1 /* $NetBSD: kern_exec.c,v 1.509 2021/09/28 15:35:44 thorpej Exp $ */ 2 3 /*- 4 * Copyright (c) 2008, 2019, 2020 The NetBSD Foundation, Inc. 5 * All rights reserved. 6 * 7 * This code is derived from software contributed to The NetBSD Foundation 8 * by Andrew Doran. 9 * 10 * Redistribution and use in source and binary forms, with or without 11 * modification, are permitted provided that the following conditions 12 * are met: 13 * 1. Redistributions of source code must retain the above copyright 14 * notice, this list of conditions and the following disclaimer. 15 * 2. Redistributions in binary form must reproduce the above copyright 16 * notice, this list of conditions and the following disclaimer in the 17 * documentation and/or other materials provided with the distribution. 18 * 19 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS 20 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED 21 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR 22 * PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS 23 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR 24 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF 25 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS 26 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN 27 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) 28 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE 29 * POSSIBILITY OF SUCH DAMAGE. 30 */ 31 32 /*- 33 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou 34 * Copyright (C) 1992 Wolfgang Solfrank. 35 * Copyright (C) 1992 TooLs GmbH. 36 * All rights reserved. 37 * 38 * Redistribution and use in source and binary forms, with or without 39 * modification, are permitted provided that the following conditions 40 * are met: 41 * 1. Redistributions of source code must retain the above copyright 42 * notice, this list of conditions and the following disclaimer. 43 * 2. Redistributions in binary form must reproduce the above copyright 44 * notice, this list of conditions and the following disclaimer in the 45 * documentation and/or other materials provided with the distribution. 46 * 3. All advertising materials mentioning features or use of this software 47 * must display the following acknowledgement: 48 * This product includes software developed by TooLs GmbH. 49 * 4. The name of TooLs GmbH may not be used to endorse or promote products 50 * derived from this software without specific prior written permission. 51 * 52 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 53 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 54 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 55 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 56 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 57 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 58 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 59 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 60 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 61 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 62 */ 63 64 #include <sys/cdefs.h> 65 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.509 2021/09/28 15:35:44 thorpej Exp $"); 66 67 #include "opt_exec.h" 68 #include "opt_execfmt.h" 69 #include "opt_ktrace.h" 70 #include "opt_modular.h" 71 #include "opt_syscall_debug.h" 72 #include "veriexec.h" 73 #include "opt_pax.h" 74 75 #include <sys/param.h> 76 #include <sys/systm.h> 77 #include <sys/filedesc.h> 78 #include <sys/kernel.h> 79 #include <sys/proc.h> 80 #include <sys/ptrace.h> 81 #include <sys/mount.h> 82 #include <sys/kmem.h> 83 #include <sys/namei.h> 84 #include <sys/vnode.h> 85 #include <sys/file.h> 86 #include <sys/filedesc.h> 87 #include <sys/acct.h> 88 #include <sys/atomic.h> 89 #include <sys/exec.h> 90 #include <sys/futex.h> 91 #include <sys/ktrace.h> 92 #include <sys/uidinfo.h> 93 #include <sys/wait.h> 94 #include <sys/mman.h> 95 #include <sys/ras.h> 96 #include <sys/signalvar.h> 97 #include <sys/stat.h> 98 #include <sys/syscall.h> 99 #include <sys/kauth.h> 100 #include <sys/lwpctl.h> 101 #include <sys/pax.h> 102 #include <sys/cpu.h> 103 #include <sys/module.h> 104 #include <sys/syscallvar.h> 105 #include <sys/syscallargs.h> 106 #if NVERIEXEC > 0 107 #include <sys/verified_exec.h> 108 #endif /* NVERIEXEC > 0 */ 109 #include <sys/sdt.h> 110 #include <sys/spawn.h> 111 #include <sys/prot.h> 112 #include <sys/cprng.h> 113 114 #include <uvm/uvm_extern.h> 115 116 #include <machine/reg.h> 117 118 #include <compat/common/compat_util.h> 119 120 #ifndef MD_TOPDOWN_INIT 121 #ifdef __USE_TOPDOWN_VM 122 #define MD_TOPDOWN_INIT(epp) (epp)->ep_flags |= EXEC_TOPDOWN_VM 123 #else 124 #define MD_TOPDOWN_INIT(epp) 125 #endif 126 #endif 127 128 struct execve_data; 129 130 extern int user_va0_disable; 131 132 static size_t calcargs(struct execve_data * restrict, const size_t); 133 static size_t calcstack(struct execve_data * restrict, const size_t); 134 static int copyoutargs(struct execve_data * restrict, struct lwp *, 135 char * const); 136 static int copyoutpsstrs(struct execve_data * restrict, struct proc *); 137 static int copyinargs(struct execve_data * restrict, char * const *, 138 char * const *, execve_fetch_element_t, char **); 139 static int copyinargstrs(struct execve_data * restrict, char * const *, 140 execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t)); 141 static int exec_sigcode_map(struct proc *, const struct emul *); 142 143 #if defined(DEBUG) && !defined(DEBUG_EXEC) 144 #define DEBUG_EXEC 145 #endif 146 #ifdef DEBUG_EXEC 147 #define DPRINTF(a) printf a 148 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \ 149 __LINE__, (s), (a), (b)) 150 static void dump_vmcmds(const struct exec_package * const, size_t, int); 151 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0) 152 #else 153 #define DPRINTF(a) 154 #define COPYPRINTF(s, a, b) 155 #define DUMPVMCMDS(p, x, e) do {} while (0) 156 #endif /* DEBUG_EXEC */ 157 158 /* 159 * DTrace SDT provider definitions 160 */ 161 SDT_PROVIDER_DECLARE(proc); 162 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *"); 163 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *"); 164 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int"); 165 166 /* 167 * Exec function switch: 168 * 169 * Note that each makecmds function is responsible for loading the 170 * exec package with the necessary functions for any exec-type-specific 171 * handling. 172 * 173 * Functions for specific exec types should be defined in their own 174 * header file. 175 */ 176 static const struct execsw **execsw = NULL; 177 static int nexecs; 178 179 u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */ 180 181 /* list of dynamically loaded execsw entries */ 182 static LIST_HEAD(execlist_head, exec_entry) ex_head = 183 LIST_HEAD_INITIALIZER(ex_head); 184 struct exec_entry { 185 LIST_ENTRY(exec_entry) ex_list; 186 SLIST_ENTRY(exec_entry) ex_slist; 187 const struct execsw *ex_sw; 188 }; 189 190 #ifndef __HAVE_SYSCALL_INTERN 191 void syscall(void); 192 #endif 193 194 /* NetBSD autoloadable syscalls */ 195 #ifdef MODULAR 196 #include <kern/syscalls_autoload.c> 197 #endif 198 199 /* NetBSD emul struct */ 200 struct emul emul_netbsd = { 201 .e_name = "netbsd", 202 #ifdef EMUL_NATIVEROOT 203 .e_path = EMUL_NATIVEROOT, 204 #else 205 .e_path = NULL, 206 #endif 207 #ifndef __HAVE_MINIMAL_EMUL 208 .e_flags = EMUL_HAS_SYS___syscall, 209 .e_errno = NULL, 210 .e_nosys = SYS_syscall, 211 .e_nsysent = SYS_NSYSENT, 212 #endif 213 #ifdef MODULAR 214 .e_sc_autoload = netbsd_syscalls_autoload, 215 #endif 216 .e_sysent = sysent, 217 .e_nomodbits = sysent_nomodbits, 218 #ifdef SYSCALL_DEBUG 219 .e_syscallnames = syscallnames, 220 #else 221 .e_syscallnames = NULL, 222 #endif 223 .e_sendsig = sendsig, 224 .e_trapsignal = trapsignal, 225 .e_sigcode = NULL, 226 .e_esigcode = NULL, 227 .e_sigobject = NULL, 228 .e_setregs = setregs, 229 .e_proc_exec = NULL, 230 .e_proc_fork = NULL, 231 .e_proc_exit = NULL, 232 .e_lwp_fork = NULL, 233 .e_lwp_exit = NULL, 234 #ifdef __HAVE_SYSCALL_INTERN 235 .e_syscall_intern = syscall_intern, 236 #else 237 .e_syscall = syscall, 238 #endif 239 .e_sysctlovly = NULL, 240 .e_vm_default_addr = uvm_default_mapaddr, 241 .e_usertrap = NULL, 242 .e_ucsize = sizeof(ucontext_t), 243 .e_startlwp = startlwp 244 }; 245 246 /* 247 * Exec lock. Used to control access to execsw[] structures. 248 * This must not be static so that netbsd32 can access it, too. 249 */ 250 krwlock_t exec_lock __cacheline_aligned; 251 252 static kmutex_t sigobject_lock __cacheline_aligned; 253 254 /* 255 * Data used between a loadvm and execve part of an "exec" operation 256 */ 257 struct execve_data { 258 struct exec_package ed_pack; 259 struct pathbuf *ed_pathbuf; 260 struct vattr ed_attr; 261 struct ps_strings ed_arginfo; 262 char *ed_argp; 263 const char *ed_pathstring; 264 char *ed_resolvedname; 265 size_t ed_ps_strings_sz; 266 int ed_szsigcode; 267 size_t ed_argslen; 268 long ed_argc; 269 long ed_envc; 270 }; 271 272 /* 273 * data passed from parent lwp to child during a posix_spawn() 274 */ 275 struct spawn_exec_data { 276 struct execve_data sed_exec; 277 struct posix_spawn_file_actions 278 *sed_actions; 279 struct posix_spawnattr *sed_attrs; 280 struct proc *sed_parent; 281 kcondvar_t sed_cv_child_ready; 282 kmutex_t sed_mtx_child; 283 int sed_error; 284 volatile uint32_t sed_refcnt; 285 }; 286 287 static struct vm_map *exec_map; 288 static struct pool exec_pool; 289 290 static void * 291 exec_pool_alloc(struct pool *pp, int flags) 292 { 293 294 return (void *)uvm_km_alloc(exec_map, NCARGS, 0, 295 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA); 296 } 297 298 static void 299 exec_pool_free(struct pool *pp, void *addr) 300 { 301 302 uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE); 303 } 304 305 static struct pool_allocator exec_palloc = { 306 .pa_alloc = exec_pool_alloc, 307 .pa_free = exec_pool_free, 308 .pa_pagesz = NCARGS 309 }; 310 311 static void 312 exec_path_free(struct execve_data *data) 313 { 314 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring); 315 pathbuf_destroy(data->ed_pathbuf); 316 if (data->ed_resolvedname) 317 PNBUF_PUT(data->ed_resolvedname); 318 } 319 320 static int 321 exec_resolvename(struct lwp *l, struct exec_package *epp, struct vnode *vp, 322 char **rpath) 323 { 324 int error; 325 char *p; 326 327 KASSERT(rpath != NULL); 328 329 *rpath = PNBUF_GET(); 330 error = vnode_to_path(*rpath, MAXPATHLEN, vp, l, l->l_proc); 331 if (error) { 332 DPRINTF(("%s: can't resolve name for %s, error %d\n", 333 __func__, epp->ep_kname, error)); 334 PNBUF_PUT(*rpath); 335 *rpath = NULL; 336 return error; 337 } 338 epp->ep_resolvedname = *rpath; 339 if ((p = strrchr(*rpath, '/')) != NULL) 340 epp->ep_kname = p + 1; 341 return 0; 342 } 343 344 345 /* 346 * check exec: 347 * given an "executable" described in the exec package's namei info, 348 * see what we can do with it. 349 * 350 * ON ENTRY: 351 * exec package with appropriate namei info 352 * lwp pointer of exec'ing lwp 353 * NO SELF-LOCKED VNODES 354 * 355 * ON EXIT: 356 * error: nothing held, etc. exec header still allocated. 