1 /* $NetBSD: kern_exec.c,v 1.510 2021/10/10 18:07:51 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 1.510 2021/10/10 18:07:51 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 /* 1371 * Notify anyone who might care that we've exec'd. 1372 * 1373 * This is slightly racy; someone could sneak in and 1374 * attach a knote after we've decided not to notify, 1375 * or vice-versa, but that's not particularly bothersome. 1376 * knote_proc_exec() will acquire p->p_lock as needed. 1377 */ 1378 if (!SLIST_EMPTY(&p->p_klist)) { 1379 knote_proc_exec(p); 1380 } 1381 1382 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1383 1384 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0); 1385 1386 emulexec(l, epp); 1387 1388 /* Allow new references from the debugger/procfs. */ 1389 rw_exit(&p->p_reflock); 1390 if (!no_local_exec_lock) 1391 rw_exit(&exec_lock); 1392 1393 mutex_enter(&proc_lock); 1394 1395 /* posix_spawn(3) reports a single event with implied exec(3) */ 1396 if ((p->p_slflag & PSL_TRACED) && !is_spawn) { 1397 mutex_enter(p->p_lock); 1398 eventswitch(TRAP_EXEC, 0, 0); 1399 mutex_enter(&proc_lock); 1400 } 1401 1402 if (p->p_sflag & PS_STOPEXEC) { 1403 ksiginfoq_t kq; 1404 1405 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 1406 p->p_pptr->p_nstopchild++; 1407 p->p_waited = 0; 1408 mutex_enter(p->p_lock); 1409 ksiginfo_queue_init(&kq); 1410 sigclearall(p, &contsigmask, &kq); 1411 lwp_lock(l); 1412 l->l_stat = LSSTOP; 1413 p->p_stat = SSTOP; 1414 p->p_nrlwps--; 1415 lwp_unlock(l); 1416 mutex_exit(p->p_lock); 1417 mutex_exit(&proc_lock); 1418 lwp_lock(l); 1419 spc_lock(l->l_cpu); 1420 mi_switch(l); 1421 ksiginfo_queue_drain(&kq); 1422 } else { 1423 mutex_exit(&proc_lock); 1424 } 1425 1426 exec_path_free(data); 1427 #ifdef TRACE_EXEC 1428 DPRINTF(("%s finished\n", __func__)); 1429 #endif 1430 return EJUSTRETURN; 1431 1432 exec_abort: 1433 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 1434 rw_exit(&p->p_reflock); 1435 if (!no_local_exec_lock) 1436 rw_exit(&exec_lock); 1437 1438 exec_path_free(data); 1439 1440 /* 1441 * the old process doesn't exist anymore. exit gracefully. 1442 * get rid of the (new) address space we have created, if any, get rid 1443 * of our namei data and vnode, and exit noting failure 1444 */ 1445 if (vm != NULL) { 1446 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 1447 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 1448 } 1449 1450 exec_free_emul_arg(epp); 1451 pool_put(&exec_pool, data->ed_argp); 1452 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1453 if (epp->ep_emul_root != NULL) 1454 vrele(epp->ep_emul_root); 1455 if (epp->ep_interp != NULL) 1456 vrele(epp->ep_interp); 1457 1458 /* Acquire the sched-state mutex (exit1() will release it). */ 1459 if (!is_spawn) { 1460 mutex_enter(p->p_lock); 1461 exit1(l, error, SIGABRT); 1462 } 1463 1464 return error; 1465 } 1466 1467 int 1468 execve1(struct lwp *l, bool has_path, const char *path, int fd, 1469 char * const *args, char * const *envs, 1470 execve_fetch_element_t fetch_element) 1471 { 1472 struct execve_data data; 1473 int error; 1474 1475 error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element, 1476 &data); 1477 if (error) 1478 return error; 1479 error = execve_runproc(l, &data, false, false); 1480 return error; 1481 } 1482 1483 static size_t 1484 fromptrsz(const struct exec_package *epp) 1485 { 1486 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *); 1487 } 1488 1489 static size_t 1490 ptrsz(const struct exec_package *epp) 1491 { 1492 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *); 1493 } 1494 1495 static size_t 1496 calcargs(struct execve_data * restrict data, const size_t argenvstrlen) 1497 { 1498 struct exec_package * const epp = &data->ed_pack; 1499 1500 const size_t nargenvptrs = 1501 1 + /* long argc */ 1502 data->ed_argc + /* char *argv[] */ 1503 1 + /* \0 */ 1504 data->ed_envc + /* char *env[] */ 1505 1; /* \0 */ 1506 1507 return (nargenvptrs * ptrsz(epp)) /* pointers */ 1508 + argenvstrlen /* strings */ 1509 + epp->ep_esch->es_arglen; /* auxinfo */ 1510 } 1511 1512 static size_t 1513 calcstack(struct execve_data * restrict data, const size_t gaplen) 1514 { 1515 struct exec_package * const epp = &data->ed_pack; 1516 1517 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode - 1518 epp->ep_esch->es_emul->e_sigcode; 1519 1520 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ? 1521 sizeof(struct ps_strings32) : sizeof(struct ps_strings); 1522 1523 const size_t sigcode_psstr_sz = 1524 data->ed_szsigcode + /* sigcode */ 1525 data->ed_ps_strings_sz + /* ps_strings */ 1526 STACK_PTHREADSPACE; /* pthread space */ 1527 1528 const size_t stacklen = 1529 data->ed_argslen + 1530 gaplen + 1531 sigcode_psstr_sz; 1532 1533 /* make the stack "safely" aligned */ 1534 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES); 1535 } 1536 1537 static int 1538 copyoutargs(struct execve_data * restrict data, struct lwp *l, 1539 char * const newstack) 1540 { 1541 struct exec_package * const epp = &data->ed_pack; 1542 struct proc *p = l->l_proc; 1543 int error; 1544 1545 memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo)); 1546 1547 /* remember information about the process */ 1548 data->ed_arginfo.ps_nargvstr = data->ed_argc; 1549 data->ed_arginfo.ps_nenvstr = data->ed_envc; 1550 1551 /* 1552 * Allocate the stack address passed to the newly execve()'ed process. 