1 /* $NetBSD: kern_exec.c,v 1.185 2004/03/26 17:13:37 drochner 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 1.185 2004/03/26 17:13:37 drochner 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, p); 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 /* 352 * exec system call 353 */ 354 /* ARGSUSED */ 355 int 356 sys_execve(struct lwp *l, void *v, register_t *retval) 357 { 358 struct sys_execve_args /* { 359 syscallarg(const char *) path; 360 syscallarg(char * const *) argp; 361 syscallarg(char * const *) envp; 362 } */ *uap = v; 363 int error; 364 u_int i; 365 struct exec_package pack; 366 struct nameidata nid; 367 struct vattr attr; 368 struct proc *p; 369 struct ucred *cred; 370 char *argp; 371 char * const *cpp; 372 char *dp, *sp; 373 long argc, envc; 374 size_t len; 375 char *stack; 376 struct ps_strings arginfo; 377 struct vmspace *vm; 378 char **tmpfap; 379 int szsigcode; 380 struct exec_vmcmd *base_vcp; 381 int oldlwpflags; 382 383 /* Disable scheduler activation upcalls. */ 384 oldlwpflags = l->l_flag & (L_SA | L_SA_UPCALL); 385 if (l->l_flag & L_SA) 386 l->l_flag &= ~(L_SA | L_SA_UPCALL); 387 388 p = l->l_proc; 389 /* 390 * Lock the process and set the P_INEXEC flag to indicate that 391 * it should be left alone until we're done here. This is 392 * necessary to avoid race conditions - e.g. in ptrace() - 393 * that might allow a local user to illicitly obtain elevated 394 * privileges. 395 */ 396 p->p_flag |= P_INEXEC; 397 398 cred = p->p_ucred; 399 base_vcp = NULL; 400 /* 401 * Init the namei data to point the file user's program name. 402 * This is done here rather than in check_exec(), so that it's 403 * possible to override this settings if any of makecmd/probe 404 * functions call check_exec() recursively - for example, 405 * see exec_script_makecmds(). 406 */ 407 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 408 409 /* 410 * initialize the fields of the exec package. 411 */ 412 pack.ep_name = SCARG(uap, path); 413 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 414 pack.ep_hdrlen = exec_maxhdrsz; 415 pack.ep_hdrvalid = 0; 416 pack.ep_ndp = &nid; 417 pack.ep_emul_arg = NULL; 418 pack.ep_vmcmds.evs_cnt = 0; 419 pack.ep_vmcmds.evs_used = 0; 420 pack.ep_vap = &attr; 421 pack.ep_flags = 0; 422 423 #ifdef LKM 424 lockmgr(&exec_lock, LK_SHARED, NULL); 425 #endif 426 427 /* see if we can run it. */ 428 #ifdef VERIFIED_EXEC 429 if ((error = check_exec(p, &pack, 1)) != 0) 430 /* if ((error = check_exec(p, &pack, 0)) != 0) */ 431 #else 432 if ((error = check_exec(p, &pack)) != 0) 433 #endif 434 goto freehdr; 435 436 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 437 438 /* allocate an argument buffer */ 439 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS); 440 #ifdef DIAGNOSTIC 441 if (argp == (vaddr_t) 0) 442 panic("execve: argp == NULL"); 443 #endif 444 dp = argp; 445 argc = 0; 446 447 /* copy the fake args list, if there's one, freeing it as we go */ 448 if (pack.ep_flags & EXEC_HASARGL) { 449 tmpfap = pack.ep_fa; 450 while (*tmpfap != NULL) { 451 char *cp; 452 453 cp = *tmpfap; 454 while (*cp) 455 *dp++ = *cp++; 456 dp++; 457 458 FREE(*tmpfap, M_EXEC); 459 tmpfap++; argc++; 460 } 461 FREE(pack.ep_fa, M_EXEC); 462 pack.ep_flags &= ~EXEC_HASARGL; 463 } 464 465 /* Now get argv & environment */ 466 if (!