1 /* $NetBSD: kern_exec.c,v 1.505 2021/05/02 10:23:55 martin 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 1.505 2021/05/02 10:23:55 martin 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 #ifdef DIAGNOSTIC 502 #define LMSG "%s: rejecting due to %s limit (%ju > %ju)\n" 503 #endif 504 #ifdef MAXTSIZ 505 if (epp->ep_tsize > MAXTSIZ) { 506 #ifdef DIAGNOSTIC 507 printf(LMSG, __func__, "text", 508 (uintmax_t)epp->ep_tsize, 509 (uintmax_t)MAXTSIZ); 510 #endif 511 error = ENOMEM; 512 break; 513 } 514 #endif 515 vsize_t dlimit = 516 (vsize_t)l->l_proc->p_rlimit[RLIMIT_DATA].rlim_cur; 517 if (epp->ep_dsize > dlimit) { 518 #ifdef DIAGNOSTIC 519 printf(LMSG, __func__, "data", 520 (uintmax_t)epp->ep_dsize, 521 (uintmax_t)dlimit); 522 #endif 523 error = ENOMEM; 524 break; 525 } 526 return 0; 527 } 528 529 /* 530 * Reset all the fields that may have been modified by the 531 * loader. 532 */ 533 KASSERT(epp->ep_emul_arg == NULL); 534 if (epp->ep_emul_root != NULL) { 535 vrele(epp->ep_emul_root); 536 epp->ep_emul_root = NULL; 537 } 538 if (epp->ep_interp != NULL) { 539 vrele(epp->ep_interp); 540 epp->ep_interp = NULL; 541 } 542 epp->ep_pax_flags = 0; 543 544 /* make sure the first "interesting" error code is saved. */ 545 if (error == ENOEXEC) 546 error = newerror; 547 548 if (epp->ep_flags & EXEC_DESTR) 549 /* Error from "#!" code, tidied up by recursive call */ 550 return error; 551 } 552 553 /* not found, error */ 554 555 /* 556 * free any vmspace-creation commands, 557 * and release their references 558 */ 559 kill_vmcmds(&epp->ep_vmcmds); 560 561 #if NVERIEXEC > 0 || defined(PAX_SEGVGUARD) 562 bad2: 563 #endif 564 /* 565 * close and release the vnode, restore the old one, free the 566 * pathname buf, and punt. 567 */ 568 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 569 VOP_CLOSE(vp, FREAD, l->l_cred); 570 vput(vp); 571 return error; 572 573 bad1: 574 /* 575 * free the namei pathname buffer, and put the vnode 576 * (which we don't yet have open). 577 */ 578 vput(vp); /* was still locked */ 579 return error; 580 } 581 582 #ifdef __MACHINE_STACK_GROWS_UP 583 #define STACK_PTHREADSPACE NBPG 584 #else 585 #define STACK_PTHREADSPACE 0 586 #endif 587 588 static int 589 execve_fetch_element(char * const *array, size_t index, char **value) 590 { 591 return copyin(array + index, value, sizeof(*value)); 592 } 593 594 /* 595 * exec system call 596 */ 597 int 598 sys_execve(struct lwp *l, const struct sys_execve_args *uap, register_t *retval) 599 { 600 /* { 601 syscallarg(const char *) path; 602 syscallarg(char * const *) argp; 603 syscallarg(char * const *) envp; 604 } */ 605 606 return execve1(l, true, SCARG(uap, path), -1, SCARG(uap, argp), 607 SCARG(uap, envp), execve_fetch_element); 608 } 609 610 int 611 sys_fexecve(struct lwp *l, const struct sys_fexecve_args *uap, 612 register_t *retval) 613 { 614 /* { 615 syscallarg(int) fd; 616 syscallarg(char * const *) argp; 617 syscallarg(char * const *) envp; 618 } */ 619 620 return execve1(l, false, NULL, SCARG(uap, fd), SCARG(uap, argp), 621 SCARG(uap, envp), execve_fetch_element); 622 } 623 624 /* 625 * Load modules to try and execute an image that we do not understand. 626 * If no execsw entries are present, we load those likely to be needed 627 * in order to run native images only. Otherwise, we autoload all 628 * possible modules that could let us run the binary. XXX lame 629 */ 630 static void 631 exec_autoload(void) 632 { 633 #ifdef MODULAR 634 static const char * const native[] = { 635 "exec_elf32", 636 "exec_elf64", 637 "exec_script", 638 NULL 639 }; 640 static const char * const compat[] = { 641 "exec_elf32", 642 "exec_elf64", 643 "exec_script", 644 "exec_aout", 645 "exec_coff", 646 "exec_ecoff", 647 "compat_aoutm68k", 648 "compat_netbsd32", 649 #if 0 650 "compat_linux", 651 "compat_linux32", 652 #endif 653 "compat_sunos", 654 "compat_sunos32", 655 "compat_ultrix", 656 NULL 657 }; 658 char const * const *list; 659 int i; 660 661 list = nexecs == 0 ? native : compat; 662 for (i = 0; list[i] != NULL; i++) { 663 if (module_autoload(list[i], MODULE_CLASS_EXEC) != 0) { 664 continue; 665 } 666 yield(); 667 } 668 #endif 669 } 670 671 /* 672 * Copy the user or kernel supplied upath to the allocated pathbuffer pbp 673 * making it absolute in the process, by prepending the current working 674 * directory if it is not. If offs is supplied it will contain the offset 675 * where the original supplied copy of upath starts. 676 */ 677 int 678 exec_makepathbuf(struct lwp *l, const char *upath, enum uio_seg seg, 679 struct pathbuf **pbp, size_t *offs) 680 { 681 char *path, *bp; 682 size_t len, tlen; 683 int error; 684 struct cwdinfo *cwdi; 685 686 path = PNBUF_GET(); 687 if (seg == UIO_SYSSPACE) { 688 error = copystr(upath, path, MAXPATHLEN, &len); 689 } else { 690 error = copyinstr(upath, path, MAXPATHLEN, &len); 691 } 692 if (error) 693 goto err; 694 695 if (path[0] == '/') { 696 if (offs) 697 *offs = 0; 698 goto out; 699 } 700 701 len++; 702 if (len + 1 >= MAXPATHLEN) { 703 error = ENAMETOOLONG; 704 goto err; 705 } 706 bp = path + MAXPATHLEN - len; 707 memmove(bp, path, len); 708 *(--bp) = '/'; 709 710 cwdi = l->l_proc->p_cwdi; 711 rw_enter(&cwdi->cwdi_lock, RW_READER); 712 error = getcwd_common(cwdi->cwdi_cdir, NULL, &bp, path, MAXPATHLEN / 2, 713 GETCWD_CHECK_ACCESS, l); 714 rw_exit(&cwdi->cwdi_lock); 715 716 if (error) 717 goto err; 718 tlen = path + MAXPATHLEN - bp; 719 720 memmove(path, bp, tlen); 721 path[tlen - 1] = '\0'; 722 if (offs) 723 *offs = tlen - len; 724 out: 725 *pbp = pathbuf_assimilate(path); 726 return 0; 727 err: 728 PNBUF_PUT(path); 729 return error; 730 } 731 732 vaddr_t 733 exec_vm_minaddr(vaddr_t va_min) 734 { 735 /* 736 * Increase va_min if we don't want NULL to be mappable by the 737 * process. 738 */ 739 #define VM_MIN_GUARD PAGE_SIZE 740 if (user_va0_disable && (va_min < VM_MIN_GUARD)) 741 return VM_MIN_GUARD; 742 return va_min; 743 } 744 745 static int 746 execve_loadvm(struct lwp *l, bool has_path, const char *path, int fd, 747 char * const *args, char * const *envs, 748 execve_fetch_element_t fetch_element, 749 struct execve_data * restrict data) 750 { 751 struct exec_package * const epp = &data->ed_pack; 752 int error; 753 struct proc *p; 754 char *dp; 755 u_int modgen; 756 757 KASSERT(data != NULL); 758 759 p = l->l_proc; 760 modgen = 0; 761 762 SDT_PROBE(proc, kernel, , exec, path, 0, 0, 0, 0); 763 764 /* 765 * Check if we have exceeded our number of processes limit. 766 * This is so that we handle the case where a root daemon 767 * forked, ran setuid to become the desired user and is trying 768 * to exec. The obvious place to do the reference counting check 769 * is setuid(), but we don't do the reference counting check there 770 * like other OS's do because then all the programs that use setuid() 771 * must be modified to check the return code of setuid() and exit(). 772 * It is dangerous to make setuid() fail, because it fails open and 773 * the program will continue to run as root. If we make it succeed 774 * and return an error code, again we are not enforcing the limit. 775 * The best place to enforce the limit is here, when the process tries 776 * to execute a new image, because eventually the process will need 777 * to call exec in order to do something useful. 778 */ 779 retry: 780 if (p->p_flag & PK_SUGID) { 781 if (kauth_authorize_process(l->l_cred, KAUTH_PROCESS_RLIMIT, 782 p, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 783 &p->p_rlimit[RLIMIT_NPROC], 784 KAUTH_ARG(RLIMIT_NPROC)) != 0 && 785 chgproccnt(kauth_cred_getuid(l->l_cred), 0) > 786 p->p_rlimit[RLIMIT_NPROC].rlim_cur) 787 return EAGAIN; 788 } 789 790 /* 791 * Drain existing references and forbid new ones. The process 792 * should be left alone until we're done here. This is necessary 793 * to avoid race conditions - e.g. in ptrace() - that might allow 794 * a local user to illicitly obtain elevated privileges. 795 */ 796 rw_enter(&p->p_reflock, RW_WRITER); 797 798 if (has_path) { 799 size_t offs; 800 /* 801 * Init the namei data to point the file user's program name. 802 * This is done here rather than in check_exec(), so that it's 803 * possible to override this settings if any of makecmd/probe 804 * functions call check_exec() recursively - for example, 805 * see exec_script_makecmds(). 