1 /* $NetBSD: kern_exec.c,v 1.500 2020/05/07 20:02:34 kamil 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 1.500 2020/05/07 20:02:34 kamil 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/ktrace.h> 91 #include <sys/uidinfo.h> 92 #include <sys/wait.h> 93 #include <sys/mman.h> 94 #include <sys/ras.h> 95 #include <sys/signalvar.h> 96 #include <sys/stat.h> 97 #include <sys/syscall.h> 98 #include <sys/kauth.h> 99 #include <sys/lwpctl.h> 100 #include <sys/pax.h> 101 #include <sys/cpu.h> 102 #include <sys/module.h> 103 #include <sys/syscallvar.h> 104 #include <sys/syscallargs.h> 105 #if NVERIEXEC > 0 106 #include <sys/verified_exec.h> 107 #endif /* NVERIEXEC > 0 */ 108 #include <sys/sdt.h> 109 #include <sys/spawn.h> 110 #include <sys/prot.h> 111 #include <sys/cprng.h> 112 113 #include <uvm/uvm_extern.h> 114 115 #include <machine/reg.h> 116 117 #include <compat/common/compat_util.h> 118 119 #ifndef MD_TOPDOWN_INIT 120 #ifdef __USE_TOPDOWN_VM 121 #define MD_TOPDOWN_INIT(epp) (epp)->ep_flags |= EXEC_TOPDOWN_VM 122 #else 123 #define MD_TOPDOWN_INIT(epp) 124 #endif 125 #endif 126 127 struct execve_data; 128 129 extern int user_va0_disable; 130 131 static size_t calcargs(struct execve_data * restrict, const size_t); 132 static size_t calcstack(struct execve_data * restrict, const size_t); 133 static int copyoutargs(struct execve_data * restrict, struct lwp *, 134 char * const); 135 static int copyoutpsstrs(struct execve_data * restrict, struct proc *); 136 static int copyinargs(struct execve_data * restrict, char * const *, 137 char * const *, execve_fetch_element_t, char **); 138 static int copyinargstrs(struct execve_data * restrict, char * const *, 139 execve_fetch_element_t, char **, size_t *, void (*)(const void *, size_t)); 140 static int exec_sigcode_map(struct proc *, const struct emul *); 141 142 #if defined(DEBUG) && !defined(DEBUG_EXEC) 143 #define DEBUG_EXEC 144 #endif 145 #ifdef DEBUG_EXEC 146 #define DPRINTF(a) printf a 147 #define COPYPRINTF(s, a, b) printf("%s, %d: copyout%s @%p %zu\n", __func__, \ 148 __LINE__, (s), (a), (b)) 149 static void dump_vmcmds(const struct exec_package * const, size_t, int); 150 #define DUMPVMCMDS(p, x, e) do { dump_vmcmds((p), (x), (e)); } while (0) 151 #else 152 #define DPRINTF(a) 153 #define COPYPRINTF(s, a, b) 154 #define DUMPVMCMDS(p, x, e) do {} while (0) 155 #endif /* DEBUG_EXEC */ 156 157 /* 158 * DTrace SDT provider definitions 159 */ 160 SDT_PROVIDER_DECLARE(proc); 161 SDT_PROBE_DEFINE1(proc, kernel, , exec, "char *"); 162 SDT_PROBE_DEFINE1(proc, kernel, , exec__success, "char *"); 163 SDT_PROBE_DEFINE1(proc, kernel, , exec__failure, "int"); 164 165 /* 166 * Exec function switch: 167 * 168 * Note that each makecmds function is responsible for loading the 169 * exec package with the necessary functions for any exec-type-specific 170 * handling. 171 * 172 * Functions for specific exec types should be defined in their own 173 * header file. 174 */ 175 static const struct execsw **execsw = NULL; 176 static int nexecs; 177 178 u_int exec_maxhdrsz; /* must not be static - used by netbsd32 */ 179 180 /* list of dynamically loaded execsw entries */ 181 static LIST_HEAD(execlist_head, exec_entry) ex_head = 182 LIST_HEAD_INITIALIZER(ex_head); 183 struct exec_entry { 184 LIST_ENTRY(exec_entry) ex_list; 185 SLIST_ENTRY(exec_entry) ex_slist; 186 const struct execsw *ex_sw; 187 }; 188 189 #ifndef __HAVE_SYSCALL_INTERN 190 void syscall(void); 191 #endif 192 193 /* NetBSD autoloadable syscalls */ 194 #ifdef MODULAR 195 #include <kern/syscalls_autoload.c> 196 #endif 197 198 /* NetBSD emul struct */ 199 struct emul emul_netbsd = { 200 .e_name = "netbsd", 201 #ifdef EMUL_NATIVEROOT 202 .e_path = EMUL_NATIVEROOT, 203 #else 204 .e_path = NULL, 205 #endif 206 #ifndef __HAVE_MINIMAL_EMUL 207 .e_flags = EMUL_HAS_SYS___syscall, 208 .e_errno = NULL, 209 .e_nosys = SYS_syscall, 210 .e_nsysent = SYS_NSYSENT, 211 #endif 212 #ifdef MODULAR 213 .e_sc_autoload = netbsd_syscalls_autoload, 214 #endif 215 .e_sysent = sysent, 216 .e_nomodbits = sysent_nomodbits, 217 #ifdef SYSCALL_DEBUG 218 .e_syscallnames = syscallnames, 219 #else 220 .e_syscallnames = NULL, 221 #endif 222 .e_sendsig = sendsig, 223 .e_trapsignal = trapsignal, 224 .e_sigcode = NULL, 225 .e_esigcode = NULL, 226 .e_sigobject = NULL, 227 .e_setregs = setregs, 228 .e_proc_exec = NULL, 229 .e_proc_fork = NULL, 230 .e_proc_exit = NULL, 231 .e_lwp_fork = NULL, 232 .e_lwp_exit = NULL, 233 #ifdef __HAVE_SYSCALL_INTERN 234 .e_syscall_intern = syscall_intern, 235 #else 236 .e_syscall = syscall, 237 #endif 238 .e_sysctlovly = NULL, 239 .e_vm_default_addr = uvm_default_mapaddr, 240 .e_usertrap = NULL, 241 .e_ucsize = sizeof(ucontext_t), 242 .e_startlwp = startlwp 243 }; 244 245 /* 246 * Exec lock. Used to control access to execsw[] structures. 247 * This must not be static so that netbsd32 can access it, too. 248 */ 249 krwlock_t exec_lock __cacheline_aligned; 250 251 static kmutex_t sigobject_lock __cacheline_aligned; 252 253 /* 254 * Data used between a loadvm and execve part of an "exec" operation 255 */ 256 struct execve_data { 257 struct exec_package ed_pack; 258 struct pathbuf *ed_pathbuf; 259 struct vattr ed_attr; 260 struct ps_strings ed_arginfo; 261 char *ed_argp; 262 const char *ed_pathstring; 263 char *ed_resolvedname; 264 size_t ed_ps_strings_sz; 265 int ed_szsigcode; 266 size_t ed_argslen; 267 long ed_argc; 268 long ed_envc; 269 }; 270 271 /* 272 * data passed from parent lwp to child during a posix_spawn() 273 */ 274 struct spawn_exec_data { 275 struct execve_data sed_exec; 276 struct posix_spawn_file_actions 277 *sed_actions; 278 struct posix_spawnattr *sed_attrs; 279 struct proc *sed_parent; 280 kcondvar_t sed_cv_child_ready; 281 kmutex_t sed_mtx_child; 282 int sed_error; 283 volatile uint32_t sed_refcnt; 284 }; 285 286 static struct vm_map *exec_map; 287 static struct pool exec_pool; 288 289 static void * 290 exec_pool_alloc(struct pool *pp, int flags) 291 { 292 293 return (void *)uvm_km_alloc(exec_map, NCARGS, 0, 294 UVM_KMF_PAGEABLE | UVM_KMF_WAITVA); 295 } 296 297 static void 298 exec_pool_free(struct pool *pp, void *addr) 299 { 300 301 uvm_km_free(exec_map, (vaddr_t)addr, NCARGS, UVM_KMF_PAGEABLE); 302 } 303 304 static struct pool_allocator exec_palloc = { 305 .pa_alloc = exec_pool_alloc, 306 .pa_free = exec_pool_free, 307 .pa_pagesz = NCARGS 308 }; 309 310 static void 311 exec_path_free(struct execve_data *data) 312 { 313 pathbuf_stringcopy_put(data->ed_pathbuf, data->ed_pathstring); 314 pathbuf_destroy(data->ed_pathbuf); 315 if (data->ed_resolvedname) 316 PNBUF_PUT(data->ed_resolvedname); 317 } 318 319 static int 320 exec_resolvename(struct lwp *l, struct exec_package *epp, struct vnode *vp, 321 char **rpath) 322 { 323 int error; 324 char *p; 325 326 KASSERT(rpath != NULL); 327 328 *rpath = PNBUF_GET(); 329 error = vnode_to_path(*rpath, MAXPATHLEN, vp, l, l->l_proc); 330 if (error) { 331 DPRINTF(("%s: can't resolve name for %s, error %d\n", 332 __func__, epp->ep_kname, error)); 333 PNBUF_PUT(*rpath); 334 *rpath = NULL; 335 return error; 336 } 337 epp->ep_resolvedname = *rpath; 338 if ((p = strrchr(*rpath, '/')) != NULL) 339 epp->ep_kname = p + 1; 340 return 0; 341 } 342 343 344 /* 345 * check exec: 346 * given an "executable" described in the exec package's namei info, 347 * see what we can do with it. 348 * 349 * ON ENTRY: 350 * exec package with appropriate namei info 351 * lwp pointer of exec'ing lwp 352 * NO SELF-LOCKED VNODES 353 * 354 * ON EXIT: 355 * error: nothing held, etc. exec header still allocated. 356 * ok: filled exec package, executable's vnode (unlocked). 357 * 358 * EXEC SWITCH ENTRY: 359 * Locked vnode to check, exec package, proc. 360 * 361 * EXEC SWITCH EXIT: 362 * ok: return 0, filled exec package, executable's vnode (unlocked). 363 * error: destructive: 364 * everything deallocated execept exec header. 365 * non-destructive: 366 * error code, executable's vnode (unlocked), 367 * exec header unmodified. 368 */ 369 int 370 /*ARGSUSED*/ 371 check_exec(struct lwp *l, struct exec_package *epp, struct pathbuf *pb, 372 char **rpath) 373 { 374 int error, i; 375 struct vnode *vp; 376 size_t resid; 377 378 if (epp->ep_resolvedname) { 379 struct nameidata nd; 380 381 // grab the absolute pathbuf here before namei() trashes it. 382 pathbuf_copystring(pb, epp->ep_resolvedname, PATH_MAX); 383 NDINIT(&nd, LOOKUP, FOLLOW | LOCKLEAF | TRYEMULROOT, pb); 384 385 /* first get the vnode */ 386 if ((error = namei(&nd)) != 0) 387 return error; 388 389 epp->ep_vp = vp = nd.ni_vp; 390 #ifdef DIAGNOSTIC 391 /* paranoia (take this out once namei stuff stabilizes) */ 392 memset(nd.ni_pnbuf, '~', PATH_MAX); 393 #endif 394 } else { 395 struct file *fp; 396 397 if ((error = fd_getvnode(epp->ep_xfd, &fp)) != 0) 398 return error; 399 epp->ep_vp = vp = fp->f_vnode; 400 vref(vp); 401 fd_putfile(epp->ep_xfd); 402 if ((error = exec_resolvename(l, epp, vp, rpath)) != 0) 403 return error; 404 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 405 } 406 407 /* check access and type */ 408 if (vp->v_type != VREG) { 409 error = EACCES; 410 goto bad1; 411 } 412 if ((error = VOP_ACCESS(vp, VEXEC, l->l_cred)) != 0) 413 goto bad1; 414 415 /* get attributes */ 416 /* XXX VOP_GETATTR is the only thing that needs LK_EXCLUSIVE here */ 417 if ((error = VOP_GETATTR(vp, epp->ep_vap, l->l_cred)) != 0) 418 goto bad1; 419 420 /* Check mount point */ 421 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 422 error = EACCES; 423 goto bad1; 424 } 425 if (vp->v_mount->mnt_flag & MNT_NOSUID) 426 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 427 428 /* try to open it */ 429 if ((error = VOP_OPEN(vp, FREAD, l->l_cred)) != 0) 430 goto bad1; 431 432 /* now we have the file, get the exec header */ 433 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 434 UIO_SYSSPACE, IO_NODELOCKED, l->l_cred, &resid, NULL); 435 if (error) 436 goto bad1; 437 438 /* unlock vp, since we need it unlocked from here on out. */ 439 VOP_UNLOCK(vp); 440 441 #if NVERIEXEC > 0 442 error = veriexec_verify(l, vp, 443 epp->ep_resolvedname ? epp->ep_resolvedname : epp->ep_kname, 444 epp->ep_flags & EXEC_INDIR ? VERIEXEC_INDIRECT : VERIEXEC_DIRECT, 445 NULL); 446 if (error) 447 goto bad2; 448 #endif /* NVERIEXEC > 0 */ 449 450 #ifdef PAX_SEGVGUARD 451 error = pax_segvguard(l, vp, epp->ep_resolvedname, false); 452 if (error) 453 goto bad2; 454 #endif /* PAX_SEGVGUARD */ 455 456 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 457 458 /* 459 * Set up default address space limits. Can be overridden 460 * by individual exec packages. 