 } 973 } else { 974 exec_fail: 975 /* we're done here, clear P_INEXEC */ 976 PROC_LOCK(p); 977 p->p_flag &= ~P_INEXEC; 978 PROC_UNLOCK(p); 979 980 SDT_PROBE1(proc, , , exec__failure, error); 981 } 982 983 if (imgp->newcred != NULL && oldcred != NULL) 984 crfree(imgp->newcred); 985 986 #ifdef MAC 987 mac_execve_exit(imgp); 988 mac_execve_interpreter_exit(interpvplabel); 989 #endif 990 exec_free_args(args); 991 992 /* 993 * Handle deferred decrement of ref counts. 994 */ 995 if (oldtextvp != NULL) 996 vrele(oldtextvp); 997 if (oldtextdvp != NULL) 998 vrele(oldtextdvp); 999 free(oldbinname, M_PARGS); 1000 #ifdef KTRACE 1001 ktr_io_params_free(kiop); 1002 #endif 1003 pargs_drop(oldargs); 1004 pargs_drop(newargs); 1005 if (oldsigacts != NULL) 1006 sigacts_free(oldsigacts); 1007 if (euip != NULL) 1008 uifree(euip); 1009 1010 if (error && imgp->vmspace_destroyed) { 1011 /* sorry, no more process anymore. exit gracefully */ 1012 exec_cleanup(td, oldvmspace); 1013 exit1(td, 0, SIGABRT); 1014 /* NOT REACHED */ 1015 } 1016 1017 #ifdef KTRACE 1018 if (error == 0) 1019 ktrprocctor(p); 1020 #endif 1021 1022 /* 1023 * We don't want cpu_set_syscall_retval() to overwrite any of 1024 * the register values put in place by exec_setregs(). 1025 * Implementations of cpu_set_syscall_retval() will leave 1026 * registers unmodified when returning EJUSTRETURN. 1027 */ 1028 return (error == 0 ? EJUSTRETURN : error); 1029 } 1030 1031 void 1032 exec_cleanup(struct thread *td, struct vmspace *oldvmspace) 1033 { 1034 if ((td->td_pflags & TDP_EXECVMSPC) != 0) { 1035 KASSERT(td->td_proc->p_vmspace != oldvmspace, 1036 ("oldvmspace still used")); 1037 vmspace_free(oldvmspace); 1038 td->td_pflags &= ~TDP_EXECVMSPC; 1039 } 1040 } 1041 1042 int 1043 exec_map_first_page(struct image_params *imgp) 1044 { 1045 vm_object_t object; 1046 vm_page_t m; 1047 int error; 1048 1049 if (imgp->firstpage != NULL) 1050 exec_unmap_first_page(imgp); 1051 1052 object = imgp->vp->v_object; 1053 if (object == NULL) 1054 return (EACCES); 1055 #if VM_NRESERVLEVEL > 0 1056 if ((object->flags & OBJ_COLORED) == 0) { 1057 VM_OBJECT_WLOCK(object); 1058 vm_object_color(object, 0); 1059 VM_OBJECT_WUNLOCK(object); 1060 } 1061 #endif 1062 error = vm_page_grab_valid_unlocked(&m, object, 0, 1063 VM_ALLOC_COUNT(VM_INITIAL_PAGEIN) | 1064 VM_ALLOC_NORMAL | VM_ALLOC_NOBUSY | VM_ALLOC_WIRED); 1065 1066 if (error != VM_PAGER_OK) 1067 return (EIO); 1068 imgp->firstpage = sf_buf_alloc(m, 0); 1069 imgp->image_header = (char *)sf_buf_kva(imgp->firstpage); 1070 1071 return (0); 1072 } 1073 1074 void 1075 exec_unmap_first_page(struct image_params *imgp) 1076 { 1077 vm_page_t m; 1078 1079 if (imgp->firstpage != NULL) { 1080 m = sf_buf_page(imgp->firstpage); 1081 sf_buf_free(imgp->firstpage); 1082 imgp->firstpage = NULL; 1083 vm_page_unwire(m, PQ_ACTIVE); 1084 } 1085 } 1086 1087 void 1088 exec_onexec_old(struct thread *td) 1089 { 1090 sigfastblock_clear(td); 1091 umtx_exec(td->td_proc); 1092 } 1093 1094 /* 1095 * This is an optimization which removes the unmanaged shared page 1096 * mapping. In combination with pmap_remove_pages(), which cleans all 1097 * managed mappings in the process' vmspace pmap, no work will be left 1098 * for pmap_remove(min, max). 1099 */ 1100 void 1101 exec_free_abi_mappings(struct proc *p) 1102 { 1103 struct vmspace *vmspace; 1104 1105 vmspace = p->p_vmspace; 1106 if (refcount_load(&vmspace->vm_refcnt) != 1) 1107 return; 1108 1109 if (!PROC_HAS_SHP(p)) 1110 return; 1111 1112 pmap_remove(vmspace_pmap(vmspace), vmspace->vm_shp_base, 1113 vmspace->vm_shp_base + p->p_sysent->sv_shared_page_len); 1114 } 1115 1116 /* 1117 * Run down the current address space and install a new one. 1118 */ 1119 int 1120 exec_new_vmspace(struct image_params *imgp, struct sysentvec *sv) 1121 { 1122 int error; 1123 struct proc *p = imgp->proc; 1124 struct vmspace *vmspace = p->p_vmspace; 1125 struct thread *td = curthread; 