1 /* $NetBSD: kern_exec.c,v 1.152 2002/04/23 15:11:25 christos 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 1.152 2002/04/23 15:11:25 christos Exp $"); 37 38 #include "opt_ktrace.h" 39 #include "opt_syscall_debug.h" 40 41 #include <sys/param.h> 42 #include <sys/systm.h> 43 #include <sys/filedesc.h> 44 #include <sys/kernel.h> 45 #include <sys/proc.h> 46 #include <sys/mount.h> 47 #include <sys/malloc.h> 48 #include <sys/namei.h> 49 #include <sys/vnode.h> 50 #include <sys/file.h> 51 #include <sys/acct.h> 52 #include <sys/exec.h> 53 #include <sys/ktrace.h> 54 #include <sys/resourcevar.h> 55 #include <sys/wait.h> 56 #include <sys/mman.h> 57 #include <sys/signalvar.h> 58 #include <sys/stat.h> 59 #include <sys/syscall.h> 60 61 #include <sys/syscallargs.h> 62 63 #include <uvm/uvm_extern.h> 64 65 #include <machine/cpu.h> 66 #include <machine/reg.h> 67 68 #ifdef DEBUG_EXEC 69 #define DPRINTF(a) uprintf a 70 #else 71 #define DPRINTF(a) 72 #endif /* DEBUG_EXEC */ 73 74 /* 75 * Exec function switch: 76 * 77 * Note that each makecmds function is responsible for loading the 78 * exec package with the necessary functions for any exec-type-specific 79 * handling. 80 * 81 * Functions for specific exec types should be defined in their own 82 * header file. 83 */ 84 extern const struct execsw execsw_builtin[]; 85 extern int nexecs_builtin; 86 static const struct execsw **execsw = NULL; 87 static int nexecs; 88 89 int exec_maxhdrsz; /* must not be static - netbsd32 needs it */ 90 91 #ifdef LKM 92 /* list of supported emulations */ 93 static 94 LIST_HEAD(emlist_head, emul_entry) el_head = LIST_HEAD_INITIALIZER(el_head); 95 struct emul_entry { 96 LIST_ENTRY(emul_entry) el_list; 97 const struct emul *el_emul; 98 int ro_entry; 99 }; 100 101 /* list of dynamically loaded execsw entries */ 102 static 103 LIST_HEAD(execlist_head, exec_entry) ex_head = LIST_HEAD_INITIALIZER(ex_head); 104 struct exec_entry { 105 LIST_ENTRY(exec_entry) ex_list; 106 const struct execsw *es; 107 }; 108 109 /* structure used for building execw[] */ 110 struct execsw_entry { 111 struct execsw_entry *next; 112 const struct execsw *es; 113 }; 114 #endif /* LKM */ 115 116 /* NetBSD emul struct */ 117 extern char sigcode[], esigcode[]; 118 #ifdef SYSCALL_DEBUG 119 extern const char * const syscallnames[]; 120 #endif 121 #ifdef __HAVE_SYSCALL_INTERN 122 void syscall_intern(struct proc *); 123 #else 124 void syscall(void); 125 #endif 126 127 const struct emul emul_netbsd = { 128 "netbsd", 129 NULL, /* emulation path */ 130 #ifndef __HAVE_MINIMAL_EMUL 131 EMUL_HAS_SYS___syscall, 132 NULL, 133 SYS_syscall, 134 SYS_MAXSYSCALL, 135 #endif 136 sysent, 137 #ifdef SYSCALL_DEBUG 138 syscallnames, 139 #else 140 NULL, 141 #endif 142 sendsig, 143 trapsignal, 144 sigcode, 145 esigcode, 146 setregs, 147 NULL, 148 NULL, 149 NULL, 150 #ifdef __HAVE_SYSCALL_INTERN 151 syscall_intern, 152 #else 153 syscall, 154 #endif 155 }; 156 157 #ifdef LKM 158 /* 159 * Exec lock. Used to control access to execsw[] structures. 160 * This must not be static so that netbsd32 can access it, too. 161 */ 162 struct lock exec_lock; 163 164 static void link_es(struct execsw_entry **, const struct execsw *); 165 #endif /* LKM */ 166 167 /* 168 * check exec: 169 * given an "executable" described in the exec package's namei info, 170 * see what we can do with it. 171 * 172 * ON ENTRY: 173 * exec package with appropriate namei info 174 * proc pointer of exec'ing proc 175 * NO SELF-LOCKED VNODES 176 * 177 * ON EXIT: 178 * error: nothing held, etc. exec header still allocated. 179 * ok: filled exec package, executable's vnode (unlocked). 180 * 181 * EXEC SWITCH ENTRY: 182 * Locked vnode to check, exec package, proc. 183 * 184 * EXEC SWITCH EXIT: 185 * ok: return 0, filled exec package, executable's vnode (unlocked). 