461 */ 462 epp->ep_vm_minaddr = exec_vm_minaddr(VM_MIN_ADDRESS); 463 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 464 465 /* 466 * set up the vmcmds for creation of the process 467 * address space 468 */ 469 error = ENOEXEC; 470 for (i = 0; i < nexecs; i++) { 471 int newerror; 472 473 epp->ep_esch = execsw[i]; 474 newerror = (*execsw[i]->es_makecmds)(l, epp); 475 476 if (!newerror) { 477 /* Seems ok: check that entry point is not too high */ 478 if (epp->ep_entry >= epp->ep_vm_maxaddr) { 479 #ifdef DIAGNOSTIC 480 printf("%s: rejecting %p due to " 481 "too high entry address (>= %p)\n", 482 __func__, (void *)epp->ep_entry, 483 (void *)epp->ep_vm_maxaddr); 484 #endif 485 error = ENOEXEC; 486 break; 487 } 488 /* Seems ok: check that entry point is not too low */ 489 if (epp->ep_entry < epp->ep_vm_minaddr) { 490 #ifdef DIAGNOSTIC 491 printf("%s: rejecting %p due to " 492 "too low entry address (< %p)\n", 493 __func__, (void *)epp->ep_entry, 494 (void *)epp->ep_vm_minaddr); 495 #endif 496 error = ENOEXEC; 497 break; 498 } 499 500 /* check limits */ 501 if ((epp->ep_tsize > MAXTSIZ) || 502 (epp->ep_dsize > (u_quad_t)l->l_proc->p_rlimit 503 [RLIMIT_DATA].rlim_cur)) { 504 #ifdef DIAGNOSTIC 505 printf("%s: rejecting due to " 506 "limits (t=%llu > %llu || d=%llu > %llu)\n", 507 __func__, 508 (unsigned long long)epp->ep_tsize, 509 (unsigned long long)MAXTSIZ, 510 (unsigned long long)epp->ep_dsize, 511 (unsigned long long) 512 l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur); 513 #endif 514 error = ENOMEM; 515 break; 516 } 517 return 0; 518 } 519 520 /* 521 * Reset all the fields that may have been modified by the 522 * loader. 523 */ 524 KASSERT(epp->ep_emul_arg == NULL); 525 if (epp->ep_emul_root != NULL) { 526 vrele(epp->ep_emul_root); 527 epp->ep_emul_root = NULL; 528 } 529 if (epp->ep_interp != NULL) { 530 vrele(epp->ep_interp); 531 epp->ep_interp = NULL; 532 } 533 epp->ep_pax_flags = 0; 534 535 /* make sure the first "interesting" error code is saved. */ 536 if (error == ENOEXEC) 537 error = newerror; 538 539 if (epp->ep_flags & EXEC_DESTR) 540 /* Error from "#!" code, tidied up by recursive call */ 541 return error; 542 } 543 544 /* not found, error */ 545 546 /* 547 * free any vmspace-creation commands, 548 * and release their references 549 */ 550 kill_vmcmds(&epp->ep_vmcmds); 551 552 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD) 553 bad2: 554 #endif 555 /* 556 * close and release the vnode, restore the old one, free the 557 * pathname buf, and punt. 558 */ 559 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 560 VOP_CLOSE(vp, FREAD, l->l_cred); 561 vput(vp); 562 return error; 563 564 bad1: 565 /* 566 * free the namei pathname buffer, and put the vnode 567 * (which we don't yet have open). 568 */ 569 vput(vp); /* was still locked */ 570 return error; 571 } 572 573 #ifdef __MACHINE_STACK_GROWS_UP 574 #define STACK_PTHREADSPACE NBPG 575 #else 576 #define STACK_PTHREADSPACE 0 577 #endif 578 579 static int 580 execve_fetch_element(char * const *array, size_t index, char **value) 581 { 582 return copyin(array + index, value, sizeof(*value)); 583 } 584 585 /* 586 * exec system call 587 */ 588 int 589 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval) 590 { 591 /* { 592 syscallarg(const char *) path; 593 syscallarg(char * const *) argp; 594 syscallarg(char * const *) envp; 595 } */ 596 597 return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp), 598 SCARG(uap, envp), execve_fetch_element); 599 } 600 601 int 602 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap, 603 register_t *retval) 604 { 605 /* { 606 syscallarg(int) fd; 607 syscallarg(char * const *) argp; 608 syscallarg(char * const *) envp; 609 } */ 610 611 return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp), 612 SCARG(uap, envp), execve_fetch_element); 613 } 614 615 /* 616 * Load modules to try and execute an image that we do not understand. 617 * If no execsw entries are present, we load those likely to be needed 618 * in order to run native images only. Otherwise, we autoload all 619 * possible modules that could let us run the binary. XXX lame 620 */ 621 static void 622 exec_autoload(void) 623 { 624 #ifdef MODULAR 625 static const char * const native[] = { 626 "exec_elf32", 627 "exec_elf64", 628 "exec_script", 629 NULL 630 }; 631 static const char * const compat[] = { 632 "exec_elf32", 633 "exec_elf64", 634 "exec_script", 635 "exec_aout", 636 "exec_coff", 637 "exec_ecoff", 638 "compat_aoutm68k", 639 "compat_netbsd32", 640 #if 0 641 "compat_linux", 642 "compat_linux32", 643 #endif 644 "compat_sunos", 645 "compat_sunos32", 646 "compat_ultrix", 647 NULL 648 }; 649 char const * const *list; 650 int i; 651 652 list = nexecs == 0 ? native : compat; 653 for (i = 0; list[i] != NULL; i++) { 654 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) { 655 continue; 656 } 657 yield(); 658 } 659 #endif 660 } 661 662 /* 663 * Copy the user or kernel supplied upath to the allocated pathbuffer pbp 664 * making it absolute in the process, by prepending the current working 665 * directory if it is not. If offs is supplied it will contain the offset 666 * where the original supplied copy of upath starts. 667 */ 668 int 669 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg, 670 struct pathbuf **pbp, size_t *offs) 671 { 672 char *path, *bp; 673 size_t len, tlen; 674 int error; 675 struct cwdinfo *cwdi; 676 677 path = PNBUF_GET(); 678 if (seg == UIO_SYSSPACE) { 679 error = copystr(upath, path, MAXPATHLEN, &len); 680 } else { 681 error = copyinstr(upath, path, MAXPATHLEN, &len); 682 } 683 if (error) 684 goto err; 685 686 if (path[0] == '/') { 687 if (offs) 688 *offs = 0; 689 goto out; 690 } 691 692 len++; 693 if (len + 1 >= MAXPATHLEN) { 694 error = ENAMETOOLONG; 695 goto err; 696 } 697 bp = path + MAXPATHLEN - len; 698 memmove(bp, path, len); 699 *(--bp) = '/'; 700 701 cwdi = l->l_proc->p_cwdi; 702 rw_enter(&cwdi->cwdi_lock, RW_READER); 703 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2, 704 GETCWD_CHECK_ACCESS, l); 705 rw_exit(&cwdi->cwdi_lock); 706 707 if (error) 708 goto err; 709 tlen = path + MAXPATHLEN - bp; 710 711 memmove(path, bp, tlen); 712 path[tlen - 1] = '\0'; 713 if (offs) 714 *offs = tlen - len; 715 out: 716 *pbp = pathbuf_assimilate(path); 717 return 0; 718 err: 719 PNBUF_PUT(path); 720 return error; 721 } 722 723 vaddr_t 724 exec_vm_minaddr(vaddr_t va_min) 725 { 726 /* 727 * Increase va_min if we don't want NULL to be mappable by the 728 * process. 729 */ 730 #define VM_MIN_GUARD PAGE_SIZE 731 if (user_va0_disable && (va_min < VM_MIN_GUARD)) 732 return VM_MIN_GUARD; 733 return va_min; 734 } 735 736 static int 737 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd, 738 char * const *args, char * const *envs, 739 execve_fetch_element_t fetch_element, 740 struct execve_data * restrict data) 741 { 742 struct exec_package * const epp = &data->ed_pack; 743 int error; 744 struct proc *p; 745 char *dp; 746 u_int modgen; 747 748 KASSERT(data != NULL); 749 750 p = l->l_proc; 751 modgen = 0; 752 753 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0); 754 755 /* 756 * Check if we have exceeded our number of processes limit. 757 * This is so that we handle the case where a root daemon 758 * forked, ran setuid to become the desired user and is trying 759 * to exec. The obvious place to do the reference counting check 760 * is setuid(), but we don't do the reference counting check there 761 * like other OS's do because then all the programs that use setuid() 762 * must be modified to check the return code of setuid() and exit(). 763 * It is dangerous to make setuid() fail, because it fails open and 764 * the program will continue to run as root. If we make it succeed 765 * and return an error code, again we are not enforcing the limit. 766 * The best place to enforce the limit is here, when the process tries 767 * to execute a new image, because eventually the process will need 768 * to call exec in order to do something useful. 769 */ 770 retry: 771 if (p->p_flag & PK_SUGID) { 772 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT, 773 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 774 &p->p_rlimit[RLIMIT_NPROC], 775 KAUTH_ARG(RLIMIT_NPROC)) != 0 && 776 chgproccnt(kauth_cred_getuid(l->l_cred), 0) > 777 p->p_rlimit[RLIMIT_NPROC].rlim_cur) 778 return EAGAIN; 779 } 780 781 /* 782 * Drain existing references and forbid new ones. The process 783 * should be left alone until we're done here. This is necessary 784 * to avoid race conditions - e.g. in ptrace() - that might allow 785 * a local user to illicitly obtain elevated privileges. 786 */ 787 rw_enter(&p->p_reflock, RW_WRITER); 788 789 if (has_path) { 790 size_t offs; 791 /* 792 * Init the namei data to point the file user's program name. 793 * This is done here rather than in check_exec(), so that it's 794 * possible to override this settings if any of makecmd/probe 795 * functions call check_exec() recursively - for example, 796 * see exec_script_makecmds(). 