806 */ 807 if ((error = exec_makepathbuf(l, path, UIO_USERSPACE, 808 &data->ed_pathbuf, &offs)) != 0) 809 goto clrflg; 810 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf); 811 epp->ep_kname = data->ed_pathstring + offs; 812 data->ed_resolvedname = PNBUF_GET(); 813 epp->ep_resolvedname = data->ed_resolvedname; 814 epp->ep_xfd = -1; 815 } else { 816 data->ed_pathbuf = pathbuf_assimilate(strcpy(PNBUF_GET(), "/")); 817 data->ed_pathstring = pathbuf_stringcopy_get(data->ed_pathbuf); 818 epp->ep_kname = "*fexecve*"; 819 data->ed_resolvedname = NULL; 820 epp->ep_resolvedname = NULL; 821 epp->ep_xfd = fd; 822 } 823 824 825 /* 826 * initialize the fields of the exec package. 827 */ 828 epp->ep_hdr = kmem_alloc(exec_maxhdrsz, KM_SLEEP); 829 epp->ep_hdrlen = exec_maxhdrsz; 830 epp->ep_hdrvalid = 0; 831 epp->ep_emul_arg = NULL; 832 epp->ep_emul_arg_free = NULL; 833 memset(&epp->ep_vmcmds, 0, sizeof(epp->ep_vmcmds)); 834 epp->ep_vap = &data->ed_attr; 835 epp->ep_flags = (p->p_flag & PK_32) ? EXEC_FROM32 : 0; 836 MD_TOPDOWN_INIT(epp); 837 epp->ep_emul_root = NULL; 838 epp->ep_interp = NULL; 839 epp->ep_esch = NULL; 840 epp->ep_pax_flags = 0; 841 memset(epp->ep_machine_arch, 0, sizeof(epp->ep_machine_arch)); 842 843 rw_enter(&exec_lock, RW_READER); 844 845 /* see if we can run it. */ 846 if ((error = check_exec(l, epp, data->ed_pathbuf, 847 &data->ed_resolvedname)) != 0) { 848 if (error != ENOENT && error != EACCES && error != ENOEXEC) { 849 DPRINTF(("%s: check exec failed for %s, error %d\n", 850 __func__, epp->ep_kname, error)); 851 } 852 goto freehdr; 853 } 854 855 /* allocate an argument buffer */ 856 data->ed_argp = pool_get(&exec_pool, PR_WAITOK); 857 KASSERT(data->ed_argp != NULL); 858 dp = data->ed_argp; 859 860 if ((error = copyinargs(data, args, envs, fetch_element, &dp)) != 0) { 861 goto bad; 862 } 863 864 /* 865 * Calculate the new stack size. 866 */ 867 868 #ifdef __MACHINE_STACK_GROWS_UP 869 /* 870 * copyargs() fills argc/argv/envp from the lower address even on 871 * __MACHINE_STACK_GROWS_UP machines. Reserve a few words just below the SP 872 * so that _rtld() use it. 873 */ 874 #define RTLD_GAP 32 875 #else 876 #define RTLD_GAP 0 877 #endif 878 879 const size_t argenvstrlen = (char *)ALIGN(dp) - data->ed_argp; 880 881 data->ed_argslen = calcargs(data, argenvstrlen); 882 883 const size_t len = calcstack(data, pax_aslr_stack_gap(epp) + RTLD_GAP); 884 885 if (len > epp->ep_ssize) { 886 /* in effect, compare to initial limit */ 887 DPRINTF(("%s: stack limit exceeded %zu\n", __func__, len)); 888 error = ENOMEM; 889 goto bad; 890 } 891 /* adjust "active stack depth" for process VSZ */ 892 epp->ep_ssize = len; 893 894 return 0; 895 896 bad: 897 /* free the vmspace-creation commands, and release their references */ 898 kill_vmcmds(&epp->ep_vmcmds); 899 /* kill any opened file descriptor, if necessary */ 900 if (epp->ep_flags & EXEC_HASFD) { 901 epp->ep_flags &= ~EXEC_HASFD; 902 fd_close(epp->ep_fd); 903 } 904 /* close and put the exec'd file */ 905 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 906 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred); 907 vput(epp->ep_vp); 908 pool_put(&exec_pool, data->ed_argp); 909 910 freehdr: 911 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 912 if (epp->ep_emul_root != NULL) 913 vrele(epp->ep_emul_root); 914 if (epp->ep_interp != NULL) 915 vrele(epp->ep_interp); 916 917 rw_exit(&exec_lock); 918 919 exec_path_free(data); 920 921 clrflg: 922 rw_exit(&p->p_reflock); 923 924 if (modgen != module_gen && error == ENOEXEC) { 925 modgen = module_gen; 926 exec_autoload(); 927 goto retry; 928 } 929 930 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 931 return error; 932 } 933 934 static int 935 execve_dovmcmds(struct lwp *l, struct execve_data * restrict data) 936 { 937 struct exec_package * const epp = &data->ed_pack; 938 struct proc *p = l->l_proc; 939 struct exec_vmcmd *base_vcp; 940 int error = 0; 941 size_t i; 942 943 /* record proc's vnode, for use by procfs and others */ 944 if (p->p_textvp) 945 vrele(p->p_textvp); 946 vref(epp->ep_vp); 947 p->p_textvp = epp->ep_vp; 948 949 /* create the new process's VM space by running the vmcmds */ 950 KASSERTMSG(epp->ep_vmcmds.evs_used != 0, "%s: no vmcmds", __func__); 951 952 #ifdef TRACE_EXEC 953 DUMPVMCMDS(epp, 0, 0); 954 #endif 955 956 base_vcp = NULL; 957 958 for (i = 0; i < epp->ep_vmcmds.evs_used && !error; i++) { 959 struct exec_vmcmd *vcp; 960 961 vcp = &epp->ep_vmcmds.evs_cmds[i]; 962 if (vcp->ev_flags & VMCMD_RELATIVE) { 963 KASSERTMSG(base_vcp != NULL, 964 "%s: relative vmcmd with no base", __func__); 965 KASSERTMSG((vcp->ev_flags & VMCMD_BASE) == 0, 966 "%s: illegal base & relative vmcmd", __func__); 967 vcp->ev_addr += base_vcp->ev_addr; 968 } 969 error = (*vcp->ev_proc)(l, vcp); 970 if (error) 971 DUMPVMCMDS(epp, i, error); 972 if (vcp->ev_flags & VMCMD_BASE) 973 base_vcp = vcp; 974 } 975 976 /* free the vmspace-creation commands, and release their references */ 977 kill_vmcmds(&epp->ep_vmcmds); 978 979 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 980 VOP_CLOSE(epp->ep_vp, FREAD, l->l_cred); 981 vput(epp->ep_vp); 982 983 /* if an error happened, deallocate and punt */ 984 if (error != 0) { 985 DPRINTF(("%s: vmcmd %zu failed: %d\n", __func__, i - 1, error)); 986 } 987 return error; 988 } 989 990 static void 991 execve_free_data(struct execve_data *data) 992 { 993 struct exec_package * const epp = &data->ed_pack; 994 995 /* free the vmspace-creation commands, and release their references */ 996 kill_vmcmds(&epp->ep_vmcmds); 997 /* kill any opened file descriptor, if necessary */ 998 if (epp->ep_flags & EXEC_HASFD) { 999 epp->ep_flags &= ~EXEC_HASFD; 1000 fd_close(epp->ep_fd); 1001 } 1002 1003 /* close and put the exec'd file */ 1004 vn_lock(epp->ep_vp, LK_EXCLUSIVE | LK_RETRY); 1005 VOP_CLOSE(epp->ep_vp, FREAD, curlwp->l_cred); 1006 vput(epp->ep_vp); 1007 pool_put(&exec_pool, data->ed_argp); 1008 1009 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1010 if (epp->ep_emul_root != NULL) 1011 vrele(epp->ep_emul_root); 1012 if (epp->ep_interp != NULL) 1013 vrele(epp->ep_interp); 1014 1015 exec_path_free(data); 1016 } 1017 1018 static void 1019 pathexec(struct proc *p, const char *resolvedname) 1020 { 1021 /* set command name & other accounting info */ 1022 const char *cmdname; 1023 1024 if (resolvedname == NULL) { 1025 cmdname = "*fexecve*"; 1026 resolvedname = "/"; 1027 } else { 1028 cmdname = strrchr(resolvedname, '/') + 1; 1029 } 1030 KASSERTMSG(resolvedname[0] == '/', "bad resolvedname `%s'", 1031 resolvedname); 1032 1033 strlcpy(p->p_comm, cmdname, sizeof(p->p_comm)); 1034 1035 kmem_strfree(p->p_path); 1036 p->p_path = kmem_strdupsize(resolvedname, NULL, KM_SLEEP); 1037 } 1038 1039 /* XXX elsewhere */ 1040 static int 1041 credexec(struct lwp *l, struct vattr *attr) 1042 { 1043 struct proc *p = l->l_proc; 1044 int error; 1045 1046 /* 1047 * Deal with set[ug]id. MNT_NOSUID has already been used to disable 1048 * s[ug]id. It's OK to check for PSL_TRACED here as we have blocked 1049 * out additional references on the process for the moment. 1050 */ 1051 if ((p->p_slflag & PSL_TRACED) == 0 && 1052 1053 (((attr->va_mode & S_ISUID) != 0 && 1054 kauth_cred_geteuid(l->l_cred) != attr->va_uid) || 1055 1056 ((attr->va_mode & S_ISGID) != 0 && 1057 kauth_cred_getegid(l->l_cred) != attr->va_gid))) { 1058 /* 1059 * Mark the process as SUGID before we do 1060 * anything that might block. 1061 */ 1062 proc_crmod_enter(); 1063 proc_crmod_leave(NULL, NULL, true); 1064 1065 /* Make sure file descriptors 0..2 are in use. */ 1066 if ((error = fd_checkstd()) != 0) { 1067 DPRINTF(("%s: fdcheckstd failed %d\n", 1068 __func__, error)); 1069 return error; 1070 } 1071 1072 /* 1073 * Copy the credential so other references don't see our 1074 * changes. 1075 */ 1076 l->l_cred = kauth_cred_copy(l->l_cred); 1077 #ifdef KTRACE 1078 /* 1079 * If the persistent trace flag isn't set, turn off. 1080 */ 1081 if (p->p_tracep) { 1082 mutex_enter(&ktrace_lock); 1083 if (!(p->p_traceflag & KTRFAC_PERSISTENT)) 1084 ktrderef(p); 1085 mutex_exit(&ktrace_lock); 1086 } 1087 #endif 1088 if (attr->va_mode & S_ISUID) 1089 kauth_cred_seteuid(l->l_cred, attr->va_uid); 1090 if (attr->va_mode & S_ISGID) 1091 kauth_cred_setegid(l->l_cred, attr->va_gid); 1092 } else { 1093 if (kauth_cred_geteuid(l->l_cred) == 1094 kauth_cred_getuid(l->l_cred) && 1095 kauth_cred_getegid(l->l_cred) == 1096 kauth_cred_getgid(l->l_cred)) 1097 p->p_flag &= ~PK_SUGID; 1098 } 1099 1100 /* 1101 * Copy the credential so other references don't see our changes. 