1 /* $NetBSD: kern_exec.c,v 1.191 2004/10/01 16:30:52 yamt Exp $ */ 2 3 /*- 4 * Copyright (C) 1993, 1994, 1996 Christopher G. Demetriou 5 * Copyright (C) 1992 Wolfgang Solfrank. 6 * Copyright (C) 1992 TooLs GmbH. 7 * All rights reserved. 8 * 9 * Redistribution and use in source and binary forms, with or without 10 * modification, are permitted provided that the following conditions 11 * are met: 12 * 1. Redistributions of source code must retain the above copyright 13 * notice, this list of conditions and the following disclaimer. 14 * 2. Redistributions in binary form must reproduce the above copyright 15 * notice, this list of conditions and the following disclaimer in the 16 * documentation and/or other materials provided with the distribution. 17 * 3. All advertising materials mentioning features or use of this software 18 * must display the following acknowledgement: 19 * This product includes software developed by TooLs GmbH. 20 * 4. The name of TooLs GmbH may not be used to endorse or promote products 21 * derived from this software without specific prior written permission. 22 * 23 * THIS SOFTWARE IS PROVIDED BY TOOLS GMBH ``AS IS'' AND ANY EXPRESS OR 24 * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES 25 * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. 26 * IN NO EVENT SHALL TOOLS GMBH BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, 27 * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, 28 * PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; 29 * OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, 30 * WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR 31 * OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF 32 * ADVISED OF THE POSSIBILITY OF SUCH DAMAGE. 33 */ 34 35 #include <sys/cdefs.h> 36 __KERNEL_RCSID(0, "$NetBSD: kern_exec.c,v 1.191 2004/10/01 16:30:52 yamt Exp $"); 37 38 #include "opt_ktrace.h" 39 #include "opt_syscall_debug.h" 40 #include "opt_compat_netbsd.h" 41 42 #include <sys/param.h> 43 #include <sys/systm.h> 44 #include <sys/filedesc.h> 45 #include <sys/kernel.h> 46 #include <sys/proc.h> 47 #include <sys/mount.h> 48 #include <sys/malloc.h> 49 #include <sys/namei.h> 50 #include <sys/vnode.h> 51 #include <sys/file.h> 52 #include <sys/acct.h> 53 #include <sys/exec.h> 54 #include <sys/ktrace.h> 55 #include <sys/resourcevar.h> 56 #include <sys/wait.h> 57 #include <sys/mman.h> 58 #include <sys/ras.h> 59 #include <sys/signalvar.h> 60 #include <sys/stat.h> 61 #include <sys/syscall.h> 62 63 #include <sys/sa.h> 64 #include <sys/savar.h> 65 #include <sys/syscallargs.h> 66 67 #include <uvm/uvm_extern.h> 68 69 #include <machine/cpu.h> 70 #include <machine/reg.h> 71 72 static int exec_sigcode_map(struct proc *, const struct emul *); 73 74 #ifdef DEBUG_EXEC 75 #define DPRINTF(a) uprintf a 76 #else 77 #define DPRINTF(a) 78 #endif /* DEBUG_EXEC */ 79 80 MALLOC_DEFINE(M_EXEC, "exec", "argument lists & other mem used by exec"); 81 82 /* 83 * Exec function switch: 84 * 85 * Note that each makecmds function is responsible for loading the 86 * exec package with the necessary functions for any exec-type-specific 87 * handling. 88 * 89 * Functions for specific exec types should be defined in their own 90 * header file. 91 */ 92 extern const struct execsw execsw_builtin[]; 93 extern int nexecs_builtin; 94 static const struct execsw **execsw = NULL; 95 static int nexecs; 96 97 u_int exec_maxhdrsz; /* must not be static - netbsd32 needs it */ 98 99 #ifdef LKM 100 /* list of supported emulations */ 101 static 102 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head); 103 struct emul_entry { 104 LIST_ENTRY(emul_entry) el_list; 105 const struct emul *el_emul; 106 int ro_entry; 107 }; 108 109 /* list of dynamically loaded execsw entries */ 110 static 111 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head); 112 struct exec_entry { 113 LIST_ENTRY(exec_entry) ex_list; 114 const struct execsw *es; 115 }; 116 117 /* structure used for building execw[] */ 118 struct execsw_entry { 119 struct execsw_entry *next; 120 const struct execsw *es; 121 }; 122 #endif /* LKM */ 123 124 #ifdef SYSCALL_DEBUG 125 extern const char * const syscallnames[]; 126 #endif 127 #ifdef __HAVE_SYSCALL_INTERN 128 void syscall_intern(struct proc *); 129 #else 130 void syscall(void); 131 #endif 132 133 #ifdef COMPAT_16 134 extern char sigcode[], esigcode[]; 135 struct uvm_object *emul_netbsd_object; 136 #endif 137 138 /* NetBSD emul struct */ 139 const struct emul emul_netbsd = { 140 "netbsd", 141 NULL, /* emulation path */ 142 #ifndef __HAVE_MINIMAL_EMUL 143 EMUL_HAS_SYS___syscall, 144 NULL, 145 SYS_syscall, 146 SYS_NSYSENT, 147 #endif 148 sysent, 149 #ifdef SYSCALL_DEBUG 150 syscallnames, 151 #else 152 NULL, 153 #endif 154 sendsig, 155 trapsignal, 156 NULL, 157 #ifdef COMPAT_16 158 sigcode, 159 esigcode, 160 &emul_netbsd_object, 161 #else 162 NULL, 163 NULL, 164 NULL, 165 #endif 166 setregs, 167 NULL, 168 NULL, 169 NULL, 170 NULL, 171 NULL, 172 #ifdef __HAVE_SYSCALL_INTERN 173 syscall_intern, 174 #else 175 syscall, 176 #endif 177 NULL, 178 NULL, 179 }; 180 181 #ifdef LKM 182 /* 183 * Exec lock. Used to control access to execsw[] structures. 