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