797 */ 798 if ((error = exec_makepathbuf(l, path, UIO_USERSPACE, 799 &data->ed_pathbuf, &offs)) != 0) 800 goto clrflg; 801 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf); 802 epp->ep_kname = data->ed_pathstring + offs; 803 data->ed_resolvedname = PNBUF_GET(); 804 epp->ep_resolvedname = data->ed_resolvedname; 805 epp->ep_xfd = -1; 806 } else { 807 data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/")); 808 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf); 809 epp->ep_kname = "*fexecve*"; 810 data->ed_resolvedname = NULL; 811 epp->ep_resolvedname = NULL; 812 epp->ep_xfd = fd; 813 } 814 815 816 /* 817 * initialize the fields of the exec package. 818 */ 819 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP); 820 epp->ep_hdrlen = exec_maxhdrsz; 821 epp->ep_hdrvalid = 0; 822 epp->ep_emul_arg = NULL; 823 epp->ep_emul_arg_free = NULL; 824 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds)); 825 epp->ep_vap = &data->ed_attr; 826 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0; 827 MD_TOPDOWN_INIT(epp); 828 epp->ep_emul_root = NULL; 829 epp->ep_interp = NULL; 830 epp->ep_esch = NULL; 831 epp->ep_pax_flags = 0; 832 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch)); 833 834 rw_enter(&exec_lock, RW_READER); 835 836 /* see if we can run it. */ 837 if ((error = check_exec(l, epp, data->ed_pathbuf, 838 &data->ed_resolvedname)) != 0) { 839 if (error != ENOENT && error != EACCES && error != ENOEXEC) { 840 DPRINTF(("%s: check exec failed for %s, error %d\n", 841 __func__, epp->ep_kname, error)); 842 } 843 goto freehdr; 844 } 845 846 /* allocate an argument buffer */ 847 data->ed_argp = pool_get(&exec_pool, PR_WAITOK); 848 KASSERT(data->ed_argp != NULL); 849 dp = data->ed_argp; 850 851 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) { 852 goto bad; 853 } 854 855 /* 856 * Calculate the new stack size. 857 */ 858 859 #ifdef __MACHINE_STACK_GROWS_UP 860 /* 861 * copyargs() fills argc/argv/envp from the lower address even on 862 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP 863 * so that _rtld() use it. 864 */ 865 #define RTLD_GAP 32 866 #else 867 #define RTLD_GAP 0 868 #endif 869 870 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp; 871 872 data->ed_argslen = calcargs(data, argenvstrlen); 873 874 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP); 875 876 if (len > epp->ep_ssize) { 877 /* in effect, compare to initial limit */ 878 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len)); 879 error = ENOMEM; 880 goto bad; 881 } 882 /* adjust "active stack depth" for process VSZ */ 883 epp->ep_ssize = len; 884 885 return 0; 886 887 bad: 888 /* free the vmspace-creation commands, and release their references */ 889 kill_vmcmds(&epp->ep_vmcmds); 890 /* kill any opened file descriptor, if necessary */ 891 if (epp->ep_flags & EXEC_HASFD) { 892 epp->ep_flags &= ~EXEC_HASFD; 893 fd_close(epp->ep_fd); 894 } 895 /* close and put the exec'd file */ 896 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 897 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred); 898 vput(epp->ep_vp); 899 pool_put(&exec_pool, data->ed_argp); 900 901 freehdr: 902 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 903 if (epp->ep_emul_root != NULL) 904 vrele(epp->ep_emul_root); 905 if (epp->ep_interp != NULL) 906 vrele(epp->ep_interp); 907 908 rw_exit(&exec_lock); 909 910 exec_path_free(data); 911 912 clrflg: 913 rw_exit(&p->p_reflock); 914 915 if (modgen != module_gen && error == ENOEXEC) { 916 modgen = module_gen; 917 exec_autoload(); 918 goto retry; 919 } 920 921 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 922 return error; 923 } 924 925 static int 926 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data) 927 { 928 struct exec_package * const epp = &data->ed_pack; 929 struct proc *p = l->l_proc; 930 struct exec_vmcmd *base_vcp; 931 int error = 0; 932 size_t i; 933 934 /* record proc's vnode, for use by procfs and others */ 935 if (p->p_textvp) 936 vrele(p->p_textvp); 937 vref(epp->ep_vp); 938 p->p_textvp = epp->ep_vp; 939 940 /* create the new process's VM space by running the vmcmds */ 941 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__); 942 943 #ifdef TRACE_EXEC 944 DUMPVMCMDS(epp, 0, 0); 945 #endif 946 947 base_vcp = NULL; 948 949 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) { 950 struct exec_vmcmd *vcp; 951 952 vcp = &epp->ep_vmcmds.evs_cmds[i]; 953 if (vcp->ev_flags & VMCMD_RELATIVE) { 954 KASSERTMSG(base_vcp != NULL, 955 "%s: relative vmcmd with no base", __func__); 956 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0, 957 "%s: illegal base & relative vmcmd", __func__); 958 vcp->ev_addr += base_vcp->ev_addr; 959 } 960 error = (*vcp->ev_proc)(l, vcp); 961 if (error) 962 DUMPVMCMDS(epp, i, error); 963 if (vcp->ev_flags & VMCMD_BASE) 964 base_vcp = vcp; 965 } 966 967 /* free the vmspace-creation commands, and release their references */ 968 kill_vmcmds(&epp->ep_vmcmds); 969 970 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 971 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred); 972 vput(epp->ep_vp); 973 974 /* if an error happened, deallocate and punt */ 975 if (error != 0) { 976 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error)); 977 } 978 return error; 979 } 980 981 static void 982 execve_free_data(struct execve_data *data) 983 { 984 struct exec_package * const epp = &data->ed_pack; 985 986 /* free the vmspace-creation commands, and release their references */ 987 kill_vmcmds(&epp->ep_vmcmds); 988 /* kill any opened file descriptor, if necessary */ 989 if (epp->ep_flags & EXEC_HASFD) { 990 epp->ep_flags &= ~EXEC_HASFD; 991 fd_close(epp->ep_fd); 992 } 993 994 /* close and put the exec'd file */ 995 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 996 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred); 997 vput(epp->ep_vp); 998 pool_put(&exec_pool, data->ed_argp); 999 1000 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1001 if (epp->ep_emul_root != NULL) 1002 vrele(epp->ep_emul_root); 1003 if (epp->ep_interp != NULL) 1004 vrele(epp->ep_interp); 1005 1006 exec_path_free(data); 1007 } 1008 1009 static void 1010 pathexec(struct proc *p, const char *resolvedname) 1011 { 1012 /* set command name & other accounting info */ 1013 const char *cmdname; 1014 1015 if (resolvedname == NULL) { 1016 cmdname = "*fexecve*"; 1017 resolvedname = "/"; 1018 } else { 1019 cmdname = strrchr(resolvedname, '/') + 1; 1020 } 1021 KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'", 1022 resolvedname); 1023 1024 strlcpy(p->p_comm, cmdname, sizeof(p->p_comm)); 1025 1026 kmem_strfree(p->p_path); 1027 p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP); 1028 } 1029 1030 /* XXX elsewhere */ 1031 static int 1032 credexec(struct lwp *l, struct vattr *attr) 1033 { 1034 struct proc *p = l->l_proc; 1035 int error; 1036 1037 /* 1038 * Deal with set[ug]id. MNT_NOSUID has already been used to disable 1039 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked 1040 * out additional references on the process for the moment. 1041 */ 1042 if ((p->p_slflag & PSL_TRACED) == 0 && 1043 1044 (((attr->va_mode & S_ISUID) != 0 && 1045 kauth_cred_geteuid(l->l_cred) != attr->va_uid) || 1046 1047 ((attr->va_mode & S_ISGID) != 0 && 1048 kauth_cred_getegid(l->l_cred) != attr->va_gid))) { 1049 /* 1050 * Mark the process as SUGID before we do 1051 * anything that might block. 1052 */ 1053 proc_crmod_enter(); 1054 proc_crmod_leave(NULL, NULL, true); 1055 1056 /* Make sure file descriptors 0..2 are in use. */ 1057 if ((error = fd_checkstd()) != 0) { 1058 DPRINTF(("%s: fdcheckstd failed %d\n", 1059 __func__, error)); 1060 return error; 1061 } 1062 1063 /* 1064 * Copy the credential so other references don't see our 1065 * changes. 1066 */ 1067 l->l_cred = kauth_cred_copy(l->l_cred); 1068 #ifdef KTRACE 1069 /* 1070 * If the persistent trace flag isn't set, turn off. 1071 */ 1072 if (p->p_tracep) { 1073 mutex_enter(&ktrace_lock); 1074 if (!(p->p_traceflag & KTRFAC_PERSISTENT)) 1075 ktrderef(p); 1076 mutex_exit(&ktrace_lock); 1077 } 1078 #endif 1079 if (attr->va_mode & S_ISUID) 1080 kauth_cred_seteuid(l->l_cred, attr->va_uid); 1081 if (attr->va_mode & S_ISGID) 1082 kauth_cred_setegid(l->l_cred, attr->va_gid); 1083 } else { 1084 if (kauth_cred_geteuid(l->l_cred) == 1085 kauth_cred_getuid(l->l_cred) && 1086 kauth_cred_getegid(l->l_cred) == 1087 kauth_cred_getgid(l->l_cred)) 1088 p->p_flag &= ~PK_SUGID; 1089 } 1090 1091 /* 1092 * Copy the credential so other references don't see our changes. 1093 * Test to see if this is necessary first, since in the common case 1094 * we won't need a private reference. 1095 */ 1096 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) || 1097 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) { 1098 l->l_cred = kauth_cred_copy(l->l_cred); 1099 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred)); 1100 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred)); 1101 } 1102 1103 /* Update the master credentials. */ 1104 if (l->l_cred != p->p_cred) { 1105 kauth_cred_t ocred; 1106 1107 kauth_cred_hold(l->l_cred); 1108 mutex_enter(p->p_lock); 1109 ocred = p->p_cred; 1110 p->p_cred = l->l_cred; 1111 mutex_exit(p->p_lock); 1112 kauth_cred_free(ocred); 1113 } 1114 1115 return 0; 1116 } 1117 1118 static void 1119 emulexec(struct lwp *l, struct exec_package *epp) 1120 { 1121 struct proc *p = l->l_proc; 1122 1123 /* The emulation root will usually have been found when we looked 1124 * for the elf interpreter (or similar), if not look now. */ 1125 if (epp->ep_esch->es_emul->e_path != NULL && 1126 epp->ep_emul_root == NULL) 1127 emul_find_root(l, epp); 1128 1129 /* Any old emulation root got removed by fdcloseexec */ 1130 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER); 1131 p->p_cwdi->cwdi_edir = epp->ep_emul_root; 1132 rw_exit(&p->p_cwdi->cwdi_lock); 1133 epp->ep_emul_root = NULL; 1134 if (epp->ep_interp != NULL) 1135 vrele(epp->ep_interp); 1136 1137 /* 1138 * Call emulation specific exec hook. This can setup per-process 1139 * p->p_emuldata or do any other per-process stuff an emulation needs. 