184 * This must not be static so that netbsd32 can access it, too. 185 */ 186 struct lock exec_lock; 187 188 static void link_es(struct execsw_entry **, const struct execsw *); 189 #endif /* LKM */ 190 191 /* 192 * check exec: 193 * given an "executable" described in the exec package's namei info, 194 * see what we can do with it. 195 * 196 * ON ENTRY: 197 * exec package with appropriate namei info 198 * proc pointer of exec'ing proc 199 * iff verified exec enabled then flag indicating a direct exec or 200 * an indirect exec (i.e. for a shell script interpreter) 201 * NO SELF-LOCKED VNODES 202 * 203 * ON EXIT: 204 * error: nothing held, etc. exec header still allocated. 205 * ok: filled exec package, executable's vnode (unlocked). 206 * 207 * EXEC SWITCH ENTRY: 208 * Locked vnode to check, exec package, proc. 209 * 210 * EXEC SWITCH EXIT: 211 * ok: return 0, filled exec package, executable's vnode (unlocked). 212 * error: destructive: 213 * everything deallocated execept exec header. 214 * non-destructive: 215 * error code, executable's vnode (unlocked), 216 * exec header unmodified. 217 */ 218 int 219 #ifdef VERIFIED_EXEC 220 check_exec(struct proc *p, struct exec_package *epp, int direct_exec) 221 #else 222 check_exec(struct proc *p, struct exec_package *epp) 223 #endif 224 { 225 int error, i; 226 struct vnode *vp; 227 struct nameidata *ndp; 228 size_t resid; 229 230 ndp = epp->ep_ndp; 231 ndp->ni_cnd.cn_nameiop = LOOKUP; 232 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 233 /* first get the vnode */ 234 if ((error = namei(ndp)) != 0) 235 return error; 236 epp->ep_vp = vp = ndp->ni_vp; 237 238 /* check access and type */ 239 if (vp->v_type != VREG) { 240 error = EACCES; 241 goto bad1; 242 } 243 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 244 goto bad1; 245 246 /* get attributes */ 247 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 248 goto bad1; 249 250 /* Check mount point */ 251 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 252 error = EACCES; 253 goto bad1; 254 } 255 if (vp->v_mount->mnt_flag & MNT_NOSUID) 256 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 257 258 /* try to open it */ 259 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 260 goto bad1; 261 262 /* unlock vp, since we need it unlocked from here on out. */ 263 VOP_UNLOCK(vp, 0); 264 265 266 #ifdef VERIFIED_EXEC 267 /* Evaluate signature for file... */ 268 if ((error = check_veriexec(p, vp, epp, direct_exec)) != 0) 269 goto bad2; 270 #endif 271 272 /* now we have the file, get the exec header */ 273 uvn_attach(vp, VM_PROT_READ); 274 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 275 UIO_SYSSPACE, 0, p->p_ucred, &resid, NULL); 276 if (error) 277 goto bad2; 278 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 279 280 /* 281 * Set up default address space limits. Can be overridden 282 * by individual exec packages. 283 * 284 * XXX probably should be all done in the exec pakages. 285 */ 286 epp->ep_vm_minaddr = VM_MIN_ADDRESS; 287 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 288 /* 289 * set up the vmcmds for creation of the process 290 * address space 291 */ 292 error = ENOEXEC; 293 for (i = 0; i < nexecs && error != 0; i++) { 294 int newerror; 295 296 epp->ep_esch = execsw[i]; 297 newerror = (*execsw[i]->es_makecmds)(p, epp); 298 /* make sure the first "interesting" error code is saved. */ 299 if (!newerror || error == ENOEXEC) 300 error = newerror; 301 302 /* if es_makecmds call was successful, update epp->ep_es */ 303 if (!newerror && (epp->ep_flags & EXEC_HASES) == 0) 304 epp->ep_es = execsw[i]; 305 306 if (epp->ep_flags & EXEC_DESTR && error != 0) 307 return error; 308 } 309 if (!error) { 310 /* check that entry point is sane */ 311 if (epp->ep_entry > VM_MAXUSER_ADDRESS) 312 error = ENOEXEC; 313 314 /* check limits */ 315 if ((epp->ep_tsize > MAXTSIZ) || 316 (epp->ep_dsize > 317 (u_quad_t)p->p_rlimit[RLIMIT_DATA].rlim_cur)) 318 error = ENOMEM; 319 320 if (!error) 321 return (0); 322 } 323 324 /* 325 * free any vmspace-creation commands, 326 * and release their references 327 */ 328 kill_vmcmds(&epp->ep_vmcmds); 329 330 bad2: 331 /* 332 * close and release the vnode, restore the old one, free the 333 * pathname buf, and punt. 334 */ 335 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 336 VOP_CLOSE(vp, FREAD, p->p_ucred, p); 337 vput(vp); 338 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 339 return error; 340 341 bad1: 342 /* 343 * free the namei pathname buffer, and put the vnode 344 * (which we don't yet have open). 