357 * ok: filled exec package, executable's vnode (unlocked). 358 * 359 * EXEC SWITCH ENTRY: 360 * Locked vnode to check, exec package, proc. 361 * 362 * EXEC SWITCH EXIT: 363 * ok: return 0, filled exec package, executable's vnode (unlocked). 364 * error: destructive: 365 * everything deallocated execept exec header. 366 * non-destructive: 367 * error code, executable's vnode (unlocked), 368 * exec header unmodified. 369 */ 370 int 371 /*ARGSUSED*/ 372 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb, 373 char **rpath) 374 { 375 int error, i; 376 struct vnode *vp; 377 size_t resid; 378 379 if (epp->ep_resolvedname) { 380 struct nameidata nd; 381 382 // grab the absolute pathbuf here before namei() trashes it. 383 pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX); 384 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 385 386 /* first get the vnode */ 387 if ((error = namei(&nd)) != 0) 388 return error; 389 390 epp->ep_vp = vp = nd.ni_vp; 391 #ifdef DIAGNOSTIC 392 /* paranoia (take this out once namei stuff stabilizes) */ 393 memset(nd.ni_pnbuf, '~', PATH_MAX); 394 #endif 395 } else { 396 struct file *fp; 397 398 if ((error = fd_getvnode(epp->ep_xfd, &fp)) != 0) 399 return error; 400 epp->ep_vp = vp = fp->f_vnode; 401 vref(vp); 402 fd_putfile(epp->ep_xfd); 403 if ((error = exec_resolvename(l, epp, vp, rpath)) != 0) 404 return error; 405 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 406 } 407 408 /* check access and type */ 409 if (vp->v_type != VREG) { 410 error = EACCES; 411 goto bad1; 412 } 413 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0) 414 goto bad1; 415 416 /* get attributes */ 417 /* XXX VOP_GETATTR is the only thing that needs LK_EXCLUSIVE here */ 418 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0) 419 goto bad1; 420 421 /* Check mount point */ 422 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 423 error = EACCES; 424 goto bad1; 425 } 426 if (vp->v_mount->mnt_flag & MNT_NOSUID) 427 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 428 429 /* try to open it */ 430 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0) 431 goto bad1; 432 433 /* now we have the file, get the exec header */ 434 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 435 UIO_SYSSPACE, IO_NODELOCKED, l->l_cred, &resid, NULL); 436 if (error) 437 goto bad1; 438 439 /* unlock vp, since we need it unlocked from here on out. */ 440 VOP_UNLOCK(vp); 441 442 #if NVERIEXEC > 0 443 error = veriexec_verify(l, vp, 444 epp->ep_resolvedname ? epp->ep_resolvedname : epp->ep_kname, 445 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT, 446 NULL); 447 if (error) 448 goto bad2; 449 #endif /* NVERIEXEC > 0 */ 450 451 #ifdef PAX_SEGVGUARD 452 error = pax_segvguard(l, vp, epp->ep_resolvedname, false); 453 if (error) 454 goto bad2; 455 #endif /* PAX_SEGVGUARD */ 456 457 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 458 459 /* 460 * Set up default address space limits. Can be overridden 461 * by individual exec packages. 462 */ 463 epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS); 464 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 465 466 /* 467 * set up the vmcmds for creation of the process 468 * address space 469 */ 470 error = ENOEXEC; 471 for (i = 0; i < nexecs; i++) { 472 int newerror; 473 474 epp->ep_esch = execsw[i]; 475 newerror = (*execsw[i]->es_makecmds)(l, epp); 476 477 if (!newerror) { 478 /* Seems ok: check that entry point is not too high */ 479 if (epp->ep_entry >= epp->ep_vm_maxaddr) { 480 #ifdef DIAGNOSTIC 481 printf("%s: rejecting %p due to " 482 "too high entry address (>= %p)\n", 483 __func__, (void *)epp->ep_entry, 484 (void *)epp->ep_vm_maxaddr); 485 #endif 486 error = ENOEXEC; 487 break; 488 } 489 /* Seems ok: check that entry point is not too low */ 490 if (epp->ep_entry < epp->ep_vm_minaddr) { 491 #ifdef DIAGNOSTIC 492 printf("%s: rejecting %p due to " 493 "too low entry address (< %p)\n", 494 __func__, (void *)epp->ep_entry, 495 (void *)epp->ep_vm_minaddr); 496 #endif 497 error = ENOEXEC; 498 break; 499 } 500 501 /* check limits */ 502 #ifdef DIAGNOSTIC 503 #define LMSG "%s: rejecting due to %s limit (%ju > %ju)\n" 504 #endif 505 #ifdef MAXTSIZ 506 if (epp->ep_tsize > MAXTSIZ) { 507 #ifdef DIAGNOSTIC 508 printf(LMSG, __func__, "text", 509 (uintmax_t)epp->ep_tsize, 510 (uintmax_t)MAXTSIZ); 511 #endif 512 error = ENOMEM; 513 break; 514 } 515 #endif 516 vsize_t dlimit = 517 (vsize_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur; 518 if (epp->ep_dsize > dlimit) { 519 #ifdef DIAGNOSTIC 520 printf(LMSG, __func__, "data", 521 (uintmax_t)epp->ep_dsize, 522 (uintmax_t)dlimit); 523 #endif 524 error = ENOMEM; 525 break; 526 } 527 return 0; 528 } 529 530 /* 531 * Reset all the fields that may have been modified by the 532 * loader. 533 */ 534 KASSERT(epp->ep_emul_arg == NULL); 535 if (epp->ep_emul_root != NULL) { 536 vrele(epp->ep_emul_root); 537 epp->ep_emul_root = NULL; 538 } 539 if (epp->ep_interp != NULL) { 540 vrele(epp->ep_interp); 541 epp->ep_interp = NULL; 542 } 543 epp->ep_pax_flags = 0; 544 545 /* make sure the first "interesting" error code is saved. */ 546 if (error == ENOEXEC) 547 error = newerror; 548 549 if (epp->ep_flags & EXEC_DESTR) 550 /* Error from "#!" code, tidied up by recursive call */ 551 return error; 552 } 553 554 /* not found, error */ 555 556 /* 557 * free any vmspace-creation commands, 558 * and release their references 559 */ 560 kill_vmcmds(&epp->ep_vmcmds); 561 562 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD) 563 bad2: 564 #endif 565 /* 566 * close and release the vnode, restore the old one, free the 567 * pathname buf, and punt. 568 */ 569 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 570 VOP_CLOSE(vp, FREAD, l->l_cred); 571 vput(vp); 572 return error; 573 574 bad1: 575 /* 576 * free the namei pathname buffer, and put the vnode 577 * (which we don't yet have open). 578 */ 579 vput(vp); /* was still locked */ 580 return error; 581 } 582 583 #ifdef __MACHINE_STACK_GROWS_UP 584 #define STACK_PTHREADSPACE NBPG 585 #else 586 #define STACK_PTHREADSPACE 0 587 #endif 588 589 static int 590 execve_fetch_element(char * const *array, size_t index, char **value) 591 { 592 return copyin(array + index, value, sizeof(*value)); 593 } 594 595 /* 596 * exec system call 597 */ 598 int 599 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval) 600 { 601 /* { 602 syscallarg(const char *) path; 603 syscallarg(char * const *) argp; 604 syscallarg(char * const *) envp; 605 } */ 606 607 return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp), 608 SCARG(uap, envp), execve_fetch_element); 609 } 610 611 int 612 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap, 613 register_t *retval) 614 { 615 /* { 616 syscallarg(int) fd; 617 syscallarg(char * const *) argp; 618 syscallarg(char * const *) envp; 619 } */ 620 621 return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp), 622 SCARG(uap, envp), execve_fetch_element); 623 } 624 625 /* 626 * Load modules to try and execute an image that we do not understand. 627 * If no execsw entries are present, we load those likely to be needed 628 * in order to run native images only. Otherwise, we autoload all 629 * possible modules that could let us run the binary. XXX lame 630 */ 631 static void 632 exec_autoload(void) 633 { 634 #ifdef MODULAR 635 static const char * const native[] = { 636 "exec_elf32", 637 "exec_elf64", 638 "exec_script", 639 NULL 640 }; 641 static const char * const compat[] = { 642 "exec_elf32", 643 "exec_elf64", 644 "exec_script", 645 "exec_aout", 646 "exec_coff", 647 "exec_ecoff", 648 "compat_aoutm68k", 649 "compat_netbsd32", 650 #if 0 651 "compat_linux", 652 "compat_linux32", 653 #endif 654 "compat_sunos", 655 "compat_sunos32", 656 "compat_ultrix", 657 NULL 658 }; 659 char const * const *list; 660 int i; 661 662 list = nexecs == 0 ? native : compat; 663 for (i = 0; list[i] != NULL; i++) { 664 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) { 665 continue; 666 } 667 yield(); 668 } 669 #endif 670 } 671 672 /* 673 * Copy the user or kernel supplied upath to the allocated pathbuffer pbp 674 * making it absolute in the process, by prepending the current working 675 * directory if it is not. If offs is supplied it will contain the offset 676 * where the original supplied copy of upath starts. 677 */ 678 int 679 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg, 680 struct pathbuf **pbp, size_t *offs) 681 { 682 char *path, *bp; 683 size_t len, tlen; 684 int error; 685 struct cwdinfo *cwdi; 686 687 path = PNBUF_GET(); 688 if (seg == UIO_SYSSPACE) { 689 error = copystr(upath, path, MAXPATHLEN, &len); 690 } else { 691 error = copyinstr(upath, path, MAXPATHLEN, &len); 692 } 693 if (error) 694 goto err; 695 696 if (path[0] == '/') { 697 if (offs) 698 *offs = 0; 699 goto out; 700 } 701 702 len++; 703 if (len + 1 >= MAXPATHLEN) { 704 error = ENAMETOOLONG; 705 goto err; 706 } 707 bp = path + MAXPATHLEN - len; 708 memmove(bp, path, len); 709 *(--bp) = '/'; 710 711 cwdi = l->l_proc->p_cwdi; 712 rw_enter(&cwdi->cwdi_lock, RW_READER); 713 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2, 714 GETCWD_CHECK_ACCESS, l); 715 rw_exit(&cwdi->cwdi_lock); 716 717 if (error) 718 goto err; 719 tlen = path + MAXPATHLEN - bp; 720 721 memmove(path, bp, tlen); 722 path[tlen - 1] = '\0'; 723 if (offs) 724 *offs = tlen - len; 725 out: 726 *pbp = pathbuf_assimilate(path); 727 return 0; 728 err: 729 PNBUF_PUT(path); 730 return error; 731 } 732 733 vaddr_t 734 exec_vm_minaddr(vaddr_t va_min) 735 { 736 /* 737 * Increase va_min if we don't want NULL to be mappable by the 738 * process. 