186 * error: destructive: 187 * everything deallocated execept exec header. 188 * non-destructive: 189 * error code, executable's vnode (unlocked), 190 * exec header unmodified. 191 */ 192 int 193 check_exec(struct proc *p, struct exec_package *epp) 194 { 195 int error, i; 196 struct vnode *vp; 197 struct nameidata *ndp; 198 size_t resid; 199 200 ndp = epp->ep_ndp; 201 ndp->ni_cnd.cn_nameiop = LOOKUP; 202 ndp->ni_cnd.cn_flags = FOLLOW | LOCKLEAF | SAVENAME; 203 /* first get the vnode */ 204 if ((error = namei(ndp)) != 0) 205 return error; 206 epp->ep_vp = vp = ndp->ni_vp; 207 208 /* check access and type */ 209 if (vp->v_type != VREG) { 210 error = EACCES; 211 goto bad1; 212 } 213 if ((error = VOP_ACCESS(vp, VEXEC, p->p_ucred, p)) != 0) 214 goto bad1; 215 216 /* get attributes */ 217 if ((error = VOP_GETATTR(vp, epp->ep_vap, p->p_ucred, p)) != 0) 218 goto bad1; 219 220 /* Check mount point */ 221 if (vp->v_mount->mnt_flag & MNT_NOEXEC) { 222 error = EACCES; 223 goto bad1; 224 } 225 if (vp->v_mount->mnt_flag & MNT_NOSUID) 226 epp->ep_vap->va_mode &= ~(S_ISUID | S_ISGID); 227 228 /* try to open it */ 229 if ((error = VOP_OPEN(vp, FREAD, p->p_ucred, p)) != 0) 230 goto bad1; 231 232 /* unlock vp, since we need it unlocked from here on out. */ 233 VOP_UNLOCK(vp, 0); 234 235 /* now we have the file, get the exec header */ 236 uvn_attach(vp, VM_PROT_READ); 237 error = vn_rdwr(UIO_READ, vp, epp->ep_hdr, epp->ep_hdrlen, 0, 238 UIO_SYSSPACE, 0, p->p_ucred, &resid, p); 239 if (error) 240 goto bad2; 241 epp->ep_hdrvalid = epp->ep_hdrlen - resid; 242 243 /* 244 * Set up default address space limits. Can be overridden 245 * by individual exec packages. 246 * 247 * XXX probably shoul be all done in the exec pakages. 248 */ 249 epp->ep_vm_minaddr = VM_MIN_ADDRESS; 250 epp->ep_vm_maxaddr = VM_MAXUSER_ADDRESS; 251 /* 252 * set up the vmcmds for creation of the process 253 * address space 254 */ 255 error = ENOEXEC; 256 for (i = 0; i < nexecs && error != 0; i++) { 257 int newerror; 258 259 epp->ep_esch = execsw[i]; 260 newerror = (*execsw[i]->es_check)(p, epp); 261 /* make sure the first "interesting" error code is saved. */ 262 if (!newerror || error == ENOEXEC) 263 error = newerror; 264 265 /* if es_check call was successful, update epp->ep_es */ 266 if (!newerror && (epp->ep_flags & EXEC_HASES) == 0) 267 epp->ep_es = execsw[i]; 268 269 if (epp->ep_flags & EXEC_DESTR && error != 0) 270 return error; 271 } 272 if (!error) { 273 /* check that entry point is sane */ 274 if (epp->ep_entry > VM_MAXUSER_ADDRESS) 275 error = ENOEXEC; 276 277 /* check limits */ 278 if ((epp->ep_tsize > MAXTSIZ) || 279 (epp->ep_dsize > p->p_rlimit[RLIMIT_DATA].rlim_cur)) 280 error = ENOMEM; 281 282 if (!error) 283 return (0); 284 } 285 286 /* 287 * free any vmspace-creation commands, 288 * and release their references 289 */ 290 kill_vmcmds(&epp->ep_vmcmds); 291 292 bad2: 293 /* 294 * close and release the vnode, restore the old one, free the 295 * pathname buf, and punt. 296 */ 297 vn_lock(vp, LK_EXCLUSIVE | LK_RETRY); 298 VOP_CLOSE(vp, FREAD, p->p_ucred, p); 299 vput(vp); 300 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 301 return error; 302 303 bad1: 304 /* 305 * free the namei pathname buffer, and put the vnode 306 * (which we don't yet have open). 