345 */ 346 vput(vp); /* was still locked */ 347 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 348 return error; 349 } 350 351 #ifdef __MACHINE_STACK_GROWS_UP 352 #define STACK_PTHREADSPACE NBPG 353 #else 354 #define STACK_PTHREADSPACE 0 355 #endif 356 357 /* 358 * exec system call 359 */ 360 /* ARGSUSED */ 361 int 362 sys_execve(struct lwp *l, void *v, register_t *retval) 363 { 364 struct sys_execve_args /* { 365 syscallarg(const char *) path; 366 syscallarg(char * const *) argp; 367 syscallarg(char * const *) envp; 368 } */ *uap = v; 369 int error; 370 u_int i; 371 struct exec_package pack; 372 struct nameidata nid; 373 struct vattr attr; 374 struct proc *p; 375 struct ucred *cred; 376 char *argp; 377 char * const *cpp; 378 char *dp, *sp; 379 long argc, envc; 380 size_t len; 381 char *stack; 382 struct ps_strings arginfo; 383 struct vmspace *vm; 384 char **tmpfap; 385 int szsigcode; 386 struct exec_vmcmd *base_vcp; 387 int oldlwpflags; 388 389 /* Disable scheduler activation upcalls. */ 390 oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL); 391 if (l->l_flag & L_SA) 392 l->l_flag &= ~(L_SA | L_SA_UPCALL); 393 394 p = l->l_proc; 395 /* 396 * Lock the process and set the P_INEXEC flag to indicate that 397 * it should be left alone until we're done here. This is 398 * necessary to avoid race conditions - e.g. in ptrace() - 399 * that might allow a local user to illicitly obtain elevated 400 * privileges. 401 */ 402 p->p_flag |= P_INEXEC; 403 404 cred = p->p_ucred; 405 base_vcp = NULL; 406 /* 407 * Init the namei data to point the file user's program name. 408 * This is done here rather than in check_exec(), so that it's 409 * possible to override this settings if any of makecmd/probe 410 * functions call check_exec() recursively - for example, 411 * see exec_script_makecmds(). 412 */ 413 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 414 415 /* 416 * initialize the fields of the exec package. 417 */ 418 pack.ep_name = SCARG(uap, path); 419 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 420 pack.ep_hdrlen = exec_maxhdrsz; 421 pack.ep_hdrvalid = 0; 422 pack.ep_ndp = &nid; 423 pack.ep_emul_arg = NULL; 424 pack.ep_vmcmds.evs_cnt = 0; 425 pack.ep_vmcmds.evs_used = 0; 426 pack.ep_vap = &attr; 427 pack.ep_flags = 0; 428 429 #ifdef LKM 430 lockmgr(&exec_lock, LK_SHARED, NULL); 431 #endif 432 433 /* see if we can run it. */ 434 #ifdef VERIFIED_EXEC 435 if ((error = check_exec(p, &pack, 1)) != 0) 436 /* if ((error = check_exec(p, &pack, 0)) != 0) */ 437 #else 438 if ((error = check_exec(p, &pack)) != 0) 439 #endif 440 goto freehdr; 441 442 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 443 444 /* allocate an argument buffer */ 445 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS); 446 #ifdef DIAGNOSTIC 447 if (argp == (vaddr_t) 0) 448 panic("execve: argp == NULL"); 449 #endif 450 dp = argp; 451 argc = 0; 452 453 /* copy the fake args list, if there's one, freeing it as we go */ 454 if (pack.ep_flags & EXEC_HASARGL) { 455 tmpfap = pack.ep_fa; 456 while (*tmpfap != NULL) { 457 char *cp; 458 459 cp = *tmpfap; 460 while (*cp) 461 *dp++ = *cp++; 462 dp++; 463 464 FREE(*tmpfap, M_EXEC); 465 tmpfap++; argc++; 466 } 467 FREE(pack.ep_fa, M_EXEC); 468 pack.ep_flags &= ~EXEC_HASARGL; 469 } 470 471 /* Now get argv & environment */ 472 if (!(cpp = SCARG(uap, argp))) { 473 error = EINVAL; 474 goto bad; 475 } 476 477 if (pack.ep_flags & EXEC_SKIPARG) 478 cpp++; 479 480 while (1) { 481 len = argp + ARG_MAX - dp; 482 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 483 goto bad; 484 if (!sp) 485 break; 486 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 487 if (error == ENAMETOOLONG) 488 error = E2BIG; 489 goto bad; 490 } 491 #ifdef KTRACE 492 if (KTRPOINT(p, KTR_EXEC_ARG)) 493 ktrkmem(p, KTR_EXEC_ARG, dp, len - 1); 494 #endif 495 dp += len; 496 cpp++; 497 argc++; 498 } 499 500 envc = 0; 501 /* environment need not be there */ 502 if ((cpp = SCARG(uap, envp)) != NULL ) { 503 while (1) { 504 len = argp + ARG_MAX - dp; 505 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 506 goto bad; 507 if (!sp) 508 break; 509 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 510 if (error == ENAMETOOLONG) 511 error = E2BIG; 512 goto bad; 513 } 514 #ifdef KTRACE 515 if (KTRPOINT(p, KTR_EXEC_ENV)) 516 ktrkmem(p, KTR_EXEC_ENV, dp, len - 1); 517 #endif 518 dp += len; 519 cpp++; 520 