1102 * Test to see if this is necessary first, since in the common case 1103 * we won't need a private reference. 1104 */ 1105 if (kauth_cred_geteuid(l->l_cred) != kauth_cred_getsvuid(l->l_cred) || 1106 kauth_cred_getegid(l->l_cred) != kauth_cred_getsvgid(l->l_cred)) { 1107 l->l_cred = kauth_cred_copy(l->l_cred); 1108 kauth_cred_setsvuid(l->l_cred, kauth_cred_geteuid(l->l_cred)); 1109 kauth_cred_setsvgid(l->l_cred, kauth_cred_getegid(l->l_cred)); 1110 } 1111 1112 /* Update the master credentials. */ 1113 if (l->l_cred != p->p_cred) { 1114 kauth_cred_t ocred; 1115 1116 kauth_cred_hold(l->l_cred); 1117 mutex_enter(p->p_lock); 1118 ocred = p->p_cred; 1119 p->p_cred = l->l_cred; 1120 mutex_exit(p->p_lock); 1121 kauth_cred_free(ocred); 1122 } 1123 1124 return 0; 1125 } 1126 1127 static void 1128 emulexec(struct lwp *l, struct exec_package *epp) 1129 { 1130 struct proc *p = l->l_proc; 1131 1132 /* The emulation root will usually have been found when we looked 1133 * for the elf interpreter (or similar), if not look now. */ 1134 if (epp->ep_esch->es_emul->e_path != NULL && 1135 epp->ep_emul_root == NULL) 1136 emul_find_root(l, epp); 1137 1138 /* Any old emulation root got removed by fdcloseexec */ 1139 rw_enter(&p->p_cwdi->cwdi_lock, RW_WRITER); 1140 p->p_cwdi->cwdi_edir = epp->ep_emul_root; 1141 rw_exit(&p->p_cwdi->cwdi_lock); 1142 epp->ep_emul_root = NULL; 1143 if (epp->ep_interp != NULL) 1144 vrele(epp->ep_interp); 1145 1146 /* 1147 * Call emulation specific exec hook. This can setup per-process 1148 * p->p_emuldata or do any other per-process stuff an emulation needs. 1149 * 1150 * If we are executing process of different emulation than the 1151 * original forked process, call e_proc_exit() of the old emulation 1152 * first, then e_proc_exec() of new emulation. If the emulation is 1153 * same, the exec hook code should deallocate any old emulation 1154 * resources held previously by this process. 1155 */ 1156 if (p->p_emul && p->p_emul->e_proc_exit 1157 && p->p_emul != epp->ep_esch->es_emul) 1158 (*p->p_emul->e_proc_exit)(p); 1159 1160 /* 1161 * Call exec hook. Emulation code may NOT store reference to anything 1162 * from &pack. 1163 */ 1164 if (epp->ep_esch->es_emul->e_proc_exec) 1165 (*epp->ep_esch->es_emul->e_proc_exec)(p, epp); 1166 1167 /* update p_emul, the old value is no longer needed */ 1168 p->p_emul = epp->ep_esch->es_emul; 1169 1170 /* ...and the same for p_execsw */ 1171 p->p_execsw = epp->ep_esch; 1172 1173 #ifdef __HAVE_SYSCALL_INTERN 1174 (*p->p_emul->e_syscall_intern)(p); 1175 #endif 1176 ktremul(); 1177 } 1178 1179 static int 1180 execve_runproc(struct lwp *l, struct execve_data * restrict data, 1181 bool no_local_exec_lock, bool is_spawn) 1182 { 1183 struct exec_package * const epp = &data->ed_pack; 1184 int error = 0; 1185 struct proc *p; 1186 struct vmspace *vm; 1187 1188 /* 1189 * In case of a posix_spawn operation, the child doing the exec 1190 * might not hold the reader lock on exec_lock, but the parent 1191 * will do this instead. 1192 */ 1193 KASSERT(no_local_exec_lock || rw_lock_held(&exec_lock)); 1194 KASSERT(!no_local_exec_lock || is_spawn); 1195 KASSERT(data != NULL); 1196 1197 p = l->l_proc; 1198 1199 /* Get rid of other LWPs. */ 1200 if (p->p_nlwps > 1) { 1201 mutex_enter(p->p_lock); 1202 exit_lwps(l); 1203 mutex_exit(p->p_lock); 1204 } 1205 KDASSERT(p->p_nlwps == 1); 1206 1207 /* Destroy any lwpctl info. */ 1208 if (p->p_lwpctl != NULL) 1209 lwp_ctl_exit(); 1210 1211 /* Remove POSIX timers */ 1212 ptimers_free(p, TIMERS_POSIX); 1213 1214 /* Set the PaX flags. */ 1215 pax_set_flags(epp, p); 1216 1217 /* 1218 * Do whatever is necessary to prepare the address space 1219 * for remapping. Note that this might replace the current 1220 * vmspace with another! 1221 * 1222 * vfork(): do not touch any user space data in the new child 1223 * until we have awoken the parent below, or it will defeat 1224 * lazy pmap switching (on x86). 1225 */ 1226 if (is_spawn) 1227 uvmspace_spawn(l, epp->ep_vm_minaddr, 1228 epp->ep_vm_maxaddr, 1229 epp->ep_flags & EXEC_TOPDOWN_VM); 1230 else 1231 uvmspace_exec(l, epp->ep_vm_minaddr, 1232 epp->ep_vm_maxaddr, 1233 epp->ep_flags & EXEC_TOPDOWN_VM); 1234 vm = p->p_vmspace; 1235 1236 vm->vm_taddr = (void *)epp->ep_taddr; 1237 vm->vm_tsize = btoc(epp->ep_tsize); 1238 vm->vm_daddr = (void*)epp->ep_daddr; 1239 vm->vm_dsize = btoc(epp->ep_dsize); 1240 vm->vm_ssize = btoc(epp->ep_ssize); 1241 vm->vm_issize = 0; 1242 vm->vm_maxsaddr = (void *)epp->ep_maxsaddr; 1243 vm->vm_minsaddr = (void *)epp->ep_minsaddr; 1244 1245 pax_aslr_init_vm(l, vm, epp); 1246 1247 cwdexec(p); 1248 fd_closeexec(); /* handle close on exec */ 1249 1250 if (__predict_false(ktrace_on)) 1251 fd_ktrexecfd(); 1252 1253 execsigs(p); /* reset caught signals */ 1254 1255 mutex_enter(p->p_lock); 1256 l->l_ctxlink = NULL; /* reset ucontext link */ 1257 p->p_acflag &= ~AFORK; 1258 p->p_flag |= PK_EXEC; 1259 mutex_exit(p->p_lock); 1260 1261 error = credexec(l, &data->ed_attr); 1262 if (error) 1263 goto exec_abort; 1264 1265 #if defined(__HAVE_RAS) 1266 /* 1267 * Remove all RASs from the address space. 1268 */ 1269 ras_purgeall(); 1270 #endif 1271 1272 /* 1273 * Stop profiling. 1274 */ 1275 if ((p->p_stflag & PST_PROFIL) != 0) { 1276 mutex_spin_enter(&p->p_stmutex); 1277 stopprofclock(p); 1278 mutex_spin_exit(&p->p_stmutex); 1279 } 1280 1281 /* 1282 * It's OK to test PL_PPWAIT unlocked here, as other LWPs have 1283 * exited and exec()/exit() are the only places it will be cleared. 1284 * 1285 * Once the parent has been awoken, curlwp may teleport to a new CPU 1286 * in sched_vforkexec(), and it's then OK to start messing with user 1287 * data. See comment above. 1288 */ 1289 if ((p->p_lflag & PL_PPWAIT) != 0) { 1290 bool samecpu; 1291 lwp_t *lp; 1292 1293 mutex_enter(&proc_lock); 1294 lp = p->p_vforklwp; 1295 p->p_vforklwp = NULL; 1296 l->l_lwpctl = NULL; /* was on loan from blocked parent */ 1297 cv_broadcast(&lp->l_waitcv); 1298 1299 /* Clear flags after cv_broadcast() (scheduler needs them). */ 1300 p->p_lflag &= ~PL_PPWAIT; 1301 lp->l_vforkwaiting = false; 1302 1303 /* If parent is still on same CPU, teleport curlwp elsewhere. */ 1304 samecpu = (lp->l_cpu == curlwp->l_cpu); 1305 mutex_exit(&proc_lock); 1306 1307 /* Give the parent its CPU back - find a new home. */ 1308 KASSERT(!is_spawn); 1309 sched_vforkexec(l, samecpu); 1310 } 1311 1312 /* Now map address space. */ 1313 error = execve_dovmcmds(l, data); 1314 if (error != 0) 1315 goto exec_abort; 1316 1317 pathexec(p, epp->ep_resolvedname); 1318 1319 char * const newstack = STACK_GROW(vm->vm_minsaddr, epp->ep_ssize); 1320 1321 error = copyoutargs(data, l, newstack); 1322 if (error != 0) 1323 goto exec_abort; 1324 1325 doexechooks(p); 1326 1327 /* 1328 * Set initial SP at the top of the stack. 1329 * 1330 * Note that on machines where stack grows up (e.g. hppa), SP points to 1331 * the end of arg/env strings. Userland guesses the address of argc 1332 * via ps_strings::ps_argvstr. 1333 */ 1334 1335 /* Setup new registers and do misc. setup. */ 1336 (*epp->ep_esch->es_emul->e_setregs)(l, epp, (vaddr_t)newstack); 1337 if (epp->ep_esch->es_setregs) 1338 (*epp->ep_esch->es_setregs)(l, epp, (vaddr_t)newstack); 1339 1340 /* Provide a consistent LWP private setting */ 1341 (void)lwp_setprivate(l, NULL); 1342 1343 /* Discard all PCU state; need to start fresh */ 1344 pcu_discard_all(l); 1345 1346 /* map the process's signal trampoline code */ 1347 if ((error = exec_sigcode_map(p, epp->ep_esch->es_emul)) != 0) { 1348 DPRINTF(("%s: map sigcode failed %d\n", __func__, error)); 1349 goto exec_abort; 1350 } 1351 1352 pool_put(&exec_pool, data->ed_argp); 1353 1354 /* notify others that we exec'd */ 1355 KNOTE(&p->p_klist, NOTE_EXEC); 1356 1357 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1358 1359 SDT_PROBE(proc, kernel, , exec__success, epp->ep_kname, 0, 0, 0, 0); 1360 1361 emulexec(l, epp); 1362 1363 /* Allow new references from the debugger/procfs. */ 1364 rw_exit(&p->p_reflock); 1365 if (!no_local_exec_lock) 1366 rw_exit(&exec_lock); 1367 1368 mutex_enter(&proc_lock); 1369 1370 /* posix_spawn(3) reports a single event with implied exec(3) */ 1371 if ((p->p_slflag & PSL_TRACED) && !is_spawn) { 1372 mutex_enter(p->p_lock); 1373 eventswitch(TRAP_EXEC, 0, 0); 1374 mutex_enter(&proc_lock); 1375 } 1376 1377 if (p->p_sflag & PS_STOPEXEC) { 1378 ksiginfoq_t kq; 1379 1380 KERNEL_UNLOCK_ALL(l, &l->l_biglocks); 1381 p->p_pptr->p_nstopchild++; 1382 p->p_waited = 0; 1383 mutex_enter(p->p_lock); 1384 ksiginfo_queue_init(&kq); 1385 sigclearall(p, &contsigmask, &kq); 1386 lwp_lock(l); 1387 l->l_stat = LSSTOP; 1388 p->p_stat = SSTOP; 1389 p->p_nrlwps--; 1390 lwp_unlock(l); 1391 mutex_exit(p->p_lock); 1392 mutex_exit(&proc_lock); 1393 lwp_lock(l); 1394 spc_lock(l->l_cpu); 1395 mi_switch(l); 1396 ksiginfo_queue_drain(&kq); 1397 } else { 1398 mutex_exit(&proc_lock); 1399 } 1400 1401 exec_path_free(data); 1402 #ifdef TRACE_EXEC 1403 DPRINTF(("%s finished\n", __func__)); 1404 #endif 1405 return EJUSTRETURN; 1406 1407 exec_abort: 1408 SDT_PROBE(proc, kernel, , exec__failure, error, 0, 0, 0, 0); 1409 rw_exit(&p->p_reflock); 1410 if (!no_local_exec_lock) 1411 rw_exit(&exec_lock); 1412 1413 exec_path_free(data); 1414 1415 /* 1416 * the old process doesn't exist anymore. exit gracefully. 