1140 * 1141 * If we are executing process of different emulation than the 1142 * original forked process, call e_proc_exit() of the old emulation 1143 * first, then e_proc_exec() of new emulation. If the emulation is 1144 * same, the exec hook code should deallocate any old emulation 1145 * resources held previously by this process. 1146 */ 1147 if (p->p_emul && p->p_emul->e_proc_exit 1148 && p->p_emul != epp->ep_esch->es_emul) 1149 (*p->p_emul->e_proc_exit)(p); 1150 1151 /* 1152 * Call exec hook. Emulation code may NOT store reference to anything 1153 * from &pack. 1154 */ 1155 if (epp->ep_esch->es_emul->e_proc_exec) 1156 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp); 1157 1158 /* update p_emul, the old value is no longer needed */ 1159 p->p_emul = epp->ep_esch->es_emul; 1160 1161 /* ...and the same for p_execsw */ 1162 p->p_execsw = epp->ep_esch; 1163 1164 #ifdef __HAVE_SYSCALL_INTERN 1165 (*p->p_emul->e_syscall_intern)(p); 1166 #endif 1167 ktremul(); 1168 } 1169 1170 static int 1171 execve_runproc(struct lwp *l, struct execve_data * restrict data, 1172 bool no_local_exec_lock, bool is_spawn) 1173 { 1174 struct exec_package * const epp = &data->ed_pack; 1175 int error = 0; 1176 struct proc *p; 1177 struct vmspace *vm; 1178 1179 /* 1180 * In case of a posix_spawn operation, the child doing the exec 1181 * might not hold the reader lock on exec_lock, but the parent 1182 * will do this instead. 1183 */ 1184 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock)); 1185 KASSERT(!no_local_exec_lock || is_spawn); 1186 KASSERT(data != NULL); 1187 1188 p = l->l_proc; 1189 1190 /* Get rid of other LWPs. */ 1191 if (p->p_nlwps > 1) { 1192 mutex_enter(p->p_lock); 1193 exit_lwps(l); 1194 mutex_exit(p->p_lock); 1195 } 1196 KDASSERT(p->p_nlwps == 1); 1197 1198 /* Destroy any lwpctl info. */ 1199 if (p->p_lwpctl != NULL) 1200 lwp_ctl_exit(); 1201 1202 /* Remove POSIX timers */ 1203 timers_free(p, TIMERS_POSIX); 1204 1205 /* Set the PaX flags. */ 1206 pax_set_flags(epp, p); 1207 1208 /* 1209 * Do whatever is necessary to prepare the address space 1210 * for remapping. Note that this might replace the current 1211 * vmspace with another! 1212 * 1213 * vfork(): do not touch any user space data in the new child 1214 * until we have awoken the parent below, or it will defeat 1215 * lazy pmap switching (on x86). 1216 */ 1217 if (is_spawn) 1218 uvmspace_spawn(l, epp->ep_vm_minaddr, 1219 epp->ep_vm_maxaddr, 1220 epp->ep_flags & EXEC_TOPDOWN_VM); 1221 else 1222 uvmspace_exec(l, epp->ep_vm_minaddr, 1223 epp->ep_vm_maxaddr, 1224 epp->ep_flags & EXEC_TOPDOWN_VM); 1225 vm = p->p_vmspace; 1226 1227 vm->vm_taddr = (void *)epp->ep_taddr; 1228 vm->vm_tsize = btoc(epp->ep_tsize); 1229 vm->vm_daddr = (void*)epp->ep_daddr; 1230 vm->vm_dsize = btoc(epp->ep_dsize); 1231 vm->vm_ssize = btoc(epp->ep_ssize); 1232 vm->vm_issize = 0; 1233 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr; 1234 vm->vm_minsaddr = (void *)epp->ep_minsaddr; 1235 1236 pax_aslr_init_vm(l, vm, epp); 1237 1238 cwdexec(p); 1239 fd_closeexec(); /* handle close on exec */ 1240 1241 if (__predict_false(ktrace_on)) 1242 fd_ktrexecfd(); 1243 1244 execsigs(p); /* reset caught signals */ 1245 1246 mutex_enter(p->p_lock); 1247 l->l_ctxlink = NULL; /* reset ucontext link */ 1248 p->p_acflag &= ~AFORK; 1249 p->p_flag |= PK_EXEC; 1250 mutex_exit(p->p_lock); 1251 1252 error = credexec(l, &data->ed_attr); 1253 if (error) 1254 goto exec_abort; 1255 1256 #if defined(__HAVE_RAS) 1257 /* 1258 * Remove all RASs from the address space. 1259 */ 1260 ras_purgeall(); 1261 #endif 1262 1263 /* 1264 * Stop profiling. 1265 */ 1266 if ((p->p_stflag & PST_PROFIL) != 0) { 1267 mutex_spin_enter(&p->p_stmutex); 1268 stopprofclock(p); 1269 mutex_spin_exit(&p->p_stmutex); 1270 } 1271 1272 /* 1273 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have 1274 * exited and exec()/exit() are the only places it will be cleared. 1275 * 1276 * Once the parent has been awoken, curlwp may teleport to a new CPU 1277 * in sched_vforkexec(), and it's then OK to start messing with user 1278 * data. See comment above. 1279 */ 1280 if ((p->p_lflag & PL_PPWAIT) != 0) { 1281 bool samecpu; 1282 lwp_t *lp; 1283 1284 mutex_enter(proc_lock); 1285 lp = p->p_vforklwp; 1286 p->p_vforklwp = NULL; 1287 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 1288 cv_broadcast(&lp->l_waitcv); 1289 1290 /* Clear flags after cv_broadcast() (scheduler needs them). */ 1291 p->p_lflag &= ~PL_PPWAIT; 1292 lp->l_vforkwaiting = false; 1293 1294 /* If parent is still on same CPU, teleport curlwp elsewhere. */ 1295 samecpu = (lp->l_cpu == curlwp->l_cpu); 1296 mutex_exit(proc_lock); 1297 1298 /* Give the parent its CPU back - find a new home. */ 1299 KASSERT(!is_spawn); 1300 sched_vforkexec(l, samecpu); 1301 } 1302 1303 /* Now map address space. */ 1304 error = execve_dovmcmds(l, data); 1305 if (error != 0) 1306 goto exec_abort; 1307 1308 pathexec(p, epp->ep_resolvedname); 1309 1310 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize); 1311 1312 error = copyoutargs(data, l, newstack); 1313 if (error != 0) 1314 goto exec_abort; 1315 1316 doexechooks(p); 1317 1318 /* 1319 * Set initial SP at the top of the stack. 1320 * 1321 * Note that on machines where stack grows up (e.g. hppa), SP points to 1322 * the end of arg/env strings. Userland guesses the address of argc 1323 * via ps_strings::ps_argvstr. 1324 */ 1325 1326 /* Setup new registers and do misc. setup. */ 1327 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack); 1328 if (epp->ep_esch->es_setregs) 1329 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack); 1330 1331 /* Provide a consistent LWP private setting */ 1332 (void)lwp_setprivate(l, NULL); 1333 1334 /* Discard all PCU state; need to start fresh */ 1335 pcu_discard_all(l); 1336 1337 /* map the process's signal trampoline code */ 1338 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) { 1339 DPRINTF(("%s: map sigcode failed %d\n", __func__, error)); 1340 goto exec_abort; 1341 } 1342 1343 pool_put(&exec_pool, data->ed_argp); 1344 1345 /* notify others that we exec'd */ 1346 KNOTE(&p->p_klist, NOTE_EXEC); 1347 1348 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1349 1350 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0); 1351 1352 emulexec(l, epp); 1353 1354 /* Allow new references from the debugger/procfs. */ 1355 rw_exit(&p->p_reflock); 1356 if (!no_local_exec_lock) 1357 rw_exit(&exec_lock); 1358 1359 mutex_enter(proc_lock); 1360 1361 /* posix_spawn(3) reports a single event with implied exec(3) */ 1362 if ((p->p_slflag & PSL_TRACED) && !is_spawn) { 1363 mutex_enter(p->p_lock); 1364 eventswitch(TRAP_EXEC, 0, 0); 1365 mutex_enter(proc_lock); 1366 } 1367 1368 if (p->p_sflag & PS_STOPEXEC) { 1369 ksiginfoq_t kq; 1370 1371 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 1372 p->p_pptr->p_nstopchild++; 1373 p->p_waited = 0; 1374 mutex_enter(p->p_lock); 1375 ksiginfo_queue_init(&kq); 1376 sigclearall(p, &contsigmask, &kq); 1377 lwp_lock(l); 1378 l->l_stat = LSSTOP; 1379 p->p_stat = SSTOP; 1380 p->p_nrlwps--; 1381 lwp_unlock(l); 1382 mutex_exit(p->p_lock); 1383 mutex_exit(proc_lock); 1384 lwp_lock(l); 1385 spc_lock(l->l_cpu); 1386 mi_switch(l); 1387 ksiginfo_queue_drain(&kq); 1388 } else { 1389 mutex_exit(proc_lock); 1390 } 1391 1392 exec_path_free(data); 1393 #ifdef TRACE_EXEC 1394 DPRINTF(("%s finished\n", __func__)); 1395 #endif 1396 return EJUSTRETURN; 1397 1398 exec_abort: 1399 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 1400 rw_exit(&p->p_reflock); 1401 if (!no_local_exec_lock) 1402 rw_exit(&exec_lock); 1403 1404 exec_path_free(data); 1405 1406 /* 1407 * the old process doesn't exist anymore. exit gracefully. 1408 * get rid of the (new) address space we have created, if any, get rid 1409 * of our namei data and vnode, and exit noting failure 1410 */ 1411 if (vm != NULL) { 1412 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 1413 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 1414 } 1415 1416 exec_free_emul_arg(epp); 1417 pool_put(&exec_pool, data->ed_argp); 1418 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1419 if (epp->ep_emul_root != NULL) 1420 vrele(epp->ep_emul_root); 1421 if (epp->ep_interp != NULL) 1422 vrele(epp->ep_interp); 1423 1424 /* Acquire the sched-state mutex (exit1() will release it). */ 1425 if (!is_spawn) { 1426 mutex_enter(p->p_lock); 1427 exit1(l, error, SIGABRT); 1428 } 1429 1430 return error; 1431 } 1432 1433 int 1434 execve1(struct lwp *l, bool has_path, const char *path, int fd, 1435 char * const *args, char * const *envs, 1436 execve_fetch_element_t fetch_element) 1437 { 1438 struct execve_data data; 1439 int error; 1440 1441 error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element, 1442 &data); 1443 if (error) 1444 return error; 1445 error = execve_runproc(l, &data, false, false); 1446 return error; 1447 } 1448 1449 static size_t 1450 fromptrsz(const struct exec_package *epp) 1451 { 1452 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *); 1453 } 1454 1455 static size_t 1456 ptrsz(const struct exec_package *epp) 1457 { 1458 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *); 1459 } 1460 1461 static size_t 1462 calcargs(struct execve_data * restrict data, const size_t argenvstrlen) 1463 { 1464 struct exec_package * const epp = &data->ed_pack; 1465 1466 const size_t nargenvptrs = 1467 1 + /* long argc */ 1468 data->ed_argc + /* char *argv[] */ 1469 1 + /* \0 */ 1470 data->ed_envc + /* char *env[] */ 1471 1; /* \0 */ 1472 1473 return (nargenvptrs * ptrsz(epp)) /* pointers */ 1474 + argenvstrlen /* strings */ 1475 + epp->ep_esch->es_arglen; /* auxinfo */ 1476 } 1477 1478 static size_t 1479 calcstack(struct execve_data * restrict data, const size_t gaplen) 1480 { 1481 struct exec_package * const epp = &data->ed_pack; 1482 1483 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode - 1484 epp->ep_esch->es_emul->e_sigcode; 1485 1486 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ? 