1126 vm_offset_t sv_minuser; 1127 vm_map_t map; 1128 1129 imgp->vmspace_destroyed = true; 1130 imgp->sysent = sv; 1131 1132 if (p->p_sysent->sv_onexec_old != NULL) 1133 p->p_sysent->sv_onexec_old(td); 1134 itimers_exec(p); 1135 1136 EVENTHANDLER_DIRECT_INVOKE(process_exec, p, imgp); 1137 1138 /* 1139 * Blow away entire process VM, if address space not shared, 1140 * otherwise, create a new VM space so that other threads are 1141 * not disrupted 1142 */ 1143 map = &vmspace->vm_map; 1144 if (map_at_zero) 1145 sv_minuser = sv->sv_minuser; 1146 else 1147 sv_minuser = MAX(sv->sv_minuser, PAGE_SIZE); 1148 if (refcount_load(&vmspace->vm_refcnt) == 1 && 1149 vm_map_min(map) == sv_minuser && 1150 vm_map_max(map) == sv->sv_maxuser && 1151 cpu_exec_vmspace_reuse(p, map)) { 1152 exec_free_abi_mappings(p); 1153 shmexit(vmspace); 1154 pmap_remove_pages(vmspace_pmap(vmspace)); 1155 vm_map_remove(map, vm_map_min(map), vm_map_max(map)); 1156 /* 1157 * An exec terminates mlockall(MCL_FUTURE). 1158 * ASLR and W^X states must be re-evaluated. 1159 */ 1160 vm_map_lock(map); 1161 vm_map_modflags(map, 0, MAP_WIREFUTURE | MAP_ASLR | 1162 MAP_ASLR_IGNSTART | MAP_ASLR_STACK | MAP_WXORX); 1163 vm_map_unlock(map); 1164 } else { 1165 error = vmspace_exec(p, sv_minuser, sv->sv_maxuser); 1166 if (error) 1167 return (error); 1168 vmspace = p->p_vmspace; 1169 map = &vmspace->vm_map; 1170 } 1171 map->flags |= imgp->map_flags; 1172 1173 return (sv->sv_onexec != NULL ? sv->sv_onexec(p, imgp) : 0); 1174 } 1175 1176 /* 1177 * Compute the stack size limit and map the main process stack. 1178 * Map the shared page. 1179 */ 1180 int 1181 exec_map_stack(struct image_params *imgp) 1182 { 1183 struct rlimit rlim_stack; 1184 struct sysentvec *sv; 1185 struct proc *p; 1186 vm_map_t map; 1187 struct vmspace *vmspace; 1188 vm_offset_t stack_addr, stack_top; 1189 vm_offset_t sharedpage_addr; 1190 u_long ssiz; 1191 int error, find_space, stack_off; 1192 vm_prot_t stack_prot; 1193 vm_object_t obj; 1194 1195 p = imgp->proc; 1196 sv = p->p_sysent; 1197 1198 if (imgp->stack_sz != 0) { 1199 ssiz = trunc_page(imgp->stack_sz); 1200 PROC_LOCK(p); 1201 lim_rlimit_proc(p, RLIMIT_STACK, &rlim_stack); 1202 PROC_UNLOCK(p); 1203 if (ssiz > rlim_stack.rlim_max) 1204 ssiz = rlim_stack.rlim_max; 1205 if (ssiz > rlim_stack.rlim_cur) { 1206 rlim_stack.rlim_cur = ssiz; 1207 kern_setrlimit(curthread, RLIMIT_STACK, &rlim_stack); 1208 } 1209 } else if (sv->sv_maxssiz != NULL) { 1210 ssiz = *sv->sv_maxssiz; 1211 } else { 1212 ssiz = maxssiz; 1213 } 1214 1215 vmspace = p->p_vmspace; 1216 map = &vmspace->vm_map; 1217 1218 stack_prot = sv->sv_shared_page_obj != NULL && imgp->stack_prot != 0 ? 1219 imgp->stack_prot : sv->sv_stackprot; 1220 if ((map->flags & MAP_ASLR_STACK) != 0) { 1221 stack_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr + 1222 lim_max(curthread, RLIMIT_DATA)); 1223 find_space = VMFS_ANY_SPACE; 1224 } else { 1225 stack_addr = sv->sv_usrstack - ssiz; 1226 find_space = VMFS_NO_SPACE; 1227 } 1228 error = vm_map_find(map, NULL, 0, &stack_addr, (vm_size_t)ssiz, 1229 sv->sv_usrstack, find_space, stack_prot, VM_PROT_ALL, 1230 MAP_STACK_GROWS_DOWN); 1231 if (error != KERN_SUCCESS) { 1232 uprintf("exec_new_vmspace: mapping stack size %#jx prot %#x " 1233 "failed, mach error %d errno %d\n", (uintmax_t)ssiz, 1234 stack_prot, error, vm_mmap_to_errno(error)); 1235 return (vm_mmap_to_errno(error)); 1236 } 1237 1238 stack_top = stack_addr + ssiz; 1239 if ((map->flags & MAP_ASLR_STACK) != 0) { 1240 /* Randomize within the first page of the stack. */ 1241 arc4rand(&stack_off, sizeof(stack_off), 0); 1242 stack_top -= rounddown2(stack_off & PAGE_MASK, sizeof(void *)); 1243 } 1244 1245 /* Map a shared page */ 1246 obj = sv->sv_shared_page_obj; 1247 if (obj == NULL) { 1248 sharedpage_addr = 0; 1249 goto out; 1250 } 1251 1252 /* 1253 * If randomization is disabled then the shared page will 1254 * be mapped at address specified in sysentvec. 