envc++; 521 } 522 } 523 524 dp = (char *) ALIGN(dp); 525 526 szsigcode = pack.ep_es->es_emul->e_esigcode - 527 pack.ep_es->es_emul->e_sigcode; 528 529 /* Now check if args & environ fit into new stack */ 530 if (pack.ep_flags & EXEC_32) 531 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 532 sizeof(int) + sizeof(int) + dp + STACKGAPLEN + 533 szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE) 534 - argp; 535 else 536 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 537 sizeof(char *) + sizeof(int) + dp + STACKGAPLEN + 538 szsigcode + sizeof(struct ps_strings) + STACK_PTHREADSPACE) 539 - argp; 540 541 len = ALIGN(len); /* make the stack "safely" aligned */ 542 543 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 544 error = ENOMEM; 545 goto bad; 546 } 547 548 /* Get rid of other LWPs/ */ 549 p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */ 550 exit_lwps(l); 551 p->p_flag &= ~P_WEXIT; 552 KDASSERT(p->p_nlwps == 1); 553 554 /* This is now LWP 1 */ 555 l->l_lid = 1; 556 p->p_nlwpid = 1; 557 558 /* Release any SA state. */ 559 if (p->p_sa) 560 sa_release(p); 561 562 /* Remove POSIX timers */ 563 timers_free(p, TIMERS_POSIX); 564 565 /* adjust "active stack depth" for process VSZ */ 566 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 567 568 /* 569 * Do whatever is necessary to prepare the address space 570 * for remapping. Note that this might replace the current 571 * vmspace with another! 572 */ 573 uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 574 575 /* record proc's vnode, for use by procfs and others */ 576 if (p->p_textvp) 577 vrele(p->p_textvp); 578 VREF(pack.ep_vp); 579 p->p_textvp = pack.ep_vp; 580 581 /* Now map address space */ 582 vm = p->p_vmspace; 583 vm->vm_taddr = (caddr_t) pack.ep_taddr; 584 vm->vm_tsize = btoc(pack.ep_tsize); 585 vm->vm_daddr = (caddr_t) pack.ep_daddr; 586 vm->vm_dsize = btoc(pack.ep_dsize); 587 vm->vm_ssize = btoc(pack.ep_ssize); 588 vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr; 589 vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr; 590 591 /* create the new process's VM space by running the vmcmds */ 592 #ifdef DIAGNOSTIC 593 if (pack.ep_vmcmds.evs_used == 0) 594 panic("execve: no vmcmds"); 595 #endif 596 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 597 struct exec_vmcmd *vcp; 598 599 vcp = &pack.ep_vmcmds.evs_cmds[i]; 600 if (vcp->ev_flags & VMCMD_RELATIVE) { 601 #ifdef DIAGNOSTIC 602 if (base_vcp == NULL) 603 panic("execve: relative vmcmd with no base"); 604 if (vcp->ev_flags & VMCMD_BASE) 605 panic("execve: illegal base & relative vmcmd"); 606 #endif 607 vcp->ev_addr += base_vcp->ev_addr; 608 } 609 error = (*vcp->ev_proc)(p, vcp); 610 #ifdef DEBUG_EXEC 611 if (error) { 612 int j; 613 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 614 for (j = 0; j <= i; j++) 615 uprintf( 616 "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n", 617 j, vp[j].ev_addr, vp[j].ev_len, 618 vp[j].ev_offset, vp[j].ev_prot, 619 vp[j].ev_flags); 620 } 621 #endif /* DEBUG_EXEC */ 622 if (vcp->ev_flags & VMCMD_BASE) 623 base_vcp = vcp; 624 } 625 626 /* free the vmspace-creation commands, and release their references */ 627 kill_vmcmds(&pack.ep_vmcmds); 628 629 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 630 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 631 vput(pack.ep_vp); 632 633 /* if an error happened, deallocate and punt */ 634 if (error) { 635 DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error)); 636 goto exec_abort; 637 } 638 639 /* remember information about the process */ 640 arginfo.ps_nargvstr = argc; 641 arginfo.ps_nenvstr = envc; 642 643 stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 644 STACK_PTHREADSPACE + sizeof(struct ps_strings) + szsigcode), 645 len - (sizeof(struct ps_strings) + szsigcode)); 646 #ifdef __MACHINE_STACK_GROWS_UP 647 /* 648 * The copyargs call always copies into lower addresses 649 * first, moving towards higher addresses, starting with 650 * the stack pointer that we give. When the stack grows 651 * down, this puts argc/argv/envp very shallow on the 652 * stack, right at the first user stack pointer, and puts 653 * STACKGAPLEN very deep in the stack. When the stack 654 * grows up, the situation is reversed. 655 * 656 * Normally, this is no big deal. But the ld_elf.so _rtld() 657 * function expects to be called with a single pointer to 658 * a region that has a few words it can stash values into, 659 * followed by argc/argv/envp. When the stack grows down, 660 * it's easy to decrement the stack pointer a little bit to 661 * allocate the space for these few words and pass the new 662 * stack pointer to _rtld. When the stack grows up, however, 663 * a few words before argc is part of the signal trampoline, XXX 664 * so we have a problem. 