1417 * get rid of the (new) address space we have created, if any, get rid 1418 * of our namei data and vnode, and exit noting failure 1419 */ 1420 if (vm != NULL) { 1421 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 1422 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 1423 } 1424 1425 exec_free_emul_arg(epp); 1426 pool_put(&exec_pool, data->ed_argp); 1427 kmem_free(epp->ep_hdr, epp->ep_hdrlen); 1428 if (epp->ep_emul_root != NULL) 1429 vrele(epp->ep_emul_root); 1430 if (epp->ep_interp != NULL) 1431 vrele(epp->ep_interp); 1432 1433 /* Acquire the sched-state mutex (exit1() will release it). */ 1434 if (!is_spawn) { 1435 mutex_enter(p->p_lock); 1436 exit1(l, error, SIGABRT); 1437 } 1438 1439 return error; 1440 } 1441 1442 int 1443 execve1(struct lwp *l, bool has_path, const char *path, int fd, 1444 char * const *args, char * const *envs, 1445 execve_fetch_element_t fetch_element) 1446 { 1447 struct execve_data data; 1448 int error; 1449 1450 error = execve_loadvm(l, has_path, path, fd, args, envs, fetch_element, 1451 &data); 1452 if (error) 1453 return error; 1454 error = execve_runproc(l, &data, false, false); 1455 return error; 1456 } 1457 1458 static size_t 1459 fromptrsz(const struct exec_package *epp) 1460 { 1461 return (epp->ep_flags & EXEC_FROM32) ? sizeof(int) : sizeof(char *); 1462 } 1463 1464 static size_t 1465 ptrsz(const struct exec_package *epp) 1466 { 1467 return (epp->ep_flags & EXEC_32) ? sizeof(int) : sizeof(char *); 1468 } 1469 1470 static size_t 1471 calcargs(struct execve_data * restrict data, const size_t argenvstrlen) 1472 { 1473 struct exec_package * const epp = &data->ed_pack; 1474 1475 const size_t nargenvptrs = 1476 1 + /* long argc */ 1477 data->ed_argc + /* char *argv[] */ 1478 1 + /* \0 */ 1479 data->ed_envc + /* char *env[] */ 1480 1; /* \0 */ 1481 1482 return (nargenvptrs * ptrsz(epp)) /* pointers */ 1483 + argenvstrlen /* strings */ 1484 + epp->ep_esch->es_arglen; /* auxinfo */ 1485 } 1486 1487 static size_t 1488 calcstack(struct execve_data * restrict data, const size_t gaplen) 1489 { 1490 struct exec_package * const epp = &data->ed_pack; 1491 1492 data->ed_szsigcode = epp->ep_esch->es_emul->e_esigcode - 1493 epp->ep_esch->es_emul->e_sigcode; 1494 1495 data->ed_ps_strings_sz = (epp->ep_flags & EXEC_32) ? 1496 sizeof(struct ps_strings32) : sizeof(struct ps_strings); 1497 1498 const size_t sigcode_psstr_sz = 1499 data->ed_szsigcode + /* sigcode */ 1500 data->ed_ps_strings_sz + /* ps_strings */ 1501 STACK_PTHREADSPACE; /* pthread space */ 1502 1503 const size_t stacklen = 1504 data->ed_argslen + 1505 gaplen + 1506 sigcode_psstr_sz; 1507 1508 /* make the stack "safely" aligned */ 1509 return STACK_LEN_ALIGN(stacklen, STACK_ALIGNBYTES); 1510 } 1511 1512 static int 1513 copyoutargs(struct execve_data * restrict data, struct lwp *l, 1514 char * const newstack) 1515 { 1516 struct exec_package * const epp = &data->ed_pack; 1517 struct proc *p = l->l_proc; 1518 int error; 1519 1520 memset(&data->ed_arginfo, 0, sizeof(data->ed_arginfo)); 1521 1522 /* remember information about the process */ 1523 data->ed_arginfo.ps_nargvstr = data->ed_argc; 1524 data->ed_arginfo.ps_nenvstr = data->ed_envc; 1525 1526 /* 1527 * Allocate the stack address passed to the newly execve()'ed process. 1528 * 1529 * The new stack address will be set to the SP (stack pointer) register 1530 * in setregs(). 1531 */ 1532 1533 char *newargs = STACK_ALLOC( 1534 STACK_SHRINK(newstack, data->ed_argslen), data->ed_argslen); 1535 1536 error = (*epp->ep_esch->es_copyargs)(l, epp, 1537 &data->ed_arginfo, &newargs, data->ed_argp); 1538 1539 if (error) { 1540 DPRINTF(("%s: copyargs failed %d\n", __func__, error)); 1541 return error; 1542 } 1543 1544 error = copyoutpsstrs(data, p); 1545 if (error != 0) 1546 return error; 1547 1548 return 0; 1549 } 1550 1551 static int 1552 copyoutpsstrs(struct execve_data * restrict data, struct proc *p) 1553 { 1554 struct exec_package * const epp = &data->ed_pack; 1555 struct ps_strings32 arginfo32; 1556 void *aip; 1557 int error; 1558 1559 /* fill process ps_strings info */ 1560 p->p_psstrp = (vaddr_t)STACK_ALLOC(STACK_GROW(epp->ep_minsaddr, 1561 STACK_PTHREADSPACE), data->ed_ps_strings_sz); 1562 1563 if (epp->ep_flags & EXEC_32) { 1564 aip = &arginfo32; 1565 arginfo32.ps_argvstr = (vaddr_t)data->ed_arginfo.ps_argvstr; 1566 arginfo32.ps_nargvstr = data->ed_arginfo.ps_nargvstr; 1567 arginfo32.ps_envstr = (vaddr_t)data->ed_arginfo.ps_envstr; 1568 arginfo32.ps_nenvstr = data->ed_arginfo.ps_nenvstr; 1569 } else 1570 aip = &data->ed_arginfo; 1571 1572 /* copy out the process's ps_strings structure */ 1573 if ((error = copyout(aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)) 1574 != 0) { 1575 DPRINTF(("%s: ps_strings copyout %p->%p size %zu failed\n", 1576 __func__, aip, (void *)p->p_psstrp, data->ed_ps_strings_sz)); 1577 return error; 1578 } 1579 1580 return 0; 1581 } 1582 1583 static int 1584 copyinargs(struct execve_data * restrict data, char * const *args, 1585 char * const *envs, execve_fetch_element_t fetch_element, char **dpp) 1586 { 1587 struct exec_package * const epp = &data->ed_pack; 1588 char *dp; 1589 size_t i; 1590 int error; 1591 1592 dp = *dpp; 1593 1594 data->ed_argc = 0; 1595 1596 /* copy the fake args list, if there's one, freeing it as we go */ 1597 if (epp->ep_flags & EXEC_HASARGL) { 1598 struct exec_fakearg *fa = epp->ep_fa; 1599 1600 while (fa->fa_arg != NULL) { 1601 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1602 size_t len; 1603 1604 len = strlcpy(dp, fa->fa_arg, maxlen); 1605 /* Count NUL into len. */ 1606 if (len < maxlen) 1607 len++; 1608 else { 1609 while (fa->fa_arg != NULL) { 1610 kmem_free(fa->fa_arg, fa->fa_len); 1611 fa++; 1612 } 1613 kmem_free(epp->ep_fa, epp->ep_fa_len); 1614 epp->ep_flags &= ~EXEC_HASARGL; 1615 return E2BIG; 1616 } 1617 ktrexecarg(fa->fa_arg, len - 1); 1618 dp += len; 1619 1620 kmem_free(fa->fa_arg, fa->fa_len); 1621 fa++; 1622 data->ed_argc++; 1623 } 1624 kmem_free(epp->ep_fa, epp->ep_fa_len); 1625 epp->ep_flags &= ~EXEC_HASARGL; 1626 } 1627 1628 /* 1629 * Read and count argument strings from user. 1630 */ 1631 1632 if (args == NULL) { 1633 DPRINTF(("%s: null args\n", __func__)); 1634 return EINVAL; 1635 } 1636 if (epp->ep_flags & EXEC_SKIPARG) 1637 args = (const void *)((const char *)args + fromptrsz(epp)); 1638 i = 0; 1639 error = copyinargstrs(data, args, fetch_element, &dp, &i, ktr_execarg); 1640 if (error != 0) { 1641 DPRINTF(("%s: copyin arg %d\n", __func__, error)); 1642 return error; 1643 } 1644 data->ed_argc += i; 1645 1646 /* 1647 * Read and count environment strings from user. 1648 */ 1649 1650 data->ed_envc = 0; 1651 /* environment need not be there */ 1652 if (envs == NULL) 1653 goto done; 1654 i = 0; 1655 error = copyinargstrs(data, envs, fetch_element, &dp, &i, ktr_execenv); 1656 if (error != 0) { 1657 DPRINTF(("%s: copyin env %d\n", __func__, error)); 1658 return error; 1659 } 1660 data->ed_envc += i; 1661 1662 done: 1663 *dpp = dp; 1664 1665 return 0; 1666 } 1667 1668 static int 1669 copyinargstrs(struct execve_data * restrict data, char * const *strs, 1670 execve_fetch_element_t fetch_element, char **dpp, size_t *ip, 1671 void (*ktr)(const void *, size_t)) 1672 { 1673 char *dp, *sp; 1674 size_t i; 1675 int error; 1676 1677 dp = *dpp; 1678 1679 i = 0; 1680 while (1) { 1681 const size_t maxlen = ARG_MAX - (dp - data->ed_argp); 1682 size_t len; 1683 1684 if ((error = (*fetch_element)(strs, i, &sp)) != 0) { 1685 return error; 1686 } 1687 if (!sp) 1688 break; 1689 if ((error = copyinstr(sp, dp, maxlen, &len)) != 0) { 1690 if (error == ENAMETOOLONG) 1691 error = E2BIG; 1692 return error; 1693 } 1694 if (__predict_false(ktrace_on)) 1695 (*ktr)(dp, len - 1); 1696 dp += len; 1697 i++; 1698 } 1699 1700 *dpp = dp; 1701 *ip = i; 1702 1703 return 0; 1704 } 1705 1706 /* 1707 * Copy argv and env strings from kernel buffer (argp) to the new stack. 