1487 sizeof(struct ps_strings32) : sizeof(struct ps_strings); 1488 1489 const size_t sigcode_psstr_sz = 1490 data->ed_szsigcode + /* sigcode */ 1491 data->ed_ps_strings_sz + /* ps_strings */ 1492 STACK_PTHREADSPACE; /* pthread space */ 1493 1494 const size_t stacklen = 1495 data->ed_argslen + 1496 gaplen + 1497 sigcode_psstr_sz; 1498 1499 /* make the stack "safely" aligned */ 1500 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES); 1501 } 1502 1503 static int 1504 copyoutargs(struct execve_data * restrict data, struct lwp *l, 1505 char * const newstack) 1506 { 1507 struct exec_package * const epp = &data->ed_pack; 1508 struct proc *p = l->l_proc; 1509 int error; 1510 1511 memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo)); 1512 1513 /* remember information about the process */ 1514 data->ed_arginfo.ps_nargvstr = data->ed_argc; 1515 data->ed_arginfo.ps_nenvstr = data->ed_envc; 1516 1517 /* 1518 * Allocate the stack address passed to the newly execve()'ed process. 1519 * 1520 * The new stack address will be set to the SP (stack pointer) register 1521 * in setregs(). 1522 */ 1523 1524 char *newargs = STACK_ALLOC( 1525 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen); 1526 1527 error = (*epp->ep_esch->es_copyargs)(l, epp, 1528 &data->ed_arginfo, &newargs, data->ed_argp); 1529 1530 if (error) { 1531 DPRINTF(("%s: copyargs failed %d\n", __func__, error)); 1532 return error; 1533 } 1534 1535 error = copyoutpsstrs(data, p); 1536 if (error != 0) 1537 return error; 1538 1539 return 0; 1540 } 1541 1542 static int 1543 copyoutpsstrs(struct execve_data * restrict data, struct proc *p) 1544 { 1545 struct exec_package * const epp = &data->ed_pack; 1546 struct ps_strings32 arginfo32; 1547 void *aip; 1548 int error; 1549 1550 /* fill process ps_strings info */ 1551 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr, 1552 STACK_PTHREADSPACE), data->ed_ps_strings_sz); 1553 1554 if (epp->ep_flags & EXEC_32) { 1555 aip = &arginfo32; 1556 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr; 1557 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr; 1558 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr; 1559 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr; 1560 } else 1561 aip = &data->ed_arginfo; 1562 1563 /* copy out the process's ps_strings structure */ 1564 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)) 1565 != 0) { 1566 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n", 1567 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)); 1568 return error; 1569 } 1570 1571 return 0; 1572 } 1573 1574 static int 1575 copyinargs(struct execve_data * restrict data, char * const *args, 1576 char * const *envs, execve_fetch_element_t fetch_element, char **dpp) 1577 { 1578 struct exec_package * const epp = &data->ed_pack; 1579 char *dp; 1580 size_t i; 1581 int error; 1582 1583 dp = *dpp; 1584 1585 data->ed_argc = 0; 1586 1587 /* copy the fake args list, if there's one, freeing it as we go */ 1588 if (epp->ep_flags & EXEC_HASARGL) { 1589 struct exec_fakearg *fa = epp->ep_fa; 1590 1591 while (fa->fa_arg != NULL) { 1592 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1593 size_t len; 1594 1595 len = strlcpy(dp, fa->fa_arg, maxlen); 1596 /* Count NUL into len. */ 1597 if (len < maxlen) 1598 len++; 1599 else { 1600 while (fa->fa_arg != NULL) { 1601 kmem_free(fa->fa_arg, fa->fa_len); 1602 fa++; 1603 } 1604 kmem_free(epp->ep_fa, epp->ep_fa_len); 1605 epp->ep_flags &= ~EXEC_HASARGL; 1606 return E2BIG; 1607 } 1608 ktrexecarg(fa->fa_arg, len - 1); 1609 dp += len; 1610 1611 kmem_free(fa->fa_arg, fa->fa_len); 1612 fa++; 1613 data->ed_argc++; 1614 } 1615 kmem_free(epp->ep_fa, epp->ep_fa_len); 1616 epp->ep_flags &= ~EXEC_HASARGL; 1617 } 1618 1619 /* 1620 * Read and count argument strings from user. 1621 */ 1622 1623 if (args == NULL) { 1624 DPRINTF(("%s: null args\n", __func__)); 1625 return EINVAL; 1626 } 1627 if (epp->ep_flags & EXEC_SKIPARG) 1628 args = (const void *)((const char *)args + fromptrsz(epp)); 1629 i = 0; 1630 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg); 1631 if (error != 0) { 1632 DPRINTF(("%s: copyin arg %d\n", __func__, error)); 1633 return error; 1634 } 1635 data->ed_argc += i; 1636 1637 /* 1638 * Read and count environment strings from user. 1639 */ 1640 1641 data->ed_envc = 0; 1642 /* environment need not be there */ 1643 if (envs == NULL) 1644 goto done; 1645 i = 0; 1646 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv); 1647 if (error != 0) { 1648 DPRINTF(("%s: copyin env %d\n", __func__, error)); 1649 return error; 1650 } 1651 data->ed_envc += i; 1652 1653 done: 1654 *dpp = dp; 1655 1656 return 0; 1657 } 1658 1659 static int 1660 copyinargstrs(struct execve_data * restrict data, char * const *strs, 1661 execve_fetch_element_t fetch_element, char **dpp, size_t *ip, 1662 void (*ktr)(const void *, size_t)) 1663 { 1664 char *dp, *sp; 1665 size_t i; 1666 int error; 1667 1668 dp = *dpp; 1669 1670 i = 0; 1671 while (1) { 1672 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1673 size_t len; 1674 1675 if ((error = (*fetch_element)(strs, i, &sp)) != 0) { 1676 return error; 1677 } 1678 if (!sp) 1679 break; 1680 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) { 1681 if (error == ENAMETOOLONG) 1682 error = E2BIG; 1683 return error; 1684 } 1685 if (__predict_false(ktrace_on)) 1686 (*ktr)(dp, len - 1); 1687 dp += len; 1688 i++; 1689 } 1690 1691 *dpp = dp; 1692 *ip = i; 1693 1694 return 0; 1695 } 1696 1697 /* 1698 * Copy argv and env strings from kernel buffer (argp) to the new stack. 1699 * Those strings are located just after auxinfo. 1700 */ 1701 int 1702 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo, 1703 char **stackp, void *argp) 1704 { 1705 char **cpp, *dp, *sp; 1706 size_t len; 1707 void *nullp; 1708 long argc, envc; 1709 int error; 1710 1711 cpp = (char **)*stackp; 1712 nullp = NULL; 1713 argc = arginfo->ps_nargvstr; 1714 envc = arginfo->ps_nenvstr; 1715 1716 /* argc on stack is long */ 1717 CTASSERT(sizeof(*cpp) == sizeof(argc)); 1718 1719 dp = (char *)(cpp + 1720 1 + /* long argc */ 1721 argc + /* char *argv[] */ 1722 1 + /* \0 */ 1723 envc + /* char *env[] */ 1724 1) + /* \0 */ 1725 pack->ep_esch->es_arglen; /* auxinfo */ 1726 sp = argp; 1727 1728 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) { 1729 COPYPRINTF("", cpp - 1, sizeof(argc)); 1730 return error; 1731 } 1732 1733 /* XXX don't copy them out, remap them! */ 1734 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 1735 1736 for (; --argc >= 0; sp += len, dp += len) { 1737 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1738 COPYPRINTF("", cpp - 1, sizeof(dp)); 1739 return error; 1740 } 1741 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1742 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1743 return error; 1744 } 1745 } 1746 1747 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1748 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1749 return error; 1750 } 1751 1752 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 1753 1754 for (; --envc >= 0; sp += len, dp += len) { 1755 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1756 COPYPRINTF("", cpp - 1, sizeof(dp)); 1757 return error; 1758 } 1759 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1760 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1761 return error; 1762 } 1763 1764 } 1765 1766 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1767 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1768 return error; 1769 } 1770 1771 *stackp = (char *)cpp; 1772 return 0; 1773 } 1774 1775 1776 /* 1777 * Add execsw[] entries. 1778 */ 1779 int 1780 exec_add(struct execsw *esp, int count) 1781 { 1782 struct exec_entry *it; 1783 int i; 1784 1785 if (count == 0) { 1786 return 0; 1787 } 1788 1789 /* Check for duplicates. */ 1790 rw_enter(&exec_lock, RW_WRITER); 1791 for (i = 0; i < count; i++) { 1792 LIST_FOREACH(it, &ex_head, ex_list) { 1793 /* assume unique (makecmds, probe_func, emulation) */ 1794 if (it->ex_sw->es_makecmds == esp[i].es_makecmds && 1795 it->ex_sw->u.elf_probe_func == 1796 esp[i].u.elf_probe_func && 1797 it->ex_sw->es_emul == esp[i].es_emul) { 1798 rw_exit(&exec_lock); 1799 return EEXIST; 1800 } 1801 } 1802 } 1803 1804 /* Allocate new entries. */ 1805 for (i = 0; i < count; i++) { 1806 it = kmem_alloc(sizeof(*it), KM_SLEEP); 1807 it->ex_sw = &esp[i]; 1808 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1809 } 1810 1811 /* update execsw[] */ 1812 exec_init(0); 1813 rw_exit(&exec_lock); 1814 return 0; 1815 } 1816 1817 /* 1818 * Remove execsw[] entry. 1819 */ 1820 int 1821 exec_remove(struct execsw *esp, int count) 1822 { 1823 struct exec_entry *it, *next; 1824 int i; 1825 const struct proclist_desc *pd; 1826 proc_t *p; 1827 1828 if (count == 0) { 1829 return 0; 1830 } 1831 1832 /* Abort if any are busy. */ 1833 rw_enter(&exec_lock, RW_WRITER); 1834 for (i = 0; i < count; i++) { 1835 mutex_enter(proc_lock); 1836 for (pd = proclists; pd->pd_list != NULL; pd++) { 1837 PROCLIST_FOREACH(p, pd->pd_list) { 1838 if (p->p_execsw == &esp[i]) { 1839 mutex_exit(proc_lock); 1840 rw_exit(&exec_lock); 1841 return EBUSY; 1842 } 1843 } 1844 } 1845 mutex_exit(proc_lock); 1846 } 1847 1848 /* None are busy, so remove them all. */ 1849 for (i = 0; i < count; i++) { 1850 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) { 1851 next = LIST_NEXT(it, ex_list); 1852 if (it->ex_sw == &esp[i]) { 1853 LIST_REMOVE(it, ex_list); 1854 kmem_free(it, sizeof(*it)); 1855 break; 1856 } 1857 } 1858 } 1859 1860 /* update execsw[] */ 1861 exec_init(0); 1862 rw_exit(&exec_lock); 1863 return 0; 1864 } 1865 1866 /* 1867 * Initialize exec structures. If init_boot is true, also does necessary 1868 * one-time initialization (it's called from main() that way). 