665 * 666 * Instead of changing how _rtld works, we take the easy way 667 * out and steal 32 bytes before we call copyargs. This 668 * space is effectively stolen from STACKGAPLEN. 669 */ 670 stack += 32; 671 #endif /* __MACHINE_STACK_GROWS_UP */ 672 673 /* Now copy argc, args & environ to new stack */ 674 error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp); 675 if (error) { 676 DPRINTF(("execve: copyargs failed %d\n", error)); 677 goto exec_abort; 678 } 679 /* Move the stack back to original point */ 680 stack = (char *)STACK_GROW(vm->vm_minsaddr, len); 681 682 /* fill process ps_strings info */ 683 p->p_psstr = (struct ps_strings *) 684 STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, STACK_PTHREADSPACE), 685 sizeof(struct ps_strings)); 686 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 687 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 688 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 689 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 690 691 /* copy out the process's ps_strings structure */ 692 if ((error = copyout(&arginfo, (char *)p->p_psstr, 693 sizeof(arginfo))) != 0) { 694 DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n", 695 &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo))); 696 goto exec_abort; 697 } 698 699 stopprofclock(p); /* stop profiling */ 700 fdcloseexec(p); /* handle close on exec */ 701 execsigs(p); /* reset catched signals */ 702 703 l->l_ctxlink = NULL; /* reset ucontext link */ 704 705 /* set command name & other accounting info */ 706 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 707 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 708 p->p_comm[len] = 0; 709 p->p_acflag &= ~AFORK; 710 711 p->p_flag |= P_EXEC; 712 if (p->p_flag & P_PPWAIT) { 713 p->p_flag &= ~P_PPWAIT; 714 wakeup((caddr_t) p->p_pptr); 715 } 716 717 /* 718 * deal with set[ug]id. 719 * MNT_NOSUID has already been used to disable s[ug]id. 720 */ 721 if ((p->p_flag & P_TRACED) == 0 && 722 723 (((attr.va_mode & S_ISUID) != 0 && 724 p->p_ucred->cr_uid != attr.va_uid) || 725 726 ((attr.va_mode & S_ISGID) != 0 && 727 p->p_ucred->cr_gid != attr.va_gid))) { 728 /* 729 * Mark the process as SUGID before we do 730 * anything that might block. 731 */ 732 p_sugid(p); 733 734 /* Make sure file descriptors 0..2 are in use. */ 735 if ((error = fdcheckstd(p)) != 0) 736 goto exec_abort; 737 738 p->p_ucred = crcopy(cred); 739 #ifdef KTRACE 740 /* 741 * If process is being ktraced, turn off - unless 742 * root set it. 743 */ 744 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) 745 ktrderef(p); 746 #endif 747 if (attr.va_mode & S_ISUID) 748 p->p_ucred->cr_uid = attr.va_uid; 749 if (attr.va_mode & S_ISGID) 750 p->p_ucred->cr_gid = attr.va_gid; 751 } else 752 p->p_flag &= ~P_SUGID; 753 p->p_cred->p_svuid = p->p_ucred->cr_uid; 754 p->p_cred->p_svgid = p->p_ucred->cr_gid; 755 756 #if defined(__HAVE_RAS) 757 /* 758 * Remove all RASs from the address space. 759 */ 760 ras_purgeall(p); 761 #endif 762 763 doexechooks(p); 764 765 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 766 767 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 768 769 /* notify others that we exec'd */ 770 KNOTE(&p->p_klist, NOTE_EXEC); 771 772 /* setup new registers and do misc. setup. */ 773 (*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack); 774 if (pack.ep_es->es_setregs) 775 (*pack.ep_es->es_setregs)(l, &pack, (u_long) stack); 776 777 /* map the process's signal trampoline code */ 778 if (exec_sigcode_map(p, pack.ep_es->es_emul)) 779 goto exec_abort; 780 781 if (p->p_flag & P_TRACED) 782 psignal(p, SIGTRAP); 783 784 free(pack.ep_hdr, M_EXEC); 785 786 /* 787 * Call emulation specific exec hook. This can setup setup per-process 788 * p->p_emuldata or do any other per-process stuff an emulation needs. 789 * 790 * If we are executing process of different emulation than the 791 * original forked process, call e_proc_exit() of the old emulation 792 * first, then e_proc_exec() of new emulation. If the emulation is 793 * same, the exec hook code should deallocate any old emulation 794 * resources held previously by this process. 