1255 * Otherwise any address above .data section can be selected. 1256 * Same logic is used for stack address randomization. 1257 * If the address randomization is applied map a guard page 1258 * at the top of UVA. 1259 */ 1260 vm_object_reference(obj); 1261 if ((imgp->imgp_flags & IMGP_ASLR_SHARED_PAGE) != 0) { 1262 sharedpage_addr = round_page((vm_offset_t)p->p_vmspace->vm_daddr + 1263 lim_max(curthread, RLIMIT_DATA)); 1264 1265 error = vm_map_fixed(map, NULL, 0, 1266 sv->sv_maxuser - PAGE_SIZE, PAGE_SIZE, 1267 VM_PROT_NONE, VM_PROT_NONE, MAP_CREATE_GUARD); 1268 if (error != KERN_SUCCESS) { 1269 /* 1270 * This is not fatal, so let's just print a warning 1271 * and continue. 1272 */ 1273 uprintf("%s: Mapping guard page at the top of UVA failed" 1274 " mach error %d errno %d", 1275 __func__, error, vm_mmap_to_errno(error)); 1276 } 1277 1278 error = vm_map_find(map, obj, 0, 1279 &sharedpage_addr, sv->sv_shared_page_len, 1280 sv->sv_maxuser, VMFS_ANY_SPACE, 1281 VM_PROT_READ | VM_PROT_EXECUTE, 1282 VM_PROT_READ | VM_PROT_EXECUTE, 1283 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1284 } else { 1285 sharedpage_addr = sv->sv_shared_page_base; 1286 vm_map_fixed(map, obj, 0, 1287 sharedpage_addr, sv->sv_shared_page_len, 1288 VM_PROT_READ | VM_PROT_EXECUTE, 1289 VM_PROT_READ | VM_PROT_EXECUTE, 1290 MAP_INHERIT_SHARE | MAP_ACC_NO_CHARGE); 1291 } 1292 if (error != KERN_SUCCESS) { 1293 uprintf("%s: mapping shared page at addr: %p" 1294 "failed, mach error %d errno %d\n", __func__, 1295 (void *)sharedpage_addr, error, vm_mmap_to_errno(error)); 1296 vm_object_deallocate(obj); 1297 return (vm_mmap_to_errno(error)); 1298 } 1299 out: 1300 /* 1301 * vm_ssize and vm_maxsaddr are somewhat antiquated concepts, but they 1302 * are still used to enforce the stack rlimit on the process stack. 1303 */ 1304 vmspace->vm_maxsaddr = (char *)stack_addr; 1305 vmspace->vm_stacktop = stack_top; 1306 vmspace->vm_ssize = sgrowsiz >> PAGE_SHIFT; 1307 vmspace->vm_shp_base = sharedpage_addr; 1308 1309 return (0); 1310 } 1311 1312 /* 1313 * Copy out argument and environment strings from the old process address 1314 * space into the temporary string buffer. 1315 */ 1316 int 1317 exec_copyin_args(struct image_args *args, const char *fname, 1318 enum uio_seg segflg, char **argv, char **envv) 1319 { 1320 u_long arg, env; 1321 int error; 1322 1323 bzero(args, sizeof(*args)); 1324 if (argv == NULL) 1325 return (EFAULT); 1326 1327 /* 1328 * Allocate demand-paged memory for the file name, argument, and 1329 * environment strings. 1330 */ 1331 error = exec_alloc_args(args); 1332 if (error != 0) 1333 return (error); 1334 1335 /* 1336 * Copy the file name. 1337 */ 1338 error = exec_args_add_fname(args, fname, segflg); 1339 if (error != 0) 1340 goto err_exit; 1341 1342 /* 1343 * extract arguments first 1344 */ 1345 for (;;) { 1346 error = fueword(argv++, &arg); 1347 if (error == -1) { 1348 error = EFAULT; 1349 goto err_exit; 1350 } 1351 if (arg == 0) 1352 break; 1353 error = exec_args_add_arg(args, (char *)(uintptr_t)arg, 1354 UIO_USERSPACE); 1355 if (error != 0) 1356 goto err_exit; 1357 } 1358 1359 /* 1360 * extract environment strings 1361 */ 1362 if (envv) { 1363 for (;;) { 1364 error = fueword(envv++, &env); 1365 if (error == -1) { 1366 error = EFAULT; 1367 goto err_exit; 1368 } 1369 if (env == 0) 1370 break; 1371 error = exec_args_add_env(args, 1372 (char *)(uintptr_t)env, UIO_USERSPACE); 1373 if (error != 0) 1374 goto err_exit; 1375 } 1376 } 1377 1378 return (0); 1379 1380 err_exit: 1381 exec_free_args(args); 1382 return (error); 1383 } 1384 1385 struct exec_args_kva { 1386 vm_offset_t addr; 1387 u_int gen; 1388 SLIST_ENTRY(exec_args_kva) next; 1389 }; 1390 1391 DPCPU_DEFINE_STATIC(struct exec_args_kva *, exec_args_kva); 1392 1393 static SLIST_HEAD(, exec_args_kva) exec_args_kva_freelist; 1394 static struct mtx exec_args_kva_mtx; 1395 