307 */ 308 vput(vp); /* was still locked */ 309 PNBUF_PUT(ndp->ni_cnd.cn_pnbuf); 310 return error; 311 } 312 313 /* 314 * exec system call 315 */ 316 /* ARGSUSED */ 317 int 318 sys_execve(struct proc *p, void *v, register_t *retval) 319 { 320 struct sys_execve_args /* { 321 syscallarg(const char *) path; 322 syscallarg(char * const *) argp; 323 syscallarg(char * const *) envp; 324 } */ *uap = v; 325 int error, i; 326 struct exec_package pack; 327 struct nameidata nid; 328 struct vattr attr; 329 struct ucred *cred; 330 char *argp; 331 char * const *cpp; 332 char *dp, *sp; 333 long argc, envc; 334 size_t len; 335 char *stack; 336 struct ps_strings arginfo; 337 struct vmspace *vm; 338 char **tmpfap; 339 int szsigcode; 340 struct exec_vmcmd *base_vcp; 341 342 /* 343 * Lock the process and set the P_INEXEC flag to indicate that 344 * it should be left alone until we're done here. This is 345 * necessary to avoid race conditions - e.g. in ptrace() - 346 * that might allow a local user to illicitly obtain elevated 347 * privileges. 348 */ 349 p->p_flag |= P_INEXEC; 350 351 cred = p->p_ucred; 352 base_vcp = NULL; 353 /* 354 * Init the namei data to point the file user's program name. 355 * This is done here rather than in check_exec(), so that it's 356 * possible to override this settings if any of makecmd/probe 357 * functions call check_exec() recursively - for example, 358 * see exec_script_makecmds(). 359 */ 360 NDINIT(&nid, LOOKUP, NOFOLLOW, UIO_USERSPACE, SCARG(uap, path), p); 361 362 /* 363 * initialize the fields of the exec package. 364 */ 365 pack.ep_name = SCARG(uap, path); 366 pack.ep_hdr = malloc(exec_maxhdrsz, M_EXEC, M_WAITOK); 367 pack.ep_hdrlen = exec_maxhdrsz; 368 pack.ep_hdrvalid = 0; 369 pack.ep_ndp = &nid; 370 pack.ep_emul_arg = NULL; 371 pack.ep_vmcmds.evs_cnt = 0; 372 pack.ep_vmcmds.evs_used = 0; 373 pack.ep_vap = &attr; 374 pack.ep_flags = 0; 375 376 #ifdef LKM 377 lockmgr(&exec_lock, LK_SHARED, NULL); 378 #endif 379 380 /* see if we can run it. */ 381 if ((error = check_exec(p, &pack)) != 0) 382 goto freehdr; 383 384 /* XXX -- THE FOLLOWING SECTION NEEDS MAJOR CLEANUP */ 385 386 /* allocate an argument buffer */ 387 argp = (char *) uvm_km_valloc_wait(exec_map, NCARGS); 388 #ifdef DIAGNOSTIC 389 if (argp == (vaddr_t) 0) 390 panic("execve: argp == NULL"); 391 #endif 392 dp = argp; 393 argc = 0; 394 395 /* copy the fake args list, if there's one, freeing it as we go */ 396 if (pack.ep_flags & EXEC_HASARGL) { 397 tmpfap = pack.ep_fa; 398 while (*tmpfap != NULL) { 399 char *cp; 400 401 cp = *tmpfap; 402 while (*cp) 403 *dp++ = *cp++; 404 dp++; 405 406 FREE(*tmpfap, M_EXEC); 407 tmpfap++; argc++; 408 } 409 FREE(pack.ep_fa, M_EXEC); 410 pack.ep_flags &= ~EXEC_HASARGL; 411 } 412 413 /* Now get argv & environment */ 414 if (!(cpp = SCARG(uap, argp))) { 415 error = EINVAL; 416 goto bad; 417 } 418 419 if (pack.ep_flags & EXEC_SKIPARG) 420 cpp++; 421 422 while (1) { 423 len = argp + ARG_MAX - dp; 424 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 425 goto bad; 426 if (!sp) 427 break; 428 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 429 if (error == ENAMETOOLONG) 430 error = E2BIG; 431 goto bad; 432 } 433 dp += len; 434 cpp++; 435 argc++; 436 } 437 438 envc = 0; 439 /* environment need not be there */ 440 if ((cpp = SCARG(uap, envp)) != NULL ) { 441 while (1) { 442 len = argp + ARG_MAX - dp; 443 if ((error = copyin(cpp, &sp, sizeof(sp))) != 0) 444 goto bad; 445 if (!sp) 446 break; 447 if ((error = copyinstr(sp, dp, len, &len)) != 0) { 448 if (error == ENAMETOOLONG) 449 error = E2BIG; 450 goto bad; 451 } 452 dp += len; 453 cpp++; 454 envc++; 455 } 456 } 457 458 dp = (char *) ALIGN(dp); 459 460 szsigcode = pack.ep_es->es_emul->e_esigcode - 461 pack.ep_es->es_emul->e_sigcode; 462 463 /* Now check if args & environ fit into new stack */ 464 if (pack.ep_flags & EXEC_32) 465 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 466 sizeof(int) + sizeof(int) + dp + STACKGAPLEN + 467 szsigcode + sizeof(struct ps_strings)) - argp; 468 else 469 len = ((argc + envc + 2 + pack.ep_es->es_arglen) * 470 sizeof(char *) + sizeof(int) + dp + STACKGAPLEN + 471 szsigcode + sizeof(struct ps_strings)) - argp; 472 473 len = ALIGN(len); /* make the stack "safely" aligned */ 474 475 if (len > pack.ep_ssize) { /* in effect, compare to initial limit */ 476 error = ENOMEM; 477 goto bad; 478 } 479 480 /* adjust "active stack depth" for process VSZ */ 481 pack.ep_ssize = len; /* maybe should go elsewhere, but... */ 482 483 /* 484 * Do whatever is necessary to prepare the address space 485 * for remapping. Note that this might replace the current 486 * vmspace with another! 487 */ 488 uvmspace_exec(p, pack.ep_vm_minaddr, pack.ep_vm_maxaddr); 489 490 /* Now map address space */ 491 vm = p->p_vmspace; 492 vm->vm_taddr = (char *) pack.ep_taddr; 493 vm->vm_tsize = btoc(pack.ep_tsize); 494 vm->vm_daddr = (char *) pack.ep_daddr; 495 vm->vm_dsize = btoc(pack.ep_dsize); 496 vm->vm_ssize = btoc(pack.ep_ssize); 497 vm->vm_maxsaddr = (char *) pack.ep_maxsaddr; 498 vm->vm_minsaddr = (char *) pack.ep_minsaddr; 499 500 /* create the new process's VM space by running the vmcmds */ 501 #ifdef DIAGNOSTIC 502 if (pack.ep_vmcmds.evs_used == 0) 503 panic("execve: no vmcmds"); 504 #endif 505 for (i = 0; i < pack.ep_vmcmds.evs_used && !error; i++) { 506 struct exec_vmcmd *vcp; 507 508 vcp = &pack.ep_vmcmds.evs_cmds[i]; 509 if (vcp->ev_flags & VMCMD_RELATIVE) { 510 #ifdef DIAGNOSTIC 511 if (base_vcp == NULL) 512 panic("execve: relative vmcmd with no base"); 513 if (vcp->ev_flags & VMCMD_BASE) 514 panic("execve: illegal base & relative vmcmd"); 515 #endif 516 vcp->ev_addr += base_vcp->ev_addr; 517 } 518 error = (*vcp->ev_proc)(p, vcp); 519 #ifdef DEBUG_EXEC 520 if (error) { 521 int j; 522 struct exec_vmcmd *vp = &pack.ep_vmcmds.evs_cmds[0]; 523 for (j = 0; j <= i; j++) 524 uprintf( 525 "vmcmd[%d] = %#lx/%#lx fd@%#lx prot=0%o flags=%d\n", 526 j, vp[j].ev_addr, vp[j].ev_len, 527 vp[j].ev_offset, vp[j].ev_prot, 528 vp[j].ev_flags); 529 } 530 #endif /* DEBUG_EXEC */ 531 if (vcp->ev_flags & VMCMD_BASE) 532 base_vcp = vcp; 533 } 534 535 /* free the vmspace-creation commands, and release their references */ 536 kill_vmcmds(&pack.ep_vmcmds); 537 538 /* if an error happened, deallocate and punt */ 539 if (error) { 540 DPRINTF(("execve: vmcmd %i failed: %d\n", i - 1, error)); 541 goto exec_abort; 542 } 543 544 /* remember information about the process */ 545 arginfo.ps_nargvstr = argc; 546 arginfo.ps_nenvstr = envc; 547 548 stack = (char *) (vm->vm_minsaddr - len); 549 /* Now copy argc, args & environ to new stack */ 550 error = (*pack.ep_es->es_copyargs)(&pack, &arginfo, &stack, argp); 551 if (error) { 552 DPRINTF(("execve: copyargs failed %d\n", error)); 553 goto exec_abort; 554 } 555 /* Move the stack back to original point */ 556 stack = (char *) (vm->vm_minsaddr - len); 557 558 /* fill process ps_strings info */ 559 p->p_psstr = (struct ps_strings *)(vm->vm_minsaddr 560 - sizeof(struct ps_strings)); 561 p->p_psargv = offsetof(struct ps_strings, ps_argvstr); 562 p->p_psnargv = offsetof(struct ps_strings, ps_nargvstr); 563 p->p_psenv = offsetof(struct ps_strings, ps_envstr); 564 p->p_psnenv = offsetof(struct ps_strings, ps_nenvstr); 565 566 /* copy out the process's ps_strings structure */ 567 if ((error = copyout(&arginfo, (char *)p->p_psstr, 568 sizeof(arginfo))) != 0) { 569 DPRINTF(("execve: ps_strings copyout %p->%p size %ld failed\n", 570 &arginfo, (char *)p->p_psstr, (long)sizeof(arginfo))); 571 goto exec_abort; 572 } 573 574 /* copy out the process's signal trapoline code */ 575 if (szsigcode) { 576 if ((error = copyout((char *)pack.ep_es->es_emul->e_sigcode, 577 p->p_sigctx.ps_sigcode = (char *)p->p_psstr - szsigcode, 578 szsigcode)) != 