1553 * 1554 * The new stack address will be set to the SP (stack pointer) register 1555 * in setregs(). 1556 */ 1557 1558 char *newargs = STACK_ALLOC( 1559 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen); 1560 1561 error = (*epp->ep_esch->es_copyargs)(l, epp, 1562 &data->ed_arginfo, &newargs, data->ed_argp); 1563 1564 if (error) { 1565 DPRINTF(("%s: copyargs failed %d\n", __func__, error)); 1566 return error; 1567 } 1568 1569 error = copyoutpsstrs(data, p); 1570 if (error != 0) 1571 return error; 1572 1573 return 0; 1574 } 1575 1576 static int 1577 copyoutpsstrs(struct execve_data * restrict data, struct proc *p) 1578 { 1579 struct exec_package * const epp = &data->ed_pack; 1580 struct ps_strings32 arginfo32; 1581 void *aip; 1582 int error; 1583 1584 /* fill process ps_strings info */ 1585 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr, 1586 STACK_PTHREADSPACE), data->ed_ps_strings_sz); 1587 1588 if (epp->ep_flags & EXEC_32) { 1589 aip = &arginfo32; 1590 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr; 1591 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr; 1592 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr; 1593 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr; 1594 } else 1595 aip = &data->ed_arginfo; 1596 1597 /* copy out the process's ps_strings structure */ 1598 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)) 1599 != 0) { 1600 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n", 1601 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)); 1602 return error; 1603 } 1604 1605 return 0; 1606 } 1607 1608 static int 1609 copyinargs(struct execve_data * restrict data, char * const *args, 1610 char * const *envs, execve_fetch_element_t fetch_element, char **dpp) 1611 { 1612 struct exec_package * const epp = &data->ed_pack; 1613 char *dp; 1614 size_t i; 1615 int error; 1616 1617 dp = *dpp; 1618 1619 data->ed_argc = 0; 1620 1621 /* copy the fake args list, if there's one, freeing it as we go */ 1622 if (epp->ep_flags & EXEC_HASARGL) { 1623 struct exec_fakearg *fa = epp->ep_fa; 1624 1625 while (fa->fa_arg != NULL) { 1626 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1627 size_t len; 1628 1629 len = strlcpy(dp, fa->fa_arg, maxlen); 1630 /* Count NUL into len. */ 1631 if (len < maxlen) 1632 len++; 1633 else { 1634 while (fa->fa_arg != NULL) { 1635 kmem_free(fa->fa_arg, fa->fa_len); 1636 fa++; 1637 } 1638 kmem_free(epp->ep_fa, epp->ep_fa_len); 1639 epp->ep_flags &= ~EXEC_HASARGL; 1640 return E2BIG; 1641 } 1642 ktrexecarg(fa->fa_arg, len - 1); 1643 dp += len; 1644 1645 kmem_free(fa->fa_arg, fa->fa_len); 1646 fa++; 1647 data->ed_argc++; 1648 } 1649 kmem_free(epp->ep_fa, epp->ep_fa_len); 1650 epp->ep_flags &= ~EXEC_HASARGL; 1651 } 1652 1653 /* 1654 * Read and count argument strings from user. 1655 */ 1656 1657 if (args == NULL) { 1658 DPRINTF(("%s: null args\n", __func__)); 1659 return EINVAL; 1660 } 1661 if (epp->ep_flags & EXEC_SKIPARG) 1662 args = (const void *)((const char *)args + fromptrsz(epp)); 1663 i = 0; 1664 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg); 1665 if (error != 0) { 1666 DPRINTF(("%s: copyin arg %d\n", __func__, error)); 1667 return error; 1668 } 1669 data->ed_argc += i; 1670 1671 /* 1672 * Read and count environment strings from user. 1673 */ 1674 1675 data->ed_envc = 0; 1676 /* environment need not be there */ 1677 if (envs == NULL) 1678 goto done; 1679 i = 0; 1680 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv); 1681 if (error != 0) { 1682 DPRINTF(("%s: copyin env %d\n", __func__, error)); 1683 return error; 1684 } 1685 data->ed_envc += i; 1686 1687 done: 1688 *dpp = dp; 1689 1690 return 0; 1691 } 1692 1693 static int 1694 copyinargstrs(struct execve_data * restrict data, char * const *strs, 1695 execve_fetch_element_t fetch_element, char **dpp, size_t *ip, 1696 void (*ktr)(const void *, size_t)) 1697 { 1698 char *dp, *sp; 1699 size_t i; 1700 int error; 1701 1702 dp = *dpp; 1703 1704 i = 0; 1705 while (1) { 1706 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1707 size_t len; 1708 1709 if ((error = (*fetch_element)(strs, i, &sp)) != 0) { 1710 return error; 1711 } 1712 if (!sp) 1713 break; 1714 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) { 1715 if (error == ENAMETOOLONG) 1716 error = E2BIG; 1717 return error; 1718 } 1719 if (__predict_false(ktrace_on)) 1720 (*ktr)(dp, len - 1); 1721 dp += len; 1722 i++; 1723 } 1724 1725 *dpp = dp; 1726 *ip = i; 1727 1728 return 0; 1729 } 1730 1731 /* 1732 * Copy argv and env strings from kernel buffer (argp) to the new stack. 1733 * Those strings are located just after auxinfo. 