739 */ 740 #define VM_MIN_GUARD PAGE_SIZE 741 if (user_va0_disable && (va_min < VM_MIN_GUARD)) 742 return VM_MIN_GUARD; 743 return va_min; 744 } 745 746 static int 747 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd, 748 char * const *args, char * const *envs, 749 execve_fetch_element_t fetch_element, 750 struct execve_data * restrict data) 751 { 752 struct exec_package * const epp = &data->ed_pack; 753 int error; 754 struct proc *p; 755 char *dp; 756 u_int modgen; 757 758 KASSERT(data != NULL); 759 760 p = l->l_proc; 761 modgen = 0; 762 763 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0); 764 765 /* 766 * Check if we have exceeded our number of processes limit. 767 * This is so that we handle the case where a root daemon 768 * forked, ran setuid to become the desired user and is trying 769 * to exec. The obvious place to do the reference counting check 770 * is setuid(), but we don't do the reference counting check there 771 * like other OS's do because then all the programs that use setuid() 772 * must be modified to check the return code of setuid() and exit(). 773 * It is dangerous to make setuid() fail, because it fails open and 774 * the program will continue to run as root. If we make it succeed 775 * and return an error code, again we are not enforcing the limit. 776 * The best place to enforce the limit is here, when the process tries 777 * to execute a new image, because eventually the process will need 778 * to call exec in order to do something useful. 779 */ 780 retry: 781 if (p->p_flag & PK_SUGID) { 782 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT, 783 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 784 &p->p_rlimit[RLIMIT_NPROC], 785 KAUTH_ARG(RLIMIT_NPROC)) != 0 && 786 chgproccnt(kauth_cred_getuid(l->l_cred), 0) > 787 p->p_rlimit[RLIMIT_NPROC].rlim_cur) 788 return EAGAIN; 789 } 790 791 /* 792 * Drain existing references and forbid new ones. The process 793 * should be left alone until we're done here. This is necessary 794 * to avoid race conditions - e.g. in ptrace() - that might allow 795 * a local user to illicitly obtain elevated privileges. 796 */ 797 rw_enter(&p->p_reflock, RW_WRITER); 798 799 if (has_path) { 800 size_t offs; 801 /* 802 * Init the namei data to point the file user's program name. 803 * This is done here rather than in check_exec(), so that it's 804 * possible to override this settings if any of makecmd/probe 805 * functions call check_exec() recursively - for example, 806 * see exec_script_makecmds(). 807 */ 808 if ((error = exec_makepathbuf(l, path, UIO_USERSPACE, 809 &data->ed_pathbuf, &offs)) != 0) 810 goto clrflg; 811 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf); 812 epp->ep_kname = data->ed_pathstring + offs; 813 data->ed_resolvedname = PNBUF_GET(); 814 epp->ep_resolvedname = data->ed_resolvedname; 815 epp->ep_xfd = -1; 816 } else { 817 data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/")); 818 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf); 819 epp->ep_kname = "*fexecve*"; 820 data->ed_resolvedname = NULL; 821 epp->ep_resolvedname = NULL; 822 epp->ep_xfd = fd; 823 } 824 825 826 /* 827 * initialize the fields of the exec package. 828 */ 829 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP); 830 epp->ep_hdrlen = exec_maxhdrsz; 831 epp->ep_hdrvalid = 0; 832 epp->ep_emul_arg = NULL; 833 epp->ep_emul_arg_free = NULL; 834 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds)); 835 epp->ep_vap = &data->ed_attr; 836 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0; 837 MD_TOPDOWN_INIT(epp); 838 epp->ep_emul_root = NULL; 839 epp->ep_interp = NULL; 840 epp->ep_esch = NULL; 841 epp->ep_pax_flags = 0; 842 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch)); 843 844 rw_enter(&exec_lock, RW_READER); 845 846 /* see if we can run it. */ 847 if ((error = check_exec(l, epp, data->ed_pathbuf, 848 &data->ed_resolvedname)) != 0) { 849 if (error != ENOENT && error != EACCES && error != ENOEXEC) { 850 DPRINTF(("%s: check exec failed for %s, error %d\n", 851 __func__, epp->ep_kname, error)); 852 } 853 goto freehdr; 854 } 855 856 /* allocate an argument buffer */ 857 data->ed_argp = pool_get(&exec_pool, PR_WAITOK); 858 KASSERT(data->ed_argp != NULL); 859 dp = data->ed_argp; 860 861 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) { 862 goto bad; 863 } 864 865 /* 866 * Calculate the new stack size. 867 */ 868 869 #ifdef __MACHINE_STACK_GROWS_UP 870 /* 871 * copyargs() fills argc/argv/envp from the lower address even on 872 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP 873 * so that _rtld() use it. 874 */ 875 #define RTLD_GAP 32 876 #else 877 #define RTLD_GAP 0 878 #endif 879 880 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp; 881 882 data->ed_argslen = calcargs(data, argenvstrlen); 883 884 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP); 885 886 if (len > epp->ep_ssize) { 887 /* in effect, compare to initial limit */ 888 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len)); 889 error = ENOMEM; 890 goto bad; 891 } 892 /* adjust "active stack depth" for process VSZ */ 893 epp->ep_ssize = len; 894 895 return 0; 896 897 bad: 898 /* free the vmspace-creation commands, and release their references */ 899 kill_vmcmds(&epp->ep_vmcmds); 900 /* kill any opened file descriptor, if necessary */ 901 if (epp->ep_flags & EXEC_HASFD) { 902 epp->ep_flags &= ~EXEC_HASFD; 903 fd_close(epp->ep_fd); 904 } 905 /* close and put the exec'd file */ 906 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 907 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred); 908 vput(epp->ep_vp); 909 pool_put(&exec_pool, data->ed_argp); 910 911 freehdr: 912 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 913 if (epp->ep_emul_root != NULL) 914 vrele(epp->ep_emul_root); 915 if (epp->ep_interp != NULL) 916 vrele(epp->ep_interp); 917 918 rw_exit(&exec_lock); 919 920 exec_path_free(data); 921 922 clrflg: 923 rw_exit(&p->p_reflock); 924 925 if (modgen != module_gen && error == ENOEXEC) { 926 modgen = module_gen; 927 exec_autoload(); 928 goto retry; 929 } 930 931 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 932 return error; 933 } 934 935 static int 936 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data) 937 { 938 struct exec_package * const epp = &data->ed_pack; 939 struct proc *p = l->l_proc; 940 struct exec_vmcmd *base_vcp; 941 int error = 0; 942 size_t i; 943 944 /* record proc's vnode, for use by procfs and others */ 945 if (p->p_textvp) 946 vrele(p->p_textvp); 947 vref(epp->ep_vp); 948 p->p_textvp = epp->ep_vp; 949 950 /* create the new process's VM space by running the vmcmds */ 951 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__); 952 953 #ifdef TRACE_EXEC 954 DUMPVMCMDS(epp, 0, 0); 955 #endif 956 957 base_vcp = NULL; 958 959 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) { 960 struct exec_vmcmd *vcp; 961 962 vcp = &epp->ep_vmcmds.evs_cmds[i]; 963 if (vcp->ev_flags & VMCMD_RELATIVE) { 964 KASSERTMSG(base_vcp != NULL, 965 "%s: relative vmcmd with no base", __func__); 966 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0, 967 "%s: illegal base & relative vmcmd", __func__); 968 vcp->ev_addr += base_vcp->ev_addr; 969 } 970 error = (*vcp->ev_proc)(l, vcp); 971 if (error) 972 DUMPVMCMDS(epp, i, error); 973 if (vcp->ev_flags & VMCMD_BASE) 974 base_vcp = vcp; 975 } 976 977 /* free the vmspace-creation commands, and release their references */ 978 kill_vmcmds(&epp->ep_vmcmds); 979 980 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 981 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred); 982 vput(epp->ep_vp); 983 984 /* if an error happened, deallocate and punt */ 985 if (error != 0) { 986 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error)); 987 } 988 return error; 989 } 990 991 static void 992 execve_free_data(struct execve_data *data) 993 { 994 struct exec_package * const epp = &data->ed_pack; 995 996 /* free the vmspace-creation commands, and release their references */ 997 kill_vmcmds(&epp->ep_vmcmds); 998 /* kill any opened file descriptor, if necessary */ 999 if (epp->ep_flags & EXEC_HASFD) { 1000 epp->ep_flags &= ~EXEC_HASFD; 1001 fd_close(epp->ep_fd); 1002 } 1003 1004 /* close and put the exec'd file */ 1005 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 1006 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred); 1007 vput(epp->ep_vp); 1008 pool_put(&exec_pool, data->ed_argp); 1009 1010 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1011 if (epp->ep_emul_root != NULL) 1012 vrele(epp->ep_emul_root); 1013 if (epp->ep_interp != NULL) 1014 vrele(epp->ep_interp); 1015 1016 exec_path_free(data); 1017 } 1018 1019 static void 1020 pathexec(struct proc *p, const char *resolvedname) 1021 { 1022 /* set command name & other accounting info */ 1023 const char *cmdname; 1024 1025 if (resolvedname == NULL) { 1026 cmdname = "*fexecve*"; 1027 resolvedname = "/"; 1028 } else { 1029 cmdname = strrchr(resolvedname, '/') + 1; 1030 } 1031 KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'", 1032 resolvedname); 1033 1034 strlcpy(p->p_comm, cmdname, sizeof(p->p_comm)); 1035 1036 kmem_strfree(p->p_path); 1037 p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP); 1038 } 1039 1040 /* XXX elsewhere */ 1041 static int 1042 credexec(struct lwp *l, struct vattr *attr) 1043 { 1044 struct proc *p = l->l_proc; 1045 int error; 1046 1047 /* 1048 * Deal with set[ug]id. MNT_NOSUID has already been used to disable 1049 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked 1050 * out additional references on the process for the moment. 1051 */ 1052 if ((p->p_slflag & PSL_TRACED) == 0 && 1053 1054 (((attr->va_mode & S_ISUID) != 0 && 1055 kauth_cred_geteuid(l->l_cred) != attr->va_uid) || 1056 1057 ((attr->va_mode & S_ISGID) != 0 && 1058 kauth_cred_getegid(l->l_cred) != attr->va_gid))) { 1059 /* 1060 * Mark the process as SUGID before we do 1061 * anything that might block. 1062 */ 1063 proc_crmod_enter(); 1064 proc_crmod_leave(NULL, NULL, true); 1065 1066 /* Make sure file descriptors 0..2 are in use. */ 1067 if ((error = fd_checkstd()) != 0) { 1068 DPRINTF(("%s: fdcheckstd failed %d\n", 1069 __func__, error)); 1070 return error; 1071 } 1072 1073 /* 1074 * Copy the credential so other references don't see our 1075 * changes. 