1869 * Once system is multiuser, this should be called with exec_lock held, 1870 * i.e. via exec_{add|remove}(). 1871 */ 1872 int 1873 exec_init(int init_boot) 1874 { 1875 const struct execsw **sw; 1876 struct exec_entry *ex; 1877 SLIST_HEAD(,exec_entry) first; 1878 SLIST_HEAD(,exec_entry) any; 1879 SLIST_HEAD(,exec_entry) last; 1880 int i, sz; 1881 1882 if (init_boot) { 1883 /* do one-time initializations */ 1884 vaddr_t vmin = 0, vmax; 1885 1886 rw_init(&exec_lock); 1887 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE); 1888 exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax, 1889 maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL); 1890 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH, 1891 "execargs", &exec_palloc, IPL_NONE); 1892 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0); 1893 } else { 1894 KASSERT(rw_write_held(&exec_lock)); 1895 } 1896 1897 /* Sort each entry onto the appropriate queue. */ 1898 SLIST_INIT(&first); 1899 SLIST_INIT(&any); 1900 SLIST_INIT(&last); 1901 sz = 0; 1902 LIST_FOREACH(ex, &ex_head, ex_list) { 1903 switch(ex->ex_sw->es_prio) { 1904 case EXECSW_PRIO_FIRST: 1905 SLIST_INSERT_HEAD(&first, ex, ex_slist); 1906 break; 1907 case EXECSW_PRIO_ANY: 1908 SLIST_INSERT_HEAD(&any, ex, ex_slist); 1909 break; 1910 case EXECSW_PRIO_LAST: 1911 SLIST_INSERT_HEAD(&last, ex, ex_slist); 1912 break; 1913 default: 1914 panic("%s", __func__); 1915 break; 1916 } 1917 sz++; 1918 } 1919 1920 /* 1921 * Create new execsw[]. Ensure we do not try a zero-sized 1922 * allocation. 1923 */ 1924 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP); 1925 i = 0; 1926 SLIST_FOREACH(ex, &first, ex_slist) { 1927 sw[i++] = ex->ex_sw; 1928 } 1929 SLIST_FOREACH(ex, &any, ex_slist) { 1930 sw[i++] = ex->ex_sw; 1931 } 1932 SLIST_FOREACH(ex, &last, ex_slist) { 1933 sw[i++] = ex->ex_sw; 1934 } 1935 1936 /* Replace old execsw[] and free used memory. */ 1937 if (execsw != NULL) { 1938 kmem_free(__UNCONST(execsw), 1939 nexecs * sizeof(struct execsw *) + 1); 1940 } 1941 execsw = sw; 1942 nexecs = sz; 1943 1944 /* Figure out the maximum size of an exec header. */ 1945 exec_maxhdrsz = sizeof(int); 1946 for (i = 0; i < nexecs; i++) { 1947 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1948 exec_maxhdrsz = execsw[i]->es_hdrsz; 1949 } 1950 1951 return 0; 1952 } 1953 1954 static int 1955 exec_sigcode_map(struct proc *p, const struct emul *e) 1956 { 1957 vaddr_t va; 1958 vsize_t sz; 1959 int error; 1960 struct uvm_object *uobj; 1961 1962 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1963 1964 if (e->e_sigobject == NULL || sz == 0) { 1965 return 0; 1966 } 1967 1968 /* 1969 * If we don't have a sigobject for this emulation, create one. 1970 * 1971 * sigobject is an anonymous memory object (just like SYSV shared 1972 * memory) that we keep a permanent reference to and that we map 1973 * in all processes that need this sigcode. The creation is simple, 1974 * we create an object, add a permanent reference to it, map it in 1975 * kernel space, copy out the sigcode to it and unmap it. 1976 * We map it with PROT_READ|PROT_EXEC into the process just 1977 * the way sys_mmap() would map it. 1978 */ 1979 1980 uobj = *e->e_sigobject; 1981 if (uobj == NULL) { 1982 mutex_enter(&sigobject_lock); 1983 if ((uobj = *e->e_sigobject) == NULL) { 1984 uobj = uao_create(sz, 0); 1985 (*uobj->pgops->pgo_reference)(uobj); 1986 va = vm_map_min(kernel_map); 1987 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1988 uobj, 0, 0, 1989 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1990 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 1991 printf("kernel mapping failed %d\n", error); 1992 (*uobj->pgops->pgo_detach)(uobj); 1993 mutex_exit(&sigobject_lock); 1994 return error; 1995 } 1996 memcpy((void *)va, e->e_sigcode, sz); 1997 #ifdef PMAP_NEED_PROCWR 1998 pmap_procwr(&proc0, va, sz); 1999 #endif 2000 uvm_unmap(kernel_map, va, va + round_page(sz)); 2001 *e->e_sigobject = uobj; 2002 } 2003 mutex_exit(&sigobject_lock); 2004 } 2005 2006 /* Just a hint to uvm_map where to put it. */ 2007 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr, 2008 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN); 2009 2010 #ifdef __alpha__ 2011 /* 2012 * Tru64 puts /sbin/loader at the end of user virtual memory, 2013 * which causes the above calculation to put the sigcode at 2014 * an invalid address. Put it just below the text instead. 2015 */ 2016 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) { 2017 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 2018 } 2019 #endif 2020 2021 (*uobj->pgops->pgo_reference)(uobj); 2022 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 2023 uobj, 0, 0, 2024 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 2025 UVM_ADV_RANDOM, 0)); 2026 if (error) { 2027 DPRINTF(("%s, %d: map %p " 2028 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n", 2029 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz), 2030 va, error)); 2031 (*uobj->pgops->pgo_detach)(uobj); 2032 return error; 2033 } 2034 p->p_sigctx.ps_sigcode = (void *)va; 2035 return 0; 2036 } 2037 2038 /* 2039 * Release a refcount on spawn_exec_data and destroy memory, if this 2040 * was the last one. 2041 */ 2042 static void 2043 spawn_exec_data_release(struct spawn_exec_data *data) 2044 { 2045 if (atomic_dec_32_nv(&data->sed_refcnt) != 0) 2046 return; 2047 2048 cv_destroy(&data->sed_cv_child_ready); 2049 mutex_destroy(&data->sed_mtx_child); 2050 2051 if (data->sed_actions) 2052 posix_spawn_fa_free(data->sed_actions, 2053 data->sed_actions->len); 2054 if (data->sed_attrs) 2055 kmem_free(data->sed_attrs, 2056 sizeof(*data->sed_attrs)); 2057 kmem_free(data, sizeof(*data)); 2058 } 2059 2060 static int 2061 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions) 2062 { 2063 struct lwp *l = curlwp; 2064 register_t retval; 2065 int error, newfd; 2066 2067 if (actions == NULL) 2068 return 0; 2069 2070 for (size_t i = 0; i < actions->len; i++) { 2071 const struct posix_spawn_file_actions_entry *fae = 2072 &actions->fae[i]; 2073 switch (fae->fae_action) { 2074 case FAE_OPEN: 2075 if (fd_getfile(fae->fae_fildes) != NULL) { 2076 error = fd_close(fae->fae_fildes); 2077 if (error) 2078 return error; 2079 } 2080 error = fd_open(fae->fae_path, fae->fae_oflag, 2081 fae->fae_mode, &newfd); 2082 if (error) 2083 return error; 2084 if (newfd != fae->fae_fildes) { 2085 error = dodup(l, newfd, 2086 fae->fae_fildes, 0, &retval); 2087 if (fd_getfile(newfd) != NULL) 2088 fd_close(newfd); 2089 } 2090 break; 2091 case FAE_DUP2: 2092 error = dodup(l, fae->fae_fildes, 2093 fae->fae_newfildes, 0, &retval); 2094 break; 2095 case FAE_CLOSE: 2096 if (fd_getfile(fae->fae_fildes) == NULL) { 2097 return EBADF; 2098 } 2099 error = fd_close(fae->fae_fildes); 2100 break; 2101 } 2102 if (error) 2103 return error; 2104 } 2105 return 0; 2106 } 2107 2108 static int 2109 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent) 2110 { 2111 struct sigaction sigact; 2112 int error; 2113 struct proc *p = curproc; 2114 struct lwp *l = curlwp; 2115 2116 if (attrs == NULL) 2117 return 0; 2118 2119 memset(&sigact, 0, sizeof(sigact)); 2120 sigact._sa_u._sa_handler = SIG_DFL; 2121 sigact.sa_flags = 0; 2122 2123 /* 2124 * set state to SSTOP so that this proc can be found by pid. 2125 * see proc_enterprp, do_sched_setparam below 2126 */ 2127 mutex_enter(proc_lock); 2128 /* 2129 * p_stat should be SACTIVE, so we need to adjust the 2130 * parent's p_nstopchild here. For safety, just make 2131 * we're on the good side of SDEAD before we adjust. 2132 */ 2133 int ostat = p->p_stat; 2134 KASSERT(ostat < SSTOP); 2135 p->p_stat = SSTOP; 2136 p->p_waited = 0; 2137 p->p_pptr->p_nstopchild++; 2138 mutex_exit(proc_lock); 2139 2140 /* Set process group */ 2141 if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) { 2142 pid_t mypid = p->p_pid; 2143 pid_t pgrp = attrs->sa_pgroup; 2144 2145 if (pgrp == 0) 2146 pgrp = mypid; 2147 2148 error = proc_enterpgrp(parent, mypid, pgrp, false); 2149 if (error) 2150 goto out; 2151 } 2152 2153 /* Set scheduler policy */ 2154 if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER) 2155 error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy, 2156 &attrs->sa_schedparam); 2157 else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) { 2158 error = do_sched_setparam(parent->p_pid, 0, 2159 SCHED_NONE, &attrs->sa_schedparam); 2160 } 2161 if (error) 2162 goto out; 2163 2164 /* Reset user ID's */ 2165 if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) { 2166 error = do_setresuid(l, -1, kauth_cred_getgid(l->l_cred), -1, 2167 ID_E_EQ_R | ID_E_EQ_S); 2168 if (error) 2169 return error; 2170 error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1, 2171 ID_E_EQ_R | ID_E_EQ_S); 2172 if (error) 2173 goto out; 2174 } 2175 2176 /* Set signal masks/defaults */ 2177 if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) { 2178 mutex_enter(p->p_lock); 2179 error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL); 2180 mutex_exit(p->p_lock); 2181 if (error) 2182 goto out; 2183 } 2184 2185 if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) { 2186 /* 2187 * The following sigaction call is using a sigaction 2188 * version 0 trampoline which is in the compatibility 2189 * code only. This is not a problem because for SIG_DFL 2190 * and SIG_IGN, the trampolines are now ignored. If they 2191 * were not, this would be a problem because we are 2192 * holding the exec_lock, and the compat code needs 2193 * to do the same in order to replace the trampoline 2194 * code of the process. 