1734 */ 1735 int 1736 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo, 1737 char **stackp, void *argp) 1738 { 1739 char **cpp, *dp, *sp; 1740 size_t len; 1741 void *nullp; 1742 long argc, envc; 1743 int error; 1744 1745 cpp = (char **)*stackp; 1746 nullp = NULL; 1747 argc = arginfo->ps_nargvstr; 1748 envc = arginfo->ps_nenvstr; 1749 1750 /* argc on stack is long */ 1751 CTASSERT(sizeof(*cpp) == sizeof(argc)); 1752 1753 dp = (char *)(cpp + 1754 1 + /* long argc */ 1755 argc + /* char *argv[] */ 1756 1 + /* \0 */ 1757 envc + /* char *env[] */ 1758 1) + /* \0 */ 1759 pack->ep_esch->es_arglen; /* auxinfo */ 1760 sp = argp; 1761 1762 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) { 1763 COPYPRINTF("", cpp - 1, sizeof(argc)); 1764 return error; 1765 } 1766 1767 /* XXX don't copy them out, remap them! */ 1768 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 1769 1770 for (; --argc >= 0; sp += len, dp += len) { 1771 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1772 COPYPRINTF("", cpp - 1, sizeof(dp)); 1773 return error; 1774 } 1775 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1776 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1777 return error; 1778 } 1779 } 1780 1781 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1782 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1783 return error; 1784 } 1785 1786 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 1787 1788 for (; --envc >= 0; sp += len, dp += len) { 1789 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1790 COPYPRINTF("", cpp - 1, sizeof(dp)); 1791 return error; 1792 } 1793 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1794 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1795 return error; 1796 } 1797 1798 } 1799 1800 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1801 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1802 return error; 1803 } 1804 1805 *stackp = (char *)cpp; 1806 return 0; 1807 } 1808 1809 1810 /* 1811 * Add execsw[] entries. 1812 */ 1813 int 1814 exec_add(struct execsw *esp, int count) 1815 { 1816 struct exec_entry *it; 1817 int i; 1818 1819 if (count == 0) { 1820 return 0; 1821 } 1822 1823 /* Check for duplicates. */ 1824 rw_enter(&exec_lock, RW_WRITER); 1825 for (i = 0; i < count; i++) { 1826 LIST_FOREACH(it, &ex_head, ex_list) { 1827 /* assume unique (makecmds, probe_func, emulation) */ 1828 if (it->ex_sw->es_makecmds == esp[i].es_makecmds && 1829 it->ex_sw->u.elf_probe_func == 1830 esp[i].u.elf_probe_func && 1831 it->ex_sw->es_emul == esp[i].es_emul) { 1832 rw_exit(&exec_lock); 1833 return EEXIST; 1834 } 1835 } 1836 } 1837 1838 /* Allocate new entries. */ 1839 for (i = 0; i < count; i++) { 1840 it = kmem_alloc(sizeof(*it), KM_SLEEP); 1841 it->ex_sw = &esp[i]; 1842 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1843 } 1844 1845 /* update execsw[] */ 1846 exec_init(0); 1847 rw_exit(&exec_lock); 1848 return 0; 1849 } 1850 1851 /* 1852 * Remove execsw[] entry. 1853 */ 1854 int 1855 exec_remove(struct execsw *esp, int count) 1856 { 1857 struct exec_entry *it, *next; 1858 int i; 1859 const struct proclist_desc *pd; 1860 proc_t *p; 1861 1862 if (count == 0) { 1863 return 0; 1864 } 1865 1866 /* Abort if any are busy. */ 1867 rw_enter(&exec_lock, RW_WRITER); 1868 for (i = 0; i < count; i++) { 1869 mutex_enter(&proc_lock); 1870 for (pd = proclists; pd->pd_list != NULL; pd++) { 1871 PROCLIST_FOREACH(p, pd->pd_list) { 1872 if (p->p_execsw == &esp[i]) { 1873 mutex_exit(&proc_lock); 1874 rw_exit(&exec_lock); 1875 return EBUSY; 1876 } 1877 } 1878 } 1879 mutex_exit(&proc_lock); 1880 } 1881 1882 /* None are busy, so remove them all. */ 1883 for (i = 0; i < count; i++) { 1884 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) { 1885 next = LIST_NEXT(it, ex_list); 1886 if (it->ex_sw == &esp[i]) { 1887 LIST_REMOVE(it, ex_list); 1888 kmem_free(it, sizeof(*it)); 1889 break; 1890 } 1891 } 1892 } 1893 1894 /* update execsw[] */ 1895 exec_init(0); 1896 rw_exit(&exec_lock); 1897 return 0; 1898 } 1899 1900 /* 1901 * Initialize exec structures. If init_boot is true, also does necessary 1902 * one-time initialization (it's called from main() that way). 1903 * Once system is multiuser, this should be called with exec_lock held, 1904 * i.e. via exec_{add|remove}(). 1905 */ 1906 int 1907 exec_init(int init_boot) 1908 { 1909 const struct execsw **sw; 1910 struct exec_entry *ex; 1911 SLIST_HEAD(,exec_entry) first; 1912 SLIST_HEAD(,exec_entry) any; 1913 SLIST_HEAD(,exec_entry) last; 1914 int i, sz; 1915 1916 if (init_boot) { 1917 /* do one-time initializations */ 1918 vaddr_t vmin = 0, vmax; 1919 1920 rw_init(&exec_lock); 1921 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE); 1922 exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax, 1923 maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL); 1924 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH, 1925 "execargs", &exec_palloc, IPL_NONE); 1926 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0); 1927 } else { 1928 KASSERT(rw_write_held(&exec_lock)); 1929 } 1930 1931 /* Sort each entry onto the appropriate queue. */ 1932 SLIST_INIT(&first); 1933 SLIST_INIT(&any); 1934 SLIST_INIT(&last); 1935 sz = 0; 1936 LIST_FOREACH(ex, &ex_head, ex_list) { 1937 switch(ex->ex_sw->es_prio) { 1938 case EXECSW_PRIO_FIRST: 1939 SLIST_INSERT_HEAD(&first, ex, ex_slist); 1940 break; 1941 case EXECSW_PRIO_ANY: 1942 SLIST_INSERT_HEAD(&any, ex, ex_slist); 1943 break; 1944 case EXECSW_PRIO_LAST: 1945 SLIST_INSERT_HEAD(&last, ex, ex_slist); 1946 break; 1947 default: 1948 panic("%s", __func__); 1949 break; 1950 } 1951 sz++; 1952 } 1953 1954 /* 1955 * Create new execsw[]. Ensure we do not try a zero-sized 1956 * allocation. 