1076 */ 1077 l->l_cred = kauth_cred_copy(l->l_cred); 1078 #ifdef KTRACE 1079 /* 1080 * If the persistent trace flag isn't set, turn off. 1081 */ 1082 if (p->p_tracep) { 1083 mutex_enter(&ktrace_lock); 1084 if (!(p->p_traceflag & KTRFAC_PERSISTENT)) 1085 ktrderef(p); 1086 mutex_exit(&ktrace_lock); 1087 } 1088 #endif 1089 if (attr->va_mode & S_ISUID) 1090 kauth_cred_seteuid(l->l_cred, attr->va_uid); 1091 if (attr->va_mode & S_ISGID) 1092 kauth_cred_setegid(l->l_cred, attr->va_gid); 1093 } else { 1094 if (kauth_cred_geteuid(l->l_cred) == 1095 kauth_cred_getuid(l->l_cred) && 1096 kauth_cred_getegid(l->l_cred) == 1097 kauth_cred_getgid(l->l_cred)) 1098 p->p_flag &= ~PK_SUGID; 1099 } 1100 1101 /* 1102 * Copy the credential so other references don't see our changes. 1103 * Test to see if this is necessary first, since in the common case 1104 * we won't need a private reference. 1105 */ 1106 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) || 1107 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) { 1108 l->l_cred = kauth_cred_copy(l->l_cred); 1109 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred)); 1110 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred)); 1111 } 1112 1113 /* Update the master credentials. */ 1114 if (l->l_cred != p->p_cred) { 1115 kauth_cred_t ocred; 1116 1117 kauth_cred_hold(l->l_cred); 1118 mutex_enter(p->p_lock); 1119 ocred = p->p_cred; 1120 p->p_cred = l->l_cred; 1121 mutex_exit(p->p_lock); 1122 kauth_cred_free(ocred); 1123 } 1124 1125 return 0; 1126 } 1127 1128 static void 1129 emulexec(struct lwp *l, struct exec_package *epp) 1130 { 1131 struct proc *p = l->l_proc; 1132 1133 /* The emulation root will usually have been found when we looked 1134 * for the elf interpreter (or similar), if not look now. */ 1135 if (epp->ep_esch->es_emul->e_path != NULL && 1136 epp->ep_emul_root == NULL) 1137 emul_find_root(l, epp); 1138 1139 /* Any old emulation root got removed by fdcloseexec */ 1140 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER); 1141 p->p_cwdi->cwdi_edir = epp->ep_emul_root; 1142 rw_exit(&p->p_cwdi->cwdi_lock); 1143 epp->ep_emul_root = NULL; 1144 if (epp->ep_interp != NULL) 1145 vrele(epp->ep_interp); 1146 1147 /* 1148 * Call emulation specific exec hook. This can setup per-process 1149 * p->p_emuldata or do any other per-process stuff an emulation needs. 1150 * 1151 * If we are executing process of different emulation than the 1152 * original forked process, call e_proc_exit() of the old emulation 1153 * first, then e_proc_exec() of new emulation. If the emulation is 1154 * same, the exec hook code should deallocate any old emulation 1155 * resources held previously by this process. 1156 */ 1157 if (p->p_emul && p->p_emul->e_proc_exit 1158 && p->p_emul != epp->ep_esch->es_emul) 1159 (*p->p_emul->e_proc_exit)(p); 1160 1161 /* 1162 * Call exec hook. Emulation code may NOT store reference to anything 1163 * from &pack. 1164 */ 1165 if (epp->ep_esch->es_emul->e_proc_exec) 1166 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp); 1167 1168 /* update p_emul, the old value is no longer needed */ 1169 p->p_emul = epp->ep_esch->es_emul; 1170 1171 /* ...and the same for p_execsw */ 1172 p->p_execsw = epp->ep_esch; 1173 1174 #ifdef __HAVE_SYSCALL_INTERN 1175 (*p->p_emul->e_syscall_intern)(p); 1176 #endif 1177 ktremul(); 1178 } 1179 1180 static int 1181 execve_runproc(struct lwp *l, struct execve_data * restrict data, 1182 bool no_local_exec_lock, bool is_spawn) 1183 { 1184 struct exec_package * const epp = &data->ed_pack; 1185 int error = 0; 1186 struct proc *p; 1187 struct vmspace *vm; 1188 1189 /* 1190 * In case of a posix_spawn operation, the child doing the exec 1191 * might not hold the reader lock on exec_lock, but the parent 1192 * will do this instead. 1193 */ 1194 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock)); 1195 KASSERT(!no_local_exec_lock || is_spawn); 1196 KASSERT(data != NULL); 1197 1198 p = l->l_proc; 1199 1200 /* Get rid of other LWPs. */ 1201 if (p->p_nlwps > 1) { 1202 mutex_enter(p->p_lock); 1203 exit_lwps(l); 1204 mutex_exit(p->p_lock); 1205 } 1206 KDASSERT(p->p_nlwps == 1); 1207 1208 /* 1209 * All of the other LWPs got rid of their robust futexes 1210 * when they exited above, but we might still have some 1211 * to dispose of. Do that now. 1212 */ 1213 if (__predict_false(l->l_robust_head != 0)) { 1214 futex_release_all_lwp(l); 1215 /* 1216 * Since this LWP will live on with a different 1217 * program image, we need to clear the robust 1218 * futex list pointer here. 1219 */ 1220 l->l_robust_head = 0; 1221 } 1222 1223 /* Destroy any lwpctl info. */ 1224 if (p->p_lwpctl != NULL) 1225 lwp_ctl_exit(); 1226 1227 /* Remove POSIX timers */ 1228 ptimers_free(p, TIMERS_POSIX); 1229 1230 /* Set the PaX flags. */ 1231 pax_set_flags(epp, p); 1232 1233 /* 1234 * Do whatever is necessary to prepare the address space 1235 * for remapping. Note that this might replace the current 1236 * vmspace with another! 1237 * 1238 * vfork(): do not touch any user space data in the new child 1239 * until we have awoken the parent below, or it will defeat 1240 * lazy pmap switching (on x86). 1241 */ 1242 if (is_spawn) 1243 uvmspace_spawn(l, epp->ep_vm_minaddr, 1244 epp->ep_vm_maxaddr, 1245 epp->ep_flags & EXEC_TOPDOWN_VM); 1246 else 1247 uvmspace_exec(l, epp->ep_vm_minaddr, 1248 epp->ep_vm_maxaddr, 1249 epp->ep_flags & EXEC_TOPDOWN_VM); 1250 vm = p->p_vmspace; 1251 1252 vm->vm_taddr = (void *)epp->ep_taddr; 1253 vm->vm_tsize = btoc(epp->ep_tsize); 1254 vm->vm_daddr = (void*)epp->ep_daddr; 1255 vm->vm_dsize = btoc(epp->ep_dsize); 1256 vm->vm_ssize = btoc(epp->ep_ssize); 1257 vm->vm_issize = 0; 1258 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr; 1259 vm->vm_minsaddr = (void *)epp->ep_minsaddr; 1260 1261 pax_aslr_init_vm(l, vm, epp); 1262 1263 cwdexec(p); 1264 fd_closeexec(); /* handle close on exec */ 1265 1266 if (__predict_false(ktrace_on)) 1267 fd_ktrexecfd(); 1268 1269 execsigs(p); /* reset caught signals */ 1270 1271 mutex_enter(p->p_lock); 1272 l->l_ctxlink = NULL; /* reset ucontext link */ 1273 p->p_acflag &= ~AFORK; 1274 p->p_flag |= PK_EXEC; 1275 mutex_exit(p->p_lock); 1276 1277 error = credexec(l, &data->ed_attr); 1278 if (error) 1279 goto exec_abort; 1280 1281 #if defined(__HAVE_RAS) 1282 /* 1283 * Remove all RASs from the address space. 1284 */ 1285 ras_purgeall(); 1286 #endif 1287 1288 /* 1289 * Stop profiling. 1290 */ 1291 if ((p->p_stflag & PST_PROFIL) != 0) { 1292 mutex_spin_enter(&p->p_stmutex); 1293 stopprofclock(p); 1294 mutex_spin_exit(&p->p_stmutex); 1295 } 1296 1297 /* 1298 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have 1299 * exited and exec()/exit() are the only places it will be cleared. 1300 * 1301 * Once the parent has been awoken, curlwp may teleport to a new CPU 1302 * in sched_vforkexec(), and it's then OK to start messing with user 1303 * data. See comment above. 1304 */ 1305 if ((p->p_lflag & PL_PPWAIT) != 0) { 1306 bool samecpu; 1307 lwp_t *lp; 1308 1309 mutex_enter(&proc_lock); 1310 lp = p->p_vforklwp; 1311 p->p_vforklwp = NULL; 1312 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 1313 cv_broadcast(&lp->l_waitcv); 1314 1315 /* Clear flags after cv_broadcast() (scheduler needs them). */ 1316 p->p_lflag &= ~PL_PPWAIT; 1317 lp->l_vforkwaiting = false; 1318 1319 /* If parent is still on same CPU, teleport curlwp elsewhere. */ 1320 samecpu = (lp->l_cpu == curlwp->l_cpu); 1321 mutex_exit(&proc_lock); 1322 1323 /* Give the parent its CPU back - find a new home. */ 1324 KASSERT(!is_spawn); 1325 sched_vforkexec(l, samecpu); 1326 } 1327 1328 /* Now map address space. */ 1329 error = execve_dovmcmds(l, data); 1330 if (error != 0) 1331 goto exec_abort; 1332 1333 pathexec(p, epp->ep_resolvedname); 1334 1335 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize); 1336 1337 error = copyoutargs(data, l, newstack); 1338 if (error != 0) 1339 goto exec_abort; 1340 1341 doexechooks(p); 1342 1343 /* 1344 * Set initial SP at the top of the stack. 1345 * 1346 * Note that on machines where stack grows up (e.g. hppa), SP points to 1347 * the end of arg/env strings. Userland guesses the address of argc 1348 * via ps_strings::ps_argvstr. 1349 */ 1350 1351 /* Setup new registers and do misc. setup. */ 1352 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack); 1353 if (epp->ep_esch->es_setregs) 1354 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack); 1355 1356 /* Provide a consistent LWP private setting */ 1357 (void)lwp_setprivate(l, NULL); 1358 1359 /* Discard all PCU state; need to start fresh */ 1360 pcu_discard_all(l); 1361 1362 /* map the process's signal trampoline code */ 1363 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) { 1364 DPRINTF(("%s: map sigcode failed %d\n", __func__, error)); 1365 goto exec_abort; 1366 } 1367 1368 pool_put(&exec_pool, data->ed_argp); 1369 1370 /* notify others that we exec'd */ 1371 KNOTE(&p->p_klist, NOTE_EXEC); 1372 1373 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1374 1375 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0); 1376 1377 emulexec(l, epp); 1378 1379 /* Allow new references from the debugger/procfs. */ 1380 rw_exit(&p->p_reflock); 1381 if (!no_local_exec_lock) 1382 rw_exit(&exec_lock); 1383 1384 mutex_enter(&proc_lock); 1385 1386 /* posix_spawn(3) reports a single event with implied exec(3) */ 1387 if ((p->p_slflag & PSL_TRACED) && !is_spawn) { 1388 mutex_enter(p->p_lock); 1389 eventswitch(TRAP_EXEC, 0, 0); 1390 mutex_enter(&proc_lock); 1391 } 1392 1393 if (p->p_sflag & PS_STOPEXEC) { 1394 ksiginfoq_t kq; 1395 1396 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 1397 p->p_pptr->p_nstopchild++; 1398 p->p_waited = 0; 1399 mutex_enter(p->p_lock); 1400 ksiginfo_queue_init(&kq); 1401 sigclearall(p, &contsigmask, &kq); 1402 lwp_lock(l); 1403 l->l_stat = LSSTOP; 1404 p->p_stat = SSTOP; 1405 p->p_nrlwps--; 1406 lwp_unlock(l); 1407 mutex_exit(p->p_lock); 1408 mutex_exit(&proc_lock); 1409 lwp_lock(l); 1410 spc_lock(l->l_cpu); 1411 mi_switch(l); 1412 ksiginfo_queue_drain(&kq); 1413 } else { 1414 mutex_exit(&proc_lock); 1415 } 1416 1417 exec_path_free(data); 1418 #ifdef TRACE_EXEC 1419 DPRINTF(("%s finished\n", __func__)); 1420 #endif 1421 return EJUSTRETURN; 1422 1423 exec_abort: 1424 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 1425 rw_exit(&p->p_reflock); 1426 if (!no_local_exec_lock) 1427 rw_exit(&exec_lock); 1428 1429 exec_path_free(data); 1430 1431 /* 1432 * the old process doesn't exist anymore. exit gracefully. 