static u_int exec_args_gen; 1396 1397 static void 1398 exec_prealloc_args_kva(void *arg __unused) 1399 { 1400 struct exec_args_kva *argkva; 1401 u_int i; 1402 1403 SLIST_INIT(&exec_args_kva_freelist); 1404 mtx_init(&exec_args_kva_mtx, "exec args kva", NULL, MTX_DEF); 1405 for (i = 0; i < exec_map_entries; i++) { 1406 argkva = malloc(sizeof(*argkva), M_PARGS, M_WAITOK); 1407 argkva->addr = kmap_alloc_wait(exec_map, exec_map_entry_size); 1408 argkva->gen = exec_args_gen; 1409 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1410 } 1411 } 1412 SYSINIT(exec_args_kva, SI_SUB_EXEC, SI_ORDER_ANY, exec_prealloc_args_kva, NULL); 1413 1414 static vm_offset_t 1415 exec_alloc_args_kva(void **cookie) 1416 { 1417 struct exec_args_kva *argkva; 1418 1419 argkva = (void *)atomic_readandclear_ptr( 1420 (uintptr_t *)DPCPU_PTR(exec_args_kva)); 1421 if (argkva == NULL) { 1422 mtx_lock(&exec_args_kva_mtx); 1423 while ((argkva = SLIST_FIRST(&exec_args_kva_freelist)) == NULL) 1424 (void)mtx_sleep(&exec_args_kva_freelist, 1425 &exec_args_kva_mtx, 0, "execkva", 0); 1426 SLIST_REMOVE_HEAD(&exec_args_kva_freelist, next); 1427 mtx_unlock(&exec_args_kva_mtx); 1428 } 1429 kasan_mark((void *)argkva->addr, exec_map_entry_size, 1430 exec_map_entry_size, 0); 1431 *(struct exec_args_kva **)cookie = argkva; 1432 return (argkva->addr); 1433 } 1434 1435 static void 1436 exec_release_args_kva(struct exec_args_kva *argkva, u_int gen) 1437 { 1438 vm_offset_t base; 1439 1440 base = argkva->addr; 1441 kasan_mark((void *)argkva->addr, 0, exec_map_entry_size, 1442 KASAN_EXEC_ARGS_FREED); 1443 if (argkva->gen != gen) { 1444 (void)vm_map_madvise(exec_map, base, base + exec_map_entry_size, 1445 MADV_FREE); 1446 argkva->gen = gen; 1447 } 1448 if (!atomic_cmpset_ptr((uintptr_t *)DPCPU_PTR(exec_args_kva), 1449 (uintptr_t)NULL, (uintptr_t)argkva)) { 1450 mtx_lock(&exec_args_kva_mtx); 1451 SLIST_INSERT_HEAD(&exec_args_kva_freelist, argkva, next); 1452 wakeup_one(&exec_args_kva_freelist); 1453 mtx_unlock(&exec_args_kva_mtx); 1454 } 1455 } 1456 1457 static void 1458 exec_free_args_kva(void *cookie) 1459 { 1460 1461 exec_release_args_kva(cookie, exec_args_gen); 1462 } 1463 1464 static void 1465 exec_args_kva_lowmem(void *arg __unused) 1466 { 1467 SLIST_HEAD(, exec_args_kva) head; 1468 struct exec_args_kva *argkva; 1469 u_int gen; 1470 int i; 1471 1472 gen = atomic_fetchadd_int(&exec_args_gen, 1) + 1; 1473 1474 /* 1475 * Force an madvise of each KVA range. Any currently allocated ranges 1476 * will have MADV_FREE applied once they are freed. 1477 */ 1478 SLIST_INIT(&head); 1479 mtx_lock(&exec_args_kva_mtx); 1480 SLIST_SWAP(&head, &exec_args_kva_freelist, exec_args_kva); 1481 mtx_unlock(&exec_args_kva_mtx); 1482 while ((argkva = SLIST_FIRST(&head)) != NULL) { 1483 SLIST_REMOVE_HEAD(&head, next); 1484 exec_release_args_kva(argkva, gen); 1485 } 1486 1487 CPU_FOREACH(i) { 1488 argkva = (void *)atomic_readandclear_ptr( 1489 (uintptr_t *)DPCPU_ID_PTR(i, exec_args_kva)); 1490 if (argkva != NULL) 1491 exec_release_args_kva(argkva, gen); 1492 } 1493 } 1494 EVENTHANDLER_DEFINE(vm_lowmem, exec_args_kva_lowmem, NULL, 1495 EVENTHANDLER_PRI_ANY); 1496 1497 /* 1498 * Allocate temporary demand-paged, zero-filled memory for the file name, 1499 * argument, and environment strings. 1500 */ 1501 int 1502 exec_alloc_args(struct image_args *args) 1503 { 1504 1505 args->buf = (char *)exec_alloc_args_kva(&args->bufkva); 1506 return (0); 1507 } 1508 1509 void 1510 exec_free_args(struct image_args *args) 1511 { 1512 1513 if (args->buf != NULL) { 1514 exec_free_args_kva(args->bufkva); 1515 args->buf = NULL; 1516 } 1517 if (args->fname_buf != NULL) { 1518 free(args->fname_buf, M_TEMP); 1519 args->fname_buf = NULL; 1520 } 1521 } 1522 1523 /* 1524 * A set to functions to fill struct image args. 1525 * 1526 * NOTE: exec_args_add_fname() must be called (possibly with a NULL 1527 * fname) before the other functions. All exec_args_add_arg() calls must 1528 * be made before any exec_args_add_env() calls. exec_args_adjust_args() 1529 * may be called any time after exec_args_add_fname(). 