1957 */ 1958 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP); 1959 i = 0; 1960 SLIST_FOREACH(ex, &first, ex_slist) { 1961 sw[i++] = ex->ex_sw; 1962 } 1963 SLIST_FOREACH(ex, &any, ex_slist) { 1964 sw[i++] = ex->ex_sw; 1965 } 1966 SLIST_FOREACH(ex, &last, ex_slist) { 1967 sw[i++] = ex->ex_sw; 1968 } 1969 1970 /* Replace old execsw[] and free used memory. */ 1971 if (execsw != NULL) { 1972 kmem_free(__UNCONST(execsw), 1973 nexecs * sizeof(struct execsw *) + 1); 1974 } 1975 execsw = sw; 1976 nexecs = sz; 1977 1978 /* Figure out the maximum size of an exec header. */ 1979 exec_maxhdrsz = sizeof(int); 1980 for (i = 0; i < nexecs; i++) { 1981 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1982 exec_maxhdrsz = execsw[i]->es_hdrsz; 1983 } 1984 1985 return 0; 1986 } 1987 1988 static int 1989 exec_sigcode_map(struct proc *p, const struct emul *e) 1990 { 1991 vaddr_t va; 1992 vsize_t sz; 1993 int error; 1994 struct uvm_object *uobj; 1995 1996 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1997 1998 if (e->e_sigobject == NULL || sz == 0) { 1999 return 0; 2000 } 2001 2002 /* 2003 * If we don't have a sigobject for this emulation, create one. 2004 * 2005 * sigobject is an anonymous memory object (just like SYSV shared 2006 * memory) that we keep a permanent reference to and that we map 2007 * in all processes that need this sigcode. The creation is simple, 2008 * we create an object, add a permanent reference to it, map it in 2009 * kernel space, copy out the sigcode to it and unmap it. 2010 * We map it with PROT_READ|PROT_EXEC into the process just 2011 * the way sys_mmap() would map it. 2012 */ 2013 2014 uobj = *e->e_sigobject; 2015 if (uobj == NULL) { 2016 mutex_enter(&sigobject_lock); 2017 if ((uobj = *e->e_sigobject) == NULL) { 2018 uobj = uao_create(sz, 0); 2019 (*uobj->pgops->pgo_reference)(uobj); 2020 va = vm_map_min(kernel_map); 2021 if ((error = uvm_map(kernel_map, &va, round_page(sz), 2022 uobj, 0, 0, 2023 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 2024 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 2025 printf("kernel mapping failed %d\n", error); 2026 (*uobj->pgops->pgo_detach)(uobj); 2027 mutex_exit(&sigobject_lock); 2028 return error; 2029 } 2030 memcpy((void *)va, e->e_sigcode, sz); 2031 #ifdef PMAP_NEED_PROCWR 2032 pmap_procwr(&proc0, va, sz); 2033 #endif 2034 uvm_unmap(kernel_map, va, va + round_page(sz)); 2035 *e->e_sigobject = uobj; 2036 } 2037 mutex_exit(&sigobject_lock); 2038 } 2039 2040 /* Just a hint to uvm_map where to put it. */ 2041 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr, 2042 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN); 2043 2044 #ifdef __alpha__ 2045 /* 2046 * Tru64 puts /sbin/loader at the end of user virtual memory, 2047 * which causes the above calculation to put the sigcode at 2048 * an invalid address. Put it just below the text instead. 2049 */ 2050 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) { 2051 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 2052 } 2053 #endif 2054 2055 (*uobj->pgops->pgo_reference)(uobj); 2056 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 2057 uobj, 0, 0, 2058 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 2059 UVM_ADV_RANDOM, 0)); 2060 if (error) { 2061 DPRINTF(("%s, %d: map %p " 2062 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n", 2063 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz), 2064 va, error)); 2065 (*uobj->pgops->pgo_detach)(uobj); 2066 return error; 2067 } 2068 p->p_sigctx.ps_sigcode = (void *)va; 2069 return 0; 2070 } 2071 2072 /* 2073 * Release a refcount on spawn_exec_data and destroy memory, if this 2074 * was the last one. 2075 */ 2076 static void 2077 spawn_exec_data_release(struct spawn_exec_data *data) 2078 { 2079 if (atomic_dec_32_nv(&data->sed_refcnt) != 0) 2080 return; 2081 2082 cv_destroy(&data->sed_cv_child_ready); 2083 mutex_destroy(&data->sed_mtx_child); 2084 2085 if (data->sed_actions) 2086 posix_spawn_fa_free(data->sed_actions, 2087 data->sed_actions->len); 2088 if (data->sed_attrs) 2089 kmem_free(data->sed_attrs, 2090 sizeof(*data->sed_attrs)); 2091 kmem_free(data, sizeof(*data)); 2092 } 2093 2094 static int 2095 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions) 2096 { 2097 struct lwp *l = curlwp; 2098 register_t retval; 2099 int error, newfd; 2100 2101 if (actions == NULL) 2102 return 0; 2103 2104 for (size_t i = 0; i < actions->len; i++) { 2105 const struct posix_spawn_file_actions_entry *fae = 2106 &actions->fae[i]; 2107 switch (fae->fae_action) { 2108 case FAE_OPEN: 2109 if (fd_getfile(fae->fae_fildes) != NULL) { 2110 error = fd_close(fae->fae_fildes); 2111 if (error) 2112 return error; 2113 } 2114 error = fd_open(fae->fae_path, fae->fae_oflag, 2115 fae->fae_mode, &newfd); 2116 if (error) 2117 return error; 2118 if (newfd != fae->fae_fildes) { 2119 error = dodup(l, newfd, 2120 fae->fae_fildes, 0, &retval); 2121 if (fd_getfile(newfd) != NULL) 2122 fd_close(newfd); 2123 } 2124 break; 2125 case FAE_DUP2: 2126 error = dodup(l, fae->fae_fildes, 2127 fae->fae_newfildes, 0, &retval); 2128 break; 2129 case FAE_CLOSE: 2130 if (fd_getfile(fae->fae_fildes) == NULL) { 2131 return EBADF; 2132 } 2133 error = fd_close(fae->fae_fildes); 2134 break; 2135 } 2136 if (error) 2137 return error; 2138 } 2139 return 0; 2140 } 2141 2142 static int 2143 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent) 2144 { 2145 struct sigaction sigact; 2146 int error; 2147 struct proc *p = curproc; 2148 struct lwp *l = curlwp; 2149 2150 if (attrs == NULL) 2151 return 0; 2152 2153 memset(&sigact, 0, sizeof(sigact)); 2154 sigact._sa_u._sa_handler = SIG_DFL; 2155 sigact.sa_flags = 0; 2156 2157 /* 2158 * set state to SSTOP so that this proc can be found by pid. 2159 * see proc_enterprp, do_sched_setparam below 2160 */ 2161 mutex_enter(&proc_lock); 2162 /* 2163 * p_stat should be SACTIVE, so we need to adjust the 2164 * parent's p_nstopchild here. For safety, just make 2165 * we're on the good side of SDEAD before we adjust. 