1433 * get rid of the (new) address space we have created, if any, get rid 1434 * of our namei data and vnode, and exit noting failure 1435 */ 1436 if (vm != NULL) { 1437 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 1438 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 1439 } 1440 1441 exec_free_emul_arg(epp); 1442 pool_put(&exec_pool, data->ed_argp); 1443 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1444 if (epp->ep_emul_root != NULL) 1445 vrele(epp->ep_emul_root); 1446 if (epp->ep_interp != NULL) 1447 vrele(epp->ep_interp); 1448 1449 /* Acquire the sched-state mutex (exit1() will release it). */ 1450 if (!is_spawn) { 1451 mutex_enter(p->p_lock); 1452 exit1(l, error, SIGABRT); 1453 } 1454 1455 return error; 1456 } 1457 1458 int 1459 execve1(struct lwp *l, bool has_path, const char *path, int fd, 1460 char * const *args, char * const *envs, 1461 execve_fetch_element_t fetch_element) 1462 { 1463 struct execve_data data; 1464 int error; 1465 1466 error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element, 1467 &data); 1468 if (error) 1469 return error; 1470 error = execve_runproc(l, &data, false, false); 1471 return error; 1472 } 1473 1474 static size_t 1475 fromptrsz(const struct exec_package *epp) 1476 { 1477 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *); 1478 } 1479 1480 static size_t 1481 ptrsz(const struct exec_package *epp) 1482 { 1483 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *); 1484 } 1485 1486 static size_t 1487 calcargs(struct execve_data * restrict data, const size_t argenvstrlen) 1488 { 1489 struct exec_package * const epp = &data->ed_pack; 1490 1491 const size_t nargenvptrs = 1492 1 + /* long argc */ 1493 data->ed_argc + /* char *argv[] */ 1494 1 + /* \0 */ 1495 data->ed_envc + /* char *env[] */ 1496 1; /* \0 */ 1497 1498 return (nargenvptrs * ptrsz(epp)) /* pointers */ 1499 + argenvstrlen /* strings */ 1500 + epp->ep_esch->es_arglen; /* auxinfo */ 1501 } 1502 1503 static size_t 1504 calcstack(struct execve_data * restrict data, const size_t gaplen) 1505 { 1506 struct exec_package * const epp = &data->ed_pack; 1507 1508 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode - 1509 epp->ep_esch->es_emul->e_sigcode; 1510 1511 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ? 1512 sizeof(struct ps_strings32) : sizeof(struct ps_strings); 1513 1514 const size_t sigcode_psstr_sz = 1515 data->ed_szsigcode + /* sigcode */ 1516 data->ed_ps_strings_sz + /* ps_strings */ 1517 STACK_PTHREADSPACE; /* pthread space */ 1518 1519 const size_t stacklen = 1520 data->ed_argslen + 1521 gaplen + 1522 sigcode_psstr_sz; 1523 1524 /* make the stack "safely" aligned */ 1525 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES); 1526 } 1527 1528 static int 1529 copyoutargs(struct execve_data * restrict data, struct lwp *l, 1530 char * const newstack) 1531 { 1532 struct exec_package * const epp = &data->ed_pack; 1533 struct proc *p = l->l_proc; 1534 int error; 1535 1536 memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo)); 1537 1538 /* remember information about the process */ 1539 data->ed_arginfo.ps_nargvstr = data->ed_argc; 1540 data->ed_arginfo.ps_nenvstr = data->ed_envc; 1541 1542 /* 1543 * Allocate the stack address passed to the newly execve()'ed process. 1544 * 1545 * The new stack address will be set to the SP (stack pointer) register 1546 * in setregs(). 1547 */ 1548 1549 char *newargs = STACK_ALLOC( 1550 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen); 1551 1552 error = (*epp->ep_esch->es_copyargs)(l, epp, 1553 &data->ed_arginfo, &newargs, data->ed_argp); 1554 1555 if (error) { 1556 DPRINTF(("%s: copyargs failed %d\n", __func__, error)); 1557 return error; 1558 } 1559 1560 error = copyoutpsstrs(data, p); 1561 if (error != 0) 1562 return error; 1563 1564 return 0; 1565 } 1566 1567 static int 1568 copyoutpsstrs(struct execve_data * restrict data, struct proc *p) 1569 { 1570 struct exec_package * const epp = &data->ed_pack; 1571 struct ps_strings32 arginfo32; 1572 void *aip; 1573 int error; 1574 1575 /* fill process ps_strings info */ 1576 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr, 1577 STACK_PTHREADSPACE), data->ed_ps_strings_sz); 1578 1579 if (epp->ep_flags & EXEC_32) { 1580 aip = &arginfo32; 1581 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr; 1582 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr; 1583 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr; 1584 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr; 1585 } else 1586 aip = &data->ed_arginfo; 1587 1588 /* copy out the process's ps_strings structure */ 1589 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)) 1590 != 0) { 1591 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n", 1592 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)); 1593 return error; 1594 } 1595 1596 return 0; 1597 } 1598 1599 static int 1600 copyinargs(struct execve_data * restrict data, char * const *args, 1601 char * const *envs, execve_fetch_element_t fetch_element, char **dpp) 1602 { 1603 struct exec_package * const epp = &data->ed_pack; 1604 char *dp; 1605 size_t i; 1606 int error; 1607 1608 dp = *dpp; 1609 1610 data->ed_argc = 0; 1611 1612 /* copy the fake args list, if there's one, freeing it as we go */ 1613 if (epp->ep_flags & EXEC_HASARGL) { 1614 struct exec_fakearg *fa = epp->ep_fa; 1615 1616 while (fa->fa_arg != NULL) { 1617 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1618 size_t len; 1619 1620 len = strlcpy(dp, fa->fa_arg, maxlen); 1621 /* Count NUL into len. */ 1622 if (len < maxlen) 1623 len++; 1624 else { 1625 while (fa->fa_arg != NULL) { 1626 kmem_free(fa->fa_arg, fa->fa_len); 1627 fa++; 1628 } 1629 kmem_free(epp->ep_fa, epp->ep_fa_len); 1630 epp->ep_flags &= ~EXEC_HASARGL; 1631 return E2BIG; 1632 } 1633 ktrexecarg(fa->fa_arg, len - 1); 1634 dp += len; 1635 1636 kmem_free(fa->fa_arg, fa->fa_len); 1637 fa++; 1638 data->ed_argc++; 1639 } 1640 kmem_free(epp->ep_fa, epp->ep_fa_len); 1641 epp->ep_flags &= ~EXEC_HASARGL; 1642 } 1643 1644 /* 1645 * Read and count argument strings from user. 1646 */ 1647 1648 if (args == NULL) { 1649 DPRINTF(("%s: null args\n", __func__)); 1650 return EINVAL; 1651 } 1652 if (epp->ep_flags & EXEC_SKIPARG) 1653 args = (const void *)((const char *)args + fromptrsz(epp)); 1654 i = 0; 1655 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg); 1656 if (error != 0) { 1657 DPRINTF(("%s: copyin arg %d\n", __func__, error)); 1658 return error; 1659 } 1660 data->ed_argc += i; 1661 1662 /* 1663 * Read and count environment strings from user. 1664 */ 1665 1666 data->ed_envc = 0; 1667 /* environment need not be there */ 1668 if (envs == NULL) 1669 goto done; 1670 i = 0; 1671 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv); 1672 if (error != 0) { 1673 DPRINTF(("%s: copyin env %d\n", __func__, error)); 1674 return error; 1675 } 1676 data->ed_envc += i; 1677 1678 done: 1679 *dpp = dp; 1680 1681 return 0; 1682 } 1683 1684 static int 1685 copyinargstrs(struct execve_data * restrict data, char * const *strs, 1686 execve_fetch_element_t fetch_element, char **dpp, size_t *ip, 1687 void (*ktr)(const void *, size_t)) 1688 { 1689 char *dp, *sp; 1690 size_t i; 1691 int error; 1692 1693 dp = *dpp; 1694 1695 i = 0; 1696 while (1) { 1697 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1698 size_t len; 1699 1700 if ((error = (*fetch_element)(strs, i, &sp)) != 0) { 1701 return error; 1702 } 1703 if (!sp) 1704 break; 1705 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) { 1706 if (error == ENAMETOOLONG) 1707 error = E2BIG; 1708 return error; 1709 } 1710 if (__predict_false(ktrace_on)) 1711 (*ktr)(dp, len - 1); 1712 dp += len; 1713 i++; 1714 } 1715 1716 *dpp = dp; 1717 *ip = i; 1718 1719 return 0; 1720 } 1721 1722 /* 1723 * Copy argv and env strings from kernel buffer (argp) to the new stack. 1724 * Those strings are located just after auxinfo. 1725 */ 1726 int 1727 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo, 1728 char **stackp, void *argp) 1729 { 1730 char **cpp, *dp, *sp; 1731 size_t len; 1732 void *nullp; 1733 long argc, envc; 1734 int error; 1735 1736 cpp = (char **)*stackp; 1737 nullp = NULL; 1738 argc = arginfo->ps_nargvstr; 1739 envc = arginfo->ps_nenvstr; 1740 1741 /* argc on stack is long */ 1742 CTASSERT(sizeof(*cpp) == sizeof(argc)); 1743 1744 dp = (char *)(cpp + 1745 1 + /* long argc */ 1746 argc + /* char *argv[] */ 1747 1 + /* \0 */ 1748 envc + /* char *env[] */ 1749 1) + /* \0 */ 1750 pack->ep_esch->es_arglen; /* auxinfo */ 1751 sp = argp; 1752 1753 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) { 1754 COPYPRINTF("", cpp - 1, sizeof(argc)); 1755 return error; 1756 } 1757 1758 /* XXX don't copy them out, remap them! */ 1759 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 1760 1761 for (; --argc >= 0; sp += len, dp += len) { 1762 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1763 COPYPRINTF("", cpp - 1, sizeof(dp)); 1764 return error; 1765 } 1766 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1767 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1768 return error; 1769 } 1770 } 1771 1772 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1773 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1774 return error; 1775 } 1776 1777 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 1778 1779 for (; --envc >= 0; sp += len, dp += len) { 1780 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1781 COPYPRINTF("", cpp - 1, sizeof(dp)); 1782 return error; 1783 } 1784 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1785 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1786 return error; 1787 } 1788 1789 } 1790 1791 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1792 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1793 return error; 1794 } 1795 1796 *stackp = (char *)cpp; 1797 return 0; 1798 } 1799 1800 1801 /* 1802 * Add execsw[] entries. 