(cpp = SCARG(uap, argp))) { 467 error = EINVAL; 468 goto bad; 469 } 470 471 if (pack.ep_flags & EXEC_SKIPARG) 472 cpp++; 473 474 while (1) { 475 len = argp + ARG_MAX - dp; 476 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 477 goto bad; 478 if (!sp) 479 break; 480 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 481 if (error == ENAMETOOLONG) 482 error = E2BIG; 483 goto bad; 484 } 485 #ifdef KTRACE 486 if (KTRPOINT(p, KTR_EXEC_ARG)) 487 ktrkmem(p, KTR_EXEC_ARG, dp, len - 1); 488 #endif 489 dp += len; 490 cpp++; 491 argc++; 492 } 493 494 envc = 0; 495 /* environment need not be there */ 496 if ((cpp = SCARG(uap, envp)) != NULL ) { 497 while (1) { 498 len = argp + ARG_MAX - dp; 499 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 500 goto bad; 501 if (!sp) 502 break; 503 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 504 if (error == ENAMETOOLONG) 505 error = E2BIG; 506 goto bad; 507 } 508 #ifdef KTRACE 509 if (KTRPOINT(p, KTR_EXEC_ENV)) 510 ktrkmem(p, KTR_EXEC_ENV, dp, len - 1); 511 #endif 512 dp += len; 513 cpp++; 514 envc++; 515 } 516 } 517 518 dp = (char *) ALIGN(dp); 519 520 szsigcode = pack.ep_es->es_emul->e_esigcode - 521 pack.ep_es->es_emul->e_sigcode; 522 523 /* Now check if args & environ fit into new stack */ 524 if (pack.ep_flags & EXEC_32) 525 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 526 sizeof(int) + sizeof(int) + dp + STACKGAPLEN + 527 szsigcode + sizeof(struct ps_strings)) - argp; 528 else 529 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 530 sizeof(char *) + sizeof(int) + dp + STACKGAPLEN + 531 szsigcode + sizeof(struct ps_strings)) - argp; 532 533 len = ALIGN(len); /* make the stack "safely" aligned */ 534 535 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 536 error = ENOMEM; 537 goto bad; 538 } 539 540 /* Get rid of other LWPs/ */ 541 p->p_flag |= P_WEXIT; /* XXX hack. lwp-exit stuff wants to see it. */ 542 exit_lwps(l); 543 p->p_flag &= ~P_WEXIT; 544 KDASSERT(p->p_nlwps == 1); 545 546 /* This is now LWP 1 */ 547 l->l_lid = 1; 548 p->p_nlwpid = 1; 549 550 /* Release any SA state. */ 551 if (p->p_sa) 552 sa_release(p); 553 554 /* Remove POSIX timers */ 555 timers_free(p, TIMERS_POSIX); 556 557 /* adjust "active stack depth" for process VSZ */ 558 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 559 560 /* 561 * Do whatever is necessary to prepare the address space 562 * for remapping. Note that this might replace the current 563 * vmspace with another! 564 */ 565 uvmspace_exec(l, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 566 567 /* Now map address space */ 568 vm = p->p_vmspace; 569 vm->vm_taddr = (caddr_t) pack.ep_taddr; 570 vm->vm_tsize = btoc(pack.ep_tsize); 571 vm->vm_daddr = (caddr_t) pack.ep_daddr; 572 vm->vm_dsize = btoc(pack.ep_dsize); 573 vm->vm_ssize = btoc(pack.ep_ssize); 574 vm->vm_maxsaddr = (caddr_t) pack.ep_maxsaddr; 575 vm->vm_minsaddr = (caddr_t) pack.ep_minsaddr; 576 577 /* create the new process's VM space by running the vmcmds */ 578 #ifdef DIAGNOSTIC 579 if (pack.ep_vmcmds.evs_used == 0) 580 panic("execve: no vmcmds"); 581 #endif 582 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 583 struct exec_vmcmd *vcp; 584 585 vcp = &pack.ep_vmcmds.evs_cmds[i]; 586 if (vcp->ev_flags & VMCMD_RELATIVE) { 587 #ifdef DIAGNOSTIC 588 if (base_vcp == NULL) 589 panic("execve: relative vmcmd with no base"); 590 if (vcp->ev_flags & VMCMD_BASE) 591 panic("execve: illegal base & relative vmcmd"); 592 #endif 593 vcp->ev_addr += base_vcp->ev_addr; 594 } 595 error = (*vcp->ev_proc)(p, vcp); 596 #ifdef DEBUG_EXEC 