795 */ 796 if (p->p_emul && p->p_emul->e_proc_exit 797 && p->p_emul != pack.ep_es->es_emul) 798 (*p->p_emul->e_proc_exit)(p); 799 800 /* 801 * Call exec hook. Emulation code may NOT store reference to anything 802 * from &pack. 803 */ 804 if (pack.ep_es->es_emul->e_proc_exec) 805 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack); 806 807 /* update p_emul, the old value is no longer needed */ 808 p->p_emul = pack.ep_es->es_emul; 809 810 /* ...and the same for p_execsw */ 811 p->p_execsw = pack.ep_es; 812 813 #ifdef __HAVE_SYSCALL_INTERN 814 (*p->p_emul->e_syscall_intern)(p); 815 #endif 816 #ifdef KTRACE 817 if (KTRPOINT(p, KTR_EMUL)) 818 ktremul(p); 819 #endif 820 821 #ifdef LKM 822 lockmgr(&exec_lock, LK_RELEASE, NULL); 823 #endif 824 p->p_flag &= ~P_INEXEC; 825 826 if (p->p_flag & P_STOPEXEC) { 827 int s; 828 829 sigminusset(&contsigmask, &p->p_sigctx.ps_siglist); 830 SCHED_LOCK(s); 831 p->p_pptr->p_nstopchild++; 832 p->p_stat = SSTOP; 833 l->l_stat = LSSTOP; 834 p->p_nrlwps--; 835 mi_switch(l, NULL); 836 SCHED_ASSERT_UNLOCKED(); 837 splx(s); 838 } 839 840 return (EJUSTRETURN); 841 842 bad: 843 p->p_flag &= ~P_INEXEC; 844 /* free the vmspace-creation commands, and release their references */ 845 kill_vmcmds(&pack.ep_vmcmds); 846 /* kill any opened file descriptor, if necessary */ 847 if (pack.ep_flags & EXEC_HASFD) { 848 pack.ep_flags &= ~EXEC_HASFD; 849 (void) fdrelease(p, pack.ep_fd); 850 } 851 /* close and put the exec'd file */ 852 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 853 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 854 vput(pack.ep_vp); 855 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 856 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 857 858 freehdr: 859 l->l_flag |= oldlwpflags; 860 p->p_flag &= ~P_INEXEC; 861 #ifdef LKM 862 lockmgr(&exec_lock, LK_RELEASE, NULL); 863 #endif 864 865 free(pack.ep_hdr, M_EXEC); 866 return error; 867 868 exec_abort: 869 p->p_flag &= ~P_INEXEC; 870 #ifdef LKM 871 lockmgr(&exec_lock, LK_RELEASE, NULL); 872 #endif 873 874 /* 875 * the old process doesn't exist anymore. exit gracefully. 876 * get rid of the (new) address space we have created, if any, get rid 877 * of our namei data and vnode, and exit noting failure 878 */ 879 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 880 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 881 if (pack.ep_emul_arg) 882 FREE(pack.ep_emul_arg, M_TEMP); 883 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 884 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 885 free(pack.ep_hdr, M_EXEC); 886 exit1(l, W_EXITCODE(error, SIGABRT)); 887 888 /* NOTREACHED */ 889 return 0; 890 } 891 892 893 int 894 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo, 895 char **stackp, void *argp) 896 { 897 char **cpp, *dp, *sp; 898 size_t len; 899 void *nullp; 900 long argc, envc; 901 int error; 902 903 cpp = (char **)*stackp; 904 nullp = NULL; 905 argc = arginfo->ps_nargvstr; 906 envc = arginfo->ps_nenvstr; 907 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) 908 return error; 909 910 dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen); 911 sp = argp; 912 913 /* XXX don't copy them out, remap them! */ 914 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 915 916 for (; --argc >= 0; sp += len, dp += len) 917 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 918 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 919 return error; 920 921 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 922 return error; 923 924 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 925 926 for (; --envc >= 0; sp += len, dp += len) 927 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 928 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 929 return error; 930 931 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 932 return error; 933 934 *stackp = (char *)cpp; 935 return 0; 936 } 937 938 #ifdef LKM 939 /* 940 * Find an emulation of given name in list of emulations. 941 * Needs to be called with the exec_lock held. 942 */ 943 const struct emul * 944 emul_search(const char *name) 945 { 946 struct emul_entry *it; 947 948 LIST_FOREACH(it, &el_head, el_list) { 949 if (strcmp(name, it->el_emul->e_name) == 0) 950 return it->el_emul; 951 } 952 953 return NULL; 954 } 955 956 /* 957 * Add an emulation to list, if it's not there already. 958 */ 959 int 960 emul_register(const struct emul *emul, int ro_entry) 961 { 962 struct emul_entry *ee; 963 int error; 964 965 error = 0; 966 lockmgr(&exec_lock, LK_SHARED, NULL); 967 968 if (emul_search(emul->e_name)) { 969 error = EEXIST; 970 goto out; 971 } 972 973 MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry), 974 M_EXEC, M_WAITOK); 975 ee->el_emul = emul; 976 ee->ro_entry = ro_entry; 977 LIST_INSERT_HEAD(&el_head, ee, el_list); 978 979 out: 980 lockmgr(&exec_lock, LK_RELEASE, NULL); 981 return error; 982 } 983 984 /* 985 * Remove emulation with name 'name' from list of supported emulations. 