1803 */ 1804 int 1805 exec_add(struct execsw *esp, int count) 1806 { 1807 struct exec_entry *it; 1808 int i; 1809 1810 if (count == 0) { 1811 return 0; 1812 } 1813 1814 /* Check for duplicates. */ 1815 rw_enter(&exec_lock, RW_WRITER); 1816 for (i = 0; i < count; i++) { 1817 LIST_FOREACH(it, &ex_head, ex_list) { 1818 /* assume unique (makecmds, probe_func, emulation) */ 1819 if (it->ex_sw->es_makecmds == esp[i].es_makecmds && 1820 it->ex_sw->u.elf_probe_func == 1821 esp[i].u.elf_probe_func && 1822 it->ex_sw->es_emul == esp[i].es_emul) { 1823 rw_exit(&exec_lock); 1824 return EEXIST; 1825 } 1826 } 1827 } 1828 1829 /* Allocate new entries. */ 1830 for (i = 0; i < count; i++) { 1831 it = kmem_alloc(sizeof(*it), KM_SLEEP); 1832 it->ex_sw = &esp[i]; 1833 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1834 } 1835 1836 /* update execsw[] */ 1837 exec_init(0); 1838 rw_exit(&exec_lock); 1839 return 0; 1840 } 1841 1842 /* 1843 * Remove execsw[] entry. 1844 */ 1845 int 1846 exec_remove(struct execsw *esp, int count) 1847 { 1848 struct exec_entry *it, *next; 1849 int i; 1850 const struct proclist_desc *pd; 1851 proc_t *p; 1852 1853 if (count == 0) { 1854 return 0; 1855 } 1856 1857 /* Abort if any are busy. */ 1858 rw_enter(&exec_lock, RW_WRITER); 1859 for (i = 0; i < count; i++) { 1860 mutex_enter(&proc_lock); 1861 for (pd = proclists; pd->pd_list != NULL; pd++) { 1862 PROCLIST_FOREACH(p, pd->pd_list) { 1863 if (p->p_execsw == &esp[i]) { 1864 mutex_exit(&proc_lock); 1865 rw_exit(&exec_lock); 1866 return EBUSY; 1867 } 1868 } 1869 } 1870 mutex_exit(&proc_lock); 1871 } 1872 1873 /* None are busy, so remove them all. */ 1874 for (i = 0; i < count; i++) { 1875 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) { 1876 next = LIST_NEXT(it, ex_list); 1877 if (it->ex_sw == &esp[i]) { 1878 LIST_REMOVE(it, ex_list); 1879 kmem_free(it, sizeof(*it)); 1880 break; 1881 } 1882 } 1883 } 1884 1885 /* update execsw[] */ 1886 exec_init(0); 1887 rw_exit(&exec_lock); 1888 return 0; 1889 } 1890 1891 /* 1892 * Initialize exec structures. If init_boot is true, also does necessary 1893 * one-time initialization (it's called from main() that way). 1894 * Once system is multiuser, this should be called with exec_lock held, 1895 * i.e. via exec_{add|remove}(). 1896 */ 1897 int 1898 exec_init(int init_boot) 1899 { 1900 const struct execsw **sw; 1901 struct exec_entry *ex; 1902 SLIST_HEAD(,exec_entry) first; 1903 SLIST_HEAD(,exec_entry) any; 1904 SLIST_HEAD(,exec_entry) last; 1905 int i, sz; 1906 1907 if (init_boot) { 1908 /* do one-time initializations */ 1909 vaddr_t vmin = 0, vmax; 1910 1911 rw_init(&exec_lock); 1912 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE); 1913 exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax, 1914 maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL); 1915 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH, 1916 "execargs", &exec_palloc, IPL_NONE); 1917 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0); 1918 } else { 1919 KASSERT(rw_write_held(&exec_lock)); 1920 } 1921 1922 /* Sort each entry onto the appropriate queue. */ 1923 SLIST_INIT(&first); 1924 SLIST_INIT(&any); 1925 SLIST_INIT(&last); 1926 sz = 0; 1927 LIST_FOREACH(ex, &ex_head, ex_list) { 1928 switch(ex->ex_sw->es_prio) { 1929 case EXECSW_PRIO_FIRST: 1930 SLIST_INSERT_HEAD(&first, ex, ex_slist); 1931 break; 1932 case EXECSW_PRIO_ANY: 1933 SLIST_INSERT_HEAD(&any, ex, ex_slist); 1934 break; 1935 case EXECSW_PRIO_LAST: 1936 SLIST_INSERT_HEAD(&last, ex, ex_slist); 1937 break; 1938 default: 1939 panic("%s", __func__); 1940 break; 1941 } 1942 sz++; 1943 } 1944 1945 /* 1946 * Create new execsw[]. Ensure we do not try a zero-sized 1947 * allocation. 1948 */ 1949 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP); 1950 i = 0; 1951 SLIST_FOREACH(ex, &first, ex_slist) { 1952 sw[i++] = ex->ex_sw; 1953 } 1954 SLIST_FOREACH(ex, &any, ex_slist) { 1955 sw[i++] = ex->ex_sw; 1956 } 1957 SLIST_FOREACH(ex, &last, ex_slist) { 1958 sw[i++] = ex->ex_sw; 1959 } 1960 1961 /* Replace old execsw[] and free used memory. */ 1962 if (execsw != NULL) { 1963 kmem_free(__UNCONST(execsw), 1964 nexecs * sizeof(struct execsw *) + 1); 1965 } 1966 execsw = sw; 1967 nexecs = sz; 1968 1969 /* Figure out the maximum size of an exec header. */ 1970 exec_maxhdrsz = sizeof(int); 1971 for (i = 0; i < nexecs; i++) { 1972 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1973 exec_maxhdrsz = execsw[i]->es_hdrsz; 1974 } 1975 1976 return 0; 1977 } 1978 1979 static int 1980 exec_sigcode_map(struct proc *p, const struct emul *e) 1981 { 1982 vaddr_t va; 1983 vsize_t sz; 1984 int error; 1985 struct uvm_object *uobj; 1986 1987 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1988 1989 if (e->e_sigobject == NULL || sz == 0) { 1990 return 0; 1991 } 1992 1993 /* 1994 * If we don't have a sigobject for this emulation, create one. 1995 * 1996 * sigobject is an anonymous memory object (just like SYSV shared 1997 * memory) that we keep a permanent reference to and that we map 1998 * in all processes that need this sigcode. The creation is simple, 1999 * we create an object, add a permanent reference to it, map it in 2000 * kernel space, copy out the sigcode to it and unmap it. 2001 * We map it with PROT_READ|PROT_EXEC into the process just 2002 * the way sys_mmap() would map it. 2003 */ 2004 2005 uobj = *e->e_sigobject; 2006 if (uobj == NULL) { 2007 mutex_enter(&sigobject_lock); 2008 if ((uobj = *e->e_sigobject) == NULL) { 2009 uobj = uao_create(sz, 0); 2010 (*uobj->pgops->pgo_reference)(uobj); 2011 va = vm_map_min(kernel_map); 2012 if ((error = uvm_map(kernel_map, &va, round_page(sz), 2013 uobj, 0, 0, 2014 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 2015 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 2016 printf("kernel mapping failed %d\n", error); 2017 (*uobj->pgops->pgo_detach)(uobj); 2018 mutex_exit(&sigobject_lock); 2019 return error; 2020 } 2021 memcpy((void *)va, e->e_sigcode, sz); 2022 #ifdef PMAP_NEED_PROCWR 2023 pmap_procwr(&proc0, va, sz); 2024 #endif 2025 uvm_unmap(kernel_map, va, va + round_page(sz)); 2026 *e->e_sigobject = uobj; 2027 } 2028 mutex_exit(&sigobject_lock); 2029 } 2030 2031 /* Just a hint to uvm_map where to put it. */ 2032 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr, 2033 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN); 2034 2035 #ifdef __alpha__ 2036 /* 2037 * Tru64 puts /sbin/loader at the end of user virtual memory, 2038 * which causes the above calculation to put the sigcode at 2039 * an invalid address. Put it just below the text instead. 2040 */ 2041 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) { 2042 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 2043 } 2044 #endif 2045 2046 (*uobj->pgops->pgo_reference)(uobj); 2047 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 2048 uobj, 0, 0, 2049 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 2050 UVM_ADV_RANDOM, 0)); 2051 if (error) { 2052 DPRINTF(("%s, %d: map %p " 2053 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n", 2054 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz), 2055 va, error)); 2056 (*uobj->pgops->pgo_detach)(uobj); 2057 return error; 2058 } 2059 p->p_sigctx.ps_sigcode = (void *)va; 2060 return 0; 2061 } 2062 2063 /* 2064 * Release a refcount on spawn_exec_data and destroy memory, if this 2065 * was the last one. 2066 */ 2067 static void 2068 spawn_exec_data_release(struct spawn_exec_data *data) 2069 { 2070 if (atomic_dec_32_nv(&data->sed_refcnt) != 0) 2071 return; 2072 2073 cv_destroy(&data->sed_cv_child_ready); 2074 mutex_destroy(&data->sed_mtx_child); 2075 2076 if (data->sed_actions) 2077 posix_spawn_fa_free(data->sed_actions, 2078 data->sed_actions->len); 2079 if (data->sed_attrs) 2080 kmem_free(data->sed_attrs, 2081 sizeof(*data->sed_attrs)); 2082 kmem_free(data, sizeof(*data)); 2083 } 2084 2085 static int 2086 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions) 2087 { 2088 struct lwp *l = curlwp; 2089 register_t retval; 2090 int error, newfd; 2091 2092 if (actions == NULL) 2093 return 0; 2094 2095 for (size_t i = 0; i < actions->len; i++) { 2096 const struct posix_spawn_file_actions_entry *fae = 2097 &actions->fae[i]; 2098 switch (fae->fae_action) { 2099 case FAE_OPEN: 2100 if (fd_getfile(fae->fae_fildes) != NULL) { 2101 error = fd_close(fae->fae_fildes); 2102 if (error) 2103 return error; 2104 } 2105 error = fd_open(fae->fae_path, fae->fae_oflag, 2106 fae->fae_mode, &newfd); 2107 if (error) 2108 return error; 2109 if (newfd != fae->fae_fildes) { 2110 error = dodup(l, newfd, 2111 fae->fae_fildes, 0, &retval); 2112 if (fd_getfile(newfd) != NULL) 2113 fd_close(newfd); 2114 } 2115 break; 2116 case FAE_DUP2: 2117 error = dodup(l, fae->fae_fildes, 2118 fae->fae_newfildes, 0, &retval); 2119 break; 2120 case FAE_CLOSE: 2121 if (fd_getfile(fae->fae_fildes) == NULL) { 2122 return EBADF; 2123 } 2124 error = fd_close(fae->fae_fildes); 2125 break; 2126 } 2127 if (error) 2128 return error; 2129 } 2130 return 0; 2131 } 2132 2133 static int 2134 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent) 2135 { 2136 struct sigaction sigact; 2137 int error; 2138 struct proc *p = curproc; 2139 struct lwp *l = curlwp; 2140 2141 if (attrs == NULL) 2142 return 0; 2143 2144 memset(&sigact, 0, sizeof(sigact)); 2145 sigact._sa_u._sa_handler = SIG_DFL; 2146 sigact.sa_flags = 0; 2147 2148 /* 2149 * set state to SSTOP so that this proc can be found by pid. 2150 * see proc_enterprp, do_sched_setparam below 2151 */ 2152 mutex_enter(&proc_lock); 2153 /* 2154 * p_stat should be SACTIVE, so we need to adjust the 2155 * parent's p_nstopchild here. For safety, just make 2156 * we're on the good side of SDEAD before we adjust. 