1530 * 1531 * exec_args_add_fname() - install path to be executed 1532 * exec_args_add_arg() - append an argument string 1533 * exec_args_add_env() - append an env string 1534 * exec_args_adjust_args() - adjust location of the argument list to 1535 * allow new arguments to be prepended 1536 */ 1537 int 1538 exec_args_add_fname(struct image_args *args, const char *fname, 1539 enum uio_seg segflg) 1540 { 1541 int error; 1542 size_t length; 1543 1544 KASSERT(args->fname == NULL, ("fname already appended")); 1545 KASSERT(args->endp == NULL, ("already appending to args")); 1546 1547 if (fname != NULL) { 1548 args->fname = args->buf; 1549 error = segflg == UIO_SYSSPACE ? 1550 copystr(fname, args->fname, PATH_MAX, &length) : 1551 copyinstr(fname, args->fname, PATH_MAX, &length); 1552 if (error != 0) 1553 return (error == ENAMETOOLONG ? E2BIG : error); 1554 } else 1555 length = 0; 1556 1557 /* Set up for _arg_*()/_env_*() */ 1558 args->endp = args->buf + length; 1559 /* begin_argv must be set and kept updated */ 1560 args->begin_argv = args->endp; 1561 KASSERT(exec_map_entry_size - length >= ARG_MAX, 1562 ("too little space remaining for arguments %zu < %zu", 1563 exec_map_entry_size - length, (size_t)ARG_MAX)); 1564 args->stringspace = ARG_MAX; 1565 1566 return (0); 1567 } 1568 1569 static int 1570 exec_args_add_str(struct image_args *args, const char *str, 1571 enum uio_seg segflg, int *countp) 1572 { 1573 int error; 1574 size_t length; 1575 1576 KASSERT(args->endp != NULL, ("endp not initialized")); 1577 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1578 1579 error = (segflg == UIO_SYSSPACE) ? 1580 copystr(str, args->endp, args->stringspace, &length) : 1581 copyinstr(str, args->endp, args->stringspace, &length); 1582 if (error != 0) 1583 return (error == ENAMETOOLONG ? E2BIG : error); 1584 args->stringspace -= length; 1585 args->endp += length; 1586 (*countp)++; 1587 1588 return (0); 1589 } 1590 1591 int 1592 exec_args_add_arg(struct image_args *args, const char *argp, 1593 enum uio_seg segflg) 1594 { 1595 1596 KASSERT(args->envc == 0, ("appending args after env")); 1597 1598 return (exec_args_add_str(args, argp, segflg, &args->argc)); 1599 } 1600 1601 int 1602 exec_args_add_env(struct image_args *args, const char *envp, 1603 enum uio_seg segflg) 1604 { 1605 1606 if (args->envc == 0) 1607 args->begin_envv = args->endp; 1608 1609 return (exec_args_add_str(args, envp, segflg, &args->envc)); 1610 } 1611 1612 int 1613 exec_args_adjust_args(struct image_args *args, size_t consume, ssize_t extend) 1614 { 1615 ssize_t offset; 1616 1617 KASSERT(args->endp != NULL, ("endp not initialized")); 1618 KASSERT(args->begin_argv != NULL, ("begin_argp not initialized")); 1619 1620 offset = extend - consume; 1621 if (args->stringspace < offset) 1622 return (E2BIG); 1623 memmove(args->begin_argv + extend, args->begin_argv + consume, 1624 args->endp - args->begin_argv + consume); 1625 if (args->envc > 0) 1626 args->begin_envv += offset; 1627 args->endp += offset; 1628 args->stringspace -= offset; 1629 return (0); 1630 } 1631 1632 char * 1633 exec_args_get_begin_envv(struct image_args *args) 1634 { 1635 1636 KASSERT(args->endp != NULL, ("endp not initialized")); 1637 1638 if (args->envc > 0) 1639 return (args->begin_envv); 1640 return (args->endp); 1641 } 1642 1643 /* 1644 * Copy strings out to the new process address space, constructing new arg 1645 * and env vector tables. Return a pointer to the base so that it can be used 1646 * as the initial stack pointer. 