2195 */ 2196 for (int i = 1; i <= NSIG; i++) { 2197 if (sigismember(&attrs->sa_sigdefault, i)) 2198 sigaction1(l, i, &sigact, NULL, NULL, 0); 2199 } 2200 } 2201 error = 0; 2202 out: 2203 mutex_enter(proc_lock); 2204 p->p_stat = ostat; 2205 p->p_pptr->p_nstopchild--; 2206 mutex_exit(proc_lock); 2207 return error; 2208 } 2209 2210 /* 2211 * A child lwp of a posix_spawn operation starts here and ends up in 2212 * cpu_spawn_return, dealing with all filedescriptor and scheduler 2213 * manipulations in between. 2214 * The parent waits for the child, as it is not clear whether the child 2215 * will be able to acquire its own exec_lock. If it can, the parent can 2216 * be released early and continue running in parallel. If not (or if the 2217 * magic debug flag is passed in the scheduler attribute struct), the 2218 * child rides on the parent's exec lock until it is ready to return to 2219 * to userland - and only then releases the parent. This method loses 2220 * concurrency, but improves error reporting. 2221 */ 2222 static void 2223 spawn_return(void *arg) 2224 { 2225 struct spawn_exec_data *spawn_data = arg; 2226 struct lwp *l = curlwp; 2227 struct proc *p = l->l_proc; 2228 int error; 2229 bool have_reflock; 2230 bool parent_is_waiting = true; 2231 2232 /* 2233 * Check if we can release parent early. 2234 * We either need to have no sed_attrs, or sed_attrs does not 2235 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require 2236 * safe access to the parent proc (passed in sed_parent). 2237 * We then try to get the exec_lock, and only if that works, we can 2238 * release the parent here already. 2239 */ 2240 struct posix_spawnattr *attrs = spawn_data->sed_attrs; 2241 if ((!attrs || (attrs->sa_flags 2242 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0) 2243 && rw_tryenter(&exec_lock, RW_READER)) { 2244 parent_is_waiting = false; 2245 mutex_enter(&spawn_data->sed_mtx_child); 2246 cv_signal(&spawn_data->sed_cv_child_ready); 2247 mutex_exit(&spawn_data->sed_mtx_child); 2248 } 2249 2250 /* don't allow debugger access yet */ 2251 rw_enter(&p->p_reflock, RW_WRITER); 2252 have_reflock = true; 2253 2254 /* handle posix_spawn_file_actions */ 2255 error = handle_posix_spawn_file_actions(spawn_data->sed_actions); 2256 if (error) 2257 goto report_error; 2258 2259 /* handle posix_spawnattr */ 2260 error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent); 2261 if (error) 2262 goto report_error; 2263 2264 /* now do the real exec */ 2265 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting, 2266 true); 2267 have_reflock = false; 2268 if (error == EJUSTRETURN) 2269 error = 0; 2270 else if (error) 2271 goto report_error; 2272 2273 if (parent_is_waiting) { 2274 mutex_enter(&spawn_data->sed_mtx_child); 2275 cv_signal(&spawn_data->sed_cv_child_ready); 2276 mutex_exit(&spawn_data->sed_mtx_child); 2277 } 2278 2279 /* release our refcount on the data */ 2280 spawn_exec_data_release(spawn_data); 2281 2282 if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) == 2283 (PSL_TRACED|PSL_TRACEDCHILD)) { 2284 eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN); 2285 } 2286 2287 /* and finally: leave to userland for the first time */ 2288 cpu_spawn_return(l); 2289 2290 /* NOTREACHED */ 2291 return; 2292 2293 report_error: 2294 if (have_reflock) { 2295 /* 2296 * We have not passed through execve_runproc(), 2297 * which would have released the p_reflock and also 2298 * taken ownership of the sed_exec part of spawn_data, 2299 * so release/free both here. 2300 */ 2301 rw_exit(&p->p_reflock); 2302 execve_free_data(&spawn_data->sed_exec); 2303 } 2304 2305 if (parent_is_waiting) { 2306 /* pass error to parent */ 2307 mutex_enter(&spawn_data->sed_mtx_child); 2308 spawn_data->sed_error = error; 2309 cv_signal(&spawn_data->sed_cv_child_ready); 2310 mutex_exit(&spawn_data->sed_mtx_child); 2311 } else { 2312 rw_exit(&exec_lock); 2313 } 2314 2315 /* release our refcount on the data */ 2316 spawn_exec_data_release(spawn_data); 2317 2318 /* done, exit */ 2319 mutex_enter(p->p_lock); 2320 /* 2321 * Posix explicitly asks for an exit code of 127 if we report 2322 * errors from the child process - so, unfortunately, there 2323 * is no way to report a more exact error code. 2324 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as 2325 * flag bit in the attrp argument to posix_spawn(2), see above. 2326 */ 2327 exit1(l, 127, 0); 2328 } 2329 2330 void 2331 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len) 2332 { 2333 2334 for (size_t i = 0; i < len; i++) { 2335 struct posix_spawn_file_actions_entry *fae = &fa->fae[i]; 2336 if (fae->fae_action != FAE_OPEN) 2337 continue; 2338 kmem_strfree(fae->fae_path); 2339 } 2340 if (fa->len > 0) 2341 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len); 2342 kmem_free(fa, sizeof(*fa)); 2343 } 2344 2345 static int 2346 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap, 2347 const struct posix_spawn_file_actions *ufa, rlim_t lim) 2348 { 2349 struct posix_spawn_file_actions *fa; 2350 struct posix_spawn_file_actions_entry *fae; 2351 char *pbuf = NULL; 2352 int error; 2353 size_t i = 0; 2354 2355 fa = kmem_alloc(sizeof(*fa), KM_SLEEP); 2356 error = copyin(ufa, fa, sizeof(*fa)); 2357 if (error || fa->len == 0) { 2358 kmem_free(fa, sizeof(*fa)); 2359 return error; /* 0 if not an error, and len == 0 */ 2360 } 2361 2362 if (fa->len > lim) { 2363 kmem_free(fa, sizeof(*fa)); 2364 return EINVAL; 2365 } 2366 2367 fa->size = fa->len; 2368 size_t fal = fa->len * sizeof(*fae); 2369 fae = fa->fae; 2370 fa->fae = kmem_alloc(fal, KM_SLEEP); 2371 error = copyin(fae, fa->fae, fal); 2372 if (error) 2373 goto out; 2374 2375 pbuf = PNBUF_GET(); 2376 for (; i < fa->len; i++) { 2377 fae = &fa->fae[i]; 2378 if (fae->fae_action != FAE_OPEN) 2379 continue; 2380 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal); 2381 if (error) 2382 goto out; 2383 fae->fae_path = kmem_alloc(fal, KM_SLEEP); 2384 memcpy(fae->fae_path, pbuf, fal); 2385 } 2386 PNBUF_PUT(pbuf); 2387 2388 *fap = fa; 2389 return 0; 2390 out: 2391 if (pbuf) 2392 PNBUF_PUT(pbuf); 2393 posix_spawn_fa_free(fa, i); 2394 return error; 2395 } 2396 2397 /* 2398 * N.B. increments nprocs upon success. Callers need to drop nprocs if 2399 * they fail for some other reason. 2400 */ 2401 int 2402 check_posix_spawn(struct lwp *l1) 2403 { 2404 int error, tnprocs, count; 2405 uid_t uid; 2406 struct proc *p1; 2407 2408 p1 = l1->l_proc; 2409 uid = kauth_cred_getuid(l1->l_cred); 2410 tnprocs = atomic_inc_uint_nv(&nprocs); 2411 2412 /* 2413 * Although process entries are dynamically created, we still keep 2414 * a global limit on the maximum number we will create. 2415 */ 2416 if (__predict_false(tnprocs >= maxproc)) 2417 error = -1; 2418 else 2419 error = kauth_authorize_process(l1->l_cred, 2420 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL); 2421 2422 if (error) { 2423 atomic_dec_uint(&nprocs); 2424 return EAGAIN; 2425 } 2426 2427 /* 2428 * Enforce limits. 2429 */ 2430 count = chgproccnt(uid, 1); 2431 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT, 2432 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 2433 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 && 2434 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 2435 (void)chgproccnt(uid, -1); 2436 atomic_dec_uint(&nprocs); 2437 return EAGAIN; 2438 } 2439 2440 return 0; 2441 } 2442 2443 int 2444 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path, 2445 struct posix_spawn_file_actions *fa, 2446 struct posix_spawnattr *sa, 2447 char *const *argv, char *const *envp, 2448 execve_fetch_element_t fetch) 2449 { 2450 2451 struct proc *p1, *p2; 2452 struct lwp *l2; 2453 int error; 2454 struct spawn_exec_data *spawn_data; 2455 vaddr_t uaddr; 2456 pid_t pid; 2457 bool have_exec_lock = false; 2458 2459 p1 = l1->l_proc; 2460 2461 /* Allocate and init spawn_data */ 2462 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP); 2463 spawn_data->sed_refcnt = 1; /* only parent so far */ 2464 cv_init(&spawn_data->sed_cv_child_ready, "pspawn"); 2465 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE); 2466 mutex_enter(&spawn_data->sed_mtx_child); 2467 2468 /* 2469 * Do the first part of the exec now, collect state 2470 * in spawn_data. 2471 */ 2472 error = execve_loadvm(l1, true, path, -1, argv, 2473 envp, fetch, &spawn_data->sed_exec); 2474 if (error == EJUSTRETURN) 2475 error = 0; 2476 else if (error) 2477 goto error_exit; 2478 2479 have_exec_lock = true; 2480 2481 /* 2482 * Allocate virtual address space for the U-area now, while it 2483 * is still easy to abort the fork operation if we're out of 2484 * kernel virtual address space. 2485 */ 2486 uaddr = uvm_uarea_alloc(); 2487 if (__predict_false(uaddr == 0)) { 2488 error = ENOMEM; 2489 goto error_exit; 2490 } 2491 2492 /* 2493 * Allocate new proc. Borrow proc0 vmspace for it, we will 2494 * replace it with its own before returning to userland 2495 * in the child. 2496 */ 2497 p2 = proc_alloc(); 2498 if (p2 == NULL) { 2499 /* We were unable to allocate a process ID. */ 2500 error = EAGAIN; 2501 goto error_exit; 2502 } 2503 2504 /* 2505 * This is a point of no return, we will have to go through 2506 * the child proc to properly clean it up past this point. 2507 */ 2508 pid = p2->p_pid; 2509 2510 /* 2511 * Make a proc table entry for the new process. 2512 * Start by zeroing the section of proc that is zero-initialized, 2513 * then copy the section that is copied directly from the parent. 