1708 * Those strings are located just after auxinfo. 1709 */ 1710 int 1711 copyargs(struct lwp *l, struct exec_package *pack, struct ps_strings *arginfo, 1712 char **stackp, void *argp) 1713 { 1714 char **cpp, *dp, *sp; 1715 size_t len; 1716 void *nullp; 1717 long argc, envc; 1718 int error; 1719 1720 cpp = (char **)*stackp; 1721 nullp = NULL; 1722 argc = arginfo->ps_nargvstr; 1723 envc = arginfo->ps_nenvstr; 1724 1725 /* argc on stack is long */ 1726 CTASSERT(sizeof(*cpp) == sizeof(argc)); 1727 1728 dp = (char *)(cpp + 1729 1 + /* long argc */ 1730 argc + /* char *argv[] */ 1731 1 + /* \0 */ 1732 envc + /* char *env[] */ 1733 1) + /* \0 */ 1734 pack->ep_esch->es_arglen; /* auxinfo */ 1735 sp = argp; 1736 1737 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) { 1738 COPYPRINTF("", cpp - 1, sizeof(argc)); 1739 return error; 1740 } 1741 1742 /* XXX don't copy them out, remap them! */ 1743 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 1744 1745 for (; --argc >= 0; sp += len, dp += len) { 1746 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1747 COPYPRINTF("", cpp - 1, sizeof(dp)); 1748 return error; 1749 } 1750 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1751 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1752 return error; 1753 } 1754 } 1755 1756 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1757 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1758 return error; 1759 } 1760 1761 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 1762 1763 for (; --envc >= 0; sp += len, dp += len) { 1764 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0) { 1765 COPYPRINTF("", cpp - 1, sizeof(dp)); 1766 return error; 1767 } 1768 if ((error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) { 1769 COPYPRINTF("str", dp, (size_t)ARG_MAX); 1770 return error; 1771 } 1772 1773 } 1774 1775 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) { 1776 COPYPRINTF("", cpp - 1, sizeof(nullp)); 1777 return error; 1778 } 1779 1780 *stackp = (char *)cpp; 1781 return 0; 1782 } 1783 1784 1785 /* 1786 * Add execsw[] entries. 1787 */ 1788 int 1789 exec_add(struct execsw *esp, int count) 1790 { 1791 struct exec_entry *it; 1792 int i; 1793 1794 if (count == 0) { 1795 return 0; 1796 } 1797 1798 /* Check for duplicates. */ 1799 rw_enter(&exec_lock, RW_WRITER); 1800 for (i = 0; i < count; i++) { 1801 LIST_FOREACH(it, &ex_head, ex_list) { 1802 /* assume unique (makecmds, probe_func, emulation) */ 1803 if (it->ex_sw->es_makecmds == esp[i].es_makecmds && 1804 it->ex_sw->u.elf_probe_func == 1805 esp[i].u.elf_probe_func && 1806 it->ex_sw->es_emul == esp[i].es_emul) { 1807 rw_exit(&exec_lock); 1808 return EEXIST; 1809 } 1810 } 1811 } 1812 1813 /* Allocate new entries. */ 1814 for (i = 0; i < count; i++) { 1815 it = kmem_alloc(sizeof(*it), KM_SLEEP); 1816 it->ex_sw = &esp[i]; 1817 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1818 } 1819 1820 /* update execsw[] */ 1821 exec_init(0); 1822 rw_exit(&exec_lock); 1823 return 0; 1824 } 1825 1826 /* 1827 * Remove execsw[] entry. 1828 */ 1829 int 1830 exec_remove(struct execsw *esp, int count) 1831 { 1832 struct exec_entry *it, *next; 1833 int i; 1834 const struct proclist_desc *pd; 1835 proc_t *p; 1836 1837 if (count == 0) { 1838 return 0; 1839 } 1840 1841 /* Abort if any are busy. */ 1842 rw_enter(&exec_lock, RW_WRITER); 1843 for (i = 0; i < count; i++) { 1844 mutex_enter(&proc_lock); 1845 for (pd = proclists; pd->pd_list != NULL; pd++) { 1846 PROCLIST_FOREACH(p, pd->pd_list) { 1847 if (p->p_execsw == &esp[i]) { 1848 mutex_exit(&proc_lock); 1849 rw_exit(&exec_lock); 1850 return EBUSY; 1851 } 1852 } 1853 } 1854 mutex_exit(&proc_lock); 1855 } 1856 1857 /* None are busy, so remove them all. */ 1858 for (i = 0; i < count; i++) { 1859 for (it = LIST_FIRST(&ex_head); it != NULL; it = next) { 1860 next = LIST_NEXT(it, ex_list); 1861 if (it->ex_sw == &esp[i]) { 1862 LIST_REMOVE(it, ex_list); 1863 kmem_free(it, sizeof(*it)); 1864 break; 1865 } 1866 } 1867 } 1868 1869 /* update execsw[] */ 1870 exec_init(0); 1871 rw_exit(&exec_lock); 1872 return 0; 1873 } 1874 1875 /* 1876 * Initialize exec structures. If init_boot is true, also does necessary 1877 * one-time initialization (it's called from main() that way). 1878 * Once system is multiuser, this should be called with exec_lock held, 1879 * i.e. via exec_{add|remove}(). 1880 */ 1881 int 1882 exec_init(int init_boot) 1883 { 1884 const struct execsw **sw; 1885 struct exec_entry *ex; 1886 SLIST_HEAD(,exec_entry) first; 1887 SLIST_HEAD(,exec_entry) any; 1888 SLIST_HEAD(,exec_entry) last; 1889 int i, sz; 1890 1891 if (init_boot) { 1892 /* do one-time initializations */ 1893 vaddr_t vmin = 0, vmax; 1894 1895 rw_init(&exec_lock); 1896 mutex_init(&sigobject_lock, MUTEX_DEFAULT, IPL_NONE); 1897 exec_map = uvm_km_suballoc(kernel_map, &vmin, &vmax, 1898 maxexec*NCARGS, VM_MAP_PAGEABLE, false, NULL); 1899 pool_init(&exec_pool, NCARGS, 0, 0, PR_NOALIGN|PR_NOTOUCH, 1900 "execargs", &exec_palloc, IPL_NONE); 1901 pool_sethardlimit(&exec_pool, maxexec, "should not happen", 0); 1902 } else { 1903 KASSERT(rw_write_held(&exec_lock)); 1904 } 1905 1906 /* Sort each entry onto the appropriate queue. */ 1907 SLIST_INIT(&first); 1908 SLIST_INIT(&any); 1909 SLIST_INIT(&last); 1910 sz = 0; 1911 LIST_FOREACH(ex, &ex_head, ex_list) { 1912 switch(ex->ex_sw->es_prio) { 1913 case EXECSW_PRIO_FIRST: 1914 SLIST_INSERT_HEAD(&first, ex, ex_slist); 1915 break; 1916 case EXECSW_PRIO_ANY: 1917 SLIST_INSERT_HEAD(&any, ex, ex_slist); 1918 break; 1919 case EXECSW_PRIO_LAST: 1920 SLIST_INSERT_HEAD(&last, ex, ex_slist); 1921 break; 1922 default: 1923 panic("%s", __func__); 1924 break; 1925 } 1926 sz++; 1927 } 1928 1929 /* 1930 * Create new execsw[]. Ensure we do not try a zero-sized 1931 * allocation. 1932 */ 1933 sw = kmem_alloc(sz * sizeof(struct execsw *) + 1, KM_SLEEP); 1934 i = 0; 1935 SLIST_FOREACH(ex, &first, ex_slist) { 1936 sw[i++] = ex->ex_sw; 1937 } 1938 SLIST_FOREACH(ex, &any, ex_slist) { 1939 sw[i++] = ex->ex_sw; 1940 } 1941 SLIST_FOREACH(ex, &last, ex_slist) { 1942 sw[i++] = ex->ex_sw; 1943 } 1944 1945 /* Replace old execsw[] and free used memory. */ 1946 if (execsw != NULL) { 1947 kmem_free(__UNCONST(execsw), 1948 nexecs * sizeof(struct execsw *) + 1); 1949 } 1950 execsw = sw; 1951 nexecs = sz; 1952 1953 /* Figure out the maximum size of an exec header. */ 1954 exec_maxhdrsz = sizeof(int); 1955 for (i = 0; i < nexecs; i++) { 1956 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1957 exec_maxhdrsz = execsw[i]->es_hdrsz; 1958 } 1959 1960 return 0; 1961 } 1962 1963 static int 1964 exec_sigcode_map(struct proc *p, const struct emul *e) 1965 { 1966 vaddr_t va; 1967 vsize_t sz; 1968 int error; 1969 struct uvm_object *uobj; 1970 1971 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1972 1973 if (e->e_sigobject == NULL || sz == 0) { 1974 return 0; 1975 } 1976 1977 /* 1978 * If we don't have a sigobject for this emulation, create one. 1979 * 1980 * sigobject is an anonymous memory object (just like SYSV shared 1981 * memory) that we keep a permanent reference to and that we map 1982 * in all processes that need this sigcode. The creation is simple, 1983 * we create an object, add a permanent reference to it, map it in 1984 * kernel space, copy out the sigcode to it and unmap it. 1985 * We map it with PROT_READ|PROT_EXEC into the process just 1986 * the way sys_mmap() would map it. 1987 */ 1988 1989 uobj = *e->e_sigobject; 1990 if (uobj == NULL) { 1991 mutex_enter(&sigobject_lock); 1992 if ((uobj = *e->e_sigobject) == NULL) { 1993 uobj = uao_create(sz, 0); 1994 (*uobj->pgops->pgo_reference)(uobj); 1995 va = vm_map_min(kernel_map); 1996 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1997 uobj, 0, 0, 1998 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1999 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 2000 printf("kernel mapping failed %d\n", error); 2001 (*uobj->pgops->pgo_detach)(uobj); 2002 mutex_exit(&sigobject_lock); 2003 return error; 2004 } 2005 memcpy((void *)va, e->e_sigcode, sz); 2006 #ifdef PMAP_NEED_PROCWR 2007 pmap_procwr(&proc0, va, sz); 2008 #endif 2009 uvm_unmap(kernel_map, va, va + round_page(sz)); 2010 *e->e_sigobject = uobj; 2011 } 2012 mutex_exit(&sigobject_lock); 2013 } 2014 2015 /* Just a hint to uvm_map where to put it. */ 2016 va = e->e_vm_default_addr(p, (vaddr_t)p->p_vmspace->vm_daddr, 2017 round_page(sz), p->p_vmspace->vm_map.flags & VM_MAP_TOPDOWN); 2018 2019 #ifdef __alpha__ 2020 /* 2021 * Tru64 puts /sbin/loader at the end of user virtual memory, 2022 * which causes the above calculation to put the sigcode at 2023 * an invalid address. Put it just below the text instead. 