2166 */ 2167 int ostat = p->p_stat; 2168 KASSERT(ostat < SSTOP); 2169 p->p_stat = SSTOP; 2170 p->p_waited = 0; 2171 p->p_pptr->p_nstopchild++; 2172 mutex_exit(&proc_lock); 2173 2174 /* Set process group */ 2175 if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) { 2176 pid_t mypid = p->p_pid; 2177 pid_t pgrp = attrs->sa_pgroup; 2178 2179 if (pgrp == 0) 2180 pgrp = mypid; 2181 2182 error = proc_enterpgrp(parent, mypid, pgrp, false); 2183 if (error) 2184 goto out; 2185 } 2186 2187 /* Set scheduler policy */ 2188 if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER) 2189 error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy, 2190 &attrs->sa_schedparam); 2191 else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) { 2192 error = do_sched_setparam(parent->p_pid, 0, 2193 SCHED_NONE, &attrs->sa_schedparam); 2194 } 2195 if (error) 2196 goto out; 2197 2198 /* Reset user ID's */ 2199 if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) { 2200 error = do_setresgid(l, -1, kauth_cred_getgid(l->l_cred), -1, 2201 ID_E_EQ_R | ID_E_EQ_S); 2202 if (error) 2203 return error; 2204 error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1, 2205 ID_E_EQ_R | ID_E_EQ_S); 2206 if (error) 2207 goto out; 2208 } 2209 2210 /* Set signal masks/defaults */ 2211 if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) { 2212 mutex_enter(p->p_lock); 2213 error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL); 2214 mutex_exit(p->p_lock); 2215 if (error) 2216 goto out; 2217 } 2218 2219 if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) { 2220 /* 2221 * The following sigaction call is using a sigaction 2222 * version 0 trampoline which is in the compatibility 2223 * code only. This is not a problem because for SIG_DFL 2224 * and SIG_IGN, the trampolines are now ignored. If they 2225 * were not, this would be a problem because we are 2226 * holding the exec_lock, and the compat code needs 2227 * to do the same in order to replace the trampoline 2228 * code of the process. 2229 */ 2230 for (int i = 1; i <= NSIG; i++) { 2231 if (sigismember(&attrs->sa_sigdefault, i)) 2232 sigaction1(l, i, &sigact, NULL, NULL, 0); 2233 } 2234 } 2235 error = 0; 2236 out: 2237 mutex_enter(&proc_lock); 2238 p->p_stat = ostat; 2239 p->p_pptr->p_nstopchild--; 2240 mutex_exit(&proc_lock); 2241 return error; 2242 } 2243 2244 /* 2245 * A child lwp of a posix_spawn operation starts here and ends up in 2246 * cpu_spawn_return, dealing with all filedescriptor and scheduler 2247 * manipulations in between. 2248 * The parent waits for the child, as it is not clear whether the child 2249 * will be able to acquire its own exec_lock. If it can, the parent can 2250 * be released early and continue running in parallel. If not (or if the 2251 * magic debug flag is passed in the scheduler attribute struct), the 2252 * child rides on the parent's exec lock until it is ready to return to 2253 * to userland - and only then releases the parent. This method loses 2254 * concurrency, but improves error reporting. 2255 */ 2256 static void 2257 spawn_return(void *arg) 2258 { 2259 struct spawn_exec_data *spawn_data = arg; 2260 struct lwp *l = curlwp; 2261 struct proc *p = l->l_proc; 2262 int error; 2263 bool have_reflock; 2264 bool parent_is_waiting = true; 2265 2266 /* 2267 * Check if we can release parent early. 2268 * We either need to have no sed_attrs, or sed_attrs does not 2269 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require 2270 * safe access to the parent proc (passed in sed_parent). 2271 * We then try to get the exec_lock, and only if that works, we can 2272 * release the parent here already. 2273 */ 2274 struct posix_spawnattr *attrs = spawn_data->sed_attrs; 2275 if ((!attrs || (attrs->sa_flags 2276 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0) 2277 && rw_tryenter(&exec_lock, RW_READER)) { 2278 parent_is_waiting = false; 2279 mutex_enter(&spawn_data->sed_mtx_child); 2280 cv_signal(&spawn_data->sed_cv_child_ready); 2281 mutex_exit(&spawn_data->sed_mtx_child); 2282 } 2283 2284 /* don't allow debugger access yet */ 2285 rw_enter(&p->p_reflock, RW_WRITER); 2286 have_reflock = true; 2287 2288 /* handle posix_spawnattr */ 2289 error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent); 2290 if (error) 2291 goto report_error; 2292 2293 /* handle posix_spawn_file_actions */ 2294 error = handle_posix_spawn_file_actions(spawn_data->sed_actions); 2295 if (error) 2296 goto report_error; 2297 2298 /* now do the real exec */ 2299 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting, 2300 true); 2301 have_reflock = false; 2302 if (error == EJUSTRETURN) 2303 error = 0; 2304 else if (error) 2305 goto report_error; 2306 2307 if (parent_is_waiting) { 2308 mutex_enter(&spawn_data->sed_mtx_child); 2309 cv_signal(&spawn_data->sed_cv_child_ready); 2310 mutex_exit(&spawn_data->sed_mtx_child); 2311 } 2312 2313 /* release our refcount on the data */ 2314 spawn_exec_data_release(spawn_data); 2315 2316 if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) == 2317 (PSL_TRACED|PSL_TRACEDCHILD)) { 2318 eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN); 2319 } 2320 2321 /* and finally: leave to userland for the first time */ 2322 cpu_spawn_return(l); 2323 2324 /* NOTREACHED */ 2325 return; 2326 2327 report_error: 2328 if (have_reflock) { 2329 /* 2330 * We have not passed through execve_runproc(), 2331 * which would have released the p_reflock and also 2332 * taken ownership of the sed_exec part of spawn_data, 2333 * so release/free both here. 2334 */ 2335 rw_exit(&p->p_reflock); 2336 execve_free_data(&spawn_data->sed_exec); 2337 } 2338 2339 if (parent_is_waiting) { 2340 /* pass error to parent */ 2341 mutex_enter(&spawn_data->sed_mtx_child); 2342 spawn_data->sed_error = error; 2343 cv_signal(&spawn_data->sed_cv_child_ready); 2344 mutex_exit(&spawn_data->sed_mtx_child); 2345 } else { 2346 rw_exit(&exec_lock); 2347 } 2348 2349 /* release our refcount on the data */ 2350 spawn_exec_data_release(spawn_data); 2351 2352 /* done, exit */ 2353 mutex_enter(p->p_lock); 2354 /* 2355 * Posix explicitly asks for an exit code of 127 if we report 2356 * errors from the child process - so, unfortunately, there 2357 * is no way to report a more exact error code. 2358 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as 2359 * flag bit in the attrp argument to posix_spawn(2), see above. 2360 */ 2361 exit1(l, 127, 0); 2362 } 2363 2364 void 2365 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len) 2366 { 2367 2368 for (size_t i = 0; i < len; i++) { 2369 struct posix_spawn_file_actions_entry *fae = &fa->fae[i]; 2370 if (fae->fae_action != FAE_OPEN) 2371 continue; 2372 kmem_strfree(fae->fae_path); 2373 } 2374 if (fa->len > 0) 2375 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len); 2376 kmem_free(fa, sizeof(*fa)); 2377 } 2378 2379 static int 2380 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap, 2381 const struct posix_spawn_file_actions *ufa, rlim_t lim) 2382 { 2383 struct posix_spawn_file_actions *fa; 2384 struct posix_spawn_file_actions_entry *fae; 2385 char *pbuf = NULL; 2386 int error; 2387 size_t i = 0; 2388 2389 fa = kmem_alloc(sizeof(*fa), KM_SLEEP); 2390 error = copyin(ufa, fa, sizeof(*fa)); 2391 if (error || fa->len == 0) { 2392 kmem_free(fa, sizeof(*fa)); 2393 return error; /* 0 if not an error, and len == 0 */ 2394 } 2395 2396 if (fa->len > lim) { 2397 kmem_free(fa, sizeof(*fa)); 2398 return EINVAL; 2399 } 2400 2401 fa->size = fa->len; 2402 size_t fal = fa->len * sizeof(*fae); 2403 fae = fa->fae; 2404 fa->fae = kmem_alloc(fal, KM_SLEEP); 2405 error = copyin(fae, fa->fae, fal); 2406 if (error) 2407 goto out; 2408 2409 pbuf = PNBUF_GET(); 2410 for (; i < fa->len; i++) { 2411 fae = &fa->fae[i]; 2412 if (fae->fae_action != FAE_OPEN) 2413 continue; 2414 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal); 2415 if (error) 2416 goto out; 2417 fae->fae_path = kmem_alloc(fal, KM_SLEEP); 2418 memcpy(fae->fae_path, pbuf, fal); 2419 } 2420 PNBUF_PUT(pbuf); 2421 2422 *fap = fa; 2423 return 0; 2424 out: 2425 if (pbuf) 2426 PNBUF_PUT(pbuf); 2427 posix_spawn_fa_free(fa, i); 2428 return error; 2429 } 2430 2431 /* 2432 * N.B. increments nprocs upon success. Callers need to drop nprocs if 2433 * they fail for some other reason. 2434 */ 2435 int 2436 check_posix_spawn(struct lwp *l1) 2437 { 2438 int error, tnprocs, count; 2439 uid_t uid; 2440 struct proc *p1; 2441 2442 p1 = l1->l_proc; 2443 uid = kauth_cred_getuid(l1->l_cred); 2444 tnprocs = atomic_inc_uint_nv(&nprocs); 2445 2446 /* 2447 * Although process entries are dynamically created, we still keep 2448 * a global limit on the maximum number we will create. 2449 */ 2450 if (__predict_false(tnprocs >= maxproc)) 2451 error = -1; 2452 else 2453 error = kauth_authorize_process(l1->l_cred, 2454 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL); 2455 2456 if (error) { 2457 atomic_dec_uint(&nprocs); 2458 return EAGAIN; 2459 } 2460 2461 /* 2462 * Enforce limits. 2463 */ 2464 count = chgproccnt(uid, 1); 2465 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT, 2466 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 2467 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 && 2468 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 2469 (void)chgproccnt(uid, -1); 2470 atomic_dec_uint(&nprocs); 2471 return EAGAIN; 2472 } 2473 2474 return 0; 2475 } 2476 2477 int 2478 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path, 2479 struct posix_spawn_file_actions *fa, 2480 struct posix_spawnattr *sa, 2481 char *const *argv, char *const *envp, 2482 execve_fetch_element_t fetch) 2483 { 2484 2485 struct proc *p1, *p2; 2486 struct lwp *l2; 2487 int error; 2488 struct spawn_exec_data *spawn_data; 2489 vaddr_t uaddr; 2490 pid_t pid; 2491 bool have_exec_lock = false; 2492 2493 p1 = l1->l_proc; 2494 2495 /* Allocate and init spawn_data */ 2496 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP); 2497 spawn_data->sed_refcnt = 1; /* only parent so far */ 2498 cv_init(&spawn_data->sed_cv_child_ready, "pspawn"); 2499 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE); 2500 mutex_enter(&spawn_data->sed_mtx_child); 2501 2502 /* 2503 * Do the first part of the exec now, collect state 2504 * in spawn_data. 2505 */ 2506 error = execve_loadvm(l1, true, path, -1, argv, 2507 envp, fetch, &spawn_data->sed_exec); 2508 if (error == EJUSTRETURN) 2509 error = 0; 2510 else if (error) 2511 goto error_exit; 2512 2513 have_exec_lock = true; 2514 2515 /* 2516 * Allocate virtual address space for the U-area now, while it 2517 * is still easy to abort the fork operation if we're out of 2518 * kernel virtual address space. 