597 if (error) { 598 int j; 599 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 600 for (j = 0; j <= i; j++) 601 uprintf( 602 "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n", 603 j, vp[j].ev_addr, vp[j].ev_len, 604 vp[j].ev_offset, vp[j].ev_prot, 605 vp[j].ev_flags); 606 } 607 #endif /* DEBUG_EXEC */ 608 if (vcp->ev_flags & VMCMD_BASE) 609 base_vcp = vcp; 610 } 611 612 /* free the vmspace-creation commands, and release their references */ 613 kill_vmcmds(&pack.ep_vmcmds); 614 615 /* if an error happened, deallocate and punt */ 616 if (error) { 617 DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error)); 618 goto exec_abort; 619 } 620 621 /* remember information about the process */ 622 arginfo.ps_nargvstr = argc; 623 arginfo.ps_nenvstr = envc; 624 625 stack = (char *)STACK_ALLOC(STACK_GROW(vm->vm_minsaddr, 626 sizeof(struct ps_strings) + szsigcode), 627 len - (sizeof(struct ps_strings) + szsigcode)); 628 #ifdef __MACHINE_STACK_GROWS_UP 629 /* 630 * The copyargs call always copies into lower addresses 631 * first, moving towards higher addresses, starting with 632 * the stack pointer that we give. When the stack grows 633 * down, this puts argc/argv/envp very shallow on the 634 * stack, right at the first user stack pointer, and puts 635 * STACKGAPLEN very deep in the stack. When the stack 636 * grows up, the situation is reversed. 637 * 638 * Normally, this is no big deal. But the ld_elf.so _rtld() 639 * function expects to be called with a single pointer to 640 * a region that has a few words it can stash values into, 641 * followed by argc/argv/envp. When the stack grows down, 642 * it's easy to decrement the stack pointer a little bit to 643 * allocate the space for these few words and pass the new 644 * stack pointer to _rtld. When the stack grows up, however, 645 * a few words before argc is part of the signal trampoline, XXX 646 * so we have a problem. 647 * 648 * Instead of changing how _rtld works, we take the easy way 649 * out and steal 32 bytes before we call copyargs. This 650 * space is effectively stolen from STACKGAPLEN. 651 */ 652 stack += 32; 653 #endif /* __MACHINE_STACK_GROWS_UP */ 654 655 /* Now copy argc, args & environ to new stack */ 656 error = (*pack.ep_es->es_copyargs)(p, &pack, &arginfo, &stack, argp); 657 if (error) { 658 DPRINTF(("execve: copyargs failed %d\n", error)); 659 goto exec_abort; 660 } 661 /* Move the stack back to original point */ 662 stack = (char *)STACK_GROW(vm->vm_minsaddr, len); 663 664 /* fill process ps_strings info */ 665 p->p_psstr = (struct ps_strings *)STACK_ALLOC(vm->vm_minsaddr, 666 sizeof(struct ps_strings)); 667 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 668 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 669 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 670 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 671 672 /* copy out the process's ps_strings structure */ 673 if ((error = copyout(&arginfo, (char *)p->p_psstr, 674 sizeof(arginfo))) != 0) { 675 DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n", 676 &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo))); 677 goto exec_abort; 678 } 679 680 stopprofclock(p); /* stop profiling */ 681 fdcloseexec(p); /* handle close on exec */ 682 execsigs(p); /* reset catched signals */ 683 684 l->l_ctxlink = NULL; /* reset ucontext link */ 685 686 /* set command name & other accounting info */ 687 