2024 */ 2025 if (va == (vaddr_t)vm_map_max(&p->p_vmspace->vm_map)) { 2026 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 2027 } 2028 #endif 2029 2030 (*uobj->pgops->pgo_reference)(uobj); 2031 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 2032 uobj, 0, 0, 2033 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 2034 UVM_ADV_RANDOM, 0)); 2035 if (error) { 2036 DPRINTF(("%s, %d: map %p " 2037 "uvm_map %#"PRIxVSIZE"@%#"PRIxVADDR" failed %d\n", 2038 __func__, __LINE__, &p->p_vmspace->vm_map, round_page(sz), 2039 va, error)); 2040 (*uobj->pgops->pgo_detach)(uobj); 2041 return error; 2042 } 2043 p->p_sigctx.ps_sigcode = (void *)va; 2044 return 0; 2045 } 2046 2047 /* 2048 * Release a refcount on spawn_exec_data and destroy memory, if this 2049 * was the last one. 2050 */ 2051 static void 2052 spawn_exec_data_release(struct spawn_exec_data *data) 2053 { 2054 if (atomic_dec_32_nv(&data->sed_refcnt) != 0) 2055 return; 2056 2057 cv_destroy(&data->sed_cv_child_ready); 2058 mutex_destroy(&data->sed_mtx_child); 2059 2060 if (data->sed_actions) 2061 posix_spawn_fa_free(data->sed_actions, 2062 data->sed_actions->len); 2063 if (data->sed_attrs) 2064 kmem_free(data->sed_attrs, 2065 sizeof(*data->sed_attrs)); 2066 kmem_free(data, sizeof(*data)); 2067 } 2068 2069 static int 2070 handle_posix_spawn_file_actions(struct posix_spawn_file_actions *actions) 2071 { 2072 struct lwp *l = curlwp; 2073 register_t retval; 2074 int error, newfd; 2075 2076 if (actions == NULL) 2077 return 0; 2078 2079 for (size_t i = 0; i < actions->len; i++) { 2080 const struct posix_spawn_file_actions_entry *fae = 2081 &actions->fae[i]; 2082 switch (fae->fae_action) { 2083 case FAE_OPEN: 2084 if (fd_getfile(fae->fae_fildes) != NULL) { 2085 error = fd_close(fae->fae_fildes); 2086 if (error) 2087 return error; 2088 } 2089 error = fd_open(fae->fae_path, fae->fae_oflag, 2090 fae->fae_mode, &newfd); 2091 if (error) 2092 return error; 2093 if (newfd != fae->fae_fildes) { 2094 error = dodup(l, newfd, 2095 fae->fae_fildes, 0, &retval); 2096 if (fd_getfile(newfd) != NULL) 2097 fd_close(newfd); 2098 } 2099 break; 2100 case FAE_DUP2: 2101 error = dodup(l, fae->fae_fildes, 2102 fae->fae_newfildes, 0, &retval); 2103 break; 2104 case FAE_CLOSE: 2105 if (fd_getfile(fae->fae_fildes) == NULL) { 2106 return EBADF; 2107 } 2108 error = fd_close(fae->fae_fildes); 2109 break; 2110 } 2111 if (error) 2112 return error; 2113 } 2114 return 0; 2115 } 2116 2117 static int 2118 handle_posix_spawn_attrs(struct posix_spawnattr *attrs, struct proc *parent) 2119 { 2120 struct sigaction sigact; 2121 int error; 2122 struct proc *p = curproc; 2123 struct lwp *l = curlwp; 2124 2125 if (attrs == NULL) 2126 return 0; 2127 2128 memset(&sigact, 0, sizeof(sigact)); 2129 sigact._sa_u._sa_handler = SIG_DFL; 2130 sigact.sa_flags = 0; 2131 2132 /* 2133 * set state to SSTOP so that this proc can be found by pid. 2134 * see proc_enterprp, do_sched_setparam below 2135 */ 2136 mutex_enter(&proc_lock); 2137 /* 2138 * p_stat should be SACTIVE, so we need to adjust the 2139 * parent's p_nstopchild here. For safety, just make 2140 * we're on the good side of SDEAD before we adjust. 2141 */ 2142 int ostat = p->p_stat; 2143 KASSERT(ostat < SSTOP); 2144 p->p_stat = SSTOP; 2145 p->p_waited = 0; 2146 p->p_pptr->p_nstopchild++; 2147 mutex_exit(&proc_lock); 2148 2149 /* Set process group */ 2150 if (attrs->sa_flags & POSIX_SPAWN_SETPGROUP) { 2151 pid_t mypid = p->p_pid; 2152 pid_t pgrp = attrs->sa_pgroup; 2153 2154 if (pgrp == 0) 2155 pgrp = mypid; 2156 2157 error = proc_enterpgrp(parent, mypid, pgrp, false); 2158 if (error) 2159 goto out; 2160 } 2161 2162 /* Set scheduler policy */ 2163 if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDULER) 2164 error = do_sched_setparam(p->p_pid, 0, attrs->sa_schedpolicy, 2165 &attrs->sa_schedparam); 2166 else if (attrs->sa_flags & POSIX_SPAWN_SETSCHEDPARAM) { 2167 error = do_sched_setparam(parent->p_pid, 0, 2168 SCHED_NONE, &attrs->sa_schedparam); 2169 } 2170 if (error) 2171 goto out; 2172 2173 /* Reset user ID's */ 2174 if (attrs->sa_flags & POSIX_SPAWN_RESETIDS) { 2175 error = do_setresgid(l, -1, kauth_cred_getgid(l->l_cred), -1, 2176 ID_E_EQ_R | ID_E_EQ_S); 2177 if (error) 2178 return error; 2179 error = do_setresuid(l, -1, kauth_cred_getuid(l->l_cred), -1, 2180 ID_E_EQ_R | ID_E_EQ_S); 2181 if (error) 2182 goto out; 2183 } 2184 2185 /* Set signal masks/defaults */ 2186 if (attrs->sa_flags & POSIX_SPAWN_SETSIGMASK) { 2187 mutex_enter(p->p_lock); 2188 error = sigprocmask1(l, SIG_SETMASK, &attrs->sa_sigmask, NULL); 2189 mutex_exit(p->p_lock); 2190 if (error) 2191 goto out; 2192 } 2193 2194 if (attrs->sa_flags & POSIX_SPAWN_SETSIGDEF) { 2195 /* 2196 * The following sigaction call is using a sigaction 2197 * version 0 trampoline which is in the compatibility 2198 * code only. This is not a problem because for SIG_DFL 2199 * and SIG_IGN, the trampolines are now ignored. If they 2200 * were not, this would be a problem because we are 2201 * holding the exec_lock, and the compat code needs 2202 * to do the same in order to replace the trampoline 2203 * code of the process. 2204 */ 2205 for (int i = 1; i <= NSIG; i++) { 2206 if (sigismember(&attrs->sa_sigdefault, i)) 2207 sigaction1(l, i, &sigact, NULL, NULL, 0); 2208 } 2209 } 2210 error = 0; 2211 out: 2212 mutex_enter(&proc_lock); 2213 p->p_stat = ostat; 2214 p->p_pptr->p_nstopchild--; 2215 mutex_exit(&proc_lock); 2216 return error; 2217 } 2218 2219 /* 2220 * A child lwp of a posix_spawn operation starts here and ends up in 2221 * cpu_spawn_return, dealing with all filedescriptor and scheduler 2222 * manipulations in between. 2223 * The parent waits for the child, as it is not clear whether the child 2224 * will be able to acquire its own exec_lock. If it can, the parent can 2225 * be released early and continue running in parallel. If not (or if the 2226 * magic debug flag is passed in the scheduler attribute struct), the 2227 * child rides on the parent's exec lock until it is ready to return to 2228 * to userland - and only then releases the parent. This method loses 2229 * concurrency, but improves error reporting. 2230 */ 2231 static void 2232 spawn_return(void *arg) 2233 { 2234 struct spawn_exec_data *spawn_data = arg; 2235 struct lwp *l = curlwp; 2236 struct proc *p = l->l_proc; 2237 int error; 2238 bool have_reflock; 2239 bool parent_is_waiting = true; 2240 2241 /* 2242 * Check if we can release parent early. 2243 * We either need to have no sed_attrs, or sed_attrs does not 2244 * have POSIX_SPAWN_RETURNERROR or one of the flags, that require 2245 * safe access to the parent proc (passed in sed_parent). 2246 * We then try to get the exec_lock, and only if that works, we can 2247 * release the parent here already. 