2157 */ 2158 int ostat = p->p_stat; 2159 KASSERT(ostat < SSTOP); 2160 p->p_stat = SSTOP; 2161 p->p_waited = 0; 2162 p->p_pptr->p_nstopchild++; 2163 mutex_exit(&proc_lock); 2164 2165 /* Set process group */ 2166 if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) { 2167 pid_t mypid = p->p_pid; 2168 pid_t pgrp = attrs->sa_pgroup; 2169 2170 if (pgrp == 0) 2171 pgrp = mypid; 2172 2173 error = proc_enterpgrp(parent, mypid, pgrp, false); 2174 if (error) 2175 goto out; 2176 } 2177 2178 /* Set scheduler policy */ 2179 if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER) 2180 error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy, 2181 &attrs->sa_schedparam); 2182 else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) { 2183 error = do_sched_setparam(parent->p_pid, 0, 2184 SCHED_NONE, &attrs->sa_schedparam); 2185 } 2186 if (error) 2187 goto out; 2188 2189 /* Reset user ID's */ 2190 if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) { 2191 error = do_setresgid(l, -1, kauth_cred_getgid(l->l_cred), -1, 2192 ID_E_EQ_R | ID_E_EQ_S); 2193 if (error) 2194 return error; 2195 error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1, 2196 ID_E_EQ_R | ID_E_EQ_S); 2197 if (error) 2198 goto out; 2199 } 2200 2201 /* Set signal masks/defaults */ 2202 if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) { 2203 mutex_enter(p->p_lock); 2204 error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL); 2205 mutex_exit(p->p_lock); 2206 if (error) 2207 goto out; 2208 } 2209 2210 if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) { 2211 /* 2212 * The following sigaction call is using a sigaction 2213 * version 0 trampoline which is in the compatibility 2214 * code only. This is not a problem because for SIG_DFL 2215 * and SIG_IGN, the trampolines are now ignored. If they 2216 * were not, this would be a problem because we are 2217 * holding the exec_lock, and the compat code needs 2218 * to do the same in order to replace the trampoline 2219 * code of the process. 2220 */ 2221 for (int i = 1; i <= NSIG; i++) { 2222 if (sigismember(&attrs->sa_sigdefault, i)) 2223 sigaction1(l, i, &sigact, NULL, NULL, 0); 2224 } 2225 } 2226 error = 0; 2227 out: 2228 mutex_enter(&proc_lock); 2229 p->p_stat = ostat; 2230 p->p_pptr->p_nstopchild--; 2231 mutex_exit(&proc_lock); 2232 return error; 2233 } 2234 2235 /* 2236 * A child lwp of a posix_spawn operation starts here and ends up in 2237 * cpu_spawn_return, dealing with all filedescriptor and scheduler 2238 * manipulations in between. 2239 * The parent waits for the child, as it is not clear whether the child 2240 * will be able to acquire its own exec_lock. If it can, the parent can 2241 * be released early and continue running in parallel. If not (or if the 2242 * magic debug flag is passed in the scheduler attribute struct), the 2243 * child rides on the parent's exec lock until it is ready to return to 2244 * to userland - and only then releases the parent. This method loses 2245 * concurrency, but improves error reporting. 2246 */ 2247 static void 2248 spawn_return(void *arg) 2249 { 2250 struct spawn_exec_data *spawn_data = arg; 2251 struct lwp *l = curlwp; 2252 struct proc *p = l->l_proc; 2253 int error; 2254 bool have_reflock; 2255 bool parent_is_waiting = true; 2256 2257 /* 2258 * Check if we can release parent early. 2259 * We either need to have no sed_attrs, or sed_attrs does not 2260 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require 2261 * safe access to the parent proc (passed in sed_parent). 2262 * We then try to get the exec_lock, and only if that works, we can 2263 * release the parent here already. 2264 */ 2265 struct posix_spawnattr *attrs = spawn_data->sed_attrs; 2266 if ((!attrs || (attrs->sa_flags 2267 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0) 2268 && rw_tryenter(&exec_lock, RW_READER)) { 2269 parent_is_waiting = false; 2270 mutex_enter(&spawn_data->sed_mtx_child); 2271 cv_signal(&spawn_data->sed_cv_child_ready); 2272 mutex_exit(&spawn_data->sed_mtx_child); 2273 } 2274 2275 /* don't allow debugger access yet */ 2276 rw_enter(&p->p_reflock, RW_WRITER); 2277 have_reflock = true; 2278 2279 /* handle posix_spawnattr */ 2280 error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent); 2281 if (error) 2282 goto report_error; 2283 2284 /* handle posix_spawn_file_actions */ 2285 error = handle_posix_spawn_file_actions(spawn_data->sed_actions); 2286 if (error) 2287 goto report_error; 2288 2289 /* now do the real exec */ 2290 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting, 2291 true); 2292 have_reflock = false; 2293 if (error == EJUSTRETURN) 2294 error = 0; 2295 else if (error) 2296 goto report_error; 2297 2298 if (parent_is_waiting) { 2299 mutex_enter(&spawn_data->sed_mtx_child); 2300 cv_signal(&spawn_data->sed_cv_child_ready); 2301 mutex_exit(&spawn_data->sed_mtx_child); 2302 } 2303 2304 /* release our refcount on the data */ 2305 spawn_exec_data_release(spawn_data); 2306 2307 if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) == 2308 (PSL_TRACED|PSL_TRACEDCHILD)) { 2309 eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN); 2310 } 2311 2312 /* and finally: leave to userland for the first time */ 2313 cpu_spawn_return(l); 2314 2315 /* NOTREACHED */ 2316 return; 2317 2318 report_error: 2319 if (have_reflock) { 2320 /* 2321 * We have not passed through execve_runproc(), 2322 * which would have released the p_reflock and also 2323 * taken ownership of the sed_exec part of spawn_data, 2324 * so release/free both here. 2325 */ 2326 rw_exit(&p->p_reflock); 2327 execve_free_data(&spawn_data->sed_exec); 2328 } 2329 2330 if (parent_is_waiting) { 2331 /* pass error to parent */ 2332 mutex_enter(&spawn_data->sed_mtx_child); 2333 spawn_data->sed_error = error; 2334 cv_signal(&spawn_data->sed_cv_child_ready); 2335 mutex_exit(&spawn_data->sed_mtx_child); 2336 } else { 2337 rw_exit(&exec_lock); 2338 } 2339 2340 /* release our refcount on the data */ 2341 spawn_exec_data_release(spawn_data); 2342 2343 /* done, exit */ 2344 mutex_enter(p->p_lock); 2345 /* 2346 * Posix explicitly asks for an exit code of 127 if we report 2347 * errors from the child process - so, unfortunately, there 2348 * is no way to report a more exact error code. 2349 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as 2350 * flag bit in the attrp argument to posix_spawn(2), see above. 2351 */ 2352 exit1(l, 127, 0); 2353 } 2354 2355 void 2356 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len) 2357 { 2358 2359 for (size_t i = 0; i < len; i++) { 2360 struct posix_spawn_file_actions_entry *fae = &fa->fae[i]; 2361 if (fae->fae_action != FAE_OPEN) 2362 continue; 2363 kmem_strfree(fae->fae_path); 2364 } 2365 if (fa->len > 0) 2366 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len); 2367 kmem_free(fa, sizeof(*fa)); 2368 } 2369 2370 static int 2371 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap, 2372 const struct posix_spawn_file_actions *ufa, rlim_t lim) 2373 { 2374 struct posix_spawn_file_actions *fa; 2375 struct posix_spawn_file_actions_entry *fae; 2376 char *pbuf = NULL; 2377 int error; 2378 size_t i = 0; 2379 2380 fa = kmem_alloc(sizeof(*fa), KM_SLEEP); 2381 error = copyin(ufa, fa, sizeof(*fa)); 2382 if (error || fa->len == 0) { 2383 kmem_free(fa, sizeof(*fa)); 2384 return error; /* 0 if not an error, and len == 0 */ 2385 } 2386 2387 if (fa->len > lim) { 2388 kmem_free(fa, sizeof(*fa)); 2389 return EINVAL; 2390 } 2391 2392 fa->size = fa->len; 2393 size_t fal = fa->len * sizeof(*fae); 2394 fae = fa->fae; 2395 fa->fae = kmem_alloc(fal, KM_SLEEP); 2396 error = copyin(fae, fa->fae, fal); 2397 if (error) 2398 goto out; 2399 2400 pbuf = PNBUF_GET(); 2401 for (; i < fa->len; i++) { 2402 fae = &fa->fae[i]; 2403 if (fae->fae_action != FAE_OPEN) 2404 continue; 2405 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal); 2406 if (error) 2407 goto out; 2408 fae->fae_path = kmem_alloc(fal, KM_SLEEP); 2409 memcpy(fae->fae_path, pbuf, fal); 2410 } 2411 PNBUF_PUT(pbuf); 2412 2413 *fap = fa; 2414 return 0; 2415 out: 2416 if (pbuf) 2417 PNBUF_PUT(pbuf); 2418 posix_spawn_fa_free(fa, i); 2419 return error; 2420 } 2421 2422 /* 2423 * N.B. increments nprocs upon success. Callers need to drop nprocs if 2424 * they fail for some other reason. 2425 */ 2426 int 2427 check_posix_spawn(struct lwp *l1) 2428 { 2429 int error, tnprocs, count; 2430 uid_t uid; 2431 struct proc *p1; 2432 2433 p1 = l1->l_proc; 2434 uid = kauth_cred_getuid(l1->l_cred); 2435 tnprocs = atomic_inc_uint_nv(&nprocs); 2436 2437 /* 2438 * Although process entries are dynamically created, we still keep 2439 * a global limit on the maximum number we will create. 2440 */ 2441 if (__predict_false(tnprocs >= maxproc)) 2442 error = -1; 2443 else 2444 error = kauth_authorize_process(l1->l_cred, 2445 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL); 2446 2447 if (error) { 2448 atomic_dec_uint(&nprocs); 2449 return EAGAIN; 2450 } 2451 2452 /* 2453 * Enforce limits. 2454 */ 2455 count = chgproccnt(uid, 1); 2456 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT, 2457 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 2458 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 && 2459 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 2460 (void)chgproccnt(uid, -1); 2461 atomic_dec_uint(&nprocs); 2462 return EAGAIN; 2463 } 2464 2465 return 0; 2466 } 2467 2468 int 2469 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path, 2470 struct posix_spawn_file_actions *fa, 2471 struct posix_spawnattr *sa, 2472 char *const *argv, char *const *envp, 2473 execve_fetch_element_t fetch) 2474 { 2475 2476 struct proc *p1, *p2; 2477 struct lwp *l2; 2478 int error; 2479 struct spawn_exec_data *spawn_data; 2480 vaddr_t uaddr; 2481 pid_t pid; 2482 bool have_exec_lock = false; 2483 2484 p1 = l1->l_proc; 2485 2486 /* Allocate and init spawn_data */ 2487 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP); 2488 spawn_data->sed_refcnt = 1; /* only parent so far */ 2489 cv_init(&spawn_data->sed_cv_child_ready, "pspawn"); 2490 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE); 2491 mutex_enter(&spawn_data->sed_mtx_child); 2492 2493 /* 2494 * Do the first part of the exec now, collect state 2495 * in spawn_data. 2496 */ 2497 error = execve_loadvm(l1, true, path, -1, argv, 2498 envp, fetch, &spawn_data->sed_exec); 2499 if (error == EJUSTRETURN) 2500 error = 0; 2501 else if (error) 2502 goto error_exit; 2503 2504 have_exec_lock = true; 2505 2506 /* 2507 * Allocate virtual address space for the U-area now, while it 2508 * is still easy to abort the fork operation if we're out of 2509 * kernel virtual address space. 2510 */ 2511 uaddr = uvm_uarea_alloc(); 2512 if (__predict_false(uaddr == 0)) { 2513 error = ENOMEM; 2514 goto error_exit; 2515 } 2516 2517 /* 2518 * Allocate new proc. Borrow proc0 vmspace for it, we will 2519 * replace it with its own before returning to userland 2520 * in the child. 