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 688 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 689 p->p_comm[len] = 0; 690 p->p_acflag &= ~AFORK; 691 692 /* record proc's vnode, for use by procfs and others */ 693 if (p->p_textvp) 694 vrele(p->p_textvp); 695 VREF(pack.ep_vp); 696 p->p_textvp = pack.ep_vp; 697 698 p->p_flag |= P_EXEC; 699 if (p->p_flag & P_PPWAIT) { 700 p->p_flag &= ~P_PPWAIT; 701 wakeup((caddr_t) p->p_pptr); 702 } 703 704 /* 705 * deal with set[ug]id. 706 * MNT_NOSUID has already been used to disable s[ug]id. 707 */ 708 if ((p->p_flag & P_TRACED) == 0 && 709 710 (((attr.va_mode & S_ISUID) != 0 && 711 p->p_ucred->cr_uid != attr.va_uid) || 712 713 ((attr.va_mode & S_ISGID) != 0 && 714 p->p_ucred->cr_gid != attr.va_gid))) { 715 /* 716 * Mark the process as SUGID before we do 717 * anything that might block. 718 */ 719 p_sugid(p); 720 721 /* Make sure file descriptors 0..2 are in use. */ 722 if ((error = fdcheckstd(p)) != 0) 723 goto exec_abort; 724 725 p->p_ucred = crcopy(cred); 726 #ifdef KTRACE 727 /* 728 * If process is being ktraced, turn off - unless 729 * root set it. 730 */ 731 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) 732 ktrderef(p); 733 #endif 734 if (attr.va_mode & S_ISUID) 735 p->p_ucred->cr_uid = attr.va_uid; 736 if (attr.va_mode & S_ISGID) 737 p->p_ucred->cr_gid = attr.va_gid; 738 } else 739 p->p_flag &= ~P_SUGID; 740 p->p_cred->p_svuid = p->p_ucred->cr_uid; 741 p->p_cred->p_svgid = p->p_ucred->cr_gid; 742 743 #if defined(__HAVE_RAS) 744 /* 745 * Remove all RASs from the address space. 746 */ 747 ras_purgeall(p); 748 #endif 749 750 doexechooks(p); 751 752 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 753 754 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 755 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 756 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 757 vput(pack.ep_vp); 758 759 /* notify others that we exec'd */ 760 KNOTE(&p->p_klist, NOTE_EXEC); 761 762 /* setup new registers and do misc. setup. */ 763 (*pack.ep_es->es_emul->e_setregs)(l, &pack, (u_long) stack); 764 if (pack.ep_es->es_setregs) 765 (*pack.ep_es->es_setregs)(l, &pack, (u_long) stack); 766 767 /* map the process's signal trampoline code */ 768 if (exec_sigcode_map(p, pack.ep_es->es_emul)) 769 goto exec_abort; 770 771 if (p->p_flag & P_TRACED) 772 psignal(p, SIGTRAP); 773 774 free(pack.ep_hdr, M_EXEC); 775 776 /* 777 * Call emulation specific exec hook. This can setup setup per-process 778 * p->p_emuldata or do any other per-process stuff an emulation needs. 779 * 780 * If we are executing process of different emulation than the 781 * original forked process, call e_proc_exit() of the old emulation 782 * first, then e_proc_exec() of new emulation. If the emulation is 783 * same, the exec hook code should deallocate any old emulation 784 * resources held previously by this process. 785 */ 786 if (p->p_emul && p->p_emul->e_proc_exit 787 && p->p_emul != pack.ep_es->es_emul) 788 (*p->p_emul->e_proc_exit)(p); 789 790 /* 791 * Call exec hook. Emulation code may NOT store reference to anything 792 * from &pack. 793 */ 794 if (pack.ep_es->es_emul->e_proc_exec) 795 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack); 796 797 /* update p_emul, the old value is no longer needed */ 798 p->p_emul = pack.ep_es->es_emul; 799 800 /* ...and the same for p_execsw */ 801 p->p_execsw = pack.ep_es; 802 803 #ifdef __HAVE_SYSCALL_INTERN 804 (*p->p_emul->e_syscall_intern)(p); 805 #endif 806 #ifdef KTRACE 807 if (KTRPOINT(p, KTR_EMUL)) 808 ktremul(p); 809 #endif 810 811 #ifdef LKM 812 lockmgr(&exec_lock, LK_RELEASE, NULL); 813 #endif 814 p->p_flag &= ~P_INEXEC; 815 816 if (p->p_flag & P_STOPEXEC) { 817 int s; 818 819 sigminusset(&contsigmask, &p->p_sigctx.ps_siglist); 820 SCHED_LOCK(s); 821 p->p_pptr->p_nstopchild++; 822 p->p_stat = SSTOP; 823 l->l_stat = LSSTOP; 824 p->p_nrlwps--; 825 mi_switch(l, NULL); 826 SCHED_ASSERT_UNLOCKED(); 827 splx(s); 828 } 829 830 return (EJUSTRETURN); 831 832 bad: 833 p->p_flag &= ~P_INEXEC; 834 /* free the vmspace-creation commands, and release their references */ 835 kill_vmcmds(&pack.ep_vmcmds); 836 /* kill any opened file descriptor, if necessary */ 837 if (pack.ep_flags & EXEC_HASFD) { 838 pack.ep_flags &= ~EXEC_HASFD; 839 (void) fdrelease(p, pack.ep_fd); 840 } 841 /* close and put the exec'd file */ 842 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 843 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 844 vput(pack.ep_vp); 845 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 846 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 847 848 freehdr: 849 l->l_flag |= oldlwpflags; 850 p->p_flag &= ~P_INEXEC; 851 #ifdef LKM 852 lockmgr(&exec_lock, LK_RELEASE, NULL); 853 #endif 854 855 free(pack.ep_hdr, M_EXEC); 856 return error; 857 858 exec_abort: 859 p->p_flag &= ~P_INEXEC; 860 #ifdef LKM 861 lockmgr(&exec_lock, LK_RELEASE, NULL); 862 #endif 863 864 /* 865 * the old process doesn't exist anymore. exit gracefully. 866 * get rid of the (new) address space we have created, if any, get rid 867 * of our namei data and vnode, and exit noting failure 868 */ 869 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 870 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 871 if (pack.ep_emul_arg) 872 FREE(pack.ep_emul_arg, M_TEMP); 873 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 874 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 875 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 876 vput(pack.ep_vp); 877 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 878 free(pack.ep_hdr, M_EXEC); 879 exit1(l, W_EXITCODE(error, SIGABRT)); 880 881 /* NOTREACHED */ 882 return 0; 883 } 884 885 886 int 887 copyargs(struct proc *p, struct exec_package *pack, struct ps_strings *arginfo, 888 char **stackp, void *argp) 889 { 890 char **cpp, *dp, *sp; 891 size_t len; 892 void *nullp; 893 long argc, envc; 894 int error; 895 896 cpp = (char **)*stackp; 897 nullp = NULL; 898 argc = arginfo->ps_nargvstr; 899 envc = arginfo->ps_nenvstr; 900 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) 901 return error; 902 903 dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen); 904 sp = argp; 905 906 /* XXX don't copy them out, remap them! */ 907 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 908 909 for (; --argc >= 0; sp += len, dp += len) 910 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 911 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 912 return error; 913 914 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 915 return error; 916 917 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 918 919 for (; --envc >= 0; sp += len, dp += len) 920 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 921 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 922 return error; 923 924 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 925 return error; 926 927 *stackp = (char *)cpp; 928 return 0; 929 } 930 931 #ifdef LKM 932 /* 933 * Find an emulation of given name in list of emulations. 