986 */ 987 int 988 emul_unregister(const char *name) 989 { 990 const struct proclist_desc *pd; 991 struct emul_entry *it; 992 int i, error; 993 struct proc *ptmp; 994 995 error = 0; 996 lockmgr(&exec_lock, LK_SHARED, NULL); 997 998 LIST_FOREACH(it, &el_head, el_list) { 999 if (strcmp(it->el_emul->e_name, name) == 0) 1000 break; 1001 } 1002 1003 if (!it) { 1004 error = ENOENT; 1005 goto out; 1006 } 1007 1008 if (it->ro_entry) { 1009 error = EBUSY; 1010 goto out; 1011 } 1012 1013 /* test if any execw[] entry is still using this */ 1014 for(i=0; i < nexecs; i++) { 1015 if (execsw[i]->es_emul == it->el_emul) { 1016 error = EBUSY; 1017 goto out; 1018 } 1019 } 1020 1021 /* 1022 * Test if any process is running under this emulation - since 1023 * emul_unregister() is running quite sendomly, it's better 1024 * to do expensive check here than to use any locking. 1025 */ 1026 proclist_lock_read(); 1027 for (pd = proclists; pd->pd_list != NULL && !error; pd++) { 1028 PROCLIST_FOREACH(ptmp, pd->pd_list) { 1029 if (ptmp->p_emul == it->el_emul) { 1030 error = EBUSY; 1031 break; 1032 } 1033 } 1034 } 1035 proclist_unlock_read(); 1036 1037 if (error) 1038 goto out; 1039 1040 1041 /* entry is not used, remove it */ 1042 LIST_REMOVE(it, el_list); 1043 FREE(it, M_EXEC); 1044 1045 out: 1046 lockmgr(&exec_lock, LK_RELEASE, NULL); 1047 return error; 1048 } 1049 1050 /* 1051 * Add execsw[] entry. 1052 */ 1053 int 1054 exec_add(struct execsw *esp, const char *e_name) 1055 { 1056 struct exec_entry *it; 1057 int error; 1058 1059 error = 0; 1060 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1061 1062 if (!esp->es_emul) { 1063 esp->es_emul = emul_search(e_name); 1064 if (!esp->es_emul) { 1065 error = ENOENT; 1066 goto out; 1067 } 1068 } 1069 1070 LIST_FOREACH(it, &ex_head, ex_list) { 1071 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1072 if (it->es->es_makecmds == esp->es_makecmds 1073 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1074 && it->es->es_emul == esp->es_emul) { 1075 error = EEXIST; 1076 goto out; 1077 } 1078 } 1079 1080 /* if we got here, the entry doesn't exist yet */ 1081 MALLOC(it, struct exec_entry *, sizeof(struct exec_entry), 1082 M_EXEC, M_WAITOK); 1083 it->es = esp; 1084 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1085 1086 /* update execsw[] */ 1087 exec_init(0); 1088 1089 out: 1090 lockmgr(&exec_lock, LK_RELEASE, NULL); 1091 return error; 1092 } 1093 1094 /* 1095 * Remove execsw[] entry. 1096 */ 1097 int 1098 exec_remove(const struct execsw *esp) 1099 { 1100 struct exec_entry *it; 1101 int error; 1102 1103 error = 0; 1104 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1105 1106 LIST_FOREACH(it, &ex_head, ex_list) { 1107 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1108 if (it->es->es_makecmds == esp->es_makecmds 1109 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1110 && it->es->es_emul == esp->es_emul) 1111 break; 1112 } 1113 if (!it) { 1114 error = ENOENT; 1115 goto out; 1116 } 1117 1118 /* remove item from list and free resources */ 1119 LIST_REMOVE(it, ex_list); 1120 FREE(it, M_EXEC); 1121 1122 /* update execsw[] */ 1123 exec_init(0); 1124 1125 out: 1126 lockmgr(&exec_lock, LK_RELEASE, NULL); 1127 return error; 1128 } 1129 1130 static void 1131 link_es(struct execsw_entry **listp, const struct execsw *esp) 1132 { 1133 struct execsw_entry *et, *e1; 1134 1135 MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry), 1136 M_TEMP, M_WAITOK); 1137 et->next = NULL; 1138 et->es = esp; 1139 if (*listp == NULL) { 1140 *listp = et; 1141 return; 1142 } 1143 1144 switch(et->es->es_prio) { 1145 case EXECSW_PRIO_FIRST: 1146 /* put new entry as the first */ 1147 et->next = *listp; 1148 *listp = et; 1149 break; 1150 case EXECSW_PRIO_ANY: 1151 /* put new entry after all *_FIRST and *_ANY entries */ 1152 for(e1 = *listp; e1->next 1153 && e1->next->es->es_prio != EXECSW_PRIO_LAST; 1154 e1 = e1->next); 1155 et->next = e1->next; 1156 e1->next = et; 1157 break; 1158 case EXECSW_PRIO_LAST: 1159 /* put new entry as the last one */ 1160 for(e1 = *listp; e1->next; e1 = e1->next); 1161 e1->next = et; 1162 break; 1163 default: 1164 #ifdef DIAGNOSTIC 1165 panic("execw[] entry with unknown priority %d found", 1166 et->es->es_prio); 1167 #endif 1168 break; 1169 } 1170 } 1171 1172 /* 1173 * Initialize exec structures. If init_boot is true, also does necessary 1174 * one-time initialization (it's called from main() that way). 1175 * Once system is multiuser, this should be called with exec_lock held, 1176 * i.e. via exec_{add|remove}(). 1177 */ 1178 int 1179 exec_init(int init_boot) 1180 { 1181 const struct execsw **new_es, * const *old_es; 1182 struct execsw_entry *list, *e1; 1183 struct exec_entry *e2; 1184 int i, es_sz; 1185 1186 if (init_boot) { 1187 /* do one-time initializations */ 1188 lockinit(&exec_lock, PWAIT, "execlck", 0, 0); 1189 1190 /* register compiled-in emulations */ 1191 for(i=0; i < nexecs_builtin; i++) { 1192 if (execsw_builtin[i].es_emul) 1193 emul_register(execsw_builtin[i].es_emul, 1); 1194 } 1195 #ifdef DIAGNOSTIC 1196 if (i == 0) 1197 panic("no emulations found in execsw_builtin[]"); 1198 #endif 1199 } 1200 1201 /* 1202 * Build execsw[] array from builtin entries and entries added 1203 * at runtime. 