2519 */ 2520 uaddr = uvm_uarea_alloc(); 2521 if (__predict_false(uaddr == 0)) { 2522 error = ENOMEM; 2523 goto error_exit; 2524 } 2525 2526 /* 2527 * Allocate new proc. Borrow proc0 vmspace for it, we will 2528 * replace it with its own before returning to userland 2529 * in the child. 2530 */ 2531 p2 = proc_alloc(); 2532 if (p2 == NULL) { 2533 /* We were unable to allocate a process ID. */ 2534 error = EAGAIN; 2535 goto error_exit; 2536 } 2537 2538 /* 2539 * This is a point of no return, we will have to go through 2540 * the child proc to properly clean it up past this point. 2541 */ 2542 pid = p2->p_pid; 2543 2544 /* 2545 * Make a proc table entry for the new process. 2546 * Start by zeroing the section of proc that is zero-initialized, 2547 * then copy the section that is copied directly from the parent. 2548 */ 2549 memset(&p2->p_startzero, 0, 2550 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero)); 2551 memcpy(&p2->p_startcopy, &p1->p_startcopy, 2552 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy)); 2553 p2->p_vmspace = proc0.p_vmspace; 2554 2555 TAILQ_INIT(&p2->p_sigpend.sp_info); 2556 2557 LIST_INIT(&p2->p_lwps); 2558 LIST_INIT(&p2->p_sigwaiters); 2559 2560 /* 2561 * Duplicate sub-structures as needed. 2562 * Increase reference counts on shared objects. 2563 * Inherit flags we want to keep. The flags related to SIGCHLD 2564 * handling are important in order to keep a consistent behaviour 2565 * for the child after the fork. If we are a 32-bit process, the 2566 * child will be too. 2567 */ 2568 p2->p_flag = 2569 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32); 2570 p2->p_emul = p1->p_emul; 2571 p2->p_execsw = p1->p_execsw; 2572 2573 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH); 2574 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE); 2575 rw_init(&p2->p_reflock); 2576 cv_init(&p2->p_waitcv, "wait"); 2577 cv_init(&p2->p_lwpcv, "lwpwait"); 2578 2579 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 2580 2581 kauth_proc_fork(p1, p2); 2582 2583 p2->p_raslist = NULL; 2584 p2->p_fd = fd_copy(); 2585 2586 /* XXX racy */ 2587 p2->p_mqueue_cnt = p1->p_mqueue_cnt; 2588 2589 p2->p_cwdi = cwdinit(); 2590 2591 /* 2592 * Note: p_limit (rlimit stuff) is copy-on-write, so normally 2593 * we just need increase pl_refcnt. 2594 */ 2595 if (!p1->p_limit->pl_writeable) { 2596 lim_addref(p1->p_limit); 2597 p2->p_limit = p1->p_limit; 2598 } else { 2599 p2->p_limit = lim_copy(p1->p_limit); 2600 } 2601 2602 p2->p_lflag = 0; 2603 l1->l_vforkwaiting = false; 2604 p2->p_sflag = 0; 2605 p2->p_slflag = 0; 2606 p2->p_pptr = p1; 2607 p2->p_ppid = p1->p_pid; 2608 LIST_INIT(&p2->p_children); 2609 2610 p2->p_aio = NULL; 2611 2612 #ifdef KTRACE 2613 /* 2614 * Copy traceflag and tracefile if enabled. 2615 * If not inherited, these were zeroed above. 2616 */ 2617 if (p1->p_traceflag & KTRFAC_INHERIT) { 2618 mutex_enter(&ktrace_lock); 2619 p2->p_traceflag = p1->p_traceflag; 2620 if ((p2->p_tracep = p1->p_tracep) != NULL) 2621 ktradref(p2); 2622 mutex_exit(&ktrace_lock); 2623 } 2624 #endif 2625 2626 /* 2627 * Create signal actions for the child process. 2628 */ 2629 p2->p_sigacts = sigactsinit(p1, 0); 2630 mutex_enter(p1->p_lock); 2631 p2->p_sflag |= 2632 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 2633 sched_proc_fork(p1, p2); 2634 mutex_exit(p1->p_lock); 2635 2636 p2->p_stflag = p1->p_stflag; 2637 2638 /* 2639 * p_stats. 2640 * Copy parts of p_stats, and zero out the rest. 2641 */ 2642 p2->p_stats = pstatscopy(p1->p_stats); 2643 2644 /* copy over machdep flags to the new proc */ 2645 cpu_proc_fork(p1, p2); 2646 2647 /* 2648 * Prepare remaining parts of spawn data 2649 */ 2650 spawn_data->sed_actions = fa; 2651 spawn_data->sed_attrs = sa; 2652 2653 spawn_data->sed_parent = p1; 2654 2655 /* create LWP */ 2656 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data, 2657 &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk); 2658 l2->l_ctxlink = NULL; /* reset ucontext link */ 2659 2660 /* 2661 * Copy the credential so other references don't see our changes. 2662 * Test to see if this is necessary first, since in the common case 2663 * we won't need a private reference. 2664 */ 2665 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) || 2666 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) { 2667 l2->l_cred = kauth_cred_copy(l2->l_cred); 2668 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred)); 2669 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred)); 2670 } 2671 2672 /* Update the master credentials. */ 2673 if (l2->l_cred != p2->p_cred) { 2674 kauth_cred_t ocred; 2675 2676 kauth_cred_hold(l2->l_cred); 2677 mutex_enter(p2->p_lock); 2678 ocred = p2->p_cred; 2679 p2->p_cred = l2->l_cred; 2680 mutex_exit(p2->p_lock); 2681 kauth_cred_free(ocred); 2682 } 2683 2684 *child_ok = true; 2685 spawn_data->sed_refcnt = 2; /* child gets it as well */ 2686 #if 0 2687 l2->l_nopreempt = 1; /* start it non-preemptable */ 2688 #endif 2689 2690 /* 2691 * It's now safe for the scheduler and other processes to see the 2692 * child process. 