934 * Needs to be called with the exec_lock held. 935 */ 936 const struct emul * 937 emul_search(const char *name) 938 { 939 struct emul_entry *it; 940 941 LIST_FOREACH(it, &el_head, el_list) { 942 if (strcmp(name, it->el_emul->e_name) == 0) 943 return it->el_emul; 944 } 945 946 return NULL; 947 } 948 949 /* 950 * Add an emulation to list, if it's not there already. 951 */ 952 int 953 emul_register(const struct emul *emul, int ro_entry) 954 { 955 struct emul_entry *ee; 956 int error; 957 958 error = 0; 959 lockmgr(&exec_lock, LK_SHARED, NULL); 960 961 if (emul_search(emul->e_name)) { 962 error = EEXIST; 963 goto out; 964 } 965 966 MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry), 967 M_EXEC, M_WAITOK); 968 ee->el_emul = emul; 969 ee->ro_entry = ro_entry; 970 LIST_INSERT_HEAD(&el_head, ee, el_list); 971 972 out: 973 lockmgr(&exec_lock, LK_RELEASE, NULL); 974 return error; 975 } 976 977 /* 978 * Remove emulation with name 'name' from list of supported emulations. 979 */ 980 int 981 emul_unregister(const char *name) 982 { 983 const struct proclist_desc *pd; 984 struct emul_entry *it; 985 int i, error; 986 struct proc *ptmp; 987 988 error = 0; 989 lockmgr(&exec_lock, LK_SHARED, NULL); 990 991 LIST_FOREACH(it, &el_head, el_list) { 992 if (strcmp(it->el_emul->e_name, name) == 0) 993 break; 994 } 995 996 if (!it) { 997 error = ENOENT; 998 goto out; 999 } 1000 1001 if (it->ro_entry) { 1002 error = EBUSY; 1003 goto out; 1004 } 1005 1006 /* test if any execw[] entry is still using this */ 1007 for(i=0; i < nexecs; i++) { 1008 if (execsw[i]->es_emul == it->el_emul) { 1009 error = EBUSY; 1010 goto out; 1011 } 1012 } 1013 1014 /* 1015 * Test if any process is running under this emulation - since 1016 * emul_unregister() is running quite sendomly, it's better 1017 * to do expensive check here than to use any locking. 1018 */ 1019 proclist_lock_read(); 1020 for (pd = proclists; pd->pd_list != NULL && !error; pd++) { 1021 LIST_FOREACH(ptmp, pd->pd_list, p_list) { 1022 if (ptmp->p_emul == it->el_emul) { 1023 error = EBUSY; 1024 break; 1025 } 1026 } 1027 } 1028 proclist_unlock_read(); 1029 1030 if (error) 1031 goto out; 1032 1033 1034 /* entry is not used, remove it */ 1035 LIST_REMOVE(it, el_list); 1036 FREE(it, M_EXEC); 1037 1038 out: 1039 lockmgr(&exec_lock, LK_RELEASE, NULL); 1040 return error; 1041 } 1042 1043 /* 1044 * Add execsw[] entry. 1045 */ 1046 int 1047 exec_add(struct execsw *esp, const char *e_name) 1048 { 1049 struct exec_entry *it; 1050 int error; 1051 1052 error = 0; 1053 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1054 1055 if (!esp->es_emul) { 1056 esp->es_emul = emul_search(e_name); 1057 if (!esp->es_emul) { 1058 error = ENOENT; 1059 goto out; 1060 } 1061 } 1062 1063 LIST_FOREACH(it, &ex_head, ex_list) { 1064 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1065 if (it->es->es_makecmds == esp->es_makecmds 1066 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1067 && it->es->es_emul == esp->es_emul) { 1068 error = EEXIST; 1069 goto out; 1070 } 1071 } 1072 1073 /* if we got here, the entry doesn't exist yet */ 1074 MALLOC(it, struct exec_entry *, sizeof(struct exec_entry), 1075 M_EXEC, M_WAITOK); 1076 it->es = esp; 1077 LIST_INSERT_HEAD(&ex_head, it, ex_list); 1078 1079 /* update execsw[] */ 1080 exec_init(0); 1081 1082 out: 1083 lockmgr(&exec_lock, LK_RELEASE, NULL); 1084 return error; 1085 } 1086 1087 /* 1088 * Remove execsw[] entry. 