1204 */ 1205 list = NULL; 1206 for(i=0; i < nexecs_builtin; i++) 1207 link_es(&list, &execsw_builtin[i]); 1208 1209 /* Add dynamically loaded entries */ 1210 es_sz = nexecs_builtin; 1211 LIST_FOREACH(e2, &ex_head, ex_list) { 1212 link_es(&list, e2->es); 1213 es_sz++; 1214 } 1215 1216 /* 1217 * Now that we have sorted all execw entries, create new execsw[] 1218 * and free no longer needed memory in the process. 1219 */ 1220 new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK); 1221 for(i=0; list; i++) { 1222 new_es[i] = list->es; 1223 e1 = list->next; 1224 FREE(list, M_TEMP); 1225 list = e1; 1226 } 1227 1228 /* 1229 * New execsw[] array built, now replace old execsw[] and free 1230 * used memory. 1231 */ 1232 old_es = execsw; 1233 execsw = new_es; 1234 nexecs = es_sz; 1235 if (old_es) 1236 free((void *)old_es, M_EXEC); 1237 1238 /* 1239 * Figure out the maximum size of an exec header. 1240 */ 1241 exec_maxhdrsz = 0; 1242 for (i = 0; i < nexecs; i++) { 1243 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1244 exec_maxhdrsz = execsw[i]->es_hdrsz; 1245 } 1246 1247 return 0; 1248 } 1249 #endif 1250 1251 #ifndef LKM 1252 /* 1253 * Simplified exec_init() for kernels without LKMs. Only initialize 1254 * exec_maxhdrsz and execsw[]. 1255 */ 1256 int 1257 exec_init(int init_boot) 1258 { 1259 int i; 1260 1261 #ifdef DIAGNOSTIC 1262 if (!init_boot) 1263 panic("exec_init(): called with init_boot == 0"); 1264 #endif 1265 1266 /* do one-time initializations */ 1267 nexecs = nexecs_builtin; 1268 execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK); 1269 1270 /* 1271 * Fill in execsw[] and figure out the maximum size of an exec header. 1272 */ 1273 exec_maxhdrsz = 0; 1274 for(i=0; i < nexecs; i++) { 1275 execsw[i] = &execsw_builtin[i]; 1276 if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz) 1277 exec_maxhdrsz = execsw_builtin[i].es_hdrsz; 1278 } 1279 1280 return 0; 1281 1282 } 1283 #endif /* !LKM */ 1284 1285 static int 1286 exec_sigcode_map(struct proc *p, const struct emul *e) 1287 { 1288 vaddr_t va; 1289 vsize_t sz; 1290 int error; 1291 struct uvm_object *uobj; 1292 1293 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1294 1295 if (e->e_sigobject == NULL || sz == 0) { 1296 return 0; 1297 } 1298 1299 /* 1300 * If we don't have a sigobject for this emulation, create one. 1301 * 1302 * sigobject is an anonymous memory object (just like SYSV shared 1303 * memory) that we keep a permanent reference to and that we map 1304 * in all processes that need this sigcode. The creation is simple, 1305 * we create an object, add a permanent reference to it, map it in 1306 * kernel space, copy out the sigcode to it and unmap it. 1307 * We map it with PROT_READ|PROT_EXEC into the process just 1308 * the way sys_mmap() would map it. 1309 */ 1310 1311 uobj = *e->e_sigobject; 1312 if (uobj == NULL) { 1313 uobj = uao_create(sz, 0); 1314 (*uobj->pgops->pgo_reference)(uobj); 1315 va = vm_map_min(kernel_map); 1316 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1317 uobj, 0, 0, 1318 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1319 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 1320 printf("kernel mapping failed %d\n", error); 1321 (*uobj->pgops->pgo_detach)(uobj); 1322 return (error); 1323 } 1324 memcpy((void *)va, e->e_sigcode, sz); 1325 #ifdef PMAP_NEED_PROCWR 1326 pmap_procwr(&proc0, va, sz); 1327 #endif 1328 uvm_unmap(kernel_map, va, va + round_page(sz)); 1329 *e->e_sigobject = uobj; 1330 } 1331 1332 /* Just a hint to uvm_map where to put it. */ 1333 va = VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, round_page(sz)); 1334 1335 #ifdef __alpha__ 1336 /* 1337 * Tru64 puts /sbin/loader at the end of user virtual memory, 1338 * which causes the above calculation to put the sigcode at 1339 * an invalid address. Put it just below the text instead. 1340 */ 1341 if (va == (vaddr_t)p->p_vmspace->vm_map.max_offset) { 1342 va = (vaddr_t)p->p_vmspace->vm_taddr - round_page(sz); 1343 } 1344 #endif 1345 1346 (*uobj->pgops->pgo_reference)(uobj); 1347 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 1348 uobj, 0, 0, 1349 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 1350 UVM_ADV_RANDOM, 0)); 1351 if (error) { 1352 (*uobj->pgops->pgo_detach)(uobj); 1353 return (error); 1354 } 1355 p->p_sigctx.ps_sigcode = (void *)va; 1356 return (0); 1357 } 1358