1647 */ 1648 int 1649 exec_copyout_strings(struct image_params *imgp, uintptr_t *stack_base) 1650 { 1651 int argc, envc; 1652 char **vectp; 1653 char *stringp; 1654 uintptr_t destp, ustringp; 1655 struct ps_strings *arginfo; 1656 struct proc *p; 1657 struct sysentvec *sysent; 1658 size_t execpath_len; 1659 int error, szsigcode; 1660 char canary[sizeof(long) * 8]; 1661 1662 p = imgp->proc; 1663 sysent = p->p_sysent; 1664 1665 destp = PROC_PS_STRINGS(p); 1666 arginfo = imgp->ps_strings = (void *)destp; 1667 1668 /* 1669 * Install sigcode. 1670 */ 1671 if (sysent->sv_shared_page_base == 0 && sysent->sv_szsigcode != NULL) { 1672 szsigcode = *(sysent->sv_szsigcode); 1673 destp -= szsigcode; 1674 destp = rounddown2(destp, sizeof(void *)); 1675 error = copyout(sysent->sv_sigcode, (void *)destp, szsigcode); 1676 if (error != 0) 1677 return (error); 1678 } 1679 1680 /* 1681 * Copy the image path for the rtld. 1682 */ 1683 if (imgp->execpath != NULL && imgp->auxargs != NULL) { 1684 execpath_len = strlen(imgp->execpath) + 1; 1685 destp -= execpath_len; 1686 destp = rounddown2(destp, sizeof(void *)); 1687 imgp->execpathp = (void *)destp; 1688 error = copyout(imgp->execpath, imgp->execpathp, execpath_len); 1689 if (error != 0) 1690 return (error); 1691 } 1692 1693 /* 1694 * Prepare the canary for SSP. 1695 */ 1696 arc4rand(canary, sizeof(canary), 0); 1697 destp -= sizeof(canary); 1698 imgp->canary = (void *)destp; 1699 error = copyout(canary, imgp->canary, sizeof(canary)); 1700 if (error != 0) 1701 return (error); 1702 imgp->canarylen = sizeof(canary); 1703 1704 /* 1705 * Prepare the pagesizes array. 1706 */ 1707 imgp->pagesizeslen = sizeof(pagesizes[0]) * MAXPAGESIZES; 1708 destp -= imgp->pagesizeslen; 1709 destp = rounddown2(destp, sizeof(void *)); 1710 imgp->pagesizes = (void *)destp; 1711 error = copyout(pagesizes, imgp->pagesizes, imgp->pagesizeslen); 1712 if (error != 0) 1713 return (error); 1714 1715 /* 1716 * Allocate room for the argument and environment strings. 1717 */ 1718 destp -= ARG_MAX - imgp->args->stringspace; 1719 destp = rounddown2(destp, sizeof(void *)); 1720 ustringp = destp; 1721 1722 if (imgp->auxargs) { 1723 /* 1724 * Allocate room on the stack for the ELF auxargs 1725 * array. It has up to AT_COUNT entries. 1726 */ 1727 destp -= AT_COUNT * sizeof(Elf_Auxinfo); 1728 destp = rounddown2(destp, sizeof(void *)); 1729 } 1730 1731 vectp = (char **)destp; 1732 1733 /* 1734 * Allocate room for the argv[] and env vectors including the 1735 * terminating NULL pointers. 1736 */ 1737 vectp -= imgp->args->argc + 1 + imgp->args->envc + 1; 1738 1739 /* 1740 * vectp also becomes our initial stack base 1741 */ 1742 *stack_base = (uintptr_t)vectp; 1743 1744 stringp = imgp->args->begin_argv; 1745 argc = imgp->args->argc; 1746 envc = imgp->args->envc; 1747 1748 /* 1749 * Copy out strings - arguments and environment. 1750 */ 1751 error = copyout(stringp, (void *)ustringp, 1752 ARG_MAX - imgp->args->stringspace); 1753 if (error != 0) 1754 return (error); 1755 1756 /* 1757 * Fill in "ps_strings" struct for ps, w, etc. 1758 */ 1759 imgp->argv = vectp; 1760 if (suword(&arginfo->ps_argvstr, (long)(intptr_t)vectp) != 0 || 1761 suword32(&arginfo->ps_nargvstr, argc) != 0) 1762 return (EFAULT); 1763 1764 /* 1765 * Fill in argument portion of vector table. 1766 */ 1767 for (; argc > 0; --argc) { 1768 if (suword(vectp++, ustringp) != 0) 1769 return (EFAULT); 1770 while (*stringp++ != 0) 1771 ustringp++; 1772 ustringp++; 1773 } 1774 1775 /* a null vector table pointer separates the argp's from the envp's */ 1776 if (suword(vectp++, 0) != 0) 1777 return (EFAULT); 1778 1779 imgp->envv = vectp; 1780 if (suword(&arginfo->ps_envstr, (long)(intptr_t)vectp) != 0 || 1781 suword32(&arginfo->ps_nenvstr, envc) != 0) 1782 return (EFAULT); 1783 1784 /* 1785 * Fill in environment portion of vector table. 1786 */ 1787 for (; envc > 0; --envc) { 1788 if (suword(vectp++, ustringp) != 0) 1789 return (EFAULT); 1790 while (*stringp++ != 0) 1791 ustringp++; 1792 ustringp++; 1793 } 1794 1795 /* end of vector table is a null pointer */ 1796 if (suword(vectp, 0) != 0) 1797 return (EFAULT); 1798 1799 if (imgp->auxargs) { 1800 vectp++; 1801 error = imgp->sysent->sv_copyout_auxargs(imgp, 1802 (uintptr_t)vectp); 1803 if (error != 0) 1804 return (error); 1805 } 1806 1807 return (0); 1808 } 1809 1810 /* 1811 * Check permissions of file to execute. 