0) { 579 DPRINTF(("execve: sig trampoline copyout failed\n")); 580 goto exec_abort; 581 } 582 #ifdef PMAP_NEED_PROCWR 583 /* This is code. Let the pmap do what is needed. */ 584 pmap_procwr(p, (vaddr_t)p->p_sigctx.ps_sigcode, szsigcode); 585 #endif 586 } 587 588 stopprofclock(p); /* stop profiling */ 589 fdcloseexec(p); /* handle close on exec */ 590 execsigs(p); /* reset catched signals */ 591 p->p_ctxlink = NULL; /* reset ucontext link */ 592 593 /* set command name & other accounting info */ 594 len = min(nid.ni_cnd.cn_namelen, MAXCOMLEN); 595 memcpy(p->p_comm, nid.ni_cnd.cn_nameptr, len); 596 p->p_comm[len] = 0; 597 p->p_acflag &= ~AFORK; 598 599 /* record proc's vnode, for use by procfs and others */ 600 if (p->p_textvp) 601 vrele(p->p_textvp); 602 VREF(pack.ep_vp); 603 p->p_textvp = pack.ep_vp; 604 605 p->p_flag |= P_EXEC; 606 if (p->p_flag & P_PPWAIT) { 607 p->p_flag &= ~P_PPWAIT; 608 wakeup((caddr_t) p->p_pptr); 609 } 610 611 /* 612 * deal with set[ug]id. 613 * MNT_NOSUID has already been used to disable s[ug]id. 614 */ 615 if ((p->p_flag & P_TRACED) == 0 && 616 617 (((attr.va_mode & S_ISUID) != 0 && 618 p->p_ucred->cr_uid != attr.va_uid) || 619 620 ((attr.va_mode & S_ISGID) != 0 && 621 p->p_ucred->cr_gid != attr.va_gid))) { 622 /* 623 * Mark the process as SUGID before we do 624 * anything that might block. 625 */ 626 p_sugid(p); 627 628 /* Make sure file descriptors 0..2 are in use. */ 629 if ((error = fdcheckstd(p)) != 0) 630 goto exec_abort; 631 632 p->p_ucred = crcopy(cred); 633 #ifdef KTRACE 634 /* 635 * If process is being ktraced, turn off - unless 636 * root set it. 637 */ 638 if (p->p_tracep && !(p->p_traceflag & KTRFAC_ROOT)) 639 ktrderef(p); 640 #endif 641 if (attr.va_mode & S_ISUID) 642 p->p_ucred->cr_uid = attr.va_uid; 643 if (attr.va_mode & S_ISGID) 644 p->p_ucred->cr_gid = attr.va_gid; 645 } else 646 p->p_flag &= ~P_SUGID; 647 p->p_cred->p_svuid = p->p_ucred->cr_uid; 648 p->p_cred->p_svgid = p->p_ucred->cr_gid; 649 650 doexechooks(p); 651 652 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 653 654 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 655 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 656 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 657 vput(pack.ep_vp); 658 659 /* setup new registers and do misc. setup. */ 660 (*pack.ep_es->es_emul->e_setregs)(p, &pack, (u_long) stack); 661 if (pack.ep_es->es_setregs) 662 (*pack.ep_es->es_setregs)(p, &pack, (u_long) stack); 663 664 if (p->p_flag & P_TRACED) 665 psignal(p, SIGTRAP); 666 667 free(pack.ep_hdr, M_EXEC); 668 669 /* 670 * Call emulation specific exec hook. This can setup setup per-process 671 * p->p_emuldata or do any other per-process stuff an emulation needs. 672 * 673 * If we are executing process of different emulation than the 674 * original forked process, call e_proc_exit() of the old emulation 675 * first, then e_proc_exec() of new emulation. If the emulation is 676 * same, the exec hook code should deallocate any old emulation 677 * resources held previously by this process. 678 */ 679 if (p->p_emul && p->p_emul->e_proc_exit 680 && p->p_emul != pack.ep_es->es_emul) 681 (*p->p_emul->e_proc_exit)(p); 682 683 /* 684 * Call exec hook. Emulation code may NOT store reference to anything 685 * from &pack. 686 */ 687 if (pack.ep_es->es_emul->e_proc_exec) 688 (*pack.ep_es->es_emul->e_proc_exec)(p, &pack); 689 690 /* update p_emul, the old value is no longer needed */ 691 p->p_emul = pack.ep_es->es_emul; 692 693 /* ...and the same for p_execsw */ 694 p->p_execsw = pack.ep_es; 695 696 #ifdef __HAVE_SYSCALL_INTERN 697 (*p->p_emul->e_syscall_intern)(p); 698 #endif 699 #ifdef KTRACE 700 if (KTRPOINT(p, KTR_EMUL)) 701 ktremul(p); 702 #endif 703 704 #ifdef LKM 705 lockmgr(&exec_lock, LK_RELEASE, NULL); 706 #endif 707 p->p_flag &= ~P_INEXEC; 708 return (EJUSTRETURN); 709 710 bad: 711 p->p_flag &= ~P_INEXEC; 712 /* free the vmspace-creation commands, and release their references */ 713 kill_vmcmds(&pack.ep_vmcmds); 714 /* kill any opened file descriptor, if necessary */ 715 if (pack.ep_flags & EXEC_HASFD) { 716 pack.ep_flags &= ~EXEC_HASFD; 717 (void) fdrelease(p, pack.ep_fd); 718 } 719 /* close and put the exec'd file */ 720 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 721 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 722 vput(pack.ep_vp); 723 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 724 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 725 726 freehdr: 727 p->p_flag &= ~P_INEXEC; 728 #ifdef LKM 729 lockmgr(&exec_lock, LK_RELEASE, NULL); 730 #endif 731 732 free(pack.ep_hdr, M_EXEC); 733 return error; 734 735 exec_abort: 736 p->p_flag &= ~P_INEXEC; 737 #ifdef LKM 738 lockmgr(&exec_lock, LK_RELEASE, NULL); 739 #endif 740 741 /* 742 * the old process doesn't exist anymore. exit gracefully. 743 * get rid of the (new) address space we have created, if any, get rid 744 * of our namei data and vnode, and exit noting failure 745 */ 746 uvm_deallocate(&vm->vm_map, VM_MIN_ADDRESS, 747 VM_MAXUSER_ADDRESS - VM_MIN_ADDRESS); 748 if (pack.ep_emul_arg) 749 FREE(pack.ep_emul_arg, M_TEMP); 750 PNBUF_PUT(nid.ni_cnd.cn_pnbuf); 751 vn_lock(pack.ep_vp, LK_EXCLUSIVE | LK_RETRY); 752 VOP_CLOSE(pack.ep_vp, FREAD, cred, p); 753 vput(pack.ep_vp); 754 uvm_km_free_wakeup(exec_map, (vaddr_t) argp, NCARGS); 755 free(pack.ep_hdr, M_EXEC); 756 exit1(p, W_EXITCODE(error, SIGABRT)); 757 758 /* NOTREACHED */ 759 return 0; 760 } 761 762 763 int 764 copyargs(struct exec_package *pack, struct ps_strings *arginfo, 765 char **stackp, void *argp) 766 { 767 char **cpp, *dp, *sp; 768 size_t len; 769 void *nullp; 770 long argc, envc; 771 int error; 772 773 cpp = (char **)*stackp; 774 nullp = NULL; 775 argc = arginfo->ps_nargvstr; 776 envc = arginfo->ps_nenvstr; 777 if ((error = copyout(&argc, cpp++, sizeof(argc))) != 0) 778 return error; 779 780 dp = (char *) (cpp + argc + envc + 2 + pack->ep_es->es_arglen); 781 sp = argp; 782 783 /* XXX don't copy them out, remap them! */ 784 arginfo->ps_argvstr = cpp; /* remember location of argv for later */ 785 786 for (; --argc >= 0; sp += len, dp += len) 787 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 788 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 789 return error; 790 791 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 792 return error; 793 794 arginfo->ps_envstr = cpp; /* remember location of envp for later */ 795 796 for (; --envc >= 0; sp += len, dp += len) 797 if ((error = copyout(&dp, cpp++, sizeof(dp))) != 0 || 798 (error = copyoutstr(sp, dp, ARG_MAX, &len)) != 0) 799 return error; 800 801 if ((error = copyout(&nullp, cpp++, sizeof(nullp))) != 0) 802 return error; 803 804 *stackp = (char *)cpp; 805 return 0; 806 } 807 808 #ifdef LKM 809 /* 810 * Find an emulation of given name in list of emulations. 811 * Needs to be called with the exec_lock held. 812 */ 813 const struct emul * 814 emul_search(const char *name) 815 { 816 struct emul_entry *it; 817 818 LIST_FOREACH(it, &el_head, el_list) { 819 if (strcmp(name, it->el_emul->e_name) == 0) 820 return it->el_emul; 821 } 822 823 return NULL; 824 } 825 826 /* 827 * Add an emulation to list, if it's not there already. 