1089 */ 1090 int 1091 exec_remove(const struct execsw *esp) 1092 { 1093 struct exec_entry *it; 1094 int error; 1095 1096 error = 0; 1097 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 1098 1099 LIST_FOREACH(it, &ex_head, ex_list) { 1100 /* assume tuple (makecmds, probe_func, emulation) is unique */ 1101 if (it->es->es_makecmds == esp->es_makecmds 1102 && it->es->u.elf_probe_func == esp->u.elf_probe_func 1103 && it->es->es_emul == esp->es_emul) 1104 break; 1105 } 1106 if (!it) { 1107 error = ENOENT; 1108 goto out; 1109 } 1110 1111 /* remove item from list and free resources */ 1112 LIST_REMOVE(it, ex_list); 1113 FREE(it, M_EXEC); 1114 1115 /* update execsw[] */ 1116 exec_init(0); 1117 1118 out: 1119 lockmgr(&exec_lock, LK_RELEASE, NULL); 1120 return error; 1121 } 1122 1123 static void 1124 link_es(struct execsw_entry **listp, const struct execsw *esp) 1125 { 1126 struct execsw_entry *et, *e1; 1127 1128 MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry), 1129 M_TEMP, M_WAITOK); 1130 et->next = NULL; 1131 et->es = esp; 1132 if (*listp == NULL) { 1133 *listp = et; 1134 return; 1135 } 1136 1137 switch(et->es->es_prio) { 1138 case EXECSW_PRIO_FIRST: 1139 /* put new entry as the first */ 1140 et->next = *listp; 1141 *listp = et; 1142 break; 1143 case EXECSW_PRIO_ANY: 1144 /* put new entry after all *_FIRST and *_ANY entries */ 1145 for(e1 = *listp; e1->next 1146 && e1->next->es->es_prio != EXECSW_PRIO_LAST; 1147 e1 = e1->next); 1148 et->next = e1->next; 1149 e1->next = et; 1150 break; 1151 case EXECSW_PRIO_LAST: 1152 /* put new entry as the last one */ 1153 for(e1 = *listp; e1->next; e1 = e1->next); 1154 e1->next = et; 1155 break; 1156 default: 1157 #ifdef DIAGNOSTIC 1158 panic("execw[] entry with unknown priority %d found", 1159 et->es->es_prio); 1160 #endif 1161 break; 1162 } 1163 } 1164 1165 /* 1166 * Initialize exec structures. If init_boot is true, also does necessary 1167 * one-time initialization (it's called from main() that way). 1168 * Once system is multiuser, this should be called with exec_lock held, 1169 * i.e. via exec_{add|remove}(). 1170 */ 1171 int 1172 exec_init(int init_boot) 1173 { 1174 const struct execsw **new_es, * const *old_es; 1175 struct execsw_entry *list, *e1; 1176 struct exec_entry *e2; 1177 int i, es_sz; 1178 1179 if (init_boot) { 1180 /* do one-time initializations */ 1181 lockinit(&exec_lock, PWAIT, "execlck", 0, 0); 1182 1183 /* register compiled-in emulations */ 1184 for(i=0; i < nexecs_builtin; i++) { 1185 if (execsw_builtin[i].es_emul) 1186 emul_register(execsw_builtin[i].es_emul, 1); 1187 } 1188 #ifdef DIAGNOSTIC 1189 if (i == 0) 1190 panic("no emulations found in execsw_builtin[]"); 1191 #endif 1192 } 1193 1194 /* 1195 * Build execsw[] array from builtin entries and entries added 1196 * at runtime. 1197 */ 1198 list = NULL; 1199 for(i=0; i < nexecs_builtin; i++) 1200 link_es(&list, &execsw_builtin[i]); 1201 1202 /* Add dynamically loaded entries */ 1203 es_sz = nexecs_builtin; 1204 LIST_FOREACH(e2, &ex_head, ex_list) { 1205 link_es(&list, e2->es); 1206 es_sz++; 1207 } 1208 1209 /* 1210 * Now that we have sorted all execw entries, create new execsw[] 1211 * and free no longer needed memory in the process. 