2248 */ 2249 struct posix_spawnattr *attrs = spawn_data->sed_attrs; 2250 if ((!attrs || (attrs->sa_flags 2251 & (POSIX_SPAWN_RETURNERROR|POSIX_SPAWN_SETPGROUP)) == 0) 2252 && rw_tryenter(&exec_lock, RW_READER)) { 2253 parent_is_waiting = false; 2254 mutex_enter(&spawn_data->sed_mtx_child); 2255 cv_signal(&spawn_data->sed_cv_child_ready); 2256 mutex_exit(&spawn_data->sed_mtx_child); 2257 } 2258 2259 /* don't allow debugger access yet */ 2260 rw_enter(&p->p_reflock, RW_WRITER); 2261 have_reflock = true; 2262 2263 /* handle posix_spawn_file_actions */ 2264 error = handle_posix_spawn_file_actions(spawn_data->sed_actions); 2265 if (error) 2266 goto report_error; 2267 2268 /* handle posix_spawnattr */ 2269 error = handle_posix_spawn_attrs(attrs, spawn_data->sed_parent); 2270 if (error) 2271 goto report_error; 2272 2273 /* now do the real exec */ 2274 error = execve_runproc(l, &spawn_data->sed_exec, parent_is_waiting, 2275 true); 2276 have_reflock = false; 2277 if (error == EJUSTRETURN) 2278 error = 0; 2279 else if (error) 2280 goto report_error; 2281 2282 if (parent_is_waiting) { 2283 mutex_enter(&spawn_data->sed_mtx_child); 2284 cv_signal(&spawn_data->sed_cv_child_ready); 2285 mutex_exit(&spawn_data->sed_mtx_child); 2286 } 2287 2288 /* release our refcount on the data */ 2289 spawn_exec_data_release(spawn_data); 2290 2291 if ((p->p_slflag & (PSL_TRACED|PSL_TRACEDCHILD)) == 2292 (PSL_TRACED|PSL_TRACEDCHILD)) { 2293 eventswitchchild(p, TRAP_CHLD, PTRACE_POSIX_SPAWN); 2294 } 2295 2296 /* and finally: leave to userland for the first time */ 2297 cpu_spawn_return(l); 2298 2299 /* NOTREACHED */ 2300 return; 2301 2302 report_error: 2303 if (have_reflock) { 2304 /* 2305 * We have not passed through execve_runproc(), 2306 * which would have released the p_reflock and also 2307 * taken ownership of the sed_exec part of spawn_data, 2308 * so release/free both here. 2309 */ 2310 rw_exit(&p->p_reflock); 2311 execve_free_data(&spawn_data->sed_exec); 2312 } 2313 2314 if (parent_is_waiting) { 2315 /* pass error to parent */ 2316 mutex_enter(&spawn_data->sed_mtx_child); 2317 spawn_data->sed_error = error; 2318 cv_signal(&spawn_data->sed_cv_child_ready); 2319 mutex_exit(&spawn_data->sed_mtx_child); 2320 } else { 2321 rw_exit(&exec_lock); 2322 } 2323 2324 /* release our refcount on the data */ 2325 spawn_exec_data_release(spawn_data); 2326 2327 /* done, exit */ 2328 mutex_enter(p->p_lock); 2329 /* 2330 * Posix explicitly asks for an exit code of 127 if we report 2331 * errors from the child process - so, unfortunately, there 2332 * is no way to report a more exact error code. 2333 * A NetBSD specific workaround is POSIX_SPAWN_RETURNERROR as 2334 * flag bit in the attrp argument to posix_spawn(2), see above. 2335 */ 2336 exit1(l, 127, 0); 2337 } 2338 2339 void 2340 posix_spawn_fa_free(struct posix_spawn_file_actions *fa, size_t len) 2341 { 2342 2343 for (size_t i = 0; i < len; i++) { 2344 struct posix_spawn_file_actions_entry *fae = &fa->fae[i]; 2345 if (fae->fae_action != FAE_OPEN) 2346 continue; 2347 kmem_strfree(fae->fae_path); 2348 } 2349 if (fa->len > 0) 2350 kmem_free(fa->fae, sizeof(*fa->fae) * fa->len); 2351 kmem_free(fa, sizeof(*fa)); 2352 } 2353 2354 static int 2355 posix_spawn_fa_alloc(struct posix_spawn_file_actions **fap, 2356 const struct posix_spawn_file_actions *ufa, rlim_t lim) 2357 { 2358 struct posix_spawn_file_actions *fa; 2359 struct posix_spawn_file_actions_entry *fae; 2360 char *pbuf = NULL; 2361 int error; 2362 size_t i = 0; 2363 2364 fa = kmem_alloc(sizeof(*fa), KM_SLEEP); 2365 error = copyin(ufa, fa, sizeof(*fa)); 2366 if (error || fa->len == 0) { 2367 kmem_free(fa, sizeof(*fa)); 2368 return error; /* 0 if not an error, and len == 0 */ 2369 } 2370 2371 if (fa->len > lim) { 2372 kmem_free(fa, sizeof(*fa)); 2373 return EINVAL; 2374 } 2375 2376 fa->size = fa->len; 2377 size_t fal = fa->len * sizeof(*fae); 2378 fae = fa->fae; 2379 fa->fae = kmem_alloc(fal, KM_SLEEP); 2380 error = copyin(fae, fa->fae, fal); 2381 if (error) 2382 goto out; 2383 2384 pbuf = PNBUF_GET(); 2385 for (; i < fa->len; i++) { 2386 fae = &fa->fae[i]; 2387 if (fae->fae_action != FAE_OPEN) 2388 continue; 2389 error = copyinstr(fae->fae_path, pbuf, MAXPATHLEN, &fal); 2390 if (error) 2391 goto out; 2392 fae->fae_path = kmem_alloc(fal, KM_SLEEP); 2393 memcpy(fae->fae_path, pbuf, fal); 2394 } 2395 PNBUF_PUT(pbuf); 2396 2397 *fap = fa; 2398 return 0; 2399 out: 2400 if (pbuf) 2401 PNBUF_PUT(pbuf); 2402 posix_spawn_fa_free(fa, i); 2403 return error; 2404 } 2405 2406 /* 2407 * N.B. increments nprocs upon success. Callers need to drop nprocs if 2408 * they fail for some other reason. 2409 */ 2410 int 2411 check_posix_spawn(struct lwp *l1) 2412 { 2413 int error, tnprocs, count; 2414 uid_t uid; 2415 struct proc *p1; 2416 2417 p1 = l1->l_proc; 2418 uid = kauth_cred_getuid(l1->l_cred); 2419 tnprocs = atomic_inc_uint_nv(&nprocs); 2420 2421 /* 2422 * Although process entries are dynamically created, we still keep 2423 * a global limit on the maximum number we will create. 2424 */ 2425 if (__predict_false(tnprocs >= maxproc)) 2426 error = -1; 2427 else 2428 error = kauth_authorize_process(l1->l_cred, 2429 KAUTH_PROCESS_FORK, p1, KAUTH_ARG(tnprocs), NULL, NULL); 2430 2431 if (error) { 2432 atomic_dec_uint(&nprocs); 2433 return EAGAIN; 2434 } 2435 2436 /* 2437 * Enforce limits. 2438 */ 2439 count = chgproccnt(uid, 1); 2440 if (kauth_authorize_process(l1->l_cred, KAUTH_PROCESS_RLIMIT, 2441 p1, KAUTH_ARG(KAUTH_REQ_PROCESS_RLIMIT_BYPASS), 2442 &p1->p_rlimit[RLIMIT_NPROC], KAUTH_ARG(RLIMIT_NPROC)) != 0 && 2443 __predict_false(count > p1->p_rlimit[RLIMIT_NPROC].rlim_cur)) { 2444 (void)chgproccnt(uid, -1); 2445 atomic_dec_uint(&nprocs); 2446 return EAGAIN; 2447 } 2448 2449 return 0; 2450 } 2451 2452 int 2453 do_posix_spawn(struct lwp *l1, pid_t *pid_res, bool *child_ok, const char *path, 2454 struct posix_spawn_file_actions *fa, 2455 struct posix_spawnattr *sa, 2456 char *const *argv, char *const *envp, 2457 execve_fetch_element_t fetch) 2458 { 2459 2460 struct proc *p1, *p2; 2461 struct lwp *l2; 2462 int error; 2463 struct spawn_exec_data *spawn_data; 2464 vaddr_t uaddr; 2465 pid_t pid; 2466 bool have_exec_lock = false; 2467 2468 p1 = l1->l_proc; 2469 2470 /* Allocate and init spawn_data */ 2471 spawn_data = kmem_zalloc(sizeof(*spawn_data), KM_SLEEP); 2472 spawn_data->sed_refcnt = 1; /* only parent so far */ 2473 cv_init(&spawn_data->sed_cv_child_ready, "pspawn"); 2474 mutex_init(&spawn_data->sed_mtx_child, MUTEX_DEFAULT, IPL_NONE); 2475 mutex_enter(&spawn_data->sed_mtx_child); 2476 2477 /* 2478 * Do the first part of the exec now, collect state 2479 * in spawn_data. 2480 */ 2481 error = execve_loadvm(l1, true, path, -1, argv, 2482 envp, fetch, &spawn_data->sed_exec); 2483 if (error == EJUSTRETURN) 2484 error = 0; 2485 else if (error) 2486 goto error_exit; 2487 2488 have_exec_lock = true; 2489 2490 /* 2491 * Allocate virtual address space for the U-area now, while it 2492 * is still easy to abort the fork operation if we're out of 2493 * kernel virtual address space. 2494 */ 2495 uaddr = uvm_uarea_alloc(); 2496 if (__predict_false(uaddr == 0)) { 2497 error = ENOMEM; 2498 goto error_exit; 2499 } 2500 2501 /* 2502 * Allocate new proc. Borrow proc0 vmspace for it, we will 2503 * replace it with its own before returning to userland 2504 * in the child. 2505 */ 2506 p2 = proc_alloc(); 2507 if (p2 == NULL) { 2508 /* We were unable to allocate a process ID. */ 2509 error = EAGAIN; 2510 goto error_exit; 2511 } 2512 2513 /* 2514 * This is a point of no return, we will have to go through 2515 * the child proc to properly clean it up past this point. 2516 */ 2517 pid = p2->p_pid; 2518 2519 /* 2520 * Make a proc table entry for the new process. 2521 * Start by zeroing the section of proc that is zero-initialized, 2522 * then copy the section that is copied directly from the parent. 2523 */ 2524 memset(&p2->p_startzero, 0, 2525 (unsigned) ((char *)&p2->p_endzero - (char *)&p2->p_startzero)); 2526 memcpy(&p2->p_startcopy, &p1->p_startcopy, 2527 (unsigned) ((char *)&p2->p_endcopy - (char *)&p2->p_startcopy)); 2528 p2->p_vmspace = proc0.p_vmspace; 2529 2530 TAILQ_INIT(&p2->p_sigpend.sp_info); 2531 2532 LIST_INIT(&p2->p_lwps); 2533 LIST_INIT(&p2->p_sigwaiters); 2534 2535 /* 2536 * Duplicate sub-structures as needed. 2537 * Increase reference counts on shared objects. 2538 * Inherit flags we want to keep. The flags related to SIGCHLD 2539 * handling are important in order to keep a consistent behaviour 2540 * for the child after the fork. If we are a 32-bit process, the 2541 * child will be too. 2542 */ 2543 p2->p_flag = 2544 p1->p_flag & (PK_SUGID | PK_NOCLDWAIT | PK_CLDSIGIGN | PK_32); 2545 p2->p_emul = p1->p_emul; 2546 p2->p_execsw = p1->p_execsw; 2547 2548 mutex_init(&p2->p_stmutex, MUTEX_DEFAULT, IPL_HIGH); 2549 mutex_init(&p2->p_auxlock, MUTEX_DEFAULT, IPL_NONE); 2550 rw_init(&p2->p_reflock); 2551 cv_init(&p2->p_waitcv, "wait"); 2552 cv_init(&p2->p_lwpcv, "lwpwait"); 2553 2554 p2->p_lock = mutex_obj_alloc(MUTEX_DEFAULT, IPL_NONE); 2555 2556 kauth_proc_fork(p1, p2); 2557 2558 p2->p_raslist = NULL; 2559 p2->p_fd = fd_copy(); 2560 2561 /* XXX racy */ 2562 p2->p_mqueue_cnt = p1->p_mqueue_cnt; 2563 2564 p2->p_cwdi = cwdinit(); 2565 2566 /* 2567 * Note: p_limit (rlimit stuff) is copy-on-write, so normally 2568 * we just need increase pl_refcnt. 