828 */ 829 int 830 emul_register(const struct emul *emul, int ro_entry) 831 { 832 struct emul_entry *ee; 833 int error; 834 835 error = 0; 836 lockmgr(&exec_lock, LK_SHARED, NULL); 837 838 if (emul_search(emul->e_name)) { 839 error = EEXIST; 840 goto out; 841 } 842 843 MALLOC(ee, struct emul_entry *, sizeof(struct emul_entry), 844 M_EXEC, M_WAITOK); 845 ee->el_emul = emul; 846 ee->ro_entry = ro_entry; 847 LIST_INSERT_HEAD(&el_head, ee, el_list); 848 849 out: 850 lockmgr(&exec_lock, LK_RELEASE, NULL); 851 return error; 852 } 853 854 /* 855 * Remove emulation with name 'name' from list of supported emulations. 856 */ 857 int 858 emul_unregister(const char *name) 859 { 860 const struct proclist_desc *pd; 861 struct emul_entry *it; 862 int i, error; 863 struct proc *ptmp; 864 865 error = 0; 866 lockmgr(&exec_lock, LK_SHARED, NULL); 867 868 LIST_FOREACH(it, &el_head, el_list) { 869 if (strcmp(it->el_emul->e_name, name) == 0) 870 break; 871 } 872 873 if (!it) { 874 error = ENOENT; 875 goto out; 876 } 877 878 if (it->ro_entry) { 879 error = EBUSY; 880 goto out; 881 } 882 883 /* test if any execw[] entry is still using this */ 884 for(i=0; i < nexecs; i++) { 885 if (execsw[i]->es_emul == it->el_emul) { 886 error = EBUSY; 887 goto out; 888 } 889 } 890 891 /* 892 * Test if any process is running under this emulation - since 893 * emul_unregister() is running quite sendomly, it's better 894 * to do expensive check here than to use any locking. 895 */ 896 proclist_lock_read(); 897 for (pd = proclists; pd->pd_list != NULL && !error; pd++) { 898 LIST_FOREACH(ptmp, pd->pd_list, p_list) { 899 if (ptmp->p_emul == it->el_emul) { 900 error = EBUSY; 901 break; 902 } 903 } 904 } 905 proclist_unlock_read(); 906 907 if (error) 908 goto out; 909 910 911 /* entry is not used, remove it */ 912 LIST_REMOVE(it, el_list); 913 FREE(it, M_EXEC); 914 915 out: 916 lockmgr(&exec_lock, LK_RELEASE, NULL); 917 return error; 918 } 919 920 /* 921 * Add execsw[] entry. 922 */ 923 int 924 exec_add(struct execsw *esp, const char *e_name) 925 { 926 struct exec_entry *it; 927 int error; 928 929 error = 0; 930 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 931 932 if (!esp->es_emul) { 933 esp->es_emul = emul_search(e_name); 934 if (!esp->es_emul) { 935 error = ENOENT; 936 goto out; 937 } 938 } 939 940 LIST_FOREACH(it, &ex_head, ex_list) { 941 /* assume tuple (makecmds, probe_func, emulation) is unique */ 942 if (it->es->es_check == esp->es_check 943 && it->es->u.elf_probe_func == esp->u.elf_probe_func 944 && it->es->es_emul == esp->es_emul) { 945 error = EEXIST; 946 goto out; 947 } 948 } 949 950 /* if we got here, the entry doesn't exist yet */ 951 MALLOC(it, struct exec_entry *, sizeof(struct exec_entry), 952 M_EXEC, M_WAITOK); 953 it->es = esp; 954 LIST_INSERT_HEAD(&ex_head, it, ex_list); 955 956 /* update execsw[] */ 957 exec_init(0); 958 959 out: 960 lockmgr(&exec_lock, LK_RELEASE, NULL); 961 return error; 962 } 963 964 /* 965 * Remove execsw[] entry. 966 */ 967 int 968 exec_remove(const struct execsw *esp) 969 { 970 struct exec_entry *it; 971 int error; 972 973 error = 0; 974 lockmgr(&exec_lock, LK_EXCLUSIVE, NULL); 975 976 LIST_FOREACH(it, &ex_head, ex_list) { 977 /* assume tuple (makecmds, probe_func, emulation) is unique */ 978 if (it->es->es_check == esp->es_check 979 && it->es->u.elf_probe_func == esp->u.elf_probe_func 980 && it->es->es_emul == esp->es_emul) 981 break; 982 } 983 if (!it) { 984 error = ENOENT; 985 goto out; 986 } 987 988 /* remove item from list and free resources */ 989 LIST_REMOVE(it, ex_list); 990 FREE(it, M_EXEC); 991 992 /* update execsw[] */ 993 exec_init(0); 994 995 out: 996 lockmgr(&exec_lock, LK_RELEASE, NULL); 997 return error; 998 } 999 1000 static void 1001 link_es(struct execsw_entry **listp, const struct execsw *esp) 1002 { 1003 struct execsw_entry *et, *e1; 1004 1005 MALLOC(et, struct execsw_entry *, sizeof(struct execsw_entry), 1006 M_TEMP, M_WAITOK); 1007 et->next = NULL; 1008 