2521 */ 2522 p2 = proc_alloc(); 2523 if (p2 == NULL) { 2524 /* We were unable to allocate a process ID. */ 2525 error = EAGAIN; 2526 goto error_exit; 2527 } 2528 2529 /* 2530 * This is a point of no return, we will have to go through 2531 * the child proc to properly clean it up past this point. 2532 */ 2533 pid = p2->p_pid; 2534 2535 /* 2536 * Make a proc table entry for the new process. 2537 * Start by zeroing the section of proc that is zero-initialized, 2538 * then copy the section that is copied directly from the parent. 2539 */ 2540 memset(&p2->p_startzero, 0, 2541 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero)); 2542 memcpy(&p2->p_startcopy, &p1->p_startcopy, 2543 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy)); 2544 p2->p_vmspace = proc0.p_vmspace; 2545 2546 TAILQ_INIT(&p2->p_sigpend.sp_info); 2547 2548 LIST_INIT(&p2->p_lwps); 2549 LIST_INIT(&p2->p_sigwaiters); 2550 2551 /* 2552 * Duplicate sub-structures as needed. 2553 * Increase reference counts on shared objects. 2554 * Inherit flags we want to keep. The flags related to SIGCHLD 2555 * handling are important in order to keep a consistent behaviour 2556 * for the child after the fork. If we are a 32-bit process, the 2557 * child will be too. 2558 */ 2559 p2->p_flag = 2560 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32); 2561 p2->p_emul = p1->p_emul; 2562 p2->p_execsw = p1->p_execsw; 2563 2564 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH); 2565 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE); 2566 rw_init(&p2->p_reflock); 2567 cv_init(&p2->p_waitcv, "wait"); 2568 cv_init(&p2->p_lwpcv, "lwpwait"); 2569 2570 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 2571 2572 kauth_proc_fork(p1, p2); 2573 2574 p2->p_raslist = NULL; 2575 p2->p_fd = fd_copy(); 2576 2577 /* XXX racy */ 2578 p2->p_mqueue_cnt = p1->p_mqueue_cnt; 2579 2580 p2->p_cwdi = cwdinit(); 2581 2582 /* 2583 * Note: p_limit (rlimit stuff) is copy-on-write, so normally 2584 * we just need increase pl_refcnt. 2585 */ 2586 if (!p1->p_limit->pl_writeable) { 2587 lim_addref(p1->p_limit); 2588 p2->p_limit = p1->p_limit; 2589 } else { 2590 p2->p_limit = lim_copy(p1->p_limit); 2591 } 2592 2593 p2->p_lflag = 0; 2594 l1->l_vforkwaiting = false; 2595 p2->p_sflag = 0; 2596 p2->p_slflag = 0; 2597 p2->p_pptr = p1; 2598 p2->p_ppid = p1->p_pid; 2599 LIST_INIT(&p2->p_children); 2600 2601 p2->p_aio = NULL; 2602 2603 #ifdef KTRACE 2604 /* 2605 * Copy traceflag and tracefile if enabled. 2606 * If not inherited, these were zeroed above. 2607 */ 2608 if (p1->p_traceflag & KTRFAC_INHERIT) { 2609 mutex_enter(&ktrace_lock); 2610 p2->p_traceflag = p1->p_traceflag; 2611 if ((p2->p_tracep = p1->p_tracep) != NULL) 2612 ktradref(p2); 2613 mutex_exit(&ktrace_lock); 2614 } 2615 #endif 2616 2617 /* 2618 * Create signal actions for the child process. 2619 */ 2620 p2->p_sigacts = sigactsinit(p1, 0); 2621 mutex_enter(p1->p_lock); 2622 p2->p_sflag |= 2623 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 2624 sched_proc_fork(p1, p2); 2625 mutex_exit(p1->p_lock); 2626 2627 p2->p_stflag = p1->p_stflag; 2628 2629 /* 2630 * p_stats. 2631 * Copy parts of p_stats, and zero out the rest. 2632 */ 2633 p2->p_stats = pstatscopy(p1->p_stats); 2634 2635 /* copy over machdep flags to the new proc */ 2636 cpu_proc_fork(p1, p2); 2637 2638 /* 2639 * Prepare remaining parts of spawn data 2640 */ 2641 spawn_data->sed_actions = fa; 2642 spawn_data->sed_attrs = sa; 2643 2644 spawn_data->sed_parent = p1; 2645 2646 /* create LWP */ 2647 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data, 2648 &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk); 2649 l2->l_ctxlink = NULL; /* reset ucontext link */ 2650 2651 /* 2652 * Copy the credential so other references don't see our changes. 2653 * Test to see if this is necessary first, since in the common case 2654 * we won't need a private reference. 2655 */ 2656 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) || 2657 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) { 2658 l2->l_cred = kauth_cred_copy(l2->l_cred); 2659 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred)); 2660 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred)); 2661 } 2662 2663 /* Update the master credentials. */ 2664 if (l2->l_cred != p2->p_cred) { 2665 kauth_cred_t ocred; 2666 2667 kauth_cred_hold(l2->l_cred); 2668 mutex_enter(p2->p_lock); 2669 ocred = p2->p_cred; 2670 p2->p_cred = l2->l_cred; 2671 mutex_exit(p2->p_lock); 2672 kauth_cred_free(ocred); 2673 } 2674 2675 *child_ok = true; 2676 spawn_data->sed_refcnt = 2; /* child gets it as well */ 2677 #if 0 2678 l2->l_nopreempt = 1; /* start it non-preemptable */ 2679 #endif 2680 2681 /* 2682 * It's now safe for the scheduler and other processes to see the 2683 * child process. 2684 */ 2685 mutex_enter(&proc_lock); 2686 2687 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 2688 p2->p_lflag |= PL_CONTROLT; 2689 2690 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 2691 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */ 2692 2693 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) == 2694 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2695 proc_changeparent(p2, p1->p_pptr); 2696 SET(p2->p_slflag, PSL_TRACEDCHILD); 2697 } 2698 2699 p2->p_oppid = p1->p_pid; /* Remember the original parent id. */ 2700 2701 LIST_INSERT_AFTER(p1, p2, p_pglist); 2702 LIST_INSERT_HEAD(&allproc, p2, p_list); 2703 2704 p2->p_trace_enabled = trace_is_enabled(p2); 2705 #ifdef __HAVE_SYSCALL_INTERN 2706 (*p2->p_emul->e_syscall_intern)(p2); 2707 #endif 2708 2709 /* 2710 * Make child runnable, set start time, and add to run queue except 2711 * if the parent requested the child to start in SSTOP state. 2712 */ 2713 mutex_enter(p2->p_lock); 2714 2715 getmicrotime(&p2->p_stats->p_start); 2716 2717 lwp_lock(l2); 2718 KASSERT(p2->p_nrlwps == 1); 2719 KASSERT(l2->l_stat == LSIDL); 2720 p2->p_nrlwps = 1; 2721 p2->p_stat = SACTIVE; 2722 setrunnable(l2); 2723 /* LWP now unlocked */ 2724 2725 mutex_exit(p2->p_lock); 2726 mutex_exit(&proc_lock); 2727 2728 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child); 2729 error = spawn_data->sed_error; 2730 mutex_exit(&spawn_data->sed_mtx_child); 2731 spawn_exec_data_release(spawn_data); 2732 2733 rw_exit(&p1->p_reflock); 2734 rw_exit(&exec_lock); 2735 have_exec_lock = false; 2736 2737 *pid_res = pid; 2738 2739 if (error) 2740 return error; 2741 2742 if (p1->p_slflag & PSL_TRACED) { 2743 /* Paranoid check */ 2744 mutex_enter(&proc_lock); 2745 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) != 2746 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2747 mutex_exit(&proc_lock); 2748 return 0; 2749 } 2750 2751 mutex_enter(p1->p_lock); 2752 eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid); 2753 } 2754 return 0; 2755 2756 error_exit: 2757 if (have_exec_lock) { 2758 execve_free_data(&spawn_data->sed_exec); 2759 rw_exit(&p1->p_reflock); 2760 rw_exit(&exec_lock); 2761 } 2762 mutex_exit(&spawn_data->sed_mtx_child); 2763 spawn_exec_data_release(spawn_data); 2764 2765 return error; 2766 } 2767 2768 int 2769 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap, 2770 register_t *retval) 2771 { 2772 /* { 2773 syscallarg(pid_t *) pid; 2774 syscallarg(const char *) path; 2775 syscallarg(const struct posix_spawn_file_actions *) file_actions; 2776 syscallarg(const struct posix_spawnattr *) attrp; 2777 syscallarg(char *const *) argv; 2778 syscallarg(char *const *) envp; 2779 } */ 2780 2781 int error; 2782 struct posix_spawn_file_actions *fa = NULL; 2783 struct posix_spawnattr *sa = NULL; 2784 pid_t pid; 2785 bool child_ok = false; 2786 rlim_t max_fileactions; 2787 proc_t *p = l1->l_proc; 2788 2789 /* check_posix_spawn() increments nprocs for us. */ 2790 error = check_posix_spawn(l1); 2791 if (error) { 2792 *retval = error; 2793 return 0; 2794 } 2795 2796 /* copy in file_actions struct */ 2797 if (SCARG(uap, file_actions) != NULL) { 2798 max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur, 2799 maxfiles); 2800 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions), 2801 max_fileactions); 2802 if (error) 2803 goto error_exit; 2804 } 2805 2806 /* copyin posix_spawnattr struct */ 2807 if (SCARG(uap, attrp) != NULL) { 2808 sa = kmem_alloc(sizeof(*sa), KM_SLEEP); 2809 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa)); 2810 if (error) 2811 goto error_exit; 2812 } 2813 2814 /* 2815 * Do the spawn 2816 */ 2817 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa, 2818 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element); 2819 if (error) 2820 goto error_exit; 2821 2822 if (error == 0 && SCARG(uap, pid) != NULL) 2823 error = copyout(&pid, SCARG(uap, pid), sizeof(pid)); 2824 2825 *retval = error; 2826 return 0; 2827 2828 error_exit: 2829 if (!child_ok) { 2830 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1); 2831 atomic_dec_uint(&nprocs); 2832 2833 if (sa) 2834 kmem_free(sa, sizeof(*sa)); 2835 if (fa) 2836 posix_spawn_fa_free(fa, fa->len); 2837 } 2838 2839 *retval = error; 2840 return 0; 2841 } 2842 2843 void 2844 exec_free_emul_arg(struct exec_package *epp) 2845 { 2846 if (epp->ep_emul_arg_free != NULL) { 2847 KASSERT(epp->ep_emul_arg != NULL); 2848 (*epp->ep_emul_arg_free)(epp->ep_emul_arg); 2849 epp->ep_emul_arg_free = NULL; 2850 epp->ep_emul_arg = NULL; 2851 } else { 2852 KASSERT(epp->ep_emul_arg == NULL); 2853 } 2854 } 2855 2856 #ifdef DEBUG_EXEC 2857 static void 2858 dump_vmcmds(const struct exec_package * const epp, size_t x, int error) 2859 { 2860 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0]; 2861 size_t j; 2862 2863 if (error == 0) 2864 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used)); 2865 else 2866 DPRINTF(("vmcmds %zu/%u, error %d\n", x, 2867 epp->ep_vmcmds.evs_used, error)); 2868 2869 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) { 2870 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#" 2871 PRIxVADDR"/%#"PRIxVSIZE" fd@%#" 2872 PRIxVSIZE" prot=0%o flags=%d\n", j, 2873 vp[j].ev_proc == vmcmd_map_pagedvn ? 2874 "pagedvn" : 2875 vp[j].ev_proc == vmcmd_map_readvn ? 2876 "readvn" : 2877 vp[j].ev_proc == vmcmd_map_zero ? 2878 "zero" : "*unknown*", 2879 vp[j].ev_addr, vp[j].ev_len, 2880 vp[j].ev_offset, vp[j].ev_prot, 2881 vp[j].ev_flags)); 2882 if (error != 0 && j == x) 2883 DPRINTF((" ^--- failed\n")); 2884 } 2885 } 2886 #endif 2887