2569 */ 2570 if (!p1->p_limit->pl_writeable) { 2571 lim_addref(p1->p_limit); 2572 p2->p_limit = p1->p_limit; 2573 } else { 2574 p2->p_limit = lim_copy(p1->p_limit); 2575 } 2576 2577 p2->p_lflag = 0; 2578 l1->l_vforkwaiting = false; 2579 p2->p_sflag = 0; 2580 p2->p_slflag = 0; 2581 p2->p_pptr = p1; 2582 p2->p_ppid = p1->p_pid; 2583 LIST_INIT(&p2->p_children); 2584 2585 p2->p_aio = NULL; 2586 2587 #ifdef KTRACE 2588 /* 2589 * Copy traceflag and tracefile if enabled. 2590 * If not inherited, these were zeroed above. 2591 */ 2592 if (p1->p_traceflag & KTRFAC_INHERIT) { 2593 mutex_enter(&ktrace_lock); 2594 p2->p_traceflag = p1->p_traceflag; 2595 if ((p2->p_tracep = p1->p_tracep) != NULL) 2596 ktradref(p2); 2597 mutex_exit(&ktrace_lock); 2598 } 2599 #endif 2600 2601 /* 2602 * Create signal actions for the child process. 2603 */ 2604 p2->p_sigacts = sigactsinit(p1, 0); 2605 mutex_enter(p1->p_lock); 2606 p2->p_sflag |= 2607 (p1->p_sflag & (PS_STOPFORK | PS_STOPEXEC | PS_NOCLDSTOP)); 2608 sched_proc_fork(p1, p2); 2609 mutex_exit(p1->p_lock); 2610 2611 p2->p_stflag = p1->p_stflag; 2612 2613 /* 2614 * p_stats. 2615 * Copy parts of p_stats, and zero out the rest. 2616 */ 2617 p2->p_stats = pstatscopy(p1->p_stats); 2618 2619 /* copy over machdep flags to the new proc */ 2620 cpu_proc_fork(p1, p2); 2621 2622 /* 2623 * Prepare remaining parts of spawn data 2624 */ 2625 spawn_data->sed_actions = fa; 2626 spawn_data->sed_attrs = sa; 2627 2628 spawn_data->sed_parent = p1; 2629 2630 /* create LWP */ 2631 lwp_create(l1, p2, uaddr, 0, NULL, 0, spawn_return, spawn_data, 2632 &l2, l1->l_class, &l1->l_sigmask, &l1->l_sigstk); 2633 l2->l_ctxlink = NULL; /* reset ucontext link */ 2634 2635 /* 2636 * Copy the credential so other references don't see our changes. 2637 * Test to see if this is necessary first, since in the common case 2638 * we won't need a private reference. 2639 */ 2640 if (kauth_cred_geteuid(l2->l_cred) != kauth_cred_getsvuid(l2->l_cred) || 2641 kauth_cred_getegid(l2->l_cred) != kauth_cred_getsvgid(l2->l_cred)) { 2642 l2->l_cred = kauth_cred_copy(l2->l_cred); 2643 kauth_cred_setsvuid(l2->l_cred, kauth_cred_geteuid(l2->l_cred)); 2644 kauth_cred_setsvgid(l2->l_cred, kauth_cred_getegid(l2->l_cred)); 2645 } 2646 2647 /* Update the master credentials. */ 2648 if (l2->l_cred != p2->p_cred) { 2649 kauth_cred_t ocred; 2650 2651 kauth_cred_hold(l2->l_cred); 2652 mutex_enter(p2->p_lock); 2653 ocred = p2->p_cred; 2654 p2->p_cred = l2->l_cred; 2655 mutex_exit(p2->p_lock); 2656 kauth_cred_free(ocred); 2657 } 2658 2659 *child_ok = true; 2660 spawn_data->sed_refcnt = 2; /* child gets it as well */ 2661 #if 0 2662 l2->l_nopreempt = 1; /* start it non-preemptable */ 2663 #endif 2664 2665 /* 2666 * It's now safe for the scheduler and other processes to see the 2667 * child process. 2668 */ 2669 mutex_enter(&proc_lock); 2670 2671 if (p1->p_session->s_ttyvp != NULL && p1->p_lflag & PL_CONTROLT) 2672 p2->p_lflag |= PL_CONTROLT; 2673 2674 LIST_INSERT_HEAD(&p1->p_children, p2, p_sibling); 2675 p2->p_exitsig = SIGCHLD; /* signal for parent on exit */ 2676 2677 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) == 2678 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2679 proc_changeparent(p2, p1->p_pptr); 2680 SET(p2->p_slflag, PSL_TRACEDCHILD); 2681 } 2682 2683 p2->p_oppid = p1->p_pid; /* Remember the original parent id. */ 2684 2685 LIST_INSERT_AFTER(p1, p2, p_pglist); 2686 LIST_INSERT_HEAD(&allproc, p2, p_list); 2687 2688 p2->p_trace_enabled = trace_is_enabled(p2); 2689 #ifdef __HAVE_SYSCALL_INTERN 2690 (*p2->p_emul->e_syscall_intern)(p2); 2691 #endif 2692 2693 /* 2694 * Make child runnable, set start time, and add to run queue except 2695 * if the parent requested the child to start in SSTOP state. 2696 */ 2697 mutex_enter(p2->p_lock); 2698 2699 getmicrotime(&p2->p_stats->p_start); 2700 2701 lwp_lock(l2); 2702 KASSERT(p2->p_nrlwps == 1); 2703 KASSERT(l2->l_stat == LSIDL); 2704 p2->p_nrlwps = 1; 2705 p2->p_stat = SACTIVE; 2706 setrunnable(l2); 2707 /* LWP now unlocked */ 2708 2709 mutex_exit(p2->p_lock); 2710 mutex_exit(&proc_lock); 2711 2712 cv_wait(&spawn_data->sed_cv_child_ready, &spawn_data->sed_mtx_child); 2713 error = spawn_data->sed_error; 2714 mutex_exit(&spawn_data->sed_mtx_child); 2715 spawn_exec_data_release(spawn_data); 2716 2717 rw_exit(&p1->p_reflock); 2718 rw_exit(&exec_lock); 2719 have_exec_lock = false; 2720 2721 *pid_res = pid; 2722 2723 if (error) 2724 return error; 2725 2726 if (p1->p_slflag & PSL_TRACED) { 2727 /* Paranoid check */ 2728 mutex_enter(&proc_lock); 2729 if ((p1->p_slflag & (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) != 2730 (PSL_TRACEPOSIX_SPAWN|PSL_TRACED)) { 2731 mutex_exit(&proc_lock); 2732 return 0; 2733 } 2734 2735 mutex_enter(p1->p_lock); 2736 eventswitch(TRAP_CHLD, PTRACE_POSIX_SPAWN, pid); 2737 } 2738 return 0; 2739 2740 error_exit: 2741 if (have_exec_lock) { 2742 execve_free_data(&spawn_data->sed_exec); 2743 rw_exit(&p1->p_reflock); 2744 rw_exit(&exec_lock); 2745 } 2746 mutex_exit(&spawn_data->sed_mtx_child); 2747 spawn_exec_data_release(spawn_data); 2748 2749 return error; 2750 } 2751 2752 int 2753 sys_posix_spawn(struct lwp *l1, const struct sys_posix_spawn_args *uap, 2754 register_t *retval) 2755 { 2756 /* { 2757 syscallarg(pid_t *) pid; 2758 syscallarg(const char *) path; 2759 syscallarg(const struct posix_spawn_file_actions *) file_actions; 2760 syscallarg(const struct posix_spawnattr *) attrp; 2761 syscallarg(char *const *) argv; 2762 syscallarg(char *const *) envp; 2763 } */ 2764 2765 int error; 2766 struct posix_spawn_file_actions *fa = NULL; 2767 struct posix_spawnattr *sa = NULL; 2768 pid_t pid; 2769 bool child_ok = false; 2770 rlim_t max_fileactions; 2771 proc_t *p = l1->l_proc; 2772 2773 /* check_posix_spawn() increments nprocs for us. */ 2774 error = check_posix_spawn(l1); 2775 if (error) { 2776 *retval = error; 2777 return 0; 2778 } 2779 2780 /* copy in file_actions struct */ 2781 if (SCARG(uap, file_actions) != NULL) { 2782 max_fileactions = 2 * uimin(p->p_rlimit[RLIMIT_NOFILE].rlim_cur, 2783 maxfiles); 2784 error = posix_spawn_fa_alloc(&fa, SCARG(uap, file_actions), 2785 max_fileactions); 2786 if (error) 2787 goto error_exit; 2788 } 2789 2790 /* copyin posix_spawnattr struct */ 2791 if (SCARG(uap, attrp) != NULL) { 2792 sa = kmem_alloc(sizeof(*sa), KM_SLEEP); 2793 error = copyin(SCARG(uap, attrp), sa, sizeof(*sa)); 2794 if (error) 2795 goto error_exit; 2796 } 2797 2798 /* 2799 * Do the spawn 2800 */ 2801 error = do_posix_spawn(l1, &pid, &child_ok, SCARG(uap, path), fa, sa, 2802 SCARG(uap, argv), SCARG(uap, envp), execve_fetch_element); 2803 if (error) 2804 goto error_exit; 2805 2806 if (error == 0 && SCARG(uap, pid) != NULL) 2807 error = copyout(&pid, SCARG(uap, pid), sizeof(pid)); 2808 2809 *retval = error; 2810 return 0; 2811 2812 error_exit: 2813 if (!child_ok) { 2814 (void)chgproccnt(kauth_cred_getuid(l1->l_cred), -1); 2815 atomic_dec_uint(&nprocs); 2816 2817 if (sa) 2818 kmem_free(sa, sizeof(*sa)); 2819 if (fa) 2820 posix_spawn_fa_free(fa, fa->len); 2821 } 2822 2823 *retval = error; 2824 return 0; 2825 } 2826 2827 void 2828 exec_free_emul_arg(struct exec_package *epp) 2829 { 2830 if (epp->ep_emul_arg_free != NULL) { 2831 KASSERT(epp->ep_emul_arg != NULL); 2832 (*epp->ep_emul_arg_free)(epp->ep_emul_arg); 2833 epp->ep_emul_arg_free = NULL; 2834 epp->ep_emul_arg = NULL; 2835 } else { 2836 KASSERT(epp->ep_emul_arg == NULL); 2837 } 2838 } 2839 2840 #ifdef DEBUG_EXEC 2841 static void 2842 dump_vmcmds(const struct exec_package * const epp, size_t x, int error) 2843 { 2844 struct exec_vmcmd *vp = &epp->ep_vmcmds.evs_cmds[0]; 2845 size_t j; 2846 2847 if (error == 0) 2848 DPRINTF(("vmcmds %u\n", epp->ep_vmcmds.evs_used)); 2849 else 2850 DPRINTF(("vmcmds %zu/%u, error %d\n", x, 2851 epp->ep_vmcmds.evs_used, error)); 2852 2853 for (j = 0; j < epp->ep_vmcmds.evs_used; j++) { 2854 DPRINTF(("vmcmd[%zu] = vmcmd_map_%s %#" 2855 PRIxVADDR"/%#"PRIxVSIZE" fd@%#" 2856 PRIxVSIZE" prot=0%o flags=%d\n", j, 2857 vp[j].ev_proc == vmcmd_map_pagedvn ? 2858 "pagedvn" : 2859 vp[j].ev_proc == vmcmd_map_readvn ? 2860 "readvn" : 2861 vp[j].ev_proc == vmcmd_map_zero ? 2862 "zero" : "*unknown*", 2863 vp[j].ev_addr, vp[j].ev_len, 2864 vp[j].ev_offset, vp[j].ev_prot, 2865 vp[j].ev_flags)); 2866 if (error != 0 && j == x) 2867 DPRINTF((" ^--- failed\n")); 2868 } 2869 } 2870 #endif 2871