2514 */ 2515 memset(&p2->p_startzero, 0, 2516 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero)); 2517 memcpy(&p2->p_startcopy, &p1->p_startcopy, 2518 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy)); 2519 p2->p_vmspace = proc0.p_vmspace; 2520 2521 TAILQ_INIT(&p2->p_sigpend.sp_info); 2522 2523 LIST_INIT(&p2->p_lwps); 2524 LIST_INIT(&p2->p_sigwaiters); 2525 2526 /* 2527 * Duplicate sub-structures as needed. 2528 * Increase reference counts on shared objects. 2529 * Inherit flags we want to keep. The flags related to SIGCHLD 2530 * handling are important in order to keep a consistent behaviour 2531 * for the child after the fork. If we are a 32-bit process, the 2532 * child will be too. 2533 */ 2534 p2->p_flag = 2535 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32); 2536 p2->p_emul = p1->p_emul; 2537 p2->p_execsw = p1->p_execsw; 2538 2539 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH); 2540 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE); 2541 rw_init(&p2->p_reflock); 2542 cv_init(&p2->p_waitcv, "wait"); 2543 cv_init(&p2->p_lwpcv, "lwpwait"); 2544 2545 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 2546 2547 kauth_proc_fork(p1, p2); 2548 2549 p2->p_raslist = NULL; 2550 p2->p_fd = fd_copy(); 2551 2552 /* XXX racy */ 2553 p2->p_mqueue_cnt = p1->p_mqueue_cnt; 2554 2555 p2->p_cwdi = cwdinit(); 2556 2557 /* 2558 * Note: p_limit (rlimit stuff) is copy-on-write, so normally 2559 * we just need increase pl_refcnt. 2560 */ 2561 if (!p1->p_limit->pl_writeable) { 2562 lim_addref(p1->p_limit); 2563 p2->p_limit = p1->p_limit; 2564 } else { 2565 p2->p_limit = lim_copy(p1->p_limit); 2566 } 2567 2568 p2->p_lflag = 0; 2569 l1->l_vforkwaiting = false; 2570 p2->p_sflag = 0; 2571 p2->p_slflag = 0; 2572 p2->p_pptr = p1; 2573 p2->p_ppid = p1->p_pid; 2574 LIST_INIT(&p2->p_children); 2575 2576 p2->p_aio = NULL; 2577 2578 #ifdef KTRACE 2579 /* 2580 * Copy traceflag and tracefile if enabled. 2581 * If not inherited, these were zeroed above. 2582 */ 2583 if (p1->p_traceflag & KTRFAC_INHERIT) { 2584 mutex_enter(&ktrace_lock); 2585 p2->p_traceflag = p1->p_traceflag; 2586 if ((p2->p_tracep = p1->p_tracep) != NULL) 2587 ktradref(p2); 2588 mutex_exit(&ktrace_lock); 2589 } 2590 #endif 2591 2592 /* 2593 * Create signal actions for the child process. 2594 */ 2595 p2->p_sigacts = sigactsinit(p1, 0); 2596 mutex_enter(p1->p_lock); 2597 p2->p_sflag |= 2598 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 2599 sched_proc_fork(p1, p2); 2600 mutex_exit(p1->p_lock); 2601 2602 p2->p_stflag = p1->p_stflag; 2603 2604 /* 2605 * p_stats. 2606 * Copy parts of p_stats, and zero out the rest. 2607 */ 2608 p2->p_stats = pstatscopy(p1->p_stats); 2609 2610 /* copy over machdep flags to the new proc */ 2611 cpu_proc_fork(p1, p2); 2612 2613 /* 2614 * Prepare remaining parts of spawn data 2615 */ 2616 spawn_data->sed_actions = fa; 2617 spawn_data->sed_attrs = sa; 2618 2619 spawn_data->sed_parent = p1; 2620 2621 /* create LWP */ 2622 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data, 2623 &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk); 2624 l2->l_ctxlink = NULL; /* reset ucontext link */ 2625 2626 /* 2627 * Copy the credential so other references don't see our changes. 2628 * Test to see if this is necessary first, since in the common case 2629 * we won't need a private reference. 2630 */ 2631 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) || 2632 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) { 2633 l2->l_cred = kauth_cred_copy(l2->l_cred); 2634 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred)); 2635 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred)); 2636 } 2637 2638 /* Update the master credentials. */ 2639 if (l2->l_cred != p2->p_cred) { 2640 kauth_cred_t ocred; 2641 2642 kauth_cred_hold(l2->l_cred); 2643 mutex_enter(p2->p_lock); 2644 ocred = p2->p_cred; 2645 p2->p_cred = l2->l_cred; 2646 mutex_exit(p2->p_lock); 2647 kauth_cred_free(ocred); 2648 } 2649 2650 *child_ok = true; 2651 spawn_data->sed_refcnt = 2; /* child gets it as well */ 2652 #if 0 2653 l2->l_nopreempt = 1; /* start it non-preemptable */ 2654 #endif 2655 2656 /* 2657 * It's now safe for the scheduler and other processes to see the 2658 * child process. 2659 */ 2660 mutex_enter(proc_lock); 2661 2662 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 2663 p2->p_lflag |= PL_CONTROLT; 2664 2665 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 2666 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */ 2667 2668 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) == 2669 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2670 proc_changeparent(p2, p1->p_pptr); 2671 SET(p2->p_slflag, PSL_TRACEDCHILD); 2672 } 2673 2674 p2->p_oppid = p1->p_pid; /* Remember the original parent id. */ 2675 2676 LIST_INSERT_AFTER(p1, p2, p_pglist); 2677 LIST_INSERT_HEAD(&allproc, p2, p_list); 2678 2679 p2->p_trace_enabled = trace_is_enabled(p2); 2680 #ifdef __HAVE_SYSCALL_INTERN 2681 (*p2->p_emul->e_syscall_intern)(p2); 2682 #endif 2683 2684 /* 2685 * Make child runnable, set start time, and add to run queue except 2686 * if the parent requested the child to start in SSTOP state. 2687 */ 2688 mutex_enter(p2->p_lock); 2689 2690 getmicrotime(&p2->p_stats->p_start); 2691 2692 lwp_lock(l2); 2693 KASSERT(p2->p_nrlwps == 1); 2694 KASSERT(l2->l_stat == LSIDL); 2695 p2->p_nrlwps = 1; 2696 p2->p_stat = SACTIVE; 2697 setrunnable(l2); 2698 /* LWP now unlocked */ 2699 2700 mutex_exit(p2->p_lock); 2701 mutex_exit(proc_lock); 2702 2703 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child); 2704 error = spawn_data->sed_error; 2705 mutex_exit(&spawn_data->sed_mtx_child); 2706 spawn_exec_data_release(spawn_data); 2707 2708 rw_exit(&p1->p_reflock); 2709 rw_exit(&exec_lock); 2710 have_exec_lock = false; 2711 2712 *pid_res = pid; 2713 2714 if (error) 2715 return error; 2716 2717 if (p1->p_slflag & PSL_TRACED) { 2718 /* Paranoid check */ 2719 mutex_enter(proc_lock); 2720 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) != 2721 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2722 mutex_exit(proc_lock); 2723 return 0; 2724 } 2725 2726 mutex_enter(p1->p_lock); 2727 eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid); 2728 } 2729 return 0; 2730 2731 error_exit: 2732 if (have_exec_lock) { 2733 execve_free_data(&spawn_data->sed_exec); 2734 rw_exit(&p1->p_reflock); 2735 rw_exit(&exec_lock); 2736 } 2737 mutex_exit(&spawn_data->sed_mtx_child); 2738 spawn_exec_data_release(spawn_data); 2739 2740 return error; 2741 } 2742 2743 int 2744 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap, 2745 register_t *retval) 2746 { 2747 /* { 2748 syscallarg(pid_t *) pid; 2749 syscallarg(const char *) path; 2750 syscallarg(const struct posix_spawn_file_actions *) file_actions; 2751 syscallarg(const struct posix_spawnattr *) attrp; 2752 syscallarg(char *const *) argv; 2753 syscallarg(char *const *) envp; 2754 } */ 2755 2756 int error; 2757 struct posix_spawn_file_actions *fa = NULL; 2758 struct posix_spawnattr *sa = NULL; 2759 pid_t pid; 2760 bool child_ok = false; 2761 rlim_t max_fileactions; 2762 proc_t *p = l1->l_proc; 2763 2764 /* check_posix_spawn() increments nprocs for us. */ 2765 error = check_posix_spawn(l1); 2766 if (error) { 2767 *retval = error; 2768 return 0; 2769 } 2770 2771 /* copy in file_actions struct */ 2772 if (SCARG(uap, file_actions) != NULL) { 2773 max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur, 2774 maxfiles); 2775 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions), 2776 max_fileactions); 2777 if (error) 2778 goto error_exit; 2779 } 2780 2781 /* copyin posix_spawnattr struct */ 2782 if (SCARG(uap, attrp) != NULL) { 2783 sa = kmem_alloc(sizeof(*sa), KM_SLEEP); 2784 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa)); 2785 if (error) 2786 goto error_exit; 2787 } 2788 2789 /* 2790 * Do the spawn 2791 */ 2792 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa, 2793 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element); 2794 if (error) 2795 goto error_exit; 2796 2797 if (error == 0 && SCARG(uap, pid) != NULL) 2798 error = copyout(&pid, SCARG(uap, pid), sizeof(pid)); 2799 2800 *retval = error; 2801 return 0; 2802 2803 error_exit: 2804 if (!child_ok) { 2805 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1); 2806 atomic_dec_uint(&nprocs); 2807 2808 if (sa) 2809 kmem_free(sa, sizeof(*sa)); 2810 if (fa) 2811 posix_spawn_fa_free(fa, fa->len); 2812 } 2813 2814 *retval = error; 2815 return 0; 2816 } 2817 2818 void 2819 exec_free_emul_arg(struct exec_package *epp) 2820 { 2821 if (epp->ep_emul_arg_free != NULL) { 2822 KASSERT(epp->ep_emul_arg != NULL); 2823 (*epp->ep_emul_arg_free)(epp->ep_emul_arg); 2824 epp->ep_emul_arg_free = NULL; 2825 epp->ep_emul_arg = NULL; 2826 } else { 2827 KASSERT(epp->ep_emul_arg == NULL); 2828 } 2829 } 2830 2831 #ifdef DEBUG_EXEC 2832 static void 2833 dump_vmcmds(const struct exec_package * const epp, size_t x, int error) 2834 { 2835 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0]; 2836 size_t j; 2837 2838 if (error == 0) 2839 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used)); 2840 else 2841 DPRINTF(("vmcmds %zu/%u, error %d\n", x, 2842 epp->ep_vmcmds.evs_used, error)); 2843 2844 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) { 2845 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#" 2846 PRIxVADDR"/%#"PRIxVSIZE" fd@%#" 2847 PRIxVSIZE" prot=0%o flags=%d\n", j, 2848 vp[j].ev_proc == vmcmd_map_pagedvn ? 2849 "pagedvn" : 2850 vp[j].ev_proc == vmcmd_map_readvn ? 2851 "readvn" : 2852 vp[j].ev_proc == vmcmd_map_zero ? 2853 "zero" : "*unknown*", 2854 vp[j].ev_addr, vp[j].ev_len, 2855 vp[j].ev_offset, vp[j].ev_prot, 2856 vp[j].ev_flags)); 2857 if (error != 0 && j == x) 2858 DPRINTF((" ^--- failed\n")); 2859 } 2860 } 2861 #endif 2862