et->es = esp; 1009 if (*listp == NULL) { 1010 *listp = et; 1011 return; 1012 } 1013 1014 switch(et->es->es_prio) { 1015 case EXECSW_PRIO_FIRST: 1016 /* put new entry as the first */ 1017 et->next = *listp; 1018 *listp = et; 1019 break; 1020 case EXECSW_PRIO_ANY: 1021 /* put new entry after all *_FIRST and *_ANY entries */ 1022 for(e1 = *listp; e1->next 1023 && e1->next->es->es_prio != EXECSW_PRIO_LAST; 1024 e1 = e1->next); 1025 et->next = e1->next; 1026 e1->next = et; 1027 break; 1028 case EXECSW_PRIO_LAST: 1029 /* put new entry as the last one */ 1030 for(e1 = *listp; e1->next; e1 = e1->next); 1031 e1->next = et; 1032 break; 1033 default: 1034 #ifdef DIAGNOSTIC 1035 panic("execw[] entry with unknown priority %d found\n", 1036 et->es->es_prio); 1037 #endif 1038 break; 1039 } 1040 } 1041 1042 /* 1043 * Initialize exec structures. If init_boot is true, also does necessary 1044 * one-time initialization (it's called from main() that way). 1045 * Once system is multiuser, this should be called with exec_lock held, 1046 * i.e. via exec_{add|remove}(). 1047 */ 1048 int 1049 exec_init(int init_boot) 1050 { 1051 const struct execsw **new_es, * const *old_es; 1052 struct execsw_entry *list, *e1; 1053 struct exec_entry *e2; 1054 int i, es_sz; 1055 1056 if (init_boot) { 1057 /* do one-time initializations */ 1058 lockinit(&exec_lock, PWAIT, "execlck", 0, 0); 1059 1060 /* register compiled-in emulations */ 1061 for(i=0; i < nexecs_builtin; i++) { 1062 if (execsw_builtin[i].es_emul) 1063 emul_register(execsw_builtin[i].es_emul, 1); 1064 } 1065 #ifdef DIAGNOSTIC 1066 if (i == 0) 1067 panic("no emulations found in execsw_builtin[]\n"); 1068 #endif 1069 } 1070 1071 /* 1072 * Build execsw[] array from builtin entries and entries added 1073 * at runtime. 1074 */ 1075 list = NULL; 1076 for(i=0; i < nexecs_builtin; i++) 1077 link_es(&list, &execsw_builtin[i]); 1078 1079 /* Add dynamically loaded entries */ 1080 es_sz = nexecs_builtin; 1081 LIST_FOREACH(e2, &ex_head, ex_list) { 1082 link_es(&list, e2->es); 1083 es_sz++; 1084 } 1085 1086 /* 1087 * Now that we have sorted all execw entries, create new execsw[] 1088 * and free no longer needed memory in the process. 1089 */ 1090 new_es = malloc(es_sz * sizeof(struct execsw *), M_EXEC, M_WAITOK); 1091 for(i=0; list; i++) { 1092 new_es[i] = list->es; 1093 e1 = list->next; 1094 FREE(list, M_TEMP); 1095 list = e1; 1096 } 1097 1098 /* 1099 * New execsw[] array built, now replace old execsw[] and free 1100 * used memory. 1101 */ 1102 old_es = execsw; 1103 execsw = new_es; 1104 nexecs = es_sz; 1105 if (old_es) 1106 free((void *)old_es, M_EXEC); 1107 1108 /* 1109 * Figure out the maximum size of an exec header. 1110 */ 1111 exec_maxhdrsz = 0; 1112 for (i = 0; i < nexecs; i++) { 1113 if (execsw[i]->es_hdrsz > exec_maxhdrsz) 1114 exec_maxhdrsz = execsw[i]->es_hdrsz; 1115 } 1116 1117 return 0; 1118 } 1119 #endif 1120 1121 #ifndef LKM 1122 /* 1123 * Simplified exec_init() for kernels without LKMs. Only initialize 1124 * exec_maxhdrsz and execsw[]. 1125 */ 1126 int 1127 exec_init(int init_boot) 1128 { 1129 int i; 1130 1131 #ifdef DIAGNOSTIC 1132 if (!init_boot) 1133 panic("exec_init(): called with init_boot == 0"); 1134 #endif 1135 1136 /* do one-time initializations */ 1137 nexecs = nexecs_builtin; 1138 execsw = malloc(nexecs*sizeof(struct execsw *), M_EXEC, M_WAITOK); 1139 1140 /* 1141 * Fill in execsw[] and figure out the maximum size of an exec header. 1142 */ 1143 exec_maxhdrsz = 0; 1144 for(i=0; i < nexecs; i++) { 1145 execsw[i] = &execsw_builtin[i]; 1146 if (execsw_builtin[i].es_hdrsz > exec_maxhdrsz) 1147 exec_maxhdrsz = execsw_builtin[i].es_hdrsz; 1148 } 1149 1150 return 0; 1151 1152 } 1153 #endif /* !LKM */ 1154