1812 * Called with imgp->vp locked. 1813 * Return 0 for success or error code on failure. 1814 */ 1815 int 1816 exec_check_permissions(struct image_params *imgp) 1817 { 1818 struct vnode *vp = imgp->vp; 1819 struct vattr *attr = imgp->attr; 1820 struct thread *td; 1821 int error; 1822 1823 td = curthread; 1824 1825 /* Get file attributes */ 1826 error = VOP_GETATTR(vp, attr, td->td_ucred); 1827 if (error) 1828 return (error); 1829 1830 #ifdef MAC 1831 error = mac_vnode_check_exec(td->td_ucred, imgp->vp, imgp); 1832 if (error) 1833 return (error); 1834 #endif 1835 1836 /* 1837 * 1) Check if file execution is disabled for the filesystem that 1838 * this file resides on. 1839 * 2) Ensure that at least one execute bit is on. Otherwise, a 1840 * privileged user will always succeed, and we don't want this 1841 * to happen unless the file really is executable. 1842 * 3) Ensure that the file is a regular file. 1843 */ 1844 if ((vp->v_mount->mnt_flag & MNT_NOEXEC) || 1845 (attr->va_mode & (S_IXUSR | S_IXGRP | S_IXOTH)) == 0 || 1846 (attr->va_type != VREG)) 1847 return (EACCES); 1848 1849 /* 1850 * Zero length files can't be exec'd 1851 */ 1852 if (attr->va_size == 0) 1853 return (ENOEXEC); 1854 1855 /* 1856 * Check for execute permission to file based on current credentials. 1857 */ 1858 error = VOP_ACCESS(vp, VEXEC, td->td_ucred, td); 1859 if (error) 1860 return (error); 1861 1862 /* 1863 * Check number of open-for-writes on the file and deny execution 1864 * if there are any. 1865 * 1866 * Add a text reference now so no one can write to the 1867 * executable while we're activating it. 1868 * 1869 * Remember if this was set before and unset it in case this is not 1870 * actually an executable image. 1871 */ 1872 error = VOP_SET_TEXT(vp); 1873 if (error != 0) 1874 return (error); 1875 imgp->textset = true; 1876 1877 /* 1878 * Call filesystem specific open routine (which does nothing in the 1879 * general case). 1880 */ 1881 error = VOP_OPEN(vp, FREAD, td->td_ucred, td, NULL); 1882 if (error == 0) 1883 imgp->opened = true; 1884 return (error); 1885 } 1886 1887 /* 1888 * Exec handler registration 1889 */ 1890 int 1891 exec_register(const struct execsw *execsw_arg) 1892 { 1893 const struct execsw **es, **xs, **newexecsw; 1894 u_int count = 2; /* New slot and trailing NULL */ 1895 1896 if (execsw) 1897 for (es = execsw; *es; es++) 1898 count++; 1899 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1900 xs = newexecsw; 1901 if (execsw) 1902 for (es = execsw; *es; es++) 1903 *xs++ = *es; 1904 *xs++ = execsw_arg; 1905 *xs = NULL; 1906 if (execsw) 1907 free(execsw, M_TEMP); 1908 execsw = newexecsw; 1909 return (0); 1910 } 1911 1912 int 1913 exec_unregister(const struct execsw *execsw_arg) 1914 { 1915 const struct execsw **es, **xs, **newexecsw; 1916 int count = 1; 1917 1918 if (execsw == NULL) 1919 panic("unregister with no handlers left?\n"); 1920 1921 for (es = execsw; *es; es++) { 1922 if (*es == execsw_arg) 1923 break; 1924 } 1925 if (*es == NULL) 1926 return (ENOENT); 1927 for (es = execsw; *es; es++) 1928 if (*es != execsw_arg) 1929 count++; 1930 newexecsw = malloc(count * sizeof(*es), M_TEMP, M_WAITOK); 1931 xs = newexecsw; 1932 for (es = execsw; *es; es++) 1933 if (*es != execsw_arg) 1934 *xs++ = *es; 1935 *xs = NULL; 1936 if (execsw) 1937 free(execsw, M_TEMP); 1938 execsw = newexecsw; 1939 return (0); 1940 } 1941 1942 /* 1943 * Write out a core segment to the compression stream. 1944 */ 1945 static int 1946 compress_chunk(struct coredump_params *cp, char *base, char *buf, size_t len) 1947 { 1948 size_t chunk_len; 1949 int error; 1950 1951 while (len > 0) { 1952 chunk_len = MIN(len, CORE_BUF_SIZE); 1953 1954 /* 1955 * We can get EFAULT error here. 1956 * In that case zero out the current chunk of the segment. 