2693 */ 2694 mutex_enter(&proc_lock); 2695 2696 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 2697 p2->p_lflag |= PL_CONTROLT; 2698 2699 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 2700 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */ 2701 2702 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) == 2703 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2704 proc_changeparent(p2, p1->p_pptr); 2705 SET(p2->p_slflag, PSL_TRACEDCHILD); 2706 } 2707 2708 p2->p_oppid = p1->p_pid; /* Remember the original parent id. */ 2709 2710 LIST_INSERT_AFTER(p1, p2, p_pglist); 2711 LIST_INSERT_HEAD(&allproc, p2, p_list); 2712 2713 p2->p_trace_enabled = trace_is_enabled(p2); 2714 #ifdef __HAVE_SYSCALL_INTERN 2715 (*p2->p_emul->e_syscall_intern)(p2); 2716 #endif 2717 2718 /* 2719 * Make child runnable, set start time, and add to run queue except 2720 * if the parent requested the child to start in SSTOP state. 2721 */ 2722 mutex_enter(p2->p_lock); 2723 2724 getmicrotime(&p2->p_stats->p_start); 2725 2726 lwp_lock(l2); 2727 KASSERT(p2->p_nrlwps == 1); 2728 KASSERT(l2->l_stat == LSIDL); 2729 p2->p_nrlwps = 1; 2730 p2->p_stat = SACTIVE; 2731 setrunnable(l2); 2732 /* LWP now unlocked */ 2733 2734 mutex_exit(p2->p_lock); 2735 mutex_exit(&proc_lock); 2736 2737 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child); 2738 error = spawn_data->sed_error; 2739 mutex_exit(&spawn_data->sed_mtx_child); 2740 spawn_exec_data_release(spawn_data); 2741 2742 rw_exit(&p1->p_reflock); 2743 rw_exit(&exec_lock); 2744 have_exec_lock = false; 2745 2746 *pid_res = pid; 2747 2748 if (error) 2749 return error; 2750 2751 if (p1->p_slflag & PSL_TRACED) { 2752 /* Paranoid check */ 2753 mutex_enter(&proc_lock); 2754 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) != 2755 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2756 mutex_exit(&proc_lock); 2757 return 0; 2758 } 2759 2760 mutex_enter(p1->p_lock); 2761 eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid); 2762 } 2763 return 0; 2764 2765 error_exit: 2766 if (have_exec_lock) { 2767 execve_free_data(&spawn_data->sed_exec); 2768 rw_exit(&p1->p_reflock); 2769 rw_exit(&exec_lock); 2770 } 2771 mutex_exit(&spawn_data->sed_mtx_child); 2772 spawn_exec_data_release(spawn_data); 2773 2774 return error; 2775 } 2776 2777 int 2778 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap, 2779 register_t *retval) 2780 { 2781 /* { 2782 syscallarg(pid_t *) pid; 2783 syscallarg(const char *) path; 2784 syscallarg(const struct posix_spawn_file_actions *) file_actions; 2785 syscallarg(const struct posix_spawnattr *) attrp; 2786 syscallarg(char *const *) argv; 2787 syscallarg(char *const *) envp; 2788 } */ 2789 2790 int error; 2791 struct posix_spawn_file_actions *fa = NULL; 2792 struct posix_spawnattr *sa = NULL; 2793 pid_t pid; 2794 bool child_ok = false; 2795 rlim_t max_fileactions; 2796 proc_t *p = l1->l_proc; 2797 2798 /* check_posix_spawn() increments nprocs for us. */ 2799 error = check_posix_spawn(l1); 2800 if (error) { 2801 *retval = error; 2802 return 0; 2803 } 2804 2805 /* copy in file_actions struct */ 2806 if (SCARG(uap, file_actions) != NULL) { 2807 max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur, 2808 maxfiles); 2809 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions), 2810 max_fileactions); 2811 if (error) 2812 goto error_exit; 2813 } 2814 2815 /* copyin posix_spawnattr struct */ 2816 if (SCARG(uap, attrp) != NULL) { 2817 sa = kmem_alloc(sizeof(*sa), KM_SLEEP); 2818 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa)); 2819 if (error) 2820 goto error_exit; 2821 } 2822 2823 /* 2824 * Do the spawn 2825 */ 2826 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa, 2827 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element); 2828 if (error) 2829 goto error_exit; 2830 2831 if (error == 0 && SCARG(uap, pid) != NULL) 2832 error = copyout(&pid, SCARG(uap, pid), sizeof(pid)); 2833 2834 *retval = error; 2835 return 0; 2836 2837 error_exit: 2838 if (!child_ok) { 2839 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1); 2840 atomic_dec_uint(&nprocs); 2841 2842 if (sa) 2843 kmem_free(sa, sizeof(*sa)); 2844 if (fa) 2845 posix_spawn_fa_free(fa, fa->len); 2846 } 2847 2848 *retval = error; 2849 return 0; 2850 } 2851 2852 void 2853 exec_free_emul_arg(struct exec_package *epp) 2854 { 2855 if (epp->ep_emul_arg_free != NULL) { 2856 KASSERT(epp->ep_emul_arg != NULL); 2857 (*epp->ep_emul_arg_free)(epp->ep_emul_arg); 2858 epp->ep_emul_arg_free = NULL; 2859 epp->ep_emul_arg = NULL; 2860 } else { 2861 KASSERT(epp->ep_emul_arg == NULL); 2862 } 2863 } 2864 2865 #ifdef DEBUG_EXEC 2866 static void 2867 dump_vmcmds(const struct exec_package * const epp, size_t x, int error) 2868 { 2869 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0]; 2870 size_t j; 2871 2872 if (error == 0) 2873 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used)); 2874 else 2875 DPRINTF(("vmcmds %zu/%u, error %d\n", x, 2876 epp->ep_vmcmds.evs_used, error)); 2877 2878 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) { 2879 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#" 2880 PRIxVADDR"/%#"PRIxVSIZE" fd@%#" 2881 PRIxVSIZE" prot=0%o flags=%d\n", j, 2882 vp[j].ev_proc == vmcmd_map_pagedvn ? 2883 "pagedvn" : 2884 vp[j].ev_proc == vmcmd_map_readvn ? 2885 "readvn" : 2886 vp[j].ev_proc == vmcmd_map_zero ? 2887 "zero" : "*unknown*", 2888 vp[j].ev_addr, vp[j].ev_len, 2889 vp[j].ev_offset, vp[j].ev_prot, 2890 vp[j].ev_flags)); 2891 if (error != 0 && j == x) 2892 DPRINTF((" ^--- failed\n")); 2893 } 2894 } 2895 #endif 2896