1212 */ 1213 new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK); 1214 for(i=0; list; i++) { 1215 new_es[i] = list->es; 1216 e1 = list->next; 1217 FREE(list, M_TEMP); 1218 list = e1; 1219 } 1220 1221 /* 1222 * New execsw[] array built, now replace old execsw[] and free 1223 * used memory. 1224 */ 1225 old_es = execsw; 1226 execsw = new_es; 1227 nexecs = es_sz; 1228 if (old_es) 1229 free((void *)old_es, M_EXEC); 1230 1231 /* 1232 * Figure out the maximum size of an exec header. 1233 */ 1234 exec_maxhdrsz = 0; 1235 for (i = 0; i < nexecs; i++) { 1236 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1237 exec_maxhdrsz = execsw[i]->es_hdrsz; 1238 } 1239 1240 return 0; 1241 } 1242 #endif 1243 1244 #ifndef LKM 1245 /* 1246 * Simplified exec_init() for kernels without LKMs. Only initialize 1247 * exec_maxhdrsz and execsw[]. 1248 */ 1249 int 1250 exec_init(int init_boot) 1251 { 1252 int i; 1253 1254 #ifdef DIAGNOSTIC 1255 if (!init_boot) 1256 panic("exec_init(): called with init_boot == 0"); 1257 #endif 1258 1259 /* do one-time initializations */ 1260 nexecs = nexecs_builtin; 1261 execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK); 1262 1263 /* 1264 * Fill in execsw[] and figure out the maximum size of an exec header. 1265 */ 1266 exec_maxhdrsz = 0; 1267 for(i=0; i < nexecs; i++) { 1268 execsw[i] = &execsw_builtin[i]; 1269 if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz) 1270 exec_maxhdrsz = execsw_builtin[i].es_hdrsz; 1271 } 1272 1273 return 0; 1274 1275 } 1276 #endif /* !LKM */ 1277 1278 static int 1279 exec_sigcode_map(struct proc *p, const struct emul *e) 1280 { 1281 vaddr_t va; 1282 vsize_t sz; 1283 int error; 1284 struct uvm_object *uobj; 1285 1286 sz = (vaddr_t)e->e_esigcode - (vaddr_t)e->e_sigcode; 1287 1288 if (e->e_sigobject == NULL || sz == 0) { 1289 return 0; 1290 } 1291 1292 /* 1293 * If we don't have a sigobject for this emulation, create one. 1294 * 1295 * sigobject is an anonymous memory object (just like SYSV shared 1296 * memory) that we keep a permanent reference to and that we map 1297 * in all processes that need this sigcode. The creation is simple, 1298 * we create an object, add a permanent reference to it, map it in 1299 * kernel space, copy out the sigcode to it and unmap it. 1300 * The we map it with PROT_READ|PROT_EXEC into the process just 1301 * the way sys_mmap would map it. 1302 */ 1303 1304 uobj = *e->e_sigobject; 1305 if (uobj == NULL) { 1306 uobj = uao_create(sz, 0); 1307 (*uobj->pgops->pgo_reference)(uobj); 1308 va = vm_map_min(kernel_map); 1309 if ((error = uvm_map(kernel_map, &va, round_page(sz), 1310 uobj, 0, 0, 1311 UVM_MAPFLAG(UVM_PROT_RW, UVM_PROT_RW, 1312 UVM_INH_SHARE, UVM_ADV_RANDOM, 0)))) { 1313 printf("kernel mapping failed %d\n", error); 1314 (*uobj->pgops->pgo_detach)(uobj); 1315 return (error); 1316 } 1317 memcpy((void *)va, e->e_sigcode, sz); 1318 #ifdef PMAP_NEED_PROCWR 1319 pmap_procwr(&proc0, va, sz); 1320 #endif 1321 uvm_unmap(kernel_map, va, va + round_page(sz)); 1322 *e->e_sigobject = uobj; 1323 } 1324 1325 /* Just a hint to uvm_map where to put it. */ 1326 va = VM_DEFAULT_ADDRESS(p->p_vmspace->vm_daddr, round_page(sz)); 1327 (*uobj->pgops->pgo_reference)(uobj); 1328 error = uvm_map(&p->p_vmspace->vm_map, &va, round_page(sz), 1329 uobj, 0, 0, 1330 UVM_MAPFLAG(UVM_PROT_RX, UVM_PROT_RX, UVM_INH_SHARE, 1331 UVM_ADV_RANDOM, 0)); 1332 if (error) { 1333 (*uobj->pgops->pgo_detach)(uobj); 1334 return (error); 1335 } 1336 p->p_sigctx.ps_sigcode = (void *)va; 1337 return (0); 1338 } 1339