1957 */ 1958 error = copyin(base, buf, chunk_len); 1959 if (error != 0) 1960 bzero(buf, chunk_len); 1961 error = compressor_write(cp->comp, buf, chunk_len); 1962 if (error != 0) 1963 break; 1964 base += chunk_len; 1965 len -= chunk_len; 1966 } 1967 return (error); 1968 } 1969 1970 int 1971 core_write(struct coredump_params *cp, const void *base, size_t len, 1972 off_t offset, enum uio_seg seg, size_t *resid) 1973 { 1974 1975 return (vn_rdwr_inchunks(UIO_WRITE, cp->vp, __DECONST(void *, base), 1976 len, offset, seg, IO_UNIT | IO_DIRECT | IO_RANGELOCKED, 1977 cp->active_cred, cp->file_cred, resid, cp->td)); 1978 } 1979 1980 int 1981 core_output(char *base, size_t len, off_t offset, struct coredump_params *cp, 1982 void *tmpbuf) 1983 { 1984 vm_map_t map; 1985 struct mount *mp; 1986 size_t resid, runlen; 1987 int error; 1988 bool success; 1989 1990 KASSERT((uintptr_t)base % PAGE_SIZE == 0, 1991 ("%s: user address %p is not page-aligned", __func__, base)); 1992 1993 if (cp->comp != NULL) 1994 return (compress_chunk(cp, base, tmpbuf, len)); 1995 1996 map = &cp->td->td_proc->p_vmspace->vm_map; 1997 for (; len > 0; base += runlen, offset += runlen, len -= runlen) { 1998 /* 1999 * Attempt to page in all virtual pages in the range. If a 2000 * virtual page is not backed by the pager, it is represented as 2001 * a hole in the file. This can occur with zero-filled 2002 * anonymous memory or truncated files, for example. 2003 */ 2004 for (runlen = 0; runlen < len; runlen += PAGE_SIZE) { 2005 if (core_dump_can_intr && curproc_sigkilled()) 2006 return (EINTR); 2007 error = vm_fault(map, (uintptr_t)base + runlen, 2008 VM_PROT_READ, VM_FAULT_NOFILL, NULL); 2009 if (runlen == 0) 2010 success = error == KERN_SUCCESS; 2011 else if ((error == KERN_SUCCESS) != success) 2012 break; 2013 } 2014 2015 if (success) { 2016 error = core_write(cp, base, runlen, offset, 2017 UIO_USERSPACE, &resid); 2018 if (error != 0) { 2019 if (error != EFAULT) 2020 break; 2021 2022 /* 2023 * EFAULT may be returned if the user mapping 2024 * could not be accessed, e.g., because a mapped 2025 * file has been truncated. Skip the page if no 2026 * progress was made, to protect against a 2027 * hypothetical scenario where vm_fault() was 2028 * successful but core_write() returns EFAULT 2029 * anyway. 2030 */ 2031 runlen -= resid; 2032 if (runlen == 0) { 2033 success = false; 2034 runlen = PAGE_SIZE; 2035 } 2036 } 2037 } 2038 if (!success) { 2039 error = vn_start_write(cp->vp, &mp, V_WAIT); 2040 if (error != 0) 2041 break; 2042 vn_lock(cp->vp, LK_EXCLUSIVE | LK_RETRY); 2043 error = vn_truncate_locked(cp->vp, offset + runlen, 2044 false, cp->td->td_ucred); 2045 VOP_UNLOCK(cp->vp); 2046 vn_finished_write(mp); 2047 if (error != 0) 2048 break; 2049 } 2050 } 2051 return (error); 2052 } 2053 2054 /* 2055 * Drain into a core file. 2056 */ 2057 int 2058 sbuf_drain_core_output(void *arg, const char *data, int len) 2059 { 2060 struct coredump_params *cp; 2061 struct proc *p; 2062 int error, locked; 2063 2064 cp = arg; 2065 p = cp->td->td_proc; 2066 2067 /* 2068 * Some kern_proc out routines that print to this sbuf may 2069 * call us with the process lock held. Draining with the 2070 * non-sleepable lock held is unsafe. The lock is needed for 2071 * those routines when dumping a live process. In our case we 2072 * can safely release the lock before draining and acquire 2073 * again after. 2074 */ 2075 locked = PROC_LOCKED(p); 2076 if (locked) 2077 PROC_UNLOCK(p); 2078 if (cp->comp != NULL) 2079 error = compressor_write(cp->comp, __DECONST(char *, data), 2080 len); 2081 else 2082 error = core_write(cp, __DECONST(void *, data), len, cp->offset, 2083 UIO_SYSSPACE, NULL); 2084 if (locked) 2085 PROC_LOCK(p); 2086 if (error != 0) 2087 return (-error); 2088 cp->offset += len; 2089 return (len); 2090 } 2091